[SCSI] libiscsi: fix data corruption when target has to resend data-in packets
[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 struct scsi_transport_template *iscsi_tcp_scsi_transport;
68 static struct scsi_host_template iscsi_sht;
69 static struct iscsi_transport iscsi_tcp_transport;
70
71 static unsigned int iscsi_max_lun = 512;
72 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
73
74 static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
75                                    struct iscsi_segment *segment);
76
77 /*
78  * Scatterlist handling: inside the iscsi_segment, we
79  * remember an index into the scatterlist, and set data/size
80  * to the current scatterlist entry. For highmem pages, we
81  * kmap as needed.
82  *
83  * Note that the page is unmapped when we return from
84  * TCP's data_ready handler, so we may end up mapping and
85  * unmapping the same page repeatedly. The whole reason
86  * for this is that we shouldn't keep the page mapped
87  * outside the softirq.
88  */
89
90 /**
91  * iscsi_tcp_segment_init_sg - init indicated scatterlist entry
92  * @segment: the buffer object
93  * @sg: scatterlist
94  * @offset: byte offset into that sg entry
95  *
96  * This function sets up the segment so that subsequent
97  * data is copied to the indicated sg entry, at the given
98  * offset.
99  */
100 static inline void
101 iscsi_tcp_segment_init_sg(struct iscsi_segment *segment,
102                           struct scatterlist *sg, unsigned int offset)
103 {
104         segment->sg = sg;
105         segment->sg_offset = offset;
106         segment->size = min(sg->length - offset,
107                             segment->total_size - segment->total_copied);
108         segment->data = NULL;
109 }
110
111 /**
112  * iscsi_tcp_segment_map - map the current S/G page
113  * @segment: iscsi_segment
114  * @recv: 1 if called from recv path
115  *
116  * We only need to possibly kmap data if scatter lists are being used,
117  * because the iscsi passthrough and internal IO paths will never use high
118  * mem pages.
119  */
120 static inline void
121 iscsi_tcp_segment_map(struct iscsi_segment *segment, int recv)
122 {
123         struct scatterlist *sg;
124
125         if (segment->data != NULL || !segment->sg)
126                 return;
127
128         sg = segment->sg;
129         BUG_ON(segment->sg_mapped);
130         BUG_ON(sg->length == 0);
131
132         /*
133          * If the page count is greater than one it is ok to send
134          * to the network layer's zero copy send path. If not we
135          * have to go the slow sendmsg path. We always map for the
136          * recv path.
137          */
138         if (page_count(sg_page(sg)) >= 1 && !recv)
139                 return;
140
141         debug_tcp("iscsi_tcp_segment_map %s %p\n", recv ? "recv" : "xmit",
142                   segment);
143         segment->sg_mapped = kmap_atomic(sg_page(sg), KM_SOFTIRQ0);
144         segment->data = segment->sg_mapped + sg->offset + segment->sg_offset;
145 }
146
147 static inline void
148 iscsi_tcp_segment_unmap(struct iscsi_segment *segment)
149 {
150         debug_tcp("iscsi_tcp_segment_unmap %p\n", segment);
151
152         if (segment->sg_mapped) {
153                 debug_tcp("iscsi_tcp_segment_unmap valid\n");
154                 kunmap_atomic(segment->sg_mapped, KM_SOFTIRQ0);
155                 segment->sg_mapped = NULL;
156                 segment->data = NULL;
157         }
158 }
159
160 /*
161  * Splice the digest buffer into the buffer
162  */
163 static inline void
164 iscsi_tcp_segment_splice_digest(struct iscsi_segment *segment, void *digest)
165 {
166         segment->data = digest;
167         segment->digest_len = ISCSI_DIGEST_SIZE;
168         segment->total_size += ISCSI_DIGEST_SIZE;
169         segment->size = ISCSI_DIGEST_SIZE;
170         segment->copied = 0;
171         segment->sg = NULL;
172         segment->hash = NULL;
173 }
174
175 /**
176  * iscsi_tcp_segment_done - check whether the segment is complete
177  * @segment: iscsi segment to check
178  * @recv: set to one of this is called from the recv path
179  * @copied: number of bytes copied
180  *
181  * Check if we're done receiving this segment. If the receive
182  * buffer is full but we expect more data, move on to the
183  * next entry in the scatterlist.
184  *
185  * If the amount of data we received isn't a multiple of 4,
186  * we will transparently receive the pad bytes, too.
187  *
188  * This function must be re-entrant.
189  */
190 static inline int
191 iscsi_tcp_segment_done(struct iscsi_segment *segment, int recv, unsigned copied)
192 {
193         static unsigned char padbuf[ISCSI_PAD_LEN];
194         struct scatterlist sg;
195         unsigned int pad;
196
197         debug_tcp("copied %u %u size %u %s\n", segment->copied, copied,
198                   segment->size, recv ? "recv" : "xmit");
199         if (segment->hash && copied) {
200                 /*
201                  * If a segment is kmapd we must unmap it before sending
202                  * to the crypto layer since that will try to kmap it again.
203                  */
204                 iscsi_tcp_segment_unmap(segment);
205
206                 if (!segment->data) {
207                         sg_init_table(&sg, 1);
208                         sg_set_page(&sg, sg_page(segment->sg), copied,
209                                     segment->copied + segment->sg_offset +
210                                                         segment->sg->offset);
211                 } else
212                         sg_init_one(&sg, segment->data + segment->copied,
213                                     copied);
214                 crypto_hash_update(segment->hash, &sg, copied);
215         }
216
217         segment->copied += copied;
218         if (segment->copied < segment->size) {
219                 iscsi_tcp_segment_map(segment, recv);
220                 return 0;
221         }
222
223         segment->total_copied += segment->copied;
224         segment->copied = 0;
225         segment->size = 0;
226
227         /* Unmap the current scatterlist page, if there is one. */
228         iscsi_tcp_segment_unmap(segment);
229
230         /* Do we have more scatterlist entries? */
231         debug_tcp("total copied %u total size %u\n", segment->total_copied,
232                    segment->total_size);
233         if (segment->total_copied < segment->total_size) {
234                 /* Proceed to the next entry in the scatterlist. */
235                 iscsi_tcp_segment_init_sg(segment, sg_next(segment->sg),
236                                           0);
237                 iscsi_tcp_segment_map(segment, recv);
238                 BUG_ON(segment->size == 0);
239                 return 0;
240         }
241
242         /* Do we need to handle padding? */
243         pad = iscsi_padding(segment->total_copied);
244         if (pad != 0) {
245                 debug_tcp("consume %d pad bytes\n", pad);
246                 segment->total_size += pad;
247                 segment->size = pad;
248                 segment->data = padbuf;
249                 return 0;
250         }
251
252         /*
253          * Set us up for transferring the data digest. hdr digest
254          * is completely handled in hdr done function.
255          */
256         if (segment->hash) {
257                 crypto_hash_final(segment->hash, segment->digest);
258                 iscsi_tcp_segment_splice_digest(segment,
259                                  recv ? segment->recv_digest : segment->digest);
260                 return 0;
261         }
262
263         return 1;
264 }
265
266 /**
267  * iscsi_tcp_xmit_segment - transmit segment
268  * @tcp_conn: the iSCSI TCP connection
269  * @segment: the buffer to transmnit
270  *
271  * This function transmits as much of the buffer as
272  * the network layer will accept, and returns the number of
273  * bytes transmitted.
274  *
275  * If CRC hashing is enabled, the function will compute the
276  * hash as it goes. When the entire segment has been transmitted,
277  * it will retrieve the hash value and send it as well.
278  */
279 static int
280 iscsi_tcp_xmit_segment(struct iscsi_tcp_conn *tcp_conn,
281                        struct iscsi_segment *segment)
282 {
283         struct socket *sk = tcp_conn->sock;
284         unsigned int copied = 0;
285         int r = 0;
286
287         while (!iscsi_tcp_segment_done(segment, 0, r)) {
288                 struct scatterlist *sg;
289                 unsigned int offset, copy;
290                 int flags = 0;
291
292                 r = 0;
293                 offset = segment->copied;
294                 copy = segment->size - offset;
295
296                 if (segment->total_copied + segment->size < segment->total_size)
297                         flags |= MSG_MORE;
298
299                 /* Use sendpage if we can; else fall back to sendmsg */
300                 if (!segment->data) {
301                         sg = segment->sg;
302                         offset += segment->sg_offset + sg->offset;
303                         r = tcp_conn->sendpage(sk, sg_page(sg), offset, copy,
304                                                flags);
305                 } else {
306                         struct msghdr msg = { .msg_flags = flags };
307                         struct kvec iov = {
308                                 .iov_base = segment->data + offset,
309                                 .iov_len = copy
310                         };
311
312                         r = kernel_sendmsg(sk, &msg, &iov, 1, copy);
313                 }
314
315                 if (r < 0) {
316                         iscsi_tcp_segment_unmap(segment);
317                         if (copied || r == -EAGAIN)
318                                 break;
319                         return r;
320                 }
321                 copied += r;
322         }
323         return copied;
324 }
325
326 /**
327  * iscsi_tcp_segment_recv - copy data to segment
328  * @tcp_conn: the iSCSI TCP connection
329  * @segment: the buffer to copy to
330  * @ptr: data pointer
331  * @len: amount of data available
332  *
333  * This function copies up to @len bytes to the
334  * given buffer, and returns the number of bytes
335  * consumed, which can actually be less than @len.
336  *
337  * If hash digest is enabled, the function will update the
338  * hash while copying.
339  * Combining these two operations doesn't buy us a lot (yet),
340  * but in the future we could implement combined copy+crc,
341  * just way we do for network layer checksums.
342  */
343 static int
344 iscsi_tcp_segment_recv(struct iscsi_tcp_conn *tcp_conn,
345                        struct iscsi_segment *segment, const void *ptr,
346                        unsigned int len)
347 {
348         unsigned int copy = 0, copied = 0;
349
350         while (!iscsi_tcp_segment_done(segment, 1, copy)) {
351                 if (copied == len) {
352                         debug_tcp("iscsi_tcp_segment_recv copied %d bytes\n",
353                                   len);
354                         break;
355                 }
356
357                 copy = min(len - copied, segment->size - segment->copied);
358                 debug_tcp("iscsi_tcp_segment_recv copying %d\n", copy);
359                 memcpy(segment->data + segment->copied, ptr + copied, copy);
360                 copied += copy;
361         }
362         return copied;
363 }
364
365 static inline void
366 iscsi_tcp_dgst_header(struct hash_desc *hash, const void *hdr, size_t hdrlen,
367                       unsigned char digest[ISCSI_DIGEST_SIZE])
368 {
369         struct scatterlist sg;
370
371         sg_init_one(&sg, hdr, hdrlen);
372         crypto_hash_digest(hash, &sg, hdrlen, digest);
373 }
374
375 static inline int
376 iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
377                       struct iscsi_segment *segment)
378 {
379         if (!segment->digest_len)
380                 return 1;
381
382         if (memcmp(segment->recv_digest, segment->digest,
383                    segment->digest_len)) {
384                 debug_scsi("digest mismatch\n");
385                 return 0;
386         }
387
388         return 1;
389 }
390
391 /*
392  * Helper function to set up segment buffer
393  */
394 static inline void
395 __iscsi_segment_init(struct iscsi_segment *segment, size_t size,
396                      iscsi_segment_done_fn_t *done, struct hash_desc *hash)
397 {
398         memset(segment, 0, sizeof(*segment));
399         segment->total_size = size;
400         segment->done = done;
401
402         if (hash) {
403                 segment->hash = hash;
404                 crypto_hash_init(hash);
405         }
406 }
407
408 static inline void
409 iscsi_segment_init_linear(struct iscsi_segment *segment, void *data,
410                           size_t size, iscsi_segment_done_fn_t *done,
411                           struct hash_desc *hash)
412 {
413         __iscsi_segment_init(segment, size, done, hash);
414         segment->data = data;
415         segment->size = size;
416 }
417
418 static inline int
419 iscsi_segment_seek_sg(struct iscsi_segment *segment,
420                       struct scatterlist *sg_list, unsigned int sg_count,
421                       unsigned int offset, size_t size,
422                       iscsi_segment_done_fn_t *done, struct hash_desc *hash)
423 {
424         struct scatterlist *sg;
425         unsigned int i;
426
427         debug_scsi("iscsi_segment_seek_sg offset %u size %llu\n",
428                   offset, size);
429         __iscsi_segment_init(segment, size, done, hash);
430         for_each_sg(sg_list, sg, sg_count, i) {
431                 debug_scsi("sg %d, len %u offset %u\n", i, sg->length,
432                            sg->offset);
433                 if (offset < sg->length) {
434                         iscsi_tcp_segment_init_sg(segment, sg, offset);
435                         return 0;
436                 }
437                 offset -= sg->length;
438         }
439
440         return ISCSI_ERR_DATA_OFFSET;
441 }
442
443 /**
444  * iscsi_tcp_hdr_recv_prep - prep segment for hdr reception
445  * @tcp_conn: iscsi connection to prep for
446  *
447  * This function always passes NULL for the hash argument, because when this
448  * function is called we do not yet know the final size of the header and want
449  * to delay the digest processing until we know that.
450  */
451 static void
452 iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
453 {
454         debug_tcp("iscsi_tcp_hdr_recv_prep(%p%s)\n", tcp_conn,
455                   tcp_conn->iscsi_conn->hdrdgst_en ? ", digest enabled" : "");
456         iscsi_segment_init_linear(&tcp_conn->in.segment,
457                                 tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
458                                 iscsi_tcp_hdr_recv_done, NULL);
459 }
460
461 /*
462  * Handle incoming reply to any other type of command
463  */
464 static int
465 iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
466                          struct iscsi_segment *segment)
467 {
468         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
469         int rc = 0;
470
471         if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
472                 return ISCSI_ERR_DATA_DGST;
473
474         rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr,
475                         conn->data, tcp_conn->in.datalen);
476         if (rc)
477                 return rc;
478
479         iscsi_tcp_hdr_recv_prep(tcp_conn);
480         return 0;
481 }
482
483 static void
484 iscsi_tcp_data_recv_prep(struct iscsi_tcp_conn *tcp_conn)
485 {
486         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
487         struct hash_desc *rx_hash = NULL;
488
489         if (conn->datadgst_en)
490                 rx_hash = &tcp_conn->rx_hash;
491
492         iscsi_segment_init_linear(&tcp_conn->in.segment,
493                                 conn->data, tcp_conn->in.datalen,
494                                 iscsi_tcp_data_recv_done, rx_hash);
495 }
496
497 /*
498  * must be called with session lock
499  */
500 static void
501 iscsi_tcp_cleanup_task(struct iscsi_conn *conn, struct iscsi_task *task)
502 {
503         struct iscsi_tcp_task *tcp_task = task->dd_data;
504         struct iscsi_r2t_info *r2t;
505
506         /* nothing to do for mgmt tasks */
507         if (!task->sc)
508                 return;
509
510         /* flush task's r2t queues */
511         while (__kfifo_get(tcp_task->r2tqueue, (void*)&r2t, sizeof(void*))) {
512                 __kfifo_put(tcp_task->r2tpool.queue, (void*)&r2t,
513                             sizeof(void*));
514                 debug_scsi("iscsi_tcp_cleanup_task pending r2t dropped\n");
515         }
516
517         r2t = tcp_task->r2t;
518         if (r2t != NULL) {
519                 __kfifo_put(tcp_task->r2tpool.queue, (void*)&r2t,
520                             sizeof(void*));
521                 tcp_task->r2t = NULL;
522         }
523 }
524
525 /**
526  * iscsi_data_in - SCSI Data-In Response processing
527  * @conn: iscsi connection
528  * @task: scsi command task
529  **/
530 static int
531 iscsi_data_in(struct iscsi_conn *conn, struct iscsi_task *task)
532 {
533         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
534         struct iscsi_tcp_task *tcp_task = task->dd_data;
535         struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
536         int datasn = be32_to_cpu(rhdr->datasn);
537         unsigned total_in_length = scsi_in(task->sc)->length;
538
539         iscsi_update_cmdsn(conn->session, (struct iscsi_nopin*)rhdr);
540         if (tcp_conn->in.datalen == 0)
541                 return 0;
542
543         if (tcp_task->exp_datasn != datasn) {
544                 debug_tcp("%s: task->exp_datasn(%d) != rhdr->datasn(%d)\n",
545                           __func__, tcp_task->exp_datasn, datasn);
546                 return ISCSI_ERR_DATASN;
547         }
548
549         tcp_task->exp_datasn++;
550
551         tcp_task->data_offset = be32_to_cpu(rhdr->offset);
552         if (tcp_task->data_offset + tcp_conn->in.datalen > total_in_length) {
553                 debug_tcp("%s: data_offset(%d) + data_len(%d) > total_length_in(%d)\n",
554                           __func__, tcp_task->data_offset,
555                           tcp_conn->in.datalen, total_in_length);
556                 return ISCSI_ERR_DATA_OFFSET;
557         }
558
559         conn->datain_pdus_cnt++;
560         return 0;
561 }
562
563 /**
564  * iscsi_solicit_data_init - initialize first Data-Out
565  * @conn: iscsi connection
566  * @task: scsi command task
567  * @r2t: R2T info
568  *
569  * Notes:
570  *      Initialize first Data-Out within this R2T sequence and finds
571  *      proper data_offset within this SCSI command.
572  *
573  *      This function is called with connection lock taken.
574  **/
575 static void
576 iscsi_solicit_data_init(struct iscsi_conn *conn, struct iscsi_task *task,
577                         struct iscsi_r2t_info *r2t)
578 {
579         struct iscsi_data *hdr;
580
581         hdr = &r2t->dtask.hdr;
582         memset(hdr, 0, sizeof(struct iscsi_data));
583         hdr->ttt = r2t->ttt;
584         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
585         r2t->solicit_datasn++;
586         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
587         memcpy(hdr->lun, task->hdr->lun, sizeof(hdr->lun));
588         hdr->itt = task->hdr->itt;
589         hdr->exp_statsn = r2t->exp_statsn;
590         hdr->offset = cpu_to_be32(r2t->data_offset);
591         if (r2t->data_length > conn->max_xmit_dlength) {
592                 hton24(hdr->dlength, conn->max_xmit_dlength);
593                 r2t->data_count = conn->max_xmit_dlength;
594                 hdr->flags = 0;
595         } else {
596                 hton24(hdr->dlength, r2t->data_length);
597                 r2t->data_count = r2t->data_length;
598                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
599         }
600         conn->dataout_pdus_cnt++;
601
602         r2t->sent = 0;
603 }
604
605 /**
606  * iscsi_r2t_rsp - iSCSI R2T Response processing
607  * @conn: iscsi connection
608  * @task: scsi command task
609  **/
610 static int
611 iscsi_r2t_rsp(struct iscsi_conn *conn, struct iscsi_task *task)
612 {
613         struct iscsi_r2t_info *r2t;
614         struct iscsi_session *session = conn->session;
615         struct iscsi_tcp_task *tcp_task = task->dd_data;
616         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
617         struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
618         int r2tsn = be32_to_cpu(rhdr->r2tsn);
619         int rc;
620
621         if (tcp_conn->in.datalen) {
622                 iscsi_conn_printk(KERN_ERR, conn,
623                                   "invalid R2t with datalen %d\n",
624                                   tcp_conn->in.datalen);
625                 return ISCSI_ERR_DATALEN;
626         }
627
628         if (tcp_task->exp_datasn != r2tsn){
629                 debug_tcp("%s: task->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
630                           __func__, tcp_task->exp_datasn, r2tsn);
631                 return ISCSI_ERR_R2TSN;
632         }
633
634         /* fill-in new R2T associated with the task */
635         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
636
637         if (!task->sc || session->state != ISCSI_STATE_LOGGED_IN) {
638                 iscsi_conn_printk(KERN_INFO, conn,
639                                   "dropping R2T itt %d in recovery.\n",
640                                   task->itt);
641                 return 0;
642         }
643
644         rc = __kfifo_get(tcp_task->r2tpool.queue, (void*)&r2t, sizeof(void*));
645         BUG_ON(!rc);
646
647         r2t->exp_statsn = rhdr->statsn;
648         r2t->data_length = be32_to_cpu(rhdr->data_length);
649         if (r2t->data_length == 0) {
650                 iscsi_conn_printk(KERN_ERR, conn,
651                                   "invalid R2T with zero data len\n");
652                 __kfifo_put(tcp_task->r2tpool.queue, (void*)&r2t,
653                             sizeof(void*));
654                 return ISCSI_ERR_DATALEN;
655         }
656
657         if (r2t->data_length > session->max_burst)
658                 debug_scsi("invalid R2T with data len %u and max burst %u."
659                            "Attempting to execute request.\n",
660                             r2t->data_length, session->max_burst);
661
662         r2t->data_offset = be32_to_cpu(rhdr->data_offset);
663         if (r2t->data_offset + r2t->data_length > scsi_out(task->sc)->length) {
664                 iscsi_conn_printk(KERN_ERR, conn,
665                                   "invalid R2T with data len %u at offset %u "
666                                   "and total length %d\n", r2t->data_length,
667                                   r2t->data_offset, scsi_out(task->sc)->length);
668                 __kfifo_put(tcp_task->r2tpool.queue, (void*)&r2t,
669                             sizeof(void*));
670                 return ISCSI_ERR_DATALEN;
671         }
672
673         r2t->ttt = rhdr->ttt; /* no flip */
674         r2t->solicit_datasn = 0;
675
676         iscsi_solicit_data_init(conn, task, r2t);
677
678         tcp_task->exp_datasn = r2tsn + 1;
679         __kfifo_put(tcp_task->r2tqueue, (void*)&r2t, sizeof(void*));
680         conn->r2t_pdus_cnt++;
681
682         iscsi_requeue_task(task);
683         return 0;
684 }
685
686 /*
687  * Handle incoming reply to DataIn command
688  */
689 static int
690 iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
691                           struct iscsi_segment *segment)
692 {
693         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
694         struct iscsi_hdr *hdr = tcp_conn->in.hdr;
695         int rc;
696
697         if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
698                 return ISCSI_ERR_DATA_DGST;
699
700         /* check for non-exceptional status */
701         if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
702                 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
703                 if (rc)
704                         return rc;
705         }
706
707         iscsi_tcp_hdr_recv_prep(tcp_conn);
708         return 0;
709 }
710
711 /**
712  * iscsi_tcp_hdr_dissect - process PDU header
713  * @conn: iSCSI connection
714  * @hdr: PDU header
715  *
716  * This function analyzes the header of the PDU received,
717  * and performs several sanity checks. If the PDU is accompanied
718  * by data, the receive buffer is set up to copy the incoming data
719  * to the correct location.
720  */
721 static int
722 iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
723 {
724         int rc = 0, opcode, ahslen;
725         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
726         struct iscsi_task *task;
727
728         /* verify PDU length */
729         tcp_conn->in.datalen = ntoh24(hdr->dlength);
730         if (tcp_conn->in.datalen > conn->max_recv_dlength) {
731                 iscsi_conn_printk(KERN_ERR, conn,
732                                   "iscsi_tcp: datalen %d > %d\n",
733                                   tcp_conn->in.datalen, conn->max_recv_dlength);
734                 return ISCSI_ERR_DATALEN;
735         }
736
737         /* Additional header segments. So far, we don't
738          * process additional headers.
739          */
740         ahslen = hdr->hlength << 2;
741
742         opcode = hdr->opcode & ISCSI_OPCODE_MASK;
743         /* verify itt (itt encoding: age+cid+itt) */
744         rc = iscsi_verify_itt(conn, hdr->itt);
745         if (rc)
746                 return rc;
747
748         debug_tcp("opcode 0x%x ahslen %d datalen %d\n",
749                   opcode, ahslen, tcp_conn->in.datalen);
750
751         switch(opcode) {
752         case ISCSI_OP_SCSI_DATA_IN:
753                 spin_lock(&conn->session->lock);
754                 task = iscsi_itt_to_ctask(conn, hdr->itt);
755                 if (!task)
756                         rc = ISCSI_ERR_BAD_ITT;
757                 else
758                         rc = iscsi_data_in(conn, task);
759                 if (rc) {
760                         spin_unlock(&conn->session->lock);
761                         break;
762                 }
763
764                 if (tcp_conn->in.datalen) {
765                         struct iscsi_tcp_task *tcp_task = task->dd_data;
766                         struct hash_desc *rx_hash = NULL;
767                         struct scsi_data_buffer *sdb = scsi_in(task->sc);
768
769                         /*
770                          * Setup copy of Data-In into the Scsi_Cmnd
771                          * Scatterlist case:
772                          * We set up the iscsi_segment to point to the next
773                          * scatterlist entry to copy to. As we go along,
774                          * we move on to the next scatterlist entry and
775                          * update the digest per-entry.
776                          */
777                         if (conn->datadgst_en)
778                                 rx_hash = &tcp_conn->rx_hash;
779
780                         debug_tcp("iscsi_tcp_begin_data_in(%p, offset=%d, "
781                                   "datalen=%d)\n", tcp_conn,
782                                   tcp_task->data_offset,
783                                   tcp_conn->in.datalen);
784                         rc = iscsi_segment_seek_sg(&tcp_conn->in.segment,
785                                                    sdb->table.sgl,
786                                                    sdb->table.nents,
787                                                    tcp_task->data_offset,
788                                                    tcp_conn->in.datalen,
789                                                    iscsi_tcp_process_data_in,
790                                                    rx_hash);
791                         spin_unlock(&conn->session->lock);
792                         return rc;
793                 }
794                 rc = __iscsi_complete_pdu(conn, hdr, NULL, 0);
795                 spin_unlock(&conn->session->lock);
796                 break;
797         case ISCSI_OP_SCSI_CMD_RSP:
798                 if (tcp_conn->in.datalen) {
799                         iscsi_tcp_data_recv_prep(tcp_conn);
800                         return 0;
801                 }
802                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
803                 break;
804         case ISCSI_OP_R2T:
805                 spin_lock(&conn->session->lock);
806                 task = iscsi_itt_to_ctask(conn, hdr->itt);
807                 if (!task)
808                         rc = ISCSI_ERR_BAD_ITT;
809                 else if (ahslen)
810                         rc = ISCSI_ERR_AHSLEN;
811                 else if (task->sc->sc_data_direction == DMA_TO_DEVICE)
812                         rc = iscsi_r2t_rsp(conn, task);
813                 else
814                         rc = ISCSI_ERR_PROTO;
815                 spin_unlock(&conn->session->lock);
816                 break;
817         case ISCSI_OP_LOGIN_RSP:
818         case ISCSI_OP_TEXT_RSP:
819         case ISCSI_OP_REJECT:
820         case ISCSI_OP_ASYNC_EVENT:
821                 /*
822                  * It is possible that we could get a PDU with a buffer larger
823                  * than 8K, but there are no targets that currently do this.
824                  * For now we fail until we find a vendor that needs it
825                  */
826                 if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
827                         iscsi_conn_printk(KERN_ERR, conn,
828                                           "iscsi_tcp: received buffer of "
829                                           "len %u but conn buffer is only %u "
830                                           "(opcode %0x)\n",
831                                           tcp_conn->in.datalen,
832                                           ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
833                         rc = ISCSI_ERR_PROTO;
834                         break;
835                 }
836
837                 /* If there's data coming in with the response,
838                  * receive it to the connection's buffer.
839                  */
840                 if (tcp_conn->in.datalen) {
841                         iscsi_tcp_data_recv_prep(tcp_conn);
842                         return 0;
843                 }
844         /* fall through */
845         case ISCSI_OP_LOGOUT_RSP:
846         case ISCSI_OP_NOOP_IN:
847         case ISCSI_OP_SCSI_TMFUNC_RSP:
848                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
849                 break;
850         default:
851                 rc = ISCSI_ERR_BAD_OPCODE;
852                 break;
853         }
854
855         if (rc == 0) {
856                 /* Anything that comes with data should have
857                  * been handled above. */
858                 if (tcp_conn->in.datalen)
859                         return ISCSI_ERR_PROTO;
860                 iscsi_tcp_hdr_recv_prep(tcp_conn);
861         }
862
863         return rc;
864 }
865
866 /**
867  * iscsi_tcp_hdr_recv_done - process PDU header
868  *
869  * This is the callback invoked when the PDU header has
870  * been received. If the header is followed by additional
871  * header segments, we go back for more data.
872  */
873 static int
874 iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
875                         struct iscsi_segment *segment)
876 {
877         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
878         struct iscsi_hdr *hdr;
879
880         /* Check if there are additional header segments
881          * *prior* to computing the digest, because we
882          * may need to go back to the caller for more.
883          */
884         hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
885         if (segment->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
886                 /* Bump the header length - the caller will
887                  * just loop around and get the AHS for us, and
888                  * call again. */
889                 unsigned int ahslen = hdr->hlength << 2;
890
891                 /* Make sure we don't overflow */
892                 if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
893                         return ISCSI_ERR_AHSLEN;
894
895                 segment->total_size += ahslen;
896                 segment->size += ahslen;
897                 return 0;
898         }
899
900         /* We're done processing the header. See if we're doing
901          * header digests; if so, set up the recv_digest buffer
902          * and go back for more. */
903         if (conn->hdrdgst_en) {
904                 if (segment->digest_len == 0) {
905                         iscsi_tcp_segment_splice_digest(segment,
906                                                         segment->recv_digest);
907                         return 0;
908                 }
909                 iscsi_tcp_dgst_header(&tcp_conn->rx_hash, hdr,
910                                       segment->total_copied - ISCSI_DIGEST_SIZE,
911                                       segment->digest);
912
913                 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
914                         return ISCSI_ERR_HDR_DGST;
915         }
916
917         tcp_conn->in.hdr = hdr;
918         return iscsi_tcp_hdr_dissect(conn, hdr);
919 }
920
921 /**
922  * iscsi_tcp_recv - TCP receive in sendfile fashion
923  * @rd_desc: read descriptor
924  * @skb: socket buffer
925  * @offset: offset in skb
926  * @len: skb->len - offset
927  **/
928 static int
929 iscsi_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
930                unsigned int offset, size_t len)
931 {
932         struct iscsi_conn *conn = rd_desc->arg.data;
933         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
934         struct iscsi_segment *segment = &tcp_conn->in.segment;
935         struct skb_seq_state seq;
936         unsigned int consumed = 0;
937         int rc = 0;
938
939         debug_tcp("in %d bytes\n", skb->len - offset);
940
941         if (unlikely(conn->suspend_rx)) {
942                 debug_tcp("conn %d Rx suspended!\n", conn->id);
943                 return 0;
944         }
945
946         skb_prepare_seq_read(skb, offset, skb->len, &seq);
947         while (1) {
948                 unsigned int avail;
949                 const u8 *ptr;
950
951                 avail = skb_seq_read(consumed, &ptr, &seq);
952                 if (avail == 0) {
953                         debug_tcp("no more data avail. Consumed %d\n",
954                                   consumed);
955                         break;
956                 }
957                 BUG_ON(segment->copied >= segment->size);
958
959                 debug_tcp("skb %p ptr=%p avail=%u\n", skb, ptr, avail);
960                 rc = iscsi_tcp_segment_recv(tcp_conn, segment, ptr, avail);
961                 BUG_ON(rc == 0);
962                 consumed += rc;
963
964                 if (segment->total_copied >= segment->total_size) {
965                         debug_tcp("segment done\n");
966                         rc = segment->done(tcp_conn, segment);
967                         if (rc != 0) {
968                                 skb_abort_seq_read(&seq);
969                                 goto error;
970                         }
971
972                         /* The done() functions sets up the
973                          * next segment. */
974                 }
975         }
976         skb_abort_seq_read(&seq);
977         conn->rxdata_octets += consumed;
978         return consumed;
979
980 error:
981         debug_tcp("Error receiving PDU, errno=%d\n", rc);
982         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
983         return 0;
984 }
985
986 static void
987 iscsi_tcp_data_ready(struct sock *sk, int flag)
988 {
989         struct iscsi_conn *conn = sk->sk_user_data;
990         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
991         read_descriptor_t rd_desc;
992
993         read_lock(&sk->sk_callback_lock);
994
995         /*
996          * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
997          * We set count to 1 because we want the network layer to
998          * hand us all the skbs that are available. iscsi_tcp_recv
999          * handled pdus that cross buffers or pdus that still need data.
1000          */
1001         rd_desc.arg.data = conn;
1002         rd_desc.count = 1;
1003         tcp_read_sock(sk, &rd_desc, iscsi_tcp_recv);
1004
1005         read_unlock(&sk->sk_callback_lock);
1006
1007         /* If we had to (atomically) map a highmem page,
1008          * unmap it now. */
1009         iscsi_tcp_segment_unmap(&tcp_conn->in.segment);
1010 }
1011
1012 static void
1013 iscsi_tcp_state_change(struct sock *sk)
1014 {
1015         struct iscsi_tcp_conn *tcp_conn;
1016         struct iscsi_conn *conn;
1017         struct iscsi_session *session;
1018         void (*old_state_change)(struct sock *);
1019
1020         read_lock(&sk->sk_callback_lock);
1021
1022         conn = (struct iscsi_conn*)sk->sk_user_data;
1023         session = conn->session;
1024
1025         if ((sk->sk_state == TCP_CLOSE_WAIT ||
1026              sk->sk_state == TCP_CLOSE) &&
1027             !atomic_read(&sk->sk_rmem_alloc)) {
1028                 debug_tcp("iscsi_tcp_state_change: TCP_CLOSE|TCP_CLOSE_WAIT\n");
1029                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1030         }
1031
1032         tcp_conn = conn->dd_data;
1033         old_state_change = tcp_conn->old_state_change;
1034
1035         read_unlock(&sk->sk_callback_lock);
1036
1037         old_state_change(sk);
1038 }
1039
1040 /**
1041  * iscsi_write_space - Called when more output buffer space is available
1042  * @sk: socket space is available for
1043  **/
1044 static void
1045 iscsi_write_space(struct sock *sk)
1046 {
1047         struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
1048         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1049
1050         tcp_conn->old_write_space(sk);
1051         debug_tcp("iscsi_write_space: cid %d\n", conn->id);
1052         scsi_queue_work(conn->session->host, &conn->xmitwork);
1053 }
1054
1055 static void
1056 iscsi_conn_set_callbacks(struct iscsi_conn *conn)
1057 {
1058         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1059         struct sock *sk = tcp_conn->sock->sk;
1060
1061         /* assign new callbacks */
1062         write_lock_bh(&sk->sk_callback_lock);
1063         sk->sk_user_data = conn;
1064         tcp_conn->old_data_ready = sk->sk_data_ready;
1065         tcp_conn->old_state_change = sk->sk_state_change;
1066         tcp_conn->old_write_space = sk->sk_write_space;
1067         sk->sk_data_ready = iscsi_tcp_data_ready;
1068         sk->sk_state_change = iscsi_tcp_state_change;
1069         sk->sk_write_space = iscsi_write_space;
1070         write_unlock_bh(&sk->sk_callback_lock);
1071 }
1072
1073 static void
1074 iscsi_conn_restore_callbacks(struct iscsi_tcp_conn *tcp_conn)
1075 {
1076         struct sock *sk = tcp_conn->sock->sk;
1077
1078         /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
1079         write_lock_bh(&sk->sk_callback_lock);
1080         sk->sk_user_data    = NULL;
1081         sk->sk_data_ready   = tcp_conn->old_data_ready;
1082         sk->sk_state_change = tcp_conn->old_state_change;
1083         sk->sk_write_space  = tcp_conn->old_write_space;
1084         sk->sk_no_check  = 0;
1085         write_unlock_bh(&sk->sk_callback_lock);
1086 }
1087
1088 /**
1089  * iscsi_xmit - TCP transmit
1090  **/
1091 static int
1092 iscsi_xmit(struct iscsi_conn *conn)
1093 {
1094         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1095         struct iscsi_segment *segment = &tcp_conn->out.segment;
1096         unsigned int consumed = 0;
1097         int rc = 0;
1098
1099         while (1) {
1100                 rc = iscsi_tcp_xmit_segment(tcp_conn, segment);
1101                 if (rc < 0)
1102                         goto error;
1103                 if (rc == 0)
1104                         break;
1105
1106                 consumed += rc;
1107
1108                 if (segment->total_copied >= segment->total_size) {
1109                         if (segment->done != NULL) {
1110                                 rc = segment->done(tcp_conn, segment);
1111                                 if (rc < 0)
1112                                         goto error;
1113                         }
1114                 }
1115         }
1116
1117         debug_tcp("xmit %d bytes\n", consumed);
1118
1119         conn->txdata_octets += consumed;
1120         return consumed;
1121
1122 error:
1123         /* Transmit error. We could initiate error recovery
1124          * here. */
1125         debug_tcp("Error sending PDU, errno=%d\n", rc);
1126         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1127         return rc;
1128 }
1129
1130 /**
1131  * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
1132  */
1133 static inline int
1134 iscsi_tcp_xmit_qlen(struct iscsi_conn *conn)
1135 {
1136         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1137         struct iscsi_segment *segment = &tcp_conn->out.segment;
1138
1139         return segment->total_copied - segment->total_size;
1140 }
1141
1142 static inline int
1143 iscsi_tcp_flush(struct iscsi_conn *conn)
1144 {
1145         int rc;
1146
1147         while (iscsi_tcp_xmit_qlen(conn)) {
1148                 rc = iscsi_xmit(conn);
1149                 if (rc == 0)
1150                         return -EAGAIN;
1151                 if (rc < 0)
1152                         return rc;
1153         }
1154
1155         return 0;
1156 }
1157
1158 /*
1159  * This is called when we're done sending the header.
1160  * Simply copy the data_segment to the send segment, and return.
1161  */
1162 static int
1163 iscsi_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
1164                         struct iscsi_segment *segment)
1165 {
1166         tcp_conn->out.segment = tcp_conn->out.data_segment;
1167         debug_tcp("Header done. Next segment size %u total_size %u\n",
1168                   tcp_conn->out.segment.size, tcp_conn->out.segment.total_size);
1169         return 0;
1170 }
1171
1172 static void
1173 iscsi_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr, size_t hdrlen)
1174 {
1175         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1176
1177         debug_tcp("%s(%p%s)\n", __func__, tcp_conn,
1178                         conn->hdrdgst_en? ", digest enabled" : "");
1179
1180         /* Clear the data segment - needs to be filled in by the
1181          * caller using iscsi_tcp_send_data_prep() */
1182         memset(&tcp_conn->out.data_segment, 0, sizeof(struct iscsi_segment));
1183
1184         /* If header digest is enabled, compute the CRC and
1185          * place the digest into the same buffer. We make
1186          * sure that both iscsi_tcp_task and mtask have
1187          * sufficient room.
1188          */
1189         if (conn->hdrdgst_en) {
1190                 iscsi_tcp_dgst_header(&tcp_conn->tx_hash, hdr, hdrlen,
1191                                       hdr + hdrlen);
1192                 hdrlen += ISCSI_DIGEST_SIZE;
1193         }
1194
1195         /* Remember header pointer for later, when we need
1196          * to decide whether there's a payload to go along
1197          * with the header. */
1198         tcp_conn->out.hdr = hdr;
1199
1200         iscsi_segment_init_linear(&tcp_conn->out.segment, hdr, hdrlen,
1201                                 iscsi_tcp_send_hdr_done, NULL);
1202 }
1203
1204 /*
1205  * Prepare the send buffer for the payload data.
1206  * Padding and checksumming will all be taken care
1207  * of by the iscsi_segment routines.
1208  */
1209 static int
1210 iscsi_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
1211                          unsigned int count, unsigned int offset,
1212                          unsigned int len)
1213 {
1214         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1215         struct hash_desc *tx_hash = NULL;
1216         unsigned int hdr_spec_len;
1217
1218         debug_tcp("%s(%p, offset=%d, datalen=%d%s)\n", __func__,
1219                         tcp_conn, offset, len,
1220                         conn->datadgst_en? ", digest enabled" : "");
1221
1222         /* Make sure the datalen matches what the caller
1223            said he would send. */
1224         hdr_spec_len = ntoh24(tcp_conn->out.hdr->dlength);
1225         WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
1226
1227         if (conn->datadgst_en)
1228                 tx_hash = &tcp_conn->tx_hash;
1229
1230         return iscsi_segment_seek_sg(&tcp_conn->out.data_segment,
1231                                    sg, count, offset, len,
1232                                    NULL, tx_hash);
1233 }
1234
1235 static void
1236 iscsi_tcp_send_linear_data_prepare(struct iscsi_conn *conn, void *data,
1237                                    size_t len)
1238 {
1239         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1240         struct hash_desc *tx_hash = NULL;
1241         unsigned int hdr_spec_len;
1242
1243         debug_tcp("%s(%p, datalen=%d%s)\n", __func__, tcp_conn, len,
1244                   conn->datadgst_en? ", digest enabled" : "");
1245
1246         /* Make sure the datalen matches what the caller
1247            said he would send. */
1248         hdr_spec_len = ntoh24(tcp_conn->out.hdr->dlength);
1249         WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
1250
1251         if (conn->datadgst_en)
1252                 tx_hash = &tcp_conn->tx_hash;
1253
1254         iscsi_segment_init_linear(&tcp_conn->out.data_segment,
1255                                 data, len, NULL, tx_hash);
1256 }
1257
1258 /**
1259  * iscsi_solicit_data_cont - initialize next Data-Out
1260  * @conn: iscsi connection
1261  * @task: scsi command task
1262  * @r2t: R2T info
1263  * @left: bytes left to transfer
1264  *
1265  * Notes:
1266  *      Initialize next Data-Out within this R2T sequence and continue
1267  *      to process next Scatter-Gather element(if any) of this SCSI command.
1268  *
1269  *      Called under connection lock.
1270  **/
1271 static int
1272 iscsi_solicit_data_cont(struct iscsi_conn *conn, struct iscsi_task *task,
1273                         struct iscsi_r2t_info *r2t)
1274 {
1275         struct iscsi_data *hdr;
1276         int new_offset, left;
1277
1278         BUG_ON(r2t->data_length - r2t->sent < 0);
1279         left = r2t->data_length - r2t->sent;
1280         if (left == 0)
1281                 return 0;
1282
1283         hdr = &r2t->dtask.hdr;
1284         memset(hdr, 0, sizeof(struct iscsi_data));
1285         hdr->ttt = r2t->ttt;
1286         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
1287         r2t->solicit_datasn++;
1288         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
1289         memcpy(hdr->lun, task->hdr->lun, sizeof(hdr->lun));
1290         hdr->itt = task->hdr->itt;
1291         hdr->exp_statsn = r2t->exp_statsn;
1292         new_offset = r2t->data_offset + r2t->sent;
1293         hdr->offset = cpu_to_be32(new_offset);
1294         if (left > conn->max_xmit_dlength) {
1295                 hton24(hdr->dlength, conn->max_xmit_dlength);
1296                 r2t->data_count = conn->max_xmit_dlength;
1297         } else {
1298                 hton24(hdr->dlength, left);
1299                 r2t->data_count = left;
1300                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
1301         }
1302
1303         conn->dataout_pdus_cnt++;
1304         return 1;
1305 }
1306
1307 /**
1308  * iscsi_tcp_task - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
1309  * @conn: iscsi connection
1310  * @task: scsi command task
1311  * @sc: scsi command
1312  **/
1313 static int
1314 iscsi_tcp_task_init(struct iscsi_task *task)
1315 {
1316         struct iscsi_tcp_task *tcp_task = task->dd_data;
1317         struct iscsi_conn *conn = task->conn;
1318         struct scsi_cmnd *sc = task->sc;
1319         int err;
1320
1321         if (!sc) {
1322                 /*
1323                  * mgmt tasks do not have a scatterlist since they come
1324                  * in from the iscsi interface.
1325                  */
1326                 debug_scsi("mtask deq [cid %d itt 0x%x]\n", conn->id,
1327                            task->itt);
1328
1329                 /* Prepare PDU, optionally w/ immediate data */
1330                 iscsi_tcp_send_hdr_prep(conn, task->hdr, sizeof(*task->hdr));
1331
1332                 /* If we have immediate data, attach a payload */
1333                 if (task->data_count)
1334                         iscsi_tcp_send_linear_data_prepare(conn, task->data,
1335                                                            task->data_count);
1336                 return 0;
1337         }
1338
1339         BUG_ON(__kfifo_len(tcp_task->r2tqueue));
1340         tcp_task->sent = 0;
1341         tcp_task->exp_datasn = 0;
1342
1343         /* Prepare PDU, optionally w/ immediate data */
1344         debug_scsi("task deq [cid %d itt 0x%x imm %d unsol %d]\n",
1345                     conn->id, task->itt, task->imm_count,
1346                     task->unsol_count);
1347         iscsi_tcp_send_hdr_prep(conn, task->hdr, task->hdr_len);
1348
1349         if (!task->imm_count)
1350                 return 0;
1351
1352         /* If we have immediate data, attach a payload */
1353         err = iscsi_tcp_send_data_prep(conn, scsi_out(sc)->table.sgl,
1354                                        scsi_out(sc)->table.nents,
1355                                        0, task->imm_count);
1356         if (err)
1357                 return err;
1358         tcp_task->sent += task->imm_count;
1359         task->imm_count = 0;
1360         return 0;
1361 }
1362
1363 /*
1364  * iscsi_tcp_task_xmit - xmit normal PDU task
1365  * @task: iscsi command task
1366  *
1367  * We're expected to return 0 when everything was transmitted succesfully,
1368  * -EAGAIN if there's still data in the queue, or != 0 for any other kind
1369  * of error.
1370  */
1371 static int
1372 iscsi_tcp_task_xmit(struct iscsi_task *task)
1373 {
1374         struct iscsi_conn *conn = task->conn;
1375         struct iscsi_tcp_task *tcp_task = task->dd_data;
1376         struct scsi_cmnd *sc = task->sc;
1377         struct scsi_data_buffer *sdb;
1378         int rc = 0;
1379
1380 flush:
1381         /* Flush any pending data first. */
1382         rc = iscsi_tcp_flush(conn);
1383         if (rc < 0)
1384                 return rc;
1385
1386         /* mgmt command */
1387         if (!sc) {
1388                 if (task->hdr->itt == RESERVED_ITT)
1389                         iscsi_put_task(task);
1390                 return 0;
1391         }
1392
1393         /* Are we done already? */
1394         if (sc->sc_data_direction != DMA_TO_DEVICE)
1395                 return 0;
1396
1397         sdb = scsi_out(sc);
1398         if (task->unsol_count != 0) {
1399                 struct iscsi_data *hdr = &tcp_task->unsol_dtask.hdr;
1400
1401                 /* Prepare a header for the unsolicited PDU.
1402                  * The amount of data we want to send will be
1403                  * in task->data_count.
1404                  * FIXME: return the data count instead.
1405                  */
1406                 iscsi_prep_unsolicit_data_pdu(task, hdr);
1407
1408                 debug_tcp("unsol dout [itt 0x%x doff %d dlen %d]\n",
1409                                 task->itt, tcp_task->sent, task->data_count);
1410
1411                 iscsi_tcp_send_hdr_prep(conn, hdr, sizeof(*hdr));
1412                 rc = iscsi_tcp_send_data_prep(conn, sdb->table.sgl,
1413                                               sdb->table.nents, tcp_task->sent,
1414                                               task->data_count);
1415                 if (rc)
1416                         goto fail;
1417                 tcp_task->sent += task->data_count;
1418                 task->unsol_count -= task->data_count;
1419                 goto flush;
1420         } else {
1421                 struct iscsi_session *session = conn->session;
1422                 struct iscsi_r2t_info *r2t;
1423
1424                 /* All unsolicited PDUs sent. Check for solicited PDUs.
1425                  */
1426                 spin_lock_bh(&session->lock);
1427                 r2t = tcp_task->r2t;
1428                 if (r2t != NULL) {
1429                         /* Continue with this R2T? */
1430                         if (!iscsi_solicit_data_cont(conn, task, r2t)) {
1431                                 debug_scsi("  done with r2t %p\n", r2t);
1432
1433                                 __kfifo_put(tcp_task->r2tpool.queue,
1434                                             (void*)&r2t, sizeof(void*));
1435                                 tcp_task->r2t = r2t = NULL;
1436                         }
1437                 }
1438
1439                 if (r2t == NULL) {
1440                         __kfifo_get(tcp_task->r2tqueue, (void*)&tcp_task->r2t,
1441                                     sizeof(void*));
1442                         r2t = tcp_task->r2t;
1443                 }
1444                 spin_unlock_bh(&session->lock);
1445
1446                 /* Waiting for more R2Ts to arrive. */
1447                 if (r2t == NULL) {
1448                         debug_tcp("no R2Ts yet\n");
1449                         return 0;
1450                 }
1451
1452                 debug_scsi("sol dout %p [dsn %d itt 0x%x doff %d dlen %d]\n",
1453                         r2t, r2t->solicit_datasn - 1, task->itt,
1454                         r2t->data_offset + r2t->sent, r2t->data_count);
1455
1456                 iscsi_tcp_send_hdr_prep(conn, &r2t->dtask.hdr,
1457                                         sizeof(struct iscsi_hdr));
1458
1459                 rc = iscsi_tcp_send_data_prep(conn, sdb->table.sgl,
1460                                               sdb->table.nents,
1461                                               r2t->data_offset + r2t->sent,
1462                                               r2t->data_count);
1463                 if (rc)
1464                         goto fail;
1465                 tcp_task->sent += r2t->data_count;
1466                 r2t->sent += r2t->data_count;
1467                 goto flush;
1468         }
1469         return 0;
1470 fail:
1471         iscsi_conn_failure(conn, rc);
1472         return -EIO;
1473 }
1474
1475 static struct iscsi_cls_conn *
1476 iscsi_tcp_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
1477 {
1478         struct iscsi_conn *conn;
1479         struct iscsi_cls_conn *cls_conn;
1480         struct iscsi_tcp_conn *tcp_conn;
1481
1482         cls_conn = iscsi_conn_setup(cls_session, sizeof(*tcp_conn), conn_idx);
1483         if (!cls_conn)
1484                 return NULL;
1485         conn = cls_conn->dd_data;
1486         /*
1487          * due to strange issues with iser these are not set
1488          * in iscsi_conn_setup
1489          */
1490         conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1491
1492         tcp_conn = conn->dd_data;
1493         tcp_conn->iscsi_conn = conn;
1494
1495         tcp_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1496                                                   CRYPTO_ALG_ASYNC);
1497         tcp_conn->tx_hash.flags = 0;
1498         if (IS_ERR(tcp_conn->tx_hash.tfm))
1499                 goto free_conn;
1500
1501         tcp_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1502                                                   CRYPTO_ALG_ASYNC);
1503         tcp_conn->rx_hash.flags = 0;
1504         if (IS_ERR(tcp_conn->rx_hash.tfm))
1505                 goto free_tx_tfm;
1506
1507         return cls_conn;
1508
1509 free_tx_tfm:
1510         crypto_free_hash(tcp_conn->tx_hash.tfm);
1511 free_conn:
1512         iscsi_conn_printk(KERN_ERR, conn,
1513                           "Could not create connection due to crc32c "
1514                           "loading error. Make sure the crc32c "
1515                           "module is built as a module or into the "
1516                           "kernel\n");
1517         iscsi_conn_teardown(cls_conn);
1518         return NULL;
1519 }
1520
1521 static void
1522 iscsi_tcp_release_conn(struct iscsi_conn *conn)
1523 {
1524         struct iscsi_session *session = conn->session;
1525         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1526         struct socket *sock = tcp_conn->sock;
1527
1528         if (!sock)
1529                 return;
1530
1531         sock_hold(sock->sk);
1532         iscsi_conn_restore_callbacks(tcp_conn);
1533         sock_put(sock->sk);
1534
1535         spin_lock_bh(&session->lock);
1536         tcp_conn->sock = NULL;
1537         spin_unlock_bh(&session->lock);
1538         sockfd_put(sock);
1539 }
1540
1541 static void
1542 iscsi_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
1543 {
1544         struct iscsi_conn *conn = cls_conn->dd_data;
1545         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1546
1547         iscsi_tcp_release_conn(conn);
1548
1549         if (tcp_conn->tx_hash.tfm)
1550                 crypto_free_hash(tcp_conn->tx_hash.tfm);
1551         if (tcp_conn->rx_hash.tfm)
1552                 crypto_free_hash(tcp_conn->rx_hash.tfm);
1553
1554         iscsi_conn_teardown(cls_conn);
1555 }
1556
1557 static void
1558 iscsi_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
1559 {
1560         struct iscsi_conn *conn = cls_conn->dd_data;
1561         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1562
1563         /* userspace may have goofed up and not bound us */
1564         if (!tcp_conn->sock)
1565                 return;
1566         /*
1567          * Make sure our recv side is stopped.
1568          * Older tools called conn stop before ep_disconnect
1569          * so IO could still be coming in.
1570          */
1571         write_lock_bh(&tcp_conn->sock->sk->sk_callback_lock);
1572         set_bit(ISCSI_SUSPEND_BIT, &conn->suspend_rx);
1573         write_unlock_bh(&tcp_conn->sock->sk->sk_callback_lock);
1574
1575         iscsi_conn_stop(cls_conn, flag);
1576         iscsi_tcp_release_conn(conn);
1577 }
1578
1579 static int iscsi_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
1580                               char *buf, int *port,
1581                               int (*getname)(struct socket *, struct sockaddr *,
1582                                         int *addrlen))
1583 {
1584         struct sockaddr_storage *addr;
1585         struct sockaddr_in6 *sin6;
1586         struct sockaddr_in *sin;
1587         int rc = 0, len;
1588
1589         addr = kmalloc(sizeof(*addr), GFP_KERNEL);
1590         if (!addr)
1591                 return -ENOMEM;
1592
1593         if (getname(sock, (struct sockaddr *) addr, &len)) {
1594                 rc = -ENODEV;
1595                 goto free_addr;
1596         }
1597
1598         switch (addr->ss_family) {
1599         case AF_INET:
1600                 sin = (struct sockaddr_in *)addr;
1601                 spin_lock_bh(&conn->session->lock);
1602                 sprintf(buf, NIPQUAD_FMT, NIPQUAD(sin->sin_addr.s_addr));
1603                 *port = be16_to_cpu(sin->sin_port);
1604                 spin_unlock_bh(&conn->session->lock);
1605                 break;
1606         case AF_INET6:
1607                 sin6 = (struct sockaddr_in6 *)addr;
1608                 spin_lock_bh(&conn->session->lock);
1609                 sprintf(buf, NIP6_FMT, NIP6(sin6->sin6_addr));
1610                 *port = be16_to_cpu(sin6->sin6_port);
1611                 spin_unlock_bh(&conn->session->lock);
1612                 break;
1613         }
1614 free_addr:
1615         kfree(addr);
1616         return rc;
1617 }
1618
1619 static int
1620 iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
1621                     struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
1622                     int is_leading)
1623 {
1624         struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
1625         struct iscsi_host *ihost = shost_priv(shost);
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, ihost->local_address,
1650                                 &ihost->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         iscsi_conn_set_callbacks(conn);
1668         tcp_conn->sendpage = tcp_conn->sock->ops->sendpage;
1669         /*
1670          * set receive state machine into initial state
1671          */
1672         iscsi_tcp_hdr_recv_prep(tcp_conn);
1673         return 0;
1674
1675 free_socket:
1676         sockfd_put(sock);
1677         return err;
1678 }
1679
1680 static int
1681 iscsi_r2tpool_alloc(struct iscsi_session *session)
1682 {
1683         int i;
1684         int cmd_i;
1685
1686         /*
1687          * initialize per-task: R2T pool and xmit queue
1688          */
1689         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1690                 struct iscsi_task *task = session->cmds[cmd_i];
1691                 struct iscsi_tcp_task *tcp_task = task->dd_data;
1692
1693                 /*
1694                  * pre-allocated x4 as much r2ts to handle race when
1695                  * target acks DataOut faster than we data_xmit() queues
1696                  * could replenish r2tqueue.
1697                  */
1698
1699                 /* R2T pool */
1700                 if (iscsi_pool_init(&tcp_task->r2tpool, session->max_r2t * 4, NULL,
1701                                     sizeof(struct iscsi_r2t_info))) {
1702                         goto r2t_alloc_fail;
1703                 }
1704
1705                 /* R2T xmit queue */
1706                 tcp_task->r2tqueue = kfifo_alloc(
1707                       session->max_r2t * 4 * sizeof(void*), GFP_KERNEL, NULL);
1708                 if (tcp_task->r2tqueue == ERR_PTR(-ENOMEM)) {
1709                         iscsi_pool_free(&tcp_task->r2tpool);
1710                         goto r2t_alloc_fail;
1711                 }
1712         }
1713
1714         return 0;
1715
1716 r2t_alloc_fail:
1717         for (i = 0; i < cmd_i; i++) {
1718                 struct iscsi_task *task = session->cmds[i];
1719                 struct iscsi_tcp_task *tcp_task = task->dd_data;
1720
1721                 kfifo_free(tcp_task->r2tqueue);
1722                 iscsi_pool_free(&tcp_task->r2tpool);
1723         }
1724         return -ENOMEM;
1725 }
1726
1727 static void
1728 iscsi_r2tpool_free(struct iscsi_session *session)
1729 {
1730         int i;
1731
1732         for (i = 0; i < session->cmds_max; i++) {
1733                 struct iscsi_task *task = session->cmds[i];
1734                 struct iscsi_tcp_task *tcp_task = task->dd_data;
1735
1736                 kfifo_free(tcp_task->r2tqueue);
1737                 iscsi_pool_free(&tcp_task->r2tpool);
1738         }
1739 }
1740
1741 static int
1742 iscsi_conn_set_param(struct iscsi_cls_conn *cls_conn, enum iscsi_param param,
1743                      char *buf, int buflen)
1744 {
1745         struct iscsi_conn *conn = cls_conn->dd_data;
1746         struct iscsi_session *session = conn->session;
1747         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1748         int value;
1749
1750         switch(param) {
1751         case ISCSI_PARAM_HDRDGST_EN:
1752                 iscsi_set_param(cls_conn, param, buf, buflen);
1753                 break;
1754         case ISCSI_PARAM_DATADGST_EN:
1755                 iscsi_set_param(cls_conn, param, buf, buflen);
1756                 tcp_conn->sendpage = conn->datadgst_en ?
1757                         sock_no_sendpage : tcp_conn->sock->ops->sendpage;
1758                 break;
1759         case ISCSI_PARAM_MAX_R2T:
1760                 sscanf(buf, "%d", &value);
1761                 if (value <= 0 || !is_power_of_2(value))
1762                         return -EINVAL;
1763                 if (session->max_r2t == value)
1764                         break;
1765                 iscsi_r2tpool_free(session);
1766                 iscsi_set_param(cls_conn, param, buf, buflen);
1767                 if (iscsi_r2tpool_alloc(session))
1768                         return -ENOMEM;
1769                 break;
1770         default:
1771                 return iscsi_set_param(cls_conn, param, buf, buflen);
1772         }
1773
1774         return 0;
1775 }
1776
1777 static int
1778 iscsi_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
1779                          enum iscsi_param param, char *buf)
1780 {
1781         struct iscsi_conn *conn = cls_conn->dd_data;
1782         int len;
1783
1784         switch(param) {
1785         case ISCSI_PARAM_CONN_PORT:
1786                 spin_lock_bh(&conn->session->lock);
1787                 len = sprintf(buf, "%hu\n", conn->portal_port);
1788                 spin_unlock_bh(&conn->session->lock);
1789                 break;
1790         case ISCSI_PARAM_CONN_ADDRESS:
1791                 spin_lock_bh(&conn->session->lock);
1792                 len = sprintf(buf, "%s\n", conn->portal_address);
1793                 spin_unlock_bh(&conn->session->lock);
1794                 break;
1795         default:
1796                 return iscsi_conn_get_param(cls_conn, param, buf);
1797         }
1798
1799         return len;
1800 }
1801
1802 static void
1803 iscsi_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
1804 {
1805         struct iscsi_conn *conn = cls_conn->dd_data;
1806         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1807
1808         stats->txdata_octets = conn->txdata_octets;
1809         stats->rxdata_octets = conn->rxdata_octets;
1810         stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
1811         stats->dataout_pdus = conn->dataout_pdus_cnt;
1812         stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
1813         stats->datain_pdus = conn->datain_pdus_cnt;
1814         stats->r2t_pdus = conn->r2t_pdus_cnt;
1815         stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
1816         stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
1817         stats->custom_length = 3;
1818         strcpy(stats->custom[0].desc, "tx_sendpage_failures");
1819         stats->custom[0].value = tcp_conn->sendpage_failures_cnt;
1820         strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
1821         stats->custom[1].value = tcp_conn->discontiguous_hdr_cnt;
1822         strcpy(stats->custom[2].desc, "eh_abort_cnt");
1823         stats->custom[2].value = conn->eh_abort_cnt;
1824 }
1825
1826 static struct iscsi_cls_session *
1827 iscsi_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,
1828                          uint16_t qdepth, uint32_t initial_cmdsn,
1829                          uint32_t *hostno)
1830 {
1831         struct iscsi_cls_session *cls_session;
1832         struct iscsi_session *session;
1833         struct Scsi_Host *shost;
1834         int cmd_i;
1835
1836         if (ep) {
1837                 printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep);
1838                 return NULL;
1839         }
1840
1841         shost = iscsi_host_alloc(&iscsi_sht, 0, qdepth);
1842         if (!shost)
1843                 return NULL;
1844         shost->transportt = iscsi_tcp_scsi_transport;
1845         shost->max_lun = iscsi_max_lun;
1846         shost->max_id = 0;
1847         shost->max_channel = 0;
1848         shost->max_cmd_len = SCSI_MAX_VARLEN_CDB_SIZE;
1849
1850         if (iscsi_host_add(shost, NULL))
1851                 goto free_host;
1852         *hostno = shost->host_no;
1853
1854         cls_session = iscsi_session_setup(&iscsi_tcp_transport, shost, cmds_max,
1855                                           sizeof(struct iscsi_tcp_task),
1856                                           initial_cmdsn, 0);
1857         if (!cls_session)
1858                 goto remove_host;
1859         session = cls_session->dd_data;
1860
1861         shost->can_queue = session->scsi_cmds_max;
1862         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1863                 struct iscsi_task *task = session->cmds[cmd_i];
1864                 struct iscsi_tcp_task *tcp_task = task->dd_data;
1865
1866                 task->hdr = &tcp_task->hdr.cmd_hdr;
1867                 task->hdr_max = sizeof(tcp_task->hdr) - ISCSI_DIGEST_SIZE;
1868         }
1869
1870         if (iscsi_r2tpool_alloc(session))
1871                 goto remove_session;
1872         return cls_session;
1873
1874 remove_session:
1875         iscsi_session_teardown(cls_session);
1876 remove_host:
1877         iscsi_host_remove(shost);
1878 free_host:
1879         iscsi_host_free(shost);
1880         return NULL;
1881 }
1882
1883 static void iscsi_tcp_session_destroy(struct iscsi_cls_session *cls_session)
1884 {
1885         struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
1886
1887         iscsi_r2tpool_free(cls_session->dd_data);
1888
1889         iscsi_host_remove(shost);
1890         iscsi_host_free(shost);
1891 }
1892
1893 static int iscsi_tcp_slave_configure(struct scsi_device *sdev)
1894 {
1895         blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
1896         blk_queue_dma_alignment(sdev->request_queue, 0);
1897         return 0;
1898 }
1899
1900 static struct scsi_host_template iscsi_sht = {
1901         .module                 = THIS_MODULE,
1902         .name                   = "iSCSI Initiator over TCP/IP",
1903         .queuecommand           = iscsi_queuecommand,
1904         .change_queue_depth     = iscsi_change_queue_depth,
1905         .can_queue              = ISCSI_DEF_XMIT_CMDS_MAX - 1,
1906         .sg_tablesize           = 4096,
1907         .max_sectors            = 0xFFFF,
1908         .cmd_per_lun            = ISCSI_DEF_CMD_PER_LUN,
1909         .eh_abort_handler       = iscsi_eh_abort,
1910         .eh_device_reset_handler= iscsi_eh_device_reset,
1911         .eh_host_reset_handler  = iscsi_eh_host_reset,
1912         .use_clustering         = DISABLE_CLUSTERING,
1913         .slave_configure        = iscsi_tcp_slave_configure,
1914         .proc_name              = "iscsi_tcp",
1915         .this_id                = -1,
1916 };
1917
1918 static struct iscsi_transport iscsi_tcp_transport = {
1919         .owner                  = THIS_MODULE,
1920         .name                   = "tcp",
1921         .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
1922                                   | CAP_DATADGST,
1923         .param_mask             = ISCSI_MAX_RECV_DLENGTH |
1924                                   ISCSI_MAX_XMIT_DLENGTH |
1925                                   ISCSI_HDRDGST_EN |
1926                                   ISCSI_DATADGST_EN |
1927                                   ISCSI_INITIAL_R2T_EN |
1928                                   ISCSI_MAX_R2T |
1929                                   ISCSI_IMM_DATA_EN |
1930                                   ISCSI_FIRST_BURST |
1931                                   ISCSI_MAX_BURST |
1932                                   ISCSI_PDU_INORDER_EN |
1933                                   ISCSI_DATASEQ_INORDER_EN |
1934                                   ISCSI_ERL |
1935                                   ISCSI_CONN_PORT |
1936                                   ISCSI_CONN_ADDRESS |
1937                                   ISCSI_EXP_STATSN |
1938                                   ISCSI_PERSISTENT_PORT |
1939                                   ISCSI_PERSISTENT_ADDRESS |
1940                                   ISCSI_TARGET_NAME | ISCSI_TPGT |
1941                                   ISCSI_USERNAME | ISCSI_PASSWORD |
1942                                   ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
1943                                   ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
1944                                   ISCSI_LU_RESET_TMO |
1945                                   ISCSI_PING_TMO | ISCSI_RECV_TMO |
1946                                   ISCSI_IFACE_NAME | ISCSI_INITIATOR_NAME,
1947         .host_param_mask        = ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
1948                                   ISCSI_HOST_INITIATOR_NAME |
1949                                   ISCSI_HOST_NETDEV_NAME,
1950         /* session management */
1951         .create_session         = iscsi_tcp_session_create,
1952         .destroy_session        = iscsi_tcp_session_destroy,
1953         /* connection management */
1954         .create_conn            = iscsi_tcp_conn_create,
1955         .bind_conn              = iscsi_tcp_conn_bind,
1956         .destroy_conn           = iscsi_tcp_conn_destroy,
1957         .set_param              = iscsi_conn_set_param,
1958         .get_conn_param         = iscsi_tcp_conn_get_param,
1959         .get_session_param      = iscsi_session_get_param,
1960         .start_conn             = iscsi_conn_start,
1961         .stop_conn              = iscsi_tcp_conn_stop,
1962         /* iscsi host params */
1963         .get_host_param         = iscsi_host_get_param,
1964         .set_host_param         = iscsi_host_set_param,
1965         /* IO */
1966         .send_pdu               = iscsi_conn_send_pdu,
1967         .get_stats              = iscsi_conn_get_stats,
1968         .init_task              = iscsi_tcp_task_init,
1969         .xmit_task              = iscsi_tcp_task_xmit,
1970         .cleanup_task           = iscsi_tcp_cleanup_task,
1971         /* recovery */
1972         .session_recovery_timedout = iscsi_session_recovery_timedout,
1973 };
1974
1975 static int __init
1976 iscsi_tcp_init(void)
1977 {
1978         if (iscsi_max_lun < 1) {
1979                 printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
1980                        iscsi_max_lun);
1981                 return -EINVAL;
1982         }
1983
1984         iscsi_tcp_scsi_transport = iscsi_register_transport(
1985                                                         &iscsi_tcp_transport);
1986         if (!iscsi_tcp_scsi_transport)
1987                 return -ENODEV;
1988
1989         return 0;
1990 }
1991
1992 static void __exit
1993 iscsi_tcp_exit(void)
1994 {
1995         iscsi_unregister_transport(&iscsi_tcp_transport);
1996 }
1997
1998 module_init(iscsi_tcp_init);
1999 module_exit(iscsi_tcp_exit);