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/slab.h>
33#include <linux/file.h>
34#include <linux/blkdev.h>
35#include <linux/crypto.h>
36#include <linux/delay.h>
37#include <linux/kfifo.h>
38#include <linux/scatterlist.h>
39#include <linux/module.h>
40#include <net/tcp.h>
41#include <scsi/scsi_cmnd.h>
42#include <scsi/scsi_device.h>
43#include <scsi/scsi_host.h>
44#include <scsi/scsi.h>
45#include <scsi/scsi_transport_iscsi.h>
46
47#include "iscsi_tcp.h"
48
49MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
50	      "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
51	      "Alex Aizman <itn780@yahoo.com>");
52MODULE_DESCRIPTION("iSCSI/TCP data-path");
53MODULE_LICENSE("GPL");
54
55static int iscsi_dbg_libtcp;
56module_param_named(debug_libiscsi_tcp, iscsi_dbg_libtcp, int,
57		   S_IRUGO | S_IWUSR);
58MODULE_PARM_DESC(debug_libiscsi_tcp, "Turn on debugging for libiscsi_tcp "
59		 "module. Set to 1 to turn on, and zero to turn off. Default "
60		 "is off.");
61
62#define ISCSI_DBG_TCP(_conn, dbg_fmt, arg...)			\
63	do {							\
64		if (iscsi_dbg_libtcp)				\
65			iscsi_conn_printk(KERN_INFO, _conn,	\
66					     "%s " dbg_fmt,	\
67					     __func__, ##arg);	\
68	} while (0);
69
70static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
71				   struct iscsi_segment *segment);
72
73/*
74 * Scatterlist handling: inside the iscsi_segment, we
75 * remember an index into the scatterlist, and set data/size
76 * to the current scatterlist entry. For highmem pages, we
77 * kmap as needed.
78 *
79 * Note that the page is unmapped when we return from
80 * TCP's data_ready handler, so we may end up mapping and
81 * unmapping the same page repeatedly. The whole reason
82 * for this is that we shouldn't keep the page mapped
83 * outside the softirq.
84 */
85
86/**
87 * iscsi_tcp_segment_init_sg - init indicated scatterlist entry
88 * @segment: the buffer object
89 * @sg: scatterlist
90 * @offset: byte offset into that sg entry
91 *
92 * This function sets up the segment so that subsequent
93 * data is copied to the indicated sg entry, at the given
94 * offset.
95 */
96static inline void
97iscsi_tcp_segment_init_sg(struct iscsi_segment *segment,
98			  struct scatterlist *sg, unsigned int offset)
99{
100	segment->sg = sg;
101	segment->sg_offset = offset;
102	segment->size = min(sg->length - offset,
103			    segment->total_size - segment->total_copied);
104	segment->data = NULL;
105}
106
107/**
108 * iscsi_tcp_segment_map - map the current S/G page
109 * @segment: iscsi_segment
110 * @recv: 1 if called from recv path
111 *
112 * We only need to possibly kmap data if scatter lists are being used,
113 * because the iscsi passthrough and internal IO paths will never use high
114 * mem pages.
115 */
116static void iscsi_tcp_segment_map(struct iscsi_segment *segment, int recv)
117{
118	struct scatterlist *sg;
119
120	if (segment->data != NULL || !segment->sg)
121		return;
122
123	sg = segment->sg;
124	BUG_ON(segment->sg_mapped);
125	BUG_ON(sg->length == 0);
126
127	/*
128	 * If the page count is greater than one it is ok to send
129	 * to the network layer's zero copy send path. If not we
130	 * have to go the slow sendmsg path. We always map for the
131	 * recv path.
132	 */
133	if (page_count(sg_page(sg)) >= 1 && !recv)
134		return;
135
136	if (recv) {
137		segment->atomic_mapped = true;
138		segment->sg_mapped = kmap_atomic(sg_page(sg));
139	} else {
140		segment->atomic_mapped = false;
141		/* the xmit path can sleep with the page mapped so use kmap */
142		segment->sg_mapped = kmap(sg_page(sg));
143	}
144
145	segment->data = segment->sg_mapped + sg->offset + segment->sg_offset;
146}
147
148void iscsi_tcp_segment_unmap(struct iscsi_segment *segment)
149{
150	if (segment->sg_mapped) {
151		if (segment->atomic_mapped)
152			kunmap_atomic(segment->sg_mapped);
153		else
154			kunmap(sg_page(segment->sg));
155		segment->sg_mapped = NULL;
156		segment->data = NULL;
157	}
158}
159EXPORT_SYMBOL_GPL(iscsi_tcp_segment_unmap);
160
161/*
162 * Splice the digest buffer into the buffer
163 */
164static inline void
165iscsi_tcp_segment_splice_digest(struct iscsi_segment *segment, void *digest)
166{
167	segment->data = digest;
168	segment->digest_len = ISCSI_DIGEST_SIZE;
169	segment->total_size += ISCSI_DIGEST_SIZE;
170	segment->size = ISCSI_DIGEST_SIZE;
171	segment->copied = 0;
172	segment->sg = NULL;
173	segment->hash = NULL;
174}
175
176/**
177 * iscsi_tcp_segment_done - check whether the segment is complete
178 * @tcp_conn: iscsi tcp connection
179 * @segment: iscsi segment to check
180 * @recv: set to one of this is called from the recv path
181 * @copied: number of bytes copied
182 *
183 * Check if we're done receiving this segment. If the receive
184 * buffer is full but we expect more data, move on to the
185 * next entry in the scatterlist.
186 *
187 * If the amount of data we received isn't a multiple of 4,
188 * we will transparently receive the pad bytes, too.
189 *
190 * This function must be re-entrant.
191 */
192int iscsi_tcp_segment_done(struct iscsi_tcp_conn *tcp_conn,
193			   struct iscsi_segment *segment, int recv,
194			   unsigned copied)
195{
196	struct scatterlist sg;
197	unsigned int pad;
198
199	ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copied %u %u size %u %s\n",
200		      segment->copied, copied, segment->size,
201		      recv ? "recv" : "xmit");
202	if (segment->hash && copied) {
203		/*
204		 * If a segment is kmapd we must unmap it before sending
205		 * to the crypto layer since that will try to kmap it again.
206		 */
207		iscsi_tcp_segment_unmap(segment);
208
209		if (!segment->data) {
210			sg_init_table(&sg, 1);
211			sg_set_page(&sg, sg_page(segment->sg), copied,
212				    segment->copied + segment->sg_offset +
213							segment->sg->offset);
214		} else
215			sg_init_one(&sg, segment->data + segment->copied,
216				    copied);
217		crypto_hash_update(segment->hash, &sg, copied);
218	}
219
220	segment->copied += copied;
221	if (segment->copied < segment->size) {
222		iscsi_tcp_segment_map(segment, recv);
223		return 0;
224	}
225
226	segment->total_copied += segment->copied;
227	segment->copied = 0;
228	segment->size = 0;
229
230	/* Unmap the current scatterlist page, if there is one. */
231	iscsi_tcp_segment_unmap(segment);
232
233	/* Do we have more scatterlist entries? */
234	ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "total copied %u total size %u\n",
235		      segment->total_copied, segment->total_size);
236	if (segment->total_copied < segment->total_size) {
237		/* Proceed to the next entry in the scatterlist. */
238		iscsi_tcp_segment_init_sg(segment, sg_next(segment->sg),
239					  0);
240		iscsi_tcp_segment_map(segment, recv);
241		BUG_ON(segment->size == 0);
242		return 0;
243	}
244
245	/* Do we need to handle padding? */
246	if (!(tcp_conn->iscsi_conn->session->tt->caps & CAP_PADDING_OFFLOAD)) {
247		pad = iscsi_padding(segment->total_copied);
248		if (pad != 0) {
249			ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
250				      "consume %d pad bytes\n", pad);
251			segment->total_size += pad;
252			segment->size = pad;
253			segment->data = segment->padbuf;
254			return 0;
255		}
256	}
257
258	/*
259	 * Set us up for transferring the data digest. hdr digest
260	 * is completely handled in hdr done function.
261	 */
262	if (segment->hash) {
263		crypto_hash_final(segment->hash, segment->digest);
264		iscsi_tcp_segment_splice_digest(segment,
265				 recv ? segment->recv_digest : segment->digest);
266		return 0;
267	}
268
269	return 1;
270}
271EXPORT_SYMBOL_GPL(iscsi_tcp_segment_done);
272
273/**
274 * iscsi_tcp_segment_recv - copy data to segment
275 * @tcp_conn: the iSCSI TCP connection
276 * @segment: the buffer to copy to
277 * @ptr: data pointer
278 * @len: amount of data available
279 *
280 * This function copies up to @len bytes to the
281 * given buffer, and returns the number of bytes
282 * consumed, which can actually be less than @len.
283 *
284 * If hash digest is enabled, the function will update the
285 * hash while copying.
286 * Combining these two operations doesn't buy us a lot (yet),
287 * but in the future we could implement combined copy+crc,
288 * just way we do for network layer checksums.
289 */
290static int
291iscsi_tcp_segment_recv(struct iscsi_tcp_conn *tcp_conn,
292		       struct iscsi_segment *segment, const void *ptr,
293		       unsigned int len)
294{
295	unsigned int copy = 0, copied = 0;
296
297	while (!iscsi_tcp_segment_done(tcp_conn, segment, 1, copy)) {
298		if (copied == len) {
299			ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
300				      "copied %d bytes\n", len);
301			break;
302		}
303
304		copy = min(len - copied, segment->size - segment->copied);
305		ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copying %d\n", copy);
306		memcpy(segment->data + segment->copied, ptr + copied, copy);
307		copied += copy;
308	}
309	return copied;
310}
311
312inline void
313iscsi_tcp_dgst_header(struct hash_desc *hash, const void *hdr, size_t hdrlen,
314		      unsigned char digest[ISCSI_DIGEST_SIZE])
315{
316	struct scatterlist sg;
317
318	sg_init_one(&sg, hdr, hdrlen);
319	crypto_hash_digest(hash, &sg, hdrlen, digest);
320}
321EXPORT_SYMBOL_GPL(iscsi_tcp_dgst_header);
322
323static inline int
324iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
325		      struct iscsi_segment *segment)
326{
327	if (!segment->digest_len)
328		return 1;
329
330	if (memcmp(segment->recv_digest, segment->digest,
331		   segment->digest_len)) {
332		ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "digest mismatch\n");
333		return 0;
334	}
335
336	return 1;
337}
338
339/*
340 * Helper function to set up segment buffer
341 */
342static inline void
343__iscsi_segment_init(struct iscsi_segment *segment, size_t size,
344		     iscsi_segment_done_fn_t *done, struct hash_desc *hash)
345{
346	memset(segment, 0, sizeof(*segment));
347	segment->total_size = size;
348	segment->done = done;
349
350	if (hash) {
351		segment->hash = hash;
352		crypto_hash_init(hash);
353	}
354}
355
356inline void
357iscsi_segment_init_linear(struct iscsi_segment *segment, void *data,
358			  size_t size, iscsi_segment_done_fn_t *done,
359			  struct hash_desc *hash)
360{
361	__iscsi_segment_init(segment, size, done, hash);
362	segment->data = data;
363	segment->size = size;
364}
365EXPORT_SYMBOL_GPL(iscsi_segment_init_linear);
366
367inline int
368iscsi_segment_seek_sg(struct iscsi_segment *segment,
369		      struct scatterlist *sg_list, unsigned int sg_count,
370		      unsigned int offset, size_t size,
371		      iscsi_segment_done_fn_t *done, struct hash_desc *hash)
372{
373	struct scatterlist *sg;
374	unsigned int i;
375
376	__iscsi_segment_init(segment, size, done, hash);
377	for_each_sg(sg_list, sg, sg_count, i) {
378		if (offset < sg->length) {
379			iscsi_tcp_segment_init_sg(segment, sg, offset);
380			return 0;
381		}
382		offset -= sg->length;
383	}
384
385	return ISCSI_ERR_DATA_OFFSET;
386}
387EXPORT_SYMBOL_GPL(iscsi_segment_seek_sg);
388
389/**
390 * iscsi_tcp_hdr_recv_prep - prep segment 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 */
397void iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
398{
399	ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
400		      "(%s)\n", tcp_conn->iscsi_conn->hdrdgst_en ?
401		      "digest enabled" : "digest disabled");
402	iscsi_segment_init_linear(&tcp_conn->in.segment,
403				tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
404				iscsi_tcp_hdr_recv_done, NULL);
405}
406EXPORT_SYMBOL_GPL(iscsi_tcp_hdr_recv_prep);
407
408/*
409 * Handle incoming reply to any other type of command
410 */
411static int
412iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
413			 struct iscsi_segment *segment)
414{
415	struct iscsi_conn *conn = tcp_conn->iscsi_conn;
416	int rc = 0;
417
418	if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
419		return ISCSI_ERR_DATA_DGST;
420
421	rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr,
422			conn->data, tcp_conn->in.datalen);
423	if (rc)
424		return rc;
425
426	iscsi_tcp_hdr_recv_prep(tcp_conn);
427	return 0;
428}
429
430static void
431iscsi_tcp_data_recv_prep(struct iscsi_tcp_conn *tcp_conn)
432{
433	struct iscsi_conn *conn = tcp_conn->iscsi_conn;
434	struct hash_desc *rx_hash = NULL;
435
436	if (conn->datadgst_en &&
437	    !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
438		rx_hash = tcp_conn->rx_hash;
439
440	iscsi_segment_init_linear(&tcp_conn->in.segment,
441				conn->data, tcp_conn->in.datalen,
442				iscsi_tcp_data_recv_done, rx_hash);
443}
444
445/**
446 * iscsi_tcp_cleanup_task - free tcp_task resources
447 * @task: iscsi task
448 *
449 * must be called with session back_lock
450 */
451void iscsi_tcp_cleanup_task(struct iscsi_task *task)
452{
453	struct iscsi_tcp_task *tcp_task = task->dd_data;
454	struct iscsi_r2t_info *r2t;
455
456	/* nothing to do for mgmt */
457	if (!task->sc)
458		return;
459
460	spin_lock_bh(&tcp_task->queue2pool);
461	/* flush task's r2t queues */
462	while (kfifo_out(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*))) {
463		kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
464			    sizeof(void*));
465		ISCSI_DBG_TCP(task->conn, "pending r2t dropped\n");
466	}
467
468	r2t = tcp_task->r2t;
469	if (r2t != NULL) {
470		kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
471			    sizeof(void*));
472		tcp_task->r2t = NULL;
473	}
474	spin_unlock_bh(&tcp_task->queue2pool);
475}
476EXPORT_SYMBOL_GPL(iscsi_tcp_cleanup_task);
477
478/**
479 * iscsi_tcp_data_in - SCSI Data-In Response processing
480 * @conn: iscsi connection
481 * @task: scsi command task
482 */
483static int iscsi_tcp_data_in(struct iscsi_conn *conn, struct iscsi_task *task)
484{
485	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
486	struct iscsi_tcp_task *tcp_task = task->dd_data;
487	struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
488	int datasn = be32_to_cpu(rhdr->datasn);
489	unsigned total_in_length = scsi_in(task->sc)->length;
490
491	/*
492	 * lib iscsi will update this in the completion handling if there
493	 * is status.
494	 */
495	if (!(rhdr->flags & ISCSI_FLAG_DATA_STATUS))
496		iscsi_update_cmdsn(conn->session, (struct iscsi_nopin*)rhdr);
497
498	if (tcp_conn->in.datalen == 0)
499		return 0;
500
501	if (tcp_task->exp_datasn != datasn) {
502		ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->datasn(%d)"
503			      "\n", tcp_task->exp_datasn, datasn);
504		return ISCSI_ERR_DATASN;
505	}
506
507	tcp_task->exp_datasn++;
508
509	tcp_task->data_offset = be32_to_cpu(rhdr->offset);
510	if (tcp_task->data_offset + tcp_conn->in.datalen > total_in_length) {
511		ISCSI_DBG_TCP(conn, "data_offset(%d) + data_len(%d) > "
512			      "total_length_in(%d)\n", tcp_task->data_offset,
513			      tcp_conn->in.datalen, total_in_length);
514		return ISCSI_ERR_DATA_OFFSET;
515	}
516
517	conn->datain_pdus_cnt++;
518	return 0;
519}
520
521/**
522 * iscsi_tcp_r2t_rsp - iSCSI R2T Response processing
523 * @conn: iscsi connection
524 * @task: scsi command task
525 */
526static int iscsi_tcp_r2t_rsp(struct iscsi_conn *conn, struct iscsi_task *task)
527{
528	struct iscsi_session *session = conn->session;
529	struct iscsi_tcp_task *tcp_task = task->dd_data;
530	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
531	struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
532	struct iscsi_r2t_info *r2t;
533	int r2tsn = be32_to_cpu(rhdr->r2tsn);
534	u32 data_length;
535	u32 data_offset;
536	int rc;
537
538	if (tcp_conn->in.datalen) {
539		iscsi_conn_printk(KERN_ERR, conn,
540				  "invalid R2t with datalen %d\n",
541				  tcp_conn->in.datalen);
542		return ISCSI_ERR_DATALEN;
543	}
544
545	if (tcp_task->exp_datasn != r2tsn){
546		ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
547			      tcp_task->exp_datasn, r2tsn);
548		return ISCSI_ERR_R2TSN;
549	}
550
551	/* fill-in new R2T associated with the task */
552	iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
553
554	if (!task->sc || session->state != ISCSI_STATE_LOGGED_IN) {
555		iscsi_conn_printk(KERN_INFO, conn,
556				  "dropping R2T itt %d in recovery.\n",
557				  task->itt);
558		return 0;
559	}
560
561	data_length = be32_to_cpu(rhdr->data_length);
562	if (data_length == 0) {
563		iscsi_conn_printk(KERN_ERR, conn,
564				  "invalid R2T with zero data len\n");
565		return ISCSI_ERR_DATALEN;
566	}
567
568	if (data_length > session->max_burst)
569		ISCSI_DBG_TCP(conn, "invalid R2T with data len %u and max "
570			      "burst %u. Attempting to execute request.\n",
571			      data_length, session->max_burst);
572
573	data_offset = be32_to_cpu(rhdr->data_offset);
574	if (data_offset + data_length > scsi_out(task->sc)->length) {
575		iscsi_conn_printk(KERN_ERR, conn,
576				  "invalid R2T with data len %u at offset %u "
577				  "and total length %d\n", data_length,
578				  data_offset, scsi_out(task->sc)->length);
579		return ISCSI_ERR_DATALEN;
580	}
581
582	spin_lock(&tcp_task->pool2queue);
583	rc = kfifo_out(&tcp_task->r2tpool.queue, (void *)&r2t, sizeof(void *));
584	if (!rc) {
585		iscsi_conn_printk(KERN_ERR, conn, "Could not allocate R2T. "
586				  "Target has sent more R2Ts than it "
587				  "negotiated for or driver has leaked.\n");
588		spin_unlock(&tcp_task->pool2queue);
589		return ISCSI_ERR_PROTO;
590	}
591
592	r2t->exp_statsn = rhdr->statsn;
593	r2t->data_length = data_length;
594	r2t->data_offset = data_offset;
595
596	r2t->ttt = rhdr->ttt; /* no flip */
597	r2t->datasn = 0;
598	r2t->sent = 0;
599
600	tcp_task->exp_datasn = r2tsn + 1;
601	kfifo_in(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*));
602	conn->r2t_pdus_cnt++;
603	spin_unlock(&tcp_task->pool2queue);
604
605	iscsi_requeue_task(task);
606	return 0;
607}
608
609/*
610 * Handle incoming reply to DataIn command
611 */
612static int
613iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
614			  struct iscsi_segment *segment)
615{
616	struct iscsi_conn *conn = tcp_conn->iscsi_conn;
617	struct iscsi_hdr *hdr = tcp_conn->in.hdr;
618	int rc;
619
620	if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
621		return ISCSI_ERR_DATA_DGST;
622
623	/* check for non-exceptional status */
624	if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
625		rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
626		if (rc)
627			return rc;
628	}
629
630	iscsi_tcp_hdr_recv_prep(tcp_conn);
631	return 0;
632}
633
634/**
635 * iscsi_tcp_hdr_dissect - process PDU header
636 * @conn: iSCSI connection
637 * @hdr: PDU header
638 *
639 * This function analyzes the header of the PDU received,
640 * and performs several sanity checks. If the PDU is accompanied
641 * by data, the receive buffer is set up to copy the incoming data
642 * to the correct location.
643 */
644static int
645iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
646{
647	int rc = 0, opcode, ahslen;
648	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
649	struct iscsi_task *task;
650
651	/* verify PDU length */
652	tcp_conn->in.datalen = ntoh24(hdr->dlength);
653	if (tcp_conn->in.datalen > conn->max_recv_dlength) {
654		iscsi_conn_printk(KERN_ERR, conn,
655				  "iscsi_tcp: datalen %d > %d\n",
656				  tcp_conn->in.datalen, conn->max_recv_dlength);
657		return ISCSI_ERR_DATALEN;
658	}
659
660	/* Additional header segments. So far, we don't
661	 * process additional headers.
662	 */
663	ahslen = hdr->hlength << 2;
664
665	opcode = hdr->opcode & ISCSI_OPCODE_MASK;
666	/* verify itt (itt encoding: age+cid+itt) */
667	rc = iscsi_verify_itt(conn, hdr->itt);
668	if (rc)
669		return rc;
670
671	ISCSI_DBG_TCP(conn, "opcode 0x%x ahslen %d datalen %d\n",
672		      opcode, ahslen, tcp_conn->in.datalen);
673
674	switch(opcode) {
675	case ISCSI_OP_SCSI_DATA_IN:
676		spin_lock(&conn->session->back_lock);
677		task = iscsi_itt_to_ctask(conn, hdr->itt);
678		if (!task)
679			rc = ISCSI_ERR_BAD_ITT;
680		else
681			rc = iscsi_tcp_data_in(conn, task);
682		if (rc) {
683			spin_unlock(&conn->session->back_lock);
684			break;
685		}
686
687		if (tcp_conn->in.datalen) {
688			struct iscsi_tcp_task *tcp_task = task->dd_data;
689			struct hash_desc *rx_hash = NULL;
690			struct scsi_data_buffer *sdb = scsi_in(task->sc);
691
692			/*
693			 * Setup copy of Data-In into the Scsi_Cmnd
694			 * Scatterlist case:
695			 * We set up the iscsi_segment to point to the next
696			 * scatterlist entry to copy to. As we go along,
697			 * we move on to the next scatterlist entry and
698			 * update the digest per-entry.
699			 */
700			if (conn->datadgst_en &&
701			    !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
702				rx_hash = tcp_conn->rx_hash;
703
704			ISCSI_DBG_TCP(conn, "iscsi_tcp_begin_data_in( "
705				     "offset=%d, datalen=%d)\n",
706				      tcp_task->data_offset,
707				      tcp_conn->in.datalen);
708			task->last_xfer = jiffies;
709			rc = iscsi_segment_seek_sg(&tcp_conn->in.segment,
710						   sdb->table.sgl,
711						   sdb->table.nents,
712						   tcp_task->data_offset,
713						   tcp_conn->in.datalen,
714						   iscsi_tcp_process_data_in,
715						   rx_hash);
716			spin_unlock(&conn->session->back_lock);
717			return rc;
718		}
719		rc = __iscsi_complete_pdu(conn, hdr, NULL, 0);
720		spin_unlock(&conn->session->back_lock);
721		break;
722	case ISCSI_OP_SCSI_CMD_RSP:
723		if (tcp_conn->in.datalen) {
724			iscsi_tcp_data_recv_prep(tcp_conn);
725			return 0;
726		}
727		rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
728		break;
729	case ISCSI_OP_R2T:
730		spin_lock(&conn->session->back_lock);
731		task = iscsi_itt_to_ctask(conn, hdr->itt);
732		spin_unlock(&conn->session->back_lock);
733		if (!task)
734			rc = ISCSI_ERR_BAD_ITT;
735		else if (ahslen)
736			rc = ISCSI_ERR_AHSLEN;
737		else if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
738			task->last_xfer = jiffies;
739			spin_lock(&conn->session->frwd_lock);
740			rc = iscsi_tcp_r2t_rsp(conn, task);
741			spin_unlock(&conn->session->frwd_lock);
742		} else
743			rc = ISCSI_ERR_PROTO;
744		break;
745	case ISCSI_OP_LOGIN_RSP:
746	case ISCSI_OP_TEXT_RSP:
747	case ISCSI_OP_REJECT:
748	case ISCSI_OP_ASYNC_EVENT:
749		/*
750		 * It is possible that we could get a PDU with a buffer larger
751		 * than 8K, but there are no targets that currently do this.
752		 * For now we fail until we find a vendor that needs it
753		 */
754		if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
755			iscsi_conn_printk(KERN_ERR, conn,
756					  "iscsi_tcp: received buffer of "
757					  "len %u but conn buffer is only %u "
758					  "(opcode %0x)\n",
759					  tcp_conn->in.datalen,
760					  ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
761			rc = ISCSI_ERR_PROTO;
762			break;
763		}
764
765		/* If there's data coming in with the response,
766		 * receive it to the connection's buffer.
767		 */
768		if (tcp_conn->in.datalen) {
769			iscsi_tcp_data_recv_prep(tcp_conn);
770			return 0;
771		}
772	/* fall through */
773	case ISCSI_OP_LOGOUT_RSP:
774	case ISCSI_OP_NOOP_IN:
775	case ISCSI_OP_SCSI_TMFUNC_RSP:
776		rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
777		break;
778	default:
779		rc = ISCSI_ERR_BAD_OPCODE;
780		break;
781	}
782
783	if (rc == 0) {
784		/* Anything that comes with data should have
785		 * been handled above. */
786		if (tcp_conn->in.datalen)
787			return ISCSI_ERR_PROTO;
788		iscsi_tcp_hdr_recv_prep(tcp_conn);
789	}
790
791	return rc;
792}
793
794/**
795 * iscsi_tcp_hdr_recv_done - process PDU header
796 *
797 * This is the callback invoked when the PDU header has
798 * been received. If the header is followed by additional
799 * header segments, we go back for more data.
800 */
801static int
802iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
803			struct iscsi_segment *segment)
804{
805	struct iscsi_conn *conn = tcp_conn->iscsi_conn;
806	struct iscsi_hdr *hdr;
807
808	/* Check if there are additional header segments
809	 * *prior* to computing the digest, because we
810	 * may need to go back to the caller for more.
811	 */
812	hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
813	if (segment->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
814		/* Bump the header length - the caller will
815		 * just loop around and get the AHS for us, and
816		 * call again. */
817		unsigned int ahslen = hdr->hlength << 2;
818
819		/* Make sure we don't overflow */
820		if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
821			return ISCSI_ERR_AHSLEN;
822
823		segment->total_size += ahslen;
824		segment->size += ahslen;
825		return 0;
826	}
827
828	/* We're done processing the header. See if we're doing
829	 * header digests; if so, set up the recv_digest buffer
830	 * and go back for more. */
831	if (conn->hdrdgst_en &&
832	    !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD)) {
833		if (segment->digest_len == 0) {
834			/*
835			 * Even if we offload the digest processing we
836			 * splice it in so we can increment the skb/segment
837			 * counters in preparation for the data segment.
838			 */
839			iscsi_tcp_segment_splice_digest(segment,
840							segment->recv_digest);
841			return 0;
842		}
843
844		iscsi_tcp_dgst_header(tcp_conn->rx_hash, hdr,
845				      segment->total_copied - ISCSI_DIGEST_SIZE,
846				      segment->digest);
847
848		if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
849			return ISCSI_ERR_HDR_DGST;
850	}
851
852	tcp_conn->in.hdr = hdr;
853	return iscsi_tcp_hdr_dissect(conn, hdr);
854}
855
856/**
857 * iscsi_tcp_recv_segment_is_hdr - tests if we are reading in a header
858 * @tcp_conn: iscsi tcp conn
859 *
860 * returns non zero if we are currently processing or setup to process
861 * a header.
862 */
863inline int iscsi_tcp_recv_segment_is_hdr(struct iscsi_tcp_conn *tcp_conn)
864{
865	return tcp_conn->in.segment.done == iscsi_tcp_hdr_recv_done;
866}
867EXPORT_SYMBOL_GPL(iscsi_tcp_recv_segment_is_hdr);
868
869/**
870 * iscsi_tcp_recv_skb - Process skb
871 * @conn: iscsi connection
872 * @skb: network buffer with header and/or data segment
873 * @offset: offset in skb
874 * @offload: bool indicating if transfer was offloaded
875 *
876 * Will return status of transfer in status. And will return
877 * number of bytes copied.
878 */
879int iscsi_tcp_recv_skb(struct iscsi_conn *conn, struct sk_buff *skb,
880		       unsigned int offset, bool offloaded, int *status)
881{
882	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
883	struct iscsi_segment *segment = &tcp_conn->in.segment;
884	struct skb_seq_state seq;
885	unsigned int consumed = 0;
886	int rc = 0;
887
888	ISCSI_DBG_TCP(conn, "in %d bytes\n", skb->len - offset);
889	/*
890	 * Update for each skb instead of pdu, because over slow networks a
891	 * data_in's data could take a while to read in. We also want to
892	 * account for r2ts.
893	 */
894	conn->last_recv = jiffies;
895
896	if (unlikely(conn->suspend_rx)) {
897		ISCSI_DBG_TCP(conn, "Rx suspended!\n");
898		*status = ISCSI_TCP_SUSPENDED;
899		return 0;
900	}
901
902	if (offloaded) {
903		segment->total_copied = segment->total_size;
904		goto segment_done;
905	}
906
907	skb_prepare_seq_read(skb, offset, skb->len, &seq);
908	while (1) {
909		unsigned int avail;
910		const u8 *ptr;
911
912		avail = skb_seq_read(consumed, &ptr, &seq);
913		if (avail == 0) {
914			ISCSI_DBG_TCP(conn, "no more data avail. Consumed %d\n",
915				      consumed);
916			*status = ISCSI_TCP_SKB_DONE;
917			goto skb_done;
918		}
919		BUG_ON(segment->copied >= segment->size);
920
921		ISCSI_DBG_TCP(conn, "skb %p ptr=%p avail=%u\n", skb, ptr,
922			      avail);
923		rc = iscsi_tcp_segment_recv(tcp_conn, segment, ptr, avail);
924		BUG_ON(rc == 0);
925		consumed += rc;
926
927		if (segment->total_copied >= segment->total_size) {
928			skb_abort_seq_read(&seq);
929			goto segment_done;
930		}
931	}
932
933segment_done:
934	*status = ISCSI_TCP_SEGMENT_DONE;
935	ISCSI_DBG_TCP(conn, "segment done\n");
936	rc = segment->done(tcp_conn, segment);
937	if (rc != 0) {
938		*status = ISCSI_TCP_CONN_ERR;
939		ISCSI_DBG_TCP(conn, "Error receiving PDU, errno=%d\n", rc);
940		iscsi_conn_failure(conn, rc);
941		return 0;
942	}
943	/* The done() functions sets up the next segment. */
944
945skb_done:
946	conn->rxdata_octets += consumed;
947	return consumed;
948}
949EXPORT_SYMBOL_GPL(iscsi_tcp_recv_skb);
950
951/**
952 * iscsi_tcp_task_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
953 * @conn: iscsi connection
954 * @task: scsi command task
955 * @sc: scsi command
956 */
957int iscsi_tcp_task_init(struct iscsi_task *task)
958{
959	struct iscsi_tcp_task *tcp_task = task->dd_data;
960	struct iscsi_conn *conn = task->conn;
961	struct scsi_cmnd *sc = task->sc;
962	int err;
963
964	if (!sc) {
965		/*
966		 * mgmt tasks do not have a scatterlist since they come
967		 * in from the iscsi interface.
968		 */
969		ISCSI_DBG_TCP(conn, "mtask deq [itt 0x%x]\n", task->itt);
970
971		return conn->session->tt->init_pdu(task, 0, task->data_count);
972	}
973
974	BUG_ON(kfifo_len(&tcp_task->r2tqueue));
975	tcp_task->exp_datasn = 0;
976
977	/* Prepare PDU, optionally w/ immediate data */
978	ISCSI_DBG_TCP(conn, "task deq [itt 0x%x imm %d unsol %d]\n",
979		      task->itt, task->imm_count, task->unsol_r2t.data_length);
980
981	err = conn->session->tt->init_pdu(task, 0, task->imm_count);
982	if (err)
983		return err;
984	task->imm_count = 0;
985	return 0;
986}
987EXPORT_SYMBOL_GPL(iscsi_tcp_task_init);
988
989static struct iscsi_r2t_info *iscsi_tcp_get_curr_r2t(struct iscsi_task *task)
990{
991	struct iscsi_tcp_task *tcp_task = task->dd_data;
992	struct iscsi_r2t_info *r2t = NULL;
993
994	if (iscsi_task_has_unsol_data(task))
995		r2t = &task->unsol_r2t;
996	else {
997		spin_lock_bh(&tcp_task->queue2pool);
998		if (tcp_task->r2t) {
999			r2t = tcp_task->r2t;
1000			/* Continue with this R2T? */
1001			if (r2t->data_length <= r2t->sent) {
1002				ISCSI_DBG_TCP(task->conn,
1003					      "  done with r2t %p\n", r2t);
1004				kfifo_in(&tcp_task->r2tpool.queue,
1005					    (void *)&tcp_task->r2t,
1006					    sizeof(void *));
1007				tcp_task->r2t = r2t = NULL;
1008			}
1009		}
1010
1011		if (r2t == NULL) {
1012			if (kfifo_out(&tcp_task->r2tqueue,
1013			    (void *)&tcp_task->r2t, sizeof(void *)) !=
1014			    sizeof(void *))
1015				r2t = NULL;
1016			else
1017				r2t = tcp_task->r2t;
1018		}
1019		spin_unlock_bh(&tcp_task->queue2pool);
1020	}
1021
1022	return r2t;
1023}
1024
1025/**
1026 * iscsi_tcp_task_xmit - xmit normal PDU task
1027 * @task: iscsi command task
1028 *
1029 * We're expected to return 0 when everything was transmitted successfully,
1030 * -EAGAIN if there's still data in the queue, or != 0 for any other kind
1031 * of error.
1032 */
1033int iscsi_tcp_task_xmit(struct iscsi_task *task)
1034{
1035	struct iscsi_conn *conn = task->conn;
1036	struct iscsi_session *session = conn->session;
1037	struct iscsi_r2t_info *r2t;
1038	int rc = 0;
1039
1040flush:
1041	/* Flush any pending data first. */
1042	rc = session->tt->xmit_pdu(task);
1043	if (rc < 0)
1044		return rc;
1045
1046	/* mgmt command */
1047	if (!task->sc) {
1048		if (task->hdr->itt == RESERVED_ITT)
1049			iscsi_put_task(task);
1050		return 0;
1051	}
1052
1053	/* Are we done already? */
1054	if (task->sc->sc_data_direction != DMA_TO_DEVICE)
1055		return 0;
1056
1057	r2t = iscsi_tcp_get_curr_r2t(task);
1058	if (r2t == NULL) {
1059		/* Waiting for more R2Ts to arrive. */
1060		ISCSI_DBG_TCP(conn, "no R2Ts yet\n");
1061		return 0;
1062	}
1063
1064	rc = conn->session->tt->alloc_pdu(task, ISCSI_OP_SCSI_DATA_OUT);
1065	if (rc)
1066		return rc;
1067	iscsi_prep_data_out_pdu(task, r2t, (struct iscsi_data *) task->hdr);
1068
1069	ISCSI_DBG_TCP(conn, "sol dout %p [dsn %d itt 0x%x doff %d dlen %d]\n",
1070		      r2t, r2t->datasn - 1, task->hdr->itt,
1071		      r2t->data_offset + r2t->sent, r2t->data_count);
1072
1073	rc = conn->session->tt->init_pdu(task, r2t->data_offset + r2t->sent,
1074					 r2t->data_count);
1075	if (rc) {
1076		iscsi_conn_failure(conn, ISCSI_ERR_XMIT_FAILED);
1077		return rc;
1078	}
1079
1080	r2t->sent += r2t->data_count;
1081	goto flush;
1082}
1083EXPORT_SYMBOL_GPL(iscsi_tcp_task_xmit);
1084
1085struct iscsi_cls_conn *
1086iscsi_tcp_conn_setup(struct iscsi_cls_session *cls_session, int dd_data_size,
1087		      uint32_t conn_idx)
1088
1089{
1090	struct iscsi_conn *conn;
1091	struct iscsi_cls_conn *cls_conn;
1092	struct iscsi_tcp_conn *tcp_conn;
1093
1094	cls_conn = iscsi_conn_setup(cls_session,
1095				    sizeof(*tcp_conn) + dd_data_size, conn_idx);
1096	if (!cls_conn)
1097		return NULL;
1098	conn = cls_conn->dd_data;
1099	/*
1100	 * due to strange issues with iser these are not set
1101	 * in iscsi_conn_setup
1102	 */
1103	conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1104
1105	tcp_conn = conn->dd_data;
1106	tcp_conn->iscsi_conn = conn;
1107	tcp_conn->dd_data = conn->dd_data + sizeof(*tcp_conn);
1108	return cls_conn;
1109}
1110EXPORT_SYMBOL_GPL(iscsi_tcp_conn_setup);
1111
1112void iscsi_tcp_conn_teardown(struct iscsi_cls_conn *cls_conn)
1113{
1114	iscsi_conn_teardown(cls_conn);
1115}
1116EXPORT_SYMBOL_GPL(iscsi_tcp_conn_teardown);
1117
1118int iscsi_tcp_r2tpool_alloc(struct iscsi_session *session)
1119{
1120	int i;
1121	int cmd_i;
1122
1123	/*
1124	 * initialize per-task: R2T pool and xmit queue
1125	 */
1126	for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1127	        struct iscsi_task *task = session->cmds[cmd_i];
1128		struct iscsi_tcp_task *tcp_task = task->dd_data;
1129
1130		/*
1131		 * pre-allocated x2 as much r2ts to handle race when
1132		 * target acks DataOut faster than we data_xmit() queues
1133		 * could replenish r2tqueue.
1134		 */
1135
1136		/* R2T pool */
1137		if (iscsi_pool_init(&tcp_task->r2tpool,
1138				    session->max_r2t * 2, NULL,
1139				    sizeof(struct iscsi_r2t_info))) {
1140			goto r2t_alloc_fail;
1141		}
1142
1143		/* R2T xmit queue */
1144		if (kfifo_alloc(&tcp_task->r2tqueue,
1145		      session->max_r2t * 4 * sizeof(void*), GFP_KERNEL)) {
1146			iscsi_pool_free(&tcp_task->r2tpool);
1147			goto r2t_alloc_fail;
1148		}
1149		spin_lock_init(&tcp_task->pool2queue);
1150		spin_lock_init(&tcp_task->queue2pool);
1151	}
1152
1153	return 0;
1154
1155r2t_alloc_fail:
1156	for (i = 0; i < cmd_i; i++) {
1157		struct iscsi_task *task = session->cmds[i];
1158		struct iscsi_tcp_task *tcp_task = task->dd_data;
1159
1160		kfifo_free(&tcp_task->r2tqueue);
1161		iscsi_pool_free(&tcp_task->r2tpool);
1162	}
1163	return -ENOMEM;
1164}
1165EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_alloc);
1166
1167void iscsi_tcp_r2tpool_free(struct iscsi_session *session)
1168{
1169	int i;
1170
1171	for (i = 0; i < session->cmds_max; i++) {
1172		struct iscsi_task *task = session->cmds[i];
1173		struct iscsi_tcp_task *tcp_task = task->dd_data;
1174
1175		kfifo_free(&tcp_task->r2tqueue);
1176		iscsi_pool_free(&tcp_task->r2tpool);
1177	}
1178}
1179EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_free);
1180
1181int iscsi_tcp_set_max_r2t(struct iscsi_conn *conn, char *buf)
1182{
1183	struct iscsi_session *session = conn->session;
1184	unsigned short r2ts = 0;
1185
1186	sscanf(buf, "%hu", &r2ts);
1187	if (session->max_r2t == r2ts)
1188		return 0;
1189
1190	if (!r2ts || !is_power_of_2(r2ts))
1191		return -EINVAL;
1192
1193	session->max_r2t = r2ts;
1194	iscsi_tcp_r2tpool_free(session);
1195	return iscsi_tcp_r2tpool_alloc(session);
1196}
1197EXPORT_SYMBOL_GPL(iscsi_tcp_set_max_r2t);
1198
1199void iscsi_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
1200			      struct iscsi_stats *stats)
1201{
1202	struct iscsi_conn *conn = cls_conn->dd_data;
1203
1204	stats->txdata_octets = conn->txdata_octets;
1205	stats->rxdata_octets = conn->rxdata_octets;
1206	stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
1207	stats->dataout_pdus = conn->dataout_pdus_cnt;
1208	stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
1209	stats->datain_pdus = conn->datain_pdus_cnt;
1210	stats->r2t_pdus = conn->r2t_pdus_cnt;
1211	stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
1212	stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
1213}
1214EXPORT_SYMBOL_GPL(iscsi_tcp_conn_get_stats);
1215