[SCSI] iscsi_tcp: fix handling of data buffer padding
[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 <net/tcp.h>
38 #include <scsi/scsi_cmnd.h>
39 #include <scsi/scsi_host.h>
40 #include <scsi/scsi.h>
41 #include <scsi/scsi_transport_iscsi.h>
42
43 #include "iscsi_tcp.h"
44
45 MODULE_AUTHOR("Dmitry Yusupov <dmitry_yus@yahoo.com>, "
46               "Alex Aizman <itn780@yahoo.com>");
47 MODULE_DESCRIPTION("iSCSI/TCP data-path");
48 MODULE_LICENSE("GPL");
49 /* #define DEBUG_TCP */
50 #define DEBUG_ASSERT
51
52 #ifdef DEBUG_TCP
53 #define debug_tcp(fmt...) printk(KERN_INFO "tcp: " fmt)
54 #else
55 #define debug_tcp(fmt...)
56 #endif
57
58 #ifndef DEBUG_ASSERT
59 #ifdef BUG_ON
60 #undef BUG_ON
61 #endif
62 #define BUG_ON(expr)
63 #endif
64
65 static unsigned int iscsi_max_lun = 512;
66 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
67
68 static inline void
69 iscsi_buf_init_iov(struct iscsi_buf *ibuf, char *vbuf, int size)
70 {
71         ibuf->sg.page = virt_to_page(vbuf);
72         ibuf->sg.offset = offset_in_page(vbuf);
73         ibuf->sg.length = size;
74         ibuf->sent = 0;
75         ibuf->use_sendmsg = 1;
76 }
77
78 static inline void
79 iscsi_buf_init_sg(struct iscsi_buf *ibuf, struct scatterlist *sg)
80 {
81         ibuf->sg.page = sg->page;
82         ibuf->sg.offset = sg->offset;
83         ibuf->sg.length = sg->length;
84         /*
85          * Fastpath: sg element fits into single page
86          */
87         if (sg->length + sg->offset <= PAGE_SIZE && !PageSlab(sg->page))
88                 ibuf->use_sendmsg = 0;
89         else
90                 ibuf->use_sendmsg = 1;
91         ibuf->sent = 0;
92 }
93
94 static inline int
95 iscsi_buf_left(struct iscsi_buf *ibuf)
96 {
97         int rc;
98
99         rc = ibuf->sg.length - ibuf->sent;
100         BUG_ON(rc < 0);
101         return rc;
102 }
103
104 static inline void
105 iscsi_hdr_digest(struct iscsi_conn *conn, struct iscsi_buf *buf,
106                  u8* crc)
107 {
108         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
109
110         crypto_hash_digest(&tcp_conn->tx_hash, &buf->sg, buf->sg.length, crc);
111         buf->sg.length += sizeof(u32);
112 }
113
114 static inline int
115 iscsi_hdr_extract(struct iscsi_tcp_conn *tcp_conn)
116 {
117         struct sk_buff *skb = tcp_conn->in.skb;
118
119         tcp_conn->in.zero_copy_hdr = 0;
120
121         if (tcp_conn->in.copy >= tcp_conn->hdr_size &&
122             tcp_conn->in_progress == IN_PROGRESS_WAIT_HEADER) {
123                 /*
124                  * Zero-copy PDU Header: using connection context
125                  * to store header pointer.
126                  */
127                 if (skb_shinfo(skb)->frag_list == NULL &&
128                     !skb_shinfo(skb)->nr_frags) {
129                         tcp_conn->in.hdr = (struct iscsi_hdr *)
130                                 ((char*)skb->data + tcp_conn->in.offset);
131                         tcp_conn->in.zero_copy_hdr = 1;
132                 } else {
133                         /* ignoring return code since we checked
134                          * in.copy before */
135                         skb_copy_bits(skb, tcp_conn->in.offset,
136                                 &tcp_conn->hdr, tcp_conn->hdr_size);
137                         tcp_conn->in.hdr = &tcp_conn->hdr;
138                 }
139                 tcp_conn->in.offset += tcp_conn->hdr_size;
140                 tcp_conn->in.copy -= tcp_conn->hdr_size;
141         } else {
142                 int hdr_remains;
143                 int copylen;
144
145                 /*
146                  * PDU header scattered across SKB's,
147                  * copying it... This'll happen quite rarely.
148                  */
149
150                 if (tcp_conn->in_progress == IN_PROGRESS_WAIT_HEADER)
151                         tcp_conn->in.hdr_offset = 0;
152
153                 hdr_remains = tcp_conn->hdr_size - tcp_conn->in.hdr_offset;
154                 BUG_ON(hdr_remains <= 0);
155
156                 copylen = min(tcp_conn->in.copy, hdr_remains);
157                 skb_copy_bits(skb, tcp_conn->in.offset,
158                         (char*)&tcp_conn->hdr + tcp_conn->in.hdr_offset,
159                         copylen);
160
161                 debug_tcp("PDU gather offset %d bytes %d in.offset %d "
162                        "in.copy %d\n", tcp_conn->in.hdr_offset, copylen,
163                        tcp_conn->in.offset, tcp_conn->in.copy);
164
165                 tcp_conn->in.offset += copylen;
166                 tcp_conn->in.copy -= copylen;
167                 if (copylen < hdr_remains)  {
168                         tcp_conn->in_progress = IN_PROGRESS_HEADER_GATHER;
169                         tcp_conn->in.hdr_offset += copylen;
170                         return -EAGAIN;
171                 }
172                 tcp_conn->in.hdr = &tcp_conn->hdr;
173                 tcp_conn->discontiguous_hdr_cnt++;
174                 tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
175         }
176
177         return 0;
178 }
179
180 /*
181  * must be called with session lock
182  */
183 static void
184 iscsi_tcp_cleanup_ctask(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
185 {
186         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
187         struct iscsi_r2t_info *r2t;
188         struct scsi_cmnd *sc;
189
190         /* flush ctask's r2t queues */
191         while (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*))) {
192                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
193                             sizeof(void*));
194                 debug_scsi("iscsi_tcp_cleanup_ctask pending r2t dropped\n");
195         }
196
197         sc = ctask->sc;
198         if (unlikely(!sc))
199                 return;
200
201         tcp_ctask->xmstate = XMSTATE_IDLE;
202         tcp_ctask->r2t = NULL;
203 }
204
205 /**
206  * iscsi_data_rsp - SCSI Data-In Response processing
207  * @conn: iscsi connection
208  * @ctask: scsi command task
209  **/
210 static int
211 iscsi_data_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
212 {
213         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
214         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
215         struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
216         struct iscsi_session *session = conn->session;
217         struct scsi_cmnd *sc = ctask->sc;
218         int datasn = be32_to_cpu(rhdr->datasn);
219
220         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
221         /*
222          * setup Data-In byte counter (gets decremented..)
223          */
224         ctask->data_count = tcp_conn->in.datalen;
225
226         if (tcp_conn->in.datalen == 0)
227                 return 0;
228
229         if (tcp_ctask->exp_datasn != datasn) {
230                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->datasn(%d)\n",
231                           __FUNCTION__, tcp_ctask->exp_datasn, datasn);
232                 return ISCSI_ERR_DATASN;
233         }
234
235         tcp_ctask->exp_datasn++;
236
237         tcp_ctask->data_offset = be32_to_cpu(rhdr->offset);
238         if (tcp_ctask->data_offset + tcp_conn->in.datalen > sc->request_bufflen) {
239                 debug_tcp("%s: data_offset(%d) + data_len(%d) > total_length_in(%d)\n",
240                           __FUNCTION__, tcp_ctask->data_offset,
241                           tcp_conn->in.datalen, sc->request_bufflen);
242                 return ISCSI_ERR_DATA_OFFSET;
243         }
244
245         if (rhdr->flags & ISCSI_FLAG_DATA_STATUS) {
246                 conn->exp_statsn = be32_to_cpu(rhdr->statsn) + 1;
247                 if (rhdr->flags & ISCSI_FLAG_DATA_UNDERFLOW) {
248                         int res_count = be32_to_cpu(rhdr->residual_count);
249
250                         if (res_count > 0 &&
251                             res_count <= sc->request_bufflen) {
252                                 sc->resid = res_count;
253                                 sc->result = (DID_OK << 16) | rhdr->cmd_status;
254                         } else
255                                 sc->result = (DID_BAD_TARGET << 16) |
256                                         rhdr->cmd_status;
257                 } else if (rhdr->flags & ISCSI_FLAG_DATA_OVERFLOW) {
258                         sc->resid = be32_to_cpu(rhdr->residual_count);
259                         sc->result = (DID_OK << 16) | rhdr->cmd_status;
260                 } else
261                         sc->result = (DID_OK << 16) | rhdr->cmd_status;
262         }
263
264         conn->datain_pdus_cnt++;
265         return 0;
266 }
267
268 /**
269  * iscsi_solicit_data_init - initialize first Data-Out
270  * @conn: iscsi connection
271  * @ctask: scsi command task
272  * @r2t: R2T info
273  *
274  * Notes:
275  *      Initialize first Data-Out within this R2T sequence and finds
276  *      proper data_offset within this SCSI command.
277  *
278  *      This function is called with connection lock taken.
279  **/
280 static void
281 iscsi_solicit_data_init(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
282                         struct iscsi_r2t_info *r2t)
283 {
284         struct iscsi_data *hdr;
285         struct scsi_cmnd *sc = ctask->sc;
286
287         hdr = &r2t->dtask.hdr;
288         memset(hdr, 0, sizeof(struct iscsi_data));
289         hdr->ttt = r2t->ttt;
290         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
291         r2t->solicit_datasn++;
292         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
293         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
294         hdr->itt = ctask->hdr->itt;
295         hdr->exp_statsn = r2t->exp_statsn;
296         hdr->offset = cpu_to_be32(r2t->data_offset);
297         if (r2t->data_length > conn->max_xmit_dlength) {
298                 hton24(hdr->dlength, conn->max_xmit_dlength);
299                 r2t->data_count = conn->max_xmit_dlength;
300                 hdr->flags = 0;
301         } else {
302                 hton24(hdr->dlength, r2t->data_length);
303                 r2t->data_count = r2t->data_length;
304                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
305         }
306         conn->dataout_pdus_cnt++;
307
308         r2t->sent = 0;
309
310         iscsi_buf_init_iov(&r2t->headbuf, (char*)hdr,
311                            sizeof(struct iscsi_hdr));
312
313         if (sc->use_sg) {
314                 int i, sg_count = 0;
315                 struct scatterlist *sg = sc->request_buffer;
316
317                 r2t->sg = NULL;
318                 for (i = 0; i < sc->use_sg; i++, sg += 1) {
319                         /* FIXME: prefetch ? */
320                         if (sg_count + sg->length > r2t->data_offset) {
321                                 int page_offset;
322
323                                 /* sg page found! */
324
325                                 /* offset within this page */
326                                 page_offset = r2t->data_offset - sg_count;
327
328                                 /* fill in this buffer */
329                                 iscsi_buf_init_sg(&r2t->sendbuf, sg);
330                                 r2t->sendbuf.sg.offset += page_offset;
331                                 r2t->sendbuf.sg.length -= page_offset;
332
333                                 /* xmit logic will continue with next one */
334                                 r2t->sg = sg + 1;
335                                 break;
336                         }
337                         sg_count += sg->length;
338                 }
339                 BUG_ON(r2t->sg == NULL);
340         } else {
341                 iscsi_buf_init_iov(&r2t->sendbuf,
342                             (char*)sc->request_buffer + r2t->data_offset,
343                             r2t->data_count);
344                 r2t->sg = NULL;
345         }
346 }
347
348 /**
349  * iscsi_r2t_rsp - iSCSI R2T Response processing
350  * @conn: iscsi connection
351  * @ctask: scsi command task
352  **/
353 static int
354 iscsi_r2t_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
355 {
356         struct iscsi_r2t_info *r2t;
357         struct iscsi_session *session = conn->session;
358         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
359         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
360         struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
361         int r2tsn = be32_to_cpu(rhdr->r2tsn);
362         int rc;
363
364         if (tcp_conn->in.datalen) {
365                 printk(KERN_ERR "iscsi_tcp: invalid R2t with datalen %d\n",
366                        tcp_conn->in.datalen);
367                 return ISCSI_ERR_DATALEN;
368         }
369
370         if (tcp_ctask->exp_datasn != r2tsn){
371                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
372                           __FUNCTION__, tcp_ctask->exp_datasn, r2tsn);
373                 return ISCSI_ERR_R2TSN;
374         }
375
376         /* fill-in new R2T associated with the task */
377         spin_lock(&session->lock);
378         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
379
380         if (!ctask->sc || ctask->mtask ||
381              session->state != ISCSI_STATE_LOGGED_IN) {
382                 printk(KERN_INFO "iscsi_tcp: dropping R2T itt %d in "
383                        "recovery...\n", ctask->itt);
384                 spin_unlock(&session->lock);
385                 return 0;
386         }
387
388         rc = __kfifo_get(tcp_ctask->r2tpool.queue, (void*)&r2t, sizeof(void*));
389         BUG_ON(!rc);
390
391         r2t->exp_statsn = rhdr->statsn;
392         r2t->data_length = be32_to_cpu(rhdr->data_length);
393         if (r2t->data_length == 0) {
394                 printk(KERN_ERR "iscsi_tcp: invalid R2T with zero data len\n");
395                 spin_unlock(&session->lock);
396                 return ISCSI_ERR_DATALEN;
397         }
398
399         if (r2t->data_length > session->max_burst)
400                 debug_scsi("invalid R2T with data len %u and max burst %u."
401                            "Attempting to execute request.\n",
402                             r2t->data_length, session->max_burst);
403
404         r2t->data_offset = be32_to_cpu(rhdr->data_offset);
405         if (r2t->data_offset + r2t->data_length > ctask->sc->request_bufflen) {
406                 spin_unlock(&session->lock);
407                 printk(KERN_ERR "iscsi_tcp: invalid R2T with data len %u at "
408                        "offset %u and total length %d\n", r2t->data_length,
409                        r2t->data_offset, ctask->sc->request_bufflen);
410                 return ISCSI_ERR_DATALEN;
411         }
412
413         r2t->ttt = rhdr->ttt; /* no flip */
414         r2t->solicit_datasn = 0;
415
416         iscsi_solicit_data_init(conn, ctask, r2t);
417
418         tcp_ctask->exp_datasn = r2tsn + 1;
419         __kfifo_put(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*));
420         tcp_ctask->xmstate |= XMSTATE_SOL_HDR_INIT;
421         list_move_tail(&ctask->running, &conn->xmitqueue);
422
423         scsi_queue_work(session->host, &conn->xmitwork);
424         conn->r2t_pdus_cnt++;
425         spin_unlock(&session->lock);
426
427         return 0;
428 }
429
430 static int
431 iscsi_tcp_hdr_recv(struct iscsi_conn *conn)
432 {
433         int rc = 0, opcode, ahslen;
434         struct iscsi_hdr *hdr;
435         struct iscsi_session *session = conn->session;
436         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
437         uint32_t cdgst, rdgst = 0, itt;
438
439         hdr = tcp_conn->in.hdr;
440
441         /* verify PDU length */
442         tcp_conn->in.datalen = ntoh24(hdr->dlength);
443         if (tcp_conn->in.datalen > conn->max_recv_dlength) {
444                 printk(KERN_ERR "iscsi_tcp: datalen %d > %d\n",
445                        tcp_conn->in.datalen, conn->max_recv_dlength);
446                 return ISCSI_ERR_DATALEN;
447         }
448         tcp_conn->data_copied = 0;
449
450         /* read AHS */
451         ahslen = hdr->hlength << 2;
452         tcp_conn->in.offset += ahslen;
453         tcp_conn->in.copy -= ahslen;
454         if (tcp_conn->in.copy < 0) {
455                 printk(KERN_ERR "iscsi_tcp: can't handle AHS with length "
456                        "%d bytes\n", ahslen);
457                 return ISCSI_ERR_AHSLEN;
458         }
459
460         /* calculate read padding */
461         tcp_conn->in.padding = tcp_conn->in.datalen & (ISCSI_PAD_LEN-1);
462         if (tcp_conn->in.padding) {
463                 tcp_conn->in.padding = ISCSI_PAD_LEN - tcp_conn->in.padding;
464                 debug_scsi("read padding %d bytes\n", tcp_conn->in.padding);
465         }
466
467         if (conn->hdrdgst_en) {
468                 struct scatterlist sg;
469
470                 sg_init_one(&sg, (u8 *)hdr,
471                             sizeof(struct iscsi_hdr) + ahslen);
472                 crypto_hash_digest(&tcp_conn->rx_hash, &sg, sg.length,
473                                    (u8 *)&cdgst);
474                 rdgst = *(uint32_t*)((char*)hdr + sizeof(struct iscsi_hdr) +
475                                      ahslen);
476                 if (cdgst != rdgst) {
477                         printk(KERN_ERR "iscsi_tcp: hdrdgst error "
478                                "recv 0x%x calc 0x%x\n", rdgst, cdgst);
479                         return ISCSI_ERR_HDR_DGST;
480                 }
481         }
482
483         opcode = hdr->opcode & ISCSI_OPCODE_MASK;
484         /* verify itt (itt encoding: age+cid+itt) */
485         rc = iscsi_verify_itt(conn, hdr, &itt);
486         if (rc == ISCSI_ERR_NO_SCSI_CMD) {
487                 tcp_conn->in.datalen = 0; /* force drop */
488                 return 0;
489         } else if (rc)
490                 return rc;
491
492         debug_tcp("opcode 0x%x offset %d copy %d ahslen %d datalen %d\n",
493                   opcode, tcp_conn->in.offset, tcp_conn->in.copy,
494                   ahslen, tcp_conn->in.datalen);
495
496         switch(opcode) {
497         case ISCSI_OP_SCSI_DATA_IN:
498                 tcp_conn->in.ctask = session->cmds[itt];
499                 rc = iscsi_data_rsp(conn, tcp_conn->in.ctask);
500                 if (rc)
501                         return rc;
502                 /* fall through */
503         case ISCSI_OP_SCSI_CMD_RSP:
504                 tcp_conn->in.ctask = session->cmds[itt];
505                 if (tcp_conn->in.datalen)
506                         goto copy_hdr;
507
508                 spin_lock(&session->lock);
509                 rc = __iscsi_complete_pdu(conn, hdr, NULL, 0);
510                 spin_unlock(&session->lock);
511                 break;
512         case ISCSI_OP_R2T:
513                 tcp_conn->in.ctask = session->cmds[itt];
514                 if (ahslen)
515                         rc = ISCSI_ERR_AHSLEN;
516                 else if (tcp_conn->in.ctask->sc->sc_data_direction ==
517                                                                 DMA_TO_DEVICE)
518                         rc = iscsi_r2t_rsp(conn, tcp_conn->in.ctask);
519                 else
520                         rc = ISCSI_ERR_PROTO;
521                 break;
522         case ISCSI_OP_LOGIN_RSP:
523         case ISCSI_OP_TEXT_RSP:
524         case ISCSI_OP_REJECT:
525         case ISCSI_OP_ASYNC_EVENT:
526                 /*
527                  * It is possible that we could get a PDU with a buffer larger
528                  * than 8K, but there are no targets that currently do this.
529                  * For now we fail until we find a vendor that needs it
530                  */
531                 if (ISCSI_DEF_MAX_RECV_SEG_LEN <
532                     tcp_conn->in.datalen) {
533                         printk(KERN_ERR "iscsi_tcp: received buffer of len %u "
534                               "but conn buffer is only %u (opcode %0x)\n",
535                               tcp_conn->in.datalen,
536                               ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
537                         rc = ISCSI_ERR_PROTO;
538                         break;
539                 }
540
541                 if (tcp_conn->in.datalen)
542                         goto copy_hdr;
543         /* fall through */
544         case ISCSI_OP_LOGOUT_RSP:
545         case ISCSI_OP_NOOP_IN:
546         case ISCSI_OP_SCSI_TMFUNC_RSP:
547                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
548                 break;
549         default:
550                 rc = ISCSI_ERR_BAD_OPCODE;
551                 break;
552         }
553
554         return rc;
555
556 copy_hdr:
557         /*
558          * if we did zero copy for the header but we will need multiple
559          * skbs to complete the command then we have to copy the header
560          * for later use
561          */
562         if (tcp_conn->in.zero_copy_hdr && tcp_conn->in.copy <=
563            (tcp_conn->in.datalen + tcp_conn->in.padding +
564             (conn->datadgst_en ? 4 : 0))) {
565                 debug_tcp("Copying header for later use. in.copy %d in.datalen"
566                           " %d\n", tcp_conn->in.copy, tcp_conn->in.datalen);
567                 memcpy(&tcp_conn->hdr, tcp_conn->in.hdr,
568                        sizeof(struct iscsi_hdr));
569                 tcp_conn->in.hdr = &tcp_conn->hdr;
570                 tcp_conn->in.zero_copy_hdr = 0;
571         }
572         return 0;
573 }
574
575 /**
576  * iscsi_ctask_copy - copy skb bits to the destanation cmd task
577  * @conn: iscsi tcp connection
578  * @ctask: scsi command task
579  * @buf: buffer to copy to
580  * @buf_size: size of buffer
581  * @offset: offset within the buffer
582  *
583  * Notes:
584  *      The function calls skb_copy_bits() and updates per-connection and
585  *      per-cmd byte counters.
586  *
587  *      Read counters (in bytes):
588  *
589  *      conn->in.offset         offset within in progress SKB
590  *      conn->in.copy           left to copy from in progress SKB
591  *                              including padding
592  *      conn->in.copied         copied already from in progress SKB
593  *      conn->data_copied       copied already from in progress buffer
594  *      ctask->sent             total bytes sent up to the MidLayer
595  *      ctask->data_count       left to copy from in progress Data-In
596  *      buf_left                left to copy from in progress buffer
597  **/
598 static inline int
599 iscsi_ctask_copy(struct iscsi_tcp_conn *tcp_conn, struct iscsi_cmd_task *ctask,
600                 void *buf, int buf_size, int offset)
601 {
602         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
603         int buf_left = buf_size - (tcp_conn->data_copied + offset);
604         unsigned size = min(tcp_conn->in.copy, buf_left);
605         int rc;
606
607         size = min(size, ctask->data_count);
608
609         debug_tcp("ctask_copy %d bytes at offset %d copied %d\n",
610                size, tcp_conn->in.offset, tcp_conn->in.copied);
611
612         BUG_ON(size <= 0);
613         BUG_ON(tcp_ctask->sent + size > ctask->sc->request_bufflen);
614
615         rc = skb_copy_bits(tcp_conn->in.skb, tcp_conn->in.offset,
616                            (char*)buf + (offset + tcp_conn->data_copied), size);
617         /* must fit into skb->len */
618         BUG_ON(rc);
619
620         tcp_conn->in.offset += size;
621         tcp_conn->in.copy -= size;
622         tcp_conn->in.copied += size;
623         tcp_conn->data_copied += size;
624         tcp_ctask->sent += size;
625         ctask->data_count -= size;
626
627         BUG_ON(tcp_conn->in.copy < 0);
628         BUG_ON(ctask->data_count < 0);
629
630         if (buf_size != (tcp_conn->data_copied + offset)) {
631                 if (!ctask->data_count) {
632                         BUG_ON(buf_size - tcp_conn->data_copied < 0);
633                         /* done with this PDU */
634                         return buf_size - tcp_conn->data_copied;
635                 }
636                 return -EAGAIN;
637         }
638
639         /* done with this buffer or with both - PDU and buffer */
640         tcp_conn->data_copied = 0;
641         return 0;
642 }
643
644 /**
645  * iscsi_tcp_copy - copy skb bits to the destanation buffer
646  * @conn: iscsi tcp connection
647  *
648  * Notes:
649  *      The function calls skb_copy_bits() and updates per-connection
650  *      byte counters.
651  **/
652 static inline int
653 iscsi_tcp_copy(struct iscsi_conn *conn, int buf_size)
654 {
655         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
656         int buf_left = buf_size - tcp_conn->data_copied;
657         int size = min(tcp_conn->in.copy, buf_left);
658         int rc;
659
660         debug_tcp("tcp_copy %d bytes at offset %d copied %d\n",
661                size, tcp_conn->in.offset, tcp_conn->data_copied);
662         BUG_ON(size <= 0);
663
664         rc = skb_copy_bits(tcp_conn->in.skb, tcp_conn->in.offset,
665                            (char*)conn->data + tcp_conn->data_copied, size);
666         BUG_ON(rc);
667
668         tcp_conn->in.offset += size;
669         tcp_conn->in.copy -= size;
670         tcp_conn->in.copied += size;
671         tcp_conn->data_copied += size;
672
673         if (buf_size != tcp_conn->data_copied)
674                 return -EAGAIN;
675
676         return 0;
677 }
678
679 static inline void
680 partial_sg_digest_update(struct hash_desc *desc, struct scatterlist *sg,
681                          int offset, int length)
682 {
683         struct scatterlist temp;
684
685         memcpy(&temp, sg, sizeof(struct scatterlist));
686         temp.offset = offset;
687         temp.length = length;
688         crypto_hash_update(desc, &temp, length);
689 }
690
691 static void
692 iscsi_recv_digest_update(struct iscsi_tcp_conn *tcp_conn, char* buf, int len)
693 {
694         struct scatterlist tmp;
695
696         sg_init_one(&tmp, buf, len);
697         crypto_hash_update(&tcp_conn->rx_hash, &tmp, len);
698 }
699
700 static int iscsi_scsi_data_in(struct iscsi_conn *conn)
701 {
702         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
703         struct iscsi_cmd_task *ctask = tcp_conn->in.ctask;
704         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
705         struct scsi_cmnd *sc = ctask->sc;
706         struct scatterlist *sg;
707         int i, offset, rc = 0;
708
709         BUG_ON((void*)ctask != sc->SCp.ptr);
710
711         /*
712          * copying Data-In into the Scsi_Cmnd
713          */
714         if (!sc->use_sg) {
715                 i = ctask->data_count;
716                 rc = iscsi_ctask_copy(tcp_conn, ctask, sc->request_buffer,
717                                       sc->request_bufflen,
718                                       tcp_ctask->data_offset);
719                 if (rc == -EAGAIN)
720                         return rc;
721                 if (conn->datadgst_en)
722                         iscsi_recv_digest_update(tcp_conn, sc->request_buffer,
723                                                  i);
724                 rc = 0;
725                 goto done;
726         }
727
728         offset = tcp_ctask->data_offset;
729         sg = sc->request_buffer;
730
731         if (tcp_ctask->data_offset)
732                 for (i = 0; i < tcp_ctask->sg_count; i++)
733                         offset -= sg[i].length;
734         /* we've passed through partial sg*/
735         if (offset < 0)
736                 offset = 0;
737
738         for (i = tcp_ctask->sg_count; i < sc->use_sg; i++) {
739                 char *dest;
740
741                 dest = kmap_atomic(sg[i].page, KM_SOFTIRQ0);
742                 rc = iscsi_ctask_copy(tcp_conn, ctask, dest + sg[i].offset,
743                                       sg[i].length, offset);
744                 kunmap_atomic(dest, KM_SOFTIRQ0);
745                 if (rc == -EAGAIN)
746                         /* continue with the next SKB/PDU */
747                         return rc;
748                 if (!rc) {
749                         if (conn->datadgst_en) {
750                                 if (!offset)
751                                         crypto_hash_update(
752                                                         &tcp_conn->rx_hash,
753                                                         &sg[i], sg[i].length);
754                                 else
755                                         partial_sg_digest_update(
756                                                         &tcp_conn->rx_hash,
757                                                         &sg[i],
758                                                         sg[i].offset + offset,
759                                                         sg[i].length - offset);
760                         }
761                         offset = 0;
762                         tcp_ctask->sg_count++;
763                 }
764
765                 if (!ctask->data_count) {
766                         if (rc && conn->datadgst_en)
767                                 /*
768                                  * data-in is complete, but buffer not...
769                                  */
770                                 partial_sg_digest_update(&tcp_conn->rx_hash,
771                                                          &sg[i],
772                                                          sg[i].offset,
773                                                          sg[i].length-rc);
774                         rc = 0;
775                         break;
776                 }
777
778                 if (!tcp_conn->in.copy)
779                         return -EAGAIN;
780         }
781         BUG_ON(ctask->data_count);
782
783 done:
784         /* check for non-exceptional status */
785         if (tcp_conn->in.hdr->flags & ISCSI_FLAG_DATA_STATUS) {
786                 debug_scsi("done [sc %lx res %d itt 0x%x flags 0x%x]\n",
787                            (long)sc, sc->result, ctask->itt,
788                            tcp_conn->in.hdr->flags);
789                 spin_lock(&conn->session->lock);
790                 __iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
791                 spin_unlock(&conn->session->lock);
792         }
793
794         return rc;
795 }
796
797 static int
798 iscsi_data_recv(struct iscsi_conn *conn)
799 {
800         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
801         int rc = 0, opcode;
802
803         opcode = tcp_conn->in.hdr->opcode & ISCSI_OPCODE_MASK;
804         switch (opcode) {
805         case ISCSI_OP_SCSI_DATA_IN:
806                 rc = iscsi_scsi_data_in(conn);
807                 break;
808         case ISCSI_OP_SCSI_CMD_RSP:
809         case ISCSI_OP_TEXT_RSP:
810         case ISCSI_OP_LOGIN_RSP:
811         case ISCSI_OP_ASYNC_EVENT:
812         case ISCSI_OP_REJECT:
813                 /*
814                  * Collect data segment to the connection's data
815                  * placeholder
816                  */
817                 if (iscsi_tcp_copy(conn, tcp_conn->in.datalen)) {
818                         rc = -EAGAIN;
819                         goto exit;
820                 }
821
822                 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, conn->data,
823                                         tcp_conn->in.datalen);
824                 if (!rc && conn->datadgst_en && opcode != ISCSI_OP_LOGIN_RSP)
825                         iscsi_recv_digest_update(tcp_conn, conn->data,
826                                                 tcp_conn->in.datalen);
827                 break;
828         default:
829                 BUG_ON(1);
830         }
831 exit:
832         return rc;
833 }
834
835 /**
836  * iscsi_tcp_data_recv - TCP receive in sendfile fashion
837  * @rd_desc: read descriptor
838  * @skb: socket buffer
839  * @offset: offset in skb
840  * @len: skb->len - offset
841  **/
842 static int
843 iscsi_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
844                 unsigned int offset, size_t len)
845 {
846         int rc;
847         struct iscsi_conn *conn = rd_desc->arg.data;
848         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
849         int processed;
850         char pad[ISCSI_PAD_LEN];
851         struct scatterlist sg;
852
853         /*
854          * Save current SKB and its offset in the corresponding
855          * connection context.
856          */
857         tcp_conn->in.copy = skb->len - offset;
858         tcp_conn->in.offset = offset;
859         tcp_conn->in.skb = skb;
860         tcp_conn->in.len = tcp_conn->in.copy;
861         BUG_ON(tcp_conn->in.copy <= 0);
862         debug_tcp("in %d bytes\n", tcp_conn->in.copy);
863
864 more:
865         tcp_conn->in.copied = 0;
866         rc = 0;
867
868         if (unlikely(conn->suspend_rx)) {
869                 debug_tcp("conn %d Rx suspended!\n", conn->id);
870                 return 0;
871         }
872
873         if (tcp_conn->in_progress == IN_PROGRESS_WAIT_HEADER ||
874             tcp_conn->in_progress == IN_PROGRESS_HEADER_GATHER) {
875                 rc = iscsi_hdr_extract(tcp_conn);
876                 if (rc) {
877                        if (rc == -EAGAIN)
878                                 goto nomore;
879                        else {
880                                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
881                                 return 0;
882                        }
883                 }
884
885                 /*
886                  * Verify and process incoming PDU header.
887                  */
888                 rc = iscsi_tcp_hdr_recv(conn);
889                 if (!rc && tcp_conn->in.datalen) {
890                         if (conn->datadgst_en)
891                                 crypto_hash_init(&tcp_conn->rx_hash);
892                         tcp_conn->in_progress = IN_PROGRESS_DATA_RECV;
893                 } else if (rc) {
894                         iscsi_conn_failure(conn, rc);
895                         return 0;
896                 }
897         }
898
899         if (tcp_conn->in_progress == IN_PROGRESS_DDIGEST_RECV &&
900             tcp_conn->in.copy) {
901                 uint32_t recv_digest;
902
903                 debug_tcp("extra data_recv offset %d copy %d\n",
904                           tcp_conn->in.offset, tcp_conn->in.copy);
905
906                 if (!tcp_conn->data_copied) {
907                         if (tcp_conn->in.padding) {
908                                 debug_tcp("padding -> %d\n",
909                                           tcp_conn->in.padding);
910                                 memset(pad, 0, tcp_conn->in.padding);
911                                 sg_init_one(&sg, pad, tcp_conn->in.padding);
912                                 crypto_hash_update(&tcp_conn->rx_hash,
913                                                    &sg, sg.length);
914                         }
915                         crypto_hash_final(&tcp_conn->rx_hash,
916                                           (u8 *) &tcp_conn->in.datadgst);
917                         debug_tcp("rx digest 0x%x\n", tcp_conn->in.datadgst);
918                 }
919
920                 rc = iscsi_tcp_copy(conn, sizeof(uint32_t));
921                 if (rc) {
922                         if (rc == -EAGAIN)
923                                 goto again;
924                         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
925                         return 0;
926                 }
927
928                 memcpy(&recv_digest, conn->data, sizeof(uint32_t));
929                 if (recv_digest != tcp_conn->in.datadgst) {
930                         debug_tcp("iscsi_tcp: data digest error!"
931                                   "0x%x != 0x%x\n", recv_digest,
932                                   tcp_conn->in.datadgst);
933                         iscsi_conn_failure(conn, ISCSI_ERR_DATA_DGST);
934                         return 0;
935                 } else {
936                         debug_tcp("iscsi_tcp: data digest match!"
937                                   "0x%x == 0x%x\n", recv_digest,
938                                   tcp_conn->in.datadgst);
939                         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
940                 }
941         }
942
943         if (tcp_conn->in_progress == IN_PROGRESS_DATA_RECV &&
944             tcp_conn->in.copy) {
945                 debug_tcp("data_recv offset %d copy %d\n",
946                        tcp_conn->in.offset, tcp_conn->in.copy);
947
948                 rc = iscsi_data_recv(conn);
949                 if (rc) {
950                         if (rc == -EAGAIN)
951                                 goto again;
952                         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
953                         return 0;
954                 }
955
956                 if (tcp_conn->in.padding)
957                         tcp_conn->in_progress = IN_PROGRESS_PAD_RECV;
958                 else if (conn->datadgst_en)
959                         tcp_conn->in_progress = IN_PROGRESS_DDIGEST_RECV;
960                 else
961                         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
962                 tcp_conn->data_copied = 0;
963         }
964
965         if (tcp_conn->in_progress == IN_PROGRESS_PAD_RECV &&
966             tcp_conn->in.copy) {
967                 int copylen = min(tcp_conn->in.padding - tcp_conn->data_copied,
968                                   tcp_conn->in.copy);
969
970                 tcp_conn->in.copy -= copylen;
971                 tcp_conn->in.offset += copylen;
972                 tcp_conn->data_copied += copylen;
973
974                 if (tcp_conn->data_copied != tcp_conn->in.padding)
975                         tcp_conn->in_progress = IN_PROGRESS_PAD_RECV;
976                 else if (conn->datadgst_en)
977                         tcp_conn->in_progress = IN_PROGRESS_DDIGEST_RECV;
978                 else
979                         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
980                 tcp_conn->data_copied = 0;
981         }
982
983         debug_tcp("f, processed %d from out of %d padding %d\n",
984                tcp_conn->in.offset - offset, (int)len, tcp_conn->in.padding);
985         BUG_ON(tcp_conn->in.offset - offset > len);
986
987         if (tcp_conn->in.offset - offset != len) {
988                 debug_tcp("continue to process %d bytes\n",
989                        (int)len - (tcp_conn->in.offset - offset));
990                 goto more;
991         }
992
993 nomore:
994         processed = tcp_conn->in.offset - offset;
995         BUG_ON(processed == 0);
996         return processed;
997
998 again:
999         processed = tcp_conn->in.offset - offset;
1000         debug_tcp("c, processed %d from out of %d rd_desc_cnt %d\n",
1001                   processed, (int)len, (int)rd_desc->count);
1002         BUG_ON(processed == 0);
1003         BUG_ON(processed > len);
1004
1005         conn->rxdata_octets += processed;
1006         return processed;
1007 }
1008
1009 static void
1010 iscsi_tcp_data_ready(struct sock *sk, int flag)
1011 {
1012         struct iscsi_conn *conn = sk->sk_user_data;
1013         read_descriptor_t rd_desc;
1014
1015         read_lock(&sk->sk_callback_lock);
1016
1017         /*
1018          * Use rd_desc to pass 'conn' to iscsi_tcp_data_recv.
1019          * We set count to 1 because we want the network layer to
1020          * hand us all the skbs that are available. iscsi_tcp_data_recv
1021          * handled pdus that cross buffers or pdus that still need data.
1022          */
1023         rd_desc.arg.data = conn;
1024         rd_desc.count = 1;
1025         tcp_read_sock(sk, &rd_desc, iscsi_tcp_data_recv);
1026
1027         read_unlock(&sk->sk_callback_lock);
1028 }
1029
1030 static void
1031 iscsi_tcp_state_change(struct sock *sk)
1032 {
1033         struct iscsi_tcp_conn *tcp_conn;
1034         struct iscsi_conn *conn;
1035         struct iscsi_session *session;
1036         void (*old_state_change)(struct sock *);
1037
1038         read_lock(&sk->sk_callback_lock);
1039
1040         conn = (struct iscsi_conn*)sk->sk_user_data;
1041         session = conn->session;
1042
1043         if ((sk->sk_state == TCP_CLOSE_WAIT ||
1044              sk->sk_state == TCP_CLOSE) &&
1045             !atomic_read(&sk->sk_rmem_alloc)) {
1046                 debug_tcp("iscsi_tcp_state_change: TCP_CLOSE|TCP_CLOSE_WAIT\n");
1047                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1048         }
1049
1050         tcp_conn = conn->dd_data;
1051         old_state_change = tcp_conn->old_state_change;
1052
1053         read_unlock(&sk->sk_callback_lock);
1054
1055         old_state_change(sk);
1056 }
1057
1058 /**
1059  * iscsi_write_space - Called when more output buffer space is available
1060  * @sk: socket space is available for
1061  **/
1062 static void
1063 iscsi_write_space(struct sock *sk)
1064 {
1065         struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
1066         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1067
1068         tcp_conn->old_write_space(sk);
1069         debug_tcp("iscsi_write_space: cid %d\n", conn->id);
1070         scsi_queue_work(conn->session->host, &conn->xmitwork);
1071 }
1072
1073 static void
1074 iscsi_conn_set_callbacks(struct iscsi_conn *conn)
1075 {
1076         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1077         struct sock *sk = tcp_conn->sock->sk;
1078
1079         /* assign new callbacks */
1080         write_lock_bh(&sk->sk_callback_lock);
1081         sk->sk_user_data = conn;
1082         tcp_conn->old_data_ready = sk->sk_data_ready;
1083         tcp_conn->old_state_change = sk->sk_state_change;
1084         tcp_conn->old_write_space = sk->sk_write_space;
1085         sk->sk_data_ready = iscsi_tcp_data_ready;
1086         sk->sk_state_change = iscsi_tcp_state_change;
1087         sk->sk_write_space = iscsi_write_space;
1088         write_unlock_bh(&sk->sk_callback_lock);
1089 }
1090
1091 static void
1092 iscsi_conn_restore_callbacks(struct iscsi_tcp_conn *tcp_conn)
1093 {
1094         struct sock *sk = tcp_conn->sock->sk;
1095
1096         /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
1097         write_lock_bh(&sk->sk_callback_lock);
1098         sk->sk_user_data    = NULL;
1099         sk->sk_data_ready   = tcp_conn->old_data_ready;
1100         sk->sk_state_change = tcp_conn->old_state_change;
1101         sk->sk_write_space  = tcp_conn->old_write_space;
1102         sk->sk_no_check  = 0;
1103         write_unlock_bh(&sk->sk_callback_lock);
1104 }
1105
1106 /**
1107  * iscsi_send - generic send routine
1108  * @sk: kernel's socket
1109  * @buf: buffer to write from
1110  * @size: actual size to write
1111  * @flags: socket's flags
1112  */
1113 static inline int
1114 iscsi_send(struct iscsi_conn *conn, struct iscsi_buf *buf, int size, int flags)
1115 {
1116         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1117         struct socket *sk = tcp_conn->sock;
1118         int offset = buf->sg.offset + buf->sent, res;
1119
1120         /*
1121          * if we got use_sg=0 or are sending something we kmallocd
1122          * then we did not have to do kmap (kmap returns page_address)
1123          *
1124          * if we got use_sg > 0, but had to drop down, we do not
1125          * set clustering so this should only happen for that
1126          * slab case.
1127          */
1128         if (buf->use_sendmsg)
1129                 res = sock_no_sendpage(sk, buf->sg.page, offset, size, flags);
1130         else
1131                 res = tcp_conn->sendpage(sk, buf->sg.page, offset, size, flags);
1132
1133         if (res >= 0) {
1134                 conn->txdata_octets += res;
1135                 buf->sent += res;
1136                 return res;
1137         }
1138
1139         tcp_conn->sendpage_failures_cnt++;
1140         if (res == -EAGAIN)
1141                 res = -ENOBUFS;
1142         else
1143                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1144         return res;
1145 }
1146
1147 /**
1148  * iscsi_sendhdr - send PDU Header via tcp_sendpage()
1149  * @conn: iscsi connection
1150  * @buf: buffer to write from
1151  * @datalen: lenght of data to be sent after the header
1152  *
1153  * Notes:
1154  *      (Tx, Fast Path)
1155  **/
1156 static inline int
1157 iscsi_sendhdr(struct iscsi_conn *conn, struct iscsi_buf *buf, int datalen)
1158 {
1159         int flags = 0; /* MSG_DONTWAIT; */
1160         int res, size;
1161
1162         size = buf->sg.length - buf->sent;
1163         BUG_ON(buf->sent + size > buf->sg.length);
1164         if (buf->sent + size != buf->sg.length || datalen)
1165                 flags |= MSG_MORE;
1166
1167         res = iscsi_send(conn, buf, size, flags);
1168         debug_tcp("sendhdr %d bytes, sent %d res %d\n", size, buf->sent, res);
1169         if (res >= 0) {
1170                 if (size != res)
1171                         return -EAGAIN;
1172                 return 0;
1173         }
1174
1175         return res;
1176 }
1177
1178 /**
1179  * iscsi_sendpage - send one page of iSCSI Data-Out.
1180  * @conn: iscsi connection
1181  * @buf: buffer to write from
1182  * @count: remaining data
1183  * @sent: number of bytes sent
1184  *
1185  * Notes:
1186  *      (Tx, Fast Path)
1187  **/
1188 static inline int
1189 iscsi_sendpage(struct iscsi_conn *conn, struct iscsi_buf *buf,
1190                int *count, int *sent)
1191 {
1192         int flags = 0; /* MSG_DONTWAIT; */
1193         int res, size;
1194
1195         size = buf->sg.length - buf->sent;
1196         BUG_ON(buf->sent + size > buf->sg.length);
1197         if (size > *count)
1198                 size = *count;
1199         if (buf->sent + size != buf->sg.length || *count != size)
1200                 flags |= MSG_MORE;
1201
1202         res = iscsi_send(conn, buf, size, flags);
1203         debug_tcp("sendpage: %d bytes, sent %d left %d sent %d res %d\n",
1204                   size, buf->sent, *count, *sent, res);
1205         if (res >= 0) {
1206                 *count -= res;
1207                 *sent += res;
1208                 if (size != res)
1209                         return -EAGAIN;
1210                 return 0;
1211         }
1212
1213         return res;
1214 }
1215
1216 static inline void
1217 iscsi_data_digest_init(struct iscsi_tcp_conn *tcp_conn,
1218                       struct iscsi_tcp_cmd_task *tcp_ctask)
1219 {
1220         crypto_hash_init(&tcp_conn->tx_hash);
1221         tcp_ctask->digest_count = 4;
1222 }
1223
1224 /**
1225  * iscsi_solicit_data_cont - initialize next Data-Out
1226  * @conn: iscsi connection
1227  * @ctask: scsi command task
1228  * @r2t: R2T info
1229  * @left: bytes left to transfer
1230  *
1231  * Notes:
1232  *      Initialize next Data-Out within this R2T sequence and continue
1233  *      to process next Scatter-Gather element(if any) of this SCSI command.
1234  *
1235  *      Called under connection lock.
1236  **/
1237 static void
1238 iscsi_solicit_data_cont(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
1239                         struct iscsi_r2t_info *r2t, int left)
1240 {
1241         struct iscsi_data *hdr;
1242         struct scsi_cmnd *sc = ctask->sc;
1243         int new_offset;
1244
1245         hdr = &r2t->dtask.hdr;
1246         memset(hdr, 0, sizeof(struct iscsi_data));
1247         hdr->ttt = r2t->ttt;
1248         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
1249         r2t->solicit_datasn++;
1250         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
1251         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
1252         hdr->itt = ctask->hdr->itt;
1253         hdr->exp_statsn = r2t->exp_statsn;
1254         new_offset = r2t->data_offset + r2t->sent;
1255         hdr->offset = cpu_to_be32(new_offset);
1256         if (left > conn->max_xmit_dlength) {
1257                 hton24(hdr->dlength, conn->max_xmit_dlength);
1258                 r2t->data_count = conn->max_xmit_dlength;
1259         } else {
1260                 hton24(hdr->dlength, left);
1261                 r2t->data_count = left;
1262                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
1263         }
1264         conn->dataout_pdus_cnt++;
1265
1266         iscsi_buf_init_iov(&r2t->headbuf, (char*)hdr,
1267                            sizeof(struct iscsi_hdr));
1268
1269         if (iscsi_buf_left(&r2t->sendbuf))
1270                 return;
1271
1272         if (sc->use_sg) {
1273                 iscsi_buf_init_sg(&r2t->sendbuf, r2t->sg);
1274                 r2t->sg += 1;
1275         } else {
1276                 iscsi_buf_init_iov(&r2t->sendbuf,
1277                             (char*)sc->request_buffer + new_offset,
1278                             r2t->data_count);
1279                 r2t->sg = NULL;
1280         }
1281 }
1282
1283 static void iscsi_set_padding(struct iscsi_tcp_cmd_task *tcp_ctask,
1284                               unsigned long len)
1285 {
1286         tcp_ctask->pad_count = len & (ISCSI_PAD_LEN - 1);
1287         if (!tcp_ctask->pad_count)
1288                 return;
1289
1290         tcp_ctask->pad_count = ISCSI_PAD_LEN - tcp_ctask->pad_count;
1291         debug_scsi("write padding %d bytes\n", tcp_ctask->pad_count);
1292         tcp_ctask->xmstate |= XMSTATE_W_PAD;
1293 }
1294
1295 /**
1296  * iscsi_tcp_cmd_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
1297  * @conn: iscsi connection
1298  * @ctask: scsi command task
1299  * @sc: scsi command
1300  **/
1301 static void
1302 iscsi_tcp_cmd_init(struct iscsi_cmd_task *ctask)
1303 {
1304         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1305
1306         BUG_ON(__kfifo_len(tcp_ctask->r2tqueue));
1307         tcp_ctask->xmstate = XMSTATE_CMD_HDR_INIT;
1308 }
1309
1310 /**
1311  * iscsi_tcp_mtask_xmit - xmit management(immediate) task
1312  * @conn: iscsi connection
1313  * @mtask: task management task
1314  *
1315  * Notes:
1316  *      The function can return -EAGAIN in which case caller must
1317  *      call it again later, or recover. '0' return code means successful
1318  *      xmit.
1319  *
1320  *      Management xmit state machine consists of these states:
1321  *              XMSTATE_IMM_HDR_INIT    - calculate digest of PDU Header
1322  *              XMSTATE_IMM_HDR         - PDU Header xmit in progress
1323  *              XMSTATE_IMM_DATA        - PDU Data xmit in progress
1324  *              XMSTATE_IDLE            - management PDU is done
1325  **/
1326 static int
1327 iscsi_tcp_mtask_xmit(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1328 {
1329         struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
1330         int rc;
1331
1332         debug_scsi("mtask deq [cid %d state %x itt 0x%x]\n",
1333                 conn->id, tcp_mtask->xmstate, mtask->itt);
1334
1335         if (tcp_mtask->xmstate & XMSTATE_IMM_HDR_INIT) {
1336                 iscsi_buf_init_iov(&tcp_mtask->headbuf, (char*)mtask->hdr,
1337                                    sizeof(struct iscsi_hdr));
1338
1339                 if (mtask->data_count) {
1340                         tcp_mtask->xmstate |= XMSTATE_IMM_DATA;
1341                         iscsi_buf_init_iov(&tcp_mtask->sendbuf,
1342                                            (char*)mtask->data,
1343                                            mtask->data_count);
1344                 }
1345
1346                 if (conn->c_stage != ISCSI_CONN_INITIAL_STAGE &&
1347                     conn->stop_stage != STOP_CONN_RECOVER &&
1348                     conn->hdrdgst_en)
1349                         iscsi_hdr_digest(conn, &tcp_mtask->headbuf,
1350                                         (u8*)tcp_mtask->hdrext);
1351
1352                 tcp_mtask->sent = 0;
1353                 tcp_mtask->xmstate &= ~XMSTATE_IMM_HDR_INIT;
1354                 tcp_mtask->xmstate |= XMSTATE_IMM_HDR;
1355         }
1356
1357         if (tcp_mtask->xmstate & XMSTATE_IMM_HDR) {
1358                 rc = iscsi_sendhdr(conn, &tcp_mtask->headbuf,
1359                                    mtask->data_count);
1360                 if (rc)
1361                         return rc;
1362                 tcp_mtask->xmstate &= ~XMSTATE_IMM_HDR;
1363         }
1364
1365         if (tcp_mtask->xmstate & XMSTATE_IMM_DATA) {
1366                 BUG_ON(!mtask->data_count);
1367                 tcp_mtask->xmstate &= ~XMSTATE_IMM_DATA;
1368                 /* FIXME: implement.
1369                  * Virtual buffer could be spreaded across multiple pages...
1370                  */
1371                 do {
1372                         int rc;
1373
1374                         rc = iscsi_sendpage(conn, &tcp_mtask->sendbuf,
1375                                         &mtask->data_count, &tcp_mtask->sent);
1376                         if (rc) {
1377                                 tcp_mtask->xmstate |= XMSTATE_IMM_DATA;
1378                                 return rc;
1379                         }
1380                 } while (mtask->data_count);
1381         }
1382
1383         BUG_ON(tcp_mtask->xmstate != XMSTATE_IDLE);
1384         if (mtask->hdr->itt == RESERVED_ITT) {
1385                 struct iscsi_session *session = conn->session;
1386
1387                 spin_lock_bh(&session->lock);
1388                 list_del(&conn->mtask->running);
1389                 __kfifo_put(session->mgmtpool.queue, (void*)&conn->mtask,
1390                             sizeof(void*));
1391                 spin_unlock_bh(&session->lock);
1392         }
1393         return 0;
1394 }
1395
1396 static int
1397 iscsi_send_cmd_hdr(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1398 {
1399         struct scsi_cmnd *sc = ctask->sc;
1400         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1401         int rc = 0;
1402
1403         if (tcp_ctask->xmstate & XMSTATE_CMD_HDR_INIT) {
1404                 tcp_ctask->sent = 0;
1405                 tcp_ctask->sg_count = 0;
1406                 tcp_ctask->exp_datasn = 0;
1407
1408                 if (sc->sc_data_direction == DMA_TO_DEVICE) {
1409                         if (sc->use_sg) {
1410                                 struct scatterlist *sg = sc->request_buffer;
1411
1412                                 iscsi_buf_init_sg(&tcp_ctask->sendbuf, sg);
1413                                 tcp_ctask->sg = sg + 1;
1414                                 tcp_ctask->bad_sg = sg + sc->use_sg;
1415                         } else {
1416                                 iscsi_buf_init_iov(&tcp_ctask->sendbuf,
1417                                                    sc->request_buffer,
1418                                                    sc->request_bufflen);
1419                                 tcp_ctask->sg = NULL;
1420                                 tcp_ctask->bad_sg = NULL;
1421                         }
1422
1423                         debug_scsi("cmd [itt 0x%x total %d imm_data %d "
1424                                    "unsol count %d, unsol offset %d]\n",
1425                                    ctask->itt, sc->request_bufflen,
1426                                    ctask->imm_count, ctask->unsol_count,
1427                                    ctask->unsol_offset);
1428                 }
1429
1430                 iscsi_buf_init_iov(&tcp_ctask->headbuf, (char*)ctask->hdr,
1431                                   sizeof(struct iscsi_hdr));
1432
1433                 if (conn->hdrdgst_en)
1434                         iscsi_hdr_digest(conn, &tcp_ctask->headbuf,
1435                                          (u8*)tcp_ctask->hdrext);
1436                 tcp_ctask->xmstate &= ~XMSTATE_CMD_HDR_INIT;
1437                 tcp_ctask->xmstate |= XMSTATE_CMD_HDR_XMIT;
1438         }
1439
1440         if (tcp_ctask->xmstate & XMSTATE_CMD_HDR_XMIT) {
1441                 rc = iscsi_sendhdr(conn, &tcp_ctask->headbuf, ctask->imm_count);
1442                 if (rc)
1443                         return rc;
1444                 tcp_ctask->xmstate &= ~XMSTATE_CMD_HDR_XMIT;
1445
1446                 if (sc->sc_data_direction != DMA_TO_DEVICE)
1447                         return 0;
1448
1449                 if (ctask->imm_count) {
1450                         tcp_ctask->xmstate |= XMSTATE_IMM_DATA;
1451                         iscsi_set_padding(tcp_ctask, ctask->imm_count);
1452
1453                         if (ctask->conn->datadgst_en) {
1454                                 iscsi_data_digest_init(ctask->conn->dd_data,
1455                                                        tcp_ctask);
1456                                 tcp_ctask->immdigest = 0;
1457                         }
1458                 }
1459
1460                 if (ctask->unsol_count)
1461                         tcp_ctask->xmstate |=
1462                                         XMSTATE_UNS_HDR | XMSTATE_UNS_INIT;
1463         }
1464         return rc;
1465 }
1466
1467 static int
1468 iscsi_send_padding(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1469 {
1470         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1471         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1472         int sent = 0, rc;
1473
1474         if (tcp_ctask->xmstate & XMSTATE_W_PAD) {
1475                 iscsi_buf_init_iov(&tcp_ctask->sendbuf, (char*)&tcp_ctask->pad,
1476                                    tcp_ctask->pad_count);
1477                 if (conn->datadgst_en)
1478                         crypto_hash_update(&tcp_conn->tx_hash,
1479                                            &tcp_ctask->sendbuf.sg,
1480                                            tcp_ctask->sendbuf.sg.length);
1481         } else if (!(tcp_ctask->xmstate & XMSTATE_W_RESEND_PAD))
1482                 return 0;
1483
1484         tcp_ctask->xmstate &= ~XMSTATE_W_PAD;
1485         tcp_ctask->xmstate &= ~XMSTATE_W_RESEND_PAD;
1486         debug_scsi("sending %d pad bytes for itt 0x%x\n",
1487                    tcp_ctask->pad_count, ctask->itt);
1488         rc = iscsi_sendpage(conn, &tcp_ctask->sendbuf, &tcp_ctask->pad_count,
1489                            &sent);
1490         if (rc) {
1491                 debug_scsi("padding send failed %d\n", rc);
1492                 tcp_ctask->xmstate |= XMSTATE_W_RESEND_PAD;
1493         }
1494         return rc;
1495 }
1496
1497 static int
1498 iscsi_send_digest(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
1499                         struct iscsi_buf *buf, uint32_t *digest)
1500 {
1501         struct iscsi_tcp_cmd_task *tcp_ctask;
1502         struct iscsi_tcp_conn *tcp_conn;
1503         int rc, sent = 0;
1504
1505         if (!conn->datadgst_en)
1506                 return 0;
1507
1508         tcp_ctask = ctask->dd_data;
1509         tcp_conn = conn->dd_data;
1510
1511         if (!(tcp_ctask->xmstate & XMSTATE_W_RESEND_DATA_DIGEST)) {
1512                 crypto_hash_final(&tcp_conn->tx_hash, (u8*)digest);
1513                 iscsi_buf_init_iov(buf, (char*)digest, 4);
1514         }
1515         tcp_ctask->xmstate &= ~XMSTATE_W_RESEND_DATA_DIGEST;
1516
1517         rc = iscsi_sendpage(conn, buf, &tcp_ctask->digest_count, &sent);
1518         if (!rc)
1519                 debug_scsi("sent digest 0x%x for itt 0x%x\n", *digest,
1520                           ctask->itt);
1521         else {
1522                 debug_scsi("sending digest 0x%x failed for itt 0x%x!\n",
1523                           *digest, ctask->itt);
1524                 tcp_ctask->xmstate |= XMSTATE_W_RESEND_DATA_DIGEST;
1525         }
1526         return rc;
1527 }
1528
1529 static int
1530 iscsi_send_data(struct iscsi_cmd_task *ctask, struct iscsi_buf *sendbuf,
1531                 struct scatterlist **sg, int *sent, int *count,
1532                 struct iscsi_buf *digestbuf, uint32_t *digest)
1533 {
1534         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1535         struct iscsi_conn *conn = ctask->conn;
1536         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1537         int rc, buf_sent, offset;
1538
1539         while (*count) {
1540                 buf_sent = 0;
1541                 offset = sendbuf->sent;
1542
1543                 rc = iscsi_sendpage(conn, sendbuf, count, &buf_sent);
1544                 *sent = *sent + buf_sent;
1545                 if (buf_sent && conn->datadgst_en)
1546                         partial_sg_digest_update(&tcp_conn->tx_hash,
1547                                 &sendbuf->sg, sendbuf->sg.offset + offset,
1548                                 buf_sent);
1549                 if (!iscsi_buf_left(sendbuf) && *sg != tcp_ctask->bad_sg) {
1550                         iscsi_buf_init_sg(sendbuf, *sg);
1551                         *sg = *sg + 1;
1552                 }
1553
1554                 if (rc)
1555                         return rc;
1556         }
1557
1558         rc = iscsi_send_padding(conn, ctask);
1559         if (rc)
1560                 return rc;
1561
1562         return iscsi_send_digest(conn, ctask, digestbuf, digest);
1563 }
1564
1565 static int
1566 iscsi_send_unsol_hdr(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1567 {
1568         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1569         struct iscsi_data_task *dtask;
1570         int rc;
1571
1572         tcp_ctask->xmstate |= XMSTATE_UNS_DATA;
1573         if (tcp_ctask->xmstate & XMSTATE_UNS_INIT) {
1574                 dtask = &tcp_ctask->unsol_dtask;
1575
1576                 iscsi_prep_unsolicit_data_pdu(ctask, &dtask->hdr);
1577                 iscsi_buf_init_iov(&tcp_ctask->headbuf, (char*)&dtask->hdr,
1578                                    sizeof(struct iscsi_hdr));
1579                 if (conn->hdrdgst_en)
1580                         iscsi_hdr_digest(conn, &tcp_ctask->headbuf,
1581                                         (u8*)dtask->hdrext);
1582
1583                 tcp_ctask->xmstate &= ~XMSTATE_UNS_INIT;
1584                 iscsi_set_padding(tcp_ctask, ctask->data_count);
1585         }
1586
1587         rc = iscsi_sendhdr(conn, &tcp_ctask->headbuf, ctask->data_count);
1588         if (rc) {
1589                 tcp_ctask->xmstate &= ~XMSTATE_UNS_DATA;
1590                 tcp_ctask->xmstate |= XMSTATE_UNS_HDR;
1591                 return rc;
1592         }
1593
1594         if (conn->datadgst_en) {
1595                 dtask = &tcp_ctask->unsol_dtask;
1596                 iscsi_data_digest_init(ctask->conn->dd_data, tcp_ctask);
1597                 dtask->digest = 0;
1598         }
1599
1600         debug_scsi("uns dout [itt 0x%x dlen %d sent %d]\n",
1601                    ctask->itt, ctask->unsol_count, tcp_ctask->sent);
1602         return 0;
1603 }
1604
1605 static int
1606 iscsi_send_unsol_pdu(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1607 {
1608         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1609         int rc;
1610
1611         if (tcp_ctask->xmstate & XMSTATE_UNS_HDR) {
1612                 BUG_ON(!ctask->unsol_count);
1613                 tcp_ctask->xmstate &= ~XMSTATE_UNS_HDR;
1614 send_hdr:
1615                 rc = iscsi_send_unsol_hdr(conn, ctask);
1616                 if (rc)
1617                         return rc;
1618         }
1619
1620         if (tcp_ctask->xmstate & XMSTATE_UNS_DATA) {
1621                 struct iscsi_data_task *dtask = &tcp_ctask->unsol_dtask;
1622                 int start = tcp_ctask->sent;
1623
1624                 rc = iscsi_send_data(ctask, &tcp_ctask->sendbuf, &tcp_ctask->sg,
1625                                      &tcp_ctask->sent, &ctask->data_count,
1626                                      &dtask->digestbuf, &dtask->digest);
1627                 ctask->unsol_count -= tcp_ctask->sent - start;
1628                 if (rc)
1629                         return rc;
1630                 tcp_ctask->xmstate &= ~XMSTATE_UNS_DATA;
1631                 /*
1632                  * Done with the Data-Out. Next, check if we need
1633                  * to send another unsolicited Data-Out.
1634                  */
1635                 if (ctask->unsol_count) {
1636                         debug_scsi("sending more uns\n");
1637                         tcp_ctask->xmstate |= XMSTATE_UNS_INIT;
1638                         goto send_hdr;
1639                 }
1640         }
1641         return 0;
1642 }
1643
1644 static int iscsi_send_sol_pdu(struct iscsi_conn *conn,
1645                               struct iscsi_cmd_task *ctask)
1646 {
1647         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1648         struct iscsi_session *session = conn->session;
1649         struct iscsi_r2t_info *r2t;
1650         struct iscsi_data_task *dtask;
1651         int left, rc;
1652
1653         if (tcp_ctask->xmstate & XMSTATE_SOL_HDR_INIT) {
1654                 if (!tcp_ctask->r2t) {
1655                         spin_lock_bh(&session->lock);
1656                         __kfifo_get(tcp_ctask->r2tqueue, (void*)&tcp_ctask->r2t,
1657                                     sizeof(void*));
1658                         spin_unlock_bh(&session->lock);
1659                 }
1660 send_hdr:
1661                 r2t = tcp_ctask->r2t;
1662                 dtask = &r2t->dtask;
1663
1664                 if (conn->hdrdgst_en)
1665                         iscsi_hdr_digest(conn, &r2t->headbuf,
1666                                         (u8*)dtask->hdrext);
1667                 tcp_ctask->xmstate &= ~XMSTATE_SOL_HDR_INIT;
1668                 tcp_ctask->xmstate |= XMSTATE_SOL_HDR;
1669         }
1670
1671         if (tcp_ctask->xmstate & XMSTATE_SOL_HDR) {
1672                 r2t = tcp_ctask->r2t;
1673                 dtask = &r2t->dtask;
1674
1675                 rc = iscsi_sendhdr(conn, &r2t->headbuf, r2t->data_count);
1676                 if (rc)
1677                         return rc;
1678                 tcp_ctask->xmstate &= ~XMSTATE_SOL_HDR;
1679                 tcp_ctask->xmstate |= XMSTATE_SOL_DATA;
1680
1681                 if (conn->datadgst_en) {
1682                         iscsi_data_digest_init(conn->dd_data, tcp_ctask);
1683                         dtask->digest = 0;
1684                 }
1685
1686                 iscsi_set_padding(tcp_ctask, r2t->data_count);
1687                 debug_scsi("sol dout [dsn %d itt 0x%x dlen %d sent %d]\n",
1688                         r2t->solicit_datasn - 1, ctask->itt, r2t->data_count,
1689                         r2t->sent);
1690         }
1691
1692         if (tcp_ctask->xmstate & XMSTATE_SOL_DATA) {
1693                 r2t = tcp_ctask->r2t;
1694                 dtask = &r2t->dtask;
1695
1696                 rc = iscsi_send_data(ctask, &r2t->sendbuf, &r2t->sg,
1697                                      &r2t->sent, &r2t->data_count,
1698                                      &dtask->digestbuf, &dtask->digest);
1699                 if (rc)
1700                         return rc;
1701                 tcp_ctask->xmstate &= ~XMSTATE_SOL_DATA;
1702
1703                 /*
1704                  * Done with this Data-Out. Next, check if we have
1705                  * to send another Data-Out for this R2T.
1706                  */
1707                 BUG_ON(r2t->data_length - r2t->sent < 0);
1708                 left = r2t->data_length - r2t->sent;
1709                 if (left) {
1710                         iscsi_solicit_data_cont(conn, ctask, r2t, left);
1711                         goto send_hdr;
1712                 }
1713
1714                 /*
1715                  * Done with this R2T. Check if there are more
1716                  * outstanding R2Ts ready to be processed.
1717                  */
1718                 spin_lock_bh(&session->lock);
1719                 tcp_ctask->r2t = NULL;
1720                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
1721                             sizeof(void*));
1722                 if (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t,
1723                                 sizeof(void*))) {
1724                         tcp_ctask->r2t = r2t;
1725                         spin_unlock_bh(&session->lock);
1726                         goto send_hdr;
1727                 }
1728                 spin_unlock_bh(&session->lock);
1729         }
1730         return 0;
1731 }
1732
1733 /**
1734  * iscsi_tcp_ctask_xmit - xmit normal PDU task
1735  * @conn: iscsi connection
1736  * @ctask: iscsi command task
1737  *
1738  * Notes:
1739  *      The function can return -EAGAIN in which case caller must
1740  *      call it again later, or recover. '0' return code means successful
1741  *      xmit.
1742  *      The function is devided to logical helpers (above) for the different
1743  *      xmit stages.
1744  *
1745  *iscsi_send_cmd_hdr()
1746  *      XMSTATE_CMD_HDR_INIT - prepare Header and Data buffers Calculate
1747  *                             Header Digest
1748  *      XMSTATE_CMD_HDR_XMIT - Transmit header in progress
1749  *
1750  *iscsi_send_padding
1751  *      XMSTATE_W_PAD        - Prepare and send pading
1752  *      XMSTATE_W_RESEND_PAD - retry send pading
1753  *
1754  *iscsi_send_digest
1755  *      XMSTATE_W_RESEND_DATA_DIGEST - Finalize and send Data Digest
1756  *      XMSTATE_W_RESEND_DATA_DIGEST - retry sending digest
1757  *
1758  *iscsi_send_unsol_hdr
1759  *      XMSTATE_UNS_INIT     - prepare un-solicit data header and digest
1760  *      XMSTATE_UNS_HDR      - send un-solicit header
1761  *
1762  *iscsi_send_unsol_pdu
1763  *      XMSTATE_UNS_DATA     - send un-solicit data in progress
1764  *
1765  *iscsi_send_sol_pdu
1766  *      XMSTATE_SOL_HDR_INIT - solicit data header and digest initialize
1767  *      XMSTATE_SOL_HDR      - send solicit header
1768  *      XMSTATE_SOL_DATA     - send solicit data
1769  *
1770  *iscsi_tcp_ctask_xmit
1771  *      XMSTATE_IMM_DATA     - xmit managment data (??)
1772  **/
1773 static int
1774 iscsi_tcp_ctask_xmit(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1775 {
1776         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1777         int rc = 0;
1778
1779         debug_scsi("ctask deq [cid %d xmstate %x itt 0x%x]\n",
1780                 conn->id, tcp_ctask->xmstate, ctask->itt);
1781
1782         rc = iscsi_send_cmd_hdr(conn, ctask);
1783         if (rc)
1784                 return rc;
1785         if (ctask->sc->sc_data_direction != DMA_TO_DEVICE)
1786                 return 0;
1787
1788         if (tcp_ctask->xmstate & XMSTATE_IMM_DATA) {
1789                 rc = iscsi_send_data(ctask, &tcp_ctask->sendbuf, &tcp_ctask->sg,
1790                                      &tcp_ctask->sent, &ctask->imm_count,
1791                                      &tcp_ctask->immbuf, &tcp_ctask->immdigest);
1792                 if (rc)
1793                         return rc;
1794                 tcp_ctask->xmstate &= ~XMSTATE_IMM_DATA;
1795         }
1796
1797         rc = iscsi_send_unsol_pdu(conn, ctask);
1798         if (rc)
1799                 return rc;
1800
1801         rc = iscsi_send_sol_pdu(conn, ctask);
1802         if (rc)
1803                 return rc;
1804
1805         return rc;
1806 }
1807
1808 static struct iscsi_cls_conn *
1809 iscsi_tcp_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
1810 {
1811         struct iscsi_conn *conn;
1812         struct iscsi_cls_conn *cls_conn;
1813         struct iscsi_tcp_conn *tcp_conn;
1814
1815         cls_conn = iscsi_conn_setup(cls_session, conn_idx);
1816         if (!cls_conn)
1817                 return NULL;
1818         conn = cls_conn->dd_data;
1819         /*
1820          * due to strange issues with iser these are not set
1821          * in iscsi_conn_setup
1822          */
1823         conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1824
1825         tcp_conn = kzalloc(sizeof(*tcp_conn), GFP_KERNEL);
1826         if (!tcp_conn)
1827                 goto tcp_conn_alloc_fail;
1828
1829         conn->dd_data = tcp_conn;
1830         tcp_conn->iscsi_conn = conn;
1831         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
1832         /* initial operational parameters */
1833         tcp_conn->hdr_size = sizeof(struct iscsi_hdr);
1834
1835         tcp_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1836                                                   CRYPTO_ALG_ASYNC);
1837         tcp_conn->tx_hash.flags = 0;
1838         if (IS_ERR(tcp_conn->tx_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->tx_hash.tfm));
1843                 goto free_tcp_conn;
1844         }
1845
1846         tcp_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1847                                                   CRYPTO_ALG_ASYNC);
1848         tcp_conn->rx_hash.flags = 0;
1849         if (IS_ERR(tcp_conn->rx_hash.tfm)) {
1850                 printk(KERN_ERR "Could not create connection due to crc32c "
1851                        "loading error %ld. Make sure the crc32c module is "
1852                        "built as a module or into the kernel\n",
1853                         PTR_ERR(tcp_conn->rx_hash.tfm));
1854                 goto free_tx_tfm;
1855         }
1856
1857         return cls_conn;
1858
1859 free_tx_tfm:
1860         crypto_free_hash(tcp_conn->tx_hash.tfm);
1861 free_tcp_conn:
1862         kfree(tcp_conn);
1863 tcp_conn_alloc_fail:
1864         iscsi_conn_teardown(cls_conn);
1865         return NULL;
1866 }
1867
1868 static void
1869 iscsi_tcp_release_conn(struct iscsi_conn *conn)
1870 {
1871         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1872
1873         if (!tcp_conn->sock)
1874                 return;
1875
1876         sock_hold(tcp_conn->sock->sk);
1877         iscsi_conn_restore_callbacks(tcp_conn);
1878         sock_put(tcp_conn->sock->sk);
1879
1880         sock_release(tcp_conn->sock);
1881         tcp_conn->sock = NULL;
1882         conn->recv_lock = NULL;
1883 }
1884
1885 static void
1886 iscsi_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
1887 {
1888         struct iscsi_conn *conn = cls_conn->dd_data;
1889         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1890
1891         iscsi_tcp_release_conn(conn);
1892         iscsi_conn_teardown(cls_conn);
1893
1894         if (tcp_conn->tx_hash.tfm)
1895                 crypto_free_hash(tcp_conn->tx_hash.tfm);
1896         if (tcp_conn->rx_hash.tfm)
1897                 crypto_free_hash(tcp_conn->rx_hash.tfm);
1898
1899         kfree(tcp_conn);
1900 }
1901
1902 static void
1903 iscsi_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
1904 {
1905         struct iscsi_conn *conn = cls_conn->dd_data;
1906         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1907
1908         iscsi_conn_stop(cls_conn, flag);
1909         iscsi_tcp_release_conn(conn);
1910         tcp_conn->hdr_size = sizeof(struct iscsi_hdr);
1911 }
1912
1913 static int
1914 iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
1915                     struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
1916                     int is_leading)
1917 {
1918         struct iscsi_conn *conn = cls_conn->dd_data;
1919         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1920         struct sock *sk;
1921         struct socket *sock;
1922         int err;
1923
1924         /* lookup for existing socket */
1925         sock = sockfd_lookup((int)transport_eph, &err);
1926         if (!sock) {
1927                 printk(KERN_ERR "iscsi_tcp: sockfd_lookup failed %d\n", err);
1928                 return -EEXIST;
1929         }
1930
1931         err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
1932         if (err)
1933                 return err;
1934
1935         /* bind iSCSI connection and socket */
1936         tcp_conn->sock = sock;
1937
1938         /* setup Socket parameters */
1939         sk = sock->sk;
1940         sk->sk_reuse = 1;
1941         sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
1942         sk->sk_allocation = GFP_ATOMIC;
1943
1944         /* FIXME: disable Nagle's algorithm */
1945
1946         /*
1947          * Intercept TCP callbacks for sendfile like receive
1948          * processing.
1949          */
1950         conn->recv_lock = &sk->sk_callback_lock;
1951         iscsi_conn_set_callbacks(conn);
1952         tcp_conn->sendpage = tcp_conn->sock->ops->sendpage;
1953         /*
1954          * set receive state machine into initial state
1955          */
1956         tcp_conn->in_progress = IN_PROGRESS_WAIT_HEADER;
1957
1958         return 0;
1959 }
1960
1961 /* called with host lock */
1962 static void
1963 iscsi_tcp_mgmt_init(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1964 {
1965         struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
1966         tcp_mtask->xmstate = XMSTATE_IMM_HDR_INIT;
1967 }
1968
1969 static int
1970 iscsi_r2tpool_alloc(struct iscsi_session *session)
1971 {
1972         int i;
1973         int cmd_i;
1974
1975         /*
1976          * initialize per-task: R2T pool and xmit queue
1977          */
1978         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1979                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
1980                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1981
1982                 /*
1983                  * pre-allocated x4 as much r2ts to handle race when
1984                  * target acks DataOut faster than we data_xmit() queues
1985                  * could replenish r2tqueue.
1986                  */
1987
1988                 /* R2T pool */
1989                 if (iscsi_pool_init(&tcp_ctask->r2tpool, session->max_r2t * 4,
1990                                     (void***)&tcp_ctask->r2ts,
1991                                     sizeof(struct iscsi_r2t_info))) {
1992                         goto r2t_alloc_fail;
1993                 }
1994
1995                 /* R2T xmit queue */
1996                 tcp_ctask->r2tqueue = kfifo_alloc(
1997                       session->max_r2t * 4 * sizeof(void*), GFP_KERNEL, NULL);
1998                 if (tcp_ctask->r2tqueue == ERR_PTR(-ENOMEM)) {
1999                         iscsi_pool_free(&tcp_ctask->r2tpool,
2000                                         (void**)tcp_ctask->r2ts);
2001                         goto r2t_alloc_fail;
2002                 }
2003         }
2004
2005         return 0;
2006
2007 r2t_alloc_fail:
2008         for (i = 0; i < cmd_i; i++) {
2009                 struct iscsi_cmd_task *ctask = session->cmds[i];
2010                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2011
2012                 kfifo_free(tcp_ctask->r2tqueue);
2013                 iscsi_pool_free(&tcp_ctask->r2tpool,
2014                                 (void**)tcp_ctask->r2ts);
2015         }
2016         return -ENOMEM;
2017 }
2018
2019 static void
2020 iscsi_r2tpool_free(struct iscsi_session *session)
2021 {
2022         int i;
2023
2024         for (i = 0; i < session->cmds_max; i++) {
2025                 struct iscsi_cmd_task *ctask = session->cmds[i];
2026                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2027
2028                 kfifo_free(tcp_ctask->r2tqueue);
2029                 iscsi_pool_free(&tcp_ctask->r2tpool,
2030                                 (void**)tcp_ctask->r2ts);
2031         }
2032 }
2033
2034 static int
2035 iscsi_conn_set_param(struct iscsi_cls_conn *cls_conn, enum iscsi_param param,
2036                      char *buf, int buflen)
2037 {
2038         struct iscsi_conn *conn = cls_conn->dd_data;
2039         struct iscsi_session *session = conn->session;
2040         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2041         int value;
2042
2043         switch(param) {
2044         case ISCSI_PARAM_HDRDGST_EN:
2045                 iscsi_set_param(cls_conn, param, buf, buflen);
2046                 tcp_conn->hdr_size = sizeof(struct iscsi_hdr);
2047                 if (conn->hdrdgst_en)
2048                         tcp_conn->hdr_size += sizeof(__u32);
2049                 break;
2050         case ISCSI_PARAM_DATADGST_EN:
2051                 iscsi_set_param(cls_conn, param, buf, buflen);
2052                 tcp_conn->sendpage = conn->datadgst_en ?
2053                         sock_no_sendpage : tcp_conn->sock->ops->sendpage;
2054                 break;
2055         case ISCSI_PARAM_MAX_R2T:
2056                 sscanf(buf, "%d", &value);
2057                 if (session->max_r2t == roundup_pow_of_two(value))
2058                         break;
2059                 iscsi_r2tpool_free(session);
2060                 iscsi_set_param(cls_conn, param, buf, buflen);
2061                 if (session->max_r2t & (session->max_r2t - 1))
2062                         session->max_r2t = roundup_pow_of_two(session->max_r2t);
2063                 if (iscsi_r2tpool_alloc(session))
2064                         return -ENOMEM;
2065                 break;
2066         default:
2067                 return iscsi_set_param(cls_conn, param, buf, buflen);
2068         }
2069
2070         return 0;
2071 }
2072
2073 static int
2074 iscsi_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
2075                          enum iscsi_param param, char *buf)
2076 {
2077         struct iscsi_conn *conn = cls_conn->dd_data;
2078         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2079         struct inet_sock *inet;
2080         struct ipv6_pinfo *np;
2081         struct sock *sk;
2082         int len;
2083
2084         switch(param) {
2085         case ISCSI_PARAM_CONN_PORT:
2086                 if (!tcp_conn->sock)
2087                         return -EINVAL;
2088
2089                 inet = inet_sk(tcp_conn->sock->sk);
2090                 len = sprintf(buf, "%hu\n", be16_to_cpu(inet->dport));
2091                 break;
2092         case ISCSI_PARAM_CONN_ADDRESS:
2093                 if (!tcp_conn->sock)
2094                         return -EINVAL;
2095
2096                 sk = tcp_conn->sock->sk;
2097                 if (sk->sk_family == PF_INET) {
2098                         inet = inet_sk(sk);
2099                         len = sprintf(buf, NIPQUAD_FMT "\n",
2100                                       NIPQUAD(inet->daddr));
2101                 } else {
2102                         np = inet6_sk(sk);
2103                         len = sprintf(buf, NIP6_FMT "\n", NIP6(np->daddr));
2104                 }
2105                 break;
2106         default:
2107                 return iscsi_conn_get_param(cls_conn, param, buf);
2108         }
2109
2110         return len;
2111 }
2112
2113 static void
2114 iscsi_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
2115 {
2116         struct iscsi_conn *conn = cls_conn->dd_data;
2117         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2118
2119         stats->txdata_octets = conn->txdata_octets;
2120         stats->rxdata_octets = conn->rxdata_octets;
2121         stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
2122         stats->dataout_pdus = conn->dataout_pdus_cnt;
2123         stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
2124         stats->datain_pdus = conn->datain_pdus_cnt;
2125         stats->r2t_pdus = conn->r2t_pdus_cnt;
2126         stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
2127         stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
2128         stats->custom_length = 3;
2129         strcpy(stats->custom[0].desc, "tx_sendpage_failures");
2130         stats->custom[0].value = tcp_conn->sendpage_failures_cnt;
2131         strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
2132         stats->custom[1].value = tcp_conn->discontiguous_hdr_cnt;
2133         strcpy(stats->custom[2].desc, "eh_abort_cnt");
2134         stats->custom[2].value = conn->eh_abort_cnt;
2135 }
2136
2137 static struct iscsi_cls_session *
2138 iscsi_tcp_session_create(struct iscsi_transport *iscsit,
2139                          struct scsi_transport_template *scsit,
2140                          uint16_t cmds_max, uint16_t qdepth,
2141                          uint32_t initial_cmdsn, uint32_t *hostno)
2142 {
2143         struct iscsi_cls_session *cls_session;
2144         struct iscsi_session *session;
2145         uint32_t hn;
2146         int cmd_i;
2147
2148         cls_session = iscsi_session_setup(iscsit, scsit, cmds_max, qdepth,
2149                                          sizeof(struct iscsi_tcp_cmd_task),
2150                                          sizeof(struct iscsi_tcp_mgmt_task),
2151                                          initial_cmdsn, &hn);
2152         if (!cls_session)
2153                 return NULL;
2154         *hostno = hn;
2155
2156         session = class_to_transport_session(cls_session);
2157         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
2158                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
2159                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2160
2161                 ctask->hdr = &tcp_ctask->hdr;
2162         }
2163
2164         for (cmd_i = 0; cmd_i < session->mgmtpool_max; cmd_i++) {
2165                 struct iscsi_mgmt_task *mtask = session->mgmt_cmds[cmd_i];
2166                 struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
2167
2168                 mtask->hdr = &tcp_mtask->hdr;
2169         }
2170
2171         if (iscsi_r2tpool_alloc(class_to_transport_session(cls_session)))
2172                 goto r2tpool_alloc_fail;
2173
2174         return cls_session;
2175
2176 r2tpool_alloc_fail:
2177         iscsi_session_teardown(cls_session);
2178         return NULL;
2179 }
2180
2181 static void iscsi_tcp_session_destroy(struct iscsi_cls_session *cls_session)
2182 {
2183         iscsi_r2tpool_free(class_to_transport_session(cls_session));
2184         iscsi_session_teardown(cls_session);
2185 }
2186
2187 static struct scsi_host_template iscsi_sht = {
2188         .name                   = "iSCSI Initiator over TCP/IP",
2189         .queuecommand           = iscsi_queuecommand,
2190         .change_queue_depth     = iscsi_change_queue_depth,
2191         .can_queue              = ISCSI_DEF_XMIT_CMDS_MAX - 1,
2192         .sg_tablesize           = ISCSI_SG_TABLESIZE,
2193         .max_sectors            = 0xFFFF,
2194         .cmd_per_lun            = ISCSI_DEF_CMD_PER_LUN,
2195         .eh_abort_handler       = iscsi_eh_abort,
2196         .eh_host_reset_handler  = iscsi_eh_host_reset,
2197         .use_clustering         = DISABLE_CLUSTERING,
2198         .proc_name              = "iscsi_tcp",
2199         .this_id                = -1,
2200 };
2201
2202 static struct iscsi_transport iscsi_tcp_transport = {
2203         .owner                  = THIS_MODULE,
2204         .name                   = "tcp",
2205         .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
2206                                   | CAP_DATADGST,
2207         .param_mask             = ISCSI_MAX_RECV_DLENGTH |
2208                                   ISCSI_MAX_XMIT_DLENGTH |
2209                                   ISCSI_HDRDGST_EN |
2210                                   ISCSI_DATADGST_EN |
2211                                   ISCSI_INITIAL_R2T_EN |
2212                                   ISCSI_MAX_R2T |
2213                                   ISCSI_IMM_DATA_EN |
2214                                   ISCSI_FIRST_BURST |
2215                                   ISCSI_MAX_BURST |
2216                                   ISCSI_PDU_INORDER_EN |
2217                                   ISCSI_DATASEQ_INORDER_EN |
2218                                   ISCSI_ERL |
2219                                   ISCSI_CONN_PORT |
2220                                   ISCSI_CONN_ADDRESS |
2221                                   ISCSI_EXP_STATSN |
2222                                   ISCSI_PERSISTENT_PORT |
2223                                   ISCSI_PERSISTENT_ADDRESS |
2224                                   ISCSI_TARGET_NAME | ISCSI_TPGT |
2225                                   ISCSI_USERNAME | ISCSI_PASSWORD |
2226                                   ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN,
2227         .host_param_mask        = ISCSI_HOST_HWADDRESS |
2228                                   ISCSI_HOST_INITIATOR_NAME,
2229         .host_template          = &iscsi_sht,
2230         .conndata_size          = sizeof(struct iscsi_conn),
2231         .max_conn               = 1,
2232         .max_cmd_len            = ISCSI_TCP_MAX_CMD_LEN,
2233         /* session management */
2234         .create_session         = iscsi_tcp_session_create,
2235         .destroy_session        = iscsi_tcp_session_destroy,
2236         /* connection management */
2237         .create_conn            = iscsi_tcp_conn_create,
2238         .bind_conn              = iscsi_tcp_conn_bind,
2239         .destroy_conn           = iscsi_tcp_conn_destroy,
2240         .set_param              = iscsi_conn_set_param,
2241         .get_conn_param         = iscsi_tcp_conn_get_param,
2242         .get_session_param      = iscsi_session_get_param,
2243         .start_conn             = iscsi_conn_start,
2244         .stop_conn              = iscsi_tcp_conn_stop,
2245         /* iscsi host params */
2246         .get_host_param         = iscsi_host_get_param,
2247         .set_host_param         = iscsi_host_set_param,
2248         /* IO */
2249         .send_pdu               = iscsi_conn_send_pdu,
2250         .get_stats              = iscsi_conn_get_stats,
2251         .init_cmd_task          = iscsi_tcp_cmd_init,
2252         .init_mgmt_task         = iscsi_tcp_mgmt_init,
2253         .xmit_cmd_task          = iscsi_tcp_ctask_xmit,
2254         .xmit_mgmt_task         = iscsi_tcp_mtask_xmit,
2255         .cleanup_cmd_task       = iscsi_tcp_cleanup_ctask,
2256         /* recovery */
2257         .session_recovery_timedout = iscsi_session_recovery_timedout,
2258 };
2259
2260 static int __init
2261 iscsi_tcp_init(void)
2262 {
2263         if (iscsi_max_lun < 1) {
2264                 printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
2265                        iscsi_max_lun);
2266                 return -EINVAL;
2267         }
2268         iscsi_tcp_transport.max_lun = iscsi_max_lun;
2269
2270         if (!iscsi_register_transport(&iscsi_tcp_transport))
2271                 return -ENODEV;
2272
2273         return 0;
2274 }
2275
2276 static void __exit
2277 iscsi_tcp_exit(void)
2278 {
2279         iscsi_unregister_transport(&iscsi_tcp_transport);
2280 }
2281
2282 module_init(iscsi_tcp_init);
2283 module_exit(iscsi_tcp_exit);