[SCSI] iscsi_tcp: rewrite recv path
[linux-2.6.git] / drivers / scsi / iscsi_tcp.c
1 /*
2  * iSCSI Initiator over TCP/IP Data-Path
3  *
4  * Copyright (C) 2004 Dmitry Yusupov
5  * Copyright (C) 2004 Alex Aizman
6  * Copyright (C) 2005 - 2006 Mike Christie
7  * Copyright (C) 2006 Red Hat, Inc.  All rights reserved.
8  * maintained by open-iscsi@googlegroups.com
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published
12  * by the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful, but
16  * WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18  * General Public License for more details.
19  *
20  * See the file COPYING included with this distribution for more details.
21  *
22  * Credits:
23  *      Christoph Hellwig
24  *      FUJITA Tomonori
25  *      Arne Redlich
26  *      Zhenyu Wang
27  */
28
29 #include <linux/types.h>
30 #include <linux/list.h>
31 #include <linux/inet.h>
32 #include <linux/file.h>
33 #include <linux/blkdev.h>
34 #include <linux/crypto.h>
35 #include <linux/delay.h>
36 #include <linux/kfifo.h>
37 #include <linux/scatterlist.h>
38 #include <net/tcp.h>
39 #include <scsi/scsi_cmnd.h>
40 #include <scsi/scsi_device.h>
41 #include <scsi/scsi_host.h>
42 #include <scsi/scsi.h>
43 #include <scsi/scsi_transport_iscsi.h>
44
45 #include "iscsi_tcp.h"
46
47 MODULE_AUTHOR("Dmitry Yusupov <dmitry_yus@yahoo.com>, "
48               "Alex Aizman <itn780@yahoo.com>");
49 MODULE_DESCRIPTION("iSCSI/TCP data-path");
50 MODULE_LICENSE("GPL");
51 #undef DEBUG_TCP
52 #define DEBUG_ASSERT
53
54 #ifdef DEBUG_TCP
55 #define debug_tcp(fmt...) printk(KERN_INFO "tcp: " fmt)
56 #else
57 #define debug_tcp(fmt...)
58 #endif
59
60 #ifndef DEBUG_ASSERT
61 #ifdef BUG_ON
62 #undef BUG_ON
63 #endif
64 #define BUG_ON(expr)
65 #endif
66
67 static unsigned int iscsi_max_lun = 512;
68 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
69
70 static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
71                                    struct iscsi_chunk *chunk);
72
73 static inline void
74 iscsi_buf_init_iov(struct iscsi_buf *ibuf, char *vbuf, int size)
75 {
76         ibuf->sg.page = virt_to_page(vbuf);
77         ibuf->sg.offset = offset_in_page(vbuf);
78         ibuf->sg.length = size;
79         ibuf->sent = 0;
80         ibuf->use_sendmsg = 1;
81 }
82
83 static inline void
84 iscsi_buf_init_sg(struct iscsi_buf *ibuf, struct scatterlist *sg)
85 {
86         ibuf->sg.page = sg->page;
87         ibuf->sg.offset = sg->offset;
88         ibuf->sg.length = sg->length;
89         /*
90          * Fastpath: sg element fits into single page
91          */
92         if (sg->length + sg->offset <= PAGE_SIZE && !PageSlab(sg->page))
93                 ibuf->use_sendmsg = 0;
94         else
95                 ibuf->use_sendmsg = 1;
96         ibuf->sent = 0;
97 }
98
99 static inline int
100 iscsi_buf_left(struct iscsi_buf *ibuf)
101 {
102         int rc;
103
104         rc = ibuf->sg.length - ibuf->sent;
105         BUG_ON(rc < 0);
106         return rc;
107 }
108
109 static inline void
110 iscsi_hdr_digest(struct iscsi_conn *conn, struct iscsi_buf *buf,
111                  u8* crc)
112 {
113         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
114
115         crypto_hash_digest(&tcp_conn->tx_hash, &buf->sg, buf->sg.length, crc);
116         buf->sg.length += sizeof(u32);
117 }
118
119 /*
120  * Scatterlist handling: inside the iscsi_chunk, we
121  * remember an index into the scatterlist, and set data/size
122  * to the current scatterlist entry. For highmem pages, we
123  * kmap as needed.
124  *
125  * Note that the page is unmapped when we return from
126  * TCP's data_ready handler, so we may end up mapping and
127  * unmapping the same page repeatedly. The whole reason
128  * for this is that we shouldn't keep the page mapped
129  * outside the softirq.
130  */
131
132 /**
133  * iscsi_tcp_chunk_init_sg - init indicated scatterlist entry
134  * @chunk: the buffer object
135  * @idx: index into scatterlist
136  * @offset: byte offset into that sg entry
137  *
138  * This function sets up the chunk so that subsequent
139  * data is copied to the indicated sg entry, at the given
140  * offset.
141  */
142 static inline void
143 iscsi_tcp_chunk_init_sg(struct iscsi_chunk *chunk,
144                         unsigned int idx, unsigned int offset)
145 {
146         struct scatterlist *sg;
147
148         BUG_ON(chunk->sg == NULL);
149
150         sg = &chunk->sg[idx];
151         chunk->sg_index = idx;
152         chunk->sg_offset = offset;
153         chunk->size = min(sg->length - offset, chunk->total_size);
154         chunk->data = NULL;
155 }
156
157 /**
158  * iscsi_tcp_chunk_map - map the current S/G page
159  * @chunk: iscsi chunk
160  *
161  * We only need to possibly kmap data if scatter lists are being used,
162  * because the iscsi passthrough and internal IO paths will never use high
163  * mem pages.
164  */
165 static inline void
166 iscsi_tcp_chunk_map(struct iscsi_chunk *chunk)
167 {
168         struct scatterlist *sg;
169
170         if (chunk->data != NULL || !chunk->sg)
171                 return;
172
173         sg = &chunk->sg[chunk->sg_index];
174         BUG_ON(chunk->sg_mapped);
175         BUG_ON(sg->length == 0);
176         chunk->sg_mapped = kmap_atomic(sg->page, KM_SOFTIRQ0);
177         chunk->data = chunk->sg_mapped + sg->offset + chunk->sg_offset;
178 }
179
180 static inline void
181 iscsi_tcp_chunk_unmap(struct iscsi_chunk *chunk)
182 {
183         if (chunk->sg_mapped) {
184                 kunmap_atomic(chunk->sg_mapped, KM_SOFTIRQ0);
185                 chunk->sg_mapped = NULL;
186                 chunk->data = NULL;
187         }
188 }
189
190 /*
191  * Splice the digest buffer into the buffer
192  */
193 static inline void
194 iscsi_tcp_chunk_splice_digest(struct iscsi_chunk *chunk, void *digest)
195 {
196         chunk->data = digest;
197         chunk->digest_len = ISCSI_DIGEST_SIZE;
198         chunk->total_size += ISCSI_DIGEST_SIZE;
199         chunk->size = ISCSI_DIGEST_SIZE;
200         chunk->copied = 0;
201         chunk->sg = NULL;
202         chunk->sg_index = 0;
203         chunk->hash = NULL;
204 }
205
206 /**
207  * iscsi_tcp_chunk_done - check whether the chunk is complete
208  * @chunk: iscsi chunk to check
209  *
210  * Check if we're done receiving this chunk. If the receive
211  * buffer is full but we expect more data, move on to the
212  * next entry in the scatterlist.
213  *
214  * If the amount of data we received isn't a multiple of 4,
215  * we will transparently receive the pad bytes, too.
216  *
217  * This function must be re-entrant.
218  */
219 static inline int
220 iscsi_tcp_chunk_done(struct iscsi_chunk *chunk)
221 {
222         static unsigned char padbuf[ISCSI_PAD_LEN];
223
224         if (chunk->copied < chunk->size) {
225                 iscsi_tcp_chunk_map(chunk);
226                 return 0;
227         }
228
229         chunk->total_copied += chunk->copied;
230         chunk->copied = 0;
231         chunk->size = 0;
232
233         /* Unmap the current scatterlist page, if there is one. */
234         iscsi_tcp_chunk_unmap(chunk);
235
236         /* Do we have more scatterlist entries? */
237         if (chunk->total_copied < chunk->total_size) {
238                 /* Proceed to the next entry in the scatterlist. */
239                 iscsi_tcp_chunk_init_sg(chunk, chunk->sg_index + 1, 0);
240                 iscsi_tcp_chunk_map(chunk);
241                 BUG_ON(chunk->size == 0);
242                 return 0;
243         }
244
245         /* Do we need to handle padding? */
246         if (chunk->total_copied & (ISCSI_PAD_LEN-1)) {
247                 unsigned int pad;
248
249                 pad = ISCSI_PAD_LEN - (chunk->total_copied & (ISCSI_PAD_LEN-1));
250                 debug_tcp("consume %d pad bytes\n", pad);
251                 chunk->total_size += pad;
252                 chunk->size = pad;
253                 chunk->data = padbuf;
254                 return 0;
255         }
256
257         /*
258          * Set us up for receiving the data digest. hdr digest
259          * is completely handled in hdr done function.
260          */
261         if (chunk->hash) {
262                 if (chunk->digest_len == 0) {
263                         crypto_hash_final(chunk->hash, chunk->digest);
264                         iscsi_tcp_chunk_splice_digest(chunk,
265                                                       chunk->recv_digest);
266                         return 0;
267                 }
268         }
269
270         return 1;
271 }
272
273 /**
274  * iscsi_tcp_chunk_recv - copy data to chunk
275  * @tcp_conn: the iSCSI TCP connection
276  * @chunk: the buffer to copy to
277  * @ptr: data pointer
278  * @len: amount of data available
279  *
280  * This function copies up to @len bytes to the
281  * given buffer, and returns the number of bytes
282  * consumed, which can actually be less than @len.
283  *
284  * If hash digest is enabled, the function will update the
285  * hash while copying.
286  * Combining these two operations doesn't buy us a lot (yet),
287  * but in the future we could implement combined copy+crc,
288  * just way we do for network layer checksums.
289  */
290 static int
291 iscsi_tcp_chunk_recv(struct iscsi_tcp_conn *tcp_conn,
292                      struct iscsi_chunk *chunk, const void *ptr,
293                      unsigned int len)
294 {
295         struct scatterlist sg;
296         unsigned int copy, copied = 0;
297
298         while (!iscsi_tcp_chunk_done(chunk)) {
299                 if (copied == len)
300                         goto out;
301
302                 copy = min(len - copied, chunk->size - chunk->copied);
303                 memcpy(chunk->data + chunk->copied, ptr + copied, copy);
304
305                 if (chunk->hash) {
306                         sg_init_one(&sg, ptr + copied, copy);
307                         crypto_hash_update(chunk->hash, &sg, copy);
308                 }
309                 chunk->copied += copy;
310                 copied += copy;
311         }
312
313 out:
314         return copied;
315 }
316
317 static inline void
318 iscsi_tcp_dgst_header(struct hash_desc *hash, const void *hdr, size_t hdrlen,
319                       unsigned char digest[ISCSI_DIGEST_SIZE])
320 {
321         struct scatterlist sg;
322
323         sg_init_one(&sg, hdr, hdrlen);
324         crypto_hash_digest(hash, &sg, hdrlen, digest);
325 }
326
327 static inline int
328 iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
329                       struct iscsi_chunk *chunk)
330 {
331         if (!chunk->digest_len)
332                 return 1;
333
334         if (memcmp(chunk->recv_digest, chunk->digest, chunk->digest_len)) {
335                 debug_scsi("digest mismatch\n");
336                 return 0;
337         }
338
339         return 1;
340 }
341
342 /*
343  * Helper function to set up chunk buffer
344  */
345 static inline void
346 __iscsi_chunk_init(struct iscsi_chunk *chunk, size_t size,
347                    iscsi_chunk_done_fn_t *done, struct hash_desc *hash)
348 {
349         memset(chunk, 0, sizeof(*chunk));
350         chunk->total_size = size;
351         chunk->done = done;
352
353         if (hash) {
354                 chunk->hash = hash;
355                 crypto_hash_init(hash);
356         }
357 }
358
359 static inline void
360 iscsi_chunk_init_linear(struct iscsi_chunk *chunk, void *data, size_t size,
361                         iscsi_chunk_done_fn_t *done, struct hash_desc *hash)
362 {
363         __iscsi_chunk_init(chunk, size, done, hash);
364         chunk->data = data;
365         chunk->size = size;
366 }
367
368 static inline int
369 iscsi_chunk_seek_sg(struct iscsi_chunk *chunk,
370                     struct scatterlist *sg, unsigned int sg_count,
371                     unsigned int offset, size_t size,
372                     iscsi_chunk_done_fn_t *done, struct hash_desc *hash)
373 {
374         unsigned int i;
375
376         __iscsi_chunk_init(chunk, size, done, hash);
377         for (i = 0; i < sg_count; ++i) {
378                 if (offset < sg[i].length) {
379                         chunk->sg = sg;
380                         chunk->sg_count = sg_count;
381                         iscsi_tcp_chunk_init_sg(chunk, i, offset);
382                         return 0;
383                 }
384                 offset -= sg[i].length;
385         }
386
387         return ISCSI_ERR_DATA_OFFSET;
388 }
389
390 /**
391  * iscsi_tcp_hdr_recv_prep - prep chunk for hdr reception
392  * @tcp_conn: iscsi connection to prep for
393  *
394  * This function always passes NULL for the hash argument, because when this
395  * function is called we do not yet know the final size of the header and want
396  * to delay the digest processing until we know that.
397  */
398 static void
399 iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
400 {
401         debug_tcp("iscsi_tcp_hdr_recv_prep(%p%s)\n", tcp_conn,
402                   tcp_conn->iscsi_conn->hdrdgst_en ? ", digest enabled" : "");
403         iscsi_chunk_init_linear(&tcp_conn->in.chunk,
404                                 tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
405                                 iscsi_tcp_hdr_recv_done, NULL);
406 }
407
408 /*
409  * Handle incoming reply to any other type of command
410  */
411 static int
412 iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
413                          struct iscsi_chunk *chunk)
414 {
415         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
416         int rc = 0;
417
418         if (!iscsi_tcp_dgst_verify(tcp_conn, chunk))
419                 return ISCSI_ERR_DATA_DGST;
420
421         rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr,
422                         conn->data, tcp_conn->in.datalen);
423         if (rc)
424                 return rc;
425
426         iscsi_tcp_hdr_recv_prep(tcp_conn);
427         return 0;
428 }
429
430 static void
431 iscsi_tcp_data_recv_prep(struct iscsi_tcp_conn *tcp_conn)
432 {
433         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
434         struct hash_desc *rx_hash = NULL;
435
436         if (conn->datadgst_en)
437                 rx_hash = &tcp_conn->rx_hash;
438
439         iscsi_chunk_init_linear(&tcp_conn->in.chunk,
440                                 conn->data, tcp_conn->in.datalen,
441                                 iscsi_tcp_data_recv_done, rx_hash);
442 }
443
444 /*
445  * must be called with session lock
446  */
447 static void
448 iscsi_tcp_cleanup_ctask(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
449 {
450         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
451         struct iscsi_r2t_info *r2t;
452         struct scsi_cmnd *sc;
453
454         /* flush ctask's r2t queues */
455         while (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*))) {
456                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
457                             sizeof(void*));
458                 debug_scsi("iscsi_tcp_cleanup_ctask pending r2t dropped\n");
459         }
460
461         sc = ctask->sc;
462         if (unlikely(!sc))
463                 return;
464
465         tcp_ctask->xmstate = XMSTATE_IDLE;
466         tcp_ctask->r2t = NULL;
467 }
468
469 /**
470  * iscsi_data_rsp - SCSI Data-In Response processing
471  * @conn: iscsi connection
472  * @ctask: scsi command task
473  **/
474 static int
475 iscsi_data_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
476 {
477         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
478         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
479         struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
480         struct iscsi_session *session = conn->session;
481         struct scsi_cmnd *sc = ctask->sc;
482         int datasn = be32_to_cpu(rhdr->datasn);
483
484         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
485         /*
486          * setup Data-In byte counter (gets decremented..)
487          */
488         ctask->data_count = tcp_conn->in.datalen;
489
490         if (tcp_conn->in.datalen == 0)
491                 return 0;
492
493         if (tcp_ctask->exp_datasn != datasn) {
494                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->datasn(%d)\n",
495                           __FUNCTION__, tcp_ctask->exp_datasn, datasn);
496                 return ISCSI_ERR_DATASN;
497         }
498
499         tcp_ctask->exp_datasn++;
500
501         tcp_ctask->data_offset = be32_to_cpu(rhdr->offset);
502         if (tcp_ctask->data_offset + tcp_conn->in.datalen > scsi_bufflen(sc)) {
503                 debug_tcp("%s: data_offset(%d) + data_len(%d) > total_length_in(%d)\n",
504                           __FUNCTION__, tcp_ctask->data_offset,
505                           tcp_conn->in.datalen, scsi_bufflen(sc));
506                 return ISCSI_ERR_DATA_OFFSET;
507         }
508
509         if (rhdr->flags & ISCSI_FLAG_DATA_STATUS) {
510                 conn->exp_statsn = be32_to_cpu(rhdr->statsn) + 1;
511                 if (rhdr->flags & ISCSI_FLAG_DATA_UNDERFLOW) {
512                         int res_count = be32_to_cpu(rhdr->residual_count);
513
514                         if (res_count > 0 &&
515                             res_count <= scsi_bufflen(sc)) {
516                                 scsi_set_resid(sc, res_count);
517                                 sc->result = (DID_OK << 16) | rhdr->cmd_status;
518                         } else
519                                 sc->result = (DID_BAD_TARGET << 16) |
520                                         rhdr->cmd_status;
521                 } else if (rhdr->flags & ISCSI_FLAG_DATA_OVERFLOW) {
522                         scsi_set_resid(sc, be32_to_cpu(rhdr->residual_count));
523                         sc->result = (DID_OK << 16) | rhdr->cmd_status;
524                 } else
525                         sc->result = (DID_OK << 16) | rhdr->cmd_status;
526         }
527
528         conn->datain_pdus_cnt++;
529         return 0;
530 }
531
532 /**
533  * iscsi_solicit_data_init - initialize first Data-Out
534  * @conn: iscsi connection
535  * @ctask: scsi command task
536  * @r2t: R2T info
537  *
538  * Notes:
539  *      Initialize first Data-Out within this R2T sequence and finds
540  *      proper data_offset within this SCSI command.
541  *
542  *      This function is called with connection lock taken.
543  **/
544 static void
545 iscsi_solicit_data_init(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
546                         struct iscsi_r2t_info *r2t)
547 {
548         struct iscsi_data *hdr;
549         struct scsi_cmnd *sc = ctask->sc;
550         int i, sg_count = 0;
551         struct scatterlist *sg;
552
553         hdr = &r2t->dtask.hdr;
554         memset(hdr, 0, sizeof(struct iscsi_data));
555         hdr->ttt = r2t->ttt;
556         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
557         r2t->solicit_datasn++;
558         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
559         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
560         hdr->itt = ctask->hdr->itt;
561         hdr->exp_statsn = r2t->exp_statsn;
562         hdr->offset = cpu_to_be32(r2t->data_offset);
563         if (r2t->data_length > conn->max_xmit_dlength) {
564                 hton24(hdr->dlength, conn->max_xmit_dlength);
565                 r2t->data_count = conn->max_xmit_dlength;
566                 hdr->flags = 0;
567         } else {
568                 hton24(hdr->dlength, r2t->data_length);
569                 r2t->data_count = r2t->data_length;
570                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
571         }
572         conn->dataout_pdus_cnt++;
573
574         r2t->sent = 0;
575
576         iscsi_buf_init_iov(&r2t->headbuf, (char*)hdr,
577                            sizeof(struct iscsi_hdr));
578
579         sg = scsi_sglist(sc);
580         r2t->sg = NULL;
581         for (i = 0; i < scsi_sg_count(sc); i++, sg += 1) {
582                 /* FIXME: prefetch ? */
583                 if (sg_count + sg->length > r2t->data_offset) {
584                         int page_offset;
585
586                         /* sg page found! */
587
588                         /* offset within this page */
589                         page_offset = r2t->data_offset - sg_count;
590
591                         /* fill in this buffer */
592                         iscsi_buf_init_sg(&r2t->sendbuf, sg);
593                         r2t->sendbuf.sg.offset += page_offset;
594                         r2t->sendbuf.sg.length -= page_offset;
595
596                         /* xmit logic will continue with next one */
597                         r2t->sg = sg + 1;
598                         break;
599                 }
600                 sg_count += sg->length;
601         }
602         BUG_ON(r2t->sg == NULL);
603 }
604
605 /**
606  * iscsi_r2t_rsp - iSCSI R2T Response processing
607  * @conn: iscsi connection
608  * @ctask: scsi command task
609  **/
610 static int
611 iscsi_r2t_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
612 {
613         struct iscsi_r2t_info *r2t;
614         struct iscsi_session *session = conn->session;
615         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->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                 printk(KERN_ERR "iscsi_tcp: invalid R2t with datalen %d\n",
623                        tcp_conn->in.datalen);
624                 return ISCSI_ERR_DATALEN;
625         }
626
627         if (tcp_ctask->exp_datasn != r2tsn){
628                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
629                           __FUNCTION__, tcp_ctask->exp_datasn, r2tsn);
630                 return ISCSI_ERR_R2TSN;
631         }
632
633         /* fill-in new R2T associated with the task */
634         spin_lock(&session->lock);
635         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
636
637         if (!ctask->sc || session->state != ISCSI_STATE_LOGGED_IN) {
638                 printk(KERN_INFO "iscsi_tcp: dropping R2T itt %d in "
639                        "recovery...\n", ctask->itt);
640                 spin_unlock(&session->lock);
641                 return 0;
642         }
643
644         rc = __kfifo_get(tcp_ctask->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                 printk(KERN_ERR "iscsi_tcp: invalid R2T with zero data len\n");
651                 spin_unlock(&session->lock);
652                 return ISCSI_ERR_DATALEN;
653         }
654
655         if (r2t->data_length > session->max_burst)
656                 debug_scsi("invalid R2T with data len %u and max burst %u."
657                            "Attempting to execute request.\n",
658                             r2t->data_length, session->max_burst);
659
660         r2t->data_offset = be32_to_cpu(rhdr->data_offset);
661         if (r2t->data_offset + r2t->data_length > scsi_bufflen(ctask->sc)) {
662                 spin_unlock(&session->lock);
663                 printk(KERN_ERR "iscsi_tcp: invalid R2T with data len %u at "
664                        "offset %u and total length %d\n", r2t->data_length,
665                        r2t->data_offset, scsi_bufflen(ctask->sc));
666                 return ISCSI_ERR_DATALEN;
667         }
668
669         r2t->ttt = rhdr->ttt; /* no flip */
670         r2t->solicit_datasn = 0;
671
672         iscsi_solicit_data_init(conn, ctask, r2t);
673
674         tcp_ctask->exp_datasn = r2tsn + 1;
675         __kfifo_put(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*));
676         tcp_ctask->xmstate |= XMSTATE_SOL_HDR_INIT;
677         conn->r2t_pdus_cnt++;
678
679         iscsi_requeue_ctask(ctask);
680         spin_unlock(&session->lock);
681
682         return 0;
683 }
684
685 /*
686  * Handle incoming reply to DataIn command
687  */
688 static int
689 iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
690                           struct iscsi_chunk *chunk)
691 {
692         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
693         struct iscsi_hdr *hdr = tcp_conn->in.hdr;
694         int rc;
695
696         if (!iscsi_tcp_dgst_verify(tcp_conn, chunk))
697                 return ISCSI_ERR_DATA_DGST;
698
699         /* check for non-exceptional status */
700         if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
701                 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
702                 if (rc)
703                         return rc;
704         }
705
706         iscsi_tcp_hdr_recv_prep(tcp_conn);
707         return 0;
708 }
709
710 /**
711  * iscsi_tcp_hdr_dissect - process PDU header
712  * @conn: iSCSI connection
713  * @hdr: PDU header
714  *
715  * This function analyzes the header of the PDU received,
716  * and performs several sanity checks. If the PDU is accompanied
717  * by data, the receive buffer is set up to copy the incoming data
718  * to the correct location.
719  */
720 static int
721 iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
722 {
723         int rc = 0, opcode, ahslen;
724         struct iscsi_session *session = conn->session;
725         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
726         struct iscsi_cmd_task *ctask;
727         uint32_t itt;
728
729         /* verify PDU length */
730         tcp_conn->in.datalen = ntoh24(hdr->dlength);
731         if (tcp_conn->in.datalen > conn->max_recv_dlength) {
732                 printk(KERN_ERR "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 == ISCSI_ERR_NO_SCSI_CMD) {
746                 /* XXX: what does this do? */
747                 tcp_conn->in.datalen = 0; /* force drop */
748                 return 0;
749         } else if (rc)
750                 return rc;
751
752         debug_tcp("opcode 0x%x ahslen %d datalen %d\n",
753                   opcode, ahslen, tcp_conn->in.datalen);
754
755         switch(opcode) {
756         case ISCSI_OP_SCSI_DATA_IN:
757                 ctask = session->cmds[itt];
758                 rc = iscsi_data_rsp(conn, ctask);
759                 if (rc)
760                         return rc;
761                 if (tcp_conn->in.datalen) {
762                         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
763                         struct hash_desc *rx_hash = NULL;
764
765                         /*
766                          * Setup copy of Data-In into the Scsi_Cmnd
767                          * Scatterlist case:
768                          * We set up the iscsi_chunk to point to the next
769                          * scatterlist entry to copy to. As we go along,
770                          * we move on to the next scatterlist entry and
771                          * update the digest per-entry.
772                          */
773                         if (conn->datadgst_en)
774                                 rx_hash = &tcp_conn->rx_hash;
775
776                         debug_tcp("iscsi_tcp_begin_data_in(%p, offset=%d, "
777                                   "datalen=%d)\n", tcp_conn,
778                                   tcp_ctask->data_offset,
779                                   tcp_conn->in.datalen);
780                         return iscsi_chunk_seek_sg(&tcp_conn->in.chunk,
781                                                 scsi_sglist(ctask->sc),
782                                                 scsi_sg_count(ctask->sc),
783                                                 tcp_ctask->data_offset,
784                                                 tcp_conn->in.datalen,
785                                                 iscsi_tcp_process_data_in,
786                                                 rx_hash);
787                 }
788                 /* fall through */
789         case ISCSI_OP_SCSI_CMD_RSP:
790                 if (tcp_conn->in.datalen) {
791                         iscsi_tcp_data_recv_prep(tcp_conn);
792                         return 0;
793                 }
794                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
795                 break;
796         case ISCSI_OP_R2T:
797                 ctask = session->cmds[itt];
798                 if (ahslen)
799                         rc = ISCSI_ERR_AHSLEN;
800                 else if (ctask->sc->sc_data_direction == DMA_TO_DEVICE)
801                         rc = iscsi_r2t_rsp(conn, ctask);
802                 else
803                         rc = ISCSI_ERR_PROTO;
804                 break;
805         case ISCSI_OP_LOGIN_RSP:
806         case ISCSI_OP_TEXT_RSP:
807         case ISCSI_OP_REJECT:
808         case ISCSI_OP_ASYNC_EVENT:
809                 /*
810                  * It is possible that we could get a PDU with a buffer larger
811                  * than 8K, but there are no targets that currently do this.
812                  * For now we fail until we find a vendor that needs it
813                  */
814                 if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
815                         printk(KERN_ERR "iscsi_tcp: received buffer of len %u "
816                               "but conn buffer is only %u (opcode %0x)\n",
817                               tcp_conn->in.datalen,
818                               ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
819                         rc = ISCSI_ERR_PROTO;
820                         break;
821                 }
822
823                 /* If there's data coming in with the response,
824                  * receive it to the connection's buffer.
825                  */
826                 if (tcp_conn->in.datalen) {
827                         iscsi_tcp_data_recv_prep(tcp_conn);
828                         return 0;
829                 }
830         /* fall through */
831         case ISCSI_OP_LOGOUT_RSP:
832         case ISCSI_OP_NOOP_IN:
833         case ISCSI_OP_SCSI_TMFUNC_RSP:
834                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
835                 break;
836         default:
837                 rc = ISCSI_ERR_BAD_OPCODE;
838                 break;
839         }
840
841         if (rc == 0) {
842                 /* Anything that comes with data should have
843                  * been handled above. */
844                 if (tcp_conn->in.datalen)
845                         return ISCSI_ERR_PROTO;
846                 iscsi_tcp_hdr_recv_prep(tcp_conn);
847         }
848
849         return rc;
850 }
851
852 static inline void
853 partial_sg_digest_update(struct hash_desc *desc, struct scatterlist *sg,
854                          int offset, int length)
855 {
856         struct scatterlist temp;
857
858         sg_init_table(&temp, 1);
859         sg_set_page(&temp, sg_page(sg), length, offset);
860         crypto_hash_update(desc, &temp, length);
861 }
862
863 /**
864  * iscsi_tcp_hdr_recv_done - process PDU header
865  *
866  * This is the callback invoked when the PDU header has
867  * been received. If the header is followed by additional
868  * header segments, we go back for more data.
869  */
870 static int
871 iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
872                         struct iscsi_chunk *chunk)
873 {
874         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
875         struct iscsi_hdr *hdr;
876
877         /* Check if there are additional header segments
878          * *prior* to computing the digest, because we
879          * may need to go back to the caller for more.
880          */
881         hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
882         if (chunk->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
883                 /* Bump the header length - the caller will
884                  * just loop around and get the AHS for us, and
885                  * call again. */
886                 unsigned int ahslen = hdr->hlength << 2;
887
888                 /* Make sure we don't overflow */
889                 if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
890                         return ISCSI_ERR_AHSLEN;
891
892                 chunk->total_size += ahslen;
893                 chunk->size += ahslen;
894                 return 0;
895         }
896
897         /* We're done processing the header. See if we're doing
898          * header digests; if so, set up the recv_digest buffer
899          * and go back for more. */
900         if (conn->hdrdgst_en) {
901                 if (chunk->digest_len == 0) {
902                         iscsi_tcp_chunk_splice_digest(chunk,
903                                                       chunk->recv_digest);
904                         return 0;
905                 }
906                 iscsi_tcp_dgst_header(&tcp_conn->rx_hash, hdr,
907                                       chunk->total_copied - ISCSI_DIGEST_SIZE,
908                                       chunk->digest);
909
910                 if (!iscsi_tcp_dgst_verify(tcp_conn, chunk))
911                         return ISCSI_ERR_HDR_DGST;
912         }
913
914         tcp_conn->in.hdr = hdr;
915         return iscsi_tcp_hdr_dissect(conn, hdr);
916 }
917
918 /**
919  * iscsi_tcp_recv - TCP receive in sendfile fashion
920  * @rd_desc: read descriptor
921  * @skb: socket buffer
922  * @offset: offset in skb
923  * @len: skb->len - offset
924  **/
925 static int
926 iscsi_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
927                unsigned int offset, size_t len)
928 {
929         struct iscsi_conn *conn = rd_desc->arg.data;
930         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
931         struct iscsi_chunk *chunk = &tcp_conn->in.chunk;
932         struct skb_seq_state seq;
933         unsigned int consumed = 0;
934         int rc = 0;
935
936         debug_tcp("in %d bytes\n", skb->len - offset);
937
938         if (unlikely(conn->suspend_rx)) {
939                 debug_tcp("conn %d Rx suspended!\n", conn->id);
940                 return 0;
941         }
942
943         skb_prepare_seq_read(skb, offset, skb->len, &seq);
944         while (1) {
945                 unsigned int avail;
946                 const u8 *ptr;
947
948                 avail = skb_seq_read(consumed, &ptr, &seq);
949                 if (avail == 0)
950                         break;
951                 BUG_ON(chunk->copied >= chunk->size);
952
953                 debug_tcp("skb %p ptr=%p avail=%u\n", skb, ptr, avail);
954                 rc = iscsi_tcp_chunk_recv(tcp_conn, chunk, ptr, avail);
955                 BUG_ON(rc == 0);
956                 consumed += rc;
957
958                 if (chunk->total_copied >= chunk->total_size) {
959                         rc = chunk->done(tcp_conn, chunk);
960                         if (rc != 0) {
961                                 skb_abort_seq_read(&seq);
962                                 goto error;
963                         }
964
965                         /* The done() functions sets up the
966                          * next chunk. */
967                 }
968         }
969
970         conn->rxdata_octets += consumed;
971         return consumed;
972
973 error:
974         debug_tcp("Error receiving PDU, errno=%d\n", rc);
975         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
976         return 0;
977 }
978
979 static void
980 iscsi_tcp_data_ready(struct sock *sk, int flag)
981 {
982         struct iscsi_conn *conn = sk->sk_user_data;
983         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
984         read_descriptor_t rd_desc;
985
986         read_lock(&sk->sk_callback_lock);
987
988         /*
989          * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
990          * We set count to 1 because we want the network layer to
991          * hand us all the skbs that are available. iscsi_tcp_recv
992          * handled pdus that cross buffers or pdus that still need data.
993          */
994         rd_desc.arg.data = conn;
995         rd_desc.count = 1;
996         tcp_read_sock(sk, &rd_desc, iscsi_tcp_recv);
997
998         read_unlock(&sk->sk_callback_lock);
999
1000         /* If we had to (atomically) map a highmem page,
1001          * unmap it now. */
1002         iscsi_tcp_chunk_unmap(&tcp_conn->in.chunk);
1003 }
1004
1005 static void
1006 iscsi_tcp_state_change(struct sock *sk)
1007 {
1008         struct iscsi_tcp_conn *tcp_conn;
1009         struct iscsi_conn *conn;
1010         struct iscsi_session *session;
1011         void (*old_state_change)(struct sock *);
1012
1013         read_lock(&sk->sk_callback_lock);
1014
1015         conn = (struct iscsi_conn*)sk->sk_user_data;
1016         session = conn->session;
1017
1018         if ((sk->sk_state == TCP_CLOSE_WAIT ||
1019              sk->sk_state == TCP_CLOSE) &&
1020             !atomic_read(&sk->sk_rmem_alloc)) {
1021                 debug_tcp("iscsi_tcp_state_change: TCP_CLOSE|TCP_CLOSE_WAIT\n");
1022                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1023         }
1024
1025         tcp_conn = conn->dd_data;
1026         old_state_change = tcp_conn->old_state_change;
1027
1028         read_unlock(&sk->sk_callback_lock);
1029
1030         old_state_change(sk);
1031 }
1032
1033 /**
1034  * iscsi_write_space - Called when more output buffer space is available
1035  * @sk: socket space is available for
1036  **/
1037 static void
1038 iscsi_write_space(struct sock *sk)
1039 {
1040         struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
1041         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1042
1043         tcp_conn->old_write_space(sk);
1044         debug_tcp("iscsi_write_space: cid %d\n", conn->id);
1045         scsi_queue_work(conn->session->host, &conn->xmitwork);
1046 }
1047
1048 static void
1049 iscsi_conn_set_callbacks(struct iscsi_conn *conn)
1050 {
1051         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1052         struct sock *sk = tcp_conn->sock->sk;
1053
1054         /* assign new callbacks */
1055         write_lock_bh(&sk->sk_callback_lock);
1056         sk->sk_user_data = conn;
1057         tcp_conn->old_data_ready = sk->sk_data_ready;
1058         tcp_conn->old_state_change = sk->sk_state_change;
1059         tcp_conn->old_write_space = sk->sk_write_space;
1060         sk->sk_data_ready = iscsi_tcp_data_ready;
1061         sk->sk_state_change = iscsi_tcp_state_change;
1062         sk->sk_write_space = iscsi_write_space;
1063         write_unlock_bh(&sk->sk_callback_lock);
1064 }
1065
1066 static void
1067 iscsi_conn_restore_callbacks(struct iscsi_tcp_conn *tcp_conn)
1068 {
1069         struct sock *sk = tcp_conn->sock->sk;
1070
1071         /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
1072         write_lock_bh(&sk->sk_callback_lock);
1073         sk->sk_user_data    = NULL;
1074         sk->sk_data_ready   = tcp_conn->old_data_ready;
1075         sk->sk_state_change = tcp_conn->old_state_change;
1076         sk->sk_write_space  = tcp_conn->old_write_space;
1077         sk->sk_no_check  = 0;
1078         write_unlock_bh(&sk->sk_callback_lock);
1079 }
1080
1081 /**
1082  * iscsi_send - generic send routine
1083  * @sk: kernel's socket
1084  * @buf: buffer to write from
1085  * @size: actual size to write
1086  * @flags: socket's flags
1087  */
1088 static inline int
1089 iscsi_send(struct iscsi_conn *conn, struct iscsi_buf *buf, int size, int flags)
1090 {
1091         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1092         struct socket *sk = tcp_conn->sock;
1093         int offset = buf->sg.offset + buf->sent, res;
1094
1095         /*
1096          * if we got use_sg=0 or are sending something we kmallocd
1097          * then we did not have to do kmap (kmap returns page_address)
1098          *
1099          * if we got use_sg > 0, but had to drop down, we do not
1100          * set clustering so this should only happen for that
1101          * slab case.
1102          */
1103         if (buf->use_sendmsg)
1104                 res = sock_no_sendpage(sk, buf->sg.page, offset, size, flags);
1105         else
1106                 res = tcp_conn->sendpage(sk, buf->sg.page, offset, size, flags);
1107
1108         if (res >= 0) {
1109                 conn->txdata_octets += res;
1110                 buf->sent += res;
1111                 return res;
1112         }
1113
1114         tcp_conn->sendpage_failures_cnt++;
1115         if (res == -EAGAIN)
1116                 res = -ENOBUFS;
1117         else
1118                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1119         return res;
1120 }
1121
1122 /**
1123  * iscsi_sendhdr - send PDU Header via tcp_sendpage()
1124  * @conn: iscsi connection
1125  * @buf: buffer to write from
1126  * @datalen: lenght of data to be sent after the header
1127  *
1128  * Notes:
1129  *      (Tx, Fast Path)
1130  **/
1131 static inline int
1132 iscsi_sendhdr(struct iscsi_conn *conn, struct iscsi_buf *buf, int datalen)
1133 {
1134         int flags = 0; /* MSG_DONTWAIT; */
1135         int res, size;
1136
1137         size = buf->sg.length - buf->sent;
1138         BUG_ON(buf->sent + size > buf->sg.length);
1139         if (buf->sent + size != buf->sg.length || datalen)
1140                 flags |= MSG_MORE;
1141
1142         res = iscsi_send(conn, buf, size, flags);
1143         debug_tcp("sendhdr %d bytes, sent %d res %d\n", size, buf->sent, res);
1144         if (res >= 0) {
1145                 if (size != res)
1146                         return -EAGAIN;
1147                 return 0;
1148         }
1149
1150         return res;
1151 }
1152
1153 /**
1154  * iscsi_sendpage - send one page of iSCSI Data-Out.
1155  * @conn: iscsi connection
1156  * @buf: buffer to write from
1157  * @count: remaining data
1158  * @sent: number of bytes sent
1159  *
1160  * Notes:
1161  *      (Tx, Fast Path)
1162  **/
1163 static inline int
1164 iscsi_sendpage(struct iscsi_conn *conn, struct iscsi_buf *buf,
1165                int *count, int *sent)
1166 {
1167         int flags = 0; /* MSG_DONTWAIT; */
1168         int res, size;
1169
1170         size = buf->sg.length - buf->sent;
1171         BUG_ON(buf->sent + size > buf->sg.length);
1172         if (size > *count)
1173                 size = *count;
1174         if (buf->sent + size != buf->sg.length || *count != size)
1175                 flags |= MSG_MORE;
1176
1177         res = iscsi_send(conn, buf, size, flags);
1178         debug_tcp("sendpage: %d bytes, sent %d left %d sent %d res %d\n",
1179                   size, buf->sent, *count, *sent, res);
1180         if (res >= 0) {
1181                 *count -= res;
1182                 *sent += res;
1183                 if (size != res)
1184                         return -EAGAIN;
1185                 return 0;
1186         }
1187
1188         return res;
1189 }
1190
1191 static inline void
1192 iscsi_data_digest_init(struct iscsi_tcp_conn *tcp_conn,
1193                       struct iscsi_tcp_cmd_task *tcp_ctask)
1194 {
1195         crypto_hash_init(&tcp_conn->tx_hash);
1196         tcp_ctask->digest_count = 4;
1197 }
1198
1199 /**
1200  * iscsi_solicit_data_cont - initialize next Data-Out
1201  * @conn: iscsi connection
1202  * @ctask: scsi command task
1203  * @r2t: R2T info
1204  * @left: bytes left to transfer
1205  *
1206  * Notes:
1207  *      Initialize next Data-Out within this R2T sequence and continue
1208  *      to process next Scatter-Gather element(if any) of this SCSI command.
1209  *
1210  *      Called under connection lock.
1211  **/
1212 static void
1213 iscsi_solicit_data_cont(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
1214                         struct iscsi_r2t_info *r2t, int left)
1215 {
1216         struct iscsi_data *hdr;
1217         int new_offset;
1218
1219         hdr = &r2t->dtask.hdr;
1220         memset(hdr, 0, sizeof(struct iscsi_data));
1221         hdr->ttt = r2t->ttt;
1222         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
1223         r2t->solicit_datasn++;
1224         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
1225         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
1226         hdr->itt = ctask->hdr->itt;
1227         hdr->exp_statsn = r2t->exp_statsn;
1228         new_offset = r2t->data_offset + r2t->sent;
1229         hdr->offset = cpu_to_be32(new_offset);
1230         if (left > conn->max_xmit_dlength) {
1231                 hton24(hdr->dlength, conn->max_xmit_dlength);
1232                 r2t->data_count = conn->max_xmit_dlength;
1233         } else {
1234                 hton24(hdr->dlength, left);
1235                 r2t->data_count = left;
1236                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
1237         }
1238         conn->dataout_pdus_cnt++;
1239
1240         iscsi_buf_init_iov(&r2t->headbuf, (char*)hdr,
1241                            sizeof(struct iscsi_hdr));
1242
1243         if (iscsi_buf_left(&r2t->sendbuf))
1244                 return;
1245
1246         iscsi_buf_init_sg(&r2t->sendbuf, r2t->sg);
1247         r2t->sg += 1;
1248 }
1249
1250 static void iscsi_set_padding(struct iscsi_tcp_cmd_task *tcp_ctask,
1251                               unsigned long len)
1252 {
1253         tcp_ctask->pad_count = len & (ISCSI_PAD_LEN - 1);
1254         if (!tcp_ctask->pad_count)
1255                 return;
1256
1257         tcp_ctask->pad_count = ISCSI_PAD_LEN - tcp_ctask->pad_count;
1258         debug_scsi("write padding %d bytes\n", tcp_ctask->pad_count);
1259         tcp_ctask->xmstate |= XMSTATE_W_PAD;
1260 }
1261
1262 /**
1263  * iscsi_tcp_cmd_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
1264  * @conn: iscsi connection
1265  * @ctask: scsi command task
1266  * @sc: scsi command
1267  **/
1268 static void
1269 iscsi_tcp_cmd_init(struct iscsi_cmd_task *ctask)
1270 {
1271         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1272
1273         BUG_ON(__kfifo_len(tcp_ctask->r2tqueue));
1274         tcp_ctask->xmstate = XMSTATE_CMD_HDR_INIT;
1275 }
1276
1277 /**
1278  * iscsi_tcp_mtask_xmit - xmit management(immediate) task
1279  * @conn: iscsi connection
1280  * @mtask: task management task
1281  *
1282  * Notes:
1283  *      The function can return -EAGAIN in which case caller must
1284  *      call it again later, or recover. '0' return code means successful
1285  *      xmit.
1286  *
1287  *      Management xmit state machine consists of these states:
1288  *              XMSTATE_IMM_HDR_INIT    - calculate digest of PDU Header
1289  *              XMSTATE_IMM_HDR         - PDU Header xmit in progress
1290  *              XMSTATE_IMM_DATA        - PDU Data xmit in progress
1291  *              XMSTATE_IDLE            - management PDU is done
1292  **/
1293 static int
1294 iscsi_tcp_mtask_xmit(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1295 {
1296         struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
1297         int rc;
1298
1299         debug_scsi("mtask deq [cid %d state %x itt 0x%x]\n",
1300                 conn->id, tcp_mtask->xmstate, mtask->itt);
1301
1302         if (tcp_mtask->xmstate & XMSTATE_IMM_HDR_INIT) {
1303                 iscsi_buf_init_iov(&tcp_mtask->headbuf, (char*)mtask->hdr,
1304                                    sizeof(struct iscsi_hdr));
1305
1306                 if (mtask->data_count) {
1307                         tcp_mtask->xmstate |= XMSTATE_IMM_DATA;
1308                         iscsi_buf_init_iov(&tcp_mtask->sendbuf,
1309                                            (char*)mtask->data,
1310                                            mtask->data_count);
1311                 }
1312
1313                 if (conn->c_stage != ISCSI_CONN_INITIAL_STAGE &&
1314                     conn->stop_stage != STOP_CONN_RECOVER &&
1315                     conn->hdrdgst_en)
1316                         iscsi_hdr_digest(conn, &tcp_mtask->headbuf,
1317                                         (u8*)tcp_mtask->hdrext);
1318
1319                 tcp_mtask->sent = 0;
1320                 tcp_mtask->xmstate &= ~XMSTATE_IMM_HDR_INIT;
1321                 tcp_mtask->xmstate |= XMSTATE_IMM_HDR;
1322         }
1323
1324         if (tcp_mtask->xmstate & XMSTATE_IMM_HDR) {
1325                 rc = iscsi_sendhdr(conn, &tcp_mtask->headbuf,
1326                                    mtask->data_count);
1327                 if (rc)
1328                         return rc;
1329                 tcp_mtask->xmstate &= ~XMSTATE_IMM_HDR;
1330         }
1331
1332         if (tcp_mtask->xmstate & XMSTATE_IMM_DATA) {
1333                 BUG_ON(!mtask->data_count);
1334                 tcp_mtask->xmstate &= ~XMSTATE_IMM_DATA;
1335                 /* FIXME: implement.
1336                  * Virtual buffer could be spreaded across multiple pages...
1337                  */
1338                 do {
1339                         int rc;
1340
1341                         rc = iscsi_sendpage(conn, &tcp_mtask->sendbuf,
1342                                         &mtask->data_count, &tcp_mtask->sent);
1343                         if (rc) {
1344                                 tcp_mtask->xmstate |= XMSTATE_IMM_DATA;
1345                                 return rc;
1346                         }
1347                 } while (mtask->data_count);
1348         }
1349
1350         BUG_ON(tcp_mtask->xmstate != XMSTATE_IDLE);
1351         if (mtask->hdr->itt == RESERVED_ITT) {
1352                 struct iscsi_session *session = conn->session;
1353
1354                 spin_lock_bh(&session->lock);
1355                 list_del(&conn->mtask->running);
1356                 __kfifo_put(session->mgmtpool.queue, (void*)&conn->mtask,
1357                             sizeof(void*));
1358                 spin_unlock_bh(&session->lock);
1359         }
1360         return 0;
1361 }
1362
1363 static int
1364 iscsi_send_cmd_hdr(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1365 {
1366         struct scsi_cmnd *sc = ctask->sc;
1367         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1368         int rc = 0;
1369
1370         if (tcp_ctask->xmstate & XMSTATE_CMD_HDR_INIT) {
1371                 tcp_ctask->sent = 0;
1372                 tcp_ctask->sg_count = 0;
1373                 tcp_ctask->exp_datasn = 0;
1374
1375                 if (sc->sc_data_direction == DMA_TO_DEVICE) {
1376                         struct scatterlist *sg = scsi_sglist(sc);
1377
1378                         iscsi_buf_init_sg(&tcp_ctask->sendbuf, sg);
1379                         tcp_ctask->sg = sg + 1;
1380                         tcp_ctask->bad_sg = sg + scsi_sg_count(sc);
1381
1382                         debug_scsi("cmd [itt 0x%x total %d imm_data %d "
1383                                    "unsol count %d, unsol offset %d]\n",
1384                                    ctask->itt, scsi_bufflen(sc),
1385                                    ctask->imm_count, ctask->unsol_count,
1386                                    ctask->unsol_offset);
1387                 }
1388
1389                 iscsi_buf_init_iov(&tcp_ctask->headbuf, (char*)ctask->hdr,
1390                                   sizeof(struct iscsi_hdr));
1391
1392                 if (conn->hdrdgst_en)
1393                         iscsi_hdr_digest(conn, &tcp_ctask->headbuf,
1394                                          (u8*)tcp_ctask->hdrext);
1395                 tcp_ctask->xmstate &= ~XMSTATE_CMD_HDR_INIT;
1396                 tcp_ctask->xmstate |= XMSTATE_CMD_HDR_XMIT;
1397         }
1398
1399         if (tcp_ctask->xmstate & XMSTATE_CMD_HDR_XMIT) {
1400                 rc = iscsi_sendhdr(conn, &tcp_ctask->headbuf, ctask->imm_count);
1401                 if (rc)
1402                         return rc;
1403                 tcp_ctask->xmstate &= ~XMSTATE_CMD_HDR_XMIT;
1404
1405                 if (sc->sc_data_direction != DMA_TO_DEVICE)
1406                         return 0;
1407
1408                 if (ctask->imm_count) {
1409                         tcp_ctask->xmstate |= XMSTATE_IMM_DATA;
1410                         iscsi_set_padding(tcp_ctask, ctask->imm_count);
1411
1412                         if (ctask->conn->datadgst_en) {
1413                                 iscsi_data_digest_init(ctask->conn->dd_data,
1414                                                        tcp_ctask);
1415                                 tcp_ctask->immdigest = 0;
1416                         }
1417                 }
1418
1419                 if (ctask->unsol_count)
1420                         tcp_ctask->xmstate |=
1421                                         XMSTATE_UNS_HDR | XMSTATE_UNS_INIT;
1422         }
1423         return rc;
1424 }
1425
1426 static int
1427 iscsi_send_padding(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1428 {
1429         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1430         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1431         int sent = 0, rc;
1432
1433         if (tcp_ctask->xmstate & XMSTATE_W_PAD) {
1434                 iscsi_buf_init_iov(&tcp_ctask->sendbuf, (char*)&tcp_ctask->pad,
1435                                    tcp_ctask->pad_count);
1436                 if (conn->datadgst_en)
1437                         crypto_hash_update(&tcp_conn->tx_hash,
1438                                            &tcp_ctask->sendbuf.sg,
1439                                            tcp_ctask->sendbuf.sg.length);
1440         } else if (!(tcp_ctask->xmstate & XMSTATE_W_RESEND_PAD))
1441                 return 0;
1442
1443         tcp_ctask->xmstate &= ~XMSTATE_W_PAD;
1444         tcp_ctask->xmstate &= ~XMSTATE_W_RESEND_PAD;
1445         debug_scsi("sending %d pad bytes for itt 0x%x\n",
1446                    tcp_ctask->pad_count, ctask->itt);
1447         rc = iscsi_sendpage(conn, &tcp_ctask->sendbuf, &tcp_ctask->pad_count,
1448                            &sent);
1449         if (rc) {
1450                 debug_scsi("padding send failed %d\n", rc);
1451                 tcp_ctask->xmstate |= XMSTATE_W_RESEND_PAD;
1452         }
1453         return rc;
1454 }
1455
1456 static int
1457 iscsi_send_digest(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
1458                         struct iscsi_buf *buf, uint32_t *digest)
1459 {
1460         struct iscsi_tcp_cmd_task *tcp_ctask;
1461         struct iscsi_tcp_conn *tcp_conn;
1462         int rc, sent = 0;
1463
1464         if (!conn->datadgst_en)
1465                 return 0;
1466
1467         tcp_ctask = ctask->dd_data;
1468         tcp_conn = conn->dd_data;
1469
1470         if (!(tcp_ctask->xmstate & XMSTATE_W_RESEND_DATA_DIGEST)) {
1471                 crypto_hash_final(&tcp_conn->tx_hash, (u8*)digest);
1472                 iscsi_buf_init_iov(buf, (char*)digest, 4);
1473         }
1474         tcp_ctask->xmstate &= ~XMSTATE_W_RESEND_DATA_DIGEST;
1475
1476         rc = iscsi_sendpage(conn, buf, &tcp_ctask->digest_count, &sent);
1477         if (!rc)
1478                 debug_scsi("sent digest 0x%x for itt 0x%x\n", *digest,
1479                           ctask->itt);
1480         else {
1481                 debug_scsi("sending digest 0x%x failed for itt 0x%x!\n",
1482                           *digest, ctask->itt);
1483                 tcp_ctask->xmstate |= XMSTATE_W_RESEND_DATA_DIGEST;
1484         }
1485         return rc;
1486 }
1487
1488 static int
1489 iscsi_send_data(struct iscsi_cmd_task *ctask, struct iscsi_buf *sendbuf,
1490                 struct scatterlist **sg, int *sent, int *count,
1491                 struct iscsi_buf *digestbuf, uint32_t *digest)
1492 {
1493         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1494         struct iscsi_conn *conn = ctask->conn;
1495         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1496         int rc, buf_sent, offset;
1497
1498         while (*count) {
1499                 buf_sent = 0;
1500                 offset = sendbuf->sent;
1501
1502                 rc = iscsi_sendpage(conn, sendbuf, count, &buf_sent);
1503                 *sent = *sent + buf_sent;
1504                 if (buf_sent && conn->datadgst_en)
1505                         partial_sg_digest_update(&tcp_conn->tx_hash,
1506                                 &sendbuf->sg, sendbuf->sg.offset + offset,
1507                                 buf_sent);
1508                 if (!iscsi_buf_left(sendbuf) && *sg != tcp_ctask->bad_sg) {
1509                         iscsi_buf_init_sg(sendbuf, *sg);
1510                         *sg = *sg + 1;
1511                 }
1512
1513                 if (rc)
1514                         return rc;
1515         }
1516
1517         rc = iscsi_send_padding(conn, ctask);
1518         if (rc)
1519                 return rc;
1520
1521         return iscsi_send_digest(conn, ctask, digestbuf, digest);
1522 }
1523
1524 static int
1525 iscsi_send_unsol_hdr(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1526 {
1527         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1528         struct iscsi_data_task *dtask;
1529         int rc;
1530
1531         tcp_ctask->xmstate |= XMSTATE_UNS_DATA;
1532         if (tcp_ctask->xmstate & XMSTATE_UNS_INIT) {
1533                 dtask = &tcp_ctask->unsol_dtask;
1534
1535                 iscsi_prep_unsolicit_data_pdu(ctask, &dtask->hdr);
1536                 iscsi_buf_init_iov(&tcp_ctask->headbuf, (char*)&dtask->hdr,
1537                                    sizeof(struct iscsi_hdr));
1538                 if (conn->hdrdgst_en)
1539                         iscsi_hdr_digest(conn, &tcp_ctask->headbuf,
1540                                         (u8*)dtask->hdrext);
1541
1542                 tcp_ctask->xmstate &= ~XMSTATE_UNS_INIT;
1543                 iscsi_set_padding(tcp_ctask, ctask->data_count);
1544         }
1545
1546         rc = iscsi_sendhdr(conn, &tcp_ctask->headbuf, ctask->data_count);
1547         if (rc) {
1548                 tcp_ctask->xmstate &= ~XMSTATE_UNS_DATA;
1549                 tcp_ctask->xmstate |= XMSTATE_UNS_HDR;
1550                 return rc;
1551         }
1552
1553         if (conn->datadgst_en) {
1554                 dtask = &tcp_ctask->unsol_dtask;
1555                 iscsi_data_digest_init(ctask->conn->dd_data, tcp_ctask);
1556                 dtask->digest = 0;
1557         }
1558
1559         debug_scsi("uns dout [itt 0x%x dlen %d sent %d]\n",
1560                    ctask->itt, ctask->unsol_count, tcp_ctask->sent);
1561         return 0;
1562 }
1563
1564 static int
1565 iscsi_send_unsol_pdu(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1566 {
1567         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1568         int rc;
1569
1570         if (tcp_ctask->xmstate & XMSTATE_UNS_HDR) {
1571                 BUG_ON(!ctask->unsol_count);
1572                 tcp_ctask->xmstate &= ~XMSTATE_UNS_HDR;
1573 send_hdr:
1574                 rc = iscsi_send_unsol_hdr(conn, ctask);
1575                 if (rc)
1576                         return rc;
1577         }
1578
1579         if (tcp_ctask->xmstate & XMSTATE_UNS_DATA) {
1580                 struct iscsi_data_task *dtask = &tcp_ctask->unsol_dtask;
1581                 int start = tcp_ctask->sent;
1582
1583                 rc = iscsi_send_data(ctask, &tcp_ctask->sendbuf, &tcp_ctask->sg,
1584                                      &tcp_ctask->sent, &ctask->data_count,
1585                                      &dtask->digestbuf, &dtask->digest);
1586                 ctask->unsol_count -= tcp_ctask->sent - start;
1587                 if (rc)
1588                         return rc;
1589                 tcp_ctask->xmstate &= ~XMSTATE_UNS_DATA;
1590                 /*
1591                  * Done with the Data-Out. Next, check if we need
1592                  * to send another unsolicited Data-Out.
1593                  */
1594                 if (ctask->unsol_count) {
1595                         debug_scsi("sending more uns\n");
1596                         tcp_ctask->xmstate |= XMSTATE_UNS_INIT;
1597                         goto send_hdr;
1598                 }
1599         }
1600         return 0;
1601 }
1602
1603 static int iscsi_send_sol_pdu(struct iscsi_conn *conn,
1604                               struct iscsi_cmd_task *ctask)
1605 {
1606         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1607         struct iscsi_session *session = conn->session;
1608         struct iscsi_r2t_info *r2t;
1609         struct iscsi_data_task *dtask;
1610         int left, rc;
1611
1612         if (tcp_ctask->xmstate & XMSTATE_SOL_HDR_INIT) {
1613                 if (!tcp_ctask->r2t) {
1614                         spin_lock_bh(&session->lock);
1615                         __kfifo_get(tcp_ctask->r2tqueue, (void*)&tcp_ctask->r2t,
1616                                     sizeof(void*));
1617                         spin_unlock_bh(&session->lock);
1618                 }
1619 send_hdr:
1620                 r2t = tcp_ctask->r2t;
1621                 dtask = &r2t->dtask;
1622
1623                 if (conn->hdrdgst_en)
1624                         iscsi_hdr_digest(conn, &r2t->headbuf,
1625                                         (u8*)dtask->hdrext);
1626                 tcp_ctask->xmstate &= ~XMSTATE_SOL_HDR_INIT;
1627                 tcp_ctask->xmstate |= XMSTATE_SOL_HDR;
1628         }
1629
1630         if (tcp_ctask->xmstate & XMSTATE_SOL_HDR) {
1631                 r2t = tcp_ctask->r2t;
1632                 dtask = &r2t->dtask;
1633
1634                 rc = iscsi_sendhdr(conn, &r2t->headbuf, r2t->data_count);
1635                 if (rc)
1636                         return rc;
1637                 tcp_ctask->xmstate &= ~XMSTATE_SOL_HDR;
1638                 tcp_ctask->xmstate |= XMSTATE_SOL_DATA;
1639
1640                 if (conn->datadgst_en) {
1641                         iscsi_data_digest_init(conn->dd_data, tcp_ctask);
1642                         dtask->digest = 0;
1643                 }
1644
1645                 iscsi_set_padding(tcp_ctask, r2t->data_count);
1646                 debug_scsi("sol dout [dsn %d itt 0x%x dlen %d sent %d]\n",
1647                         r2t->solicit_datasn - 1, ctask->itt, r2t->data_count,
1648                         r2t->sent);
1649         }
1650
1651         if (tcp_ctask->xmstate & XMSTATE_SOL_DATA) {
1652                 r2t = tcp_ctask->r2t;
1653                 dtask = &r2t->dtask;
1654
1655                 rc = iscsi_send_data(ctask, &r2t->sendbuf, &r2t->sg,
1656                                      &r2t->sent, &r2t->data_count,
1657                                      &dtask->digestbuf, &dtask->digest);
1658                 if (rc)
1659                         return rc;
1660                 tcp_ctask->xmstate &= ~XMSTATE_SOL_DATA;
1661
1662                 /*
1663                  * Done with this Data-Out. Next, check if we have
1664                  * to send another Data-Out for this R2T.
1665                  */
1666                 BUG_ON(r2t->data_length - r2t->sent < 0);
1667                 left = r2t->data_length - r2t->sent;
1668                 if (left) {
1669                         iscsi_solicit_data_cont(conn, ctask, r2t, left);
1670                         goto send_hdr;
1671                 }
1672
1673                 /*
1674                  * Done with this R2T. Check if there are more
1675                  * outstanding R2Ts ready to be processed.
1676                  */
1677                 spin_lock_bh(&session->lock);
1678                 tcp_ctask->r2t = NULL;
1679                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
1680                             sizeof(void*));
1681                 if (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t,
1682                                 sizeof(void*))) {
1683                         tcp_ctask->r2t = r2t;
1684                         spin_unlock_bh(&session->lock);
1685                         goto send_hdr;
1686                 }
1687                 spin_unlock_bh(&session->lock);
1688         }
1689         return 0;
1690 }
1691
1692 /**
1693  * iscsi_tcp_ctask_xmit - xmit normal PDU task
1694  * @conn: iscsi connection
1695  * @ctask: iscsi command task
1696  *
1697  * Notes:
1698  *      The function can return -EAGAIN in which case caller must
1699  *      call it again later, or recover. '0' return code means successful
1700  *      xmit.
1701  *      The function is devided to logical helpers (above) for the different
1702  *      xmit stages.
1703  *
1704  *iscsi_send_cmd_hdr()
1705  *      XMSTATE_CMD_HDR_INIT - prepare Header and Data buffers Calculate
1706  *                             Header Digest
1707  *      XMSTATE_CMD_HDR_XMIT - Transmit header in progress
1708  *
1709  *iscsi_send_padding
1710  *      XMSTATE_W_PAD        - Prepare and send pading
1711  *      XMSTATE_W_RESEND_PAD - retry send pading
1712  *
1713  *iscsi_send_digest
1714  *      XMSTATE_W_RESEND_DATA_DIGEST - Finalize and send Data Digest
1715  *      XMSTATE_W_RESEND_DATA_DIGEST - retry sending digest
1716  *
1717  *iscsi_send_unsol_hdr
1718  *      XMSTATE_UNS_INIT     - prepare un-solicit data header and digest
1719  *      XMSTATE_UNS_HDR      - send un-solicit header
1720  *
1721  *iscsi_send_unsol_pdu
1722  *      XMSTATE_UNS_DATA     - send un-solicit data in progress
1723  *
1724  *iscsi_send_sol_pdu
1725  *      XMSTATE_SOL_HDR_INIT - solicit data header and digest initialize
1726  *      XMSTATE_SOL_HDR      - send solicit header
1727  *      XMSTATE_SOL_DATA     - send solicit data
1728  *
1729  *iscsi_tcp_ctask_xmit
1730  *      XMSTATE_IMM_DATA     - xmit managment data (??)
1731  **/
1732 static int
1733 iscsi_tcp_ctask_xmit(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1734 {
1735         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1736         int rc = 0;
1737
1738         debug_scsi("ctask deq [cid %d xmstate %x itt 0x%x]\n",
1739                 conn->id, tcp_ctask->xmstate, ctask->itt);
1740
1741         rc = iscsi_send_cmd_hdr(conn, ctask);
1742         if (rc)
1743                 return rc;
1744         if (ctask->sc->sc_data_direction != DMA_TO_DEVICE)
1745                 return 0;
1746
1747         if (tcp_ctask->xmstate & XMSTATE_IMM_DATA) {
1748                 rc = iscsi_send_data(ctask, &tcp_ctask->sendbuf, &tcp_ctask->sg,
1749                                      &tcp_ctask->sent, &ctask->imm_count,
1750                                      &tcp_ctask->immbuf, &tcp_ctask->immdigest);
1751                 if (rc)
1752                         return rc;
1753                 tcp_ctask->xmstate &= ~XMSTATE_IMM_DATA;
1754         }
1755
1756         rc = iscsi_send_unsol_pdu(conn, ctask);
1757         if (rc)
1758                 return rc;
1759
1760         rc = iscsi_send_sol_pdu(conn, ctask);
1761         if (rc)
1762                 return rc;
1763
1764         return rc;
1765 }
1766
1767 static struct iscsi_cls_conn *
1768 iscsi_tcp_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
1769 {
1770         struct iscsi_conn *conn;
1771         struct iscsi_cls_conn *cls_conn;
1772         struct iscsi_tcp_conn *tcp_conn;
1773
1774         cls_conn = iscsi_conn_setup(cls_session, conn_idx);
1775         if (!cls_conn)
1776                 return NULL;
1777         conn = cls_conn->dd_data;
1778         /*
1779          * due to strange issues with iser these are not set
1780          * in iscsi_conn_setup
1781          */
1782         conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1783
1784         tcp_conn = kzalloc(sizeof(*tcp_conn), GFP_KERNEL);
1785         if (!tcp_conn)
1786                 goto tcp_conn_alloc_fail;
1787
1788         conn->dd_data = tcp_conn;
1789         tcp_conn->iscsi_conn = conn;
1790
1791         tcp_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1792                                                   CRYPTO_ALG_ASYNC);
1793         tcp_conn->tx_hash.flags = 0;
1794         if (IS_ERR(tcp_conn->tx_hash.tfm)) {
1795                 printk(KERN_ERR "Could not create connection due to crc32c "
1796                        "loading error %ld. Make sure the crc32c module is "
1797                        "built as a module or into the kernel\n",
1798                         PTR_ERR(tcp_conn->tx_hash.tfm));
1799                 goto free_tcp_conn;
1800         }
1801
1802         tcp_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1803                                                   CRYPTO_ALG_ASYNC);
1804         tcp_conn->rx_hash.flags = 0;
1805         if (IS_ERR(tcp_conn->rx_hash.tfm)) {
1806                 printk(KERN_ERR "Could not create connection due to crc32c "
1807                        "loading error %ld. Make sure the crc32c module is "
1808                        "built as a module or into the kernel\n",
1809                         PTR_ERR(tcp_conn->rx_hash.tfm));
1810                 goto free_tx_tfm;
1811         }
1812
1813         return cls_conn;
1814
1815 free_tx_tfm:
1816         crypto_free_hash(tcp_conn->tx_hash.tfm);
1817 free_tcp_conn:
1818         kfree(tcp_conn);
1819 tcp_conn_alloc_fail:
1820         iscsi_conn_teardown(cls_conn);
1821         return NULL;
1822 }
1823
1824 static void
1825 iscsi_tcp_release_conn(struct iscsi_conn *conn)
1826 {
1827         struct iscsi_session *session = conn->session;
1828         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1829         struct socket *sock = tcp_conn->sock;
1830
1831         if (!sock)
1832                 return;
1833
1834         sock_hold(sock->sk);
1835         iscsi_conn_restore_callbacks(tcp_conn);
1836         sock_put(sock->sk);
1837
1838         spin_lock_bh(&session->lock);
1839         tcp_conn->sock = NULL;
1840         conn->recv_lock = NULL;
1841         spin_unlock_bh(&session->lock);
1842         sockfd_put(sock);
1843 }
1844
1845 static void
1846 iscsi_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
1847 {
1848         struct iscsi_conn *conn = cls_conn->dd_data;
1849         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1850
1851         iscsi_tcp_release_conn(conn);
1852         iscsi_conn_teardown(cls_conn);
1853
1854         if (tcp_conn->tx_hash.tfm)
1855                 crypto_free_hash(tcp_conn->tx_hash.tfm);
1856         if (tcp_conn->rx_hash.tfm)
1857                 crypto_free_hash(tcp_conn->rx_hash.tfm);
1858
1859         kfree(tcp_conn);
1860 }
1861
1862 static void
1863 iscsi_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
1864 {
1865         struct iscsi_conn *conn = cls_conn->dd_data;
1866
1867         iscsi_conn_stop(cls_conn, flag);
1868         iscsi_tcp_release_conn(conn);
1869 }
1870
1871 static int iscsi_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
1872                               char *buf, int *port,
1873                               int (*getname)(struct socket *, struct sockaddr *,
1874                                         int *addrlen))
1875 {
1876         struct sockaddr_storage *addr;
1877         struct sockaddr_in6 *sin6;
1878         struct sockaddr_in *sin;
1879         int rc = 0, len;
1880
1881         addr = kmalloc(sizeof(*addr), GFP_KERNEL);
1882         if (!addr)
1883                 return -ENOMEM;
1884
1885         if (getname(sock, (struct sockaddr *) addr, &len)) {
1886                 rc = -ENODEV;
1887                 goto free_addr;
1888         }
1889
1890         switch (addr->ss_family) {
1891         case AF_INET:
1892                 sin = (struct sockaddr_in *)addr;
1893                 spin_lock_bh(&conn->session->lock);
1894                 sprintf(buf, NIPQUAD_FMT, NIPQUAD(sin->sin_addr.s_addr));
1895                 *port = be16_to_cpu(sin->sin_port);
1896                 spin_unlock_bh(&conn->session->lock);
1897                 break;
1898         case AF_INET6:
1899                 sin6 = (struct sockaddr_in6 *)addr;
1900                 spin_lock_bh(&conn->session->lock);
1901                 sprintf(buf, NIP6_FMT, NIP6(sin6->sin6_addr));
1902                 *port = be16_to_cpu(sin6->sin6_port);
1903                 spin_unlock_bh(&conn->session->lock);
1904                 break;
1905         }
1906 free_addr:
1907         kfree(addr);
1908         return rc;
1909 }
1910
1911 static int
1912 iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
1913                     struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
1914                     int is_leading)
1915 {
1916         struct iscsi_conn *conn = cls_conn->dd_data;
1917         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1918         struct sock *sk;
1919         struct socket *sock;
1920         int err;
1921
1922         /* lookup for existing socket */
1923         sock = sockfd_lookup((int)transport_eph, &err);
1924         if (!sock) {
1925                 printk(KERN_ERR "iscsi_tcp: sockfd_lookup failed %d\n", err);
1926                 return -EEXIST;
1927         }
1928         /*
1929          * copy these values now because if we drop the session
1930          * userspace may still want to query the values since we will
1931          * be using them for the reconnect
1932          */
1933         err = iscsi_tcp_get_addr(conn, sock, conn->portal_address,
1934                                  &conn->portal_port, kernel_getpeername);
1935         if (err)
1936                 goto free_socket;
1937
1938         err = iscsi_tcp_get_addr(conn, sock, conn->local_address,
1939                                 &conn->local_port, kernel_getsockname);
1940         if (err)
1941                 goto free_socket;
1942
1943         err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
1944         if (err)
1945                 goto free_socket;
1946
1947         /* bind iSCSI connection and socket */
1948         tcp_conn->sock = sock;
1949
1950         /* setup Socket parameters */
1951         sk = sock->sk;
1952         sk->sk_reuse = 1;
1953         sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
1954         sk->sk_allocation = GFP_ATOMIC;
1955
1956         /* FIXME: disable Nagle's algorithm */
1957
1958         /*
1959          * Intercept TCP callbacks for sendfile like receive
1960          * processing.
1961          */
1962         conn->recv_lock = &sk->sk_callback_lock;
1963         iscsi_conn_set_callbacks(conn);
1964         tcp_conn->sendpage = tcp_conn->sock->ops->sendpage;
1965         /*
1966          * set receive state machine into initial state
1967          */
1968         iscsi_tcp_hdr_recv_prep(tcp_conn);
1969         return 0;
1970
1971 free_socket:
1972         sockfd_put(sock);
1973         return err;
1974 }
1975
1976 /* called with host lock */
1977 static void
1978 iscsi_tcp_mgmt_init(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1979 {
1980         struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
1981         tcp_mtask->xmstate = XMSTATE_IMM_HDR_INIT;
1982 }
1983
1984 static int
1985 iscsi_r2tpool_alloc(struct iscsi_session *session)
1986 {
1987         int i;
1988         int cmd_i;
1989
1990         /*
1991          * initialize per-task: R2T pool and xmit queue
1992          */
1993         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1994                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
1995                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1996
1997                 /*
1998                  * pre-allocated x4 as much r2ts to handle race when
1999                  * target acks DataOut faster than we data_xmit() queues
2000                  * could replenish r2tqueue.
2001                  */
2002
2003                 /* R2T pool */
2004                 if (iscsi_pool_init(&tcp_ctask->r2tpool, session->max_r2t * 4,
2005                                     (void***)&tcp_ctask->r2ts,
2006                                     sizeof(struct iscsi_r2t_info))) {
2007                         goto r2t_alloc_fail;
2008                 }
2009
2010                 /* R2T xmit queue */
2011                 tcp_ctask->r2tqueue = kfifo_alloc(
2012                       session->max_r2t * 4 * sizeof(void*), GFP_KERNEL, NULL);
2013                 if (tcp_ctask->r2tqueue == ERR_PTR(-ENOMEM)) {
2014                         iscsi_pool_free(&tcp_ctask->r2tpool,
2015                                         (void**)tcp_ctask->r2ts);
2016                         goto r2t_alloc_fail;
2017                 }
2018         }
2019
2020         return 0;
2021
2022 r2t_alloc_fail:
2023         for (i = 0; i < cmd_i; i++) {
2024                 struct iscsi_cmd_task *ctask = session->cmds[i];
2025                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2026
2027                 kfifo_free(tcp_ctask->r2tqueue);
2028                 iscsi_pool_free(&tcp_ctask->r2tpool,
2029                                 (void**)tcp_ctask->r2ts);
2030         }
2031         return -ENOMEM;
2032 }
2033
2034 static void
2035 iscsi_r2tpool_free(struct iscsi_session *session)
2036 {
2037         int i;
2038
2039         for (i = 0; i < session->cmds_max; i++) {
2040                 struct iscsi_cmd_task *ctask = session->cmds[i];
2041                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2042
2043                 kfifo_free(tcp_ctask->r2tqueue);
2044                 iscsi_pool_free(&tcp_ctask->r2tpool,
2045                                 (void**)tcp_ctask->r2ts);
2046         }
2047 }
2048
2049 static int
2050 iscsi_conn_set_param(struct iscsi_cls_conn *cls_conn, enum iscsi_param param,
2051                      char *buf, int buflen)
2052 {
2053         struct iscsi_conn *conn = cls_conn->dd_data;
2054         struct iscsi_session *session = conn->session;
2055         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2056         int value;
2057
2058         switch(param) {
2059         case ISCSI_PARAM_HDRDGST_EN:
2060                 iscsi_set_param(cls_conn, param, buf, buflen);
2061                 break;
2062         case ISCSI_PARAM_DATADGST_EN:
2063                 iscsi_set_param(cls_conn, param, buf, buflen);
2064                 tcp_conn->sendpage = conn->datadgst_en ?
2065                         sock_no_sendpage : tcp_conn->sock->ops->sendpage;
2066                 break;
2067         case ISCSI_PARAM_MAX_R2T:
2068                 sscanf(buf, "%d", &value);
2069                 if (session->max_r2t == roundup_pow_of_two(value))
2070                         break;
2071                 iscsi_r2tpool_free(session);
2072                 iscsi_set_param(cls_conn, param, buf, buflen);
2073                 if (session->max_r2t & (session->max_r2t - 1))
2074                         session->max_r2t = roundup_pow_of_two(session->max_r2t);
2075                 if (iscsi_r2tpool_alloc(session))
2076                         return -ENOMEM;
2077                 break;
2078         default:
2079                 return iscsi_set_param(cls_conn, param, buf, buflen);
2080         }
2081
2082         return 0;
2083 }
2084
2085 static int
2086 iscsi_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
2087                          enum iscsi_param param, char *buf)
2088 {
2089         struct iscsi_conn *conn = cls_conn->dd_data;
2090         int len;
2091
2092         switch(param) {
2093         case ISCSI_PARAM_CONN_PORT:
2094                 spin_lock_bh(&conn->session->lock);
2095                 len = sprintf(buf, "%hu\n", conn->portal_port);
2096                 spin_unlock_bh(&conn->session->lock);
2097                 break;
2098         case ISCSI_PARAM_CONN_ADDRESS:
2099                 spin_lock_bh(&conn->session->lock);
2100                 len = sprintf(buf, "%s\n", conn->portal_address);
2101                 spin_unlock_bh(&conn->session->lock);
2102                 break;
2103         default:
2104                 return iscsi_conn_get_param(cls_conn, param, buf);
2105         }
2106
2107         return len;
2108 }
2109
2110 static int
2111 iscsi_tcp_host_get_param(struct Scsi_Host *shost, enum iscsi_host_param param,
2112                          char *buf)
2113 {
2114         struct iscsi_session *session = iscsi_hostdata(shost->hostdata);
2115         int len;
2116
2117         switch (param) {
2118         case ISCSI_HOST_PARAM_IPADDRESS:
2119                 spin_lock_bh(&session->lock);
2120                 if (!session->leadconn)
2121                         len = -ENODEV;
2122                 else
2123                         len = sprintf(buf, "%s\n",
2124                                      session->leadconn->local_address);
2125                 spin_unlock_bh(&session->lock);
2126                 break;
2127         default:
2128                 return iscsi_host_get_param(shost, param, buf);
2129         }
2130         return len;
2131 }
2132
2133 static void
2134 iscsi_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
2135 {
2136         struct iscsi_conn *conn = cls_conn->dd_data;
2137         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2138
2139         stats->txdata_octets = conn->txdata_octets;
2140         stats->rxdata_octets = conn->rxdata_octets;
2141         stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
2142         stats->dataout_pdus = conn->dataout_pdus_cnt;
2143         stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
2144         stats->datain_pdus = conn->datain_pdus_cnt;
2145         stats->r2t_pdus = conn->r2t_pdus_cnt;
2146         stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
2147         stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
2148         stats->custom_length = 3;
2149         strcpy(stats->custom[0].desc, "tx_sendpage_failures");
2150         stats->custom[0].value = tcp_conn->sendpage_failures_cnt;
2151         strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
2152         stats->custom[1].value = tcp_conn->discontiguous_hdr_cnt;
2153         strcpy(stats->custom[2].desc, "eh_abort_cnt");
2154         stats->custom[2].value = conn->eh_abort_cnt;
2155 }
2156
2157 static struct iscsi_cls_session *
2158 iscsi_tcp_session_create(struct iscsi_transport *iscsit,
2159                          struct scsi_transport_template *scsit,
2160                          uint16_t cmds_max, uint16_t qdepth,
2161                          uint32_t initial_cmdsn, uint32_t *hostno)
2162 {
2163         struct iscsi_cls_session *cls_session;
2164         struct iscsi_session *session;
2165         uint32_t hn;
2166         int cmd_i;
2167
2168         cls_session = iscsi_session_setup(iscsit, scsit, cmds_max, qdepth,
2169                                          sizeof(struct iscsi_tcp_cmd_task),
2170                                          sizeof(struct iscsi_tcp_mgmt_task),
2171                                          initial_cmdsn, &hn);
2172         if (!cls_session)
2173                 return NULL;
2174         *hostno = hn;
2175
2176         session = class_to_transport_session(cls_session);
2177         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
2178                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
2179                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2180
2181                 ctask->hdr = &tcp_ctask->hdr;
2182         }
2183
2184         for (cmd_i = 0; cmd_i < session->mgmtpool_max; cmd_i++) {
2185                 struct iscsi_mgmt_task *mtask = session->mgmt_cmds[cmd_i];
2186                 struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
2187
2188                 mtask->hdr = &tcp_mtask->hdr;
2189         }
2190
2191         if (iscsi_r2tpool_alloc(class_to_transport_session(cls_session)))
2192                 goto r2tpool_alloc_fail;
2193
2194         return cls_session;
2195
2196 r2tpool_alloc_fail:
2197         iscsi_session_teardown(cls_session);
2198         return NULL;
2199 }
2200
2201 static void iscsi_tcp_session_destroy(struct iscsi_cls_session *cls_session)
2202 {
2203         iscsi_r2tpool_free(class_to_transport_session(cls_session));
2204         iscsi_session_teardown(cls_session);
2205 }
2206
2207 static int iscsi_tcp_slave_configure(struct scsi_device *sdev)
2208 {
2209         blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
2210         blk_queue_dma_alignment(sdev->request_queue, 0);
2211         return 0;
2212 }
2213
2214 static struct scsi_host_template iscsi_sht = {
2215         .module                 = THIS_MODULE,
2216         .name                   = "iSCSI Initiator over TCP/IP",
2217         .queuecommand           = iscsi_queuecommand,
2218         .change_queue_depth     = iscsi_change_queue_depth,
2219         .can_queue              = ISCSI_DEF_XMIT_CMDS_MAX - 1,
2220         .sg_tablesize           = ISCSI_SG_TABLESIZE,
2221         .max_sectors            = 0xFFFF,
2222         .cmd_per_lun            = ISCSI_DEF_CMD_PER_LUN,
2223         .eh_abort_handler       = iscsi_eh_abort,
2224         .eh_device_reset_handler= iscsi_eh_device_reset,
2225         .eh_host_reset_handler  = iscsi_eh_host_reset,
2226         .use_clustering         = DISABLE_CLUSTERING,
2227         .slave_configure        = iscsi_tcp_slave_configure,
2228         .proc_name              = "iscsi_tcp",
2229         .this_id                = -1,
2230 };
2231
2232 static struct iscsi_transport iscsi_tcp_transport = {
2233         .owner                  = THIS_MODULE,
2234         .name                   = "tcp",
2235         .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
2236                                   | CAP_DATADGST,
2237         .param_mask             = ISCSI_MAX_RECV_DLENGTH |
2238                                   ISCSI_MAX_XMIT_DLENGTH |
2239                                   ISCSI_HDRDGST_EN |
2240                                   ISCSI_DATADGST_EN |
2241                                   ISCSI_INITIAL_R2T_EN |
2242                                   ISCSI_MAX_R2T |
2243                                   ISCSI_IMM_DATA_EN |
2244                                   ISCSI_FIRST_BURST |
2245                                   ISCSI_MAX_BURST |
2246                                   ISCSI_PDU_INORDER_EN |
2247                                   ISCSI_DATASEQ_INORDER_EN |
2248                                   ISCSI_ERL |
2249                                   ISCSI_CONN_PORT |
2250                                   ISCSI_CONN_ADDRESS |
2251                                   ISCSI_EXP_STATSN |
2252                                   ISCSI_PERSISTENT_PORT |
2253                                   ISCSI_PERSISTENT_ADDRESS |
2254                                   ISCSI_TARGET_NAME | ISCSI_TPGT |
2255                                   ISCSI_USERNAME | ISCSI_PASSWORD |
2256                                   ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
2257                                   ISCSI_FAST_ABORT,
2258         .host_param_mask        = ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
2259                                   ISCSI_HOST_INITIATOR_NAME |
2260                                   ISCSI_HOST_NETDEV_NAME,
2261         .host_template          = &iscsi_sht,
2262         .conndata_size          = sizeof(struct iscsi_conn),
2263         .max_conn               = 1,
2264         .max_cmd_len            = ISCSI_TCP_MAX_CMD_LEN,
2265         /* session management */
2266         .create_session         = iscsi_tcp_session_create,
2267         .destroy_session        = iscsi_tcp_session_destroy,
2268         /* connection management */
2269         .create_conn            = iscsi_tcp_conn_create,
2270         .bind_conn              = iscsi_tcp_conn_bind,
2271         .destroy_conn           = iscsi_tcp_conn_destroy,
2272         .set_param              = iscsi_conn_set_param,
2273         .get_conn_param         = iscsi_tcp_conn_get_param,
2274         .get_session_param      = iscsi_session_get_param,
2275         .start_conn             = iscsi_conn_start,
2276         .stop_conn              = iscsi_tcp_conn_stop,
2277         /* iscsi host params */
2278         .get_host_param         = iscsi_tcp_host_get_param,
2279         .set_host_param         = iscsi_host_set_param,
2280         /* IO */
2281         .send_pdu               = iscsi_conn_send_pdu,
2282         .get_stats              = iscsi_conn_get_stats,
2283         .init_cmd_task          = iscsi_tcp_cmd_init,
2284         .init_mgmt_task         = iscsi_tcp_mgmt_init,
2285         .xmit_cmd_task          = iscsi_tcp_ctask_xmit,
2286         .xmit_mgmt_task         = iscsi_tcp_mtask_xmit,
2287         .cleanup_cmd_task       = iscsi_tcp_cleanup_ctask,
2288         /* recovery */
2289         .session_recovery_timedout = iscsi_session_recovery_timedout,
2290 };
2291
2292 static int __init
2293 iscsi_tcp_init(void)
2294 {
2295         if (iscsi_max_lun < 1) {
2296                 printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
2297                        iscsi_max_lun);
2298                 return -EINVAL;
2299         }
2300         iscsi_tcp_transport.max_lun = iscsi_max_lun;
2301
2302         if (!iscsi_register_transport(&iscsi_tcp_transport))
2303                 return -ENODEV;
2304
2305         return 0;
2306 }
2307
2308 static void __exit
2309 iscsi_tcp_exit(void)
2310 {
2311         iscsi_unregister_transport(&iscsi_tcp_transport);
2312 }
2313
2314 module_init(iscsi_tcp_init);
2315 module_exit(iscsi_tcp_exit);