[SCSI] libiscsi, iscsi_tcp, iser: add session cmds array accessor
[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_ctask(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
502 {
503         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
504         struct iscsi_r2t_info *r2t;
505
506         /* flush ctask's r2t queues */
507         while (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*))) {
508                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
509                             sizeof(void*));
510                 debug_scsi("iscsi_tcp_cleanup_ctask pending r2t dropped\n");
511         }
512
513         r2t = tcp_ctask->r2t;
514         if (r2t != NULL) {
515                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
516                             sizeof(void*));
517                 tcp_ctask->r2t = NULL;
518         }
519 }
520
521 /**
522  * iscsi_data_rsp - SCSI Data-In Response processing
523  * @conn: iscsi connection
524  * @ctask: scsi command task
525  **/
526 static int
527 iscsi_data_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
528 {
529         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
530         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
531         struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
532         struct iscsi_session *session = conn->session;
533         struct scsi_cmnd *sc = ctask->sc;
534         int datasn = be32_to_cpu(rhdr->datasn);
535         unsigned total_in_length = scsi_in(sc)->length;
536
537         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
538         if (tcp_conn->in.datalen == 0)
539                 return 0;
540
541         if (tcp_ctask->exp_datasn != datasn) {
542                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->datasn(%d)\n",
543                           __FUNCTION__, tcp_ctask->exp_datasn, datasn);
544                 return ISCSI_ERR_DATASN;
545         }
546
547         tcp_ctask->exp_datasn++;
548
549         tcp_ctask->data_offset = be32_to_cpu(rhdr->offset);
550         if (tcp_ctask->data_offset + tcp_conn->in.datalen > total_in_length) {
551                 debug_tcp("%s: data_offset(%d) + data_len(%d) > total_length_in(%d)\n",
552                           __FUNCTION__, tcp_ctask->data_offset,
553                           tcp_conn->in.datalen, total_in_length);
554                 return ISCSI_ERR_DATA_OFFSET;
555         }
556
557         if (rhdr->flags & ISCSI_FLAG_DATA_STATUS) {
558                 sc->result = (DID_OK << 16) | rhdr->cmd_status;
559                 conn->exp_statsn = be32_to_cpu(rhdr->statsn) + 1;
560                 if (rhdr->flags & (ISCSI_FLAG_DATA_UNDERFLOW |
561                                    ISCSI_FLAG_DATA_OVERFLOW)) {
562                         int res_count = be32_to_cpu(rhdr->residual_count);
563
564                         if (res_count > 0 &&
565                             (rhdr->flags & ISCSI_FLAG_CMD_OVERFLOW ||
566                              res_count <= total_in_length))
567                                 scsi_in(sc)->resid = res_count;
568                         else
569                                 sc->result = (DID_BAD_TARGET << 16) |
570                                         rhdr->cmd_status;
571                 }
572         }
573
574         conn->datain_pdus_cnt++;
575         return 0;
576 }
577
578 /**
579  * iscsi_solicit_data_init - initialize first Data-Out
580  * @conn: iscsi connection
581  * @ctask: scsi command task
582  * @r2t: R2T info
583  *
584  * Notes:
585  *      Initialize first Data-Out within this R2T sequence and finds
586  *      proper data_offset within this SCSI command.
587  *
588  *      This function is called with connection lock taken.
589  **/
590 static void
591 iscsi_solicit_data_init(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
592                         struct iscsi_r2t_info *r2t)
593 {
594         struct iscsi_data *hdr;
595
596         hdr = &r2t->dtask.hdr;
597         memset(hdr, 0, sizeof(struct iscsi_data));
598         hdr->ttt = r2t->ttt;
599         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
600         r2t->solicit_datasn++;
601         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
602         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
603         hdr->itt = ctask->hdr->itt;
604         hdr->exp_statsn = r2t->exp_statsn;
605         hdr->offset = cpu_to_be32(r2t->data_offset);
606         if (r2t->data_length > conn->max_xmit_dlength) {
607                 hton24(hdr->dlength, conn->max_xmit_dlength);
608                 r2t->data_count = conn->max_xmit_dlength;
609                 hdr->flags = 0;
610         } else {
611                 hton24(hdr->dlength, r2t->data_length);
612                 r2t->data_count = r2t->data_length;
613                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
614         }
615         conn->dataout_pdus_cnt++;
616
617         r2t->sent = 0;
618 }
619
620 /**
621  * iscsi_r2t_rsp - iSCSI R2T Response processing
622  * @conn: iscsi connection
623  * @ctask: scsi command task
624  **/
625 static int
626 iscsi_r2t_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
627 {
628         struct iscsi_r2t_info *r2t;
629         struct iscsi_session *session = conn->session;
630         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
631         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
632         struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
633         int r2tsn = be32_to_cpu(rhdr->r2tsn);
634         int rc;
635
636         if (tcp_conn->in.datalen) {
637                 iscsi_conn_printk(KERN_ERR, conn,
638                                   "invalid R2t with datalen %d\n",
639                                   tcp_conn->in.datalen);
640                 return ISCSI_ERR_DATALEN;
641         }
642
643         if (tcp_ctask->exp_datasn != r2tsn){
644                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
645                           __FUNCTION__, tcp_ctask->exp_datasn, r2tsn);
646                 return ISCSI_ERR_R2TSN;
647         }
648
649         /* fill-in new R2T associated with the task */
650         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
651
652         if (!ctask->sc || session->state != ISCSI_STATE_LOGGED_IN) {
653                 iscsi_conn_printk(KERN_INFO, conn,
654                                   "dropping R2T itt %d in recovery.\n",
655                                   ctask->itt);
656                 return 0;
657         }
658
659         rc = __kfifo_get(tcp_ctask->r2tpool.queue, (void*)&r2t, sizeof(void*));
660         BUG_ON(!rc);
661
662         r2t->exp_statsn = rhdr->statsn;
663         r2t->data_length = be32_to_cpu(rhdr->data_length);
664         if (r2t->data_length == 0) {
665                 iscsi_conn_printk(KERN_ERR, conn,
666                                   "invalid R2T with zero data len\n");
667                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
668                             sizeof(void*));
669                 return ISCSI_ERR_DATALEN;
670         }
671
672         if (r2t->data_length > session->max_burst)
673                 debug_scsi("invalid R2T with data len %u and max burst %u."
674                            "Attempting to execute request.\n",
675                             r2t->data_length, session->max_burst);
676
677         r2t->data_offset = be32_to_cpu(rhdr->data_offset);
678         if (r2t->data_offset + r2t->data_length > scsi_out(ctask->sc)->length) {
679                 iscsi_conn_printk(KERN_ERR, conn,
680                                   "invalid R2T with data len %u at offset %u "
681                                   "and total length %d\n", r2t->data_length,
682                                   r2t->data_offset, scsi_out(ctask->sc)->length);
683                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
684                             sizeof(void*));
685                 return ISCSI_ERR_DATALEN;
686         }
687
688         r2t->ttt = rhdr->ttt; /* no flip */
689         r2t->solicit_datasn = 0;
690
691         iscsi_solicit_data_init(conn, ctask, r2t);
692
693         tcp_ctask->exp_datasn = r2tsn + 1;
694         __kfifo_put(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*));
695         conn->r2t_pdus_cnt++;
696
697         iscsi_requeue_ctask(ctask);
698         return 0;
699 }
700
701 /*
702  * Handle incoming reply to DataIn command
703  */
704 static int
705 iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
706                           struct iscsi_segment *segment)
707 {
708         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
709         struct iscsi_hdr *hdr = tcp_conn->in.hdr;
710         int rc;
711
712         if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
713                 return ISCSI_ERR_DATA_DGST;
714
715         /* check for non-exceptional status */
716         if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
717                 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
718                 if (rc)
719                         return rc;
720         }
721
722         iscsi_tcp_hdr_recv_prep(tcp_conn);
723         return 0;
724 }
725
726 /**
727  * iscsi_tcp_hdr_dissect - process PDU header
728  * @conn: iSCSI connection
729  * @hdr: PDU header
730  *
731  * This function analyzes the header of the PDU received,
732  * and performs several sanity checks. If the PDU is accompanied
733  * by data, the receive buffer is set up to copy the incoming data
734  * to the correct location.
735  */
736 static int
737 iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
738 {
739         int rc = 0, opcode, ahslen;
740         struct iscsi_session *session = conn->session;
741         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
742         struct iscsi_cmd_task *ctask;
743
744         /* verify PDU length */
745         tcp_conn->in.datalen = ntoh24(hdr->dlength);
746         if (tcp_conn->in.datalen > conn->max_recv_dlength) {
747                 iscsi_conn_printk(KERN_ERR, conn,
748                                   "iscsi_tcp: datalen %d > %d\n",
749                                   tcp_conn->in.datalen, conn->max_recv_dlength);
750                 return ISCSI_ERR_DATALEN;
751         }
752
753         /* Additional header segments. So far, we don't
754          * process additional headers.
755          */
756         ahslen = hdr->hlength << 2;
757
758         opcode = hdr->opcode & ISCSI_OPCODE_MASK;
759         /* verify itt (itt encoding: age+cid+itt) */
760         rc = iscsi_verify_itt(conn, hdr->itt);
761         if (rc)
762                 return rc;
763
764         debug_tcp("opcode 0x%x ahslen %d datalen %d\n",
765                   opcode, ahslen, tcp_conn->in.datalen);
766
767         switch(opcode) {
768         case ISCSI_OP_SCSI_DATA_IN:
769                 ctask = iscsi_itt_to_ctask(conn, hdr->itt);
770                 if (!ctask)
771                         return ISCSI_ERR_BAD_ITT;
772
773                 spin_lock(&conn->session->lock);
774                 rc = iscsi_data_rsp(conn, ctask);
775                 spin_unlock(&conn->session->lock);
776                 if (rc)
777                         return rc;
778                 if (tcp_conn->in.datalen) {
779                         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
780                         struct hash_desc *rx_hash = NULL;
781                         struct scsi_data_buffer *sdb = scsi_in(ctask->sc);
782
783                         /*
784                          * Setup copy of Data-In into the Scsi_Cmnd
785                          * Scatterlist case:
786                          * We set up the iscsi_segment to point to the next
787                          * scatterlist entry to copy to. As we go along,
788                          * we move on to the next scatterlist entry and
789                          * update the digest per-entry.
790                          */
791                         if (conn->datadgst_en)
792                                 rx_hash = &tcp_conn->rx_hash;
793
794                         debug_tcp("iscsi_tcp_begin_data_in(%p, offset=%d, "
795                                   "datalen=%d)\n", tcp_conn,
796                                   tcp_ctask->data_offset,
797                                   tcp_conn->in.datalen);
798                         return iscsi_segment_seek_sg(&tcp_conn->in.segment,
799                                                      sdb->table.sgl,
800                                                      sdb->table.nents,
801                                                      tcp_ctask->data_offset,
802                                                      tcp_conn->in.datalen,
803                                                      iscsi_tcp_process_data_in,
804                                                      rx_hash);
805                 }
806                 /* fall through */
807         case ISCSI_OP_SCSI_CMD_RSP:
808                 if (tcp_conn->in.datalen) {
809                         iscsi_tcp_data_recv_prep(tcp_conn);
810                         return 0;
811                 }
812                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
813                 break;
814         case ISCSI_OP_R2T:
815                 ctask = iscsi_itt_to_ctask(conn, hdr->itt);
816                 if (!ctask)
817                         return ISCSI_ERR_BAD_ITT;
818
819                 if (ahslen)
820                         rc = ISCSI_ERR_AHSLEN;
821                 else if (ctask->sc->sc_data_direction == DMA_TO_DEVICE) {
822                         spin_lock(&session->lock);
823                         rc = iscsi_r2t_rsp(conn, ctask);
824                         spin_unlock(&session->lock);
825                 } else
826                         rc = ISCSI_ERR_PROTO;
827                 break;
828         case ISCSI_OP_LOGIN_RSP:
829         case ISCSI_OP_TEXT_RSP:
830         case ISCSI_OP_REJECT:
831         case ISCSI_OP_ASYNC_EVENT:
832                 /*
833                  * It is possible that we could get a PDU with a buffer larger
834                  * than 8K, but there are no targets that currently do this.
835                  * For now we fail until we find a vendor that needs it
836                  */
837                 if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
838                         iscsi_conn_printk(KERN_ERR, conn,
839                                           "iscsi_tcp: received buffer of "
840                                           "len %u but conn buffer is only %u "
841                                           "(opcode %0x)\n",
842                                           tcp_conn->in.datalen,
843                                           ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
844                         rc = ISCSI_ERR_PROTO;
845                         break;
846                 }
847
848                 /* If there's data coming in with the response,
849                  * receive it to the connection's buffer.
850                  */
851                 if (tcp_conn->in.datalen) {
852                         iscsi_tcp_data_recv_prep(tcp_conn);
853                         return 0;
854                 }
855         /* fall through */
856         case ISCSI_OP_LOGOUT_RSP:
857         case ISCSI_OP_NOOP_IN:
858         case ISCSI_OP_SCSI_TMFUNC_RSP:
859                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
860                 break;
861         default:
862                 rc = ISCSI_ERR_BAD_OPCODE;
863                 break;
864         }
865
866         if (rc == 0) {
867                 /* Anything that comes with data should have
868                  * been handled above. */
869                 if (tcp_conn->in.datalen)
870                         return ISCSI_ERR_PROTO;
871                 iscsi_tcp_hdr_recv_prep(tcp_conn);
872         }
873
874         return rc;
875 }
876
877 /**
878  * iscsi_tcp_hdr_recv_done - process PDU header
879  *
880  * This is the callback invoked when the PDU header has
881  * been received. If the header is followed by additional
882  * header segments, we go back for more data.
883  */
884 static int
885 iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
886                         struct iscsi_segment *segment)
887 {
888         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
889         struct iscsi_hdr *hdr;
890
891         /* Check if there are additional header segments
892          * *prior* to computing the digest, because we
893          * may need to go back to the caller for more.
894          */
895         hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
896         if (segment->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
897                 /* Bump the header length - the caller will
898                  * just loop around and get the AHS for us, and
899                  * call again. */
900                 unsigned int ahslen = hdr->hlength << 2;
901
902                 /* Make sure we don't overflow */
903                 if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
904                         return ISCSI_ERR_AHSLEN;
905
906                 segment->total_size += ahslen;
907                 segment->size += ahslen;
908                 return 0;
909         }
910
911         /* We're done processing the header. See if we're doing
912          * header digests; if so, set up the recv_digest buffer
913          * and go back for more. */
914         if (conn->hdrdgst_en) {
915                 if (segment->digest_len == 0) {
916                         iscsi_tcp_segment_splice_digest(segment,
917                                                         segment->recv_digest);
918                         return 0;
919                 }
920                 iscsi_tcp_dgst_header(&tcp_conn->rx_hash, hdr,
921                                       segment->total_copied - ISCSI_DIGEST_SIZE,
922                                       segment->digest);
923
924                 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
925                         return ISCSI_ERR_HDR_DGST;
926         }
927
928         tcp_conn->in.hdr = hdr;
929         return iscsi_tcp_hdr_dissect(conn, hdr);
930 }
931
932 /**
933  * iscsi_tcp_recv - TCP receive in sendfile fashion
934  * @rd_desc: read descriptor
935  * @skb: socket buffer
936  * @offset: offset in skb
937  * @len: skb->len - offset
938  **/
939 static int
940 iscsi_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
941                unsigned int offset, size_t len)
942 {
943         struct iscsi_conn *conn = rd_desc->arg.data;
944         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
945         struct iscsi_segment *segment = &tcp_conn->in.segment;
946         struct skb_seq_state seq;
947         unsigned int consumed = 0;
948         int rc = 0;
949
950         debug_tcp("in %d bytes\n", skb->len - offset);
951
952         if (unlikely(conn->suspend_rx)) {
953                 debug_tcp("conn %d Rx suspended!\n", conn->id);
954                 return 0;
955         }
956
957         skb_prepare_seq_read(skb, offset, skb->len, &seq);
958         while (1) {
959                 unsigned int avail;
960                 const u8 *ptr;
961
962                 avail = skb_seq_read(consumed, &ptr, &seq);
963                 if (avail == 0) {
964                         debug_tcp("no more data avail. Consumed %d\n",
965                                   consumed);
966                         break;
967                 }
968                 BUG_ON(segment->copied >= segment->size);
969
970                 debug_tcp("skb %p ptr=%p avail=%u\n", skb, ptr, avail);
971                 rc = iscsi_tcp_segment_recv(tcp_conn, segment, ptr, avail);
972                 BUG_ON(rc == 0);
973                 consumed += rc;
974
975                 if (segment->total_copied >= segment->total_size) {
976                         debug_tcp("segment done\n");
977                         rc = segment->done(tcp_conn, segment);
978                         if (rc != 0) {
979                                 skb_abort_seq_read(&seq);
980                                 goto error;
981                         }
982
983                         /* The done() functions sets up the
984                          * next segment. */
985                 }
986         }
987         skb_abort_seq_read(&seq);
988         conn->rxdata_octets += consumed;
989         return consumed;
990
991 error:
992         debug_tcp("Error receiving PDU, errno=%d\n", rc);
993         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
994         return 0;
995 }
996
997 static void
998 iscsi_tcp_data_ready(struct sock *sk, int flag)
999 {
1000         struct iscsi_conn *conn = sk->sk_user_data;
1001         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1002         read_descriptor_t rd_desc;
1003
1004         read_lock(&sk->sk_callback_lock);
1005
1006         /*
1007          * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
1008          * We set count to 1 because we want the network layer to
1009          * hand us all the skbs that are available. iscsi_tcp_recv
1010          * handled pdus that cross buffers or pdus that still need data.
1011          */
1012         rd_desc.arg.data = conn;
1013         rd_desc.count = 1;
1014         tcp_read_sock(sk, &rd_desc, iscsi_tcp_recv);
1015
1016         read_unlock(&sk->sk_callback_lock);
1017
1018         /* If we had to (atomically) map a highmem page,
1019          * unmap it now. */
1020         iscsi_tcp_segment_unmap(&tcp_conn->in.segment);
1021 }
1022
1023 static void
1024 iscsi_tcp_state_change(struct sock *sk)
1025 {
1026         struct iscsi_tcp_conn *tcp_conn;
1027         struct iscsi_conn *conn;
1028         struct iscsi_session *session;
1029         void (*old_state_change)(struct sock *);
1030
1031         read_lock(&sk->sk_callback_lock);
1032
1033         conn = (struct iscsi_conn*)sk->sk_user_data;
1034         session = conn->session;
1035
1036         if ((sk->sk_state == TCP_CLOSE_WAIT ||
1037              sk->sk_state == TCP_CLOSE) &&
1038             !atomic_read(&sk->sk_rmem_alloc)) {
1039                 debug_tcp("iscsi_tcp_state_change: TCP_CLOSE|TCP_CLOSE_WAIT\n");
1040                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1041         }
1042
1043         tcp_conn = conn->dd_data;
1044         old_state_change = tcp_conn->old_state_change;
1045
1046         read_unlock(&sk->sk_callback_lock);
1047
1048         old_state_change(sk);
1049 }
1050
1051 /**
1052  * iscsi_write_space - Called when more output buffer space is available
1053  * @sk: socket space is available for
1054  **/
1055 static void
1056 iscsi_write_space(struct sock *sk)
1057 {
1058         struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
1059         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1060
1061         tcp_conn->old_write_space(sk);
1062         debug_tcp("iscsi_write_space: cid %d\n", conn->id);
1063         scsi_queue_work(conn->session->host, &conn->xmitwork);
1064 }
1065
1066 static void
1067 iscsi_conn_set_callbacks(struct iscsi_conn *conn)
1068 {
1069         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1070         struct sock *sk = tcp_conn->sock->sk;
1071
1072         /* assign new callbacks */
1073         write_lock_bh(&sk->sk_callback_lock);
1074         sk->sk_user_data = conn;
1075         tcp_conn->old_data_ready = sk->sk_data_ready;
1076         tcp_conn->old_state_change = sk->sk_state_change;
1077         tcp_conn->old_write_space = sk->sk_write_space;
1078         sk->sk_data_ready = iscsi_tcp_data_ready;
1079         sk->sk_state_change = iscsi_tcp_state_change;
1080         sk->sk_write_space = iscsi_write_space;
1081         write_unlock_bh(&sk->sk_callback_lock);
1082 }
1083
1084 static void
1085 iscsi_conn_restore_callbacks(struct iscsi_tcp_conn *tcp_conn)
1086 {
1087         struct sock *sk = tcp_conn->sock->sk;
1088
1089         /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
1090         write_lock_bh(&sk->sk_callback_lock);
1091         sk->sk_user_data    = NULL;
1092         sk->sk_data_ready   = tcp_conn->old_data_ready;
1093         sk->sk_state_change = tcp_conn->old_state_change;
1094         sk->sk_write_space  = tcp_conn->old_write_space;
1095         sk->sk_no_check  = 0;
1096         write_unlock_bh(&sk->sk_callback_lock);
1097 }
1098
1099 /**
1100  * iscsi_xmit - TCP transmit
1101  **/
1102 static int
1103 iscsi_xmit(struct iscsi_conn *conn)
1104 {
1105         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1106         struct iscsi_segment *segment = &tcp_conn->out.segment;
1107         unsigned int consumed = 0;
1108         int rc = 0;
1109
1110         while (1) {
1111                 rc = iscsi_tcp_xmit_segment(tcp_conn, segment);
1112                 if (rc < 0)
1113                         goto error;
1114                 if (rc == 0)
1115                         break;
1116
1117                 consumed += rc;
1118
1119                 if (segment->total_copied >= segment->total_size) {
1120                         if (segment->done != NULL) {
1121                                 rc = segment->done(tcp_conn, segment);
1122                                 if (rc < 0)
1123                                         goto error;
1124                         }
1125                 }
1126         }
1127
1128         debug_tcp("xmit %d bytes\n", consumed);
1129
1130         conn->txdata_octets += consumed;
1131         return consumed;
1132
1133 error:
1134         /* Transmit error. We could initiate error recovery
1135          * here. */
1136         debug_tcp("Error sending PDU, errno=%d\n", rc);
1137         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1138         return rc;
1139 }
1140
1141 /**
1142  * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
1143  */
1144 static inline int
1145 iscsi_tcp_xmit_qlen(struct iscsi_conn *conn)
1146 {
1147         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1148         struct iscsi_segment *segment = &tcp_conn->out.segment;
1149
1150         return segment->total_copied - segment->total_size;
1151 }
1152
1153 static inline int
1154 iscsi_tcp_flush(struct iscsi_conn *conn)
1155 {
1156         int rc;
1157
1158         while (iscsi_tcp_xmit_qlen(conn)) {
1159                 rc = iscsi_xmit(conn);
1160                 if (rc == 0)
1161                         return -EAGAIN;
1162                 if (rc < 0)
1163                         return rc;
1164         }
1165
1166         return 0;
1167 }
1168
1169 /*
1170  * This is called when we're done sending the header.
1171  * Simply copy the data_segment to the send segment, and return.
1172  */
1173 static int
1174 iscsi_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
1175                         struct iscsi_segment *segment)
1176 {
1177         tcp_conn->out.segment = tcp_conn->out.data_segment;
1178         debug_tcp("Header done. Next segment size %u total_size %u\n",
1179                   tcp_conn->out.segment.size, tcp_conn->out.segment.total_size);
1180         return 0;
1181 }
1182
1183 static void
1184 iscsi_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr, size_t hdrlen)
1185 {
1186         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1187
1188         debug_tcp("%s(%p%s)\n", __FUNCTION__, tcp_conn,
1189                         conn->hdrdgst_en? ", digest enabled" : "");
1190
1191         /* Clear the data segment - needs to be filled in by the
1192          * caller using iscsi_tcp_send_data_prep() */
1193         memset(&tcp_conn->out.data_segment, 0, sizeof(struct iscsi_segment));
1194
1195         /* If header digest is enabled, compute the CRC and
1196          * place the digest into the same buffer. We make
1197          * sure that both iscsi_tcp_ctask and mtask have
1198          * sufficient room.
1199          */
1200         if (conn->hdrdgst_en) {
1201                 iscsi_tcp_dgst_header(&tcp_conn->tx_hash, hdr, hdrlen,
1202                                       hdr + hdrlen);
1203                 hdrlen += ISCSI_DIGEST_SIZE;
1204         }
1205
1206         /* Remember header pointer for later, when we need
1207          * to decide whether there's a payload to go along
1208          * with the header. */
1209         tcp_conn->out.hdr = hdr;
1210
1211         iscsi_segment_init_linear(&tcp_conn->out.segment, hdr, hdrlen,
1212                                 iscsi_tcp_send_hdr_done, NULL);
1213 }
1214
1215 /*
1216  * Prepare the send buffer for the payload data.
1217  * Padding and checksumming will all be taken care
1218  * of by the iscsi_segment routines.
1219  */
1220 static int
1221 iscsi_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
1222                          unsigned int count, unsigned int offset,
1223                          unsigned int len)
1224 {
1225         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1226         struct hash_desc *tx_hash = NULL;
1227         unsigned int hdr_spec_len;
1228
1229         debug_tcp("%s(%p, offset=%d, datalen=%d%s)\n", __FUNCTION__,
1230                         tcp_conn, offset, len,
1231                         conn->datadgst_en? ", digest enabled" : "");
1232
1233         /* Make sure the datalen matches what the caller
1234            said he would send. */
1235         hdr_spec_len = ntoh24(tcp_conn->out.hdr->dlength);
1236         WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
1237
1238         if (conn->datadgst_en)
1239                 tx_hash = &tcp_conn->tx_hash;
1240
1241         return iscsi_segment_seek_sg(&tcp_conn->out.data_segment,
1242                                    sg, count, offset, len,
1243                                    NULL, tx_hash);
1244 }
1245
1246 static void
1247 iscsi_tcp_send_linear_data_prepare(struct iscsi_conn *conn, void *data,
1248                                    size_t len)
1249 {
1250         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1251         struct hash_desc *tx_hash = NULL;
1252         unsigned int hdr_spec_len;
1253
1254         debug_tcp("%s(%p, datalen=%d%s)\n", __FUNCTION__, tcp_conn, len,
1255                   conn->datadgst_en? ", digest enabled" : "");
1256
1257         /* Make sure the datalen matches what the caller
1258            said he would send. */
1259         hdr_spec_len = ntoh24(tcp_conn->out.hdr->dlength);
1260         WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
1261
1262         if (conn->datadgst_en)
1263                 tx_hash = &tcp_conn->tx_hash;
1264
1265         iscsi_segment_init_linear(&tcp_conn->out.data_segment,
1266                                 data, len, NULL, tx_hash);
1267 }
1268
1269 /**
1270  * iscsi_solicit_data_cont - initialize next Data-Out
1271  * @conn: iscsi connection
1272  * @ctask: scsi command task
1273  * @r2t: R2T info
1274  * @left: bytes left to transfer
1275  *
1276  * Notes:
1277  *      Initialize next Data-Out within this R2T sequence and continue
1278  *      to process next Scatter-Gather element(if any) of this SCSI command.
1279  *
1280  *      Called under connection lock.
1281  **/
1282 static int
1283 iscsi_solicit_data_cont(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
1284                         struct iscsi_r2t_info *r2t)
1285 {
1286         struct iscsi_data *hdr;
1287         int new_offset, left;
1288
1289         BUG_ON(r2t->data_length - r2t->sent < 0);
1290         left = r2t->data_length - r2t->sent;
1291         if (left == 0)
1292                 return 0;
1293
1294         hdr = &r2t->dtask.hdr;
1295         memset(hdr, 0, sizeof(struct iscsi_data));
1296         hdr->ttt = r2t->ttt;
1297         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
1298         r2t->solicit_datasn++;
1299         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
1300         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
1301         hdr->itt = ctask->hdr->itt;
1302         hdr->exp_statsn = r2t->exp_statsn;
1303         new_offset = r2t->data_offset + r2t->sent;
1304         hdr->offset = cpu_to_be32(new_offset);
1305         if (left > conn->max_xmit_dlength) {
1306                 hton24(hdr->dlength, conn->max_xmit_dlength);
1307                 r2t->data_count = conn->max_xmit_dlength;
1308         } else {
1309                 hton24(hdr->dlength, left);
1310                 r2t->data_count = left;
1311                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
1312         }
1313
1314         conn->dataout_pdus_cnt++;
1315         return 1;
1316 }
1317
1318 /**
1319  * iscsi_tcp_ctask - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
1320  * @conn: iscsi connection
1321  * @ctask: scsi command task
1322  * @sc: scsi command
1323  **/
1324 static int
1325 iscsi_tcp_ctask_init(struct iscsi_cmd_task *ctask)
1326 {
1327         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1328         struct iscsi_conn *conn = ctask->conn;
1329         struct scsi_cmnd *sc = ctask->sc;
1330         int err;
1331
1332         BUG_ON(__kfifo_len(tcp_ctask->r2tqueue));
1333         tcp_ctask->sent = 0;
1334         tcp_ctask->exp_datasn = 0;
1335
1336         /* Prepare PDU, optionally w/ immediate data */
1337         debug_scsi("ctask deq [cid %d itt 0x%x imm %d unsol %d]\n",
1338                     conn->id, ctask->itt, ctask->imm_count,
1339                     ctask->unsol_count);
1340         iscsi_tcp_send_hdr_prep(conn, ctask->hdr, ctask->hdr_len);
1341
1342         if (!ctask->imm_count)
1343                 return 0;
1344
1345         /* If we have immediate data, attach a payload */
1346         err = iscsi_tcp_send_data_prep(conn, scsi_out(sc)->table.sgl,
1347                                        scsi_out(sc)->table.nents,
1348                                        0, ctask->imm_count);
1349         if (err)
1350                 return err;
1351         tcp_ctask->sent += ctask->imm_count;
1352         ctask->imm_count = 0;
1353         return 0;
1354 }
1355
1356 /**
1357  * iscsi_tcp_mtask_xmit - xmit management(immediate) task
1358  * @conn: iscsi connection
1359  * @mtask: task management task
1360  *
1361  * Notes:
1362  *      The function can return -EAGAIN in which case caller must
1363  *      call it again later, or recover. '0' return code means successful
1364  *      xmit.
1365  **/
1366 static int
1367 iscsi_tcp_mtask_xmit(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1368 {
1369         int rc;
1370
1371         /* Flush any pending data first. */
1372         rc = iscsi_tcp_flush(conn);
1373         if (rc < 0)
1374                 return rc;
1375
1376         if (mtask->hdr->itt == RESERVED_ITT) {
1377                 struct iscsi_session *session = conn->session;
1378
1379                 spin_lock_bh(&session->lock);
1380                 iscsi_free_mgmt_task(conn, mtask);
1381                 spin_unlock_bh(&session->lock);
1382         }
1383
1384         return 0;
1385 }
1386
1387 /*
1388  * iscsi_tcp_ctask_xmit - xmit normal PDU task
1389  * @conn: iscsi connection
1390  * @ctask: iscsi command task
1391  *
1392  * We're expected to return 0 when everything was transmitted succesfully,
1393  * -EAGAIN if there's still data in the queue, or != 0 for any other kind
1394  * of error.
1395  */
1396 static int
1397 iscsi_tcp_ctask_xmit(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1398 {
1399         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1400         struct scsi_cmnd *sc = ctask->sc;
1401         struct scsi_data_buffer *sdb = scsi_out(sc);
1402         int rc = 0;
1403
1404 flush:
1405         /* Flush any pending data first. */
1406         rc = iscsi_tcp_flush(conn);
1407         if (rc < 0)
1408                 return rc;
1409
1410         /* Are we done already? */
1411         if (sc->sc_data_direction != DMA_TO_DEVICE)
1412                 return 0;
1413
1414         if (ctask->unsol_count != 0) {
1415                 struct iscsi_data *hdr = &tcp_ctask->unsol_dtask.hdr;
1416
1417                 /* Prepare a header for the unsolicited PDU.
1418                  * The amount of data we want to send will be
1419                  * in ctask->data_count.
1420                  * FIXME: return the data count instead.
1421                  */
1422                 iscsi_prep_unsolicit_data_pdu(ctask, hdr);
1423
1424                 debug_tcp("unsol dout [itt 0x%x doff %d dlen %d]\n",
1425                                 ctask->itt, tcp_ctask->sent, ctask->data_count);
1426
1427                 iscsi_tcp_send_hdr_prep(conn, hdr, sizeof(*hdr));
1428                 rc = iscsi_tcp_send_data_prep(conn, sdb->table.sgl,
1429                                               sdb->table.nents, tcp_ctask->sent,
1430                                               ctask->data_count);
1431                 if (rc)
1432                         goto fail;
1433                 tcp_ctask->sent += ctask->data_count;
1434                 ctask->unsol_count -= ctask->data_count;
1435                 goto flush;
1436         } else {
1437                 struct iscsi_session *session = conn->session;
1438                 struct iscsi_r2t_info *r2t;
1439
1440                 /* All unsolicited PDUs sent. Check for solicited PDUs.
1441                  */
1442                 spin_lock_bh(&session->lock);
1443                 r2t = tcp_ctask->r2t;
1444                 if (r2t != NULL) {
1445                         /* Continue with this R2T? */
1446                         if (!iscsi_solicit_data_cont(conn, ctask, r2t)) {
1447                                 debug_scsi("  done with r2t %p\n", r2t);
1448
1449                                 __kfifo_put(tcp_ctask->r2tpool.queue,
1450                                             (void*)&r2t, sizeof(void*));
1451                                 tcp_ctask->r2t = r2t = NULL;
1452                         }
1453                 }
1454
1455                 if (r2t == NULL) {
1456                         __kfifo_get(tcp_ctask->r2tqueue, (void*)&tcp_ctask->r2t,
1457                                     sizeof(void*));
1458                         r2t = tcp_ctask->r2t;
1459                 }
1460                 spin_unlock_bh(&session->lock);
1461
1462                 /* Waiting for more R2Ts to arrive. */
1463                 if (r2t == NULL) {
1464                         debug_tcp("no R2Ts yet\n");
1465                         return 0;
1466                 }
1467
1468                 debug_scsi("sol dout %p [dsn %d itt 0x%x doff %d dlen %d]\n",
1469                         r2t, r2t->solicit_datasn - 1, ctask->itt,
1470                         r2t->data_offset + r2t->sent, r2t->data_count);
1471
1472                 iscsi_tcp_send_hdr_prep(conn, &r2t->dtask.hdr,
1473                                         sizeof(struct iscsi_hdr));
1474
1475                 rc = iscsi_tcp_send_data_prep(conn, sdb->table.sgl,
1476                                               sdb->table.nents,
1477                                               r2t->data_offset + r2t->sent,
1478                                               r2t->data_count);
1479                 if (rc)
1480                         goto fail;
1481                 tcp_ctask->sent += r2t->data_count;
1482                 r2t->sent += r2t->data_count;
1483                 goto flush;
1484         }
1485         return 0;
1486 fail:
1487         iscsi_conn_failure(conn, rc);
1488         return -EIO;
1489 }
1490
1491 static struct iscsi_cls_conn *
1492 iscsi_tcp_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
1493 {
1494         struct iscsi_conn *conn;
1495         struct iscsi_cls_conn *cls_conn;
1496         struct iscsi_tcp_conn *tcp_conn;
1497
1498         cls_conn = iscsi_conn_setup(cls_session, sizeof(*tcp_conn), conn_idx);
1499         if (!cls_conn)
1500                 return NULL;
1501         conn = cls_conn->dd_data;
1502         /*
1503          * due to strange issues with iser these are not set
1504          * in iscsi_conn_setup
1505          */
1506         conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1507
1508         tcp_conn = conn->dd_data;
1509         tcp_conn->iscsi_conn = conn;
1510
1511         tcp_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1512                                                   CRYPTO_ALG_ASYNC);
1513         tcp_conn->tx_hash.flags = 0;
1514         if (IS_ERR(tcp_conn->tx_hash.tfm))
1515                 goto free_conn;
1516
1517         tcp_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1518                                                   CRYPTO_ALG_ASYNC);
1519         tcp_conn->rx_hash.flags = 0;
1520         if (IS_ERR(tcp_conn->rx_hash.tfm))
1521                 goto free_tx_tfm;
1522
1523         return cls_conn;
1524
1525 free_tx_tfm:
1526         crypto_free_hash(tcp_conn->tx_hash.tfm);
1527 free_conn:
1528         iscsi_conn_printk(KERN_ERR, conn,
1529                           "Could not create connection due to crc32c "
1530                           "loading error. Make sure the crc32c "
1531                           "module is built as a module or into the "
1532                           "kernel\n");
1533         iscsi_conn_teardown(cls_conn);
1534         return NULL;
1535 }
1536
1537 static void
1538 iscsi_tcp_release_conn(struct iscsi_conn *conn)
1539 {
1540         struct iscsi_session *session = conn->session;
1541         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1542         struct socket *sock = tcp_conn->sock;
1543
1544         if (!sock)
1545                 return;
1546
1547         sock_hold(sock->sk);
1548         iscsi_conn_restore_callbacks(tcp_conn);
1549         sock_put(sock->sk);
1550
1551         spin_lock_bh(&session->lock);
1552         tcp_conn->sock = NULL;
1553         conn->recv_lock = NULL;
1554         spin_unlock_bh(&session->lock);
1555         sockfd_put(sock);
1556 }
1557
1558 static void
1559 iscsi_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
1560 {
1561         struct iscsi_conn *conn = cls_conn->dd_data;
1562         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1563
1564         iscsi_tcp_release_conn(conn);
1565
1566         if (tcp_conn->tx_hash.tfm)
1567                 crypto_free_hash(tcp_conn->tx_hash.tfm);
1568         if (tcp_conn->rx_hash.tfm)
1569                 crypto_free_hash(tcp_conn->rx_hash.tfm);
1570
1571         iscsi_conn_teardown(cls_conn);
1572 }
1573
1574 static void
1575 iscsi_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
1576 {
1577         struct iscsi_conn *conn = cls_conn->dd_data;
1578
1579         iscsi_conn_stop(cls_conn, flag);
1580         iscsi_tcp_release_conn(conn);
1581 }
1582
1583 static int iscsi_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
1584                               char *buf, int *port,
1585                               int (*getname)(struct socket *, struct sockaddr *,
1586                                         int *addrlen))
1587 {
1588         struct sockaddr_storage *addr;
1589         struct sockaddr_in6 *sin6;
1590         struct sockaddr_in *sin;
1591         int rc = 0, len;
1592
1593         addr = kmalloc(sizeof(*addr), GFP_KERNEL);
1594         if (!addr)
1595                 return -ENOMEM;
1596
1597         if (getname(sock, (struct sockaddr *) addr, &len)) {
1598                 rc = -ENODEV;
1599                 goto free_addr;
1600         }
1601
1602         switch (addr->ss_family) {
1603         case AF_INET:
1604                 sin = (struct sockaddr_in *)addr;
1605                 spin_lock_bh(&conn->session->lock);
1606                 sprintf(buf, NIPQUAD_FMT, NIPQUAD(sin->sin_addr.s_addr));
1607                 *port = be16_to_cpu(sin->sin_port);
1608                 spin_unlock_bh(&conn->session->lock);
1609                 break;
1610         case AF_INET6:
1611                 sin6 = (struct sockaddr_in6 *)addr;
1612                 spin_lock_bh(&conn->session->lock);
1613                 sprintf(buf, NIP6_FMT, NIP6(sin6->sin6_addr));
1614                 *port = be16_to_cpu(sin6->sin6_port);
1615                 spin_unlock_bh(&conn->session->lock);
1616                 break;
1617         }
1618 free_addr:
1619         kfree(addr);
1620         return rc;
1621 }
1622
1623 static int
1624 iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
1625                     struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
1626                     int is_leading)
1627 {
1628         struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
1629         struct iscsi_host *ihost = shost_priv(shost);
1630         struct iscsi_conn *conn = cls_conn->dd_data;
1631         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1632         struct sock *sk;
1633         struct socket *sock;
1634         int err;
1635
1636         /* lookup for existing socket */
1637         sock = sockfd_lookup((int)transport_eph, &err);
1638         if (!sock) {
1639                 iscsi_conn_printk(KERN_ERR, conn,
1640                                   "sockfd_lookup failed %d\n", err);
1641                 return -EEXIST;
1642         }
1643         /*
1644          * copy these values now because if we drop the session
1645          * userspace may still want to query the values since we will
1646          * be using them for the reconnect
1647          */
1648         err = iscsi_tcp_get_addr(conn, sock, conn->portal_address,
1649                                  &conn->portal_port, kernel_getpeername);
1650         if (err)
1651                 goto free_socket;
1652
1653         err = iscsi_tcp_get_addr(conn, sock, ihost->local_address,
1654                                 &ihost->local_port, kernel_getsockname);
1655         if (err)
1656                 goto free_socket;
1657
1658         err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
1659         if (err)
1660                 goto free_socket;
1661
1662         /* bind iSCSI connection and socket */
1663         tcp_conn->sock = sock;
1664
1665         /* setup Socket parameters */
1666         sk = sock->sk;
1667         sk->sk_reuse = 1;
1668         sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
1669         sk->sk_allocation = GFP_ATOMIC;
1670
1671         /* FIXME: disable Nagle's algorithm */
1672
1673         /*
1674          * Intercept TCP callbacks for sendfile like receive
1675          * processing.
1676          */
1677         conn->recv_lock = &sk->sk_callback_lock;
1678         iscsi_conn_set_callbacks(conn);
1679         tcp_conn->sendpage = tcp_conn->sock->ops->sendpage;
1680         /*
1681          * set receive state machine into initial state
1682          */
1683         iscsi_tcp_hdr_recv_prep(tcp_conn);
1684         return 0;
1685
1686 free_socket:
1687         sockfd_put(sock);
1688         return err;
1689 }
1690
1691 /* called with host lock */
1692 static void
1693 iscsi_tcp_mtask_init(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1694 {
1695         debug_scsi("mtask deq [cid %d itt 0x%x]\n", conn->id, mtask->itt);
1696
1697         /* Prepare PDU, optionally w/ immediate data */
1698         iscsi_tcp_send_hdr_prep(conn, mtask->hdr, sizeof(*mtask->hdr));
1699
1700         /* If we have immediate data, attach a payload */
1701         if (mtask->data_count)
1702                 iscsi_tcp_send_linear_data_prepare(conn, mtask->data,
1703                                                    mtask->data_count);
1704 }
1705
1706 static int
1707 iscsi_r2tpool_alloc(struct iscsi_session *session)
1708 {
1709         int i;
1710         int cmd_i;
1711
1712         /*
1713          * initialize per-task: R2T pool and xmit queue
1714          */
1715         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1716                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
1717                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1718
1719                 /*
1720                  * pre-allocated x4 as much r2ts to handle race when
1721                  * target acks DataOut faster than we data_xmit() queues
1722                  * could replenish r2tqueue.
1723                  */
1724
1725                 /* R2T pool */
1726                 if (iscsi_pool_init(&tcp_ctask->r2tpool, session->max_r2t * 4, NULL,
1727                                     sizeof(struct iscsi_r2t_info))) {
1728                         goto r2t_alloc_fail;
1729                 }
1730
1731                 /* R2T xmit queue */
1732                 tcp_ctask->r2tqueue = kfifo_alloc(
1733                       session->max_r2t * 4 * sizeof(void*), GFP_KERNEL, NULL);
1734                 if (tcp_ctask->r2tqueue == ERR_PTR(-ENOMEM)) {
1735                         iscsi_pool_free(&tcp_ctask->r2tpool);
1736                         goto r2t_alloc_fail;
1737                 }
1738         }
1739
1740         return 0;
1741
1742 r2t_alloc_fail:
1743         for (i = 0; i < cmd_i; i++) {
1744                 struct iscsi_cmd_task *ctask = session->cmds[i];
1745                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1746
1747                 kfifo_free(tcp_ctask->r2tqueue);
1748                 iscsi_pool_free(&tcp_ctask->r2tpool);
1749         }
1750         return -ENOMEM;
1751 }
1752
1753 static void
1754 iscsi_r2tpool_free(struct iscsi_session *session)
1755 {
1756         int i;
1757
1758         for (i = 0; i < session->cmds_max; i++) {
1759                 struct iscsi_cmd_task *ctask = session->cmds[i];
1760                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1761
1762                 kfifo_free(tcp_ctask->r2tqueue);
1763                 iscsi_pool_free(&tcp_ctask->r2tpool);
1764         }
1765 }
1766
1767 static int
1768 iscsi_conn_set_param(struct iscsi_cls_conn *cls_conn, enum iscsi_param param,
1769                      char *buf, int buflen)
1770 {
1771         struct iscsi_conn *conn = cls_conn->dd_data;
1772         struct iscsi_session *session = conn->session;
1773         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1774         int value;
1775
1776         switch(param) {
1777         case ISCSI_PARAM_HDRDGST_EN:
1778                 iscsi_set_param(cls_conn, param, buf, buflen);
1779                 break;
1780         case ISCSI_PARAM_DATADGST_EN:
1781                 iscsi_set_param(cls_conn, param, buf, buflen);
1782                 tcp_conn->sendpage = conn->datadgst_en ?
1783                         sock_no_sendpage : tcp_conn->sock->ops->sendpage;
1784                 break;
1785         case ISCSI_PARAM_MAX_R2T:
1786                 sscanf(buf, "%d", &value);
1787                 if (value <= 0 || !is_power_of_2(value))
1788                         return -EINVAL;
1789                 if (session->max_r2t == value)
1790                         break;
1791                 iscsi_r2tpool_free(session);
1792                 iscsi_set_param(cls_conn, param, buf, buflen);
1793                 if (iscsi_r2tpool_alloc(session))
1794                         return -ENOMEM;
1795                 break;
1796         default:
1797                 return iscsi_set_param(cls_conn, param, buf, buflen);
1798         }
1799
1800         return 0;
1801 }
1802
1803 static int
1804 iscsi_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
1805                          enum iscsi_param param, char *buf)
1806 {
1807         struct iscsi_conn *conn = cls_conn->dd_data;
1808         int len;
1809
1810         switch(param) {
1811         case ISCSI_PARAM_CONN_PORT:
1812                 spin_lock_bh(&conn->session->lock);
1813                 len = sprintf(buf, "%hu\n", conn->portal_port);
1814                 spin_unlock_bh(&conn->session->lock);
1815                 break;
1816         case ISCSI_PARAM_CONN_ADDRESS:
1817                 spin_lock_bh(&conn->session->lock);
1818                 len = sprintf(buf, "%s\n", conn->portal_address);
1819                 spin_unlock_bh(&conn->session->lock);
1820                 break;
1821         default:
1822                 return iscsi_conn_get_param(cls_conn, param, buf);
1823         }
1824
1825         return len;
1826 }
1827
1828 static void
1829 iscsi_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
1830 {
1831         struct iscsi_conn *conn = cls_conn->dd_data;
1832         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1833
1834         stats->txdata_octets = conn->txdata_octets;
1835         stats->rxdata_octets = conn->rxdata_octets;
1836         stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
1837         stats->dataout_pdus = conn->dataout_pdus_cnt;
1838         stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
1839         stats->datain_pdus = conn->datain_pdus_cnt;
1840         stats->r2t_pdus = conn->r2t_pdus_cnt;
1841         stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
1842         stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
1843         stats->custom_length = 3;
1844         strcpy(stats->custom[0].desc, "tx_sendpage_failures");
1845         stats->custom[0].value = tcp_conn->sendpage_failures_cnt;
1846         strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
1847         stats->custom[1].value = tcp_conn->discontiguous_hdr_cnt;
1848         strcpy(stats->custom[2].desc, "eh_abort_cnt");
1849         stats->custom[2].value = conn->eh_abort_cnt;
1850 }
1851
1852 static struct iscsi_cls_session *
1853 iscsi_tcp_session_create(struct Scsi_Host *shost, uint16_t cmds_max,
1854                          uint16_t qdepth, uint32_t initial_cmdsn,
1855                          uint32_t *hostno)
1856 {
1857         struct iscsi_cls_session *cls_session;
1858         struct iscsi_session *session;
1859         int cmd_i;
1860
1861         if (shost) {
1862                 printk(KERN_ERR "iscsi_tcp: invalid shost %d.\n",
1863                        shost->host_no);
1864                 return NULL;
1865         }
1866
1867         shost = iscsi_host_alloc(&iscsi_sht, 0, qdepth);
1868         if (!shost)
1869                 return NULL;
1870         shost->transportt = iscsi_tcp_scsi_transport;
1871         shost->max_lun = iscsi_max_lun;
1872         shost->max_id = 0;
1873         shost->max_channel = 0;
1874         shost->max_cmd_len = 16;
1875         shost->can_queue = cmds_max;
1876
1877         if (iscsi_host_add(shost, NULL))
1878                 goto free_host;
1879         *hostno = shost->host_no;
1880
1881         cls_session = iscsi_session_setup(&iscsi_tcp_transport, shost, cmds_max,
1882                                           sizeof(struct iscsi_tcp_cmd_task),
1883                                           sizeof(struct iscsi_tcp_mgmt_task),
1884                                           initial_cmdsn);
1885         if (!cls_session)
1886                 goto remove_host;
1887         session = cls_session->dd_data;
1888
1889         shost->can_queue = session->cmds_max;
1890         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1891                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
1892                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1893
1894                 ctask->hdr = &tcp_ctask->hdr.cmd_hdr;
1895                 ctask->hdr_max = sizeof(tcp_ctask->hdr) - ISCSI_DIGEST_SIZE;
1896         }
1897
1898         for (cmd_i = 0; cmd_i < session->mgmtpool_max; cmd_i++) {
1899                 struct iscsi_mgmt_task *mtask = session->mgmt_cmds[cmd_i];
1900                 struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
1901
1902                 mtask->hdr = (struct iscsi_hdr *) &tcp_mtask->hdr;
1903         }
1904
1905         if (iscsi_r2tpool_alloc(session))
1906                 goto remove_session;
1907         return cls_session;
1908
1909 remove_session:
1910         iscsi_session_teardown(cls_session);
1911 remove_host:
1912         iscsi_host_remove(shost);
1913 free_host:
1914         iscsi_host_free(shost);
1915         return NULL;
1916 }
1917
1918 static void iscsi_tcp_session_destroy(struct iscsi_cls_session *cls_session)
1919 {
1920         struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
1921
1922         iscsi_r2tpool_free(cls_session->dd_data);
1923
1924         iscsi_host_remove(shost);
1925         iscsi_host_free(shost);
1926 }
1927
1928 static int iscsi_tcp_slave_configure(struct scsi_device *sdev)
1929 {
1930         blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
1931         blk_queue_dma_alignment(sdev->request_queue, 0);
1932         return 0;
1933 }
1934
1935 static struct scsi_host_template iscsi_sht = {
1936         .module                 = THIS_MODULE,
1937         .name                   = "iSCSI Initiator over TCP/IP",
1938         .queuecommand           = iscsi_queuecommand,
1939         .change_queue_depth     = iscsi_change_queue_depth,
1940         .can_queue              = ISCSI_DEF_XMIT_CMDS_MAX - 1,
1941         .sg_tablesize           = 4096,
1942         .max_sectors            = 0xFFFF,
1943         .cmd_per_lun            = ISCSI_DEF_CMD_PER_LUN,
1944         .eh_abort_handler       = iscsi_eh_abort,
1945         .eh_device_reset_handler= iscsi_eh_device_reset,
1946         .eh_host_reset_handler  = iscsi_eh_host_reset,
1947         .use_clustering         = DISABLE_CLUSTERING,
1948         .slave_configure        = iscsi_tcp_slave_configure,
1949         .proc_name              = "iscsi_tcp",
1950         .this_id                = -1,
1951 };
1952
1953 static struct iscsi_transport iscsi_tcp_transport = {
1954         .owner                  = THIS_MODULE,
1955         .name                   = "tcp",
1956         .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
1957                                   | CAP_DATADGST,
1958         .param_mask             = ISCSI_MAX_RECV_DLENGTH |
1959                                   ISCSI_MAX_XMIT_DLENGTH |
1960                                   ISCSI_HDRDGST_EN |
1961                                   ISCSI_DATADGST_EN |
1962                                   ISCSI_INITIAL_R2T_EN |
1963                                   ISCSI_MAX_R2T |
1964                                   ISCSI_IMM_DATA_EN |
1965                                   ISCSI_FIRST_BURST |
1966                                   ISCSI_MAX_BURST |
1967                                   ISCSI_PDU_INORDER_EN |
1968                                   ISCSI_DATASEQ_INORDER_EN |
1969                                   ISCSI_ERL |
1970                                   ISCSI_CONN_PORT |
1971                                   ISCSI_CONN_ADDRESS |
1972                                   ISCSI_EXP_STATSN |
1973                                   ISCSI_PERSISTENT_PORT |
1974                                   ISCSI_PERSISTENT_ADDRESS |
1975                                   ISCSI_TARGET_NAME | ISCSI_TPGT |
1976                                   ISCSI_USERNAME | ISCSI_PASSWORD |
1977                                   ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
1978                                   ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
1979                                   ISCSI_LU_RESET_TMO |
1980                                   ISCSI_PING_TMO | ISCSI_RECV_TMO,
1981         .host_param_mask        = ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
1982                                   ISCSI_HOST_INITIATOR_NAME |
1983                                   ISCSI_HOST_NETDEV_NAME,
1984         /* session management */
1985         .create_session         = iscsi_tcp_session_create,
1986         .destroy_session        = iscsi_tcp_session_destroy,
1987         /* connection management */
1988         .create_conn            = iscsi_tcp_conn_create,
1989         .bind_conn              = iscsi_tcp_conn_bind,
1990         .destroy_conn           = iscsi_tcp_conn_destroy,
1991         .set_param              = iscsi_conn_set_param,
1992         .get_conn_param         = iscsi_tcp_conn_get_param,
1993         .get_session_param      = iscsi_session_get_param,
1994         .start_conn             = iscsi_conn_start,
1995         .stop_conn              = iscsi_tcp_conn_stop,
1996         /* iscsi host params */
1997         .get_host_param         = iscsi_host_get_param,
1998         .set_host_param         = iscsi_host_set_param,
1999         /* IO */
2000         .send_pdu               = iscsi_conn_send_pdu,
2001         .get_stats              = iscsi_conn_get_stats,
2002         .init_cmd_task          = iscsi_tcp_ctask_init,
2003         .init_mgmt_task         = iscsi_tcp_mtask_init,
2004         .xmit_cmd_task          = iscsi_tcp_ctask_xmit,
2005         .xmit_mgmt_task         = iscsi_tcp_mtask_xmit,
2006         .cleanup_cmd_task       = iscsi_tcp_cleanup_ctask,
2007         /* recovery */
2008         .session_recovery_timedout = iscsi_session_recovery_timedout,
2009 };
2010
2011 static int __init
2012 iscsi_tcp_init(void)
2013 {
2014         if (iscsi_max_lun < 1) {
2015                 printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
2016                        iscsi_max_lun);
2017                 return -EINVAL;
2018         }
2019
2020         iscsi_tcp_scsi_transport = iscsi_register_transport(
2021                                                         &iscsi_tcp_transport);
2022         if (!iscsi_tcp_scsi_transport)
2023                 return -ENODEV;
2024
2025         return 0;
2026 }
2027
2028 static void __exit
2029 iscsi_tcp_exit(void)
2030 {
2031         iscsi_unregister_transport(&iscsi_tcp_transport);
2032 }
2033
2034 module_init(iscsi_tcp_init);
2035 module_exit(iscsi_tcp_exit);