9df6b3436e4a517a52be377a6873152b2ab8b93a
[linux-2.6.git] / drivers / scsi / libiscsi_tcp.c
1 /*
2  * iSCSI over TCP/IP Data-Path lib
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("Mike Christie <michaelc@cs.wisc.edu>, "
48               "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
49               "Alex Aizman <itn780@yahoo.com>");
50 MODULE_DESCRIPTION("iSCSI/TCP data-path");
51 MODULE_LICENSE("GPL");
52 #undef DEBUG_TCP
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 static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
61                                    struct iscsi_segment *segment);
62
63 /*
64  * Scatterlist handling: inside the iscsi_segment, we
65  * remember an index into the scatterlist, and set data/size
66  * to the current scatterlist entry. For highmem pages, we
67  * kmap as needed.
68  *
69  * Note that the page is unmapped when we return from
70  * TCP's data_ready handler, so we may end up mapping and
71  * unmapping the same page repeatedly. The whole reason
72  * for this is that we shouldn't keep the page mapped
73  * outside the softirq.
74  */
75
76 /**
77  * iscsi_tcp_segment_init_sg - init indicated scatterlist entry
78  * @segment: the buffer object
79  * @sg: scatterlist
80  * @offset: byte offset into that sg entry
81  *
82  * This function sets up the segment so that subsequent
83  * data is copied to the indicated sg entry, at the given
84  * offset.
85  */
86 static inline void
87 iscsi_tcp_segment_init_sg(struct iscsi_segment *segment,
88                           struct scatterlist *sg, unsigned int offset)
89 {
90         segment->sg = sg;
91         segment->sg_offset = offset;
92         segment->size = min(sg->length - offset,
93                             segment->total_size - segment->total_copied);
94         segment->data = NULL;
95 }
96
97 /**
98  * iscsi_tcp_segment_map - map the current S/G page
99  * @segment: iscsi_segment
100  * @recv: 1 if called from recv path
101  *
102  * We only need to possibly kmap data if scatter lists are being used,
103  * because the iscsi passthrough and internal IO paths will never use high
104  * mem pages.
105  */
106 static void iscsi_tcp_segment_map(struct iscsi_segment *segment, int recv)
107 {
108         struct scatterlist *sg;
109
110         if (segment->data != NULL || !segment->sg)
111                 return;
112
113         sg = segment->sg;
114         BUG_ON(segment->sg_mapped);
115         BUG_ON(sg->length == 0);
116
117         /*
118          * If the page count is greater than one it is ok to send
119          * to the network layer's zero copy send path. If not we
120          * have to go the slow sendmsg path. We always map for the
121          * recv path.
122          */
123         if (page_count(sg_page(sg)) >= 1 && !recv)
124                 return;
125
126         debug_tcp("iscsi_tcp_segment_map %s %p\n", recv ? "recv" : "xmit",
127                   segment);
128         segment->sg_mapped = kmap_atomic(sg_page(sg), KM_SOFTIRQ0);
129         segment->data = segment->sg_mapped + sg->offset + segment->sg_offset;
130 }
131
132 void iscsi_tcp_segment_unmap(struct iscsi_segment *segment)
133 {
134         debug_tcp("iscsi_tcp_segment_unmap %p\n", segment);
135
136         if (segment->sg_mapped) {
137                 debug_tcp("iscsi_tcp_segment_unmap valid\n");
138                 kunmap_atomic(segment->sg_mapped, KM_SOFTIRQ0);
139                 segment->sg_mapped = NULL;
140                 segment->data = NULL;
141         }
142 }
143 EXPORT_SYMBOL_GPL(iscsi_tcp_segment_unmap);
144
145 /*
146  * Splice the digest buffer into the buffer
147  */
148 static inline void
149 iscsi_tcp_segment_splice_digest(struct iscsi_segment *segment, void *digest)
150 {
151         segment->data = digest;
152         segment->digest_len = ISCSI_DIGEST_SIZE;
153         segment->total_size += ISCSI_DIGEST_SIZE;
154         segment->size = ISCSI_DIGEST_SIZE;
155         segment->copied = 0;
156         segment->sg = NULL;
157         segment->hash = NULL;
158 }
159
160 /**
161  * iscsi_tcp_segment_done - check whether the segment is complete
162  * @segment: iscsi segment to check
163  * @recv: set to one of this is called from the recv path
164  * @copied: number of bytes copied
165  *
166  * Check if we're done receiving this segment. If the receive
167  * buffer is full but we expect more data, move on to the
168  * next entry in the scatterlist.
169  *
170  * If the amount of data we received isn't a multiple of 4,
171  * we will transparently receive the pad bytes, too.
172  *
173  * This function must be re-entrant.
174  */
175 int iscsi_tcp_segment_done(struct iscsi_segment *segment, int recv,
176                            unsigned copied)
177 {
178         static unsigned char padbuf[ISCSI_PAD_LEN];
179         struct scatterlist sg;
180         unsigned int pad;
181
182         debug_tcp("copied %u %u size %u %s\n", segment->copied, copied,
183                   segment->size, recv ? "recv" : "xmit");
184         if (segment->hash && copied) {
185                 /*
186                  * If a segment is kmapd we must unmap it before sending
187                  * to the crypto layer since that will try to kmap it again.
188                  */
189                 iscsi_tcp_segment_unmap(segment);
190
191                 if (!segment->data) {
192                         sg_init_table(&sg, 1);
193                         sg_set_page(&sg, sg_page(segment->sg), copied,
194                                     segment->copied + segment->sg_offset +
195                                                         segment->sg->offset);
196                 } else
197                         sg_init_one(&sg, segment->data + segment->copied,
198                                     copied);
199                 crypto_hash_update(segment->hash, &sg, copied);
200         }
201
202         segment->copied += copied;
203         if (segment->copied < segment->size) {
204                 iscsi_tcp_segment_map(segment, recv);
205                 return 0;
206         }
207
208         segment->total_copied += segment->copied;
209         segment->copied = 0;
210         segment->size = 0;
211
212         /* Unmap the current scatterlist page, if there is one. */
213         iscsi_tcp_segment_unmap(segment);
214
215         /* Do we have more scatterlist entries? */
216         debug_tcp("total copied %u total size %u\n", segment->total_copied,
217                    segment->total_size);
218         if (segment->total_copied < segment->total_size) {
219                 /* Proceed to the next entry in the scatterlist. */
220                 iscsi_tcp_segment_init_sg(segment, sg_next(segment->sg),
221                                           0);
222                 iscsi_tcp_segment_map(segment, recv);
223                 BUG_ON(segment->size == 0);
224                 return 0;
225         }
226
227         /* Do we need to handle padding? */
228         pad = iscsi_padding(segment->total_copied);
229         if (pad != 0) {
230                 debug_tcp("consume %d pad bytes\n", pad);
231                 segment->total_size += pad;
232                 segment->size = pad;
233                 segment->data = padbuf;
234                 return 0;
235         }
236
237         /*
238          * Set us up for transferring the data digest. hdr digest
239          * is completely handled in hdr done function.
240          */
241         if (segment->hash) {
242                 crypto_hash_final(segment->hash, segment->digest);
243                 iscsi_tcp_segment_splice_digest(segment,
244                                  recv ? segment->recv_digest : segment->digest);
245                 return 0;
246         }
247
248         return 1;
249 }
250 EXPORT_SYMBOL_GPL(iscsi_tcp_segment_done);
251
252 /**
253  * iscsi_tcp_segment_recv - copy data to segment
254  * @tcp_conn: the iSCSI TCP connection
255  * @segment: the buffer to copy to
256  * @ptr: data pointer
257  * @len: amount of data available
258  *
259  * This function copies up to @len bytes to the
260  * given buffer, and returns the number of bytes
261  * consumed, which can actually be less than @len.
262  *
263  * If hash digest is enabled, the function will update the
264  * hash while copying.
265  * Combining these two operations doesn't buy us a lot (yet),
266  * but in the future we could implement combined copy+crc,
267  * just way we do for network layer checksums.
268  */
269 static int
270 iscsi_tcp_segment_recv(struct iscsi_tcp_conn *tcp_conn,
271                        struct iscsi_segment *segment, const void *ptr,
272                        unsigned int len)
273 {
274         unsigned int copy = 0, copied = 0;
275
276         while (!iscsi_tcp_segment_done(segment, 1, copy)) {
277                 if (copied == len) {
278                         debug_tcp("iscsi_tcp_segment_recv copied %d bytes\n",
279                                   len);
280                         break;
281                 }
282
283                 copy = min(len - copied, segment->size - segment->copied);
284                 debug_tcp("iscsi_tcp_segment_recv copying %d\n", copy);
285                 memcpy(segment->data + segment->copied, ptr + copied, copy);
286                 copied += copy;
287         }
288         return copied;
289 }
290
291 inline void
292 iscsi_tcp_dgst_header(struct hash_desc *hash, const void *hdr, size_t hdrlen,
293                       unsigned char digest[ISCSI_DIGEST_SIZE])
294 {
295         struct scatterlist sg;
296
297         sg_init_one(&sg, hdr, hdrlen);
298         crypto_hash_digest(hash, &sg, hdrlen, digest);
299 }
300 EXPORT_SYMBOL_GPL(iscsi_tcp_dgst_header);
301
302 static inline int
303 iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
304                       struct iscsi_segment *segment)
305 {
306         if (!segment->digest_len)
307                 return 1;
308
309         if (memcmp(segment->recv_digest, segment->digest,
310                    segment->digest_len)) {
311                 debug_scsi("digest mismatch\n");
312                 return 0;
313         }
314
315         return 1;
316 }
317
318 /*
319  * Helper function to set up segment buffer
320  */
321 static inline void
322 __iscsi_segment_init(struct iscsi_segment *segment, size_t size,
323                      iscsi_segment_done_fn_t *done, struct hash_desc *hash)
324 {
325         memset(segment, 0, sizeof(*segment));
326         segment->total_size = size;
327         segment->done = done;
328
329         if (hash) {
330                 segment->hash = hash;
331                 crypto_hash_init(hash);
332         }
333 }
334
335 inline void
336 iscsi_segment_init_linear(struct iscsi_segment *segment, void *data,
337                           size_t size, iscsi_segment_done_fn_t *done,
338                           struct hash_desc *hash)
339 {
340         __iscsi_segment_init(segment, size, done, hash);
341         segment->data = data;
342         segment->size = size;
343 }
344 EXPORT_SYMBOL_GPL(iscsi_segment_init_linear);
345
346 inline int
347 iscsi_segment_seek_sg(struct iscsi_segment *segment,
348                       struct scatterlist *sg_list, unsigned int sg_count,
349                       unsigned int offset, size_t size,
350                       iscsi_segment_done_fn_t *done, struct hash_desc *hash)
351 {
352         struct scatterlist *sg;
353         unsigned int i;
354
355         debug_scsi("iscsi_segment_seek_sg offset %u size %llu\n",
356                   offset, size);
357         __iscsi_segment_init(segment, size, done, hash);
358         for_each_sg(sg_list, sg, sg_count, i) {
359                 debug_scsi("sg %d, len %u offset %u\n", i, sg->length,
360                            sg->offset);
361                 if (offset < sg->length) {
362                         iscsi_tcp_segment_init_sg(segment, sg, offset);
363                         return 0;
364                 }
365                 offset -= sg->length;
366         }
367
368         return ISCSI_ERR_DATA_OFFSET;
369 }
370 EXPORT_SYMBOL_GPL(iscsi_segment_seek_sg);
371
372 /**
373  * iscsi_tcp_hdr_recv_prep - prep segment for hdr reception
374  * @tcp_conn: iscsi connection to prep for
375  *
376  * This function always passes NULL for the hash argument, because when this
377  * function is called we do not yet know the final size of the header and want
378  * to delay the digest processing until we know that.
379  */
380 void iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
381 {
382         debug_tcp("iscsi_tcp_hdr_recv_prep(%p%s)\n", tcp_conn,
383                   tcp_conn->iscsi_conn->hdrdgst_en ? ", digest enabled" : "");
384         iscsi_segment_init_linear(&tcp_conn->in.segment,
385                                 tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
386                                 iscsi_tcp_hdr_recv_done, NULL);
387 }
388 EXPORT_SYMBOL_GPL(iscsi_tcp_hdr_recv_prep);
389
390 /*
391  * Handle incoming reply to any other type of command
392  */
393 static int
394 iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
395                          struct iscsi_segment *segment)
396 {
397         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
398         int rc = 0;
399
400         if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
401                 return ISCSI_ERR_DATA_DGST;
402
403         rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr,
404                         conn->data, tcp_conn->in.datalen);
405         if (rc)
406                 return rc;
407
408         iscsi_tcp_hdr_recv_prep(tcp_conn);
409         return 0;
410 }
411
412 static void
413 iscsi_tcp_data_recv_prep(struct iscsi_tcp_conn *tcp_conn)
414 {
415         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
416         struct hash_desc *rx_hash = NULL;
417
418         if (conn->datadgst_en &
419             !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
420                 rx_hash = tcp_conn->rx_hash;
421
422         iscsi_segment_init_linear(&tcp_conn->in.segment,
423                                 conn->data, tcp_conn->in.datalen,
424                                 iscsi_tcp_data_recv_done, rx_hash);
425 }
426
427 /**
428  * iscsi_tcp_cleanup_task - free tcp_task resources
429  * @task: iscsi task
430  *
431  * must be called with session lock
432  */
433 void iscsi_tcp_cleanup_task(struct iscsi_task *task)
434 {
435         struct iscsi_tcp_task *tcp_task = task->dd_data;
436         struct iscsi_r2t_info *r2t;
437
438         /* nothing to do for mgmt or pending tasks */
439         if (!task->sc || task->state == ISCSI_TASK_PENDING)
440                 return;
441
442         /* flush task's r2t queues */
443         while (__kfifo_get(tcp_task->r2tqueue, (void*)&r2t, sizeof(void*))) {
444                 __kfifo_put(tcp_task->r2tpool.queue, (void*)&r2t,
445                             sizeof(void*));
446                 debug_scsi("iscsi_tcp_cleanup_task pending r2t dropped\n");
447         }
448
449         r2t = tcp_task->r2t;
450         if (r2t != NULL) {
451                 __kfifo_put(tcp_task->r2tpool.queue, (void*)&r2t,
452                             sizeof(void*));
453                 tcp_task->r2t = NULL;
454         }
455 }
456 EXPORT_SYMBOL_GPL(iscsi_tcp_cleanup_task);
457
458 /**
459  * iscsi_tcp_data_in - SCSI Data-In Response processing
460  * @conn: iscsi connection
461  * @task: scsi command task
462  */
463 static int iscsi_tcp_data_in(struct iscsi_conn *conn, struct iscsi_task *task)
464 {
465         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
466         struct iscsi_tcp_task *tcp_task = task->dd_data;
467         struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
468         int datasn = be32_to_cpu(rhdr->datasn);
469         unsigned total_in_length = scsi_in(task->sc)->length;
470
471         iscsi_update_cmdsn(conn->session, (struct iscsi_nopin*)rhdr);
472         if (tcp_conn->in.datalen == 0)
473                 return 0;
474
475         if (tcp_task->exp_datasn != datasn) {
476                 debug_tcp("%s: task->exp_datasn(%d) != rhdr->datasn(%d)\n",
477                           __func__, tcp_task->exp_datasn, datasn);
478                 return ISCSI_ERR_DATASN;
479         }
480
481         tcp_task->exp_datasn++;
482
483         tcp_task->data_offset = be32_to_cpu(rhdr->offset);
484         if (tcp_task->data_offset + tcp_conn->in.datalen > total_in_length) {
485                 debug_tcp("%s: data_offset(%d) + data_len(%d) > total_length_in(%d)\n",
486                           __func__, tcp_task->data_offset,
487                           tcp_conn->in.datalen, total_in_length);
488                 return ISCSI_ERR_DATA_OFFSET;
489         }
490
491         conn->datain_pdus_cnt++;
492         return 0;
493 }
494
495 /**
496  * iscsi_tcp_r2t_rsp - iSCSI R2T Response processing
497  * @conn: iscsi connection
498  * @task: scsi command task
499  */
500 static int iscsi_tcp_r2t_rsp(struct iscsi_conn *conn, struct iscsi_task *task)
501 {
502         struct iscsi_session *session = conn->session;
503         struct iscsi_tcp_task *tcp_task = task->dd_data;
504         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
505         struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
506         struct iscsi_r2t_info *r2t;
507         int r2tsn = be32_to_cpu(rhdr->r2tsn);
508         int rc;
509
510         if (tcp_conn->in.datalen) {
511                 iscsi_conn_printk(KERN_ERR, conn,
512                                   "invalid R2t with datalen %d\n",
513                                   tcp_conn->in.datalen);
514                 return ISCSI_ERR_DATALEN;
515         }
516
517         if (tcp_task->exp_datasn != r2tsn){
518                 debug_tcp("%s: task->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
519                           __func__, tcp_task->exp_datasn, r2tsn);
520                 return ISCSI_ERR_R2TSN;
521         }
522
523         /* fill-in new R2T associated with the task */
524         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
525
526         if (!task->sc || session->state != ISCSI_STATE_LOGGED_IN) {
527                 iscsi_conn_printk(KERN_INFO, conn,
528                                   "dropping R2T itt %d in recovery.\n",
529                                   task->itt);
530                 return 0;
531         }
532
533         rc = __kfifo_get(tcp_task->r2tpool.queue, (void*)&r2t, sizeof(void*));
534         if (!rc) {
535                 iscsi_conn_printk(KERN_ERR, conn, "Could not allocate R2T. "
536                                   "Target has sent more R2Ts than it "
537                                   "negotiated for or driver has has leaked.\n");
538                 return ISCSI_ERR_PROTO;
539         }
540
541         r2t->exp_statsn = rhdr->statsn;
542         r2t->data_length = be32_to_cpu(rhdr->data_length);
543         if (r2t->data_length == 0) {
544                 iscsi_conn_printk(KERN_ERR, conn,
545                                   "invalid R2T with zero data len\n");
546                 __kfifo_put(tcp_task->r2tpool.queue, (void*)&r2t,
547                             sizeof(void*));
548                 return ISCSI_ERR_DATALEN;
549         }
550
551         if (r2t->data_length > session->max_burst)
552                 debug_scsi("invalid R2T with data len %u and max burst %u."
553                            "Attempting to execute request.\n",
554                             r2t->data_length, session->max_burst);
555
556         r2t->data_offset = be32_to_cpu(rhdr->data_offset);
557         if (r2t->data_offset + r2t->data_length > scsi_out(task->sc)->length) {
558                 iscsi_conn_printk(KERN_ERR, conn,
559                                   "invalid R2T with data len %u at offset %u "
560                                   "and total length %d\n", r2t->data_length,
561                                   r2t->data_offset, scsi_out(task->sc)->length);
562                 __kfifo_put(tcp_task->r2tpool.queue, (void*)&r2t,
563                             sizeof(void*));
564                 return ISCSI_ERR_DATALEN;
565         }
566
567         r2t->ttt = rhdr->ttt; /* no flip */
568         r2t->datasn = 0;
569         r2t->sent = 0;
570
571         tcp_task->exp_datasn = r2tsn + 1;
572         __kfifo_put(tcp_task->r2tqueue, (void*)&r2t, sizeof(void*));
573         conn->r2t_pdus_cnt++;
574
575         iscsi_requeue_task(task);
576         return 0;
577 }
578
579 /*
580  * Handle incoming reply to DataIn command
581  */
582 static int
583 iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
584                           struct iscsi_segment *segment)
585 {
586         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
587         struct iscsi_hdr *hdr = tcp_conn->in.hdr;
588         int rc;
589
590         if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
591                 return ISCSI_ERR_DATA_DGST;
592
593         /* check for non-exceptional status */
594         if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
595                 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
596                 if (rc)
597                         return rc;
598         }
599
600         iscsi_tcp_hdr_recv_prep(tcp_conn);
601         return 0;
602 }
603
604 /**
605  * iscsi_tcp_hdr_dissect - process PDU header
606  * @conn: iSCSI connection
607  * @hdr: PDU header
608  *
609  * This function analyzes the header of the PDU received,
610  * and performs several sanity checks. If the PDU is accompanied
611  * by data, the receive buffer is set up to copy the incoming data
612  * to the correct location.
613  */
614 static int
615 iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
616 {
617         int rc = 0, opcode, ahslen;
618         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
619         struct iscsi_task *task;
620
621         /* verify PDU length */
622         tcp_conn->in.datalen = ntoh24(hdr->dlength);
623         if (tcp_conn->in.datalen > conn->max_recv_dlength) {
624                 iscsi_conn_printk(KERN_ERR, conn,
625                                   "iscsi_tcp: datalen %d > %d\n",
626                                   tcp_conn->in.datalen, conn->max_recv_dlength);
627                 return ISCSI_ERR_DATALEN;
628         }
629
630         /* Additional header segments. So far, we don't
631          * process additional headers.
632          */
633         ahslen = hdr->hlength << 2;
634
635         opcode = hdr->opcode & ISCSI_OPCODE_MASK;
636         /* verify itt (itt encoding: age+cid+itt) */
637         rc = iscsi_verify_itt(conn, hdr->itt);
638         if (rc)
639                 return rc;
640
641         debug_tcp("opcode 0x%x ahslen %d datalen %d\n",
642                   opcode, ahslen, tcp_conn->in.datalen);
643
644         switch(opcode) {
645         case ISCSI_OP_SCSI_DATA_IN:
646                 spin_lock(&conn->session->lock);
647                 task = iscsi_itt_to_ctask(conn, hdr->itt);
648                 if (!task)
649                         rc = ISCSI_ERR_BAD_ITT;
650                 else
651                         rc = iscsi_tcp_data_in(conn, task);
652                 if (rc) {
653                         spin_unlock(&conn->session->lock);
654                         break;
655                 }
656
657                 if (tcp_conn->in.datalen) {
658                         struct iscsi_tcp_task *tcp_task = task->dd_data;
659                         struct hash_desc *rx_hash = NULL;
660                         struct scsi_data_buffer *sdb = scsi_in(task->sc);
661
662                         /*
663                          * Setup copy of Data-In into the Scsi_Cmnd
664                          * Scatterlist case:
665                          * We set up the iscsi_segment to point to the next
666                          * scatterlist entry to copy to. As we go along,
667                          * we move on to the next scatterlist entry and
668                          * update the digest per-entry.
669                          */
670                         if (conn->datadgst_en &&
671                             !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
672                                 rx_hash = tcp_conn->rx_hash;
673
674                         debug_tcp("iscsi_tcp_begin_data_in(%p, offset=%d, "
675                                   "datalen=%d)\n", tcp_conn,
676                                   tcp_task->data_offset,
677                                   tcp_conn->in.datalen);
678                         rc = iscsi_segment_seek_sg(&tcp_conn->in.segment,
679                                                    sdb->table.sgl,
680                                                    sdb->table.nents,
681                                                    tcp_task->data_offset,
682                                                    tcp_conn->in.datalen,
683                                                    iscsi_tcp_process_data_in,
684                                                    rx_hash);
685                         spin_unlock(&conn->session->lock);
686                         return rc;
687                 }
688                 rc = __iscsi_complete_pdu(conn, hdr, NULL, 0);
689                 spin_unlock(&conn->session->lock);
690                 break;
691         case ISCSI_OP_SCSI_CMD_RSP:
692                 if (tcp_conn->in.datalen) {
693                         iscsi_tcp_data_recv_prep(tcp_conn);
694                         return 0;
695                 }
696                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
697                 break;
698         case ISCSI_OP_R2T:
699                 spin_lock(&conn->session->lock);
700                 task = iscsi_itt_to_ctask(conn, hdr->itt);
701                 if (!task)
702                         rc = ISCSI_ERR_BAD_ITT;
703                 else if (ahslen)
704                         rc = ISCSI_ERR_AHSLEN;
705                 else if (task->sc->sc_data_direction == DMA_TO_DEVICE)
706                         rc = iscsi_tcp_r2t_rsp(conn, task);
707                 else
708                         rc = ISCSI_ERR_PROTO;
709                 spin_unlock(&conn->session->lock);
710                 break;
711         case ISCSI_OP_LOGIN_RSP:
712         case ISCSI_OP_TEXT_RSP:
713         case ISCSI_OP_REJECT:
714         case ISCSI_OP_ASYNC_EVENT:
715                 /*
716                  * It is possible that we could get a PDU with a buffer larger
717                  * than 8K, but there are no targets that currently do this.
718                  * For now we fail until we find a vendor that needs it
719                  */
720                 if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
721                         iscsi_conn_printk(KERN_ERR, conn,
722                                           "iscsi_tcp: received buffer of "
723                                           "len %u but conn buffer is only %u "
724                                           "(opcode %0x)\n",
725                                           tcp_conn->in.datalen,
726                                           ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
727                         rc = ISCSI_ERR_PROTO;
728                         break;
729                 }
730
731                 /* If there's data coming in with the response,
732                  * receive it to the connection's buffer.
733                  */
734                 if (tcp_conn->in.datalen) {
735                         iscsi_tcp_data_recv_prep(tcp_conn);
736                         return 0;
737                 }
738         /* fall through */
739         case ISCSI_OP_LOGOUT_RSP:
740         case ISCSI_OP_NOOP_IN:
741         case ISCSI_OP_SCSI_TMFUNC_RSP:
742                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
743                 break;
744         default:
745                 rc = ISCSI_ERR_BAD_OPCODE;
746                 break;
747         }
748
749         if (rc == 0) {
750                 /* Anything that comes with data should have
751                  * been handled above. */
752                 if (tcp_conn->in.datalen)
753                         return ISCSI_ERR_PROTO;
754                 iscsi_tcp_hdr_recv_prep(tcp_conn);
755         }
756
757         return rc;
758 }
759
760 /**
761  * iscsi_tcp_hdr_recv_done - process PDU header
762  *
763  * This is the callback invoked when the PDU header has
764  * been received. If the header is followed by additional
765  * header segments, we go back for more data.
766  */
767 static int
768 iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
769                         struct iscsi_segment *segment)
770 {
771         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
772         struct iscsi_hdr *hdr;
773
774         /* Check if there are additional header segments
775          * *prior* to computing the digest, because we
776          * may need to go back to the caller for more.
777          */
778         hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
779         if (segment->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
780                 /* Bump the header length - the caller will
781                  * just loop around and get the AHS for us, and
782                  * call again. */
783                 unsigned int ahslen = hdr->hlength << 2;
784
785                 /* Make sure we don't overflow */
786                 if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
787                         return ISCSI_ERR_AHSLEN;
788
789                 segment->total_size += ahslen;
790                 segment->size += ahslen;
791                 return 0;
792         }
793
794         /* We're done processing the header. See if we're doing
795          * header digests; if so, set up the recv_digest buffer
796          * and go back for more. */
797         if (conn->hdrdgst_en) {
798                 if (segment->digest_len == 0) {
799                         /*
800                          * Even if we offload the digest processing we
801                          * splice it in so we can increment the skb/segment
802                          * counters in preparation for the data segment.
803                          */
804                         iscsi_tcp_segment_splice_digest(segment,
805                                                         segment->recv_digest);
806                         return 0;
807                 }
808
809                 if (!(conn->session->tt->caps & CAP_DIGEST_OFFLOAD)) {
810                         iscsi_tcp_dgst_header(tcp_conn->rx_hash, hdr,
811                                 segment->total_copied - ISCSI_DIGEST_SIZE,
812                                 segment->digest);
813
814                         if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
815                                 return ISCSI_ERR_HDR_DGST;
816                 }
817         }
818
819         tcp_conn->in.hdr = hdr;
820         return iscsi_tcp_hdr_dissect(conn, hdr);
821 }
822
823 /**
824  * iscsi_tcp_recv_segment_is_hdr - tests if we are reading in a header
825  * @tcp_conn: iscsi tcp conn
826  *
827  * returns non zero if we are currently processing or setup to process
828  * a header.
829  */
830 inline int iscsi_tcp_recv_segment_is_hdr(struct iscsi_tcp_conn *tcp_conn)
831 {
832         return tcp_conn->in.segment.done == iscsi_tcp_hdr_recv_done;
833 }
834 EXPORT_SYMBOL_GPL(iscsi_tcp_recv_segment_is_hdr);
835
836 /**
837  * iscsi_tcp_recv_skb - Process skb
838  * @conn: iscsi connection
839  * @skb: network buffer with header and/or data segment
840  * @offset: offset in skb
841  * @offload: bool indicating if transfer was offloaded
842  *
843  * Will return status of transfer in status. And will return
844  * number of bytes copied.
845  */
846 int iscsi_tcp_recv_skb(struct iscsi_conn *conn, struct sk_buff *skb,
847                        unsigned int offset, bool offloaded, int *status)
848 {
849         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
850         struct iscsi_segment *segment = &tcp_conn->in.segment;
851         struct skb_seq_state seq;
852         unsigned int consumed = 0;
853         int rc = 0;
854
855         debug_tcp("in %d bytes\n", skb->len - offset);
856
857         if (unlikely(conn->suspend_rx)) {
858                 debug_tcp("conn %d Rx suspended!\n", conn->id);
859                 *status = ISCSI_TCP_SUSPENDED;
860                 return 0;
861         }
862
863         if (offloaded) {
864                 segment->total_copied = segment->total_size;
865                 goto segment_done;
866         }
867
868         skb_prepare_seq_read(skb, offset, skb->len, &seq);
869         while (1) {
870                 unsigned int avail;
871                 const u8 *ptr;
872
873                 avail = skb_seq_read(consumed, &ptr, &seq);
874                 if (avail == 0) {
875                         debug_tcp("no more data avail. Consumed %d\n",
876                                   consumed);
877                         *status = ISCSI_TCP_SKB_DONE;
878                         skb_abort_seq_read(&seq);
879                         goto skb_done;
880                 }
881                 BUG_ON(segment->copied >= segment->size);
882
883                 debug_tcp("skb %p ptr=%p avail=%u\n", skb, ptr, avail);
884                 rc = iscsi_tcp_segment_recv(tcp_conn, segment, ptr, avail);
885                 BUG_ON(rc == 0);
886                 consumed += rc;
887
888                 if (segment->total_copied >= segment->total_size) {
889                         skb_abort_seq_read(&seq);
890                         goto segment_done;
891                 }
892         }
893
894 segment_done:
895         *status = ISCSI_TCP_SEGMENT_DONE;
896         debug_tcp("segment done\n");
897         rc = segment->done(tcp_conn, segment);
898         if (rc != 0) {
899                 *status = ISCSI_TCP_CONN_ERR;
900                 debug_tcp("Error receiving PDU, errno=%d\n", rc);
901                 iscsi_conn_failure(conn, rc);
902                 return 0;
903         }
904         /* The done() functions sets up the next segment. */
905
906 skb_done:
907         conn->rxdata_octets += consumed;
908         return consumed;
909 }
910 EXPORT_SYMBOL_GPL(iscsi_tcp_recv_skb);
911
912 /**
913  * iscsi_tcp_task_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
914  * @conn: iscsi connection
915  * @task: scsi command task
916  * @sc: scsi command
917  */
918 int iscsi_tcp_task_init(struct iscsi_task *task)
919 {
920         struct iscsi_tcp_task *tcp_task = task->dd_data;
921         struct iscsi_conn *conn = task->conn;
922         struct scsi_cmnd *sc = task->sc;
923         int err;
924
925         if (!sc) {
926                 /*
927                  * mgmt tasks do not have a scatterlist since they come
928                  * in from the iscsi interface.
929                  */
930                 debug_scsi("mtask deq [cid %d itt 0x%x]\n", conn->id,
931                            task->itt);
932
933                 return conn->session->tt->init_pdu(task, 0, task->data_count);
934         }
935
936         BUG_ON(__kfifo_len(tcp_task->r2tqueue));
937         tcp_task->exp_datasn = 0;
938
939         /* Prepare PDU, optionally w/ immediate data */
940         debug_scsi("task deq [cid %d itt 0x%x imm %d unsol %d]\n",
941                     conn->id, task->itt, task->imm_count,
942                     task->unsol_r2t.data_length);
943
944         err = conn->session->tt->init_pdu(task, 0, task->imm_count);
945         if (err)
946                 return err;
947         task->imm_count = 0;
948         return 0;
949 }
950 EXPORT_SYMBOL_GPL(iscsi_tcp_task_init);
951
952 static struct iscsi_r2t_info *iscsi_tcp_get_curr_r2t(struct iscsi_task *task)
953 {
954         struct iscsi_session *session = task->conn->session;
955         struct iscsi_tcp_task *tcp_task = task->dd_data;
956         struct iscsi_r2t_info *r2t = NULL;
957
958         if (iscsi_task_has_unsol_data(task))
959                 r2t = &task->unsol_r2t;
960         else {
961                 spin_lock_bh(&session->lock);
962                 if (tcp_task->r2t) {
963                         r2t = tcp_task->r2t;
964                         /* Continue with this R2T? */
965                         if (r2t->data_length <= r2t->sent) {
966                                 debug_scsi("  done with r2t %p\n", r2t);
967                                 __kfifo_put(tcp_task->r2tpool.queue,
968                                             (void *)&tcp_task->r2t,
969                                             sizeof(void *));
970                                 tcp_task->r2t = r2t = NULL;
971                         }
972                 }
973
974                 if (r2t == NULL) {
975                         __kfifo_get(tcp_task->r2tqueue,
976                                     (void *)&tcp_task->r2t, sizeof(void *));
977                         r2t = tcp_task->r2t;
978                 }
979                 spin_unlock_bh(&session->lock);
980         }
981
982         return r2t;
983 }
984
985 /**
986  * iscsi_tcp_task_xmit - xmit normal PDU task
987  * @task: iscsi command task
988  *
989  * We're expected to return 0 when everything was transmitted succesfully,
990  * -EAGAIN if there's still data in the queue, or != 0 for any other kind
991  * of error.
992  */
993 int iscsi_tcp_task_xmit(struct iscsi_task *task)
994 {
995         struct iscsi_conn *conn = task->conn;
996         struct iscsi_session *session = conn->session;
997         struct iscsi_r2t_info *r2t;
998         int rc = 0;
999
1000 flush:
1001         /* Flush any pending data first. */
1002         rc = session->tt->xmit_pdu(task);
1003         if (rc < 0)
1004                 return rc;
1005
1006         /* mgmt command */
1007         if (!task->sc) {
1008                 if (task->hdr->itt == RESERVED_ITT)
1009                         iscsi_put_task(task);
1010                 return 0;
1011         }
1012
1013         /* Are we done already? */
1014         if (task->sc->sc_data_direction != DMA_TO_DEVICE)
1015                 return 0;
1016
1017         r2t = iscsi_tcp_get_curr_r2t(task);
1018         if (r2t == NULL) {
1019                 /* Waiting for more R2Ts to arrive. */
1020                 debug_tcp("no R2Ts yet\n");
1021                 return 0;
1022         }
1023
1024         rc = conn->session->tt->alloc_pdu(task, ISCSI_OP_SCSI_DATA_OUT);
1025         if (rc)
1026                 return rc;
1027         iscsi_prep_data_out_pdu(task, r2t, (struct iscsi_data *) task->hdr);
1028
1029         debug_scsi("sol dout %p [dsn %d itt 0x%x doff %d dlen %d]\n",
1030                    r2t, r2t->datasn - 1, task->hdr->itt,
1031                    r2t->data_offset + r2t->sent, r2t->data_count);
1032
1033         rc = conn->session->tt->init_pdu(task, r2t->data_offset + r2t->sent,
1034                                          r2t->data_count);
1035         if (rc)
1036                 return rc;
1037         r2t->sent += r2t->data_count;
1038         goto flush;
1039 }
1040 EXPORT_SYMBOL_GPL(iscsi_tcp_task_xmit);
1041
1042 struct iscsi_cls_conn *
1043 iscsi_tcp_conn_setup(struct iscsi_cls_session *cls_session, int dd_data_size,
1044                       uint32_t conn_idx)
1045
1046 {
1047         struct iscsi_conn *conn;
1048         struct iscsi_cls_conn *cls_conn;
1049         struct iscsi_tcp_conn *tcp_conn;
1050
1051         cls_conn = iscsi_conn_setup(cls_session, sizeof(*tcp_conn), conn_idx);
1052         if (!cls_conn)
1053                 return NULL;
1054         conn = cls_conn->dd_data;
1055         /*
1056          * due to strange issues with iser these are not set
1057          * in iscsi_conn_setup
1058          */
1059         conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1060
1061         tcp_conn = conn->dd_data;
1062         tcp_conn->iscsi_conn = conn;
1063
1064         tcp_conn->dd_data = kzalloc(dd_data_size, GFP_KERNEL);
1065         if (!tcp_conn->dd_data) {
1066                 iscsi_conn_teardown(cls_conn);
1067                 return NULL;
1068         }
1069         return cls_conn;
1070 }
1071 EXPORT_SYMBOL_GPL(iscsi_tcp_conn_setup);
1072
1073 void iscsi_tcp_conn_teardown(struct iscsi_cls_conn *cls_conn)
1074 {
1075         struct iscsi_conn *conn = cls_conn->dd_data;
1076         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1077
1078         kfree(tcp_conn->dd_data);
1079         iscsi_conn_teardown(cls_conn);
1080 }
1081 EXPORT_SYMBOL_GPL(iscsi_tcp_conn_teardown);
1082
1083 int iscsi_tcp_r2tpool_alloc(struct iscsi_session *session)
1084 {
1085         int i;
1086         int cmd_i;
1087
1088         /*
1089          * initialize per-task: R2T pool and xmit queue
1090          */
1091         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1092                 struct iscsi_task *task = session->cmds[cmd_i];
1093                 struct iscsi_tcp_task *tcp_task = task->dd_data;
1094
1095                 /*
1096                  * pre-allocated x2 as much r2ts to handle race when
1097                  * target acks DataOut faster than we data_xmit() queues
1098                  * could replenish r2tqueue.
1099                  */
1100
1101                 /* R2T pool */
1102                 if (iscsi_pool_init(&tcp_task->r2tpool,
1103                                     session->max_r2t * 2, NULL,
1104                                     sizeof(struct iscsi_r2t_info))) {
1105                         goto r2t_alloc_fail;
1106                 }
1107
1108                 /* R2T xmit queue */
1109                 tcp_task->r2tqueue = kfifo_alloc(
1110                       session->max_r2t * 4 * sizeof(void*), GFP_KERNEL, NULL);
1111                 if (tcp_task->r2tqueue == ERR_PTR(-ENOMEM)) {
1112                         iscsi_pool_free(&tcp_task->r2tpool);
1113                         goto r2t_alloc_fail;
1114                 }
1115         }
1116
1117         return 0;
1118
1119 r2t_alloc_fail:
1120         for (i = 0; i < cmd_i; i++) {
1121                 struct iscsi_task *task = session->cmds[i];
1122                 struct iscsi_tcp_task *tcp_task = task->dd_data;
1123
1124                 kfifo_free(tcp_task->r2tqueue);
1125                 iscsi_pool_free(&tcp_task->r2tpool);
1126         }
1127         return -ENOMEM;
1128 }
1129 EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_alloc);
1130
1131 void iscsi_tcp_r2tpool_free(struct iscsi_session *session)
1132 {
1133         int i;
1134
1135         for (i = 0; i < session->cmds_max; i++) {
1136                 struct iscsi_task *task = session->cmds[i];
1137                 struct iscsi_tcp_task *tcp_task = task->dd_data;
1138
1139                 kfifo_free(tcp_task->r2tqueue);
1140                 iscsi_pool_free(&tcp_task->r2tpool);
1141         }
1142 }
1143 EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_free);
1144
1145 void iscsi_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
1146                               struct iscsi_stats *stats)
1147 {
1148         struct iscsi_conn *conn = cls_conn->dd_data;
1149
1150         stats->txdata_octets = conn->txdata_octets;
1151         stats->rxdata_octets = conn->rxdata_octets;
1152         stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
1153         stats->dataout_pdus = conn->dataout_pdus_cnt;
1154         stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
1155         stats->datain_pdus = conn->datain_pdus_cnt;
1156         stats->r2t_pdus = conn->r2t_pdus_cnt;
1157         stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
1158         stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
1159 }
1160 EXPORT_SYMBOL_GPL(iscsi_tcp_conn_get_stats);