[SCSI] iscsi tcp: fix iscsi xmit state machine
[safe/jmp/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/blkdev.h>
33 #include <linux/crypto.h>
34 #include <linux/delay.h>
35 #include <linux/kfifo.h>
36 #include <linux/scatterlist.h>
37 #include <linux/mutex.h>
38 #include <net/tcp.h>
39 #include <scsi/scsi_cmnd.h>
40 #include <scsi/scsi_host.h>
41 #include <scsi/scsi.h>
42 #include <scsi/scsi_transport_iscsi.h>
43
44 #include "iscsi_tcp.h"
45
46 MODULE_AUTHOR("Dmitry Yusupov <dmitry_yus@yahoo.com>, "
47               "Alex Aizman <itn780@yahoo.com>");
48 MODULE_DESCRIPTION("iSCSI/TCP data-path");
49 MODULE_LICENSE("GPL");
50 /* #define DEBUG_TCP */
51 #define DEBUG_ASSERT
52
53 #ifdef DEBUG_TCP
54 #define debug_tcp(fmt...) printk(KERN_INFO "tcp: " fmt)
55 #else
56 #define debug_tcp(fmt...)
57 #endif
58
59 #ifndef DEBUG_ASSERT
60 #ifdef BUG_ON
61 #undef BUG_ON
62 #endif
63 #define BUG_ON(expr)
64 #endif
65
66 static unsigned int iscsi_max_lun = 512;
67 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
68
69 static inline void
70 iscsi_buf_init_iov(struct iscsi_buf *ibuf, char *vbuf, int size)
71 {
72         ibuf->sg.page = virt_to_page(vbuf);
73         ibuf->sg.offset = offset_in_page(vbuf);
74         ibuf->sg.length = size;
75         ibuf->sent = 0;
76         ibuf->use_sendmsg = 1;
77 }
78
79 static inline void
80 iscsi_buf_init_sg(struct iscsi_buf *ibuf, struct scatterlist *sg)
81 {
82         ibuf->sg.page = sg->page;
83         ibuf->sg.offset = sg->offset;
84         ibuf->sg.length = sg->length;
85         /*
86          * Fastpath: sg element fits into single page
87          */
88         if (sg->length + sg->offset <= PAGE_SIZE && !PageSlab(sg->page))
89                 ibuf->use_sendmsg = 0;
90         else
91                 ibuf->use_sendmsg = 1;
92         ibuf->sent = 0;
93 }
94
95 static inline int
96 iscsi_buf_left(struct iscsi_buf *ibuf)
97 {
98         int rc;
99
100         rc = ibuf->sg.length - ibuf->sent;
101         BUG_ON(rc < 0);
102         return rc;
103 }
104
105 static inline void
106 iscsi_hdr_digest(struct iscsi_conn *conn, struct iscsi_buf *buf,
107                  u8* crc)
108 {
109         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
110
111         crypto_hash_digest(&tcp_conn->tx_hash, &buf->sg, buf->sg.length, crc);
112         buf->sg.length += sizeof(u32);
113 }
114
115 static inline int
116 iscsi_hdr_extract(struct iscsi_tcp_conn *tcp_conn)
117 {
118         struct sk_buff *skb = tcp_conn->in.skb;
119
120         tcp_conn->in.zero_copy_hdr = 0;
121
122         if (tcp_conn->in.copy >= tcp_conn->hdr_size &&
123             tcp_conn->in_progress == IN_PROGRESS_WAIT_HEADER) {
124                 /*
125                  * Zero-copy PDU Header: using connection context
126                  * to store header pointer.
127                  */
128                 if (skb_shinfo(skb)->frag_list == NULL &&
129                     !skb_shinfo(skb)->nr_frags) {
130                         tcp_conn->in.hdr = (struct iscsi_hdr *)
131                                 ((char*)skb->data + tcp_conn->in.offset);
132                         tcp_conn->in.zero_copy_hdr = 1;
133                 } else {
134                         /* ignoring return code since we checked
135                          * in.copy before */
136                         skb_copy_bits(skb, tcp_conn->in.offset,
137                                 &tcp_conn->hdr, tcp_conn->hdr_size);
138                         tcp_conn->in.hdr = &tcp_conn->hdr;
139                 }
140                 tcp_conn->in.offset += tcp_conn->hdr_size;
141                 tcp_conn->in.copy -= tcp_conn->hdr_size;
142         } else {
143                 int hdr_remains;
144                 int copylen;
145
146                 /*
147                  * PDU header scattered across SKB's,
148                  * copying it... This'll happen quite rarely.
149                  */
150
151                 if (tcp_conn->in_progress == IN_PROGRESS_WAIT_HEADER)
152                         tcp_conn->in.hdr_offset = 0;
153
154                 hdr_remains = tcp_conn->hdr_size - tcp_conn->in.hdr_offset;
155                 BUG_ON(hdr_remains <= 0);
156
157                 copylen = min(tcp_conn->in.copy, hdr_remains);
158                 skb_copy_bits(skb, tcp_conn->in.offset,
159                         (char*)&tcp_conn->hdr + tcp_conn->in.hdr_offset,
160                         copylen);
161
162                 debug_tcp("PDU gather offset %d bytes %d in.offset %d "
163                        "in.copy %d\n", tcp_conn->in.hdr_offset, copylen,
164                        tcp_conn->in.offset, tcp_conn->in.copy);
165
166                 tcp_conn->in.offset += copylen;
167                 tcp_conn->in.copy -= copylen;
168                 if (copylen < hdr_remains)  {
169                         tcp_conn->in_progress = IN_PROGRESS_HEADER_GATHER;
170                         tcp_conn->in.hdr_offset += copylen;
171                         return -EAGAIN;
172                 }
173                 tcp_conn->in.hdr = &tcp_conn->hdr;
174                 tcp_conn->discontiguous_hdr_cnt++;
175                 tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
176         }
177
178         return 0;
179 }
180
181 /*
182  * must be called with session lock
183  */
184 static void
185 iscsi_tcp_cleanup_ctask(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
186 {
187         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
188         struct iscsi_r2t_info *r2t;
189         struct scsi_cmnd *sc;
190
191         /* flush ctask's r2t queues */
192         while (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*))) {
193                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
194                             sizeof(void*));
195                 debug_scsi("iscsi_tcp_cleanup_ctask pending r2t dropped\n");
196         }
197
198         sc = ctask->sc;
199         if (unlikely(!sc))
200                 return;
201
202         tcp_ctask->xmstate = XMSTATE_IDLE;
203         tcp_ctask->r2t = NULL;
204 }
205
206 /**
207  * iscsi_data_rsp - SCSI Data-In Response processing
208  * @conn: iscsi connection
209  * @ctask: scsi command task
210  **/
211 static int
212 iscsi_data_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
213 {
214         int rc;
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         rc = iscsi_check_assign_cmdsn(session, (struct iscsi_nopin*)rhdr);
223         if (rc)
224                 return rc;
225         /*
226          * setup Data-In byte counter (gets decremented..)
227          */
228         ctask->data_count = tcp_conn->in.datalen;
229
230         if (tcp_conn->in.datalen == 0)
231                 return 0;
232
233         if (tcp_ctask->exp_datasn != datasn) {
234                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->datasn(%d)\n",
235                           __FUNCTION__, tcp_ctask->exp_datasn, datasn);
236                 return ISCSI_ERR_DATASN;
237         }
238
239         tcp_ctask->exp_datasn++;
240
241         tcp_ctask->data_offset = be32_to_cpu(rhdr->offset);
242         if (tcp_ctask->data_offset + tcp_conn->in.datalen > sc->request_bufflen) {
243                 debug_tcp("%s: data_offset(%d) + data_len(%d) > total_length_in(%d)\n",
244                           __FUNCTION__, tcp_ctask->data_offset,
245                           tcp_conn->in.datalen, sc->request_bufflen);
246                 return ISCSI_ERR_DATA_OFFSET;
247         }
248
249         if (rhdr->flags & ISCSI_FLAG_DATA_STATUS) {
250                 conn->exp_statsn = be32_to_cpu(rhdr->statsn) + 1;
251                 if (rhdr->flags & ISCSI_FLAG_DATA_UNDERFLOW) {
252                         int res_count = be32_to_cpu(rhdr->residual_count);
253
254                         if (res_count > 0 &&
255                             res_count <= sc->request_bufflen) {
256                                 sc->resid = res_count;
257                                 sc->result = (DID_OK << 16) | rhdr->cmd_status;
258                         } else
259                                 sc->result = (DID_BAD_TARGET << 16) |
260                                         rhdr->cmd_status;
261                 } else if (rhdr->flags & ISCSI_FLAG_DATA_OVERFLOW) {
262                         sc->resid = be32_to_cpu(rhdr->residual_count);
263                         sc->result = (DID_OK << 16) | rhdr->cmd_status;
264                 } else
265                         sc->result = (DID_OK << 16) | rhdr->cmd_status;
266         }
267
268         conn->datain_pdus_cnt++;
269         return 0;
270 }
271
272 /**
273  * iscsi_solicit_data_init - initialize first Data-Out
274  * @conn: iscsi connection
275  * @ctask: scsi command task
276  * @r2t: R2T info
277  *
278  * Notes:
279  *      Initialize first Data-Out within this R2T sequence and finds
280  *      proper data_offset within this SCSI command.
281  *
282  *      This function is called with connection lock taken.
283  **/
284 static void
285 iscsi_solicit_data_init(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
286                         struct iscsi_r2t_info *r2t)
287 {
288         struct iscsi_data *hdr;
289         struct scsi_cmnd *sc = ctask->sc;
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         if (sc->use_sg) {
318                 int i, sg_count = 0;
319                 struct scatterlist *sg = sc->request_buffer;
320
321                 r2t->sg = NULL;
322                 for (i = 0; i < sc->use_sg; i++, sg += 1) {
323                         /* FIXME: prefetch ? */
324                         if (sg_count + sg->length > r2t->data_offset) {
325                                 int page_offset;
326
327                                 /* sg page found! */
328
329                                 /* offset within this page */
330                                 page_offset = r2t->data_offset - sg_count;
331
332                                 /* fill in this buffer */
333                                 iscsi_buf_init_sg(&r2t->sendbuf, sg);
334                                 r2t->sendbuf.sg.offset += page_offset;
335                                 r2t->sendbuf.sg.length -= page_offset;
336
337                                 /* xmit logic will continue with next one */
338                                 r2t->sg = sg + 1;
339                                 break;
340                         }
341                         sg_count += sg->length;
342                 }
343                 BUG_ON(r2t->sg == NULL);
344         } else {
345                 iscsi_buf_init_iov(&r2t->sendbuf,
346                             (char*)sc->request_buffer + r2t->data_offset,
347                             r2t->data_count);
348                 r2t->sg = NULL;
349         }
350 }
351
352 /**
353  * iscsi_r2t_rsp - iSCSI R2T Response processing
354  * @conn: iscsi connection
355  * @ctask: scsi command task
356  **/
357 static int
358 iscsi_r2t_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
359 {
360         struct iscsi_r2t_info *r2t;
361         struct iscsi_session *session = conn->session;
362         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
363         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
364         struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
365         int r2tsn = be32_to_cpu(rhdr->r2tsn);
366         int rc;
367
368         if (tcp_conn->in.datalen) {
369                 printk(KERN_ERR "iscsi_tcp: invalid R2t with datalen %d\n",
370                        tcp_conn->in.datalen);
371                 return ISCSI_ERR_DATALEN;
372         }
373
374         if (tcp_ctask->exp_datasn != r2tsn){
375                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
376                           __FUNCTION__, tcp_ctask->exp_datasn, r2tsn);
377                 return ISCSI_ERR_R2TSN;
378         }
379
380         rc = iscsi_check_assign_cmdsn(session, (struct iscsi_nopin*)rhdr);
381         if (rc)
382                 return rc;
383
384         /* fill-in new R2T associated with the task */
385         spin_lock(&session->lock);
386         if (!ctask->sc || ctask->mtask ||
387              session->state != ISCSI_STATE_LOGGED_IN) {
388                 printk(KERN_INFO "iscsi_tcp: dropping R2T itt %d in "
389                        "recovery...\n", ctask->itt);
390                 spin_unlock(&session->lock);
391                 return 0;
392         }
393
394         rc = __kfifo_get(tcp_ctask->r2tpool.queue, (void*)&r2t, sizeof(void*));
395         BUG_ON(!rc);
396
397         r2t->exp_statsn = rhdr->statsn;
398         r2t->data_length = be32_to_cpu(rhdr->data_length);
399         if (r2t->data_length == 0) {
400                 printk(KERN_ERR "iscsi_tcp: invalid R2T with zero data len\n");
401                 spin_unlock(&session->lock);
402                 return ISCSI_ERR_DATALEN;
403         }
404
405         if (r2t->data_length > session->max_burst)
406                 debug_scsi("invalid R2T with data len %u and max burst %u."
407                            "Attempting to execute request.\n",
408                             r2t->data_length, session->max_burst);
409
410         r2t->data_offset = be32_to_cpu(rhdr->data_offset);
411         if (r2t->data_offset + r2t->data_length > ctask->sc->request_bufflen) {
412                 spin_unlock(&session->lock);
413                 printk(KERN_ERR "iscsi_tcp: invalid R2T with data len %u at "
414                        "offset %u and total length %d\n", r2t->data_length,
415                        r2t->data_offset, ctask->sc->request_bufflen);
416                 return ISCSI_ERR_DATALEN;
417         }
418
419         r2t->ttt = rhdr->ttt; /* no flip */
420         r2t->solicit_datasn = 0;
421
422         iscsi_solicit_data_init(conn, ctask, r2t);
423
424         tcp_ctask->exp_datasn = r2tsn + 1;
425         __kfifo_put(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*));
426         tcp_ctask->xmstate |= XMSTATE_SOL_HDR_INIT;
427         list_move_tail(&ctask->running, &conn->xmitqueue);
428
429         scsi_queue_work(session->host, &conn->xmitwork);
430         conn->r2t_pdus_cnt++;
431         spin_unlock(&session->lock);
432
433         return 0;
434 }
435
436 static int
437 iscsi_tcp_hdr_recv(struct iscsi_conn *conn)
438 {
439         int rc = 0, opcode, ahslen;
440         struct iscsi_hdr *hdr;
441         struct iscsi_session *session = conn->session;
442         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
443         uint32_t cdgst, rdgst = 0, itt;
444
445         hdr = tcp_conn->in.hdr;
446
447         /* verify PDU length */
448         tcp_conn->in.datalen = ntoh24(hdr->dlength);
449         if (tcp_conn->in.datalen > conn->max_recv_dlength) {
450                 printk(KERN_ERR "iscsi_tcp: datalen %d > %d\n",
451                        tcp_conn->in.datalen, conn->max_recv_dlength);
452                 return ISCSI_ERR_DATALEN;
453         }
454         tcp_conn->data_copied = 0;
455
456         /* read AHS */
457         ahslen = hdr->hlength << 2;
458         tcp_conn->in.offset += ahslen;
459         tcp_conn->in.copy -= ahslen;
460         if (tcp_conn->in.copy < 0) {
461                 printk(KERN_ERR "iscsi_tcp: can't handle AHS with length "
462                        "%d bytes\n", ahslen);
463                 return ISCSI_ERR_AHSLEN;
464         }
465
466         /* calculate read padding */
467         tcp_conn->in.padding = tcp_conn->in.datalen & (ISCSI_PAD_LEN-1);
468         if (tcp_conn->in.padding) {
469                 tcp_conn->in.padding = ISCSI_PAD_LEN - tcp_conn->in.padding;
470                 debug_scsi("read padding %d bytes\n", tcp_conn->in.padding);
471         }
472
473         if (conn->hdrdgst_en) {
474                 struct scatterlist sg;
475
476                 sg_init_one(&sg, (u8 *)hdr,
477                             sizeof(struct iscsi_hdr) + ahslen);
478                 crypto_hash_digest(&tcp_conn->rx_hash, &sg, sg.length,
479                                    (u8 *)&cdgst);
480                 rdgst = *(uint32_t*)((char*)hdr + sizeof(struct iscsi_hdr) +
481                                      ahslen);
482                 if (cdgst != rdgst) {
483                         printk(KERN_ERR "iscsi_tcp: hdrdgst error "
484                                "recv 0x%x calc 0x%x\n", rdgst, cdgst);
485                         return ISCSI_ERR_HDR_DGST;
486                 }
487         }
488
489         opcode = hdr->opcode & ISCSI_OPCODE_MASK;
490         /* verify itt (itt encoding: age+cid+itt) */
491         rc = iscsi_verify_itt(conn, hdr, &itt);
492         if (rc == ISCSI_ERR_NO_SCSI_CMD) {
493                 tcp_conn->in.datalen = 0; /* force drop */
494                 return 0;
495         } else if (rc)
496                 return rc;
497
498         debug_tcp("opcode 0x%x offset %d copy %d ahslen %d datalen %d\n",
499                   opcode, tcp_conn->in.offset, tcp_conn->in.copy,
500                   ahslen, tcp_conn->in.datalen);
501
502         switch(opcode) {
503         case ISCSI_OP_SCSI_DATA_IN:
504                 tcp_conn->in.ctask = session->cmds[itt];
505                 rc = iscsi_data_rsp(conn, tcp_conn->in.ctask);
506                 if (rc)
507                         return rc;
508                 /* fall through */
509         case ISCSI_OP_SCSI_CMD_RSP:
510                 tcp_conn->in.ctask = session->cmds[itt];
511                 if (tcp_conn->in.datalen)
512                         goto copy_hdr;
513
514                 spin_lock(&session->lock);
515                 rc = __iscsi_complete_pdu(conn, hdr, NULL, 0);
516                 spin_unlock(&session->lock);
517                 break;
518         case ISCSI_OP_R2T:
519                 tcp_conn->in.ctask = session->cmds[itt];
520                 if (ahslen)
521                         rc = ISCSI_ERR_AHSLEN;
522                 else if (tcp_conn->in.ctask->sc->sc_data_direction ==
523                                                                 DMA_TO_DEVICE)
524                         rc = iscsi_r2t_rsp(conn, tcp_conn->in.ctask);
525                 else
526                         rc = ISCSI_ERR_PROTO;
527                 break;
528         case ISCSI_OP_LOGIN_RSP:
529         case ISCSI_OP_TEXT_RSP:
530         case ISCSI_OP_REJECT:
531         case ISCSI_OP_ASYNC_EVENT:
532                 /*
533                  * It is possible that we could get a PDU with a buffer larger
534                  * than 8K, but there are no targets that currently do this.
535                  * For now we fail until we find a vendor that needs it
536                  */
537                 if (ISCSI_DEF_MAX_RECV_SEG_LEN <
538                     tcp_conn->in.datalen) {
539                         printk(KERN_ERR "iscsi_tcp: received buffer of len %u "
540                               "but conn buffer is only %u (opcode %0x)\n",
541                               tcp_conn->in.datalen,
542                               ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
543                         rc = ISCSI_ERR_PROTO;
544                         break;
545                 }
546
547                 if (tcp_conn->in.datalen)
548                         goto copy_hdr;
549         /* fall through */
550         case ISCSI_OP_LOGOUT_RSP:
551         case ISCSI_OP_NOOP_IN:
552         case ISCSI_OP_SCSI_TMFUNC_RSP:
553                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
554                 break;
555         default:
556                 rc = ISCSI_ERR_BAD_OPCODE;
557                 break;
558         }
559
560         return rc;
561
562 copy_hdr:
563         /*
564          * if we did zero copy for the header but we will need multiple
565          * skbs to complete the command then we have to copy the header
566          * for later use
567          */
568         if (tcp_conn->in.zero_copy_hdr && tcp_conn->in.copy <=
569            (tcp_conn->in.datalen + tcp_conn->in.padding +
570             (conn->datadgst_en ? 4 : 0))) {
571                 debug_tcp("Copying header for later use. in.copy %d in.datalen"
572                           " %d\n", tcp_conn->in.copy, tcp_conn->in.datalen);
573                 memcpy(&tcp_conn->hdr, tcp_conn->in.hdr,
574                        sizeof(struct iscsi_hdr));
575                 tcp_conn->in.hdr = &tcp_conn->hdr;
576                 tcp_conn->in.zero_copy_hdr = 0;
577         }
578         return 0;
579 }
580
581 /**
582  * iscsi_ctask_copy - copy skb bits to the destanation cmd task
583  * @conn: iscsi tcp connection
584  * @ctask: scsi command task
585  * @buf: buffer to copy to
586  * @buf_size: size of buffer
587  * @offset: offset within the buffer
588  *
589  * Notes:
590  *      The function calls skb_copy_bits() and updates per-connection and
591  *      per-cmd byte counters.
592  *
593  *      Read counters (in bytes):
594  *
595  *      conn->in.offset         offset within in progress SKB
596  *      conn->in.copy           left to copy from in progress SKB
597  *                              including padding
598  *      conn->in.copied         copied already from in progress SKB
599  *      conn->data_copied       copied already from in progress buffer
600  *      ctask->sent             total bytes sent up to the MidLayer
601  *      ctask->data_count       left to copy from in progress Data-In
602  *      buf_left                left to copy from in progress buffer
603  **/
604 static inline int
605 iscsi_ctask_copy(struct iscsi_tcp_conn *tcp_conn, struct iscsi_cmd_task *ctask,
606                 void *buf, int buf_size, int offset)
607 {
608         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
609         int buf_left = buf_size - (tcp_conn->data_copied + offset);
610         unsigned size = min(tcp_conn->in.copy, buf_left);
611         int rc;
612
613         size = min(size, ctask->data_count);
614
615         debug_tcp("ctask_copy %d bytes at offset %d copied %d\n",
616                size, tcp_conn->in.offset, tcp_conn->in.copied);
617
618         BUG_ON(size <= 0);
619         BUG_ON(tcp_ctask->sent + size > ctask->sc->request_bufflen);
620
621         rc = skb_copy_bits(tcp_conn->in.skb, tcp_conn->in.offset,
622                            (char*)buf + (offset + tcp_conn->data_copied), size);
623         /* must fit into skb->len */
624         BUG_ON(rc);
625
626         tcp_conn->in.offset += size;
627         tcp_conn->in.copy -= size;
628         tcp_conn->in.copied += size;
629         tcp_conn->data_copied += size;
630         tcp_ctask->sent += size;
631         ctask->data_count -= size;
632
633         BUG_ON(tcp_conn->in.copy < 0);
634         BUG_ON(ctask->data_count < 0);
635
636         if (buf_size != (tcp_conn->data_copied + offset)) {
637                 if (!ctask->data_count) {
638                         BUG_ON(buf_size - tcp_conn->data_copied < 0);
639                         /* done with this PDU */
640                         return buf_size - tcp_conn->data_copied;
641                 }
642                 return -EAGAIN;
643         }
644
645         /* done with this buffer or with both - PDU and buffer */
646         tcp_conn->data_copied = 0;
647         return 0;
648 }
649
650 /**
651  * iscsi_tcp_copy - copy skb bits to the destanation buffer
652  * @conn: iscsi tcp connection
653  *
654  * Notes:
655  *      The function calls skb_copy_bits() and updates per-connection
656  *      byte counters.
657  **/
658 static inline int
659 iscsi_tcp_copy(struct iscsi_conn *conn, int buf_size)
660 {
661         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
662         int buf_left = buf_size - tcp_conn->data_copied;
663         int size = min(tcp_conn->in.copy, buf_left);
664         int rc;
665
666         debug_tcp("tcp_copy %d bytes at offset %d copied %d\n",
667                size, tcp_conn->in.offset, tcp_conn->data_copied);
668         BUG_ON(size <= 0);
669
670         rc = skb_copy_bits(tcp_conn->in.skb, tcp_conn->in.offset,
671                            (char*)conn->data + tcp_conn->data_copied, size);
672         BUG_ON(rc);
673
674         tcp_conn->in.offset += size;
675         tcp_conn->in.copy -= size;
676         tcp_conn->in.copied += size;
677         tcp_conn->data_copied += size;
678
679         if (buf_size != tcp_conn->data_copied)
680                 return -EAGAIN;
681
682         return 0;
683 }
684
685 static inline void
686 partial_sg_digest_update(struct hash_desc *desc, struct scatterlist *sg,
687                          int offset, int length)
688 {
689         struct scatterlist temp;
690
691         memcpy(&temp, sg, sizeof(struct scatterlist));
692         temp.offset = offset;
693         temp.length = length;
694         crypto_hash_update(desc, &temp, length);
695 }
696
697 static void
698 iscsi_recv_digest_update(struct iscsi_tcp_conn *tcp_conn, char* buf, int len)
699 {
700         struct scatterlist tmp;
701
702         sg_init_one(&tmp, buf, len);
703         crypto_hash_update(&tcp_conn->rx_hash, &tmp, len);
704 }
705
706 static int iscsi_scsi_data_in(struct iscsi_conn *conn)
707 {
708         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
709         struct iscsi_cmd_task *ctask = tcp_conn->in.ctask;
710         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
711         struct scsi_cmnd *sc = ctask->sc;
712         struct scatterlist *sg;
713         int i, offset, rc = 0;
714
715         BUG_ON((void*)ctask != sc->SCp.ptr);
716
717         /*
718          * copying Data-In into the Scsi_Cmnd
719          */
720         if (!sc->use_sg) {
721                 i = ctask->data_count;
722                 rc = iscsi_ctask_copy(tcp_conn, ctask, sc->request_buffer,
723                                       sc->request_bufflen,
724                                       tcp_ctask->data_offset);
725                 if (rc == -EAGAIN)
726                         return rc;
727                 if (conn->datadgst_en)
728                         iscsi_recv_digest_update(tcp_conn, sc->request_buffer,
729                                                  i);
730                 rc = 0;
731                 goto done;
732         }
733
734         offset = tcp_ctask->data_offset;
735         sg = sc->request_buffer;
736
737         if (tcp_ctask->data_offset)
738                 for (i = 0; i < tcp_ctask->sg_count; i++)
739                         offset -= sg[i].length;
740         /* we've passed through partial sg*/
741         if (offset < 0)
742                 offset = 0;
743
744         for (i = tcp_ctask->sg_count; i < sc->use_sg; i++) {
745                 char *dest;
746
747                 dest = kmap_atomic(sg[i].page, KM_SOFTIRQ0);
748                 rc = iscsi_ctask_copy(tcp_conn, ctask, dest + sg[i].offset,
749                                       sg[i].length, offset);
750                 kunmap_atomic(dest, KM_SOFTIRQ0);
751                 if (rc == -EAGAIN)
752                         /* continue with the next SKB/PDU */
753                         return rc;
754                 if (!rc) {
755                         if (conn->datadgst_en) {
756                                 if (!offset)
757                                         crypto_hash_update(
758                                                         &tcp_conn->rx_hash,
759                                                         &sg[i], sg[i].length);
760                                 else
761                                         partial_sg_digest_update(
762                                                         &tcp_conn->rx_hash,
763                                                         &sg[i],
764                                                         sg[i].offset + offset,
765                                                         sg[i].length - offset);
766                         }
767                         offset = 0;
768                         tcp_ctask->sg_count++;
769                 }
770
771                 if (!ctask->data_count) {
772                         if (rc && conn->datadgst_en)
773                                 /*
774                                  * data-in is complete, but buffer not...
775                                  */
776                                 partial_sg_digest_update(&tcp_conn->rx_hash,
777                                                          &sg[i],
778                                                          sg[i].offset,
779                                                          sg[i].length-rc);
780                         rc = 0;
781                         break;
782                 }
783
784                 if (!tcp_conn->in.copy)
785                         return -EAGAIN;
786         }
787         BUG_ON(ctask->data_count);
788
789 done:
790         /* check for non-exceptional status */
791         if (tcp_conn->in.hdr->flags & ISCSI_FLAG_DATA_STATUS) {
792                 debug_scsi("done [sc %lx res %d itt 0x%x flags 0x%x]\n",
793                            (long)sc, sc->result, ctask->itt,
794                            tcp_conn->in.hdr->flags);
795                 spin_lock(&conn->session->lock);
796                 __iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
797                 spin_unlock(&conn->session->lock);
798         }
799
800         return rc;
801 }
802
803 static int
804 iscsi_data_recv(struct iscsi_conn *conn)
805 {
806         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
807         int rc = 0, opcode;
808
809         opcode = tcp_conn->in.hdr->opcode & ISCSI_OPCODE_MASK;
810         switch (opcode) {
811         case ISCSI_OP_SCSI_DATA_IN:
812                 rc = iscsi_scsi_data_in(conn);
813                 break;
814         case ISCSI_OP_SCSI_CMD_RSP:
815         case ISCSI_OP_TEXT_RSP:
816         case ISCSI_OP_LOGIN_RSP:
817         case ISCSI_OP_ASYNC_EVENT:
818         case ISCSI_OP_REJECT:
819                 /*
820                  * Collect data segment to the connection's data
821                  * placeholder
822                  */
823                 if (iscsi_tcp_copy(conn, tcp_conn->in.datalen)) {
824                         rc = -EAGAIN;
825                         goto exit;
826                 }
827
828                 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, conn->data,
829                                         tcp_conn->in.datalen);
830                 if (!rc && conn->datadgst_en && opcode != ISCSI_OP_LOGIN_RSP)
831                         iscsi_recv_digest_update(tcp_conn, conn->data,
832                                                 tcp_conn->in.datalen);
833                 break;
834         default:
835                 BUG_ON(1);
836         }
837 exit:
838         return rc;
839 }
840
841 /**
842  * iscsi_tcp_data_recv - TCP receive in sendfile fashion
843  * @rd_desc: read descriptor
844  * @skb: socket buffer
845  * @offset: offset in skb
846  * @len: skb->len - offset
847  **/
848 static int
849 iscsi_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
850                 unsigned int offset, size_t len)
851 {
852         int rc;
853         struct iscsi_conn *conn = rd_desc->arg.data;
854         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
855         int processed;
856         char pad[ISCSI_PAD_LEN];
857         struct scatterlist sg;
858
859         /*
860          * Save current SKB and its offset in the corresponding
861          * connection context.
862          */
863         tcp_conn->in.copy = skb->len - offset;
864         tcp_conn->in.offset = offset;
865         tcp_conn->in.skb = skb;
866         tcp_conn->in.len = tcp_conn->in.copy;
867         BUG_ON(tcp_conn->in.copy <= 0);
868         debug_tcp("in %d bytes\n", tcp_conn->in.copy);
869
870 more:
871         tcp_conn->in.copied = 0;
872         rc = 0;
873
874         if (unlikely(conn->suspend_rx)) {
875                 debug_tcp("conn %d Rx suspended!\n", conn->id);
876                 return 0;
877         }
878
879         if (tcp_conn->in_progress == IN_PROGRESS_WAIT_HEADER ||
880             tcp_conn->in_progress == IN_PROGRESS_HEADER_GATHER) {
881                 rc = iscsi_hdr_extract(tcp_conn);
882                 if (rc) {
883                        if (rc == -EAGAIN)
884                                 goto nomore;
885                        else {
886                                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
887                                 return 0;
888                        }
889                 }
890
891                 /*
892                  * Verify and process incoming PDU header.
893                  */
894                 rc = iscsi_tcp_hdr_recv(conn);
895                 if (!rc && tcp_conn->in.datalen) {
896                         if (conn->datadgst_en)
897                                 crypto_hash_init(&tcp_conn->rx_hash);
898                         tcp_conn->in_progress = IN_PROGRESS_DATA_RECV;
899                 } else if (rc) {
900                         iscsi_conn_failure(conn, rc);
901                         return 0;
902                 }
903         }
904
905         if (tcp_conn->in_progress == IN_PROGRESS_DDIGEST_RECV) {
906                 uint32_t recv_digest;
907
908                 debug_tcp("extra data_recv offset %d copy %d\n",
909                           tcp_conn->in.offset, tcp_conn->in.copy);
910                 rc = iscsi_tcp_copy(conn, sizeof(uint32_t));
911                 if (rc) {
912                         if (rc == -EAGAIN)
913                                 goto again;
914                         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
915                         return 0;
916                 }
917
918                 memcpy(&recv_digest, conn->data, sizeof(uint32_t));
919                 if (recv_digest != tcp_conn->in.datadgst) {
920                         debug_tcp("iscsi_tcp: data digest error!"
921                                   "0x%x != 0x%x\n", recv_digest,
922                                   tcp_conn->in.datadgst);
923                         iscsi_conn_failure(conn, ISCSI_ERR_DATA_DGST);
924                         return 0;
925                 } else {
926                         debug_tcp("iscsi_tcp: data digest match!"
927                                   "0x%x == 0x%x\n", recv_digest,
928                                   tcp_conn->in.datadgst);
929                         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
930                 }
931         }
932
933         if (tcp_conn->in_progress == IN_PROGRESS_DATA_RECV &&
934            tcp_conn->in.copy) {
935
936                 debug_tcp("data_recv offset %d copy %d\n",
937                        tcp_conn->in.offset, tcp_conn->in.copy);
938
939                 rc = iscsi_data_recv(conn);
940                 if (rc) {
941                         if (rc == -EAGAIN)
942                                 goto again;
943                         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
944                         return 0;
945                 }
946                 tcp_conn->in.copy -= tcp_conn->in.padding;
947                 tcp_conn->in.offset += tcp_conn->in.padding;
948                 if (conn->datadgst_en) {
949                         if (tcp_conn->in.padding) {
950                                 debug_tcp("padding -> %d\n",
951                                           tcp_conn->in.padding);
952                                 memset(pad, 0, tcp_conn->in.padding);
953                                 sg_init_one(&sg, pad, tcp_conn->in.padding);
954                                 crypto_hash_update(&tcp_conn->rx_hash,
955                                                    &sg, sg.length);
956                         }
957                         crypto_hash_final(&tcp_conn->rx_hash,
958                                           (u8 *) &tcp_conn->in.datadgst);
959                         debug_tcp("rx digest 0x%x\n", tcp_conn->in.datadgst);
960                         tcp_conn->in_progress = IN_PROGRESS_DDIGEST_RECV;
961                         tcp_conn->data_copied = 0;
962                 } else
963                         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
964         }
965
966         debug_tcp("f, processed %d from out of %d padding %d\n",
967                tcp_conn->in.offset - offset, (int)len, tcp_conn->in.padding);
968         BUG_ON(tcp_conn->in.offset - offset > len);
969
970         if (tcp_conn->in.offset - offset != len) {
971                 debug_tcp("continue to process %d bytes\n",
972                        (int)len - (tcp_conn->in.offset - offset));
973                 goto more;
974         }
975
976 nomore:
977         processed = tcp_conn->in.offset - offset;
978         BUG_ON(processed == 0);
979         return processed;
980
981 again:
982         processed = tcp_conn->in.offset - offset;
983         debug_tcp("c, processed %d from out of %d rd_desc_cnt %d\n",
984                   processed, (int)len, (int)rd_desc->count);
985         BUG_ON(processed == 0);
986         BUG_ON(processed > len);
987
988         conn->rxdata_octets += processed;
989         return processed;
990 }
991
992 static void
993 iscsi_tcp_data_ready(struct sock *sk, int flag)
994 {
995         struct iscsi_conn *conn = sk->sk_user_data;
996         read_descriptor_t rd_desc;
997
998         read_lock(&sk->sk_callback_lock);
999
1000         /*
1001          * Use rd_desc to pass 'conn' to iscsi_tcp_data_recv.
1002          * We set count to 1 because we want the network layer to
1003          * hand us all the skbs that are available. iscsi_tcp_data_recv
1004          * handled pdus that cross buffers or pdus that still need data.
1005          */
1006         rd_desc.arg.data = conn;
1007         rd_desc.count = 1;
1008         tcp_read_sock(sk, &rd_desc, iscsi_tcp_data_recv);
1009
1010         read_unlock(&sk->sk_callback_lock);
1011 }
1012
1013 static void
1014 iscsi_tcp_state_change(struct sock *sk)
1015 {
1016         struct iscsi_tcp_conn *tcp_conn;
1017         struct iscsi_conn *conn;
1018         struct iscsi_session *session;
1019         void (*old_state_change)(struct sock *);
1020
1021         read_lock(&sk->sk_callback_lock);
1022
1023         conn = (struct iscsi_conn*)sk->sk_user_data;
1024         session = conn->session;
1025
1026         if ((sk->sk_state == TCP_CLOSE_WAIT ||
1027              sk->sk_state == TCP_CLOSE) &&
1028             !atomic_read(&sk->sk_rmem_alloc)) {
1029                 debug_tcp("iscsi_tcp_state_change: TCP_CLOSE|TCP_CLOSE_WAIT\n");
1030                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1031         }
1032
1033         tcp_conn = conn->dd_data;
1034         old_state_change = tcp_conn->old_state_change;
1035
1036         read_unlock(&sk->sk_callback_lock);
1037
1038         old_state_change(sk);
1039 }
1040
1041 /**
1042  * iscsi_write_space - Called when more output buffer space is available
1043  * @sk: socket space is available for
1044  **/
1045 static void
1046 iscsi_write_space(struct sock *sk)
1047 {
1048         struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
1049         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1050
1051         tcp_conn->old_write_space(sk);
1052         debug_tcp("iscsi_write_space: cid %d\n", conn->id);
1053         scsi_queue_work(conn->session->host, &conn->xmitwork);
1054 }
1055
1056 static void
1057 iscsi_conn_set_callbacks(struct iscsi_conn *conn)
1058 {
1059         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1060         struct sock *sk = tcp_conn->sock->sk;
1061
1062         /* assign new callbacks */
1063         write_lock_bh(&sk->sk_callback_lock);
1064         sk->sk_user_data = conn;
1065         tcp_conn->old_data_ready = sk->sk_data_ready;
1066         tcp_conn->old_state_change = sk->sk_state_change;
1067         tcp_conn->old_write_space = sk->sk_write_space;
1068         sk->sk_data_ready = iscsi_tcp_data_ready;
1069         sk->sk_state_change = iscsi_tcp_state_change;
1070         sk->sk_write_space = iscsi_write_space;
1071         write_unlock_bh(&sk->sk_callback_lock);
1072 }
1073
1074 static void
1075 iscsi_conn_restore_callbacks(struct iscsi_tcp_conn *tcp_conn)
1076 {
1077         struct sock *sk = tcp_conn->sock->sk;
1078
1079         /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
1080         write_lock_bh(&sk->sk_callback_lock);
1081         sk->sk_user_data    = NULL;
1082         sk->sk_data_ready   = tcp_conn->old_data_ready;
1083         sk->sk_state_change = tcp_conn->old_state_change;
1084         sk->sk_write_space  = tcp_conn->old_write_space;
1085         sk->sk_no_check  = 0;
1086         write_unlock_bh(&sk->sk_callback_lock);
1087 }
1088
1089 /**
1090  * iscsi_send - generic send routine
1091  * @sk: kernel's socket
1092  * @buf: buffer to write from
1093  * @size: actual size to write
1094  * @flags: socket's flags
1095  */
1096 static inline int
1097 iscsi_send(struct iscsi_conn *conn, struct iscsi_buf *buf, int size, int flags)
1098 {
1099         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1100         struct socket *sk = tcp_conn->sock;
1101         int offset = buf->sg.offset + buf->sent, res;
1102
1103         /*
1104          * if we got use_sg=0 or are sending something we kmallocd
1105          * then we did not have to do kmap (kmap returns page_address)
1106          *
1107          * if we got use_sg > 0, but had to drop down, we do not
1108          * set clustering so this should only happen for that
1109          * slab case.
1110          */
1111         if (buf->use_sendmsg)
1112                 res = sock_no_sendpage(sk, buf->sg.page, offset, size, flags);
1113         else
1114                 res = tcp_conn->sendpage(sk, buf->sg.page, offset, size, flags);
1115
1116         if (res >= 0) {
1117                 conn->txdata_octets += res;
1118                 buf->sent += res;
1119                 return res;
1120         }
1121
1122         tcp_conn->sendpage_failures_cnt++;
1123         if (res == -EAGAIN)
1124                 res = -ENOBUFS;
1125         else
1126                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1127         return res;
1128 }
1129
1130 /**
1131  * iscsi_sendhdr - send PDU Header via tcp_sendpage()
1132  * @conn: iscsi connection
1133  * @buf: buffer to write from
1134  * @datalen: lenght of data to be sent after the header
1135  *
1136  * Notes:
1137  *      (Tx, Fast Path)
1138  **/
1139 static inline int
1140 iscsi_sendhdr(struct iscsi_conn *conn, struct iscsi_buf *buf, int datalen)
1141 {
1142         int flags = 0; /* MSG_DONTWAIT; */
1143         int res, size;
1144
1145         size = buf->sg.length - buf->sent;
1146         BUG_ON(buf->sent + size > buf->sg.length);
1147         if (buf->sent + size != buf->sg.length || datalen)
1148                 flags |= MSG_MORE;
1149
1150         res = iscsi_send(conn, buf, size, flags);
1151         debug_tcp("sendhdr %d bytes, sent %d res %d\n", size, buf->sent, res);
1152         if (res >= 0) {
1153                 if (size != res)
1154                         return -EAGAIN;
1155                 return 0;
1156         }
1157
1158         return res;
1159 }
1160
1161 /**
1162  * iscsi_sendpage - send one page of iSCSI Data-Out.
1163  * @conn: iscsi connection
1164  * @buf: buffer to write from
1165  * @count: remaining data
1166  * @sent: number of bytes sent
1167  *
1168  * Notes:
1169  *      (Tx, Fast Path)
1170  **/
1171 static inline int
1172 iscsi_sendpage(struct iscsi_conn *conn, struct iscsi_buf *buf,
1173                int *count, int *sent)
1174 {
1175         int flags = 0; /* MSG_DONTWAIT; */
1176         int res, size;
1177
1178         size = buf->sg.length - buf->sent;
1179         BUG_ON(buf->sent + size > buf->sg.length);
1180         if (size > *count)
1181                 size = *count;
1182         if (buf->sent + size != buf->sg.length || *count != size)
1183                 flags |= MSG_MORE;
1184
1185         res = iscsi_send(conn, buf, size, flags);
1186         debug_tcp("sendpage: %d bytes, sent %d left %d sent %d res %d\n",
1187                   size, buf->sent, *count, *sent, res);
1188         if (res >= 0) {
1189                 *count -= res;
1190                 *sent += res;
1191                 if (size != res)
1192                         return -EAGAIN;
1193                 return 0;
1194         }
1195
1196         return res;
1197 }
1198
1199 static inline void
1200 iscsi_data_digest_init(struct iscsi_tcp_conn *tcp_conn,
1201                       struct iscsi_tcp_cmd_task *tcp_ctask)
1202 {
1203         crypto_hash_init(&tcp_conn->tx_hash);
1204         tcp_ctask->digest_count = 4;
1205 }
1206
1207 /**
1208  * iscsi_solicit_data_cont - initialize next Data-Out
1209  * @conn: iscsi connection
1210  * @ctask: scsi command task
1211  * @r2t: R2T info
1212  * @left: bytes left to transfer
1213  *
1214  * Notes:
1215  *      Initialize next Data-Out within this R2T sequence and continue
1216  *      to process next Scatter-Gather element(if any) of this SCSI command.
1217  *
1218  *      Called under connection lock.
1219  **/
1220 static void
1221 iscsi_solicit_data_cont(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
1222                         struct iscsi_r2t_info *r2t, int left)
1223 {
1224         struct iscsi_data *hdr;
1225         struct scsi_cmnd *sc = ctask->sc;
1226         int new_offset;
1227
1228         hdr = &r2t->dtask.hdr;
1229         memset(hdr, 0, sizeof(struct iscsi_data));
1230         hdr->ttt = r2t->ttt;
1231         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
1232         r2t->solicit_datasn++;
1233         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
1234         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
1235         hdr->itt = ctask->hdr->itt;
1236         hdr->exp_statsn = r2t->exp_statsn;
1237         new_offset = r2t->data_offset + r2t->sent;
1238         hdr->offset = cpu_to_be32(new_offset);
1239         if (left > conn->max_xmit_dlength) {
1240                 hton24(hdr->dlength, conn->max_xmit_dlength);
1241                 r2t->data_count = conn->max_xmit_dlength;
1242         } else {
1243                 hton24(hdr->dlength, left);
1244                 r2t->data_count = left;
1245                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
1246         }
1247         conn->dataout_pdus_cnt++;
1248
1249         iscsi_buf_init_iov(&r2t->headbuf, (char*)hdr,
1250                            sizeof(struct iscsi_hdr));
1251
1252         if (iscsi_buf_left(&r2t->sendbuf))
1253                 return;
1254
1255         if (sc->use_sg) {
1256                 iscsi_buf_init_sg(&r2t->sendbuf, r2t->sg);
1257                 r2t->sg += 1;
1258         } else {
1259                 iscsi_buf_init_iov(&r2t->sendbuf,
1260                             (char*)sc->request_buffer + new_offset,
1261                             r2t->data_count);
1262                 r2t->sg = NULL;
1263         }
1264 }
1265
1266 static void iscsi_set_padding(struct iscsi_tcp_cmd_task *tcp_ctask,
1267                               unsigned long len)
1268 {
1269         tcp_ctask->pad_count = len & (ISCSI_PAD_LEN - 1);
1270         if (!tcp_ctask->pad_count)
1271                 return;
1272
1273         tcp_ctask->pad_count = ISCSI_PAD_LEN - tcp_ctask->pad_count;
1274         debug_scsi("write padding %d bytes\n", tcp_ctask->pad_count);
1275         tcp_ctask->xmstate |= XMSTATE_W_PAD;
1276 }
1277
1278 /**
1279  * iscsi_tcp_cmd_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
1280  * @conn: iscsi connection
1281  * @ctask: scsi command task
1282  * @sc: scsi command
1283  **/
1284 static void
1285 iscsi_tcp_cmd_init(struct iscsi_cmd_task *ctask)
1286 {
1287         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1288
1289         BUG_ON(__kfifo_len(tcp_ctask->r2tqueue));
1290         tcp_ctask->xmstate = XMSTATE_CMD_HDR_INIT;
1291 }
1292
1293 /**
1294  * iscsi_tcp_mtask_xmit - xmit management(immediate) task
1295  * @conn: iscsi connection
1296  * @mtask: task management task
1297  *
1298  * Notes:
1299  *      The function can return -EAGAIN in which case caller must
1300  *      call it again later, or recover. '0' return code means successful
1301  *      xmit.
1302  *
1303  *      Management xmit state machine consists of these states:
1304  *              XMSTATE_IMM_HDR_INIT    - calculate digest of PDU Header
1305  *              XMSTATE_IMM_HDR         - PDU Header xmit in progress
1306  *              XMSTATE_IMM_DATA        - PDU Data xmit in progress
1307  *              XMSTATE_IDLE            - management PDU is done
1308  **/
1309 static int
1310 iscsi_tcp_mtask_xmit(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1311 {
1312         struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
1313         int rc;
1314
1315         debug_scsi("mtask deq [cid %d state %x itt 0x%x]\n",
1316                 conn->id, tcp_mtask->xmstate, mtask->itt);
1317
1318         if (tcp_mtask->xmstate & XMSTATE_IMM_HDR_INIT) {
1319                 iscsi_buf_init_iov(&tcp_mtask->headbuf, (char*)mtask->hdr,
1320                                    sizeof(struct iscsi_hdr));
1321
1322                 if (mtask->data_count) {
1323                         tcp_mtask->xmstate |= XMSTATE_IMM_DATA;
1324                         iscsi_buf_init_iov(&tcp_mtask->sendbuf,
1325                                            (char*)mtask->data,
1326                                            mtask->data_count);
1327                 }
1328
1329                 if (conn->c_stage != ISCSI_CONN_INITIAL_STAGE &&
1330                     conn->stop_stage != STOP_CONN_RECOVER &&
1331                     conn->hdrdgst_en)
1332                         iscsi_hdr_digest(conn, &tcp_mtask->headbuf,
1333                                         (u8*)tcp_mtask->hdrext);
1334
1335                 tcp_mtask->sent = 0;
1336                 tcp_mtask->xmstate &= ~XMSTATE_IMM_HDR_INIT;
1337                 tcp_mtask->xmstate |= XMSTATE_IMM_HDR;
1338         }
1339
1340         if (tcp_mtask->xmstate & XMSTATE_IMM_HDR) {
1341                 rc = iscsi_sendhdr(conn, &tcp_mtask->headbuf,
1342                                    mtask->data_count);
1343                 if (rc)
1344                         return rc;
1345                 tcp_mtask->xmstate &= ~XMSTATE_IMM_HDR;
1346         }
1347
1348         if (tcp_mtask->xmstate & XMSTATE_IMM_DATA) {
1349                 BUG_ON(!mtask->data_count);
1350                 tcp_mtask->xmstate &= ~XMSTATE_IMM_DATA;
1351                 /* FIXME: implement.
1352                  * Virtual buffer could be spreaded across multiple pages...
1353                  */
1354                 do {
1355                         int rc;
1356
1357                         rc = iscsi_sendpage(conn, &tcp_mtask->sendbuf,
1358                                         &mtask->data_count, &tcp_mtask->sent);
1359                         if (rc) {
1360                                 tcp_mtask->xmstate |= XMSTATE_IMM_DATA;
1361                                 return rc;
1362                         }
1363                 } while (mtask->data_count);
1364         }
1365
1366         BUG_ON(tcp_mtask->xmstate != XMSTATE_IDLE);
1367         if (mtask->hdr->itt == RESERVED_ITT) {
1368                 struct iscsi_session *session = conn->session;
1369
1370                 spin_lock_bh(&session->lock);
1371                 list_del(&conn->mtask->running);
1372                 __kfifo_put(session->mgmtpool.queue, (void*)&conn->mtask,
1373                             sizeof(void*));
1374                 spin_unlock_bh(&session->lock);
1375         }
1376         return 0;
1377 }
1378
1379 static int
1380 iscsi_send_cmd_hdr(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1381 {
1382         struct scsi_cmnd *sc = ctask->sc;
1383         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1384         int rc = 0;
1385
1386         if (tcp_ctask->xmstate & XMSTATE_CMD_HDR_INIT) {
1387                 tcp_ctask->sent = 0;
1388                 tcp_ctask->sg_count = 0;
1389                 tcp_ctask->exp_datasn = 0;
1390
1391                 if (sc->sc_data_direction == DMA_TO_DEVICE) {
1392                         if (sc->use_sg) {
1393                                 struct scatterlist *sg = sc->request_buffer;
1394
1395                                 iscsi_buf_init_sg(&tcp_ctask->sendbuf, sg);
1396                                 tcp_ctask->sg = sg + 1;
1397                                 tcp_ctask->bad_sg = sg + sc->use_sg;
1398                         } else {
1399                                 iscsi_buf_init_iov(&tcp_ctask->sendbuf,
1400                                                    sc->request_buffer,
1401                                                    sc->request_bufflen);
1402                                 tcp_ctask->sg = NULL;
1403                                 tcp_ctask->bad_sg = NULL;
1404                         }
1405
1406                         debug_scsi("cmd [itt 0x%x total %d imm_data %d "
1407                                    "unsol count %d, unsol offset %d]\n",
1408                                    ctask->itt, sc->request_bufflen,
1409                                    ctask->imm_count, ctask->unsol_count,
1410                                    ctask->unsol_offset);
1411                 }
1412
1413                 iscsi_buf_init_iov(&tcp_ctask->headbuf, (char*)ctask->hdr,
1414                                   sizeof(struct iscsi_hdr));
1415
1416                 if (conn->hdrdgst_en)
1417                         iscsi_hdr_digest(conn, &tcp_ctask->headbuf,
1418                                          (u8*)tcp_ctask->hdrext);
1419                 tcp_ctask->xmstate &= ~XMSTATE_CMD_HDR_INIT;
1420                 tcp_ctask->xmstate |= XMSTATE_CMD_HDR_XMIT;
1421         }
1422
1423         if (tcp_ctask->xmstate & XMSTATE_CMD_HDR_XMIT) {
1424                 rc = iscsi_sendhdr(conn, &tcp_ctask->headbuf, ctask->imm_count);
1425                 if (rc)
1426                         return rc;
1427                 tcp_ctask->xmstate &= ~XMSTATE_CMD_HDR_XMIT;
1428
1429                 if (sc->sc_data_direction != DMA_TO_DEVICE)
1430                         return 0;
1431
1432                 if (ctask->imm_count) {
1433                         tcp_ctask->xmstate |= XMSTATE_IMM_DATA;
1434                         iscsi_set_padding(tcp_ctask, ctask->imm_count);
1435
1436                         if (ctask->conn->datadgst_en) {
1437                                 iscsi_data_digest_init(ctask->conn->dd_data,
1438                                                        tcp_ctask);
1439                                 tcp_ctask->immdigest = 0;
1440                         }
1441                 }
1442
1443                 if (ctask->unsol_count)
1444                         tcp_ctask->xmstate |=
1445                                         XMSTATE_UNS_HDR | XMSTATE_UNS_INIT;
1446         }
1447         return rc;
1448 }
1449
1450 static int
1451 iscsi_send_padding(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1452 {
1453         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1454         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1455         int sent = 0, rc;
1456
1457         if (tcp_ctask->xmstate & XMSTATE_W_PAD) {
1458                 iscsi_buf_init_iov(&tcp_ctask->sendbuf, (char*)&tcp_ctask->pad,
1459                                    tcp_ctask->pad_count);
1460                 if (conn->datadgst_en)
1461                         crypto_hash_update(&tcp_conn->tx_hash,
1462                                            &tcp_ctask->sendbuf.sg,
1463                                            tcp_ctask->sendbuf.sg.length);
1464         } else if (!(tcp_ctask->xmstate & XMSTATE_W_RESEND_PAD))
1465                 return 0;
1466
1467         tcp_ctask->xmstate &= ~XMSTATE_W_PAD;
1468         tcp_ctask->xmstate &= ~XMSTATE_W_RESEND_PAD;
1469         debug_scsi("sending %d pad bytes for itt 0x%x\n",
1470                    tcp_ctask->pad_count, ctask->itt);
1471         rc = iscsi_sendpage(conn, &tcp_ctask->sendbuf, &tcp_ctask->pad_count,
1472                            &sent);
1473         if (rc) {
1474                 debug_scsi("padding send failed %d\n", rc);
1475                 tcp_ctask->xmstate |= XMSTATE_W_RESEND_PAD;
1476         }
1477         return rc;
1478 }
1479
1480 static int
1481 iscsi_send_digest(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
1482                         struct iscsi_buf *buf, uint32_t *digest)
1483 {
1484         struct iscsi_tcp_cmd_task *tcp_ctask;
1485         struct iscsi_tcp_conn *tcp_conn;
1486         int rc, sent = 0;
1487
1488         if (!conn->datadgst_en)
1489                 return 0;
1490
1491         tcp_ctask = ctask->dd_data;
1492         tcp_conn = conn->dd_data;
1493
1494         if (!(tcp_ctask->xmstate & XMSTATE_W_RESEND_DATA_DIGEST)) {
1495                 crypto_hash_final(&tcp_conn->tx_hash, (u8*)digest);
1496                 iscsi_buf_init_iov(buf, (char*)digest, 4);
1497         }
1498         tcp_ctask->xmstate &= ~XMSTATE_W_RESEND_DATA_DIGEST;
1499
1500         rc = iscsi_sendpage(conn, buf, &tcp_ctask->digest_count, &sent);
1501         if (!rc)
1502                 debug_scsi("sent digest 0x%x for itt 0x%x\n", *digest,
1503                           ctask->itt);
1504         else {
1505                 debug_scsi("sending digest 0x%x failed for itt 0x%x!\n",
1506                           *digest, ctask->itt);
1507                 tcp_ctask->xmstate |= XMSTATE_W_RESEND_DATA_DIGEST;
1508         }
1509         return rc;
1510 }
1511
1512 static int
1513 iscsi_send_data(struct iscsi_cmd_task *ctask, struct iscsi_buf *sendbuf,
1514                 struct scatterlist **sg, int *sent, int *count,
1515                 struct iscsi_buf *digestbuf, uint32_t *digest)
1516 {
1517         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1518         struct iscsi_conn *conn = ctask->conn;
1519         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1520         int rc, buf_sent, offset;
1521
1522         while (*count) {
1523                 buf_sent = 0;
1524                 offset = sendbuf->sent;
1525
1526                 rc = iscsi_sendpage(conn, sendbuf, count, &buf_sent);
1527                 *sent = *sent + buf_sent;
1528                 if (buf_sent && conn->datadgst_en)
1529                         partial_sg_digest_update(&tcp_conn->tx_hash,
1530                                 &sendbuf->sg, sendbuf->sg.offset + offset,
1531                                 buf_sent);
1532                 if (!iscsi_buf_left(sendbuf) && *sg != tcp_ctask->bad_sg) {
1533                         iscsi_buf_init_sg(sendbuf, *sg);
1534                         *sg = *sg + 1;
1535                 }
1536
1537                 if (rc)
1538                         return rc;
1539         }
1540
1541         rc = iscsi_send_padding(conn, ctask);
1542         if (rc)
1543                 return rc;
1544
1545         return iscsi_send_digest(conn, ctask, digestbuf, digest);
1546 }
1547
1548 static int
1549 iscsi_send_unsol_hdr(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1550 {
1551         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1552         struct iscsi_data_task *dtask;
1553         int rc;
1554
1555         tcp_ctask->xmstate |= XMSTATE_UNS_DATA;
1556         if (tcp_ctask->xmstate & XMSTATE_UNS_INIT) {
1557                 dtask = &tcp_ctask->unsol_dtask;
1558
1559                 iscsi_prep_unsolicit_data_pdu(ctask, &dtask->hdr);
1560                 iscsi_buf_init_iov(&tcp_ctask->headbuf, (char*)&dtask->hdr,
1561                                    sizeof(struct iscsi_hdr));
1562                 if (conn->hdrdgst_en)
1563                         iscsi_hdr_digest(conn, &tcp_ctask->headbuf,
1564                                         (u8*)dtask->hdrext);
1565
1566                 tcp_ctask->xmstate &= ~XMSTATE_UNS_INIT;
1567                 iscsi_set_padding(tcp_ctask, ctask->data_count);
1568         }
1569
1570         rc = iscsi_sendhdr(conn, &tcp_ctask->headbuf, ctask->data_count);
1571         if (rc) {
1572                 tcp_ctask->xmstate &= ~XMSTATE_UNS_DATA;
1573                 tcp_ctask->xmstate |= XMSTATE_UNS_HDR;
1574                 return rc;
1575         }
1576
1577         if (conn->datadgst_en) {
1578                 dtask = &tcp_ctask->unsol_dtask;
1579                 iscsi_data_digest_init(ctask->conn->dd_data, tcp_ctask);
1580                 dtask->digest = 0;
1581         }
1582
1583         debug_scsi("uns dout [itt 0x%x dlen %d sent %d]\n",
1584                    ctask->itt, ctask->unsol_count, tcp_ctask->sent);
1585         return 0;
1586 }
1587
1588 static int
1589 iscsi_send_unsol_pdu(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1590 {
1591         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1592         int rc;
1593
1594         if (tcp_ctask->xmstate & XMSTATE_UNS_HDR) {
1595                 BUG_ON(!ctask->unsol_count);
1596                 tcp_ctask->xmstate &= ~XMSTATE_UNS_HDR;
1597 send_hdr:
1598                 rc = iscsi_send_unsol_hdr(conn, ctask);
1599                 if (rc)
1600                         return rc;
1601         }
1602
1603         if (tcp_ctask->xmstate & XMSTATE_UNS_DATA) {
1604                 struct iscsi_data_task *dtask = &tcp_ctask->unsol_dtask;
1605                 int start = tcp_ctask->sent;
1606
1607                 rc = iscsi_send_data(ctask, &tcp_ctask->sendbuf, &tcp_ctask->sg,
1608                                      &tcp_ctask->sent, &ctask->data_count,
1609                                      &dtask->digestbuf, &dtask->digest);
1610                 ctask->unsol_count -= tcp_ctask->sent - start;
1611                 if (rc)
1612                         return rc;
1613                 tcp_ctask->xmstate &= ~XMSTATE_UNS_DATA;
1614                 /*
1615                  * Done with the Data-Out. Next, check if we need
1616                  * to send another unsolicited Data-Out.
1617                  */
1618                 if (ctask->unsol_count) {
1619                         debug_scsi("sending more uns\n");
1620                         tcp_ctask->xmstate |= XMSTATE_UNS_INIT;
1621                         goto send_hdr;
1622                 }
1623         }
1624         return 0;
1625 }
1626
1627 static int iscsi_send_sol_pdu(struct iscsi_conn *conn,
1628                               struct iscsi_cmd_task *ctask)
1629 {
1630         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1631         struct iscsi_session *session = conn->session;
1632         struct iscsi_r2t_info *r2t;
1633         struct iscsi_data_task *dtask;
1634         int left, rc;
1635
1636         if (tcp_ctask->xmstate & XMSTATE_SOL_HDR_INIT) {
1637                 if (!tcp_ctask->r2t) {
1638                         spin_lock_bh(&session->lock);
1639                         __kfifo_get(tcp_ctask->r2tqueue, (void*)&tcp_ctask->r2t,
1640                                     sizeof(void*));
1641                         spin_unlock_bh(&session->lock);
1642                 }
1643 send_hdr:
1644                 r2t = tcp_ctask->r2t;
1645                 dtask = &r2t->dtask;
1646
1647                 if (conn->hdrdgst_en)
1648                         iscsi_hdr_digest(conn, &r2t->headbuf,
1649                                         (u8*)dtask->hdrext);
1650                 tcp_ctask->xmstate &= ~XMSTATE_SOL_HDR_INIT;
1651                 tcp_ctask->xmstate |= XMSTATE_SOL_HDR;
1652         }
1653
1654         if (tcp_ctask->xmstate & XMSTATE_SOL_HDR) {
1655                 r2t = tcp_ctask->r2t;
1656                 dtask = &r2t->dtask;
1657
1658                 rc = iscsi_sendhdr(conn, &r2t->headbuf, r2t->data_count);
1659                 if (rc)
1660                         return rc;
1661                 tcp_ctask->xmstate &= ~XMSTATE_SOL_HDR;
1662                 tcp_ctask->xmstate |= XMSTATE_SOL_DATA;
1663
1664                 if (conn->datadgst_en) {
1665                         iscsi_data_digest_init(conn->dd_data, tcp_ctask);
1666                         dtask->digest = 0;
1667                 }
1668
1669                 iscsi_set_padding(tcp_ctask, r2t->data_count);
1670                 debug_scsi("sol dout [dsn %d itt 0x%x dlen %d sent %d]\n",
1671                         r2t->solicit_datasn - 1, ctask->itt, r2t->data_count,
1672                         r2t->sent);
1673         }
1674
1675         if (tcp_ctask->xmstate & XMSTATE_SOL_DATA) {
1676                 r2t = tcp_ctask->r2t;
1677                 dtask = &r2t->dtask;
1678
1679                 rc = iscsi_send_data(ctask, &r2t->sendbuf, &r2t->sg,
1680                                      &r2t->sent, &r2t->data_count,
1681                                      &dtask->digestbuf, &dtask->digest);
1682                 if (rc)
1683                         return rc;
1684                 tcp_ctask->xmstate &= ~XMSTATE_SOL_DATA;
1685
1686                 /*
1687                  * Done with this Data-Out. Next, check if we have
1688                  * to send another Data-Out for this R2T.
1689                  */
1690                 BUG_ON(r2t->data_length - r2t->sent < 0);
1691                 left = r2t->data_length - r2t->sent;
1692                 if (left) {
1693                         iscsi_solicit_data_cont(conn, ctask, r2t, left);
1694                         goto send_hdr;
1695                 }
1696
1697                 /*
1698                  * Done with this R2T. Check if there are more
1699                  * outstanding R2Ts ready to be processed.
1700                  */
1701                 spin_lock_bh(&session->lock);
1702                 tcp_ctask->r2t = NULL;
1703                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
1704                             sizeof(void*));
1705                 if (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t,
1706                                 sizeof(void*))) {
1707                         tcp_ctask->r2t = r2t;
1708                         spin_unlock_bh(&session->lock);
1709                         goto send_hdr;
1710                 }
1711                 spin_unlock_bh(&session->lock);
1712         }
1713         return 0;
1714 }
1715
1716 /**
1717  * iscsi_tcp_ctask_xmit - xmit normal PDU task
1718  * @conn: iscsi connection
1719  * @ctask: iscsi command task
1720  *
1721  * Notes:
1722  *      The function can return -EAGAIN in which case caller must
1723  *      call it again later, or recover. '0' return code means successful
1724  *      xmit.
1725  *      The function is devided to logical helpers (above) for the different
1726  *      xmit stages.
1727  *
1728  *iscsi_send_cmd_hdr()
1729  *      XMSTATE_CMD_HDR_INIT - prepare Header and Data buffers Calculate
1730  *                             Header Digest
1731  *      XMSTATE_CMD_HDR_XMIT - Transmit header in progress
1732  *
1733  *iscsi_send_padding
1734  *      XMSTATE_W_PAD        - Prepare and send pading
1735  *      XMSTATE_W_RESEND_PAD - retry send pading
1736  *
1737  *iscsi_send_digest
1738  *      XMSTATE_W_RESEND_DATA_DIGEST - Finalize and send Data Digest
1739  *      XMSTATE_W_RESEND_DATA_DIGEST - retry sending digest
1740  *
1741  *iscsi_send_unsol_hdr
1742  *      XMSTATE_UNS_INIT     - prepare un-solicit data header and digest
1743  *      XMSTATE_UNS_HDR      - send un-solicit header
1744  *
1745  *iscsi_send_unsol_pdu
1746  *      XMSTATE_UNS_DATA     - send un-solicit data in progress
1747  *
1748  *iscsi_send_sol_pdu
1749  *      XMSTATE_SOL_HDR_INIT - solicit data header and digest initialize
1750  *      XMSTATE_SOL_HDR      - send solicit header
1751  *      XMSTATE_SOL_DATA     - send solicit data
1752  *
1753  *iscsi_tcp_ctask_xmit
1754  *      XMSTATE_IMM_DATA     - xmit managment data (??)
1755  **/
1756 static int
1757 iscsi_tcp_ctask_xmit(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1758 {
1759         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1760         int rc = 0;
1761
1762         debug_scsi("ctask deq [cid %d xmstate %x itt 0x%x]\n",
1763                 conn->id, tcp_ctask->xmstate, ctask->itt);
1764
1765         /*
1766          * serialize with TMF AbortTask
1767          */
1768         if (ctask->mtask)
1769                 return rc;
1770
1771         rc = iscsi_send_cmd_hdr(conn, ctask);
1772         if (rc)
1773                 return rc;
1774         if (ctask->sc->sc_data_direction != DMA_TO_DEVICE)
1775                 return 0;
1776
1777         if (tcp_ctask->xmstate & XMSTATE_IMM_DATA) {
1778                 rc = iscsi_send_data(ctask, &tcp_ctask->sendbuf, &tcp_ctask->sg,
1779                                      &tcp_ctask->sent, &ctask->imm_count,
1780                                      &tcp_ctask->immbuf, &tcp_ctask->immdigest);
1781                 if (rc)
1782                         return rc;
1783                 tcp_ctask->xmstate &= ~XMSTATE_IMM_DATA;
1784         }
1785
1786         rc = iscsi_send_unsol_pdu(conn, ctask);
1787         if (rc)
1788                 return rc;
1789
1790         rc = iscsi_send_sol_pdu(conn, ctask);
1791         if (rc)
1792                 return rc;
1793
1794         return rc;
1795 }
1796
1797 static struct iscsi_cls_conn *
1798 iscsi_tcp_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
1799 {
1800         struct iscsi_conn *conn;
1801         struct iscsi_cls_conn *cls_conn;
1802         struct iscsi_tcp_conn *tcp_conn;
1803
1804         cls_conn = iscsi_conn_setup(cls_session, conn_idx);
1805         if (!cls_conn)
1806                 return NULL;
1807         conn = cls_conn->dd_data;
1808         /*
1809          * due to strange issues with iser these are not set
1810          * in iscsi_conn_setup
1811          */
1812         conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1813
1814         tcp_conn = kzalloc(sizeof(*tcp_conn), GFP_KERNEL);
1815         if (!tcp_conn)
1816                 goto tcp_conn_alloc_fail;
1817
1818         conn->dd_data = tcp_conn;
1819         tcp_conn->iscsi_conn = conn;
1820         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
1821         /* initial operational parameters */
1822         tcp_conn->hdr_size = sizeof(struct iscsi_hdr);
1823
1824         tcp_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1825                                                   CRYPTO_ALG_ASYNC);
1826         tcp_conn->tx_hash.flags = 0;
1827         if (IS_ERR(tcp_conn->tx_hash.tfm)) {
1828                 printk(KERN_ERR "Could not create connection due to crc32c "
1829                        "loading error %ld. Make sure the crc32c module is "
1830                        "built as a module or into the kernel\n",
1831                         PTR_ERR(tcp_conn->tx_hash.tfm));
1832                 goto free_tcp_conn;
1833         }
1834
1835         tcp_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1836                                                   CRYPTO_ALG_ASYNC);
1837         tcp_conn->rx_hash.flags = 0;
1838         if (IS_ERR(tcp_conn->rx_hash.tfm)) {
1839                 printk(KERN_ERR "Could not create connection due to crc32c "
1840                        "loading error %ld. Make sure the crc32c module is "
1841                        "built as a module or into the kernel\n",
1842                         PTR_ERR(tcp_conn->rx_hash.tfm));
1843                 goto free_tx_tfm;
1844         }
1845
1846         return cls_conn;
1847
1848 free_tx_tfm:
1849         crypto_free_hash(tcp_conn->tx_hash.tfm);
1850 free_tcp_conn:
1851         kfree(tcp_conn);
1852 tcp_conn_alloc_fail:
1853         iscsi_conn_teardown(cls_conn);
1854         return NULL;
1855 }
1856
1857 static void
1858 iscsi_tcp_release_conn(struct iscsi_conn *conn)
1859 {
1860         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1861
1862         if (!tcp_conn->sock)
1863                 return;
1864
1865         sock_hold(tcp_conn->sock->sk);
1866         iscsi_conn_restore_callbacks(tcp_conn);
1867         sock_put(tcp_conn->sock->sk);
1868
1869         sock_release(tcp_conn->sock);
1870         tcp_conn->sock = NULL;
1871         conn->recv_lock = NULL;
1872 }
1873
1874 static void
1875 iscsi_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
1876 {
1877         struct iscsi_conn *conn = cls_conn->dd_data;
1878         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1879
1880         iscsi_tcp_release_conn(conn);
1881         iscsi_conn_teardown(cls_conn);
1882
1883         if (tcp_conn->tx_hash.tfm)
1884                 crypto_free_hash(tcp_conn->tx_hash.tfm);
1885         if (tcp_conn->rx_hash.tfm)
1886                 crypto_free_hash(tcp_conn->rx_hash.tfm);
1887
1888         kfree(tcp_conn);
1889 }
1890
1891 static void
1892 iscsi_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
1893 {
1894         struct iscsi_conn *conn = cls_conn->dd_data;
1895         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1896
1897         iscsi_conn_stop(cls_conn, flag);
1898         iscsi_tcp_release_conn(conn);
1899         tcp_conn->hdr_size = sizeof(struct iscsi_hdr);
1900 }
1901
1902 static int
1903 iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
1904                     struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
1905                     int is_leading)
1906 {
1907         struct iscsi_conn *conn = cls_conn->dd_data;
1908         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1909         struct sock *sk;
1910         struct socket *sock;
1911         int err;
1912
1913         /* lookup for existing socket */
1914         sock = sockfd_lookup((int)transport_eph, &err);
1915         if (!sock) {
1916                 printk(KERN_ERR "iscsi_tcp: sockfd_lookup failed %d\n", err);
1917                 return -EEXIST;
1918         }
1919
1920         err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
1921         if (err)
1922                 return err;
1923
1924         /* bind iSCSI connection and socket */
1925         tcp_conn->sock = sock;
1926
1927         /* setup Socket parameters */
1928         sk = sock->sk;
1929         sk->sk_reuse = 1;
1930         sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
1931         sk->sk_allocation = GFP_ATOMIC;
1932
1933         /* FIXME: disable Nagle's algorithm */
1934
1935         /*
1936          * Intercept TCP callbacks for sendfile like receive
1937          * processing.
1938          */
1939         conn->recv_lock = &sk->sk_callback_lock;
1940         iscsi_conn_set_callbacks(conn);
1941         tcp_conn->sendpage = tcp_conn->sock->ops->sendpage;
1942         /*
1943          * set receive state machine into initial state
1944          */
1945         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
1946
1947         return 0;
1948 }
1949
1950 /* called with host lock */
1951 static void
1952 iscsi_tcp_mgmt_init(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask,
1953                     char *data, uint32_t data_size)
1954 {
1955         struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
1956         tcp_mtask->xmstate = XMSTATE_IMM_HDR_INIT;
1957 }
1958
1959 static int
1960 iscsi_r2tpool_alloc(struct iscsi_session *session)
1961 {
1962         int i;
1963         int cmd_i;
1964
1965         /*
1966          * initialize per-task: R2T pool and xmit queue
1967          */
1968         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1969                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
1970                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1971
1972                 /*
1973                  * pre-allocated x4 as much r2ts to handle race when
1974                  * target acks DataOut faster than we data_xmit() queues
1975                  * could replenish r2tqueue.
1976                  */
1977
1978                 /* R2T pool */
1979                 if (iscsi_pool_init(&tcp_ctask->r2tpool, session->max_r2t * 4,
1980                                     (void***)&tcp_ctask->r2ts,
1981                                     sizeof(struct iscsi_r2t_info))) {
1982                         goto r2t_alloc_fail;
1983                 }
1984
1985                 /* R2T xmit queue */
1986                 tcp_ctask->r2tqueue = kfifo_alloc(
1987                       session->max_r2t * 4 * sizeof(void*), GFP_KERNEL, NULL);
1988                 if (tcp_ctask->r2tqueue == ERR_PTR(-ENOMEM)) {
1989                         iscsi_pool_free(&tcp_ctask->r2tpool,
1990                                         (void**)tcp_ctask->r2ts);
1991                         goto r2t_alloc_fail;
1992                 }
1993         }
1994
1995         return 0;
1996
1997 r2t_alloc_fail:
1998         for (i = 0; i < cmd_i; i++) {
1999                 struct iscsi_cmd_task *ctask = session->cmds[i];
2000                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2001
2002                 kfifo_free(tcp_ctask->r2tqueue);
2003                 iscsi_pool_free(&tcp_ctask->r2tpool,
2004                                 (void**)tcp_ctask->r2ts);
2005         }
2006         return -ENOMEM;
2007 }
2008
2009 static void
2010 iscsi_r2tpool_free(struct iscsi_session *session)
2011 {
2012         int i;
2013
2014         for (i = 0; i < session->cmds_max; i++) {
2015                 struct iscsi_cmd_task *ctask = session->cmds[i];
2016                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2017
2018                 kfifo_free(tcp_ctask->r2tqueue);
2019                 iscsi_pool_free(&tcp_ctask->r2tpool,
2020                                 (void**)tcp_ctask->r2ts);
2021         }
2022 }
2023
2024 static int
2025 iscsi_conn_set_param(struct iscsi_cls_conn *cls_conn, enum iscsi_param param,
2026                      char *buf, int buflen)
2027 {
2028         struct iscsi_conn *conn = cls_conn->dd_data;
2029         struct iscsi_session *session = conn->session;
2030         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2031         int value;
2032
2033         switch(param) {
2034         case ISCSI_PARAM_HDRDGST_EN:
2035                 iscsi_set_param(cls_conn, param, buf, buflen);
2036                 tcp_conn->hdr_size = sizeof(struct iscsi_hdr);
2037                 if (conn->hdrdgst_en)
2038                         tcp_conn->hdr_size += sizeof(__u32);
2039                 break;
2040         case ISCSI_PARAM_DATADGST_EN:
2041                 iscsi_set_param(cls_conn, param, buf, buflen);
2042                 tcp_conn->sendpage = conn->datadgst_en ?
2043                         sock_no_sendpage : tcp_conn->sock->ops->sendpage;
2044                 break;
2045         case ISCSI_PARAM_MAX_R2T:
2046                 sscanf(buf, "%d", &value);
2047                 if (session->max_r2t == roundup_pow_of_two(value))
2048                         break;
2049                 iscsi_r2tpool_free(session);
2050                 iscsi_set_param(cls_conn, param, buf, buflen);
2051                 if (session->max_r2t & (session->max_r2t - 1))
2052                         session->max_r2t = roundup_pow_of_two(session->max_r2t);
2053                 if (iscsi_r2tpool_alloc(session))
2054                         return -ENOMEM;
2055                 break;
2056         default:
2057                 return iscsi_set_param(cls_conn, param, buf, buflen);
2058         }
2059
2060         return 0;
2061 }
2062
2063 static int
2064 iscsi_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
2065                          enum iscsi_param param, char *buf)
2066 {
2067         struct iscsi_conn *conn = cls_conn->dd_data;
2068         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2069         struct inet_sock *inet;
2070         struct ipv6_pinfo *np;
2071         struct sock *sk;
2072         int len;
2073
2074         switch(param) {
2075         case ISCSI_PARAM_CONN_PORT:
2076                 mutex_lock(&conn->xmitmutex);
2077                 if (!tcp_conn->sock) {
2078                         mutex_unlock(&conn->xmitmutex);
2079                         return -EINVAL;
2080                 }
2081
2082                 inet = inet_sk(tcp_conn->sock->sk);
2083                 len = sprintf(buf, "%hu\n", be16_to_cpu(inet->dport));
2084                 mutex_unlock(&conn->xmitmutex);
2085                 break;
2086         case ISCSI_PARAM_CONN_ADDRESS:
2087                 mutex_lock(&conn->xmitmutex);
2088                 if (!tcp_conn->sock) {
2089                         mutex_unlock(&conn->xmitmutex);
2090                         return -EINVAL;
2091                 }
2092
2093                 sk = tcp_conn->sock->sk;
2094                 if (sk->sk_family == PF_INET) {
2095                         inet = inet_sk(sk);
2096                         len = sprintf(buf, NIPQUAD_FMT "\n",
2097                                       NIPQUAD(inet->daddr));
2098                 } else {
2099                         np = inet6_sk(sk);
2100                         len = sprintf(buf, NIP6_FMT "\n", NIP6(np->daddr));
2101                 }
2102                 mutex_unlock(&conn->xmitmutex);
2103                 break;
2104         default:
2105                 return iscsi_conn_get_param(cls_conn, param, buf);
2106         }
2107
2108         return len;
2109 }
2110
2111 static void
2112 iscsi_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
2113 {
2114         struct iscsi_conn *conn = cls_conn->dd_data;
2115         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2116
2117         stats->txdata_octets = conn->txdata_octets;
2118         stats->rxdata_octets = conn->rxdata_octets;
2119         stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
2120         stats->dataout_pdus = conn->dataout_pdus_cnt;
2121         stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
2122         stats->datain_pdus = conn->datain_pdus_cnt;
2123         stats->r2t_pdus = conn->r2t_pdus_cnt;
2124         stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
2125         stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
2126         stats->custom_length = 3;
2127         strcpy(stats->custom[0].desc, "tx_sendpage_failures");
2128         stats->custom[0].value = tcp_conn->sendpage_failures_cnt;
2129         strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
2130         stats->custom[1].value = tcp_conn->discontiguous_hdr_cnt;
2131         strcpy(stats->custom[2].desc, "eh_abort_cnt");
2132         stats->custom[2].value = conn->eh_abort_cnt;
2133 }
2134
2135 static struct iscsi_cls_session *
2136 iscsi_tcp_session_create(struct iscsi_transport *iscsit,
2137                          struct scsi_transport_template *scsit,
2138                          uint32_t initial_cmdsn, uint32_t *hostno)
2139 {
2140         struct iscsi_cls_session *cls_session;
2141         struct iscsi_session *session;
2142         uint32_t hn;
2143         int cmd_i;
2144
2145         cls_session = iscsi_session_setup(iscsit, scsit,
2146                                          sizeof(struct iscsi_tcp_cmd_task),
2147                                          sizeof(struct iscsi_tcp_mgmt_task),
2148                                          initial_cmdsn, &hn);
2149         if (!cls_session)
2150                 return NULL;
2151         *hostno = hn;
2152
2153         session = class_to_transport_session(cls_session);
2154         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
2155                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
2156                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2157
2158                 ctask->hdr = &tcp_ctask->hdr;
2159         }
2160
2161         for (cmd_i = 0; cmd_i < session->mgmtpool_max; cmd_i++) {
2162                 struct iscsi_mgmt_task *mtask = session->mgmt_cmds[cmd_i];
2163                 struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
2164
2165                 mtask->hdr = &tcp_mtask->hdr;
2166         }
2167
2168         if (iscsi_r2tpool_alloc(class_to_transport_session(cls_session)))
2169                 goto r2tpool_alloc_fail;
2170
2171         return cls_session;
2172
2173 r2tpool_alloc_fail:
2174         iscsi_session_teardown(cls_session);
2175         return NULL;
2176 }
2177
2178 static void iscsi_tcp_session_destroy(struct iscsi_cls_session *cls_session)
2179 {
2180         iscsi_r2tpool_free(class_to_transport_session(cls_session));
2181         iscsi_session_teardown(cls_session);
2182 }
2183
2184 static struct scsi_host_template iscsi_sht = {
2185         .name                   = "iSCSI Initiator over TCP/IP",
2186         .queuecommand           = iscsi_queuecommand,
2187         .change_queue_depth     = iscsi_change_queue_depth,
2188         .can_queue              = ISCSI_XMIT_CMDS_MAX - 1,
2189         .sg_tablesize           = ISCSI_SG_TABLESIZE,
2190         .max_sectors            = 0xFFFF,
2191         .cmd_per_lun            = ISCSI_DEF_CMD_PER_LUN,
2192         .eh_abort_handler       = iscsi_eh_abort,
2193         .eh_host_reset_handler  = iscsi_eh_host_reset,
2194         .use_clustering         = DISABLE_CLUSTERING,
2195         .proc_name              = "iscsi_tcp",
2196         .this_id                = -1,
2197 };
2198
2199 static struct iscsi_transport iscsi_tcp_transport = {
2200         .owner                  = THIS_MODULE,
2201         .name                   = "tcp",
2202         .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
2203                                   | CAP_DATADGST,
2204         .param_mask             = ISCSI_MAX_RECV_DLENGTH |
2205                                   ISCSI_MAX_XMIT_DLENGTH |
2206                                   ISCSI_HDRDGST_EN |
2207                                   ISCSI_DATADGST_EN |
2208                                   ISCSI_INITIAL_R2T_EN |
2209                                   ISCSI_MAX_R2T |
2210                                   ISCSI_IMM_DATA_EN |
2211                                   ISCSI_FIRST_BURST |
2212                                   ISCSI_MAX_BURST |
2213                                   ISCSI_PDU_INORDER_EN |
2214                                   ISCSI_DATASEQ_INORDER_EN |
2215                                   ISCSI_ERL |
2216                                   ISCSI_CONN_PORT |
2217                                   ISCSI_CONN_ADDRESS |
2218                                   ISCSI_EXP_STATSN |
2219                                   ISCSI_PERSISTENT_PORT |
2220                                   ISCSI_PERSISTENT_ADDRESS |
2221                                   ISCSI_TARGET_NAME | ISCSI_TPGT |
2222                                   ISCSI_USERNAME | ISCSI_PASSWORD |
2223                                   ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN,
2224         .host_param_mask        = ISCSI_HOST_HWADDRESS |
2225                                   ISCSI_HOST_INITIATOR_NAME,
2226         .host_template          = &iscsi_sht,
2227         .conndata_size          = sizeof(struct iscsi_conn),
2228         .max_conn               = 1,
2229         .max_cmd_len            = ISCSI_TCP_MAX_CMD_LEN,
2230         /* session management */
2231         .create_session         = iscsi_tcp_session_create,
2232         .destroy_session        = iscsi_tcp_session_destroy,
2233         /* connection management */
2234         .create_conn            = iscsi_tcp_conn_create,
2235         .bind_conn              = iscsi_tcp_conn_bind,
2236         .destroy_conn           = iscsi_tcp_conn_destroy,
2237         .set_param              = iscsi_conn_set_param,
2238         .get_conn_param         = iscsi_tcp_conn_get_param,
2239         .get_session_param      = iscsi_session_get_param,
2240         .start_conn             = iscsi_conn_start,
2241         .stop_conn              = iscsi_tcp_conn_stop,
2242         /* iscsi host params */
2243         .get_host_param         = iscsi_host_get_param,
2244         .set_host_param         = iscsi_host_set_param,
2245         /* IO */
2246         .send_pdu               = iscsi_conn_send_pdu,
2247         .get_stats              = iscsi_conn_get_stats,
2248         .init_cmd_task          = iscsi_tcp_cmd_init,
2249         .init_mgmt_task         = iscsi_tcp_mgmt_init,
2250         .xmit_cmd_task          = iscsi_tcp_ctask_xmit,
2251         .xmit_mgmt_task         = iscsi_tcp_mtask_xmit,
2252         .cleanup_cmd_task       = iscsi_tcp_cleanup_ctask,
2253         /* recovery */
2254         .session_recovery_timedout = iscsi_session_recovery_timedout,
2255 };
2256
2257 static int __init
2258 iscsi_tcp_init(void)
2259 {
2260         if (iscsi_max_lun < 1) {
2261                 printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
2262                        iscsi_max_lun);
2263                 return -EINVAL;
2264         }
2265         iscsi_tcp_transport.max_lun = iscsi_max_lun;
2266
2267         if (!iscsi_register_transport(&iscsi_tcp_transport))
2268                 return -ENODEV;
2269
2270         return 0;
2271 }
2272
2273 static void __exit
2274 iscsi_tcp_exit(void)
2275 {
2276         iscsi_unregister_transport(&iscsi_tcp_transport);
2277 }
2278
2279 module_init(iscsi_tcp_init);
2280 module_exit(iscsi_tcp_exit);