[SCSI] libiscsi: do not block session during logout
[linux-2.6.git] / drivers / scsi / iscsi_tcp.c
1 /*
2  * iSCSI Initiator over TCP/IP Data-Path
3  *
4  * Copyright (C) 2004 Dmitry Yusupov
5  * Copyright (C) 2004 Alex Aizman
6  * Copyright (C) 2005 - 2006 Mike Christie
7  * Copyright (C) 2006 Red Hat, Inc.  All rights reserved.
8  * maintained by open-iscsi@googlegroups.com
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published
12  * by the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful, but
16  * WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18  * General Public License for more details.
19  *
20  * See the file COPYING included with this distribution for more details.
21  *
22  * Credits:
23  *      Christoph Hellwig
24  *      FUJITA Tomonori
25  *      Arne Redlich
26  *      Zhenyu Wang
27  */
28
29 #include <linux/types.h>
30 #include <linux/list.h>
31 #include <linux/inet.h>
32 #include <linux/file.h>
33 #include <linux/blkdev.h>
34 #include <linux/crypto.h>
35 #include <linux/delay.h>
36 #include <linux/kfifo.h>
37 #include <linux/scatterlist.h>
38 #include <net/tcp.h>
39 #include <scsi/scsi_cmnd.h>
40 #include <scsi/scsi_device.h>
41 #include <scsi/scsi_host.h>
42 #include <scsi/scsi.h>
43 #include <scsi/scsi_transport_iscsi.h>
44
45 #include "iscsi_tcp.h"
46
47 MODULE_AUTHOR("Dmitry Yusupov <dmitry_yus@yahoo.com>, "
48               "Alex Aizman <itn780@yahoo.com>");
49 MODULE_DESCRIPTION("iSCSI/TCP data-path");
50 MODULE_LICENSE("GPL");
51 #undef DEBUG_TCP
52 #define DEBUG_ASSERT
53
54 #ifdef DEBUG_TCP
55 #define debug_tcp(fmt...) printk(KERN_INFO "tcp: " fmt)
56 #else
57 #define debug_tcp(fmt...)
58 #endif
59
60 #ifndef DEBUG_ASSERT
61 #ifdef BUG_ON
62 #undef BUG_ON
63 #endif
64 #define BUG_ON(expr)
65 #endif
66
67 static unsigned int iscsi_max_lun = 512;
68 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
69
70 static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
71                                    struct iscsi_chunk *chunk);
72
73 static inline void
74 iscsi_buf_init_iov(struct iscsi_buf *ibuf, char *vbuf, int size)
75 {
76         ibuf->sg.page = virt_to_page(vbuf);
77         ibuf->sg.offset = offset_in_page(vbuf);
78         ibuf->sg.length = size;
79         ibuf->sent = 0;
80         ibuf->use_sendmsg = 1;
81 }
82
83 static inline void
84 iscsi_buf_init_sg(struct iscsi_buf *ibuf, struct scatterlist *sg)
85 {
86         ibuf->sg.page = sg->page;
87         ibuf->sg.offset = sg->offset;
88         ibuf->sg.length = sg->length;
89         /*
90          * Fastpath: sg element fits into single page
91          */
92         if (sg->length + sg->offset <= PAGE_SIZE && !PageSlab(sg->page))
93                 ibuf->use_sendmsg = 0;
94         else
95                 ibuf->use_sendmsg = 1;
96         ibuf->sent = 0;
97 }
98
99 static inline int
100 iscsi_buf_left(struct iscsi_buf *ibuf)
101 {
102         int rc;
103
104         rc = ibuf->sg.length - ibuf->sent;
105         BUG_ON(rc < 0);
106         return rc;
107 }
108
109 static inline void
110 iscsi_hdr_digest(struct iscsi_conn *conn, struct iscsi_buf *buf,
111                  u8* crc)
112 {
113         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
114
115         crypto_hash_digest(&tcp_conn->tx_hash, &buf->sg, buf->sg.length, crc);
116         buf->sg.length += ISCSI_DIGEST_SIZE;
117 }
118
119 /*
120  * Scatterlist handling: inside the iscsi_chunk, we
121  * remember an index into the scatterlist, and set data/size
122  * to the current scatterlist entry. For highmem pages, we
123  * kmap as needed.
124  *
125  * Note that the page is unmapped when we return from
126  * TCP's data_ready handler, so we may end up mapping and
127  * unmapping the same page repeatedly. The whole reason
128  * for this is that we shouldn't keep the page mapped
129  * outside the softirq.
130  */
131
132 /**
133  * iscsi_tcp_chunk_init_sg - init indicated scatterlist entry
134  * @chunk: the buffer object
135  * @idx: index into scatterlist
136  * @offset: byte offset into that sg entry
137  *
138  * This function sets up the chunk so that subsequent
139  * data is copied to the indicated sg entry, at the given
140  * offset.
141  */
142 static inline void
143 iscsi_tcp_chunk_init_sg(struct iscsi_chunk *chunk,
144                         unsigned int idx, unsigned int offset)
145 {
146         struct scatterlist *sg;
147
148         BUG_ON(chunk->sg == NULL);
149
150         sg = &chunk->sg[idx];
151         chunk->sg_index = idx;
152         chunk->sg_offset = offset;
153         chunk->size = min(sg->length - offset, chunk->total_size);
154         chunk->data = NULL;
155 }
156
157 /**
158  * iscsi_tcp_chunk_map - map the current S/G page
159  * @chunk: iscsi chunk
160  *
161  * We only need to possibly kmap data if scatter lists are being used,
162  * because the iscsi passthrough and internal IO paths will never use high
163  * mem pages.
164  */
165 static inline void
166 iscsi_tcp_chunk_map(struct iscsi_chunk *chunk)
167 {
168         struct scatterlist *sg;
169
170         if (chunk->data != NULL || !chunk->sg)
171                 return;
172
173         sg = &chunk->sg[chunk->sg_index];
174         BUG_ON(chunk->sg_mapped);
175         BUG_ON(sg->length == 0);
176         chunk->sg_mapped = kmap_atomic(sg->page, KM_SOFTIRQ0);
177         chunk->data = chunk->sg_mapped + sg->offset + chunk->sg_offset;
178 }
179
180 static inline void
181 iscsi_tcp_chunk_unmap(struct iscsi_chunk *chunk)
182 {
183         if (chunk->sg_mapped) {
184                 kunmap_atomic(chunk->sg_mapped, KM_SOFTIRQ0);
185                 chunk->sg_mapped = NULL;
186                 chunk->data = NULL;
187         }
188 }
189
190 /*
191  * Splice the digest buffer into the buffer
192  */
193 static inline void
194 iscsi_tcp_chunk_splice_digest(struct iscsi_chunk *chunk, void *digest)
195 {
196         chunk->data = digest;
197         chunk->digest_len = ISCSI_DIGEST_SIZE;
198         chunk->total_size += ISCSI_DIGEST_SIZE;
199         chunk->size = ISCSI_DIGEST_SIZE;
200         chunk->copied = 0;
201         chunk->sg = NULL;
202         chunk->sg_index = 0;
203         chunk->hash = NULL;
204 }
205
206 /**
207  * iscsi_tcp_chunk_done - check whether the chunk is complete
208  * @chunk: iscsi chunk to check
209  *
210  * Check if we're done receiving this chunk. If the receive
211  * buffer is full but we expect more data, move on to the
212  * next entry in the scatterlist.
213  *
214  * If the amount of data we received isn't a multiple of 4,
215  * we will transparently receive the pad bytes, too.
216  *
217  * This function must be re-entrant.
218  */
219 static inline int
220 iscsi_tcp_chunk_done(struct iscsi_chunk *chunk)
221 {
222         static unsigned char padbuf[ISCSI_PAD_LEN];
223         unsigned int pad;
224
225         if (chunk->copied < chunk->size) {
226                 iscsi_tcp_chunk_map(chunk);
227                 return 0;
228         }
229
230         chunk->total_copied += chunk->copied;
231         chunk->copied = 0;
232         chunk->size = 0;
233
234         /* Unmap the current scatterlist page, if there is one. */
235         iscsi_tcp_chunk_unmap(chunk);
236
237         /* Do we have more scatterlist entries? */
238         if (chunk->total_copied < chunk->total_size) {
239                 /* Proceed to the next entry in the scatterlist. */
240                 iscsi_tcp_chunk_init_sg(chunk, chunk->sg_index + 1, 0);
241                 iscsi_tcp_chunk_map(chunk);
242                 BUG_ON(chunk->size == 0);
243                 return 0;
244         }
245
246         /* Do we need to handle padding? */
247         pad = iscsi_padding(chunk->total_copied);
248         if (pad != 0) {
249                 debug_tcp("consume %d pad bytes\n", pad);
250                 chunk->total_size += pad;
251                 chunk->size = pad;
252                 chunk->data = padbuf;
253                 return 0;
254         }
255
256         /*
257          * Set us up for receiving the data digest. hdr digest
258          * is completely handled in hdr done function.
259          */
260         if (chunk->hash) {
261                 if (chunk->digest_len == 0) {
262                         crypto_hash_final(chunk->hash, chunk->digest);
263                         iscsi_tcp_chunk_splice_digest(chunk,
264                                                       chunk->recv_digest);
265                         return 0;
266                 }
267         }
268
269         return 1;
270 }
271
272 /**
273  * iscsi_tcp_chunk_recv - copy data to chunk
274  * @tcp_conn: the iSCSI TCP connection
275  * @chunk: the buffer to copy to
276  * @ptr: data pointer
277  * @len: amount of data available
278  *
279  * This function copies up to @len bytes to the
280  * given buffer, and returns the number of bytes
281  * consumed, which can actually be less than @len.
282  *
283  * If hash digest is enabled, the function will update the
284  * hash while copying.
285  * Combining these two operations doesn't buy us a lot (yet),
286  * but in the future we could implement combined copy+crc,
287  * just way we do for network layer checksums.
288  */
289 static int
290 iscsi_tcp_chunk_recv(struct iscsi_tcp_conn *tcp_conn,
291                      struct iscsi_chunk *chunk, const void *ptr,
292                      unsigned int len)
293 {
294         struct scatterlist sg;
295         unsigned int copy, copied = 0;
296
297         while (!iscsi_tcp_chunk_done(chunk)) {
298                 if (copied == len)
299                         goto out;
300
301                 copy = min(len - copied, chunk->size - chunk->copied);
302                 memcpy(chunk->data + chunk->copied, ptr + copied, copy);
303
304                 if (chunk->hash) {
305                         sg_init_one(&sg, ptr + copied, copy);
306                         crypto_hash_update(chunk->hash, &sg, copy);
307                 }
308                 chunk->copied += copy;
309                 copied += copy;
310         }
311
312 out:
313         return copied;
314 }
315
316 static inline void
317 iscsi_tcp_dgst_header(struct hash_desc *hash, const void *hdr, size_t hdrlen,
318                       unsigned char digest[ISCSI_DIGEST_SIZE])
319 {
320         struct scatterlist sg;
321
322         sg_init_one(&sg, hdr, hdrlen);
323         crypto_hash_digest(hash, &sg, hdrlen, digest);
324 }
325
326 static inline int
327 iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
328                       struct iscsi_chunk *chunk)
329 {
330         if (!chunk->digest_len)
331                 return 1;
332
333         if (memcmp(chunk->recv_digest, chunk->digest, chunk->digest_len)) {
334                 debug_scsi("digest mismatch\n");
335                 return 0;
336         }
337
338         return 1;
339 }
340
341 /*
342  * Helper function to set up chunk buffer
343  */
344 static inline void
345 __iscsi_chunk_init(struct iscsi_chunk *chunk, size_t size,
346                    iscsi_chunk_done_fn_t *done, struct hash_desc *hash)
347 {
348         memset(chunk, 0, sizeof(*chunk));
349         chunk->total_size = size;
350         chunk->done = done;
351
352         if (hash) {
353                 chunk->hash = hash;
354                 crypto_hash_init(hash);
355         }
356 }
357
358 static inline void
359 iscsi_chunk_init_linear(struct iscsi_chunk *chunk, void *data, size_t size,
360                         iscsi_chunk_done_fn_t *done, struct hash_desc *hash)
361 {
362         __iscsi_chunk_init(chunk, size, done, hash);
363         chunk->data = data;
364         chunk->size = size;
365 }
366
367 static inline int
368 iscsi_chunk_seek_sg(struct iscsi_chunk *chunk,
369                     struct scatterlist *sg, unsigned int sg_count,
370                     unsigned int offset, size_t size,
371                     iscsi_chunk_done_fn_t *done, struct hash_desc *hash)
372 {
373         unsigned int i;
374
375         __iscsi_chunk_init(chunk, size, done, hash);
376         for (i = 0; i < sg_count; ++i) {
377                 if (offset < sg[i].length) {
378                         chunk->sg = sg;
379                         chunk->sg_count = sg_count;
380                         iscsi_tcp_chunk_init_sg(chunk, i, offset);
381                         return 0;
382                 }
383                 offset -= sg[i].length;
384         }
385
386         return ISCSI_ERR_DATA_OFFSET;
387 }
388
389 /**
390  * iscsi_tcp_hdr_recv_prep - prep chunk for hdr reception
391  * @tcp_conn: iscsi connection to prep for
392  *
393  * This function always passes NULL for the hash argument, because when this
394  * function is called we do not yet know the final size of the header and want
395  * to delay the digest processing until we know that.
396  */
397 static void
398 iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
399 {
400         debug_tcp("iscsi_tcp_hdr_recv_prep(%p%s)\n", tcp_conn,
401                   tcp_conn->iscsi_conn->hdrdgst_en ? ", digest enabled" : "");
402         iscsi_chunk_init_linear(&tcp_conn->in.chunk,
403                                 tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
404                                 iscsi_tcp_hdr_recv_done, NULL);
405 }
406
407 /*
408  * Handle incoming reply to any other type of command
409  */
410 static int
411 iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
412                          struct iscsi_chunk *chunk)
413 {
414         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
415         int rc = 0;
416
417         if (!iscsi_tcp_dgst_verify(tcp_conn, chunk))
418                 return ISCSI_ERR_DATA_DGST;
419
420         rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr,
421                         conn->data, tcp_conn->in.datalen);
422         if (rc)
423                 return rc;
424
425         iscsi_tcp_hdr_recv_prep(tcp_conn);
426         return 0;
427 }
428
429 static void
430 iscsi_tcp_data_recv_prep(struct iscsi_tcp_conn *tcp_conn)
431 {
432         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
433         struct hash_desc *rx_hash = NULL;
434
435         if (conn->datadgst_en)
436                 rx_hash = &tcp_conn->rx_hash;
437
438         iscsi_chunk_init_linear(&tcp_conn->in.chunk,
439                                 conn->data, tcp_conn->in.datalen,
440                                 iscsi_tcp_data_recv_done, rx_hash);
441 }
442
443 /*
444  * must be called with session lock
445  */
446 static void
447 iscsi_tcp_cleanup_ctask(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
448 {
449         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
450         struct iscsi_r2t_info *r2t;
451         struct scsi_cmnd *sc;
452
453         /* flush ctask's r2t queues */
454         while (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*))) {
455                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
456                             sizeof(void*));
457                 debug_scsi("iscsi_tcp_cleanup_ctask pending r2t dropped\n");
458         }
459
460         sc = ctask->sc;
461         if (unlikely(!sc))
462                 return;
463
464         tcp_ctask->xmstate = XMSTATE_IDLE;
465         tcp_ctask->r2t = NULL;
466 }
467
468 /**
469  * iscsi_data_rsp - SCSI Data-In Response processing
470  * @conn: iscsi connection
471  * @ctask: scsi command task
472  **/
473 static int
474 iscsi_data_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
475 {
476         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
477         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
478         struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
479         struct iscsi_session *session = conn->session;
480         struct scsi_cmnd *sc = ctask->sc;
481         int datasn = be32_to_cpu(rhdr->datasn);
482
483         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
484         /*
485          * setup Data-In byte counter (gets decremented..)
486          */
487         ctask->data_count = tcp_conn->in.datalen;
488
489         if (tcp_conn->in.datalen == 0)
490                 return 0;
491
492         if (tcp_ctask->exp_datasn != datasn) {
493                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->datasn(%d)\n",
494                           __FUNCTION__, tcp_ctask->exp_datasn, datasn);
495                 return ISCSI_ERR_DATASN;
496         }
497
498         tcp_ctask->exp_datasn++;
499
500         tcp_ctask->data_offset = be32_to_cpu(rhdr->offset);
501         if (tcp_ctask->data_offset + tcp_conn->in.datalen > scsi_bufflen(sc)) {
502                 debug_tcp("%s: data_offset(%d) + data_len(%d) > total_length_in(%d)\n",
503                           __FUNCTION__, tcp_ctask->data_offset,
504                           tcp_conn->in.datalen, scsi_bufflen(sc));
505                 return ISCSI_ERR_DATA_OFFSET;
506         }
507
508         if (rhdr->flags & ISCSI_FLAG_DATA_STATUS) {
509                 sc->result = (DID_OK << 16) | rhdr->cmd_status;
510                 conn->exp_statsn = be32_to_cpu(rhdr->statsn) + 1;
511                 if (rhdr->flags & (ISCSI_FLAG_DATA_UNDERFLOW |
512                                    ISCSI_FLAG_DATA_OVERFLOW)) {
513                         int res_count = be32_to_cpu(rhdr->residual_count);
514
515                         if (res_count > 0 &&
516                             (rhdr->flags & ISCSI_FLAG_CMD_OVERFLOW ||
517                              res_count <= scsi_bufflen(sc)))
518                                 scsi_set_resid(sc, res_count);
519                         else
520                                 sc->result = (DID_BAD_TARGET << 16) |
521                                         rhdr->cmd_status;
522                 }
523         }
524
525         conn->datain_pdus_cnt++;
526         return 0;
527 }
528
529 /**
530  * iscsi_solicit_data_init - initialize first Data-Out
531  * @conn: iscsi connection
532  * @ctask: scsi command task
533  * @r2t: R2T info
534  *
535  * Notes:
536  *      Initialize first Data-Out within this R2T sequence and finds
537  *      proper data_offset within this SCSI command.
538  *
539  *      This function is called with connection lock taken.
540  **/
541 static void
542 iscsi_solicit_data_init(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
543                         struct iscsi_r2t_info *r2t)
544 {
545         struct iscsi_data *hdr;
546         struct scsi_cmnd *sc = ctask->sc;
547         int i, sg_count = 0;
548         struct scatterlist *sg;
549
550         hdr = &r2t->dtask.hdr;
551         memset(hdr, 0, sizeof(struct iscsi_data));
552         hdr->ttt = r2t->ttt;
553         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
554         r2t->solicit_datasn++;
555         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
556         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
557         hdr->itt = ctask->hdr->itt;
558         hdr->exp_statsn = r2t->exp_statsn;
559         hdr->offset = cpu_to_be32(r2t->data_offset);
560         if (r2t->data_length > conn->max_xmit_dlength) {
561                 hton24(hdr->dlength, conn->max_xmit_dlength);
562                 r2t->data_count = conn->max_xmit_dlength;
563                 hdr->flags = 0;
564         } else {
565                 hton24(hdr->dlength, r2t->data_length);
566                 r2t->data_count = r2t->data_length;
567                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
568         }
569         conn->dataout_pdus_cnt++;
570
571         r2t->sent = 0;
572
573         iscsi_buf_init_iov(&r2t->headbuf, (char*)hdr,
574                            sizeof(struct iscsi_hdr));
575
576         sg = scsi_sglist(sc);
577         r2t->sg = NULL;
578         for (i = 0; i < scsi_sg_count(sc); i++, sg += 1) {
579                 /* FIXME: prefetch ? */
580                 if (sg_count + sg->length > r2t->data_offset) {
581                         int page_offset;
582
583                         /* sg page found! */
584
585                         /* offset within this page */
586                         page_offset = r2t->data_offset - sg_count;
587
588                         /* fill in this buffer */
589                         iscsi_buf_init_sg(&r2t->sendbuf, sg);
590                         r2t->sendbuf.sg.offset += page_offset;
591                         r2t->sendbuf.sg.length -= page_offset;
592
593                         /* xmit logic will continue with next one */
594                         r2t->sg = sg + 1;
595                         break;
596                 }
597                 sg_count += sg->length;
598         }
599         BUG_ON(r2t->sg == NULL);
600 }
601
602 /**
603  * iscsi_r2t_rsp - iSCSI R2T Response processing
604  * @conn: iscsi connection
605  * @ctask: scsi command task
606  **/
607 static int
608 iscsi_r2t_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
609 {
610         struct iscsi_r2t_info *r2t;
611         struct iscsi_session *session = conn->session;
612         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
613         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
614         struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
615         int r2tsn = be32_to_cpu(rhdr->r2tsn);
616         int rc;
617
618         if (tcp_conn->in.datalen) {
619                 printk(KERN_ERR "iscsi_tcp: invalid R2t with datalen %d\n",
620                        tcp_conn->in.datalen);
621                 return ISCSI_ERR_DATALEN;
622         }
623
624         if (tcp_ctask->exp_datasn != r2tsn){
625                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
626                           __FUNCTION__, tcp_ctask->exp_datasn, r2tsn);
627                 return ISCSI_ERR_R2TSN;
628         }
629
630         /* fill-in new R2T associated with the task */
631         spin_lock(&session->lock);
632         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
633
634         if (!ctask->sc || session->state != ISCSI_STATE_LOGGED_IN) {
635                 printk(KERN_INFO "iscsi_tcp: dropping R2T itt %d in "
636                        "recovery...\n", ctask->itt);
637                 spin_unlock(&session->lock);
638                 return 0;
639         }
640
641         rc = __kfifo_get(tcp_ctask->r2tpool.queue, (void*)&r2t, sizeof(void*));
642         BUG_ON(!rc);
643
644         r2t->exp_statsn = rhdr->statsn;
645         r2t->data_length = be32_to_cpu(rhdr->data_length);
646         if (r2t->data_length == 0) {
647                 printk(KERN_ERR "iscsi_tcp: invalid R2T with zero data len\n");
648                 spin_unlock(&session->lock);
649                 return ISCSI_ERR_DATALEN;
650         }
651
652         if (r2t->data_length > session->max_burst)
653                 debug_scsi("invalid R2T with data len %u and max burst %u."
654                            "Attempting to execute request.\n",
655                             r2t->data_length, session->max_burst);
656
657         r2t->data_offset = be32_to_cpu(rhdr->data_offset);
658         if (r2t->data_offset + r2t->data_length > scsi_bufflen(ctask->sc)) {
659                 spin_unlock(&session->lock);
660                 printk(KERN_ERR "iscsi_tcp: invalid R2T with data len %u at "
661                        "offset %u and total length %d\n", r2t->data_length,
662                        r2t->data_offset, scsi_bufflen(ctask->sc));
663                 return ISCSI_ERR_DATALEN;
664         }
665
666         r2t->ttt = rhdr->ttt; /* no flip */
667         r2t->solicit_datasn = 0;
668
669         iscsi_solicit_data_init(conn, ctask, r2t);
670
671         tcp_ctask->exp_datasn = r2tsn + 1;
672         __kfifo_put(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*));
673         tcp_ctask->xmstate |= XMSTATE_SOL_HDR_INIT;
674         conn->r2t_pdus_cnt++;
675
676         iscsi_requeue_ctask(ctask);
677         spin_unlock(&session->lock);
678
679         return 0;
680 }
681
682 /*
683  * Handle incoming reply to DataIn command
684  */
685 static int
686 iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
687                           struct iscsi_chunk *chunk)
688 {
689         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
690         struct iscsi_hdr *hdr = tcp_conn->in.hdr;
691         int rc;
692
693         if (!iscsi_tcp_dgst_verify(tcp_conn, chunk))
694                 return ISCSI_ERR_DATA_DGST;
695
696         /* check for non-exceptional status */
697         if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
698                 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
699                 if (rc)
700                         return rc;
701         }
702
703         iscsi_tcp_hdr_recv_prep(tcp_conn);
704         return 0;
705 }
706
707 /**
708  * iscsi_tcp_hdr_dissect - process PDU header
709  * @conn: iSCSI connection
710  * @hdr: PDU header
711  *
712  * This function analyzes the header of the PDU received,
713  * and performs several sanity checks. If the PDU is accompanied
714  * by data, the receive buffer is set up to copy the incoming data
715  * to the correct location.
716  */
717 static int
718 iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
719 {
720         int rc = 0, opcode, ahslen;
721         struct iscsi_session *session = conn->session;
722         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
723         struct iscsi_cmd_task *ctask;
724         uint32_t itt;
725
726         /* verify PDU length */
727         tcp_conn->in.datalen = ntoh24(hdr->dlength);
728         if (tcp_conn->in.datalen > conn->max_recv_dlength) {
729                 printk(KERN_ERR "iscsi_tcp: datalen %d > %d\n",
730                        tcp_conn->in.datalen, conn->max_recv_dlength);
731                 return ISCSI_ERR_DATALEN;
732         }
733
734         /* Additional header segments. So far, we don't
735          * process additional headers.
736          */
737         ahslen = hdr->hlength << 2;
738
739         opcode = hdr->opcode & ISCSI_OPCODE_MASK;
740         /* verify itt (itt encoding: age+cid+itt) */
741         rc = iscsi_verify_itt(conn, hdr, &itt);
742         if (rc == ISCSI_ERR_NO_SCSI_CMD) {
743                 /* XXX: what does this do? */
744                 tcp_conn->in.datalen = 0; /* force drop */
745                 return 0;
746         } else if (rc)
747                 return rc;
748
749         debug_tcp("opcode 0x%x ahslen %d datalen %d\n",
750                   opcode, ahslen, tcp_conn->in.datalen);
751
752         switch(opcode) {
753         case ISCSI_OP_SCSI_DATA_IN:
754                 ctask = session->cmds[itt];
755                 rc = iscsi_data_rsp(conn, ctask);
756                 if (rc)
757                         return rc;
758                 if (tcp_conn->in.datalen) {
759                         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
760                         struct hash_desc *rx_hash = NULL;
761
762                         /*
763                          * Setup copy of Data-In into the Scsi_Cmnd
764                          * Scatterlist case:
765                          * We set up the iscsi_chunk to point to the next
766                          * scatterlist entry to copy to. As we go along,
767                          * we move on to the next scatterlist entry and
768                          * update the digest per-entry.
769                          */
770                         if (conn->datadgst_en)
771                                 rx_hash = &tcp_conn->rx_hash;
772
773                         debug_tcp("iscsi_tcp_begin_data_in(%p, offset=%d, "
774                                   "datalen=%d)\n", tcp_conn,
775                                   tcp_ctask->data_offset,
776                                   tcp_conn->in.datalen);
777                         return iscsi_chunk_seek_sg(&tcp_conn->in.chunk,
778                                                 scsi_sglist(ctask->sc),
779                                                 scsi_sg_count(ctask->sc),
780                                                 tcp_ctask->data_offset,
781                                                 tcp_conn->in.datalen,
782                                                 iscsi_tcp_process_data_in,
783                                                 rx_hash);
784                 }
785                 /* fall through */
786         case ISCSI_OP_SCSI_CMD_RSP:
787                 if (tcp_conn->in.datalen) {
788                         iscsi_tcp_data_recv_prep(tcp_conn);
789                         return 0;
790                 }
791                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
792                 break;
793         case ISCSI_OP_R2T:
794                 ctask = session->cmds[itt];
795                 if (ahslen)
796                         rc = ISCSI_ERR_AHSLEN;
797                 else if (ctask->sc->sc_data_direction == DMA_TO_DEVICE)
798                         rc = iscsi_r2t_rsp(conn, ctask);
799                 else
800                         rc = ISCSI_ERR_PROTO;
801                 break;
802         case ISCSI_OP_LOGIN_RSP:
803         case ISCSI_OP_TEXT_RSP:
804         case ISCSI_OP_REJECT:
805         case ISCSI_OP_ASYNC_EVENT:
806                 /*
807                  * It is possible that we could get a PDU with a buffer larger
808                  * than 8K, but there are no targets that currently do this.
809                  * For now we fail until we find a vendor that needs it
810                  */
811                 if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
812                         printk(KERN_ERR "iscsi_tcp: received buffer of len %u "
813                               "but conn buffer is only %u (opcode %0x)\n",
814                               tcp_conn->in.datalen,
815                               ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
816                         rc = ISCSI_ERR_PROTO;
817                         break;
818                 }
819
820                 /* If there's data coming in with the response,
821                  * receive it to the connection's buffer.
822                  */
823                 if (tcp_conn->in.datalen) {
824                         iscsi_tcp_data_recv_prep(tcp_conn);
825                         return 0;
826                 }
827         /* fall through */
828         case ISCSI_OP_LOGOUT_RSP:
829         case ISCSI_OP_NOOP_IN:
830         case ISCSI_OP_SCSI_TMFUNC_RSP:
831                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
832                 break;
833         default:
834                 rc = ISCSI_ERR_BAD_OPCODE;
835                 break;
836         }
837
838         if (rc == 0) {
839                 /* Anything that comes with data should have
840                  * been handled above. */
841                 if (tcp_conn->in.datalen)
842                         return ISCSI_ERR_PROTO;
843                 iscsi_tcp_hdr_recv_prep(tcp_conn);
844         }
845
846         return rc;
847 }
848
849 static inline void
850 partial_sg_digest_update(struct hash_desc *desc, struct scatterlist *sg,
851                          int offset, int length)
852 {
853         struct scatterlist temp;
854
855         sg_init_table(&temp, 1);
856         sg_set_page(&temp, sg_page(sg), length, offset);
857         crypto_hash_update(desc, &temp, length);
858 }
859
860 /**
861  * iscsi_tcp_hdr_recv_done - process PDU header
862  *
863  * This is the callback invoked when the PDU header has
864  * been received. If the header is followed by additional
865  * header segments, we go back for more data.
866  */
867 static int
868 iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
869                         struct iscsi_chunk *chunk)
870 {
871         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
872         struct iscsi_hdr *hdr;
873
874         /* Check if there are additional header segments
875          * *prior* to computing the digest, because we
876          * may need to go back to the caller for more.
877          */
878         hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
879         if (chunk->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
880                 /* Bump the header length - the caller will
881                  * just loop around and get the AHS for us, and
882                  * call again. */
883                 unsigned int ahslen = hdr->hlength << 2;
884
885                 /* Make sure we don't overflow */
886                 if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
887                         return ISCSI_ERR_AHSLEN;
888
889                 chunk->total_size += ahslen;
890                 chunk->size += ahslen;
891                 return 0;
892         }
893
894         /* We're done processing the header. See if we're doing
895          * header digests; if so, set up the recv_digest buffer
896          * and go back for more. */
897         if (conn->hdrdgst_en) {
898                 if (chunk->digest_len == 0) {
899                         iscsi_tcp_chunk_splice_digest(chunk,
900                                                       chunk->recv_digest);
901                         return 0;
902                 }
903                 iscsi_tcp_dgst_header(&tcp_conn->rx_hash, hdr,
904                                       chunk->total_copied - ISCSI_DIGEST_SIZE,
905                                       chunk->digest);
906
907                 if (!iscsi_tcp_dgst_verify(tcp_conn, chunk))
908                         return ISCSI_ERR_HDR_DGST;
909         }
910
911         tcp_conn->in.hdr = hdr;
912         return iscsi_tcp_hdr_dissect(conn, hdr);
913 }
914
915 /**
916  * iscsi_tcp_recv - TCP receive in sendfile fashion
917  * @rd_desc: read descriptor
918  * @skb: socket buffer
919  * @offset: offset in skb
920  * @len: skb->len - offset
921  **/
922 static int
923 iscsi_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
924                unsigned int offset, size_t len)
925 {
926         struct iscsi_conn *conn = rd_desc->arg.data;
927         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
928         struct iscsi_chunk *chunk = &tcp_conn->in.chunk;
929         struct skb_seq_state seq;
930         unsigned int consumed = 0;
931         int rc = 0;
932
933         debug_tcp("in %d bytes\n", skb->len - offset);
934
935         if (unlikely(conn->suspend_rx)) {
936                 debug_tcp("conn %d Rx suspended!\n", conn->id);
937                 return 0;
938         }
939
940         skb_prepare_seq_read(skb, offset, skb->len, &seq);
941         while (1) {
942                 unsigned int avail;
943                 const u8 *ptr;
944
945                 avail = skb_seq_read(consumed, &ptr, &seq);
946                 if (avail == 0)
947                         break;
948                 BUG_ON(chunk->copied >= chunk->size);
949
950                 debug_tcp("skb %p ptr=%p avail=%u\n", skb, ptr, avail);
951                 rc = iscsi_tcp_chunk_recv(tcp_conn, chunk, ptr, avail);
952                 BUG_ON(rc == 0);
953                 consumed += rc;
954
955                 if (chunk->total_copied >= chunk->total_size) {
956                         rc = chunk->done(tcp_conn, chunk);
957                         if (rc != 0) {
958                                 skb_abort_seq_read(&seq);
959                                 goto error;
960                         }
961
962                         /* The done() functions sets up the
963                          * next chunk. */
964                 }
965         }
966
967         conn->rxdata_octets += consumed;
968         return consumed;
969
970 error:
971         debug_tcp("Error receiving PDU, errno=%d\n", rc);
972         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
973         return 0;
974 }
975
976 static void
977 iscsi_tcp_data_ready(struct sock *sk, int flag)
978 {
979         struct iscsi_conn *conn = sk->sk_user_data;
980         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
981         read_descriptor_t rd_desc;
982
983         read_lock(&sk->sk_callback_lock);
984
985         /*
986          * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
987          * We set count to 1 because we want the network layer to
988          * hand us all the skbs that are available. iscsi_tcp_recv
989          * handled pdus that cross buffers or pdus that still need data.
990          */
991         rd_desc.arg.data = conn;
992         rd_desc.count = 1;
993         tcp_read_sock(sk, &rd_desc, iscsi_tcp_recv);
994
995         read_unlock(&sk->sk_callback_lock);
996
997         /* If we had to (atomically) map a highmem page,
998          * unmap it now. */
999         iscsi_tcp_chunk_unmap(&tcp_conn->in.chunk);
1000 }
1001
1002 static void
1003 iscsi_tcp_state_change(struct sock *sk)
1004 {
1005         struct iscsi_tcp_conn *tcp_conn;
1006         struct iscsi_conn *conn;
1007         struct iscsi_session *session;
1008         void (*old_state_change)(struct sock *);
1009
1010         read_lock(&sk->sk_callback_lock);
1011
1012         conn = (struct iscsi_conn*)sk->sk_user_data;
1013         session = conn->session;
1014
1015         if ((sk->sk_state == TCP_CLOSE_WAIT ||
1016              sk->sk_state == TCP_CLOSE) &&
1017             !atomic_read(&sk->sk_rmem_alloc)) {
1018                 debug_tcp("iscsi_tcp_state_change: TCP_CLOSE|TCP_CLOSE_WAIT\n");
1019                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1020         }
1021
1022         tcp_conn = conn->dd_data;
1023         old_state_change = tcp_conn->old_state_change;
1024
1025         read_unlock(&sk->sk_callback_lock);
1026
1027         old_state_change(sk);
1028 }
1029
1030 /**
1031  * iscsi_write_space - Called when more output buffer space is available
1032  * @sk: socket space is available for
1033  **/
1034 static void
1035 iscsi_write_space(struct sock *sk)
1036 {
1037         struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
1038         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1039
1040         tcp_conn->old_write_space(sk);
1041         debug_tcp("iscsi_write_space: cid %d\n", conn->id);
1042         scsi_queue_work(conn->session->host, &conn->xmitwork);
1043 }
1044
1045 static void
1046 iscsi_conn_set_callbacks(struct iscsi_conn *conn)
1047 {
1048         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1049         struct sock *sk = tcp_conn->sock->sk;
1050
1051         /* assign new callbacks */
1052         write_lock_bh(&sk->sk_callback_lock);
1053         sk->sk_user_data = conn;
1054         tcp_conn->old_data_ready = sk->sk_data_ready;
1055         tcp_conn->old_state_change = sk->sk_state_change;
1056         tcp_conn->old_write_space = sk->sk_write_space;
1057         sk->sk_data_ready = iscsi_tcp_data_ready;
1058         sk->sk_state_change = iscsi_tcp_state_change;
1059         sk->sk_write_space = iscsi_write_space;
1060         write_unlock_bh(&sk->sk_callback_lock);
1061 }
1062
1063 static void
1064 iscsi_conn_restore_callbacks(struct iscsi_tcp_conn *tcp_conn)
1065 {
1066         struct sock *sk = tcp_conn->sock->sk;
1067
1068         /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
1069         write_lock_bh(&sk->sk_callback_lock);
1070         sk->sk_user_data    = NULL;
1071         sk->sk_data_ready   = tcp_conn->old_data_ready;
1072         sk->sk_state_change = tcp_conn->old_state_change;
1073         sk->sk_write_space  = tcp_conn->old_write_space;
1074         sk->sk_no_check  = 0;
1075         write_unlock_bh(&sk->sk_callback_lock);
1076 }
1077
1078 /**
1079  * iscsi_send - generic send routine
1080  * @sk: kernel's socket
1081  * @buf: buffer to write from
1082  * @size: actual size to write
1083  * @flags: socket's flags
1084  */
1085 static inline int
1086 iscsi_send(struct iscsi_conn *conn, struct iscsi_buf *buf, int size, int flags)
1087 {
1088         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1089         struct socket *sk = tcp_conn->sock;
1090         int offset = buf->sg.offset + buf->sent, res;
1091
1092         /*
1093          * if we got use_sg=0 or are sending something we kmallocd
1094          * then we did not have to do kmap (kmap returns page_address)
1095          *
1096          * if we got use_sg > 0, but had to drop down, we do not
1097          * set clustering so this should only happen for that
1098          * slab case.
1099          */
1100         if (buf->use_sendmsg)
1101                 res = sock_no_sendpage(sk, buf->sg.page, offset, size, flags);
1102         else
1103                 res = tcp_conn->sendpage(sk, buf->sg.page, offset, size, flags);
1104
1105         if (res >= 0) {
1106                 conn->txdata_octets += res;
1107                 buf->sent += res;
1108                 return res;
1109         }
1110
1111         tcp_conn->sendpage_failures_cnt++;
1112         if (res == -EAGAIN)
1113                 res = -ENOBUFS;
1114         else
1115                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1116         return res;
1117 }
1118
1119 /**
1120  * iscsi_sendhdr - send PDU Header via tcp_sendpage()
1121  * @conn: iscsi connection
1122  * @buf: buffer to write from
1123  * @datalen: lenght of data to be sent after the header
1124  *
1125  * Notes:
1126  *      (Tx, Fast Path)
1127  **/
1128 static inline int
1129 iscsi_sendhdr(struct iscsi_conn *conn, struct iscsi_buf *buf, int datalen)
1130 {
1131         int flags = 0; /* MSG_DONTWAIT; */
1132         int res, size;
1133
1134         size = buf->sg.length - buf->sent;
1135         BUG_ON(buf->sent + size > buf->sg.length);
1136         if (buf->sent + size != buf->sg.length || datalen)
1137                 flags |= MSG_MORE;
1138
1139         res = iscsi_send(conn, buf, size, flags);
1140         debug_tcp("sendhdr %d bytes, sent %d res %d\n", size, buf->sent, res);
1141         if (res >= 0) {
1142                 if (size != res)
1143                         return -EAGAIN;
1144                 return 0;
1145         }
1146
1147         return res;
1148 }
1149
1150 /**
1151  * iscsi_sendpage - send one page of iSCSI Data-Out.
1152  * @conn: iscsi connection
1153  * @buf: buffer to write from
1154  * @count: remaining data
1155  * @sent: number of bytes sent
1156  *
1157  * Notes:
1158  *      (Tx, Fast Path)
1159  **/
1160 static inline int
1161 iscsi_sendpage(struct iscsi_conn *conn, struct iscsi_buf *buf,
1162                int *count, int *sent)
1163 {
1164         int flags = 0; /* MSG_DONTWAIT; */
1165         int res, size;
1166
1167         size = buf->sg.length - buf->sent;
1168         BUG_ON(buf->sent + size > buf->sg.length);
1169         if (size > *count)
1170                 size = *count;
1171         if (buf->sent + size != buf->sg.length || *count != size)
1172                 flags |= MSG_MORE;
1173
1174         res = iscsi_send(conn, buf, size, flags);
1175         debug_tcp("sendpage: %d bytes, sent %d left %d sent %d res %d\n",
1176                   size, buf->sent, *count, *sent, res);
1177         if (res >= 0) {
1178                 *count -= res;
1179                 *sent += res;
1180                 if (size != res)
1181                         return -EAGAIN;
1182                 return 0;
1183         }
1184
1185         return res;
1186 }
1187
1188 static inline void
1189 iscsi_data_digest_init(struct iscsi_tcp_conn *tcp_conn,
1190                       struct iscsi_tcp_cmd_task *tcp_ctask)
1191 {
1192         crypto_hash_init(&tcp_conn->tx_hash);
1193         tcp_ctask->digest_count = 4;
1194 }
1195
1196 /**
1197  * iscsi_solicit_data_cont - initialize next Data-Out
1198  * @conn: iscsi connection
1199  * @ctask: scsi command task
1200  * @r2t: R2T info
1201  * @left: bytes left to transfer
1202  *
1203  * Notes:
1204  *      Initialize next Data-Out within this R2T sequence and continue
1205  *      to process next Scatter-Gather element(if any) of this SCSI command.
1206  *
1207  *      Called under connection lock.
1208  **/
1209 static void
1210 iscsi_solicit_data_cont(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
1211                         struct iscsi_r2t_info *r2t, int left)
1212 {
1213         struct iscsi_data *hdr;
1214         int new_offset;
1215
1216         hdr = &r2t->dtask.hdr;
1217         memset(hdr, 0, sizeof(struct iscsi_data));
1218         hdr->ttt = r2t->ttt;
1219         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
1220         r2t->solicit_datasn++;
1221         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
1222         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
1223         hdr->itt = ctask->hdr->itt;
1224         hdr->exp_statsn = r2t->exp_statsn;
1225         new_offset = r2t->data_offset + r2t->sent;
1226         hdr->offset = cpu_to_be32(new_offset);
1227         if (left > conn->max_xmit_dlength) {
1228                 hton24(hdr->dlength, conn->max_xmit_dlength);
1229                 r2t->data_count = conn->max_xmit_dlength;
1230         } else {
1231                 hton24(hdr->dlength, left);
1232                 r2t->data_count = left;
1233                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
1234         }
1235         conn->dataout_pdus_cnt++;
1236
1237         iscsi_buf_init_iov(&r2t->headbuf, (char*)hdr,
1238                            sizeof(struct iscsi_hdr));
1239
1240         if (iscsi_buf_left(&r2t->sendbuf))
1241                 return;
1242
1243         iscsi_buf_init_sg(&r2t->sendbuf, r2t->sg);
1244         r2t->sg += 1;
1245 }
1246
1247 static void iscsi_set_padding(struct iscsi_tcp_cmd_task *tcp_ctask,
1248                               unsigned long len)
1249 {
1250         tcp_ctask->pad_count = len & (ISCSI_PAD_LEN - 1);
1251         if (!tcp_ctask->pad_count)
1252                 return;
1253
1254         tcp_ctask->pad_count = ISCSI_PAD_LEN - tcp_ctask->pad_count;
1255         debug_scsi("write padding %d bytes\n", tcp_ctask->pad_count);
1256         tcp_ctask->xmstate |= XMSTATE_W_PAD;
1257 }
1258
1259 /**
1260  * iscsi_tcp_cmd_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
1261  * @conn: iscsi connection
1262  * @ctask: scsi command task
1263  * @sc: scsi command
1264  **/
1265 static void
1266 iscsi_tcp_cmd_init(struct iscsi_cmd_task *ctask)
1267 {
1268         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1269
1270         BUG_ON(__kfifo_len(tcp_ctask->r2tqueue));
1271         tcp_ctask->xmstate = XMSTATE_CMD_HDR_INIT;
1272 }
1273
1274 /**
1275  * iscsi_tcp_mtask_xmit - xmit management(immediate) task
1276  * @conn: iscsi connection
1277  * @mtask: task management task
1278  *
1279  * Notes:
1280  *      The function can return -EAGAIN in which case caller must
1281  *      call it again later, or recover. '0' return code means successful
1282  *      xmit.
1283  *
1284  *      Management xmit state machine consists of these states:
1285  *              XMSTATE_IMM_HDR_INIT    - calculate digest of PDU Header
1286  *              XMSTATE_IMM_HDR         - PDU Header xmit in progress
1287  *              XMSTATE_IMM_DATA        - PDU Data xmit in progress
1288  *              XMSTATE_IDLE            - management PDU is done
1289  **/
1290 static int
1291 iscsi_tcp_mtask_xmit(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1292 {
1293         struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
1294         int rc;
1295
1296         debug_scsi("mtask deq [cid %d state %x itt 0x%x]\n",
1297                 conn->id, tcp_mtask->xmstate, mtask->itt);
1298
1299         if (tcp_mtask->xmstate & XMSTATE_IMM_HDR_INIT) {
1300                 iscsi_buf_init_iov(&tcp_mtask->headbuf, (char*)mtask->hdr,
1301                                    sizeof(struct iscsi_hdr));
1302
1303                 if (mtask->data_count) {
1304                         tcp_mtask->xmstate |= XMSTATE_IMM_DATA;
1305                         iscsi_buf_init_iov(&tcp_mtask->sendbuf,
1306                                            (char*)mtask->data,
1307                                            mtask->data_count);
1308                 }
1309
1310                 if (conn->c_stage != ISCSI_CONN_INITIAL_STAGE &&
1311                     conn->stop_stage != STOP_CONN_RECOVER &&
1312                     conn->hdrdgst_en)
1313                         iscsi_hdr_digest(conn, &tcp_mtask->headbuf,
1314                                         (u8*)tcp_mtask->hdrext);
1315
1316                 tcp_mtask->sent = 0;
1317                 tcp_mtask->xmstate &= ~XMSTATE_IMM_HDR_INIT;
1318                 tcp_mtask->xmstate |= XMSTATE_IMM_HDR;
1319         }
1320
1321         if (tcp_mtask->xmstate & XMSTATE_IMM_HDR) {
1322                 rc = iscsi_sendhdr(conn, &tcp_mtask->headbuf,
1323                                    mtask->data_count);
1324                 if (rc)
1325                         return rc;
1326                 tcp_mtask->xmstate &= ~XMSTATE_IMM_HDR;
1327         }
1328
1329         if (tcp_mtask->xmstate & XMSTATE_IMM_DATA) {
1330                 BUG_ON(!mtask->data_count);
1331                 tcp_mtask->xmstate &= ~XMSTATE_IMM_DATA;
1332                 /* FIXME: implement.
1333                  * Virtual buffer could be spreaded across multiple pages...
1334                  */
1335                 do {
1336                         int rc;
1337
1338                         rc = iscsi_sendpage(conn, &tcp_mtask->sendbuf,
1339                                         &mtask->data_count, &tcp_mtask->sent);
1340                         if (rc) {
1341                                 tcp_mtask->xmstate |= XMSTATE_IMM_DATA;
1342                                 return rc;
1343                         }
1344                 } while (mtask->data_count);
1345         }
1346
1347         BUG_ON(tcp_mtask->xmstate != XMSTATE_IDLE);
1348         if (mtask->hdr->itt == RESERVED_ITT) {
1349                 struct iscsi_session *session = conn->session;
1350
1351                 spin_lock_bh(&session->lock);
1352                 iscsi_free_mgmt_task(conn, mtask);
1353                 spin_unlock_bh(&session->lock);
1354         }
1355         return 0;
1356 }
1357
1358 static int
1359 iscsi_send_cmd_hdr(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1360 {
1361         struct scsi_cmnd *sc = ctask->sc;
1362         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1363         int rc = 0;
1364
1365         if (tcp_ctask->xmstate & XMSTATE_CMD_HDR_INIT) {
1366                 tcp_ctask->sent = 0;
1367                 tcp_ctask->sg_count = 0;
1368                 tcp_ctask->exp_datasn = 0;
1369
1370                 if (sc->sc_data_direction == DMA_TO_DEVICE) {
1371                         struct scatterlist *sg = scsi_sglist(sc);
1372
1373                         iscsi_buf_init_sg(&tcp_ctask->sendbuf, sg);
1374                         tcp_ctask->sg = sg + 1;
1375                         tcp_ctask->bad_sg = sg + scsi_sg_count(sc);
1376
1377                         debug_scsi("cmd [itt 0x%x total %d imm_data %d "
1378                                    "unsol count %d, unsol offset %d]\n",
1379                                    ctask->itt, scsi_bufflen(sc),
1380                                    ctask->imm_count, ctask->unsol_count,
1381                                    ctask->unsol_offset);
1382                 }
1383
1384                 iscsi_buf_init_iov(&tcp_ctask->headbuf, (char*)ctask->hdr,
1385                                   ctask->hdr_len);
1386
1387                 if (conn->hdrdgst_en)
1388                         iscsi_hdr_digest(conn, &tcp_ctask->headbuf,
1389                                          iscsi_next_hdr(ctask));
1390                 tcp_ctask->xmstate &= ~XMSTATE_CMD_HDR_INIT;
1391                 tcp_ctask->xmstate |= XMSTATE_CMD_HDR_XMIT;
1392         }
1393
1394         if (tcp_ctask->xmstate & XMSTATE_CMD_HDR_XMIT) {
1395                 rc = iscsi_sendhdr(conn, &tcp_ctask->headbuf, ctask->imm_count);
1396                 if (rc)
1397                         return rc;
1398                 tcp_ctask->xmstate &= ~XMSTATE_CMD_HDR_XMIT;
1399
1400                 if (sc->sc_data_direction != DMA_TO_DEVICE)
1401                         return 0;
1402
1403                 if (ctask->imm_count) {
1404                         tcp_ctask->xmstate |= XMSTATE_IMM_DATA;
1405                         iscsi_set_padding(tcp_ctask, ctask->imm_count);
1406
1407                         if (ctask->conn->datadgst_en) {
1408                                 iscsi_data_digest_init(ctask->conn->dd_data,
1409                                                        tcp_ctask);
1410                                 tcp_ctask->immdigest = 0;
1411                         }
1412                 }
1413
1414                 if (ctask->unsol_count)
1415                         tcp_ctask->xmstate |=
1416                                         XMSTATE_UNS_HDR | XMSTATE_UNS_INIT;
1417         }
1418         return rc;
1419 }
1420
1421 static int
1422 iscsi_send_padding(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1423 {
1424         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1425         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1426         int sent = 0, rc;
1427
1428         if (tcp_ctask->xmstate & XMSTATE_W_PAD) {
1429                 iscsi_buf_init_iov(&tcp_ctask->sendbuf, (char*)&tcp_ctask->pad,
1430                                    tcp_ctask->pad_count);
1431                 if (conn->datadgst_en)
1432                         crypto_hash_update(&tcp_conn->tx_hash,
1433                                            &tcp_ctask->sendbuf.sg,
1434                                            tcp_ctask->sendbuf.sg.length);
1435         } else if (!(tcp_ctask->xmstate & XMSTATE_W_RESEND_PAD))
1436                 return 0;
1437
1438         tcp_ctask->xmstate &= ~XMSTATE_W_PAD;
1439         tcp_ctask->xmstate &= ~XMSTATE_W_RESEND_PAD;
1440         debug_scsi("sending %d pad bytes for itt 0x%x\n",
1441                    tcp_ctask->pad_count, ctask->itt);
1442         rc = iscsi_sendpage(conn, &tcp_ctask->sendbuf, &tcp_ctask->pad_count,
1443                            &sent);
1444         if (rc) {
1445                 debug_scsi("padding send failed %d\n", rc);
1446                 tcp_ctask->xmstate |= XMSTATE_W_RESEND_PAD;
1447         }
1448         return rc;
1449 }
1450
1451 static int
1452 iscsi_send_digest(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
1453                         struct iscsi_buf *buf, uint32_t *digest)
1454 {
1455         struct iscsi_tcp_cmd_task *tcp_ctask;
1456         struct iscsi_tcp_conn *tcp_conn;
1457         int rc, sent = 0;
1458
1459         if (!conn->datadgst_en)
1460                 return 0;
1461
1462         tcp_ctask = ctask->dd_data;
1463         tcp_conn = conn->dd_data;
1464
1465         if (!(tcp_ctask->xmstate & XMSTATE_W_RESEND_DATA_DIGEST)) {
1466                 crypto_hash_final(&tcp_conn->tx_hash, (u8*)digest);
1467                 iscsi_buf_init_iov(buf, (char*)digest, 4);
1468         }
1469         tcp_ctask->xmstate &= ~XMSTATE_W_RESEND_DATA_DIGEST;
1470
1471         rc = iscsi_sendpage(conn, buf, &tcp_ctask->digest_count, &sent);
1472         if (!rc)
1473                 debug_scsi("sent digest 0x%x for itt 0x%x\n", *digest,
1474                           ctask->itt);
1475         else {
1476                 debug_scsi("sending digest 0x%x failed for itt 0x%x!\n",
1477                           *digest, ctask->itt);
1478                 tcp_ctask->xmstate |= XMSTATE_W_RESEND_DATA_DIGEST;
1479         }
1480         return rc;
1481 }
1482
1483 static int
1484 iscsi_send_data(struct iscsi_cmd_task *ctask, struct iscsi_buf *sendbuf,
1485                 struct scatterlist **sg, int *sent, int *count,
1486                 struct iscsi_buf *digestbuf, uint32_t *digest)
1487 {
1488         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1489         struct iscsi_conn *conn = ctask->conn;
1490         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1491         int rc, buf_sent, offset;
1492
1493         while (*count) {
1494                 buf_sent = 0;
1495                 offset = sendbuf->sent;
1496
1497                 rc = iscsi_sendpage(conn, sendbuf, count, &buf_sent);
1498                 *sent = *sent + buf_sent;
1499                 if (buf_sent && conn->datadgst_en)
1500                         partial_sg_digest_update(&tcp_conn->tx_hash,
1501                                 &sendbuf->sg, sendbuf->sg.offset + offset,
1502                                 buf_sent);
1503                 if (!iscsi_buf_left(sendbuf) && *sg != tcp_ctask->bad_sg) {
1504                         iscsi_buf_init_sg(sendbuf, *sg);
1505                         *sg = *sg + 1;
1506                 }
1507
1508                 if (rc)
1509                         return rc;
1510         }
1511
1512         rc = iscsi_send_padding(conn, ctask);
1513         if (rc)
1514                 return rc;
1515
1516         return iscsi_send_digest(conn, ctask, digestbuf, digest);
1517 }
1518
1519 static int
1520 iscsi_send_unsol_hdr(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1521 {
1522         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1523         struct iscsi_data_task *dtask;
1524         int rc;
1525
1526         tcp_ctask->xmstate |= XMSTATE_UNS_DATA;
1527         if (tcp_ctask->xmstate & XMSTATE_UNS_INIT) {
1528                 dtask = &tcp_ctask->unsol_dtask;
1529
1530                 iscsi_prep_unsolicit_data_pdu(ctask, &dtask->hdr);
1531                 iscsi_buf_init_iov(&tcp_ctask->headbuf, (char*)&dtask->hdr,
1532                                    sizeof(struct iscsi_hdr));
1533                 if (conn->hdrdgst_en)
1534                         iscsi_hdr_digest(conn, &tcp_ctask->headbuf,
1535                                         (u8*)dtask->hdrext);
1536
1537                 tcp_ctask->xmstate &= ~XMSTATE_UNS_INIT;
1538                 iscsi_set_padding(tcp_ctask, ctask->data_count);
1539         }
1540
1541         rc = iscsi_sendhdr(conn, &tcp_ctask->headbuf, ctask->data_count);
1542         if (rc) {
1543                 tcp_ctask->xmstate &= ~XMSTATE_UNS_DATA;
1544                 tcp_ctask->xmstate |= XMSTATE_UNS_HDR;
1545                 return rc;
1546         }
1547
1548         if (conn->datadgst_en) {
1549                 dtask = &tcp_ctask->unsol_dtask;
1550                 iscsi_data_digest_init(ctask->conn->dd_data, tcp_ctask);
1551                 dtask->digest = 0;
1552         }
1553
1554         debug_scsi("uns dout [itt 0x%x dlen %d sent %d]\n",
1555                    ctask->itt, ctask->unsol_count, tcp_ctask->sent);
1556         return 0;
1557 }
1558
1559 static int
1560 iscsi_send_unsol_pdu(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1561 {
1562         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1563         int rc;
1564
1565         if (tcp_ctask->xmstate & XMSTATE_UNS_HDR) {
1566                 BUG_ON(!ctask->unsol_count);
1567                 tcp_ctask->xmstate &= ~XMSTATE_UNS_HDR;
1568 send_hdr:
1569                 rc = iscsi_send_unsol_hdr(conn, ctask);
1570                 if (rc)
1571                         return rc;
1572         }
1573
1574         if (tcp_ctask->xmstate & XMSTATE_UNS_DATA) {
1575                 struct iscsi_data_task *dtask = &tcp_ctask->unsol_dtask;
1576                 int start = tcp_ctask->sent;
1577
1578                 rc = iscsi_send_data(ctask, &tcp_ctask->sendbuf, &tcp_ctask->sg,
1579                                      &tcp_ctask->sent, &ctask->data_count,
1580                                      &dtask->digestbuf, &dtask->digest);
1581                 ctask->unsol_count -= tcp_ctask->sent - start;
1582                 if (rc)
1583                         return rc;
1584                 tcp_ctask->xmstate &= ~XMSTATE_UNS_DATA;
1585                 /*
1586                  * Done with the Data-Out. Next, check if we need
1587                  * to send another unsolicited Data-Out.
1588                  */
1589                 if (ctask->unsol_count) {
1590                         debug_scsi("sending more uns\n");
1591                         tcp_ctask->xmstate |= XMSTATE_UNS_INIT;
1592                         goto send_hdr;
1593                 }
1594         }
1595         return 0;
1596 }
1597
1598 static int iscsi_send_sol_pdu(struct iscsi_conn *conn,
1599                               struct iscsi_cmd_task *ctask)
1600 {
1601         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1602         struct iscsi_session *session = conn->session;
1603         struct iscsi_r2t_info *r2t;
1604         struct iscsi_data_task *dtask;
1605         int left, rc;
1606
1607         if (tcp_ctask->xmstate & XMSTATE_SOL_HDR_INIT) {
1608                 if (!tcp_ctask->r2t) {
1609                         spin_lock_bh(&session->lock);
1610                         __kfifo_get(tcp_ctask->r2tqueue, (void*)&tcp_ctask->r2t,
1611                                     sizeof(void*));
1612                         spin_unlock_bh(&session->lock);
1613                 }
1614 send_hdr:
1615                 r2t = tcp_ctask->r2t;
1616                 dtask = &r2t->dtask;
1617
1618                 if (conn->hdrdgst_en)
1619                         iscsi_hdr_digest(conn, &r2t->headbuf,
1620                                         (u8*)dtask->hdrext);
1621                 tcp_ctask->xmstate &= ~XMSTATE_SOL_HDR_INIT;
1622                 tcp_ctask->xmstate |= XMSTATE_SOL_HDR;
1623         }
1624
1625         if (tcp_ctask->xmstate & XMSTATE_SOL_HDR) {
1626                 r2t = tcp_ctask->r2t;
1627                 dtask = &r2t->dtask;
1628
1629                 rc = iscsi_sendhdr(conn, &r2t->headbuf, r2t->data_count);
1630                 if (rc)
1631                         return rc;
1632                 tcp_ctask->xmstate &= ~XMSTATE_SOL_HDR;
1633                 tcp_ctask->xmstate |= XMSTATE_SOL_DATA;
1634
1635                 if (conn->datadgst_en) {
1636                         iscsi_data_digest_init(conn->dd_data, tcp_ctask);
1637                         dtask->digest = 0;
1638                 }
1639
1640                 iscsi_set_padding(tcp_ctask, r2t->data_count);
1641                 debug_scsi("sol dout [dsn %d itt 0x%x dlen %d sent %d]\n",
1642                         r2t->solicit_datasn - 1, ctask->itt, r2t->data_count,
1643                         r2t->sent);
1644         }
1645
1646         if (tcp_ctask->xmstate & XMSTATE_SOL_DATA) {
1647                 r2t = tcp_ctask->r2t;
1648                 dtask = &r2t->dtask;
1649
1650                 rc = iscsi_send_data(ctask, &r2t->sendbuf, &r2t->sg,
1651                                      &r2t->sent, &r2t->data_count,
1652                                      &dtask->digestbuf, &dtask->digest);
1653                 if (rc)
1654                         return rc;
1655                 tcp_ctask->xmstate &= ~XMSTATE_SOL_DATA;
1656
1657                 /*
1658                  * Done with this Data-Out. Next, check if we have
1659                  * to send another Data-Out for this R2T.
1660                  */
1661                 BUG_ON(r2t->data_length - r2t->sent < 0);
1662                 left = r2t->data_length - r2t->sent;
1663                 if (left) {
1664                         iscsi_solicit_data_cont(conn, ctask, r2t, left);
1665                         goto send_hdr;
1666                 }
1667
1668                 /*
1669                  * Done with this R2T. Check if there are more
1670                  * outstanding R2Ts ready to be processed.
1671                  */
1672                 spin_lock_bh(&session->lock);
1673                 tcp_ctask->r2t = NULL;
1674                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
1675                             sizeof(void*));
1676                 if (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t,
1677                                 sizeof(void*))) {
1678                         tcp_ctask->r2t = r2t;
1679                         spin_unlock_bh(&session->lock);
1680                         goto send_hdr;
1681                 }
1682                 spin_unlock_bh(&session->lock);
1683         }
1684         return 0;
1685 }
1686
1687 /**
1688  * iscsi_tcp_ctask_xmit - xmit normal PDU task
1689  * @conn: iscsi connection
1690  * @ctask: iscsi command task
1691  *
1692  * Notes:
1693  *      The function can return -EAGAIN in which case caller must
1694  *      call it again later, or recover. '0' return code means successful
1695  *      xmit.
1696  *      The function is devided to logical helpers (above) for the different
1697  *      xmit stages.
1698  *
1699  *iscsi_send_cmd_hdr()
1700  *      XMSTATE_CMD_HDR_INIT - prepare Header and Data buffers Calculate
1701  *                             Header Digest
1702  *      XMSTATE_CMD_HDR_XMIT - Transmit header in progress
1703  *
1704  *iscsi_send_padding
1705  *      XMSTATE_W_PAD        - Prepare and send pading
1706  *      XMSTATE_W_RESEND_PAD - retry send pading
1707  *
1708  *iscsi_send_digest
1709  *      XMSTATE_W_RESEND_DATA_DIGEST - Finalize and send Data Digest
1710  *      XMSTATE_W_RESEND_DATA_DIGEST - retry sending digest
1711  *
1712  *iscsi_send_unsol_hdr
1713  *      XMSTATE_UNS_INIT     - prepare un-solicit data header and digest
1714  *      XMSTATE_UNS_HDR      - send un-solicit header
1715  *
1716  *iscsi_send_unsol_pdu
1717  *      XMSTATE_UNS_DATA     - send un-solicit data in progress
1718  *
1719  *iscsi_send_sol_pdu
1720  *      XMSTATE_SOL_HDR_INIT - solicit data header and digest initialize
1721  *      XMSTATE_SOL_HDR      - send solicit header
1722  *      XMSTATE_SOL_DATA     - send solicit data
1723  *
1724  *iscsi_tcp_ctask_xmit
1725  *      XMSTATE_IMM_DATA     - xmit managment data (??)
1726  **/
1727 static int
1728 iscsi_tcp_ctask_xmit(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1729 {
1730         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1731         int rc = 0;
1732
1733         debug_scsi("ctask deq [cid %d xmstate %x itt 0x%x]\n",
1734                 conn->id, tcp_ctask->xmstate, ctask->itt);
1735
1736         rc = iscsi_send_cmd_hdr(conn, ctask);
1737         if (rc)
1738                 return rc;
1739         if (ctask->sc->sc_data_direction != DMA_TO_DEVICE)
1740                 return 0;
1741
1742         if (tcp_ctask->xmstate & XMSTATE_IMM_DATA) {
1743                 rc = iscsi_send_data(ctask, &tcp_ctask->sendbuf, &tcp_ctask->sg,
1744                                      &tcp_ctask->sent, &ctask->imm_count,
1745                                      &tcp_ctask->immbuf, &tcp_ctask->immdigest);
1746                 if (rc)
1747                         return rc;
1748                 tcp_ctask->xmstate &= ~XMSTATE_IMM_DATA;
1749         }
1750
1751         rc = iscsi_send_unsol_pdu(conn, ctask);
1752         if (rc)
1753                 return rc;
1754
1755         rc = iscsi_send_sol_pdu(conn, ctask);
1756         if (rc)
1757                 return rc;
1758
1759         return rc;
1760 }
1761
1762 static struct iscsi_cls_conn *
1763 iscsi_tcp_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
1764 {
1765         struct iscsi_conn *conn;
1766         struct iscsi_cls_conn *cls_conn;
1767         struct iscsi_tcp_conn *tcp_conn;
1768
1769         cls_conn = iscsi_conn_setup(cls_session, conn_idx);
1770         if (!cls_conn)
1771                 return NULL;
1772         conn = cls_conn->dd_data;
1773         /*
1774          * due to strange issues with iser these are not set
1775          * in iscsi_conn_setup
1776          */
1777         conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1778
1779         tcp_conn = kzalloc(sizeof(*tcp_conn), GFP_KERNEL);
1780         if (!tcp_conn)
1781                 goto tcp_conn_alloc_fail;
1782
1783         conn->dd_data = tcp_conn;
1784         tcp_conn->iscsi_conn = conn;
1785
1786         tcp_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1787                                                   CRYPTO_ALG_ASYNC);
1788         tcp_conn->tx_hash.flags = 0;
1789         if (IS_ERR(tcp_conn->tx_hash.tfm)) {
1790                 printk(KERN_ERR "Could not create connection due to crc32c "
1791                        "loading error %ld. Make sure the crc32c module is "
1792                        "built as a module or into the kernel\n",
1793                         PTR_ERR(tcp_conn->tx_hash.tfm));
1794                 goto free_tcp_conn;
1795         }
1796
1797         tcp_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1798                                                   CRYPTO_ALG_ASYNC);
1799         tcp_conn->rx_hash.flags = 0;
1800         if (IS_ERR(tcp_conn->rx_hash.tfm)) {
1801                 printk(KERN_ERR "Could not create connection due to crc32c "
1802                        "loading error %ld. Make sure the crc32c module is "
1803                        "built as a module or into the kernel\n",
1804                         PTR_ERR(tcp_conn->rx_hash.tfm));
1805                 goto free_tx_tfm;
1806         }
1807
1808         return cls_conn;
1809
1810 free_tx_tfm:
1811         crypto_free_hash(tcp_conn->tx_hash.tfm);
1812 free_tcp_conn:
1813         kfree(tcp_conn);
1814 tcp_conn_alloc_fail:
1815         iscsi_conn_teardown(cls_conn);
1816         return NULL;
1817 }
1818
1819 static void
1820 iscsi_tcp_release_conn(struct iscsi_conn *conn)
1821 {
1822         struct iscsi_session *session = conn->session;
1823         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1824         struct socket *sock = tcp_conn->sock;
1825
1826         if (!sock)
1827                 return;
1828
1829         sock_hold(sock->sk);
1830         iscsi_conn_restore_callbacks(tcp_conn);
1831         sock_put(sock->sk);
1832
1833         spin_lock_bh(&session->lock);
1834         tcp_conn->sock = NULL;
1835         conn->recv_lock = NULL;
1836         spin_unlock_bh(&session->lock);
1837         sockfd_put(sock);
1838 }
1839
1840 static void
1841 iscsi_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
1842 {
1843         struct iscsi_conn *conn = cls_conn->dd_data;
1844         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1845
1846         iscsi_tcp_release_conn(conn);
1847         iscsi_conn_teardown(cls_conn);
1848
1849         if (tcp_conn->tx_hash.tfm)
1850                 crypto_free_hash(tcp_conn->tx_hash.tfm);
1851         if (tcp_conn->rx_hash.tfm)
1852                 crypto_free_hash(tcp_conn->rx_hash.tfm);
1853
1854         kfree(tcp_conn);
1855 }
1856
1857 static void
1858 iscsi_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
1859 {
1860         struct iscsi_conn *conn = cls_conn->dd_data;
1861
1862         iscsi_conn_stop(cls_conn, flag);
1863         iscsi_tcp_release_conn(conn);
1864 }
1865
1866 static int iscsi_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
1867                               char *buf, int *port,
1868                               int (*getname)(struct socket *, struct sockaddr *,
1869                                         int *addrlen))
1870 {
1871         struct sockaddr_storage *addr;
1872         struct sockaddr_in6 *sin6;
1873         struct sockaddr_in *sin;
1874         int rc = 0, len;
1875
1876         addr = kmalloc(sizeof(*addr), GFP_KERNEL);
1877         if (!addr)
1878                 return -ENOMEM;
1879
1880         if (getname(sock, (struct sockaddr *) addr, &len)) {
1881                 rc = -ENODEV;
1882                 goto free_addr;
1883         }
1884
1885         switch (addr->ss_family) {
1886         case AF_INET:
1887                 sin = (struct sockaddr_in *)addr;
1888                 spin_lock_bh(&conn->session->lock);
1889                 sprintf(buf, NIPQUAD_FMT, NIPQUAD(sin->sin_addr.s_addr));
1890                 *port = be16_to_cpu(sin->sin_port);
1891                 spin_unlock_bh(&conn->session->lock);
1892                 break;
1893         case AF_INET6:
1894                 sin6 = (struct sockaddr_in6 *)addr;
1895                 spin_lock_bh(&conn->session->lock);
1896                 sprintf(buf, NIP6_FMT, NIP6(sin6->sin6_addr));
1897                 *port = be16_to_cpu(sin6->sin6_port);
1898                 spin_unlock_bh(&conn->session->lock);
1899                 break;
1900         }
1901 free_addr:
1902         kfree(addr);
1903         return rc;
1904 }
1905
1906 static int
1907 iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
1908                     struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
1909                     int is_leading)
1910 {
1911         struct iscsi_conn *conn = cls_conn->dd_data;
1912         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1913         struct sock *sk;
1914         struct socket *sock;
1915         int err;
1916
1917         /* lookup for existing socket */
1918         sock = sockfd_lookup((int)transport_eph, &err);
1919         if (!sock) {
1920                 printk(KERN_ERR "iscsi_tcp: sockfd_lookup failed %d\n", err);
1921                 return -EEXIST;
1922         }
1923         /*
1924          * copy these values now because if we drop the session
1925          * userspace may still want to query the values since we will
1926          * be using them for the reconnect
1927          */
1928         err = iscsi_tcp_get_addr(conn, sock, conn->portal_address,
1929                                  &conn->portal_port, kernel_getpeername);
1930         if (err)
1931                 goto free_socket;
1932
1933         err = iscsi_tcp_get_addr(conn, sock, conn->local_address,
1934                                 &conn->local_port, kernel_getsockname);
1935         if (err)
1936                 goto free_socket;
1937
1938         err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
1939         if (err)
1940                 goto free_socket;
1941
1942         /* bind iSCSI connection and socket */
1943         tcp_conn->sock = sock;
1944
1945         /* setup Socket parameters */
1946         sk = sock->sk;
1947         sk->sk_reuse = 1;
1948         sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
1949         sk->sk_allocation = GFP_ATOMIC;
1950
1951         /* FIXME: disable Nagle's algorithm */
1952
1953         /*
1954          * Intercept TCP callbacks for sendfile like receive
1955          * processing.
1956          */
1957         conn->recv_lock = &sk->sk_callback_lock;
1958         iscsi_conn_set_callbacks(conn);
1959         tcp_conn->sendpage = tcp_conn->sock->ops->sendpage;
1960         /*
1961          * set receive state machine into initial state
1962          */
1963         iscsi_tcp_hdr_recv_prep(tcp_conn);
1964         return 0;
1965
1966 free_socket:
1967         sockfd_put(sock);
1968         return err;
1969 }
1970
1971 /* called with host lock */
1972 static void
1973 iscsi_tcp_mgmt_init(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1974 {
1975         struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
1976         tcp_mtask->xmstate = XMSTATE_IMM_HDR_INIT;
1977 }
1978
1979 static int
1980 iscsi_r2tpool_alloc(struct iscsi_session *session)
1981 {
1982         int i;
1983         int cmd_i;
1984
1985         /*
1986          * initialize per-task: R2T pool and xmit queue
1987          */
1988         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1989                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
1990                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1991
1992                 /*
1993                  * pre-allocated x4 as much r2ts to handle race when
1994                  * target acks DataOut faster than we data_xmit() queues
1995                  * could replenish r2tqueue.
1996                  */
1997
1998                 /* R2T pool */
1999                 if (iscsi_pool_init(&tcp_ctask->r2tpool, session->max_r2t * 4, NULL,
2000                                     sizeof(struct iscsi_r2t_info))) {
2001                         goto r2t_alloc_fail;
2002                 }
2003
2004                 /* R2T xmit queue */
2005                 tcp_ctask->r2tqueue = kfifo_alloc(
2006                       session->max_r2t * 4 * sizeof(void*), GFP_KERNEL, NULL);
2007                 if (tcp_ctask->r2tqueue == ERR_PTR(-ENOMEM)) {
2008                         iscsi_pool_free(&tcp_ctask->r2tpool);
2009                         goto r2t_alloc_fail;
2010                 }
2011         }
2012
2013         return 0;
2014
2015 r2t_alloc_fail:
2016         for (i = 0; i < cmd_i; i++) {
2017                 struct iscsi_cmd_task *ctask = session->cmds[i];
2018                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2019
2020                 kfifo_free(tcp_ctask->r2tqueue);
2021                 iscsi_pool_free(&tcp_ctask->r2tpool);
2022         }
2023         return -ENOMEM;
2024 }
2025
2026 static void
2027 iscsi_r2tpool_free(struct iscsi_session *session)
2028 {
2029         int i;
2030
2031         for (i = 0; i < session->cmds_max; i++) {
2032                 struct iscsi_cmd_task *ctask = session->cmds[i];
2033                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2034
2035                 kfifo_free(tcp_ctask->r2tqueue);
2036                 iscsi_pool_free(&tcp_ctask->r2tpool);
2037         }
2038 }
2039
2040 static int
2041 iscsi_conn_set_param(struct iscsi_cls_conn *cls_conn, enum iscsi_param param,
2042                      char *buf, int buflen)
2043 {
2044         struct iscsi_conn *conn = cls_conn->dd_data;
2045         struct iscsi_session *session = conn->session;
2046         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2047         int value;
2048
2049         switch(param) {
2050         case ISCSI_PARAM_HDRDGST_EN:
2051                 iscsi_set_param(cls_conn, param, buf, buflen);
2052                 break;
2053         case ISCSI_PARAM_DATADGST_EN:
2054                 iscsi_set_param(cls_conn, param, buf, buflen);
2055                 tcp_conn->sendpage = conn->datadgst_en ?
2056                         sock_no_sendpage : tcp_conn->sock->ops->sendpage;
2057                 break;
2058         case ISCSI_PARAM_MAX_R2T:
2059                 sscanf(buf, "%d", &value);
2060                 if (session->max_r2t == roundup_pow_of_two(value))
2061                         break;
2062                 iscsi_r2tpool_free(session);
2063                 iscsi_set_param(cls_conn, param, buf, buflen);
2064                 if (session->max_r2t & (session->max_r2t - 1))
2065                         session->max_r2t = roundup_pow_of_two(session->max_r2t);
2066                 if (iscsi_r2tpool_alloc(session))
2067                         return -ENOMEM;
2068                 break;
2069         default:
2070                 return iscsi_set_param(cls_conn, param, buf, buflen);
2071         }
2072
2073         return 0;
2074 }
2075
2076 static int
2077 iscsi_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
2078                          enum iscsi_param param, char *buf)
2079 {
2080         struct iscsi_conn *conn = cls_conn->dd_data;
2081         int len;
2082
2083         switch(param) {
2084         case ISCSI_PARAM_CONN_PORT:
2085                 spin_lock_bh(&conn->session->lock);
2086                 len = sprintf(buf, "%hu\n", conn->portal_port);
2087                 spin_unlock_bh(&conn->session->lock);
2088                 break;
2089         case ISCSI_PARAM_CONN_ADDRESS:
2090                 spin_lock_bh(&conn->session->lock);
2091                 len = sprintf(buf, "%s\n", conn->portal_address);
2092                 spin_unlock_bh(&conn->session->lock);
2093                 break;
2094         default:
2095                 return iscsi_conn_get_param(cls_conn, param, buf);
2096         }
2097
2098         return len;
2099 }
2100
2101 static int
2102 iscsi_tcp_host_get_param(struct Scsi_Host *shost, enum iscsi_host_param param,
2103                          char *buf)
2104 {
2105         struct iscsi_session *session = iscsi_hostdata(shost->hostdata);
2106         int len;
2107
2108         switch (param) {
2109         case ISCSI_HOST_PARAM_IPADDRESS:
2110                 spin_lock_bh(&session->lock);
2111                 if (!session->leadconn)
2112                         len = -ENODEV;
2113                 else
2114                         len = sprintf(buf, "%s\n",
2115                                      session->leadconn->local_address);
2116                 spin_unlock_bh(&session->lock);
2117                 break;
2118         default:
2119                 return iscsi_host_get_param(shost, param, buf);
2120         }
2121         return len;
2122 }
2123
2124 static void
2125 iscsi_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
2126 {
2127         struct iscsi_conn *conn = cls_conn->dd_data;
2128         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2129
2130         stats->txdata_octets = conn->txdata_octets;
2131         stats->rxdata_octets = conn->rxdata_octets;
2132         stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
2133         stats->dataout_pdus = conn->dataout_pdus_cnt;
2134         stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
2135         stats->datain_pdus = conn->datain_pdus_cnt;
2136         stats->r2t_pdus = conn->r2t_pdus_cnt;
2137         stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
2138         stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
2139         stats->custom_length = 3;
2140         strcpy(stats->custom[0].desc, "tx_sendpage_failures");
2141         stats->custom[0].value = tcp_conn->sendpage_failures_cnt;
2142         strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
2143         stats->custom[1].value = tcp_conn->discontiguous_hdr_cnt;
2144         strcpy(stats->custom[2].desc, "eh_abort_cnt");
2145         stats->custom[2].value = conn->eh_abort_cnt;
2146 }
2147
2148 static struct iscsi_cls_session *
2149 iscsi_tcp_session_create(struct iscsi_transport *iscsit,
2150                          struct scsi_transport_template *scsit,
2151                          uint16_t cmds_max, uint16_t qdepth,
2152                          uint32_t initial_cmdsn, uint32_t *hostno)
2153 {
2154         struct iscsi_cls_session *cls_session;
2155         struct iscsi_session *session;
2156         uint32_t hn;
2157         int cmd_i;
2158
2159         cls_session = iscsi_session_setup(iscsit, scsit, cmds_max, qdepth,
2160                                          sizeof(struct iscsi_tcp_cmd_task),
2161                                          sizeof(struct iscsi_tcp_mgmt_task),
2162                                          initial_cmdsn, &hn);
2163         if (!cls_session)
2164                 return NULL;
2165         *hostno = hn;
2166
2167         session = class_to_transport_session(cls_session);
2168         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
2169                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
2170                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
2171
2172                 ctask->hdr = &tcp_ctask->hdr.cmd_hdr;
2173                 ctask->hdr_max = sizeof(tcp_ctask->hdr) - ISCSI_DIGEST_SIZE;
2174         }
2175
2176         for (cmd_i = 0; cmd_i < session->mgmtpool_max; cmd_i++) {
2177                 struct iscsi_mgmt_task *mtask = session->mgmt_cmds[cmd_i];
2178                 struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
2179
2180                 mtask->hdr = &tcp_mtask->hdr;
2181         }
2182
2183         if (iscsi_r2tpool_alloc(class_to_transport_session(cls_session)))
2184                 goto r2tpool_alloc_fail;
2185
2186         return cls_session;
2187
2188 r2tpool_alloc_fail:
2189         iscsi_session_teardown(cls_session);
2190         return NULL;
2191 }
2192
2193 static void iscsi_tcp_session_destroy(struct iscsi_cls_session *cls_session)
2194 {
2195         iscsi_r2tpool_free(class_to_transport_session(cls_session));
2196         iscsi_session_teardown(cls_session);
2197 }
2198
2199 static int iscsi_tcp_slave_configure(struct scsi_device *sdev)
2200 {
2201         blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
2202         blk_queue_dma_alignment(sdev->request_queue, 0);
2203         return 0;
2204 }
2205
2206 static struct scsi_host_template iscsi_sht = {
2207         .module                 = THIS_MODULE,
2208         .name                   = "iSCSI Initiator over TCP/IP",
2209         .queuecommand           = iscsi_queuecommand,
2210         .change_queue_depth     = iscsi_change_queue_depth,
2211         .can_queue              = ISCSI_DEF_XMIT_CMDS_MAX - 1,
2212         .sg_tablesize           = ISCSI_SG_TABLESIZE,
2213         .max_sectors            = 0xFFFF,
2214         .cmd_per_lun            = ISCSI_DEF_CMD_PER_LUN,
2215         .eh_abort_handler       = iscsi_eh_abort,
2216         .eh_device_reset_handler= iscsi_eh_device_reset,
2217         .eh_host_reset_handler  = iscsi_eh_host_reset,
2218         .use_clustering         = DISABLE_CLUSTERING,
2219         .slave_configure        = iscsi_tcp_slave_configure,
2220         .proc_name              = "iscsi_tcp",
2221         .this_id                = -1,
2222 };
2223
2224 static struct iscsi_transport iscsi_tcp_transport = {
2225         .owner                  = THIS_MODULE,
2226         .name                   = "tcp",
2227         .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
2228                                   | CAP_DATADGST,
2229         .param_mask             = ISCSI_MAX_RECV_DLENGTH |
2230                                   ISCSI_MAX_XMIT_DLENGTH |
2231                                   ISCSI_HDRDGST_EN |
2232                                   ISCSI_DATADGST_EN |
2233                                   ISCSI_INITIAL_R2T_EN |
2234                                   ISCSI_MAX_R2T |
2235                                   ISCSI_IMM_DATA_EN |
2236                                   ISCSI_FIRST_BURST |
2237                                   ISCSI_MAX_BURST |
2238                                   ISCSI_PDU_INORDER_EN |
2239                                   ISCSI_DATASEQ_INORDER_EN |
2240                                   ISCSI_ERL |
2241                                   ISCSI_CONN_PORT |
2242                                   ISCSI_CONN_ADDRESS |
2243                                   ISCSI_EXP_STATSN |
2244                                   ISCSI_PERSISTENT_PORT |
2245                                   ISCSI_PERSISTENT_ADDRESS |
2246                                   ISCSI_TARGET_NAME | ISCSI_TPGT |
2247                                   ISCSI_USERNAME | ISCSI_PASSWORD |
2248                                   ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
2249                                   ISCSI_FAST_ABORT,
2250         .host_param_mask        = ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
2251                                   ISCSI_HOST_INITIATOR_NAME |
2252                                   ISCSI_HOST_NETDEV_NAME,
2253         .host_template          = &iscsi_sht,
2254         .conndata_size          = sizeof(struct iscsi_conn),
2255         .max_conn               = 1,
2256         .max_cmd_len            = ISCSI_TCP_MAX_CMD_LEN,
2257         /* session management */
2258         .create_session         = iscsi_tcp_session_create,
2259         .destroy_session        = iscsi_tcp_session_destroy,
2260         /* connection management */
2261         .create_conn            = iscsi_tcp_conn_create,
2262         .bind_conn              = iscsi_tcp_conn_bind,
2263         .destroy_conn           = iscsi_tcp_conn_destroy,
2264         .set_param              = iscsi_conn_set_param,
2265         .get_conn_param         = iscsi_tcp_conn_get_param,
2266         .get_session_param      = iscsi_session_get_param,
2267         .start_conn             = iscsi_conn_start,
2268         .stop_conn              = iscsi_tcp_conn_stop,
2269         /* iscsi host params */
2270         .get_host_param         = iscsi_tcp_host_get_param,
2271         .set_host_param         = iscsi_host_set_param,
2272         /* IO */
2273         .send_pdu               = iscsi_conn_send_pdu,
2274         .get_stats              = iscsi_conn_get_stats,
2275         .init_cmd_task          = iscsi_tcp_cmd_init,
2276         .init_mgmt_task         = iscsi_tcp_mgmt_init,
2277         .xmit_cmd_task          = iscsi_tcp_ctask_xmit,
2278         .xmit_mgmt_task         = iscsi_tcp_mtask_xmit,
2279         .cleanup_cmd_task       = iscsi_tcp_cleanup_ctask,
2280         /* recovery */
2281         .session_recovery_timedout = iscsi_session_recovery_timedout,
2282 };
2283
2284 static int __init
2285 iscsi_tcp_init(void)
2286 {
2287         if (iscsi_max_lun < 1) {
2288                 printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
2289                        iscsi_max_lun);
2290                 return -EINVAL;
2291         }
2292         iscsi_tcp_transport.max_lun = iscsi_max_lun;
2293
2294         if (!iscsi_register_transport(&iscsi_tcp_transport))
2295                 return -ENODEV;
2296
2297         return 0;
2298 }
2299
2300 static void __exit
2301 iscsi_tcp_exit(void)
2302 {
2303         iscsi_unregister_transport(&iscsi_tcp_transport);
2304 }
2305
2306 module_init(iscsi_tcp_init);
2307 module_exit(iscsi_tcp_exit);