Merge git://git.infradead.org/battery-2.6
[linux-2.6] / 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 /* #define 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 inline void
71 iscsi_buf_init_iov(struct iscsi_buf *ibuf, char *vbuf, int size)
72 {
73         ibuf->sg.page = virt_to_page(vbuf);
74         ibuf->sg.offset = offset_in_page(vbuf);
75         ibuf->sg.length = size;
76         ibuf->sent = 0;
77         ibuf->use_sendmsg = 1;
78 }
79
80 static inline void
81 iscsi_buf_init_sg(struct iscsi_buf *ibuf, struct scatterlist *sg)
82 {
83         ibuf->sg.page = sg->page;
84         ibuf->sg.offset = sg->offset;
85         ibuf->sg.length = sg->length;
86         /*
87          * Fastpath: sg element fits into single page
88          */
89         if (sg->length + sg->offset <= PAGE_SIZE && !PageSlab(sg->page))
90                 ibuf->use_sendmsg = 0;
91         else
92                 ibuf->use_sendmsg = 1;
93         ibuf->sent = 0;
94 }
95
96 static inline int
97 iscsi_buf_left(struct iscsi_buf *ibuf)
98 {
99         int rc;
100
101         rc = ibuf->sg.length - ibuf->sent;
102         BUG_ON(rc < 0);
103         return rc;
104 }
105
106 static inline void
107 iscsi_hdr_digest(struct iscsi_conn *conn, struct iscsi_buf *buf,
108                  u8* crc)
109 {
110         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
111
112         crypto_hash_digest(&tcp_conn->tx_hash, &buf->sg, buf->sg.length, crc);
113         buf->sg.length += sizeof(u32);
114 }
115
116 static inline int
117 iscsi_hdr_extract(struct iscsi_tcp_conn *tcp_conn)
118 {
119         struct sk_buff *skb = tcp_conn->in.skb;
120
121         tcp_conn->in.zero_copy_hdr = 0;
122
123         if (tcp_conn->in.copy >= tcp_conn->hdr_size &&
124             tcp_conn->in_progress == IN_PROGRESS_WAIT_HEADER) {
125                 /*
126                  * Zero-copy PDU Header: using connection context
127                  * to store header pointer.
128                  */
129                 if (skb_shinfo(skb)->frag_list == NULL &&
130                     !skb_shinfo(skb)->nr_frags) {
131                         tcp_conn->in.hdr = (struct iscsi_hdr *)
132                                 ((char*)skb->data + tcp_conn->in.offset);
133                         tcp_conn->in.zero_copy_hdr = 1;
134                 } else {
135                         /* ignoring return code since we checked
136                          * in.copy before */
137                         skb_copy_bits(skb, tcp_conn->in.offset,
138                                 &tcp_conn->hdr, tcp_conn->hdr_size);
139                         tcp_conn->in.hdr = &tcp_conn->hdr;
140                 }
141                 tcp_conn->in.offset += tcp_conn->hdr_size;
142                 tcp_conn->in.copy -= tcp_conn->hdr_size;
143         } else {
144                 int hdr_remains;
145                 int copylen;
146
147                 /*
148                  * PDU header scattered across SKB's,
149                  * copying it... This'll happen quite rarely.
150                  */
151
152                 if (tcp_conn->in_progress == IN_PROGRESS_WAIT_HEADER)
153                         tcp_conn->in.hdr_offset = 0;
154
155                 hdr_remains = tcp_conn->hdr_size - tcp_conn->in.hdr_offset;
156                 BUG_ON(hdr_remains <= 0);
157
158                 copylen = min(tcp_conn->in.copy, hdr_remains);
159                 skb_copy_bits(skb, tcp_conn->in.offset,
160                         (char*)&tcp_conn->hdr + tcp_conn->in.hdr_offset,
161                         copylen);
162
163                 debug_tcp("PDU gather offset %d bytes %d in.offset %d "
164                        "in.copy %d\n", tcp_conn->in.hdr_offset, copylen,
165                        tcp_conn->in.offset, tcp_conn->in.copy);
166
167                 tcp_conn->in.offset += copylen;
168                 tcp_conn->in.copy -= copylen;
169                 if (copylen < hdr_remains)  {
170                         tcp_conn->in_progress = IN_PROGRESS_HEADER_GATHER;
171                         tcp_conn->in.hdr_offset += copylen;
172                         return -EAGAIN;
173                 }
174                 tcp_conn->in.hdr = &tcp_conn->hdr;
175                 tcp_conn->discontiguous_hdr_cnt++;
176                 tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
177         }
178
179         return 0;
180 }
181
182 /*
183  * must be called with session lock
184  */
185 static void
186 iscsi_tcp_cleanup_ctask(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
187 {
188         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
189         struct iscsi_r2t_info *r2t;
190         struct scsi_cmnd *sc;
191
192         /* flush ctask's r2t queues */
193         while (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*))) {
194                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
195                             sizeof(void*));
196                 debug_scsi("iscsi_tcp_cleanup_ctask pending r2t dropped\n");
197         }
198
199         sc = ctask->sc;
200         if (unlikely(!sc))
201                 return;
202
203         tcp_ctask->xmstate = XMSTATE_IDLE;
204         tcp_ctask->r2t = NULL;
205 }
206
207 /**
208  * iscsi_data_rsp - SCSI Data-In Response processing
209  * @conn: iscsi connection
210  * @ctask: scsi command task
211  **/
212 static int
213 iscsi_data_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
214 {
215         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
216         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
217         struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
218         struct iscsi_session *session = conn->session;
219         struct scsi_cmnd *sc = ctask->sc;
220         int datasn = be32_to_cpu(rhdr->datasn);
221
222         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
223         /*
224          * setup Data-In byte counter (gets decremented..)
225          */
226         ctask->data_count = tcp_conn->in.datalen;
227
228         if (tcp_conn->in.datalen == 0)
229                 return 0;
230
231         if (tcp_ctask->exp_datasn != datasn) {
232                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->datasn(%d)\n",
233                           __FUNCTION__, tcp_ctask->exp_datasn, datasn);
234                 return ISCSI_ERR_DATASN;
235         }
236
237         tcp_ctask->exp_datasn++;
238
239         tcp_ctask->data_offset = be32_to_cpu(rhdr->offset);
240         if (tcp_ctask->data_offset + tcp_conn->in.datalen > scsi_bufflen(sc)) {
241                 debug_tcp("%s: data_offset(%d) + data_len(%d) > total_length_in(%d)\n",
242                           __FUNCTION__, tcp_ctask->data_offset,
243                           tcp_conn->in.datalen, scsi_bufflen(sc));
244                 return ISCSI_ERR_DATA_OFFSET;
245         }
246
247         if (rhdr->flags & ISCSI_FLAG_DATA_STATUS) {
248                 conn->exp_statsn = be32_to_cpu(rhdr->statsn) + 1;
249                 if (rhdr->flags & ISCSI_FLAG_DATA_UNDERFLOW) {
250                         int res_count = be32_to_cpu(rhdr->residual_count);
251
252                         if (res_count > 0 &&
253                             res_count <= scsi_bufflen(sc)) {
254                                 scsi_set_resid(sc, res_count);
255                                 sc->result = (DID_OK << 16) | rhdr->cmd_status;
256                         } else
257                                 sc->result = (DID_BAD_TARGET << 16) |
258                                         rhdr->cmd_status;
259                 } else if (rhdr->flags & ISCSI_FLAG_DATA_OVERFLOW) {
260                         scsi_set_resid(sc, be32_to_cpu(rhdr->residual_count));
261                         sc->result = (DID_OK << 16) | rhdr->cmd_status;
262                 } else
263                         sc->result = (DID_OK << 16) | rhdr->cmd_status;
264         }
265
266         conn->datain_pdus_cnt++;
267         return 0;
268 }
269
270 /**
271  * iscsi_solicit_data_init - initialize first Data-Out
272  * @conn: iscsi connection
273  * @ctask: scsi command task
274  * @r2t: R2T info
275  *
276  * Notes:
277  *      Initialize first Data-Out within this R2T sequence and finds
278  *      proper data_offset within this SCSI command.
279  *
280  *      This function is called with connection lock taken.
281  **/
282 static void
283 iscsi_solicit_data_init(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
284                         struct iscsi_r2t_info *r2t)
285 {
286         struct iscsi_data *hdr;
287         struct scsi_cmnd *sc = ctask->sc;
288         int i, sg_count = 0;
289         struct scatterlist *sg;
290
291         hdr = &r2t->dtask.hdr;
292         memset(hdr, 0, sizeof(struct iscsi_data));
293         hdr->ttt = r2t->ttt;
294         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
295         r2t->solicit_datasn++;
296         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
297         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
298         hdr->itt = ctask->hdr->itt;
299         hdr->exp_statsn = r2t->exp_statsn;
300         hdr->offset = cpu_to_be32(r2t->data_offset);
301         if (r2t->data_length > conn->max_xmit_dlength) {
302                 hton24(hdr->dlength, conn->max_xmit_dlength);
303                 r2t->data_count = conn->max_xmit_dlength;
304                 hdr->flags = 0;
305         } else {
306                 hton24(hdr->dlength, r2t->data_length);
307                 r2t->data_count = r2t->data_length;
308                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
309         }
310         conn->dataout_pdus_cnt++;
311
312         r2t->sent = 0;
313
314         iscsi_buf_init_iov(&r2t->headbuf, (char*)hdr,
315                            sizeof(struct iscsi_hdr));
316
317         sg = scsi_sglist(sc);
318         r2t->sg = NULL;
319         for (i = 0; i < scsi_sg_count(sc); i++, sg += 1) {
320                 /* FIXME: prefetch ? */
321                 if (sg_count + sg->length > r2t->data_offset) {
322                         int page_offset;
323
324                         /* sg page found! */
325
326                         /* offset within this page */
327                         page_offset = r2t->data_offset - sg_count;
328
329                         /* fill in this buffer */
330                         iscsi_buf_init_sg(&r2t->sendbuf, sg);
331                         r2t->sendbuf.sg.offset += page_offset;
332                         r2t->sendbuf.sg.length -= page_offset;
333
334                         /* xmit logic will continue with next one */
335                         r2t->sg = sg + 1;
336                         break;
337                 }
338                 sg_count += sg->length;
339         }
340         BUG_ON(r2t->sg == NULL);
341 }
342
343 /**
344  * iscsi_r2t_rsp - iSCSI R2T Response processing
345  * @conn: iscsi connection
346  * @ctask: scsi command task
347  **/
348 static int
349 iscsi_r2t_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
350 {
351         struct iscsi_r2t_info *r2t;
352         struct iscsi_session *session = conn->session;
353         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
354         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
355         struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
356         int r2tsn = be32_to_cpu(rhdr->r2tsn);
357         int rc;
358
359         if (tcp_conn->in.datalen) {
360                 printk(KERN_ERR "iscsi_tcp: invalid R2t with datalen %d\n",
361                        tcp_conn->in.datalen);
362                 return ISCSI_ERR_DATALEN;
363         }
364
365         if (tcp_ctask->exp_datasn != r2tsn){
366                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
367                           __FUNCTION__, tcp_ctask->exp_datasn, r2tsn);
368                 return ISCSI_ERR_R2TSN;
369         }
370
371         /* fill-in new R2T associated with the task */
372         spin_lock(&session->lock);
373         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
374
375         if (!ctask->sc || ctask->mtask ||
376              session->state != ISCSI_STATE_LOGGED_IN) {
377                 printk(KERN_INFO "iscsi_tcp: dropping R2T itt %d in "
378                        "recovery...\n", ctask->itt);
379                 spin_unlock(&session->lock);
380                 return 0;
381         }
382
383         rc = __kfifo_get(tcp_ctask->r2tpool.queue, (void*)&r2t, sizeof(void*));
384         BUG_ON(!rc);
385
386         r2t->exp_statsn = rhdr->statsn;
387         r2t->data_length = be32_to_cpu(rhdr->data_length);
388         if (r2t->data_length == 0) {
389                 printk(KERN_ERR "iscsi_tcp: invalid R2T with zero data len\n");
390                 spin_unlock(&session->lock);
391                 return ISCSI_ERR_DATALEN;
392         }
393
394         if (r2t->data_length > session->max_burst)
395                 debug_scsi("invalid R2T with data len %u and max burst %u."
396                            "Attempting to execute request.\n",
397                             r2t->data_length, session->max_burst);
398
399         r2t->data_offset = be32_to_cpu(rhdr->data_offset);
400         if (r2t->data_offset + r2t->data_length > scsi_bufflen(ctask->sc)) {
401                 spin_unlock(&session->lock);
402                 printk(KERN_ERR "iscsi_tcp: invalid R2T with data len %u at "
403                        "offset %u and total length %d\n", r2t->data_length,
404                        r2t->data_offset, scsi_bufflen(ctask->sc));
405                 return ISCSI_ERR_DATALEN;
406         }
407
408         r2t->ttt = rhdr->ttt; /* no flip */
409         r2t->solicit_datasn = 0;
410
411         iscsi_solicit_data_init(conn, ctask, r2t);
412
413         tcp_ctask->exp_datasn = r2tsn + 1;
414         __kfifo_put(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*));
415         tcp_ctask->xmstate |= XMSTATE_SOL_HDR_INIT;
416         list_move_tail(&ctask->running, &conn->xmitqueue);
417
418         scsi_queue_work(session->host, &conn->xmitwork);
419         conn->r2t_pdus_cnt++;
420         spin_unlock(&session->lock);
421
422         return 0;
423 }
424
425 static int
426 iscsi_tcp_hdr_recv(struct iscsi_conn *conn)
427 {
428         int rc = 0, opcode, ahslen;
429         struct iscsi_hdr *hdr;
430         struct iscsi_session *session = conn->session;
431         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
432         uint32_t cdgst, rdgst = 0, itt;
433
434         hdr = tcp_conn->in.hdr;
435
436         /* verify PDU length */
437         tcp_conn->in.datalen = ntoh24(hdr->dlength);
438         if (tcp_conn->in.datalen > conn->max_recv_dlength) {
439                 printk(KERN_ERR "iscsi_tcp: datalen %d > %d\n",
440                        tcp_conn->in.datalen, conn->max_recv_dlength);
441                 return ISCSI_ERR_DATALEN;
442         }
443         tcp_conn->data_copied = 0;
444
445         /* read AHS */
446         ahslen = hdr->hlength << 2;
447         tcp_conn->in.offset += ahslen;
448         tcp_conn->in.copy -= ahslen;
449         if (tcp_conn->in.copy < 0) {
450                 printk(KERN_ERR "iscsi_tcp: can't handle AHS with length "
451                        "%d bytes\n", ahslen);
452                 return ISCSI_ERR_AHSLEN;
453         }
454
455         /* calculate read padding */
456         tcp_conn->in.padding = tcp_conn->in.datalen & (ISCSI_PAD_LEN-1);
457         if (tcp_conn->in.padding) {
458                 tcp_conn->in.padding = ISCSI_PAD_LEN - tcp_conn->in.padding;
459                 debug_scsi("read padding %d bytes\n", tcp_conn->in.padding);
460         }
461
462         if (conn->hdrdgst_en) {
463                 struct scatterlist sg;
464
465                 sg_init_one(&sg, (u8 *)hdr,
466                             sizeof(struct iscsi_hdr) + ahslen);
467                 crypto_hash_digest(&tcp_conn->rx_hash, &sg, sg.length,
468                                    (u8 *)&cdgst);
469                 rdgst = *(uint32_t*)((char*)hdr + sizeof(struct iscsi_hdr) +
470                                      ahslen);
471                 if (cdgst != rdgst) {
472                         printk(KERN_ERR "iscsi_tcp: hdrdgst error "
473                                "recv 0x%x calc 0x%x\n", rdgst, cdgst);
474                         return ISCSI_ERR_HDR_DGST;
475                 }
476         }
477
478         opcode = hdr->opcode & ISCSI_OPCODE_MASK;
479         /* verify itt (itt encoding: age+cid+itt) */
480         rc = iscsi_verify_itt(conn, hdr, &itt);
481         if (rc == ISCSI_ERR_NO_SCSI_CMD) {
482                 tcp_conn->in.datalen = 0; /* force drop */
483                 return 0;
484         } else if (rc)
485                 return rc;
486
487         debug_tcp("opcode 0x%x offset %d copy %d ahslen %d datalen %d\n",
488                   opcode, tcp_conn->in.offset, tcp_conn->in.copy,
489                   ahslen, tcp_conn->in.datalen);
490
491         switch(opcode) {
492         case ISCSI_OP_SCSI_DATA_IN:
493                 tcp_conn->in.ctask = session->cmds[itt];
494                 rc = iscsi_data_rsp(conn, tcp_conn->in.ctask);
495                 if (rc)
496                         return rc;
497                 /* fall through */
498         case ISCSI_OP_SCSI_CMD_RSP:
499                 tcp_conn->in.ctask = session->cmds[itt];
500                 if (tcp_conn->in.datalen)
501                         goto copy_hdr;
502
503                 spin_lock(&session->lock);
504                 rc = __iscsi_complete_pdu(conn, hdr, NULL, 0);
505                 spin_unlock(&session->lock);
506                 break;
507         case ISCSI_OP_R2T:
508                 tcp_conn->in.ctask = session->cmds[itt];
509                 if (ahslen)
510                         rc = ISCSI_ERR_AHSLEN;
511                 else if (tcp_conn->in.ctask->sc->sc_data_direction ==
512                                                                 DMA_TO_DEVICE)
513                         rc = iscsi_r2t_rsp(conn, tcp_conn->in.ctask);
514                 else
515                         rc = ISCSI_ERR_PROTO;
516                 break;
517         case ISCSI_OP_LOGIN_RSP:
518         case ISCSI_OP_TEXT_RSP:
519         case ISCSI_OP_REJECT:
520         case ISCSI_OP_ASYNC_EVENT:
521                 /*
522                  * It is possible that we could get a PDU with a buffer larger
523                  * than 8K, but there are no targets that currently do this.
524                  * For now we fail until we find a vendor that needs it
525                  */
526                 if (ISCSI_DEF_MAX_RECV_SEG_LEN <
527                     tcp_conn->in.datalen) {
528                         printk(KERN_ERR "iscsi_tcp: received buffer of len %u "
529                               "but conn buffer is only %u (opcode %0x)\n",
530                               tcp_conn->in.datalen,
531                               ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
532                         rc = ISCSI_ERR_PROTO;
533                         break;
534                 }
535
536                 if (tcp_conn->in.datalen)
537                         goto copy_hdr;
538         /* fall through */
539         case ISCSI_OP_LOGOUT_RSP:
540         case ISCSI_OP_NOOP_IN:
541         case ISCSI_OP_SCSI_TMFUNC_RSP:
542                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
543                 break;
544         default:
545                 rc = ISCSI_ERR_BAD_OPCODE;
546                 break;
547         }
548
549         return rc;
550
551 copy_hdr:
552         /*
553          * if we did zero copy for the header but we will need multiple
554          * skbs to complete the command then we have to copy the header
555          * for later use
556          */
557         if (tcp_conn->in.zero_copy_hdr && tcp_conn->in.copy <=
558            (tcp_conn->in.datalen + tcp_conn->in.padding +
559             (conn->datadgst_en ? 4 : 0))) {
560                 debug_tcp("Copying header for later use. in.copy %d in.datalen"
561                           " %d\n", tcp_conn->in.copy, tcp_conn->in.datalen);
562                 memcpy(&tcp_conn->hdr, tcp_conn->in.hdr,
563                        sizeof(struct iscsi_hdr));
564                 tcp_conn->in.hdr = &tcp_conn->hdr;
565                 tcp_conn->in.zero_copy_hdr = 0;
566         }
567         return 0;
568 }
569
570 /**
571  * iscsi_ctask_copy - copy skb bits to the destanation cmd task
572  * @conn: iscsi tcp connection
573  * @ctask: scsi command task
574  * @buf: buffer to copy to
575  * @buf_size: size of buffer
576  * @offset: offset within the buffer
577  *
578  * Notes:
579  *      The function calls skb_copy_bits() and updates per-connection and
580  *      per-cmd byte counters.
581  *
582  *      Read counters (in bytes):
583  *
584  *      conn->in.offset         offset within in progress SKB
585  *      conn->in.copy           left to copy from in progress SKB
586  *                              including padding
587  *      conn->in.copied         copied already from in progress SKB
588  *      conn->data_copied       copied already from in progress buffer
589  *      ctask->sent             total bytes sent up to the MidLayer
590  *      ctask->data_count       left to copy from in progress Data-In
591  *      buf_left                left to copy from in progress buffer
592  **/
593 static inline int
594 iscsi_ctask_copy(struct iscsi_tcp_conn *tcp_conn, struct iscsi_cmd_task *ctask,
595                 void *buf, int buf_size, int offset)
596 {
597         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
598         int buf_left = buf_size - (tcp_conn->data_copied + offset);
599         unsigned size = min(tcp_conn->in.copy, buf_left);
600         int rc;
601
602         size = min(size, ctask->data_count);
603
604         debug_tcp("ctask_copy %d bytes at offset %d copied %d\n",
605                size, tcp_conn->in.offset, tcp_conn->in.copied);
606
607         BUG_ON(size <= 0);
608         BUG_ON(tcp_ctask->sent + size > scsi_bufflen(ctask->sc));
609
610         rc = skb_copy_bits(tcp_conn->in.skb, tcp_conn->in.offset,
611                            (char*)buf + (offset + tcp_conn->data_copied), size);
612         /* must fit into skb->len */
613         BUG_ON(rc);
614
615         tcp_conn->in.offset += size;
616         tcp_conn->in.copy -= size;
617         tcp_conn->in.copied += size;
618         tcp_conn->data_copied += size;
619         tcp_ctask->sent += size;
620         ctask->data_count -= size;
621
622         BUG_ON(tcp_conn->in.copy < 0);
623         BUG_ON(ctask->data_count < 0);
624
625         if (buf_size != (tcp_conn->data_copied + offset)) {
626                 if (!ctask->data_count) {
627                         BUG_ON(buf_size - tcp_conn->data_copied < 0);
628                         /* done with this PDU */
629                         return buf_size - tcp_conn->data_copied;
630                 }
631                 return -EAGAIN;
632         }
633
634         /* done with this buffer or with both - PDU and buffer */
635         tcp_conn->data_copied = 0;
636         return 0;
637 }
638
639 /**
640  * iscsi_tcp_copy - copy skb bits to the destanation buffer
641  * @conn: iscsi tcp connection
642  *
643  * Notes:
644  *      The function calls skb_copy_bits() and updates per-connection
645  *      byte counters.
646  **/
647 static inline int
648 iscsi_tcp_copy(struct iscsi_conn *conn, int buf_size)
649 {
650         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
651         int buf_left = buf_size - tcp_conn->data_copied;
652         int size = min(tcp_conn->in.copy, buf_left);
653         int rc;
654
655         debug_tcp("tcp_copy %d bytes at offset %d copied %d\n",
656                size, tcp_conn->in.offset, tcp_conn->data_copied);
657         BUG_ON(size <= 0);
658
659         rc = skb_copy_bits(tcp_conn->in.skb, tcp_conn->in.offset,
660                            (char*)conn->data + tcp_conn->data_copied, size);
661         BUG_ON(rc);
662
663         tcp_conn->in.offset += size;
664         tcp_conn->in.copy -= size;
665         tcp_conn->in.copied += size;
666         tcp_conn->data_copied += size;
667
668         if (buf_size != tcp_conn->data_copied)
669                 return -EAGAIN;
670
671         return 0;
672 }
673
674 static inline void
675 partial_sg_digest_update(struct hash_desc *desc, struct scatterlist *sg,
676                          int offset, int length)
677 {
678         struct scatterlist temp;
679
680         memcpy(&temp, sg, sizeof(struct scatterlist));
681         temp.offset = offset;
682         temp.length = length;
683         crypto_hash_update(desc, &temp, length);
684 }
685
686 static void
687 iscsi_recv_digest_update(struct iscsi_tcp_conn *tcp_conn, char* buf, int len)
688 {
689         struct scatterlist tmp;
690
691         sg_init_one(&tmp, buf, len);
692         crypto_hash_update(&tcp_conn->rx_hash, &tmp, len);
693 }
694
695 static int iscsi_scsi_data_in(struct iscsi_conn *conn)
696 {
697         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
698         struct iscsi_cmd_task *ctask = tcp_conn->in.ctask;
699         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
700         struct scsi_cmnd *sc = ctask->sc;
701         struct scatterlist *sg;
702         int i, offset, rc = 0;
703
704         BUG_ON((void*)ctask != sc->SCp.ptr);
705
706         offset = tcp_ctask->data_offset;
707         sg = scsi_sglist(sc);
708
709         if (tcp_ctask->data_offset)
710                 for (i = 0; i < tcp_ctask->sg_count; i++)
711                         offset -= sg[i].length;
712         /* we've passed through partial sg*/
713         if (offset < 0)
714                 offset = 0;
715
716         for (i = tcp_ctask->sg_count; i < scsi_sg_count(sc); i++) {
717                 char *dest;
718
719                 dest = kmap_atomic(sg[i].page, KM_SOFTIRQ0);
720                 rc = iscsi_ctask_copy(tcp_conn, ctask, dest + sg[i].offset,
721                                       sg[i].length, offset);
722                 kunmap_atomic(dest, KM_SOFTIRQ0);
723                 if (rc == -EAGAIN)
724                         /* continue with the next SKB/PDU */
725                         return rc;
726                 if (!rc) {
727                         if (conn->datadgst_en) {
728                                 if (!offset)
729                                         crypto_hash_update(
730                                                         &tcp_conn->rx_hash,
731                                                         &sg[i], sg[i].length);
732                                 else
733                                         partial_sg_digest_update(
734                                                         &tcp_conn->rx_hash,
735                                                         &sg[i],
736                                                         sg[i].offset + offset,
737                                                         sg[i].length - offset);
738                         }
739                         offset = 0;
740                         tcp_ctask->sg_count++;
741                 }
742
743                 if (!ctask->data_count) {
744                         if (rc && conn->datadgst_en)
745                                 /*
746                                  * data-in is complete, but buffer not...
747                                  */
748                                 partial_sg_digest_update(&tcp_conn->rx_hash,
749                                                          &sg[i],
750                                                          sg[i].offset,
751                                                          sg[i].length-rc);
752                         rc = 0;
753                         break;
754                 }
755
756                 if (!tcp_conn->in.copy)
757                         return -EAGAIN;
758         }
759         BUG_ON(ctask->data_count);
760
761         /* check for non-exceptional status */
762         if (tcp_conn->in.hdr->flags & ISCSI_FLAG_DATA_STATUS) {
763                 debug_scsi("done [sc %lx res %d itt 0x%x flags 0x%x]\n",
764                            (long)sc, sc->result, ctask->itt,
765                            tcp_conn->in.hdr->flags);
766                 spin_lock(&conn->session->lock);
767                 __iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
768                 spin_unlock(&conn->session->lock);
769         }
770
771         return rc;
772 }
773
774 static int
775 iscsi_data_recv(struct iscsi_conn *conn)
776 {
777         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
778         int rc = 0, opcode;
779
780         opcode = tcp_conn->in.hdr->opcode & ISCSI_OPCODE_MASK;
781         switch (opcode) {
782         case ISCSI_OP_SCSI_DATA_IN:
783                 rc = iscsi_scsi_data_in(conn);
784                 break;
785         case ISCSI_OP_SCSI_CMD_RSP:
786         case ISCSI_OP_TEXT_RSP:
787         case ISCSI_OP_LOGIN_RSP:
788         case ISCSI_OP_ASYNC_EVENT:
789         case ISCSI_OP_REJECT:
790                 /*
791                  * Collect data segment to the connection's data
792                  * placeholder
793                  */
794                 if (iscsi_tcp_copy(conn, tcp_conn->in.datalen)) {
795                         rc = -EAGAIN;
796                         goto exit;
797                 }
798
799                 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, conn->data,
800                                         tcp_conn->in.datalen);
801                 if (!rc && conn->datadgst_en && opcode != ISCSI_OP_LOGIN_RSP)
802                         iscsi_recv_digest_update(tcp_conn, conn->data,
803                                                 tcp_conn->in.datalen);
804                 break;
805         default:
806                 BUG_ON(1);
807         }
808 exit:
809         return rc;
810 }
811
812 /**
813  * iscsi_tcp_data_recv - TCP receive in sendfile fashion
814  * @rd_desc: read descriptor
815  * @skb: socket buffer
816  * @offset: offset in skb
817  * @len: skb->len - offset
818  **/
819 static int
820 iscsi_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
821                 unsigned int offset, size_t len)
822 {
823         int rc;
824         struct iscsi_conn *conn = rd_desc->arg.data;
825         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
826         int processed;
827         char pad[ISCSI_PAD_LEN];
828         struct scatterlist sg;
829
830         /*
831          * Save current SKB and its offset in the corresponding
832          * connection context.
833          */
834         tcp_conn->in.copy = skb->len - offset;
835         tcp_conn->in.offset = offset;
836         tcp_conn->in.skb = skb;
837         tcp_conn->in.len = tcp_conn->in.copy;
838         BUG_ON(tcp_conn->in.copy <= 0);
839         debug_tcp("in %d bytes\n", tcp_conn->in.copy);
840
841 more:
842         tcp_conn->in.copied = 0;
843         rc = 0;
844
845         if (unlikely(conn->suspend_rx)) {
846                 debug_tcp("conn %d Rx suspended!\n", conn->id);
847                 return 0;
848         }
849
850         if (tcp_conn->in_progress == IN_PROGRESS_WAIT_HEADER ||
851             tcp_conn->in_progress == IN_PROGRESS_HEADER_GATHER) {
852                 rc = iscsi_hdr_extract(tcp_conn);
853                 if (rc) {
854                        if (rc == -EAGAIN)
855                                 goto nomore;
856                        else {
857                                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
858                                 return 0;
859                        }
860                 }
861
862                 /*
863                  * Verify and process incoming PDU header.
864                  */
865                 rc = iscsi_tcp_hdr_recv(conn);
866                 if (!rc && tcp_conn->in.datalen) {
867                         if (conn->datadgst_en)
868                                 crypto_hash_init(&tcp_conn->rx_hash);
869                         tcp_conn->in_progress = IN_PROGRESS_DATA_RECV;
870                 } else if (rc) {
871                         iscsi_conn_failure(conn, rc);
872                         return 0;
873                 }
874         }
875
876         if (tcp_conn->in_progress == IN_PROGRESS_DDIGEST_RECV &&
877             tcp_conn->in.copy) {
878                 uint32_t recv_digest;
879
880                 debug_tcp("extra data_recv offset %d copy %d\n",
881                           tcp_conn->in.offset, tcp_conn->in.copy);
882
883                 if (!tcp_conn->data_copied) {
884                         if (tcp_conn->in.padding) {
885                                 debug_tcp("padding -> %d\n",
886                                           tcp_conn->in.padding);
887                                 memset(pad, 0, tcp_conn->in.padding);
888                                 sg_init_one(&sg, pad, tcp_conn->in.padding);
889                                 crypto_hash_update(&tcp_conn->rx_hash,
890                                                    &sg, sg.length);
891                         }
892                         crypto_hash_final(&tcp_conn->rx_hash,
893                                           (u8 *) &tcp_conn->in.datadgst);
894                         debug_tcp("rx digest 0x%x\n", tcp_conn->in.datadgst);
895                 }
896
897                 rc = iscsi_tcp_copy(conn, sizeof(uint32_t));
898                 if (rc) {
899                         if (rc == -EAGAIN)
900                                 goto again;
901                         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
902                         return 0;
903                 }
904
905                 memcpy(&recv_digest, conn->data, sizeof(uint32_t));
906                 if (recv_digest != tcp_conn->in.datadgst) {
907                         debug_tcp("iscsi_tcp: data digest error!"
908                                   "0x%x != 0x%x\n", recv_digest,
909                                   tcp_conn->in.datadgst);
910                         iscsi_conn_failure(conn, ISCSI_ERR_DATA_DGST);
911                         return 0;
912                 } else {
913                         debug_tcp("iscsi_tcp: data digest match!"
914                                   "0x%x == 0x%x\n", recv_digest,
915                                   tcp_conn->in.datadgst);
916                         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
917                 }
918         }
919
920         if (tcp_conn->in_progress == IN_PROGRESS_DATA_RECV &&
921             tcp_conn->in.copy) {
922                 debug_tcp("data_recv offset %d copy %d\n",
923                        tcp_conn->in.offset, tcp_conn->in.copy);
924
925                 rc = iscsi_data_recv(conn);
926                 if (rc) {
927                         if (rc == -EAGAIN)
928                                 goto again;
929                         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
930                         return 0;
931                 }
932
933                 if (tcp_conn->in.padding)
934                         tcp_conn->in_progress = IN_PROGRESS_PAD_RECV;
935                 else if (conn->datadgst_en)
936                         tcp_conn->in_progress = IN_PROGRESS_DDIGEST_RECV;
937                 else
938                         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
939                 tcp_conn->data_copied = 0;
940         }
941
942         if (tcp_conn->in_progress == IN_PROGRESS_PAD_RECV &&
943             tcp_conn->in.copy) {
944                 int copylen = min(tcp_conn->in.padding - tcp_conn->data_copied,
945                                   tcp_conn->in.copy);
946
947                 tcp_conn->in.copy -= copylen;
948                 tcp_conn->in.offset += copylen;
949                 tcp_conn->data_copied += copylen;
950
951                 if (tcp_conn->data_copied != tcp_conn->in.padding)
952                         tcp_conn->in_progress = IN_PROGRESS_PAD_RECV;
953                 else if (conn->datadgst_en)
954                         tcp_conn->in_progress = IN_PROGRESS_DDIGEST_RECV;
955                 else
956                         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
957                 tcp_conn->data_copied = 0;
958         }
959
960         debug_tcp("f, processed %d from out of %d padding %d\n",
961                tcp_conn->in.offset - offset, (int)len, tcp_conn->in.padding);
962         BUG_ON(tcp_conn->in.offset - offset > len);
963
964         if (tcp_conn->in.offset - offset != len) {
965                 debug_tcp("continue to process %d bytes\n",
966                        (int)len - (tcp_conn->in.offset - offset));
967                 goto more;
968         }
969
970 nomore:
971         processed = tcp_conn->in.offset - offset;
972         BUG_ON(processed == 0);
973         return processed;
974
975 again:
976         processed = tcp_conn->in.offset - offset;
977         debug_tcp("c, processed %d from out of %d rd_desc_cnt %d\n",
978                   processed, (int)len, (int)rd_desc->count);
979         BUG_ON(processed == 0);
980         BUG_ON(processed > len);
981
982         conn->rxdata_octets += processed;
983         return processed;
984 }
985
986 static void
987 iscsi_tcp_data_ready(struct sock *sk, int flag)
988 {
989         struct iscsi_conn *conn = sk->sk_user_data;
990         read_descriptor_t rd_desc;
991
992         read_lock(&sk->sk_callback_lock);
993
994         /*
995          * Use rd_desc to pass 'conn' to iscsi_tcp_data_recv.
996          * We set count to 1 because we want the network layer to
997          * hand us all the skbs that are available. iscsi_tcp_data_recv
998          * handled pdus that cross buffers or pdus that still need data.
999          */
1000         rd_desc.arg.data = conn;
1001         rd_desc.count = 1;
1002         tcp_read_sock(sk, &rd_desc, iscsi_tcp_data_recv);
1003
1004         read_unlock(&sk->sk_callback_lock);
1005 }
1006
1007 static void
1008 iscsi_tcp_state_change(struct sock *sk)
1009 {
1010         struct iscsi_tcp_conn *tcp_conn;
1011         struct iscsi_conn *conn;
1012         struct iscsi_session *session;
1013         void (*old_state_change)(struct sock *);
1014
1015         read_lock(&sk->sk_callback_lock);
1016
1017         conn = (struct iscsi_conn*)sk->sk_user_data;
1018         session = conn->session;
1019
1020         if ((sk->sk_state == TCP_CLOSE_WAIT ||
1021              sk->sk_state == TCP_CLOSE) &&
1022             !atomic_read(&sk->sk_rmem_alloc)) {
1023                 debug_tcp("iscsi_tcp_state_change: TCP_CLOSE|TCP_CLOSE_WAIT\n");
1024                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1025         }
1026
1027         tcp_conn = conn->dd_data;
1028         old_state_change = tcp_conn->old_state_change;
1029
1030         read_unlock(&sk->sk_callback_lock);
1031
1032         old_state_change(sk);
1033 }
1034
1035 /**
1036  * iscsi_write_space - Called when more output buffer space is available
1037  * @sk: socket space is available for
1038  **/
1039 static void
1040 iscsi_write_space(struct sock *sk)
1041 {
1042         struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
1043         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1044
1045         tcp_conn->old_write_space(sk);
1046         debug_tcp("iscsi_write_space: cid %d\n", conn->id);
1047         scsi_queue_work(conn->session->host, &conn->xmitwork);
1048 }
1049
1050 static void
1051 iscsi_conn_set_callbacks(struct iscsi_conn *conn)
1052 {
1053         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1054         struct sock *sk = tcp_conn->sock->sk;
1055
1056         /* assign new callbacks */
1057         write_lock_bh(&sk->sk_callback_lock);
1058         sk->sk_user_data = conn;
1059         tcp_conn->old_data_ready = sk->sk_data_ready;
1060         tcp_conn->old_state_change = sk->sk_state_change;
1061         tcp_conn->old_write_space = sk->sk_write_space;
1062         sk->sk_data_ready = iscsi_tcp_data_ready;
1063         sk->sk_state_change = iscsi_tcp_state_change;
1064         sk->sk_write_space = iscsi_write_space;
1065         write_unlock_bh(&sk->sk_callback_lock);
1066 }
1067
1068 static void
1069 iscsi_conn_restore_callbacks(struct iscsi_tcp_conn *tcp_conn)
1070 {
1071         struct sock *sk = tcp_conn->sock->sk;
1072
1073         /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
1074         write_lock_bh(&sk->sk_callback_lock);
1075         sk->sk_user_data    = NULL;
1076         sk->sk_data_ready   = tcp_conn->old_data_ready;
1077         sk->sk_state_change = tcp_conn->old_state_change;
1078         sk->sk_write_space  = tcp_conn->old_write_space;
1079         sk->sk_no_check  = 0;
1080         write_unlock_bh(&sk->sk_callback_lock);
1081 }
1082
1083 /**
1084  * iscsi_send - generic send routine
1085  * @sk: kernel's socket
1086  * @buf: buffer to write from
1087  * @size: actual size to write
1088  * @flags: socket's flags
1089  */
1090 static inline int
1091 iscsi_send(struct iscsi_conn *conn, struct iscsi_buf *buf, int size, int flags)
1092 {
1093         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1094         struct socket *sk = tcp_conn->sock;
1095         int offset = buf->sg.offset + buf->sent, res;
1096
1097         /*
1098          * if we got use_sg=0 or are sending something we kmallocd
1099          * then we did not have to do kmap (kmap returns page_address)
1100          *
1101          * if we got use_sg > 0, but had to drop down, we do not
1102          * set clustering so this should only happen for that
1103          * slab case.
1104          */
1105         if (buf->use_sendmsg)
1106                 res = sock_no_sendpage(sk, buf->sg.page, offset, size, flags);
1107         else
1108                 res = tcp_conn->sendpage(sk, buf->sg.page, offset, size, flags);
1109
1110         if (res >= 0) {
1111                 conn->txdata_octets += res;
1112                 buf->sent += res;
1113                 return res;
1114         }
1115
1116         tcp_conn->sendpage_failures_cnt++;
1117         if (res == -EAGAIN)
1118                 res = -ENOBUFS;
1119         else
1120                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1121         return res;
1122 }
1123
1124 /**
1125  * iscsi_sendhdr - send PDU Header via tcp_sendpage()
1126  * @conn: iscsi connection
1127  * @buf: buffer to write from
1128  * @datalen: lenght of data to be sent after the header
1129  *
1130  * Notes:
1131  *      (Tx, Fast Path)
1132  **/
1133 static inline int
1134 iscsi_sendhdr(struct iscsi_conn *conn, struct iscsi_buf *buf, int datalen)
1135 {
1136         int flags = 0; /* MSG_DONTWAIT; */
1137         int res, size;
1138
1139         size = buf->sg.length - buf->sent;
1140         BUG_ON(buf->sent + size > buf->sg.length);
1141         if (buf->sent + size != buf->sg.length || datalen)
1142                 flags |= MSG_MORE;
1143
1144         res = iscsi_send(conn, buf, size, flags);
1145         debug_tcp("sendhdr %d bytes, sent %d res %d\n", size, buf->sent, res);
1146         if (res >= 0) {
1147                 if (size != res)
1148                         return -EAGAIN;
1149                 return 0;
1150         }
1151
1152         return res;
1153 }
1154
1155 /**
1156  * iscsi_sendpage - send one page of iSCSI Data-Out.
1157  * @conn: iscsi connection
1158  * @buf: buffer to write from
1159  * @count: remaining data
1160  * @sent: number of bytes sent
1161  *
1162  * Notes:
1163  *      (Tx, Fast Path)
1164  **/
1165 static inline int
1166 iscsi_sendpage(struct iscsi_conn *conn, struct iscsi_buf *buf,
1167                int *count, int *sent)
1168 {
1169         int flags = 0; /* MSG_DONTWAIT; */
1170         int res, size;
1171
1172         size = buf->sg.length - buf->sent;
1173         BUG_ON(buf->sent + size > buf->sg.length);
1174         if (size > *count)
1175                 size = *count;
1176         if (buf->sent + size != buf->sg.length || *count != size)
1177                 flags |= MSG_MORE;
1178
1179         res = iscsi_send(conn, buf, size, flags);
1180         debug_tcp("sendpage: %d bytes, sent %d left %d sent %d res %d\n",
1181                   size, buf->sent, *count, *sent, res);
1182         if (res >= 0) {
1183                 *count -= res;
1184                 *sent += res;
1185                 if (size != res)
1186                         return -EAGAIN;
1187                 return 0;
1188         }
1189
1190         return res;
1191 }
1192
1193 static inline void
1194 iscsi_data_digest_init(struct iscsi_tcp_conn *tcp_conn,
1195                       struct iscsi_tcp_cmd_task *tcp_ctask)
1196 {
1197         crypto_hash_init(&tcp_conn->tx_hash);
1198         tcp_ctask->digest_count = 4;
1199 }
1200
1201 /**
1202  * iscsi_solicit_data_cont - initialize next Data-Out
1203  * @conn: iscsi connection
1204  * @ctask: scsi command task
1205  * @r2t: R2T info
1206  * @left: bytes left to transfer
1207  *
1208  * Notes:
1209  *      Initialize next Data-Out within this R2T sequence and continue
1210  *      to process next Scatter-Gather element(if any) of this SCSI command.
1211  *
1212  *      Called under connection lock.
1213  **/
1214 static void
1215 iscsi_solicit_data_cont(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
1216                         struct iscsi_r2t_info *r2t, int left)
1217 {
1218         struct iscsi_data *hdr;
1219         int new_offset;
1220
1221         hdr = &r2t->dtask.hdr;
1222         memset(hdr, 0, sizeof(struct iscsi_data));
1223         hdr->ttt = r2t->ttt;
1224         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
1225         r2t->solicit_datasn++;
1226         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
1227         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
1228         hdr->itt = ctask->hdr->itt;
1229         hdr->exp_statsn = r2t->exp_statsn;
1230         new_offset = r2t->data_offset + r2t->sent;
1231         hdr->offset = cpu_to_be32(new_offset);
1232         if (left > conn->max_xmit_dlength) {
1233                 hton24(hdr->dlength, conn->max_xmit_dlength);
1234                 r2t->data_count = conn->max_xmit_dlength;
1235         } else {
1236                 hton24(hdr->dlength, left);
1237                 r2t->data_count = left;
1238                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
1239         }
1240         conn->dataout_pdus_cnt++;
1241
1242         iscsi_buf_init_iov(&r2t->headbuf, (char*)hdr,
1243                            sizeof(struct iscsi_hdr));
1244
1245         if (iscsi_buf_left(&r2t->sendbuf))
1246                 return;
1247
1248         iscsi_buf_init_sg(&r2t->sendbuf, r2t->sg);
1249         r2t->sg += 1;
1250 }
1251
1252 static void iscsi_set_padding(struct iscsi_tcp_cmd_task *tcp_ctask,
1253                               unsigned long len)
1254 {
1255         tcp_ctask->pad_count = len & (ISCSI_PAD_LEN - 1);
1256         if (!tcp_ctask->pad_count)
1257                 return;
1258
1259         tcp_ctask->pad_count = ISCSI_PAD_LEN - tcp_ctask->pad_count;
1260         debug_scsi("write padding %d bytes\n", tcp_ctask->pad_count);
1261         tcp_ctask->xmstate |= XMSTATE_W_PAD;
1262 }
1263
1264 /**
1265  * iscsi_tcp_cmd_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
1266  * @conn: iscsi connection
1267  * @ctask: scsi command task
1268  * @sc: scsi command
1269  **/
1270 static void
1271 iscsi_tcp_cmd_init(struct iscsi_cmd_task *ctask)
1272 {
1273         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1274
1275         BUG_ON(__kfifo_len(tcp_ctask->r2tqueue));
1276         tcp_ctask->xmstate = XMSTATE_CMD_HDR_INIT;
1277 }
1278
1279 /**
1280  * iscsi_tcp_mtask_xmit - xmit management(immediate) task
1281  * @conn: iscsi connection
1282  * @mtask: task management task
1283  *
1284  * Notes:
1285  *      The function can return -EAGAIN in which case caller must
1286  *      call it again later, or recover. '0' return code means successful
1287  *      xmit.
1288  *
1289  *      Management xmit state machine consists of these states:
1290  *              XMSTATE_IMM_HDR_INIT    - calculate digest of PDU Header
1291  *              XMSTATE_IMM_HDR         - PDU Header xmit in progress
1292  *              XMSTATE_IMM_DATA        - PDU Data xmit in progress
1293  *              XMSTATE_IDLE            - management PDU is done
1294  **/
1295 static int
1296 iscsi_tcp_mtask_xmit(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1297 {
1298         struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
1299         int rc;
1300
1301         debug_scsi("mtask deq [cid %d state %x itt 0x%x]\n",
1302                 conn->id, tcp_mtask->xmstate, mtask->itt);
1303
1304         if (tcp_mtask->xmstate & XMSTATE_IMM_HDR_INIT) {
1305                 iscsi_buf_init_iov(&tcp_mtask->headbuf, (char*)mtask->hdr,
1306                                    sizeof(struct iscsi_hdr));
1307
1308                 if (mtask->data_count) {
1309                         tcp_mtask->xmstate |= XMSTATE_IMM_DATA;
1310                         iscsi_buf_init_iov(&tcp_mtask->sendbuf,
1311                                            (char*)mtask->data,
1312                                            mtask->data_count);
1313                 }
1314
1315                 if (conn->c_stage != ISCSI_CONN_INITIAL_STAGE &&
1316                     conn->stop_stage != STOP_CONN_RECOVER &&
1317                     conn->hdrdgst_en)
1318                         iscsi_hdr_digest(conn, &tcp_mtask->headbuf,
1319                                         (u8*)tcp_mtask->hdrext);
1320
1321                 tcp_mtask->sent = 0;
1322                 tcp_mtask->xmstate &= ~XMSTATE_IMM_HDR_INIT;
1323                 tcp_mtask->xmstate |= XMSTATE_IMM_HDR;
1324         }
1325
1326         if (tcp_mtask->xmstate & XMSTATE_IMM_HDR) {
1327                 rc = iscsi_sendhdr(conn, &tcp_mtask->headbuf,
1328                                    mtask->data_count);
1329                 if (rc)
1330                         return rc;
1331                 tcp_mtask->xmstate &= ~XMSTATE_IMM_HDR;
1332         }
1333
1334         if (tcp_mtask->xmstate & XMSTATE_IMM_DATA) {
1335                 BUG_ON(!mtask->data_count);
1336                 tcp_mtask->xmstate &= ~XMSTATE_IMM_DATA;
1337                 /* FIXME: implement.
1338                  * Virtual buffer could be spreaded across multiple pages...
1339                  */
1340                 do {
1341                         int rc;
1342
1343                         rc = iscsi_sendpage(conn, &tcp_mtask->sendbuf,
1344                                         &mtask->data_count, &tcp_mtask->sent);
1345                         if (rc) {
1346                                 tcp_mtask->xmstate |= XMSTATE_IMM_DATA;
1347                                 return rc;
1348                         }
1349                 } while (mtask->data_count);
1350         }
1351
1352         BUG_ON(tcp_mtask->xmstate != XMSTATE_IDLE);
1353         if (mtask->hdr->itt == RESERVED_ITT) {
1354                 struct iscsi_session *session = conn->session;
1355
1356                 spin_lock_bh(&session->lock);
1357                 list_del(&conn->mtask->running);
1358                 __kfifo_put(session->mgmtpool.queue, (void*)&conn->mtask,
1359                             sizeof(void*));
1360                 spin_unlock_bh(&session->lock);
1361         }
1362         return 0;
1363 }
1364
1365 static int
1366 iscsi_send_cmd_hdr(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1367 {
1368         struct scsi_cmnd *sc = ctask->sc;
1369         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1370         int rc = 0;
1371
1372         if (tcp_ctask->xmstate & XMSTATE_CMD_HDR_INIT) {
1373                 tcp_ctask->sent = 0;
1374                 tcp_ctask->sg_count = 0;
1375                 tcp_ctask->exp_datasn = 0;
1376
1377                 if (sc->sc_data_direction == DMA_TO_DEVICE) {
1378                         struct scatterlist *sg = scsi_sglist(sc);
1379
1380                         iscsi_buf_init_sg(&tcp_ctask->sendbuf, sg);
1381                         tcp_ctask->sg = sg + 1;
1382                         tcp_ctask->bad_sg = sg + scsi_sg_count(sc);
1383
1384                         debug_scsi("cmd [itt 0x%x total %d imm_data %d "
1385                                    "unsol count %d, unsol offset %d]\n",
1386                                    ctask->itt, scsi_bufflen(sc),
1387                                    ctask->imm_count, ctask->unsol_count,
1388                                    ctask->unsol_offset);
1389                 }
1390
1391                 iscsi_buf_init_iov(&tcp_ctask->headbuf, (char*)ctask->hdr,
1392                                   sizeof(struct iscsi_hdr));
1393
1394                 if (conn->hdrdgst_en)
1395                         iscsi_hdr_digest(conn, &tcp_ctask->headbuf,
1396                                          (u8*)tcp_ctask->hdrext);
1397                 tcp_ctask->xmstate &= ~XMSTATE_CMD_HDR_INIT;
1398                 tcp_ctask->xmstate |= XMSTATE_CMD_HDR_XMIT;
1399         }
1400
1401         if (tcp_ctask->xmstate & XMSTATE_CMD_HDR_XMIT) {
1402                 rc = iscsi_sendhdr(conn, &tcp_ctask->headbuf, ctask->imm_count);
1403                 if (rc)
1404                         return rc;
1405                 tcp_ctask->xmstate &= ~XMSTATE_CMD_HDR_XMIT;
1406
1407                 if (sc->sc_data_direction != DMA_TO_DEVICE)
1408                         return 0;
1409
1410                 if (ctask->imm_count) {
1411                         tcp_ctask->xmstate |= XMSTATE_IMM_DATA;
1412                         iscsi_set_padding(tcp_ctask, ctask->imm_count);
1413
1414                         if (ctask->conn->datadgst_en) {
1415                                 iscsi_data_digest_init(ctask->conn->dd_data,
1416                                                        tcp_ctask);
1417                                 tcp_ctask->immdigest = 0;
1418                         }
1419                 }
1420
1421                 if (ctask->unsol_count)
1422                         tcp_ctask->xmstate |=
1423                                         XMSTATE_UNS_HDR | XMSTATE_UNS_INIT;
1424         }
1425         return rc;
1426 }
1427
1428 static int
1429 iscsi_send_padding(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1430 {
1431         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1432         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1433         int sent = 0, rc;
1434
1435         if (tcp_ctask->xmstate & XMSTATE_W_PAD) {
1436                 iscsi_buf_init_iov(&tcp_ctask->sendbuf, (char*)&tcp_ctask->pad,
1437                                    tcp_ctask->pad_count);
1438                 if (conn->datadgst_en)
1439                         crypto_hash_update(&tcp_conn->tx_hash,
1440                                            &tcp_ctask->sendbuf.sg,
1441                                            tcp_ctask->sendbuf.sg.length);
1442         } else if (!(tcp_ctask->xmstate & XMSTATE_W_RESEND_PAD))
1443                 return 0;
1444
1445         tcp_ctask->xmstate &= ~XMSTATE_W_PAD;
1446         tcp_ctask->xmstate &= ~XMSTATE_W_RESEND_PAD;
1447         debug_scsi("sending %d pad bytes for itt 0x%x\n",
1448                    tcp_ctask->pad_count, ctask->itt);
1449         rc = iscsi_sendpage(conn, &tcp_ctask->sendbuf, &tcp_ctask->pad_count,
1450                            &sent);
1451         if (rc) {
1452                 debug_scsi("padding send failed %d\n", rc);
1453                 tcp_ctask->xmstate |= XMSTATE_W_RESEND_PAD;
1454         }
1455         return rc;
1456 }
1457
1458 static int
1459 iscsi_send_digest(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
1460                         struct iscsi_buf *buf, uint32_t *digest)
1461 {
1462         struct iscsi_tcp_cmd_task *tcp_ctask;
1463         struct iscsi_tcp_conn *tcp_conn;
1464         int rc, sent = 0;
1465
1466         if (!conn->datadgst_en)
1467                 return 0;
1468
1469         tcp_ctask = ctask->dd_data;
1470         tcp_conn = conn->dd_data;
1471
1472         if (!(tcp_ctask->xmstate & XMSTATE_W_RESEND_DATA_DIGEST)) {
1473                 crypto_hash_final(&tcp_conn->tx_hash, (u8*)digest);
1474                 iscsi_buf_init_iov(buf, (char*)digest, 4);
1475         }
1476         tcp_ctask->xmstate &= ~XMSTATE_W_RESEND_DATA_DIGEST;
1477
1478         rc = iscsi_sendpage(conn, buf, &tcp_ctask->digest_count, &sent);
1479         if (!rc)
1480                 debug_scsi("sent digest 0x%x for itt 0x%x\n", *digest,
1481                           ctask->itt);
1482         else {
1483                 debug_scsi("sending digest 0x%x failed for itt 0x%x!\n",
1484                           *digest, ctask->itt);
1485                 tcp_ctask->xmstate |= XMSTATE_W_RESEND_DATA_DIGEST;
1486         }
1487         return rc;
1488 }
1489
1490 static int
1491 iscsi_send_data(struct iscsi_cmd_task *ctask, struct iscsi_buf *sendbuf,
1492                 struct scatterlist **sg, int *sent, int *count,
1493                 struct iscsi_buf *digestbuf, uint32_t *digest)
1494 {
1495         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1496         struct iscsi_conn *conn = ctask->conn;
1497         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1498         int rc, buf_sent, offset;
1499
1500         while (*count) {
1501                 buf_sent = 0;
1502                 offset = sendbuf->sent;
1503
1504                 rc = iscsi_sendpage(conn, sendbuf, count, &buf_sent);
1505                 *sent = *sent + buf_sent;
1506                 if (buf_sent && conn->datadgst_en)
1507                         partial_sg_digest_update(&tcp_conn->tx_hash,
1508                                 &sendbuf->sg, sendbuf->sg.offset + offset,
1509                                 buf_sent);
1510                 if (!iscsi_buf_left(sendbuf) && *sg != tcp_ctask->bad_sg) {
1511                         iscsi_buf_init_sg(sendbuf, *sg);
1512                         *sg = *sg + 1;
1513                 }
1514
1515                 if (rc)
1516                         return rc;
1517         }
1518
1519         rc = iscsi_send_padding(conn, ctask);
1520         if (rc)
1521                 return rc;
1522
1523         return iscsi_send_digest(conn, ctask, digestbuf, digest);
1524 }
1525
1526 static int
1527 iscsi_send_unsol_hdr(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1528 {
1529         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1530         struct iscsi_data_task *dtask;
1531         int rc;
1532
1533         tcp_ctask->xmstate |= XMSTATE_UNS_DATA;
1534         if (tcp_ctask->xmstate & XMSTATE_UNS_INIT) {
1535                 dtask = &tcp_ctask->unsol_dtask;
1536
1537                 iscsi_prep_unsolicit_data_pdu(ctask, &dtask->hdr);
1538                 iscsi_buf_init_iov(&tcp_ctask->headbuf, (char*)&dtask->hdr,
1539                                    sizeof(struct iscsi_hdr));
1540                 if (conn->hdrdgst_en)
1541                         iscsi_hdr_digest(conn, &tcp_ctask->headbuf,
1542                                         (u8*)dtask->hdrext);
1543
1544                 tcp_ctask->xmstate &= ~XMSTATE_UNS_INIT;
1545                 iscsi_set_padding(tcp_ctask, ctask->data_count);
1546         }
1547
1548         rc = iscsi_sendhdr(conn, &tcp_ctask->headbuf, ctask->data_count);
1549         if (rc) {
1550                 tcp_ctask->xmstate &= ~XMSTATE_UNS_DATA;
1551                 tcp_ctask->xmstate |= XMSTATE_UNS_HDR;
1552                 return rc;
1553         }
1554
1555         if (conn->datadgst_en) {
1556                 dtask = &tcp_ctask->unsol_dtask;
1557                 iscsi_data_digest_init(ctask->conn->dd_data, tcp_ctask);
1558                 dtask->digest = 0;
1559         }
1560
1561         debug_scsi("uns dout [itt 0x%x dlen %d sent %d]\n",
1562                    ctask->itt, ctask->unsol_count, tcp_ctask->sent);
1563         return 0;
1564 }
1565
1566 static int
1567 iscsi_send_unsol_pdu(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1568 {
1569         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1570         int rc;
1571
1572         if (tcp_ctask->xmstate & XMSTATE_UNS_HDR) {
1573                 BUG_ON(!ctask->unsol_count);
1574                 tcp_ctask->xmstate &= ~XMSTATE_UNS_HDR;
1575 send_hdr:
1576                 rc = iscsi_send_unsol_hdr(conn, ctask);
1577                 if (rc)
1578                         return rc;
1579         }
1580
1581         if (tcp_ctask->xmstate & XMSTATE_UNS_DATA) {
1582                 struct iscsi_data_task *dtask = &tcp_ctask->unsol_dtask;
1583                 int start = tcp_ctask->sent;
1584
1585                 rc = iscsi_send_data(ctask, &tcp_ctask->sendbuf, &tcp_ctask->sg,
1586                                      &tcp_ctask->sent, &ctask->data_count,
1587                                      &dtask->digestbuf, &dtask->digest);
1588                 ctask->unsol_count -= tcp_ctask->sent - start;
1589                 if (rc)
1590                         return rc;
1591                 tcp_ctask->xmstate &= ~XMSTATE_UNS_DATA;
1592                 /*
1593                  * Done with the Data-Out. Next, check if we need
1594                  * to send another unsolicited Data-Out.
1595                  */
1596                 if (ctask->unsol_count) {
1597                         debug_scsi("sending more uns\n");
1598                         tcp_ctask->xmstate |= XMSTATE_UNS_INIT;
1599                         goto send_hdr;
1600                 }
1601         }
1602         return 0;
1603 }
1604
1605 static int iscsi_send_sol_pdu(struct iscsi_conn *conn,
1606                               struct iscsi_cmd_task *ctask)
1607 {
1608         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1609         struct iscsi_session *session = conn->session;
1610         struct iscsi_r2t_info *r2t;
1611         struct iscsi_data_task *dtask;
1612         int left, rc;
1613
1614         if (tcp_ctask->xmstate & XMSTATE_SOL_HDR_INIT) {
1615                 if (!tcp_ctask->r2t) {
1616                         spin_lock_bh(&session->lock);
1617                         __kfifo_get(tcp_ctask->r2tqueue, (void*)&tcp_ctask->r2t,
1618                                     sizeof(void*));
1619                         spin_unlock_bh(&session->lock);
1620                 }
1621 send_hdr:
1622                 r2t = tcp_ctask->r2t;
1623                 dtask = &r2t->dtask;
1624
1625                 if (conn->hdrdgst_en)
1626                         iscsi_hdr_digest(conn, &r2t->headbuf,
1627                                         (u8*)dtask->hdrext);
1628                 tcp_ctask->xmstate &= ~XMSTATE_SOL_HDR_INIT;
1629                 tcp_ctask->xmstate |= XMSTATE_SOL_HDR;
1630         }
1631
1632         if (tcp_ctask->xmstate & XMSTATE_SOL_HDR) {
1633                 r2t = tcp_ctask->r2t;
1634                 dtask = &r2t->dtask;
1635
1636                 rc = iscsi_sendhdr(conn, &r2t->headbuf, r2t->data_count);
1637                 if (rc)
1638                         return rc;
1639                 tcp_ctask->xmstate &= ~XMSTATE_SOL_HDR;
1640                 tcp_ctask->xmstate |= XMSTATE_SOL_DATA;
1641
1642                 if (conn->datadgst_en) {
1643                         iscsi_data_digest_init(conn->dd_data, tcp_ctask);
1644                         dtask->digest = 0;
1645                 }
1646
1647                 iscsi_set_padding(tcp_ctask, r2t->data_count);
1648                 debug_scsi("sol dout [dsn %d itt 0x%x dlen %d sent %d]\n",
1649                         r2t->solicit_datasn - 1, ctask->itt, r2t->data_count,
1650                         r2t->sent);
1651         }
1652
1653         if (tcp_ctask->xmstate & XMSTATE_SOL_DATA) {
1654                 r2t = tcp_ctask->r2t;
1655                 dtask = &r2t->dtask;
1656
1657                 rc = iscsi_send_data(ctask, &r2t->sendbuf, &r2t->sg,
1658                                      &r2t->sent, &r2t->data_count,
1659                                      &dtask->digestbuf, &dtask->digest);
1660                 if (rc)
1661                         return rc;
1662                 tcp_ctask->xmstate &= ~XMSTATE_SOL_DATA;
1663
1664                 /*
1665                  * Done with this Data-Out. Next, check if we have
1666                  * to send another Data-Out for this R2T.
1667                  */
1668                 BUG_ON(r2t->data_length - r2t->sent < 0);
1669                 left = r2t->data_length - r2t->sent;
1670                 if (left) {
1671                         iscsi_solicit_data_cont(conn, ctask, r2t, left);
1672                         goto send_hdr;
1673                 }
1674
1675                 /*
1676                  * Done with this R2T. Check if there are more
1677                  * outstanding R2Ts ready to be processed.
1678                  */
1679                 spin_lock_bh(&session->lock);
1680                 tcp_ctask->r2t = NULL;
1681                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
1682                             sizeof(void*));
1683                 if (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t,
1684                                 sizeof(void*))) {
1685                         tcp_ctask->r2t = r2t;
1686                         spin_unlock_bh(&session->lock);
1687                         goto send_hdr;
1688                 }
1689                 spin_unlock_bh(&session->lock);
1690         }
1691         return 0;
1692 }
1693
1694 /**
1695  * iscsi_tcp_ctask_xmit - xmit normal PDU task
1696  * @conn: iscsi connection
1697  * @ctask: iscsi command task
1698  *
1699  * Notes:
1700  *      The function can return -EAGAIN in which case caller must
1701  *      call it again later, or recover. '0' return code means successful
1702  *      xmit.
1703  *      The function is devided to logical helpers (above) for the different
1704  *      xmit stages.
1705  *
1706  *iscsi_send_cmd_hdr()
1707  *      XMSTATE_CMD_HDR_INIT - prepare Header and Data buffers Calculate
1708  *                             Header Digest
1709  *      XMSTATE_CMD_HDR_XMIT - Transmit header in progress
1710  *
1711  *iscsi_send_padding
1712  *      XMSTATE_W_PAD        - Prepare and send pading
1713  *      XMSTATE_W_RESEND_PAD - retry send pading
1714  *
1715  *iscsi_send_digest
1716  *      XMSTATE_W_RESEND_DATA_DIGEST - Finalize and send Data Digest
1717  *      XMSTATE_W_RESEND_DATA_DIGEST - retry sending digest
1718  *
1719  *iscsi_send_unsol_hdr
1720  *      XMSTATE_UNS_INIT     - prepare un-solicit data header and digest
1721  *      XMSTATE_UNS_HDR      - send un-solicit header
1722  *
1723  *iscsi_send_unsol_pdu
1724  *      XMSTATE_UNS_DATA     - send un-solicit data in progress
1725  *
1726  *iscsi_send_sol_pdu
1727  *      XMSTATE_SOL_HDR_INIT - solicit data header and digest initialize
1728  *      XMSTATE_SOL_HDR      - send solicit header
1729  *      XMSTATE_SOL_DATA     - send solicit data
1730  *
1731  *iscsi_tcp_ctask_xmit
1732  *      XMSTATE_IMM_DATA     - xmit managment data (??)
1733  **/
1734 static int
1735 iscsi_tcp_ctask_xmit(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1736 {
1737         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1738         int rc = 0;
1739
1740         debug_scsi("ctask deq [cid %d xmstate %x itt 0x%x]\n",
1741                 conn->id, tcp_ctask->xmstate, ctask->itt);
1742
1743         rc = iscsi_send_cmd_hdr(conn, ctask);
1744         if (rc)
1745                 return rc;
1746         if (ctask->sc->sc_data_direction != DMA_TO_DEVICE)
1747                 return 0;
1748
1749         if (tcp_ctask->xmstate & XMSTATE_IMM_DATA) {
1750                 rc = iscsi_send_data(ctask, &tcp_ctask->sendbuf, &tcp_ctask->sg,
1751                                      &tcp_ctask->sent, &ctask->imm_count,
1752                                      &tcp_ctask->immbuf, &tcp_ctask->immdigest);
1753                 if (rc)
1754                         return rc;
1755                 tcp_ctask->xmstate &= ~XMSTATE_IMM_DATA;
1756         }
1757
1758         rc = iscsi_send_unsol_pdu(conn, ctask);
1759         if (rc)
1760                 return rc;
1761
1762         rc = iscsi_send_sol_pdu(conn, ctask);
1763         if (rc)
1764                 return rc;
1765
1766         return rc;
1767 }
1768
1769 static struct iscsi_cls_conn *
1770 iscsi_tcp_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
1771 {
1772         struct iscsi_conn *conn;
1773         struct iscsi_cls_conn *cls_conn;
1774         struct iscsi_tcp_conn *tcp_conn;
1775
1776         cls_conn = iscsi_conn_setup(cls_session, conn_idx);
1777         if (!cls_conn)
1778                 return NULL;
1779         conn = cls_conn->dd_data;
1780         /*
1781          * due to strange issues with iser these are not set
1782          * in iscsi_conn_setup
1783          */
1784         conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1785
1786         tcp_conn = kzalloc(sizeof(*tcp_conn), GFP_KERNEL);
1787         if (!tcp_conn)
1788                 goto tcp_conn_alloc_fail;
1789
1790         conn->dd_data = tcp_conn;
1791         tcp_conn->iscsi_conn = conn;
1792         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
1793         /* initial operational parameters */
1794         tcp_conn->hdr_size = sizeof(struct iscsi_hdr);
1795
1796         tcp_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1797                                                   CRYPTO_ALG_ASYNC);
1798         tcp_conn->tx_hash.flags = 0;
1799         if (IS_ERR(tcp_conn->tx_hash.tfm)) {
1800                 printk(KERN_ERR "Could not create connection due to crc32c "
1801                        "loading error %ld. Make sure the crc32c module is "
1802                        "built as a module or into the kernel\n",
1803                         PTR_ERR(tcp_conn->tx_hash.tfm));
1804                 goto free_tcp_conn;
1805         }
1806
1807         tcp_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1808                                                   CRYPTO_ALG_ASYNC);
1809         tcp_conn->rx_hash.flags = 0;
1810         if (IS_ERR(tcp_conn->rx_hash.tfm)) {
1811                 printk(KERN_ERR "Could not create connection due to crc32c "
1812                        "loading error %ld. Make sure the crc32c module is "
1813                        "built as a module or into the kernel\n",
1814                         PTR_ERR(tcp_conn->rx_hash.tfm));
1815                 goto free_tx_tfm;
1816         }
1817
1818         return cls_conn;
1819
1820 free_tx_tfm:
1821         crypto_free_hash(tcp_conn->tx_hash.tfm);
1822 free_tcp_conn:
1823         kfree(tcp_conn);
1824 tcp_conn_alloc_fail:
1825         iscsi_conn_teardown(cls_conn);
1826         return NULL;
1827 }
1828
1829 static void
1830 iscsi_tcp_release_conn(struct iscsi_conn *conn)
1831 {
1832         struct iscsi_session *session = conn->session;
1833         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1834         struct socket *sock = tcp_conn->sock;
1835
1836         if (!sock)
1837                 return;
1838
1839         sock_hold(sock->sk);
1840         iscsi_conn_restore_callbacks(tcp_conn);
1841         sock_put(sock->sk);
1842
1843         spin_lock_bh(&session->lock);
1844         tcp_conn->sock = NULL;
1845         conn->recv_lock = NULL;
1846         spin_unlock_bh(&session->lock);
1847         sockfd_put(sock);
1848 }
1849
1850 static void
1851 iscsi_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
1852 {
1853         struct iscsi_conn *conn = cls_conn->dd_data;
1854         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1855
1856         iscsi_tcp_release_conn(conn);
1857         iscsi_conn_teardown(cls_conn);
1858
1859         if (tcp_conn->tx_hash.tfm)
1860                 crypto_free_hash(tcp_conn->tx_hash.tfm);
1861         if (tcp_conn->rx_hash.tfm)
1862                 crypto_free_hash(tcp_conn->rx_hash.tfm);
1863
1864         kfree(tcp_conn);
1865 }
1866
1867 static void
1868 iscsi_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
1869 {
1870         struct iscsi_conn *conn = cls_conn->dd_data;
1871         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1872
1873         iscsi_conn_stop(cls_conn, flag);
1874         iscsi_tcp_release_conn(conn);
1875         tcp_conn->hdr_size = sizeof(struct iscsi_hdr);
1876 }
1877
1878 static int iscsi_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
1879                               char *buf, int *port,
1880                               int (*getname)(struct socket *, struct sockaddr *,
1881                                         int *addrlen))
1882 {
1883         struct sockaddr_storage *addr;
1884         struct sockaddr_in6 *sin6;
1885         struct sockaddr_in *sin;
1886         int rc = 0, len;
1887
1888         addr = kmalloc(GFP_KERNEL, sizeof(*addr));
1889         if (!addr)
1890                 return -ENOMEM;
1891
1892         if (getname(sock, (struct sockaddr *) addr, &len)) {
1893                 rc = -ENODEV;
1894                 goto free_addr;
1895         }
1896
1897         switch (addr->ss_family) {
1898         case AF_INET:
1899                 sin = (struct sockaddr_in *)addr;
1900                 spin_lock_bh(&conn->session->lock);
1901                 sprintf(buf, NIPQUAD_FMT, NIPQUAD(sin->sin_addr.s_addr));
1902                 *port = be16_to_cpu(sin->sin_port);
1903                 spin_unlock_bh(&conn->session->lock);
1904                 break;
1905         case AF_INET6:
1906                 sin6 = (struct sockaddr_in6 *)addr;
1907                 spin_lock_bh(&conn->session->lock);
1908                 sprintf(buf, NIP6_FMT, NIP6(sin6->sin6_addr));
1909                 *port = be16_to_cpu(sin6->sin6_port);
1910                 spin_unlock_bh(&conn->session->lock);
1911                 break;
1912         }
1913 free_addr:
1914         kfree(addr);
1915         return rc;
1916 }
1917
1918 static int
1919 iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
1920                     struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
1921                     int is_leading)
1922 {
1923         struct iscsi_conn *conn = cls_conn->dd_data;
1924         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1925         struct sock *sk;
1926         struct socket *sock;
1927         int err;
1928
1929         /* lookup for existing socket */
1930         sock = sockfd_lookup((int)transport_eph, &err);
1931         if (!sock) {
1932                 printk(KERN_ERR "iscsi_tcp: sockfd_lookup failed %d\n", err);
1933                 return -EEXIST;
1934         }
1935         /*
1936          * copy these values now because if we drop the session
1937          * userspace may still want to query the values since we will
1938          * be using them for the reconnect
1939          */
1940         err = iscsi_tcp_get_addr(conn, sock, conn->portal_address,
1941                                  &conn->portal_port, kernel_getpeername);
1942         if (err)
1943                 goto free_socket;
1944
1945         err = iscsi_tcp_get_addr(conn, sock, conn->local_address,
1946                                 &conn->local_port, kernel_getsockname);
1947         if (err)
1948                 goto free_socket;
1949
1950         err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
1951         if (err)
1952                 goto free_socket;
1953
1954         /* bind iSCSI connection and socket */
1955         tcp_conn->sock = sock;
1956
1957         /* setup Socket parameters */
1958         sk = sock->sk;
1959         sk->sk_reuse = 1;
1960         sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
1961         sk->sk_allocation = GFP_ATOMIC;
1962
1963         /* FIXME: disable Nagle's algorithm */
1964
1965         /*
1966          * Intercept TCP callbacks for sendfile like receive
1967          * processing.
1968          */
1969         conn->recv_lock = &sk->sk_callback_lock;
1970         iscsi_conn_set_callbacks(conn);
1971         tcp_conn->sendpage = tcp_conn->sock->ops->sendpage;
1972         /*
1973          * set receive state machine into initial state
1974          */
1975         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
1976         return 0;
1977
1978 free_socket:
1979         sockfd_put(sock);
1980         return err;
1981 }
1982
1983 /* called with host lock */
1984 static void
1985 iscsi_tcp_mgmt_init(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1986 {
1987         struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
1988         tcp_mtask->xmstate = XMSTATE_IMM_HDR_INIT;
1989 }
1990
1991 static int
1992 iscsi_r2tpool_alloc(struct iscsi_session *session)
1993 {
1994         int i;
1995         int cmd_i;
1996
1997         /*
1998          * initialize per-task: R2T pool and xmit queue
1999          */
2000         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
2001                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
2002                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2003
2004                 /*
2005                  * pre-allocated x4 as much r2ts to handle race when
2006                  * target acks DataOut faster than we data_xmit() queues
2007                  * could replenish r2tqueue.
2008                  */
2009
2010                 /* R2T pool */
2011                 if (iscsi_pool_init(&tcp_ctask->r2tpool, session->max_r2t * 4,
2012                                     (void***)&tcp_ctask->r2ts,
2013                                     sizeof(struct iscsi_r2t_info))) {
2014                         goto r2t_alloc_fail;
2015                 }
2016
2017                 /* R2T xmit queue */
2018                 tcp_ctask->r2tqueue = kfifo_alloc(
2019                       session->max_r2t * 4 * sizeof(void*), GFP_KERNEL, NULL);
2020                 if (tcp_ctask->r2tqueue == ERR_PTR(-ENOMEM)) {
2021                         iscsi_pool_free(&tcp_ctask->r2tpool,
2022                                         (void**)tcp_ctask->r2ts);
2023                         goto r2t_alloc_fail;
2024                 }
2025         }
2026
2027         return 0;
2028
2029 r2t_alloc_fail:
2030         for (i = 0; i < cmd_i; i++) {
2031                 struct iscsi_cmd_task *ctask = session->cmds[i];
2032                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2033
2034                 kfifo_free(tcp_ctask->r2tqueue);
2035                 iscsi_pool_free(&tcp_ctask->r2tpool,
2036                                 (void**)tcp_ctask->r2ts);
2037         }
2038         return -ENOMEM;
2039 }
2040
2041 static void
2042 iscsi_r2tpool_free(struct iscsi_session *session)
2043 {
2044         int i;
2045
2046         for (i = 0; i < session->cmds_max; i++) {
2047                 struct iscsi_cmd_task *ctask = session->cmds[i];
2048                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2049
2050                 kfifo_free(tcp_ctask->r2tqueue);
2051                 iscsi_pool_free(&tcp_ctask->r2tpool,
2052                                 (void**)tcp_ctask->r2ts);
2053         }
2054 }
2055
2056 static int
2057 iscsi_conn_set_param(struct iscsi_cls_conn *cls_conn, enum iscsi_param param,
2058                      char *buf, int buflen)
2059 {
2060         struct iscsi_conn *conn = cls_conn->dd_data;
2061         struct iscsi_session *session = conn->session;
2062         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2063         int value;
2064
2065         switch(param) {
2066         case ISCSI_PARAM_HDRDGST_EN:
2067                 iscsi_set_param(cls_conn, param, buf, buflen);
2068                 tcp_conn->hdr_size = sizeof(struct iscsi_hdr);
2069                 if (conn->hdrdgst_en)
2070                         tcp_conn->hdr_size += sizeof(__u32);
2071                 break;
2072         case ISCSI_PARAM_DATADGST_EN:
2073                 iscsi_set_param(cls_conn, param, buf, buflen);
2074                 tcp_conn->sendpage = conn->datadgst_en ?
2075                         sock_no_sendpage : tcp_conn->sock->ops->sendpage;
2076                 break;
2077         case ISCSI_PARAM_MAX_R2T:
2078                 sscanf(buf, "%d", &value);
2079                 if (session->max_r2t == roundup_pow_of_two(value))
2080                         break;
2081                 iscsi_r2tpool_free(session);
2082                 iscsi_set_param(cls_conn, param, buf, buflen);
2083                 if (session->max_r2t & (session->max_r2t - 1))
2084                         session->max_r2t = roundup_pow_of_two(session->max_r2t);
2085                 if (iscsi_r2tpool_alloc(session))
2086                         return -ENOMEM;
2087                 break;
2088         default:
2089                 return iscsi_set_param(cls_conn, param, buf, buflen);
2090         }
2091
2092         return 0;
2093 }
2094
2095 static int
2096 iscsi_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
2097                          enum iscsi_param param, char *buf)
2098 {
2099         struct iscsi_conn *conn = cls_conn->dd_data;
2100         int len;
2101
2102         switch(param) {
2103         case ISCSI_PARAM_CONN_PORT:
2104                 spin_lock_bh(&conn->session->lock);
2105                 len = sprintf(buf, "%hu\n", conn->portal_port);
2106                 spin_unlock_bh(&conn->session->lock);
2107                 break;
2108         case ISCSI_PARAM_CONN_ADDRESS:
2109                 spin_lock_bh(&conn->session->lock);
2110                 len = sprintf(buf, "%s\n", conn->portal_address);
2111                 spin_unlock_bh(&conn->session->lock);
2112                 break;
2113         default:
2114                 return iscsi_conn_get_param(cls_conn, param, buf);
2115         }
2116
2117         return len;
2118 }
2119
2120 static int
2121 iscsi_tcp_host_get_param(struct Scsi_Host *shost, enum iscsi_host_param param,
2122                          char *buf)
2123 {
2124         struct iscsi_session *session = iscsi_hostdata(shost->hostdata);
2125         int len;
2126
2127         switch (param) {
2128         case ISCSI_HOST_PARAM_IPADDRESS:
2129                 spin_lock_bh(&session->lock);
2130                 if (!session->leadconn)
2131                         len = -ENODEV;
2132                 else
2133                         len = sprintf(buf, "%s\n",
2134                                      session->leadconn->local_address);
2135                 spin_unlock_bh(&session->lock);
2136                 break;
2137         default:
2138                 return iscsi_host_get_param(shost, param, buf);
2139         }
2140         return len;
2141 }
2142
2143 static void
2144 iscsi_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
2145 {
2146         struct iscsi_conn *conn = cls_conn->dd_data;
2147         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2148
2149         stats->txdata_octets = conn->txdata_octets;
2150         stats->rxdata_octets = conn->rxdata_octets;
2151         stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
2152         stats->dataout_pdus = conn->dataout_pdus_cnt;
2153         stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
2154         stats->datain_pdus = conn->datain_pdus_cnt;
2155         stats->r2t_pdus = conn->r2t_pdus_cnt;
2156         stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
2157         stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
2158         stats->custom_length = 3;
2159         strcpy(stats->custom[0].desc, "tx_sendpage_failures");
2160         stats->custom[0].value = tcp_conn->sendpage_failures_cnt;
2161         strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
2162         stats->custom[1].value = tcp_conn->discontiguous_hdr_cnt;
2163         strcpy(stats->custom[2].desc, "eh_abort_cnt");
2164         stats->custom[2].value = conn->eh_abort_cnt;
2165 }
2166
2167 static struct iscsi_cls_session *
2168 iscsi_tcp_session_create(struct iscsi_transport *iscsit,
2169                          struct scsi_transport_template *scsit,
2170                          uint16_t cmds_max, uint16_t qdepth,
2171                          uint32_t initial_cmdsn, uint32_t *hostno)
2172 {
2173         struct iscsi_cls_session *cls_session;
2174         struct iscsi_session *session;
2175         uint32_t hn;
2176         int cmd_i;
2177
2178         cls_session = iscsi_session_setup(iscsit, scsit, cmds_max, qdepth,
2179                                          sizeof(struct iscsi_tcp_cmd_task),
2180                                          sizeof(struct iscsi_tcp_mgmt_task),
2181                                          initial_cmdsn, &hn);
2182         if (!cls_session)
2183                 return NULL;
2184         *hostno = hn;
2185
2186         session = class_to_transport_session(cls_session);
2187         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
2188                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
2189                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2190
2191                 ctask->hdr = &tcp_ctask->hdr;
2192         }
2193
2194         for (cmd_i = 0; cmd_i < session->mgmtpool_max; cmd_i++) {
2195                 struct iscsi_mgmt_task *mtask = session->mgmt_cmds[cmd_i];
2196                 struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
2197
2198                 mtask->hdr = &tcp_mtask->hdr;
2199         }
2200
2201         if (iscsi_r2tpool_alloc(class_to_transport_session(cls_session)))
2202                 goto r2tpool_alloc_fail;
2203
2204         return cls_session;
2205
2206 r2tpool_alloc_fail:
2207         iscsi_session_teardown(cls_session);
2208         return NULL;
2209 }
2210
2211 static void iscsi_tcp_session_destroy(struct iscsi_cls_session *cls_session)
2212 {
2213         iscsi_r2tpool_free(class_to_transport_session(cls_session));
2214         iscsi_session_teardown(cls_session);
2215 }
2216
2217 static int iscsi_tcp_slave_configure(struct scsi_device *sdev)
2218 {
2219         blk_queue_dma_alignment(sdev->request_queue, 0);
2220         return 0;
2221 }
2222
2223 static struct scsi_host_template iscsi_sht = {
2224         .name                   = "iSCSI Initiator over TCP/IP",
2225         .queuecommand           = iscsi_queuecommand,
2226         .change_queue_depth     = iscsi_change_queue_depth,
2227         .can_queue              = ISCSI_DEF_XMIT_CMDS_MAX - 1,
2228         .sg_tablesize           = ISCSI_SG_TABLESIZE,
2229         .max_sectors            = 0xFFFF,
2230         .cmd_per_lun            = ISCSI_DEF_CMD_PER_LUN,
2231         .eh_abort_handler       = iscsi_eh_abort,
2232         .eh_host_reset_handler  = iscsi_eh_host_reset,
2233         .use_clustering         = DISABLE_CLUSTERING,
2234         .slave_configure        = iscsi_tcp_slave_configure,
2235         .proc_name              = "iscsi_tcp",
2236         .this_id                = -1,
2237 };
2238
2239 static struct iscsi_transport iscsi_tcp_transport = {
2240         .owner                  = THIS_MODULE,
2241         .name                   = "tcp",
2242         .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
2243                                   | CAP_DATADGST,
2244         .param_mask             = ISCSI_MAX_RECV_DLENGTH |
2245                                   ISCSI_MAX_XMIT_DLENGTH |
2246                                   ISCSI_HDRDGST_EN |
2247                                   ISCSI_DATADGST_EN |
2248                                   ISCSI_INITIAL_R2T_EN |
2249                                   ISCSI_MAX_R2T |
2250                                   ISCSI_IMM_DATA_EN |
2251                                   ISCSI_FIRST_BURST |
2252                                   ISCSI_MAX_BURST |
2253                                   ISCSI_PDU_INORDER_EN |
2254                                   ISCSI_DATASEQ_INORDER_EN |
2255                                   ISCSI_ERL |
2256                                   ISCSI_CONN_PORT |
2257                                   ISCSI_CONN_ADDRESS |
2258                                   ISCSI_EXP_STATSN |
2259                                   ISCSI_PERSISTENT_PORT |
2260                                   ISCSI_PERSISTENT_ADDRESS |
2261                                   ISCSI_TARGET_NAME | ISCSI_TPGT |
2262                                   ISCSI_USERNAME | ISCSI_PASSWORD |
2263                                   ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN,
2264         .host_param_mask        = ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
2265                                   ISCSI_HOST_INITIATOR_NAME |
2266                                   ISCSI_HOST_NETDEV_NAME,
2267         .host_template          = &iscsi_sht,
2268         .conndata_size          = sizeof(struct iscsi_conn),
2269         .max_conn               = 1,
2270         .max_cmd_len            = ISCSI_TCP_MAX_CMD_LEN,
2271         /* session management */
2272         .create_session         = iscsi_tcp_session_create,
2273         .destroy_session        = iscsi_tcp_session_destroy,
2274         /* connection management */
2275         .create_conn            = iscsi_tcp_conn_create,
2276         .bind_conn              = iscsi_tcp_conn_bind,
2277         .destroy_conn           = iscsi_tcp_conn_destroy,
2278         .set_param              = iscsi_conn_set_param,
2279         .get_conn_param         = iscsi_tcp_conn_get_param,
2280         .get_session_param      = iscsi_session_get_param,
2281         .start_conn             = iscsi_conn_start,
2282         .stop_conn              = iscsi_tcp_conn_stop,
2283         /* iscsi host params */
2284         .get_host_param         = iscsi_tcp_host_get_param,
2285         .set_host_param         = iscsi_host_set_param,
2286         /* IO */
2287         .send_pdu               = iscsi_conn_send_pdu,
2288         .get_stats              = iscsi_conn_get_stats,
2289         .init_cmd_task          = iscsi_tcp_cmd_init,
2290         .init_mgmt_task         = iscsi_tcp_mgmt_init,
2291         .xmit_cmd_task          = iscsi_tcp_ctask_xmit,
2292         .xmit_mgmt_task         = iscsi_tcp_mtask_xmit,
2293         .cleanup_cmd_task       = iscsi_tcp_cleanup_ctask,
2294         /* recovery */
2295         .session_recovery_timedout = iscsi_session_recovery_timedout,
2296 };
2297
2298 static int __init
2299 iscsi_tcp_init(void)
2300 {
2301         if (iscsi_max_lun < 1) {
2302                 printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
2303                        iscsi_max_lun);
2304                 return -EINVAL;
2305         }
2306         iscsi_tcp_transport.max_lun = iscsi_max_lun;
2307
2308         if (!iscsi_register_transport(&iscsi_tcp_transport))
2309                 return -ENODEV;
2310
2311         return 0;
2312 }
2313
2314 static void __exit
2315 iscsi_tcp_exit(void)
2316 {
2317         iscsi_unregister_transport(&iscsi_tcp_transport);
2318 }
2319
2320 module_init(iscsi_tcp_init);
2321 module_exit(iscsi_tcp_exit);