[SCSI] iscsi class, iscsi_tcp/iser: add host arg to session creation
[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         unsigned total_in_length = scsi_in(sc)->length;
532
533         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
534         if (tcp_conn->in.datalen == 0)
535                 return 0;
536
537         if (tcp_ctask->exp_datasn != datasn) {
538                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->datasn(%d)\n",
539                           __FUNCTION__, tcp_ctask->exp_datasn, datasn);
540                 return ISCSI_ERR_DATASN;
541         }
542
543         tcp_ctask->exp_datasn++;
544
545         tcp_ctask->data_offset = be32_to_cpu(rhdr->offset);
546         if (tcp_ctask->data_offset + tcp_conn->in.datalen > total_in_length) {
547                 debug_tcp("%s: data_offset(%d) + data_len(%d) > total_length_in(%d)\n",
548                           __FUNCTION__, tcp_ctask->data_offset,
549                           tcp_conn->in.datalen, total_in_length);
550                 return ISCSI_ERR_DATA_OFFSET;
551         }
552
553         if (rhdr->flags & ISCSI_FLAG_DATA_STATUS) {
554                 sc->result = (DID_OK << 16) | rhdr->cmd_status;
555                 conn->exp_statsn = be32_to_cpu(rhdr->statsn) + 1;
556                 if (rhdr->flags & (ISCSI_FLAG_DATA_UNDERFLOW |
557                                    ISCSI_FLAG_DATA_OVERFLOW)) {
558                         int res_count = be32_to_cpu(rhdr->residual_count);
559
560                         if (res_count > 0 &&
561                             (rhdr->flags & ISCSI_FLAG_CMD_OVERFLOW ||
562                              res_count <= total_in_length))
563                                 scsi_in(sc)->resid = res_count;
564                         else
565                                 sc->result = (DID_BAD_TARGET << 16) |
566                                         rhdr->cmd_status;
567                 }
568         }
569
570         conn->datain_pdus_cnt++;
571         return 0;
572 }
573
574 /**
575  * iscsi_solicit_data_init - initialize first Data-Out
576  * @conn: iscsi connection
577  * @ctask: scsi command task
578  * @r2t: R2T info
579  *
580  * Notes:
581  *      Initialize first Data-Out within this R2T sequence and finds
582  *      proper data_offset within this SCSI command.
583  *
584  *      This function is called with connection lock taken.
585  **/
586 static void
587 iscsi_solicit_data_init(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
588                         struct iscsi_r2t_info *r2t)
589 {
590         struct iscsi_data *hdr;
591
592         hdr = &r2t->dtask.hdr;
593         memset(hdr, 0, sizeof(struct iscsi_data));
594         hdr->ttt = r2t->ttt;
595         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
596         r2t->solicit_datasn++;
597         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
598         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
599         hdr->itt = ctask->hdr->itt;
600         hdr->exp_statsn = r2t->exp_statsn;
601         hdr->offset = cpu_to_be32(r2t->data_offset);
602         if (r2t->data_length > conn->max_xmit_dlength) {
603                 hton24(hdr->dlength, conn->max_xmit_dlength);
604                 r2t->data_count = conn->max_xmit_dlength;
605                 hdr->flags = 0;
606         } else {
607                 hton24(hdr->dlength, r2t->data_length);
608                 r2t->data_count = r2t->data_length;
609                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
610         }
611         conn->dataout_pdus_cnt++;
612
613         r2t->sent = 0;
614 }
615
616 /**
617  * iscsi_r2t_rsp - iSCSI R2T Response processing
618  * @conn: iscsi connection
619  * @ctask: scsi command task
620  **/
621 static int
622 iscsi_r2t_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
623 {
624         struct iscsi_r2t_info *r2t;
625         struct iscsi_session *session = conn->session;
626         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
627         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
628         struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
629         int r2tsn = be32_to_cpu(rhdr->r2tsn);
630         int rc;
631
632         if (tcp_conn->in.datalen) {
633                 iscsi_conn_printk(KERN_ERR, conn,
634                                   "invalid R2t with datalen %d\n",
635                                   tcp_conn->in.datalen);
636                 return ISCSI_ERR_DATALEN;
637         }
638
639         if (tcp_ctask->exp_datasn != r2tsn){
640                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
641                           __FUNCTION__, tcp_ctask->exp_datasn, r2tsn);
642                 return ISCSI_ERR_R2TSN;
643         }
644
645         /* fill-in new R2T associated with the task */
646         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
647
648         if (!ctask->sc || session->state != ISCSI_STATE_LOGGED_IN) {
649                 iscsi_conn_printk(KERN_INFO, conn,
650                                   "dropping R2T itt %d in recovery.\n",
651                                   ctask->itt);
652                 return 0;
653         }
654
655         rc = __kfifo_get(tcp_ctask->r2tpool.queue, (void*)&r2t, sizeof(void*));
656         BUG_ON(!rc);
657
658         r2t->exp_statsn = rhdr->statsn;
659         r2t->data_length = be32_to_cpu(rhdr->data_length);
660         if (r2t->data_length == 0) {
661                 iscsi_conn_printk(KERN_ERR, conn,
662                                   "invalid R2T with zero data len\n");
663                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
664                             sizeof(void*));
665                 return ISCSI_ERR_DATALEN;
666         }
667
668         if (r2t->data_length > session->max_burst)
669                 debug_scsi("invalid R2T with data len %u and max burst %u."
670                            "Attempting to execute request.\n",
671                             r2t->data_length, session->max_burst);
672
673         r2t->data_offset = be32_to_cpu(rhdr->data_offset);
674         if (r2t->data_offset + r2t->data_length > scsi_out(ctask->sc)->length) {
675                 iscsi_conn_printk(KERN_ERR, conn,
676                                   "invalid R2T with data len %u at offset %u "
677                                   "and total length %d\n", r2t->data_length,
678                                   r2t->data_offset, scsi_out(ctask->sc)->length);
679                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
680                             sizeof(void*));
681                 return ISCSI_ERR_DATALEN;
682         }
683
684         r2t->ttt = rhdr->ttt; /* no flip */
685         r2t->solicit_datasn = 0;
686
687         iscsi_solicit_data_init(conn, ctask, r2t);
688
689         tcp_ctask->exp_datasn = r2tsn + 1;
690         __kfifo_put(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*));
691         conn->r2t_pdus_cnt++;
692
693         iscsi_requeue_ctask(ctask);
694         return 0;
695 }
696
697 /*
698  * Handle incoming reply to DataIn command
699  */
700 static int
701 iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
702                           struct iscsi_segment *segment)
703 {
704         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
705         struct iscsi_hdr *hdr = tcp_conn->in.hdr;
706         int rc;
707
708         if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
709                 return ISCSI_ERR_DATA_DGST;
710
711         /* check for non-exceptional status */
712         if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
713                 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
714                 if (rc)
715                         return rc;
716         }
717
718         iscsi_tcp_hdr_recv_prep(tcp_conn);
719         return 0;
720 }
721
722 /**
723  * iscsi_tcp_hdr_dissect - process PDU header
724  * @conn: iSCSI connection
725  * @hdr: PDU header
726  *
727  * This function analyzes the header of the PDU received,
728  * and performs several sanity checks. If the PDU is accompanied
729  * by data, the receive buffer is set up to copy the incoming data
730  * to the correct location.
731  */
732 static int
733 iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
734 {
735         int rc = 0, opcode, ahslen;
736         struct iscsi_session *session = conn->session;
737         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
738         struct iscsi_cmd_task *ctask;
739         uint32_t itt;
740
741         /* verify PDU length */
742         tcp_conn->in.datalen = ntoh24(hdr->dlength);
743         if (tcp_conn->in.datalen > conn->max_recv_dlength) {
744                 iscsi_conn_printk(KERN_ERR, conn,
745                                   "iscsi_tcp: datalen %d > %d\n",
746                                   tcp_conn->in.datalen, conn->max_recv_dlength);
747                 return ISCSI_ERR_DATALEN;
748         }
749
750         /* Additional header segments. So far, we don't
751          * process additional headers.
752          */
753         ahslen = hdr->hlength << 2;
754
755         opcode = hdr->opcode & ISCSI_OPCODE_MASK;
756         /* verify itt (itt encoding: age+cid+itt) */
757         rc = iscsi_verify_itt(conn, hdr, &itt);
758         if (rc)
759                 return rc;
760
761         debug_tcp("opcode 0x%x ahslen %d datalen %d\n",
762                   opcode, ahslen, tcp_conn->in.datalen);
763
764         switch(opcode) {
765         case ISCSI_OP_SCSI_DATA_IN:
766                 ctask = session->cmds[itt];
767                 spin_lock(&conn->session->lock);
768                 rc = iscsi_data_rsp(conn, ctask);
769                 spin_unlock(&conn->session->lock);
770                 if (rc)
771                         return rc;
772                 if (tcp_conn->in.datalen) {
773                         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
774                         struct hash_desc *rx_hash = NULL;
775                         struct scsi_data_buffer *sdb = scsi_in(ctask->sc);
776
777                         /*
778                          * Setup copy of Data-In into the Scsi_Cmnd
779                          * Scatterlist case:
780                          * We set up the iscsi_segment to point to the next
781                          * scatterlist entry to copy to. As we go along,
782                          * we move on to the next scatterlist entry and
783                          * update the digest per-entry.
784                          */
785                         if (conn->datadgst_en)
786                                 rx_hash = &tcp_conn->rx_hash;
787
788                         debug_tcp("iscsi_tcp_begin_data_in(%p, offset=%d, "
789                                   "datalen=%d)\n", tcp_conn,
790                                   tcp_ctask->data_offset,
791                                   tcp_conn->in.datalen);
792                         return iscsi_segment_seek_sg(&tcp_conn->in.segment,
793                                                      sdb->table.sgl,
794                                                      sdb->table.nents,
795                                                      tcp_ctask->data_offset,
796                                                      tcp_conn->in.datalen,
797                                                      iscsi_tcp_process_data_in,
798                                                      rx_hash);
799                 }
800                 /* fall through */
801         case ISCSI_OP_SCSI_CMD_RSP:
802                 if (tcp_conn->in.datalen) {
803                         iscsi_tcp_data_recv_prep(tcp_conn);
804                         return 0;
805                 }
806                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
807                 break;
808         case ISCSI_OP_R2T:
809                 ctask = session->cmds[itt];
810                 if (ahslen)
811                         rc = ISCSI_ERR_AHSLEN;
812                 else if (ctask->sc->sc_data_direction == DMA_TO_DEVICE) {
813                         spin_lock(&session->lock);
814                         rc = iscsi_r2t_rsp(conn, ctask);
815                         spin_unlock(&session->lock);
816                 } else
817                         rc = ISCSI_ERR_PROTO;
818                 break;
819         case ISCSI_OP_LOGIN_RSP:
820         case ISCSI_OP_TEXT_RSP:
821         case ISCSI_OP_REJECT:
822         case ISCSI_OP_ASYNC_EVENT:
823                 /*
824                  * It is possible that we could get a PDU with a buffer larger
825                  * than 8K, but there are no targets that currently do this.
826                  * For now we fail until we find a vendor that needs it
827                  */
828                 if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
829                         iscsi_conn_printk(KERN_ERR, conn,
830                                           "iscsi_tcp: received buffer of "
831                                           "len %u but conn buffer is only %u "
832                                           "(opcode %0x)\n",
833                                           tcp_conn->in.datalen,
834                                           ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
835                         rc = ISCSI_ERR_PROTO;
836                         break;
837                 }
838
839                 /* If there's data coming in with the response,
840                  * receive it to the connection's buffer.
841                  */
842                 if (tcp_conn->in.datalen) {
843                         iscsi_tcp_data_recv_prep(tcp_conn);
844                         return 0;
845                 }
846         /* fall through */
847         case ISCSI_OP_LOGOUT_RSP:
848         case ISCSI_OP_NOOP_IN:
849         case ISCSI_OP_SCSI_TMFUNC_RSP:
850                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
851                 break;
852         default:
853                 rc = ISCSI_ERR_BAD_OPCODE;
854                 break;
855         }
856
857         if (rc == 0) {
858                 /* Anything that comes with data should have
859                  * been handled above. */
860                 if (tcp_conn->in.datalen)
861                         return ISCSI_ERR_PROTO;
862                 iscsi_tcp_hdr_recv_prep(tcp_conn);
863         }
864
865         return rc;
866 }
867
868 /**
869  * iscsi_tcp_hdr_recv_done - process PDU header
870  *
871  * This is the callback invoked when the PDU header has
872  * been received. If the header is followed by additional
873  * header segments, we go back for more data.
874  */
875 static int
876 iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
877                         struct iscsi_segment *segment)
878 {
879         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
880         struct iscsi_hdr *hdr;
881
882         /* Check if there are additional header segments
883          * *prior* to computing the digest, because we
884          * may need to go back to the caller for more.
885          */
886         hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
887         if (segment->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
888                 /* Bump the header length - the caller will
889                  * just loop around and get the AHS for us, and
890                  * call again. */
891                 unsigned int ahslen = hdr->hlength << 2;
892
893                 /* Make sure we don't overflow */
894                 if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
895                         return ISCSI_ERR_AHSLEN;
896
897                 segment->total_size += ahslen;
898                 segment->size += ahslen;
899                 return 0;
900         }
901
902         /* We're done processing the header. See if we're doing
903          * header digests; if so, set up the recv_digest buffer
904          * and go back for more. */
905         if (conn->hdrdgst_en) {
906                 if (segment->digest_len == 0) {
907                         iscsi_tcp_segment_splice_digest(segment,
908                                                         segment->recv_digest);
909                         return 0;
910                 }
911                 iscsi_tcp_dgst_header(&tcp_conn->rx_hash, hdr,
912                                       segment->total_copied - ISCSI_DIGEST_SIZE,
913                                       segment->digest);
914
915                 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
916                         return ISCSI_ERR_HDR_DGST;
917         }
918
919         tcp_conn->in.hdr = hdr;
920         return iscsi_tcp_hdr_dissect(conn, hdr);
921 }
922
923 /**
924  * iscsi_tcp_recv - TCP receive in sendfile fashion
925  * @rd_desc: read descriptor
926  * @skb: socket buffer
927  * @offset: offset in skb
928  * @len: skb->len - offset
929  **/
930 static int
931 iscsi_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
932                unsigned int offset, size_t len)
933 {
934         struct iscsi_conn *conn = rd_desc->arg.data;
935         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
936         struct iscsi_segment *segment = &tcp_conn->in.segment;
937         struct skb_seq_state seq;
938         unsigned int consumed = 0;
939         int rc = 0;
940
941         debug_tcp("in %d bytes\n", skb->len - offset);
942
943         if (unlikely(conn->suspend_rx)) {
944                 debug_tcp("conn %d Rx suspended!\n", conn->id);
945                 return 0;
946         }
947
948         skb_prepare_seq_read(skb, offset, skb->len, &seq);
949         while (1) {
950                 unsigned int avail;
951                 const u8 *ptr;
952
953                 avail = skb_seq_read(consumed, &ptr, &seq);
954                 if (avail == 0) {
955                         debug_tcp("no more data avail. Consumed %d\n",
956                                   consumed);
957                         break;
958                 }
959                 BUG_ON(segment->copied >= segment->size);
960
961                 debug_tcp("skb %p ptr=%p avail=%u\n", skb, ptr, avail);
962                 rc = iscsi_tcp_segment_recv(tcp_conn, segment, ptr, avail);
963                 BUG_ON(rc == 0);
964                 consumed += rc;
965
966                 if (segment->total_copied >= segment->total_size) {
967                         debug_tcp("segment done\n");
968                         rc = segment->done(tcp_conn, segment);
969                         if (rc != 0) {
970                                 skb_abort_seq_read(&seq);
971                                 goto error;
972                         }
973
974                         /* The done() functions sets up the
975                          * next segment. */
976                 }
977         }
978         skb_abort_seq_read(&seq);
979         conn->rxdata_octets += consumed;
980         return consumed;
981
982 error:
983         debug_tcp("Error receiving PDU, errno=%d\n", rc);
984         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
985         return 0;
986 }
987
988 static void
989 iscsi_tcp_data_ready(struct sock *sk, int flag)
990 {
991         struct iscsi_conn *conn = sk->sk_user_data;
992         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
993         read_descriptor_t rd_desc;
994
995         read_lock(&sk->sk_callback_lock);
996
997         /*
998          * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
999          * We set count to 1 because we want the network layer to
1000          * hand us all the skbs that are available. iscsi_tcp_recv
1001          * handled pdus that cross buffers or pdus that still need data.
1002          */
1003         rd_desc.arg.data = conn;
1004         rd_desc.count = 1;
1005         tcp_read_sock(sk, &rd_desc, iscsi_tcp_recv);
1006
1007         read_unlock(&sk->sk_callback_lock);
1008
1009         /* If we had to (atomically) map a highmem page,
1010          * unmap it now. */
1011         iscsi_tcp_segment_unmap(&tcp_conn->in.segment);
1012 }
1013
1014 static void
1015 iscsi_tcp_state_change(struct sock *sk)
1016 {
1017         struct iscsi_tcp_conn *tcp_conn;
1018         struct iscsi_conn *conn;
1019         struct iscsi_session *session;
1020         void (*old_state_change)(struct sock *);
1021
1022         read_lock(&sk->sk_callback_lock);
1023
1024         conn = (struct iscsi_conn*)sk->sk_user_data;
1025         session = conn->session;
1026
1027         if ((sk->sk_state == TCP_CLOSE_WAIT ||
1028              sk->sk_state == TCP_CLOSE) &&
1029             !atomic_read(&sk->sk_rmem_alloc)) {
1030                 debug_tcp("iscsi_tcp_state_change: TCP_CLOSE|TCP_CLOSE_WAIT\n");
1031                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1032         }
1033
1034         tcp_conn = conn->dd_data;
1035         old_state_change = tcp_conn->old_state_change;
1036
1037         read_unlock(&sk->sk_callback_lock);
1038
1039         old_state_change(sk);
1040 }
1041
1042 /**
1043  * iscsi_write_space - Called when more output buffer space is available
1044  * @sk: socket space is available for
1045  **/
1046 static void
1047 iscsi_write_space(struct sock *sk)
1048 {
1049         struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
1050         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1051
1052         tcp_conn->old_write_space(sk);
1053         debug_tcp("iscsi_write_space: cid %d\n", conn->id);
1054         scsi_queue_work(conn->session->host, &conn->xmitwork);
1055 }
1056
1057 static void
1058 iscsi_conn_set_callbacks(struct iscsi_conn *conn)
1059 {
1060         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1061         struct sock *sk = tcp_conn->sock->sk;
1062
1063         /* assign new callbacks */
1064         write_lock_bh(&sk->sk_callback_lock);
1065         sk->sk_user_data = conn;
1066         tcp_conn->old_data_ready = sk->sk_data_ready;
1067         tcp_conn->old_state_change = sk->sk_state_change;
1068         tcp_conn->old_write_space = sk->sk_write_space;
1069         sk->sk_data_ready = iscsi_tcp_data_ready;
1070         sk->sk_state_change = iscsi_tcp_state_change;
1071         sk->sk_write_space = iscsi_write_space;
1072         write_unlock_bh(&sk->sk_callback_lock);
1073 }
1074
1075 static void
1076 iscsi_conn_restore_callbacks(struct iscsi_tcp_conn *tcp_conn)
1077 {
1078         struct sock *sk = tcp_conn->sock->sk;
1079
1080         /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
1081         write_lock_bh(&sk->sk_callback_lock);
1082         sk->sk_user_data    = NULL;
1083         sk->sk_data_ready   = tcp_conn->old_data_ready;
1084         sk->sk_state_change = tcp_conn->old_state_change;
1085         sk->sk_write_space  = tcp_conn->old_write_space;
1086         sk->sk_no_check  = 0;
1087         write_unlock_bh(&sk->sk_callback_lock);
1088 }
1089
1090 /**
1091  * iscsi_xmit - TCP transmit
1092  **/
1093 static int
1094 iscsi_xmit(struct iscsi_conn *conn)
1095 {
1096         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1097         struct iscsi_segment *segment = &tcp_conn->out.segment;
1098         unsigned int consumed = 0;
1099         int rc = 0;
1100
1101         while (1) {
1102                 rc = iscsi_tcp_xmit_segment(tcp_conn, segment);
1103                 if (rc < 0)
1104                         goto error;
1105                 if (rc == 0)
1106                         break;
1107
1108                 consumed += rc;
1109
1110                 if (segment->total_copied >= segment->total_size) {
1111                         if (segment->done != NULL) {
1112                                 rc = segment->done(tcp_conn, segment);
1113                                 if (rc < 0)
1114                                         goto error;
1115                         }
1116                 }
1117         }
1118
1119         debug_tcp("xmit %d bytes\n", consumed);
1120
1121         conn->txdata_octets += consumed;
1122         return consumed;
1123
1124 error:
1125         /* Transmit error. We could initiate error recovery
1126          * here. */
1127         debug_tcp("Error sending PDU, errno=%d\n", rc);
1128         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1129         return rc;
1130 }
1131
1132 /**
1133  * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
1134  */
1135 static inline int
1136 iscsi_tcp_xmit_qlen(struct iscsi_conn *conn)
1137 {
1138         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1139         struct iscsi_segment *segment = &tcp_conn->out.segment;
1140
1141         return segment->total_copied - segment->total_size;
1142 }
1143
1144 static inline int
1145 iscsi_tcp_flush(struct iscsi_conn *conn)
1146 {
1147         int rc;
1148
1149         while (iscsi_tcp_xmit_qlen(conn)) {
1150                 rc = iscsi_xmit(conn);
1151                 if (rc == 0)
1152                         return -EAGAIN;
1153                 if (rc < 0)
1154                         return rc;
1155         }
1156
1157         return 0;
1158 }
1159
1160 /*
1161  * This is called when we're done sending the header.
1162  * Simply copy the data_segment to the send segment, and return.
1163  */
1164 static int
1165 iscsi_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
1166                         struct iscsi_segment *segment)
1167 {
1168         tcp_conn->out.segment = tcp_conn->out.data_segment;
1169         debug_tcp("Header done. Next segment size %u total_size %u\n",
1170                   tcp_conn->out.segment.size, tcp_conn->out.segment.total_size);
1171         return 0;
1172 }
1173
1174 static void
1175 iscsi_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr, size_t hdrlen)
1176 {
1177         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1178
1179         debug_tcp("%s(%p%s)\n", __FUNCTION__, tcp_conn,
1180                         conn->hdrdgst_en? ", digest enabled" : "");
1181
1182         /* Clear the data segment - needs to be filled in by the
1183          * caller using iscsi_tcp_send_data_prep() */
1184         memset(&tcp_conn->out.data_segment, 0, sizeof(struct iscsi_segment));
1185
1186         /* If header digest is enabled, compute the CRC and
1187          * place the digest into the same buffer. We make
1188          * sure that both iscsi_tcp_ctask and mtask have
1189          * sufficient room.
1190          */
1191         if (conn->hdrdgst_en) {
1192                 iscsi_tcp_dgst_header(&tcp_conn->tx_hash, hdr, hdrlen,
1193                                       hdr + hdrlen);
1194                 hdrlen += ISCSI_DIGEST_SIZE;
1195         }
1196
1197         /* Remember header pointer for later, when we need
1198          * to decide whether there's a payload to go along
1199          * with the header. */
1200         tcp_conn->out.hdr = hdr;
1201
1202         iscsi_segment_init_linear(&tcp_conn->out.segment, hdr, hdrlen,
1203                                 iscsi_tcp_send_hdr_done, NULL);
1204 }
1205
1206 /*
1207  * Prepare the send buffer for the payload data.
1208  * Padding and checksumming will all be taken care
1209  * of by the iscsi_segment routines.
1210  */
1211 static int
1212 iscsi_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
1213                          unsigned int count, unsigned int offset,
1214                          unsigned int len)
1215 {
1216         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1217         struct hash_desc *tx_hash = NULL;
1218         unsigned int hdr_spec_len;
1219
1220         debug_tcp("%s(%p, offset=%d, datalen=%d%s)\n", __FUNCTION__,
1221                         tcp_conn, offset, len,
1222                         conn->datadgst_en? ", digest enabled" : "");
1223
1224         /* Make sure the datalen matches what the caller
1225            said he would send. */
1226         hdr_spec_len = ntoh24(tcp_conn->out.hdr->dlength);
1227         WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
1228
1229         if (conn->datadgst_en)
1230                 tx_hash = &tcp_conn->tx_hash;
1231
1232         return iscsi_segment_seek_sg(&tcp_conn->out.data_segment,
1233                                    sg, count, offset, len,
1234                                    NULL, tx_hash);
1235 }
1236
1237 static void
1238 iscsi_tcp_send_linear_data_prepare(struct iscsi_conn *conn, void *data,
1239                                    size_t len)
1240 {
1241         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1242         struct hash_desc *tx_hash = NULL;
1243         unsigned int hdr_spec_len;
1244
1245         debug_tcp("%s(%p, datalen=%d%s)\n", __FUNCTION__, tcp_conn, len,
1246                   conn->datadgst_en? ", digest enabled" : "");
1247
1248         /* Make sure the datalen matches what the caller
1249            said he would send. */
1250         hdr_spec_len = ntoh24(tcp_conn->out.hdr->dlength);
1251         WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
1252
1253         if (conn->datadgst_en)
1254                 tx_hash = &tcp_conn->tx_hash;
1255
1256         iscsi_segment_init_linear(&tcp_conn->out.data_segment,
1257                                 data, len, NULL, tx_hash);
1258 }
1259
1260 /**
1261  * iscsi_solicit_data_cont - initialize next Data-Out
1262  * @conn: iscsi connection
1263  * @ctask: scsi command task
1264  * @r2t: R2T info
1265  * @left: bytes left to transfer
1266  *
1267  * Notes:
1268  *      Initialize next Data-Out within this R2T sequence and continue
1269  *      to process next Scatter-Gather element(if any) of this SCSI command.
1270  *
1271  *      Called under connection lock.
1272  **/
1273 static int
1274 iscsi_solicit_data_cont(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
1275                         struct iscsi_r2t_info *r2t)
1276 {
1277         struct iscsi_data *hdr;
1278         int new_offset, left;
1279
1280         BUG_ON(r2t->data_length - r2t->sent < 0);
1281         left = r2t->data_length - r2t->sent;
1282         if (left == 0)
1283                 return 0;
1284
1285         hdr = &r2t->dtask.hdr;
1286         memset(hdr, 0, sizeof(struct iscsi_data));
1287         hdr->ttt = r2t->ttt;
1288         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
1289         r2t->solicit_datasn++;
1290         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
1291         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
1292         hdr->itt = ctask->hdr->itt;
1293         hdr->exp_statsn = r2t->exp_statsn;
1294         new_offset = r2t->data_offset + r2t->sent;
1295         hdr->offset = cpu_to_be32(new_offset);
1296         if (left > conn->max_xmit_dlength) {
1297                 hton24(hdr->dlength, conn->max_xmit_dlength);
1298                 r2t->data_count = conn->max_xmit_dlength;
1299         } else {
1300                 hton24(hdr->dlength, left);
1301                 r2t->data_count = left;
1302                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
1303         }
1304
1305         conn->dataout_pdus_cnt++;
1306         return 1;
1307 }
1308
1309 /**
1310  * iscsi_tcp_ctask - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
1311  * @conn: iscsi connection
1312  * @ctask: scsi command task
1313  * @sc: scsi command
1314  **/
1315 static int
1316 iscsi_tcp_ctask_init(struct iscsi_cmd_task *ctask)
1317 {
1318         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1319         struct iscsi_conn *conn = ctask->conn;
1320         struct scsi_cmnd *sc = ctask->sc;
1321         int err;
1322
1323         BUG_ON(__kfifo_len(tcp_ctask->r2tqueue));
1324         tcp_ctask->sent = 0;
1325         tcp_ctask->exp_datasn = 0;
1326
1327         /* Prepare PDU, optionally w/ immediate data */
1328         debug_scsi("ctask deq [cid %d itt 0x%x imm %d unsol %d]\n",
1329                     conn->id, ctask->itt, ctask->imm_count,
1330                     ctask->unsol_count);
1331         iscsi_tcp_send_hdr_prep(conn, ctask->hdr, ctask->hdr_len);
1332
1333         if (!ctask->imm_count)
1334                 return 0;
1335
1336         /* If we have immediate data, attach a payload */
1337         err = iscsi_tcp_send_data_prep(conn, scsi_out(sc)->table.sgl,
1338                                        scsi_out(sc)->table.nents,
1339                                        0, ctask->imm_count);
1340         if (err)
1341                 return err;
1342         tcp_ctask->sent += ctask->imm_count;
1343         ctask->imm_count = 0;
1344         return 0;
1345 }
1346
1347 /**
1348  * iscsi_tcp_mtask_xmit - xmit management(immediate) task
1349  * @conn: iscsi connection
1350  * @mtask: task management task
1351  *
1352  * Notes:
1353  *      The function can return -EAGAIN in which case caller must
1354  *      call it again later, or recover. '0' return code means successful
1355  *      xmit.
1356  **/
1357 static int
1358 iscsi_tcp_mtask_xmit(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1359 {
1360         int rc;
1361
1362         /* Flush any pending data first. */
1363         rc = iscsi_tcp_flush(conn);
1364         if (rc < 0)
1365                 return rc;
1366
1367         if (mtask->hdr->itt == RESERVED_ITT) {
1368                 struct iscsi_session *session = conn->session;
1369
1370                 spin_lock_bh(&session->lock);
1371                 iscsi_free_mgmt_task(conn, mtask);
1372                 spin_unlock_bh(&session->lock);
1373         }
1374
1375         return 0;
1376 }
1377
1378 /*
1379  * iscsi_tcp_ctask_xmit - xmit normal PDU task
1380  * @conn: iscsi connection
1381  * @ctask: iscsi command task
1382  *
1383  * We're expected to return 0 when everything was transmitted succesfully,
1384  * -EAGAIN if there's still data in the queue, or != 0 for any other kind
1385  * of error.
1386  */
1387 static int
1388 iscsi_tcp_ctask_xmit(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1389 {
1390         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1391         struct scsi_cmnd *sc = ctask->sc;
1392         struct scsi_data_buffer *sdb = scsi_out(sc);
1393         int rc = 0;
1394
1395 flush:
1396         /* Flush any pending data first. */
1397         rc = iscsi_tcp_flush(conn);
1398         if (rc < 0)
1399                 return rc;
1400
1401         /* Are we done already? */
1402         if (sc->sc_data_direction != DMA_TO_DEVICE)
1403                 return 0;
1404
1405         if (ctask->unsol_count != 0) {
1406                 struct iscsi_data *hdr = &tcp_ctask->unsol_dtask.hdr;
1407
1408                 /* Prepare a header for the unsolicited PDU.
1409                  * The amount of data we want to send will be
1410                  * in ctask->data_count.
1411                  * FIXME: return the data count instead.
1412                  */
1413                 iscsi_prep_unsolicit_data_pdu(ctask, hdr);
1414
1415                 debug_tcp("unsol dout [itt 0x%x doff %d dlen %d]\n",
1416                                 ctask->itt, tcp_ctask->sent, ctask->data_count);
1417
1418                 iscsi_tcp_send_hdr_prep(conn, hdr, sizeof(*hdr));
1419                 rc = iscsi_tcp_send_data_prep(conn, sdb->table.sgl,
1420                                               sdb->table.nents, tcp_ctask->sent,
1421                                               ctask->data_count);
1422                 if (rc)
1423                         goto fail;
1424                 tcp_ctask->sent += ctask->data_count;
1425                 ctask->unsol_count -= ctask->data_count;
1426                 goto flush;
1427         } else {
1428                 struct iscsi_session *session = conn->session;
1429                 struct iscsi_r2t_info *r2t;
1430
1431                 /* All unsolicited PDUs sent. Check for solicited PDUs.
1432                  */
1433                 spin_lock_bh(&session->lock);
1434                 r2t = tcp_ctask->r2t;
1435                 if (r2t != NULL) {
1436                         /* Continue with this R2T? */
1437                         if (!iscsi_solicit_data_cont(conn, ctask, r2t)) {
1438                                 debug_scsi("  done with r2t %p\n", r2t);
1439
1440                                 __kfifo_put(tcp_ctask->r2tpool.queue,
1441                                             (void*)&r2t, sizeof(void*));
1442                                 tcp_ctask->r2t = r2t = NULL;
1443                         }
1444                 }
1445
1446                 if (r2t == NULL) {
1447                         __kfifo_get(tcp_ctask->r2tqueue, (void*)&tcp_ctask->r2t,
1448                                     sizeof(void*));
1449                         r2t = tcp_ctask->r2t;
1450                 }
1451                 spin_unlock_bh(&session->lock);
1452
1453                 /* Waiting for more R2Ts to arrive. */
1454                 if (r2t == NULL) {
1455                         debug_tcp("no R2Ts yet\n");
1456                         return 0;
1457                 }
1458
1459                 debug_scsi("sol dout %p [dsn %d itt 0x%x doff %d dlen %d]\n",
1460                         r2t, r2t->solicit_datasn - 1, ctask->itt,
1461                         r2t->data_offset + r2t->sent, r2t->data_count);
1462
1463                 iscsi_tcp_send_hdr_prep(conn, &r2t->dtask.hdr,
1464                                         sizeof(struct iscsi_hdr));
1465
1466                 rc = iscsi_tcp_send_data_prep(conn, sdb->table.sgl,
1467                                               sdb->table.nents,
1468                                               r2t->data_offset + r2t->sent,
1469                                               r2t->data_count);
1470                 if (rc)
1471                         goto fail;
1472                 tcp_ctask->sent += r2t->data_count;
1473                 r2t->sent += r2t->data_count;
1474                 goto flush;
1475         }
1476         return 0;
1477 fail:
1478         iscsi_conn_failure(conn, rc);
1479         return -EIO;
1480 }
1481
1482 static struct iscsi_cls_conn *
1483 iscsi_tcp_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
1484 {
1485         struct iscsi_conn *conn;
1486         struct iscsi_cls_conn *cls_conn;
1487         struct iscsi_tcp_conn *tcp_conn;
1488
1489         cls_conn = iscsi_conn_setup(cls_session, conn_idx);
1490         if (!cls_conn)
1491                 return NULL;
1492         conn = cls_conn->dd_data;
1493         /*
1494          * due to strange issues with iser these are not set
1495          * in iscsi_conn_setup
1496          */
1497         conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1498
1499         tcp_conn = kzalloc(sizeof(*tcp_conn), GFP_KERNEL);
1500         if (!tcp_conn)
1501                 goto tcp_conn_alloc_fail;
1502
1503         conn->dd_data = tcp_conn;
1504         tcp_conn->iscsi_conn = conn;
1505
1506         tcp_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1507                                                   CRYPTO_ALG_ASYNC);
1508         tcp_conn->tx_hash.flags = 0;
1509         if (IS_ERR(tcp_conn->tx_hash.tfm))
1510                 goto free_tcp_conn;
1511
1512         tcp_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1513                                                   CRYPTO_ALG_ASYNC);
1514         tcp_conn->rx_hash.flags = 0;
1515         if (IS_ERR(tcp_conn->rx_hash.tfm))
1516                 goto free_tx_tfm;
1517
1518         return cls_conn;
1519
1520 free_tx_tfm:
1521         crypto_free_hash(tcp_conn->tx_hash.tfm);
1522 free_tcp_conn:
1523         iscsi_conn_printk(KERN_ERR, conn,
1524                           "Could not create connection due to crc32c "
1525                           "loading error. Make sure the crc32c "
1526                           "module is built as a module or into the "
1527                           "kernel\n");
1528         kfree(tcp_conn);
1529 tcp_conn_alloc_fail:
1530         iscsi_conn_teardown(cls_conn);
1531         return NULL;
1532 }
1533
1534 static void
1535 iscsi_tcp_release_conn(struct iscsi_conn *conn)
1536 {
1537         struct iscsi_session *session = conn->session;
1538         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1539         struct socket *sock = tcp_conn->sock;
1540
1541         if (!sock)
1542                 return;
1543
1544         sock_hold(sock->sk);
1545         iscsi_conn_restore_callbacks(tcp_conn);
1546         sock_put(sock->sk);
1547
1548         spin_lock_bh(&session->lock);
1549         tcp_conn->sock = NULL;
1550         conn->recv_lock = NULL;
1551         spin_unlock_bh(&session->lock);
1552         sockfd_put(sock);
1553 }
1554
1555 static void
1556 iscsi_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
1557 {
1558         struct iscsi_conn *conn = cls_conn->dd_data;
1559         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1560
1561         iscsi_tcp_release_conn(conn);
1562         iscsi_conn_teardown(cls_conn);
1563
1564         if (tcp_conn->tx_hash.tfm)
1565                 crypto_free_hash(tcp_conn->tx_hash.tfm);
1566         if (tcp_conn->rx_hash.tfm)
1567                 crypto_free_hash(tcp_conn->rx_hash.tfm);
1568
1569         kfree(tcp_conn);
1570 }
1571
1572 static void
1573 iscsi_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
1574 {
1575         struct iscsi_conn *conn = cls_conn->dd_data;
1576
1577         iscsi_conn_stop(cls_conn, flag);
1578         iscsi_tcp_release_conn(conn);
1579 }
1580
1581 static int iscsi_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
1582                               char *buf, int *port,
1583                               int (*getname)(struct socket *, struct sockaddr *,
1584                                         int *addrlen))
1585 {
1586         struct sockaddr_storage *addr;
1587         struct sockaddr_in6 *sin6;
1588         struct sockaddr_in *sin;
1589         int rc = 0, len;
1590
1591         addr = kmalloc(sizeof(*addr), GFP_KERNEL);
1592         if (!addr)
1593                 return -ENOMEM;
1594
1595         if (getname(sock, (struct sockaddr *) addr, &len)) {
1596                 rc = -ENODEV;
1597                 goto free_addr;
1598         }
1599
1600         switch (addr->ss_family) {
1601         case AF_INET:
1602                 sin = (struct sockaddr_in *)addr;
1603                 spin_lock_bh(&conn->session->lock);
1604                 sprintf(buf, NIPQUAD_FMT, NIPQUAD(sin->sin_addr.s_addr));
1605                 *port = be16_to_cpu(sin->sin_port);
1606                 spin_unlock_bh(&conn->session->lock);
1607                 break;
1608         case AF_INET6:
1609                 sin6 = (struct sockaddr_in6 *)addr;
1610                 spin_lock_bh(&conn->session->lock);
1611                 sprintf(buf, NIP6_FMT, NIP6(sin6->sin6_addr));
1612                 *port = be16_to_cpu(sin6->sin6_port);
1613                 spin_unlock_bh(&conn->session->lock);
1614                 break;
1615         }
1616 free_addr:
1617         kfree(addr);
1618         return rc;
1619 }
1620
1621 static int
1622 iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
1623                     struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
1624                     int is_leading)
1625 {
1626         struct iscsi_conn *conn = cls_conn->dd_data;
1627         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1628         struct sock *sk;
1629         struct socket *sock;
1630         int err;
1631
1632         /* lookup for existing socket */
1633         sock = sockfd_lookup((int)transport_eph, &err);
1634         if (!sock) {
1635                 iscsi_conn_printk(KERN_ERR, conn,
1636                                   "sockfd_lookup failed %d\n", err);
1637                 return -EEXIST;
1638         }
1639         /*
1640          * copy these values now because if we drop the session
1641          * userspace may still want to query the values since we will
1642          * be using them for the reconnect
1643          */
1644         err = iscsi_tcp_get_addr(conn, sock, conn->portal_address,
1645                                  &conn->portal_port, kernel_getpeername);
1646         if (err)
1647                 goto free_socket;
1648
1649         err = iscsi_tcp_get_addr(conn, sock, conn->local_address,
1650                                 &conn->local_port, kernel_getsockname);
1651         if (err)
1652                 goto free_socket;
1653
1654         err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
1655         if (err)
1656                 goto free_socket;
1657
1658         /* bind iSCSI connection and socket */
1659         tcp_conn->sock = sock;
1660
1661         /* setup Socket parameters */
1662         sk = sock->sk;
1663         sk->sk_reuse = 1;
1664         sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
1665         sk->sk_allocation = GFP_ATOMIC;
1666
1667         /* FIXME: disable Nagle's algorithm */
1668
1669         /*
1670          * Intercept TCP callbacks for sendfile like receive
1671          * processing.
1672          */
1673         conn->recv_lock = &sk->sk_callback_lock;
1674         iscsi_conn_set_callbacks(conn);
1675         tcp_conn->sendpage = tcp_conn->sock->ops->sendpage;
1676         /*
1677          * set receive state machine into initial state
1678          */
1679         iscsi_tcp_hdr_recv_prep(tcp_conn);
1680         return 0;
1681
1682 free_socket:
1683         sockfd_put(sock);
1684         return err;
1685 }
1686
1687 /* called with host lock */
1688 static void
1689 iscsi_tcp_mtask_init(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1690 {
1691         debug_scsi("mtask deq [cid %d itt 0x%x]\n", conn->id, mtask->itt);
1692
1693         /* Prepare PDU, optionally w/ immediate data */
1694         iscsi_tcp_send_hdr_prep(conn, mtask->hdr, sizeof(*mtask->hdr));
1695
1696         /* If we have immediate data, attach a payload */
1697         if (mtask->data_count)
1698                 iscsi_tcp_send_linear_data_prepare(conn, mtask->data,
1699                                                    mtask->data_count);
1700 }
1701
1702 static int
1703 iscsi_r2tpool_alloc(struct iscsi_session *session)
1704 {
1705         int i;
1706         int cmd_i;
1707
1708         /*
1709          * initialize per-task: R2T pool and xmit queue
1710          */
1711         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1712                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
1713                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1714
1715                 /*
1716                  * pre-allocated x4 as much r2ts to handle race when
1717                  * target acks DataOut faster than we data_xmit() queues
1718                  * could replenish r2tqueue.
1719                  */
1720
1721                 /* R2T pool */
1722                 if (iscsi_pool_init(&tcp_ctask->r2tpool, session->max_r2t * 4, NULL,
1723                                     sizeof(struct iscsi_r2t_info))) {
1724                         goto r2t_alloc_fail;
1725                 }
1726
1727                 /* R2T xmit queue */
1728                 tcp_ctask->r2tqueue = kfifo_alloc(
1729                       session->max_r2t * 4 * sizeof(void*), GFP_KERNEL, NULL);
1730                 if (tcp_ctask->r2tqueue == ERR_PTR(-ENOMEM)) {
1731                         iscsi_pool_free(&tcp_ctask->r2tpool);
1732                         goto r2t_alloc_fail;
1733                 }
1734         }
1735
1736         return 0;
1737
1738 r2t_alloc_fail:
1739         for (i = 0; i < cmd_i; i++) {
1740                 struct iscsi_cmd_task *ctask = session->cmds[i];
1741                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1742
1743                 kfifo_free(tcp_ctask->r2tqueue);
1744                 iscsi_pool_free(&tcp_ctask->r2tpool);
1745         }
1746         return -ENOMEM;
1747 }
1748
1749 static void
1750 iscsi_r2tpool_free(struct iscsi_session *session)
1751 {
1752         int i;
1753
1754         for (i = 0; i < session->cmds_max; i++) {
1755                 struct iscsi_cmd_task *ctask = session->cmds[i];
1756                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1757
1758                 kfifo_free(tcp_ctask->r2tqueue);
1759                 iscsi_pool_free(&tcp_ctask->r2tpool);
1760         }
1761 }
1762
1763 static int
1764 iscsi_conn_set_param(struct iscsi_cls_conn *cls_conn, enum iscsi_param param,
1765                      char *buf, int buflen)
1766 {
1767         struct iscsi_conn *conn = cls_conn->dd_data;
1768         struct iscsi_session *session = conn->session;
1769         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1770         int value;
1771
1772         switch(param) {
1773         case ISCSI_PARAM_HDRDGST_EN:
1774                 iscsi_set_param(cls_conn, param, buf, buflen);
1775                 break;
1776         case ISCSI_PARAM_DATADGST_EN:
1777                 iscsi_set_param(cls_conn, param, buf, buflen);
1778                 tcp_conn->sendpage = conn->datadgst_en ?
1779                         sock_no_sendpage : tcp_conn->sock->ops->sendpage;
1780                 break;
1781         case ISCSI_PARAM_MAX_R2T:
1782                 sscanf(buf, "%d", &value);
1783                 if (value <= 0 || !is_power_of_2(value))
1784                         return -EINVAL;
1785                 if (session->max_r2t == value)
1786                         break;
1787                 iscsi_r2tpool_free(session);
1788                 iscsi_set_param(cls_conn, param, buf, buflen);
1789                 if (iscsi_r2tpool_alloc(session))
1790                         return -ENOMEM;
1791                 break;
1792         default:
1793                 return iscsi_set_param(cls_conn, param, buf, buflen);
1794         }
1795
1796         return 0;
1797 }
1798
1799 static int
1800 iscsi_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
1801                          enum iscsi_param param, char *buf)
1802 {
1803         struct iscsi_conn *conn = cls_conn->dd_data;
1804         int len;
1805
1806         switch(param) {
1807         case ISCSI_PARAM_CONN_PORT:
1808                 spin_lock_bh(&conn->session->lock);
1809                 len = sprintf(buf, "%hu\n", conn->portal_port);
1810                 spin_unlock_bh(&conn->session->lock);
1811                 break;
1812         case ISCSI_PARAM_CONN_ADDRESS:
1813                 spin_lock_bh(&conn->session->lock);
1814                 len = sprintf(buf, "%s\n", conn->portal_address);
1815                 spin_unlock_bh(&conn->session->lock);
1816                 break;
1817         default:
1818                 return iscsi_conn_get_param(cls_conn, param, buf);
1819         }
1820
1821         return len;
1822 }
1823
1824 static int
1825 iscsi_tcp_host_get_param(struct Scsi_Host *shost, enum iscsi_host_param param,
1826                          char *buf)
1827 {
1828         struct iscsi_session *session = iscsi_hostdata(shost->hostdata);
1829         int len;
1830
1831         switch (param) {
1832         case ISCSI_HOST_PARAM_IPADDRESS:
1833                 spin_lock_bh(&session->lock);
1834                 if (!session->leadconn)
1835                         len = -ENODEV;
1836                 else
1837                         len = sprintf(buf, "%s\n",
1838                                      session->leadconn->local_address);
1839                 spin_unlock_bh(&session->lock);
1840                 break;
1841         default:
1842                 return iscsi_host_get_param(shost, param, buf);
1843         }
1844         return len;
1845 }
1846
1847 static void
1848 iscsi_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
1849 {
1850         struct iscsi_conn *conn = cls_conn->dd_data;
1851         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1852
1853         stats->txdata_octets = conn->txdata_octets;
1854         stats->rxdata_octets = conn->rxdata_octets;
1855         stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
1856         stats->dataout_pdus = conn->dataout_pdus_cnt;
1857         stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
1858         stats->datain_pdus = conn->datain_pdus_cnt;
1859         stats->r2t_pdus = conn->r2t_pdus_cnt;
1860         stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
1861         stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
1862         stats->custom_length = 3;
1863         strcpy(stats->custom[0].desc, "tx_sendpage_failures");
1864         stats->custom[0].value = tcp_conn->sendpage_failures_cnt;
1865         strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
1866         stats->custom[1].value = tcp_conn->discontiguous_hdr_cnt;
1867         strcpy(stats->custom[2].desc, "eh_abort_cnt");
1868         stats->custom[2].value = conn->eh_abort_cnt;
1869 }
1870
1871 static struct iscsi_cls_session *
1872 iscsi_tcp_session_create(struct iscsi_transport *iscsit,
1873                          struct scsi_transport_template *scsit,
1874                          struct Scsi_Host *shost, uint16_t cmds_max,
1875                          uint16_t qdepth, uint32_t initial_cmdsn,
1876                          uint32_t *hostno)
1877 {
1878         struct iscsi_cls_session *cls_session;
1879         struct iscsi_session *session;
1880         uint32_t hn;
1881         int cmd_i;
1882
1883         cls_session = iscsi_session_setup(iscsit, scsit, cmds_max, qdepth,
1884                                          sizeof(struct iscsi_tcp_cmd_task),
1885                                          sizeof(struct iscsi_tcp_mgmt_task),
1886                                          initial_cmdsn, &hn);
1887         if (!cls_session)
1888                 return NULL;
1889         *hostno = hn;
1890
1891         session = class_to_transport_session(cls_session);
1892         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1893                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
1894                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1895
1896                 ctask->hdr = &tcp_ctask->hdr.cmd_hdr;
1897                 ctask->hdr_max = sizeof(tcp_ctask->hdr) - ISCSI_DIGEST_SIZE;
1898         }
1899
1900         for (cmd_i = 0; cmd_i < session->mgmtpool_max; cmd_i++) {
1901                 struct iscsi_mgmt_task *mtask = session->mgmt_cmds[cmd_i];
1902                 struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
1903
1904                 mtask->hdr = (struct iscsi_hdr *) &tcp_mtask->hdr;
1905         }
1906
1907         if (iscsi_r2tpool_alloc(class_to_transport_session(cls_session)))
1908                 goto r2tpool_alloc_fail;
1909
1910         return cls_session;
1911
1912 r2tpool_alloc_fail:
1913         iscsi_session_teardown(cls_session);
1914         return NULL;
1915 }
1916
1917 static void iscsi_tcp_session_destroy(struct iscsi_cls_session *cls_session)
1918 {
1919         iscsi_r2tpool_free(class_to_transport_session(cls_session));
1920         iscsi_session_teardown(cls_session);
1921 }
1922
1923 static int iscsi_tcp_slave_configure(struct scsi_device *sdev)
1924 {
1925         blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
1926         blk_queue_dma_alignment(sdev->request_queue, 0);
1927         return 0;
1928 }
1929
1930 static struct scsi_host_template iscsi_sht = {
1931         .module                 = THIS_MODULE,
1932         .name                   = "iSCSI Initiator over TCP/IP",
1933         .queuecommand           = iscsi_queuecommand,
1934         .change_queue_depth     = iscsi_change_queue_depth,
1935         .can_queue              = ISCSI_DEF_XMIT_CMDS_MAX - 1,
1936         .sg_tablesize           = 4096,
1937         .max_sectors            = 0xFFFF,
1938         .cmd_per_lun            = ISCSI_DEF_CMD_PER_LUN,
1939         .eh_abort_handler       = iscsi_eh_abort,
1940         .eh_device_reset_handler= iscsi_eh_device_reset,
1941         .eh_host_reset_handler  = iscsi_eh_host_reset,
1942         .use_clustering         = DISABLE_CLUSTERING,
1943         .slave_configure        = iscsi_tcp_slave_configure,
1944         .proc_name              = "iscsi_tcp",
1945         .this_id                = -1,
1946 };
1947
1948 static struct iscsi_transport iscsi_tcp_transport = {
1949         .owner                  = THIS_MODULE,
1950         .name                   = "tcp",
1951         .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
1952                                   | CAP_DATADGST,
1953         .param_mask             = ISCSI_MAX_RECV_DLENGTH |
1954                                   ISCSI_MAX_XMIT_DLENGTH |
1955                                   ISCSI_HDRDGST_EN |
1956                                   ISCSI_DATADGST_EN |
1957                                   ISCSI_INITIAL_R2T_EN |
1958                                   ISCSI_MAX_R2T |
1959                                   ISCSI_IMM_DATA_EN |
1960                                   ISCSI_FIRST_BURST |
1961                                   ISCSI_MAX_BURST |
1962                                   ISCSI_PDU_INORDER_EN |
1963                                   ISCSI_DATASEQ_INORDER_EN |
1964                                   ISCSI_ERL |
1965                                   ISCSI_CONN_PORT |
1966                                   ISCSI_CONN_ADDRESS |
1967                                   ISCSI_EXP_STATSN |
1968                                   ISCSI_PERSISTENT_PORT |
1969                                   ISCSI_PERSISTENT_ADDRESS |
1970                                   ISCSI_TARGET_NAME | ISCSI_TPGT |
1971                                   ISCSI_USERNAME | ISCSI_PASSWORD |
1972                                   ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
1973                                   ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
1974                                   ISCSI_LU_RESET_TMO |
1975                                   ISCSI_PING_TMO | ISCSI_RECV_TMO,
1976         .host_param_mask        = ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
1977                                   ISCSI_HOST_INITIATOR_NAME |
1978                                   ISCSI_HOST_NETDEV_NAME,
1979         .host_template          = &iscsi_sht,
1980         .conndata_size          = sizeof(struct iscsi_conn),
1981         .max_conn               = 1,
1982         .max_cmd_len            = 16,
1983         /* session management */
1984         .create_session         = iscsi_tcp_session_create,
1985         .destroy_session        = iscsi_tcp_session_destroy,
1986         /* connection management */
1987         .create_conn            = iscsi_tcp_conn_create,
1988         .bind_conn              = iscsi_tcp_conn_bind,
1989         .destroy_conn           = iscsi_tcp_conn_destroy,
1990         .set_param              = iscsi_conn_set_param,
1991         .get_conn_param         = iscsi_tcp_conn_get_param,
1992         .get_session_param      = iscsi_session_get_param,
1993         .start_conn             = iscsi_conn_start,
1994         .stop_conn              = iscsi_tcp_conn_stop,
1995         /* iscsi host params */
1996         .get_host_param         = iscsi_tcp_host_get_param,
1997         .set_host_param         = iscsi_host_set_param,
1998         /* IO */
1999         .send_pdu               = iscsi_conn_send_pdu,
2000         .get_stats              = iscsi_conn_get_stats,
2001         .init_cmd_task          = iscsi_tcp_ctask_init,
2002         .init_mgmt_task         = iscsi_tcp_mtask_init,
2003         .xmit_cmd_task          = iscsi_tcp_ctask_xmit,
2004         .xmit_mgmt_task         = iscsi_tcp_mtask_xmit,
2005         .cleanup_cmd_task       = iscsi_tcp_cleanup_ctask,
2006         /* recovery */
2007         .session_recovery_timedout = iscsi_session_recovery_timedout,
2008 };
2009
2010 static int __init
2011 iscsi_tcp_init(void)
2012 {
2013         if (iscsi_max_lun < 1) {
2014                 printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
2015                        iscsi_max_lun);
2016                 return -EINVAL;
2017         }
2018         iscsi_tcp_transport.max_lun = iscsi_max_lun;
2019
2020         if (!iscsi_register_transport(&iscsi_tcp_transport))
2021                 return -ENODEV;
2022
2023         return 0;
2024 }
2025
2026 static void __exit
2027 iscsi_tcp_exit(void)
2028 {
2029         iscsi_unregister_transport(&iscsi_tcp_transport);
2030 }
2031
2032 module_init(iscsi_tcp_init);
2033 module_exit(iscsi_tcp_exit);