Loading...
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 <net/tcp.h>
40#include <scsi/scsi_cmnd.h>
41#include <scsi/scsi_device.h>
42#include <scsi/scsi_host.h>
43#include <scsi/scsi.h>
44#include <scsi/scsi_transport_iscsi.h>
45
46#include "iscsi_tcp.h"
47
48MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
49 "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
50 "Alex Aizman <itn780@yahoo.com>");
51MODULE_DESCRIPTION("iSCSI/TCP data-path");
52MODULE_LICENSE("GPL");
53
54static int iscsi_dbg_libtcp;
55module_param_named(debug_libiscsi_tcp, iscsi_dbg_libtcp, int,
56 S_IRUGO | S_IWUSR);
57MODULE_PARM_DESC(debug_libiscsi_tcp, "Turn on debugging for libiscsi_tcp "
58 "module. Set to 1 to turn on, and zero to turn off. Default "
59 "is off.");
60
61#define ISCSI_DBG_TCP(_conn, dbg_fmt, arg...) \
62 do { \
63 if (iscsi_dbg_libtcp) \
64 iscsi_conn_printk(KERN_INFO, _conn, \
65 "%s " dbg_fmt, \
66 __func__, ##arg); \
67 } while (0);
68
69static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
70 struct iscsi_segment *segment);
71
72/*
73 * Scatterlist handling: inside the iscsi_segment, we
74 * remember an index into the scatterlist, and set data/size
75 * to the current scatterlist entry. For highmem pages, we
76 * kmap as needed.
77 *
78 * Note that the page is unmapped when we return from
79 * TCP's data_ready handler, so we may end up mapping and
80 * unmapping the same page repeatedly. The whole reason
81 * for this is that we shouldn't keep the page mapped
82 * outside the softirq.
83 */
84
85/**
86 * iscsi_tcp_segment_init_sg - init indicated scatterlist entry
87 * @segment: the buffer object
88 * @sg: scatterlist
89 * @offset: byte offset into that sg entry
90 *
91 * This function sets up the segment so that subsequent
92 * data is copied to the indicated sg entry, at the given
93 * offset.
94 */
95static inline void
96iscsi_tcp_segment_init_sg(struct iscsi_segment *segment,
97 struct scatterlist *sg, unsigned int offset)
98{
99 segment->sg = sg;
100 segment->sg_offset = offset;
101 segment->size = min(sg->length - offset,
102 segment->total_size - segment->total_copied);
103 segment->data = NULL;
104}
105
106/**
107 * iscsi_tcp_segment_map - map the current S/G page
108 * @segment: iscsi_segment
109 * @recv: 1 if called from recv path
110 *
111 * We only need to possibly kmap data if scatter lists are being used,
112 * because the iscsi passthrough and internal IO paths will never use high
113 * mem pages.
114 */
115static void iscsi_tcp_segment_map(struct iscsi_segment *segment, int recv)
116{
117 struct scatterlist *sg;
118
119 if (segment->data != NULL || !segment->sg)
120 return;
121
122 sg = segment->sg;
123 BUG_ON(segment->sg_mapped);
124 BUG_ON(sg->length == 0);
125
126 /*
127 * If the page count is greater than one it is ok to send
128 * to the network layer's zero copy send path. If not we
129 * have to go the slow sendmsg path. We always map for the
130 * recv path.
131 */
132 if (page_count(sg_page(sg)) >= 1 && !recv)
133 return;
134
135 if (recv) {
136 segment->atomic_mapped = true;
137 segment->sg_mapped = kmap_atomic(sg_page(sg), KM_SOFTIRQ0);
138 } else {
139 segment->atomic_mapped = false;
140 /* the xmit path can sleep with the page mapped so use kmap */
141 segment->sg_mapped = kmap(sg_page(sg));
142 }
143
144 segment->data = segment->sg_mapped + sg->offset + segment->sg_offset;
145}
146
147void iscsi_tcp_segment_unmap(struct iscsi_segment *segment)
148{
149 if (segment->sg_mapped) {
150 if (segment->atomic_mapped)
151 kunmap_atomic(segment->sg_mapped, KM_SOFTIRQ0);
152 else
153 kunmap(sg_page(segment->sg));
154 segment->sg_mapped = NULL;
155 segment->data = NULL;
156 }
157}
158EXPORT_SYMBOL_GPL(iscsi_tcp_segment_unmap);
159
160/*
161 * Splice the digest buffer into the buffer
162 */
163static inline void
164iscsi_tcp_segment_splice_digest(struct iscsi_segment *segment, void *digest)
165{
166 segment->data = digest;
167 segment->digest_len = ISCSI_DIGEST_SIZE;
168 segment->total_size += ISCSI_DIGEST_SIZE;
169 segment->size = ISCSI_DIGEST_SIZE;
170 segment->copied = 0;
171 segment->sg = NULL;
172 segment->hash = NULL;
173}
174
175/**
176 * iscsi_tcp_segment_done - check whether the segment is complete
177 * @tcp_conn: iscsi tcp connection
178 * @segment: iscsi segment to check
179 * @recv: set to one of this is called from the recv path
180 * @copied: number of bytes copied
181 *
182 * Check if we're done receiving this segment. If the receive
183 * buffer is full but we expect more data, move on to the
184 * next entry in the scatterlist.
185 *
186 * If the amount of data we received isn't a multiple of 4,
187 * we will transparently receive the pad bytes, too.
188 *
189 * This function must be re-entrant.
190 */
191int iscsi_tcp_segment_done(struct iscsi_tcp_conn *tcp_conn,
192 struct iscsi_segment *segment, int recv,
193 unsigned copied)
194{
195 struct scatterlist sg;
196 unsigned int pad;
197
198 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copied %u %u size %u %s\n",
199 segment->copied, copied, segment->size,
200 recv ? "recv" : "xmit");
201 if (segment->hash && copied) {
202 /*
203 * If a segment is kmapd we must unmap it before sending
204 * to the crypto layer since that will try to kmap it again.
205 */
206 iscsi_tcp_segment_unmap(segment);
207
208 if (!segment->data) {
209 sg_init_table(&sg, 1);
210 sg_set_page(&sg, sg_page(segment->sg), copied,
211 segment->copied + segment->sg_offset +
212 segment->sg->offset);
213 } else
214 sg_init_one(&sg, segment->data + segment->copied,
215 copied);
216 crypto_hash_update(segment->hash, &sg, copied);
217 }
218
219 segment->copied += copied;
220 if (segment->copied < segment->size) {
221 iscsi_tcp_segment_map(segment, recv);
222 return 0;
223 }
224
225 segment->total_copied += segment->copied;
226 segment->copied = 0;
227 segment->size = 0;
228
229 /* Unmap the current scatterlist page, if there is one. */
230 iscsi_tcp_segment_unmap(segment);
231
232 /* Do we have more scatterlist entries? */
233 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "total copied %u total size %u\n",
234 segment->total_copied, segment->total_size);
235 if (segment->total_copied < segment->total_size) {
236 /* Proceed to the next entry in the scatterlist. */
237 iscsi_tcp_segment_init_sg(segment, sg_next(segment->sg),
238 0);
239 iscsi_tcp_segment_map(segment, recv);
240 BUG_ON(segment->size == 0);
241 return 0;
242 }
243
244 /* Do we need to handle padding? */
245 if (!(tcp_conn->iscsi_conn->session->tt->caps & CAP_PADDING_OFFLOAD)) {
246 pad = iscsi_padding(segment->total_copied);
247 if (pad != 0) {
248 ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
249 "consume %d pad bytes\n", pad);
250 segment->total_size += pad;
251 segment->size = pad;
252 segment->data = segment->padbuf;
253 return 0;
254 }
255 }
256
257 /*
258 * Set us up for transferring the data digest. hdr digest
259 * is completely handled in hdr done function.
260 */
261 if (segment->hash) {
262 crypto_hash_final(segment->hash, segment->digest);
263 iscsi_tcp_segment_splice_digest(segment,
264 recv ? segment->recv_digest : segment->digest);
265 return 0;
266 }
267
268 return 1;
269}
270EXPORT_SYMBOL_GPL(iscsi_tcp_segment_done);
271
272/**
273 * iscsi_tcp_segment_recv - copy data to segment
274 * @tcp_conn: the iSCSI TCP connection
275 * @segment: 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 */
289static int
290iscsi_tcp_segment_recv(struct iscsi_tcp_conn *tcp_conn,
291 struct iscsi_segment *segment, const void *ptr,
292 unsigned int len)
293{
294 unsigned int copy = 0, copied = 0;
295
296 while (!iscsi_tcp_segment_done(tcp_conn, segment, 1, copy)) {
297 if (copied == len) {
298 ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
299 "copied %d bytes\n", len);
300 break;
301 }
302
303 copy = min(len - copied, segment->size - segment->copied);
304 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copying %d\n", copy);
305 memcpy(segment->data + segment->copied, ptr + copied, copy);
306 copied += copy;
307 }
308 return copied;
309}
310
311inline void
312iscsi_tcp_dgst_header(struct hash_desc *hash, const void *hdr, size_t hdrlen,
313 unsigned char digest[ISCSI_DIGEST_SIZE])
314{
315 struct scatterlist sg;
316
317 sg_init_one(&sg, hdr, hdrlen);
318 crypto_hash_digest(hash, &sg, hdrlen, digest);
319}
320EXPORT_SYMBOL_GPL(iscsi_tcp_dgst_header);
321
322static inline int
323iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
324 struct iscsi_segment *segment)
325{
326 if (!segment->digest_len)
327 return 1;
328
329 if (memcmp(segment->recv_digest, segment->digest,
330 segment->digest_len)) {
331 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "digest mismatch\n");
332 return 0;
333 }
334
335 return 1;
336}
337
338/*
339 * Helper function to set up segment buffer
340 */
341static inline void
342__iscsi_segment_init(struct iscsi_segment *segment, size_t size,
343 iscsi_segment_done_fn_t *done, struct hash_desc *hash)
344{
345 memset(segment, 0, sizeof(*segment));
346 segment->total_size = size;
347 segment->done = done;
348
349 if (hash) {
350 segment->hash = hash;
351 crypto_hash_init(hash);
352 }
353}
354
355inline void
356iscsi_segment_init_linear(struct iscsi_segment *segment, void *data,
357 size_t size, iscsi_segment_done_fn_t *done,
358 struct hash_desc *hash)
359{
360 __iscsi_segment_init(segment, size, done, hash);
361 segment->data = data;
362 segment->size = size;
363}
364EXPORT_SYMBOL_GPL(iscsi_segment_init_linear);
365
366inline int
367iscsi_segment_seek_sg(struct iscsi_segment *segment,
368 struct scatterlist *sg_list, unsigned int sg_count,
369 unsigned int offset, size_t size,
370 iscsi_segment_done_fn_t *done, struct hash_desc *hash)
371{
372 struct scatterlist *sg;
373 unsigned int i;
374
375 __iscsi_segment_init(segment, size, done, hash);
376 for_each_sg(sg_list, sg, sg_count, i) {
377 if (offset < sg->length) {
378 iscsi_tcp_segment_init_sg(segment, sg, offset);
379 return 0;
380 }
381 offset -= sg->length;
382 }
383
384 return ISCSI_ERR_DATA_OFFSET;
385}
386EXPORT_SYMBOL_GPL(iscsi_segment_seek_sg);
387
388/**
389 * iscsi_tcp_hdr_recv_prep - prep segment for hdr reception
390 * @tcp_conn: iscsi connection to prep for
391 *
392 * This function always passes NULL for the hash argument, because when this
393 * function is called we do not yet know the final size of the header and want
394 * to delay the digest processing until we know that.
395 */
396void iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
397{
398 ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
399 "(%s)\n", tcp_conn->iscsi_conn->hdrdgst_en ?
400 "digest enabled" : "digest disabled");
401 iscsi_segment_init_linear(&tcp_conn->in.segment,
402 tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
403 iscsi_tcp_hdr_recv_done, NULL);
404}
405EXPORT_SYMBOL_GPL(iscsi_tcp_hdr_recv_prep);
406
407/*
408 * Handle incoming reply to any other type of command
409 */
410static int
411iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
412 struct iscsi_segment *segment)
413{
414 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
415 int rc = 0;
416
417 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
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
429static void
430iscsi_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 !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
437 rx_hash = tcp_conn->rx_hash;
438
439 iscsi_segment_init_linear(&tcp_conn->in.segment,
440 conn->data, tcp_conn->in.datalen,
441 iscsi_tcp_data_recv_done, rx_hash);
442}
443
444/**
445 * iscsi_tcp_cleanup_task - free tcp_task resources
446 * @task: iscsi task
447 *
448 * must be called with session lock
449 */
450void iscsi_tcp_cleanup_task(struct iscsi_task *task)
451{
452 struct iscsi_tcp_task *tcp_task = task->dd_data;
453 struct iscsi_r2t_info *r2t;
454
455 /* nothing to do for mgmt */
456 if (!task->sc)
457 return;
458
459 /* flush task's r2t queues */
460 while (kfifo_out(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*))) {
461 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
462 sizeof(void*));
463 ISCSI_DBG_TCP(task->conn, "pending r2t dropped\n");
464 }
465
466 r2t = tcp_task->r2t;
467 if (r2t != NULL) {
468 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
469 sizeof(void*));
470 tcp_task->r2t = NULL;
471 }
472}
473EXPORT_SYMBOL_GPL(iscsi_tcp_cleanup_task);
474
475/**
476 * iscsi_tcp_data_in - SCSI Data-In Response processing
477 * @conn: iscsi connection
478 * @task: scsi command task
479 */
480static int iscsi_tcp_data_in(struct iscsi_conn *conn, struct iscsi_task *task)
481{
482 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
483 struct iscsi_tcp_task *tcp_task = task->dd_data;
484 struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
485 int datasn = be32_to_cpu(rhdr->datasn);
486 unsigned total_in_length = scsi_in(task->sc)->length;
487
488 /*
489 * lib iscsi will update this in the completion handling if there
490 * is status.
491 */
492 if (!(rhdr->flags & ISCSI_FLAG_DATA_STATUS))
493 iscsi_update_cmdsn(conn->session, (struct iscsi_nopin*)rhdr);
494
495 if (tcp_conn->in.datalen == 0)
496 return 0;
497
498 if (tcp_task->exp_datasn != datasn) {
499 ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->datasn(%d)"
500 "\n", tcp_task->exp_datasn, datasn);
501 return ISCSI_ERR_DATASN;
502 }
503
504 tcp_task->exp_datasn++;
505
506 tcp_task->data_offset = be32_to_cpu(rhdr->offset);
507 if (tcp_task->data_offset + tcp_conn->in.datalen > total_in_length) {
508 ISCSI_DBG_TCP(conn, "data_offset(%d) + data_len(%d) > "
509 "total_length_in(%d)\n", tcp_task->data_offset,
510 tcp_conn->in.datalen, total_in_length);
511 return ISCSI_ERR_DATA_OFFSET;
512 }
513
514 conn->datain_pdus_cnt++;
515 return 0;
516}
517
518/**
519 * iscsi_tcp_r2t_rsp - iSCSI R2T Response processing
520 * @conn: iscsi connection
521 * @task: scsi command task
522 */
523static int iscsi_tcp_r2t_rsp(struct iscsi_conn *conn, struct iscsi_task *task)
524{
525 struct iscsi_session *session = conn->session;
526 struct iscsi_tcp_task *tcp_task = task->dd_data;
527 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
528 struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
529 struct iscsi_r2t_info *r2t;
530 int r2tsn = be32_to_cpu(rhdr->r2tsn);
531 int rc;
532
533 if (tcp_conn->in.datalen) {
534 iscsi_conn_printk(KERN_ERR, conn,
535 "invalid R2t with datalen %d\n",
536 tcp_conn->in.datalen);
537 return ISCSI_ERR_DATALEN;
538 }
539
540 if (tcp_task->exp_datasn != r2tsn){
541 ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
542 tcp_task->exp_datasn, r2tsn);
543 return ISCSI_ERR_R2TSN;
544 }
545
546 /* fill-in new R2T associated with the task */
547 iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
548
549 if (!task->sc || session->state != ISCSI_STATE_LOGGED_IN) {
550 iscsi_conn_printk(KERN_INFO, conn,
551 "dropping R2T itt %d in recovery.\n",
552 task->itt);
553 return 0;
554 }
555
556 rc = kfifo_out(&tcp_task->r2tpool.queue, (void*)&r2t, sizeof(void*));
557 if (!rc) {
558 iscsi_conn_printk(KERN_ERR, conn, "Could not allocate R2T. "
559 "Target has sent more R2Ts than it "
560 "negotiated for or driver has has leaked.\n");
561 return ISCSI_ERR_PROTO;
562 }
563
564 r2t->exp_statsn = rhdr->statsn;
565 r2t->data_length = be32_to_cpu(rhdr->data_length);
566 if (r2t->data_length == 0) {
567 iscsi_conn_printk(KERN_ERR, conn,
568 "invalid R2T with zero data len\n");
569 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
570 sizeof(void*));
571 return ISCSI_ERR_DATALEN;
572 }
573
574 if (r2t->data_length > session->max_burst)
575 ISCSI_DBG_TCP(conn, "invalid R2T with data len %u and max "
576 "burst %u. Attempting to execute request.\n",
577 r2t->data_length, session->max_burst);
578
579 r2t->data_offset = be32_to_cpu(rhdr->data_offset);
580 if (r2t->data_offset + r2t->data_length > scsi_out(task->sc)->length) {
581 iscsi_conn_printk(KERN_ERR, conn,
582 "invalid R2T with data len %u at offset %u "
583 "and total length %d\n", r2t->data_length,
584 r2t->data_offset, scsi_out(task->sc)->length);
585 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
586 sizeof(void*));
587 return ISCSI_ERR_DATALEN;
588 }
589
590 r2t->ttt = rhdr->ttt; /* no flip */
591 r2t->datasn = 0;
592 r2t->sent = 0;
593
594 tcp_task->exp_datasn = r2tsn + 1;
595 kfifo_in(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*));
596 conn->r2t_pdus_cnt++;
597
598 iscsi_requeue_task(task);
599 return 0;
600}
601
602/*
603 * Handle incoming reply to DataIn command
604 */
605static int
606iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
607 struct iscsi_segment *segment)
608{
609 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
610 struct iscsi_hdr *hdr = tcp_conn->in.hdr;
611 int rc;
612
613 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
614 return ISCSI_ERR_DATA_DGST;
615
616 /* check for non-exceptional status */
617 if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
618 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
619 if (rc)
620 return rc;
621 }
622
623 iscsi_tcp_hdr_recv_prep(tcp_conn);
624 return 0;
625}
626
627/**
628 * iscsi_tcp_hdr_dissect - process PDU header
629 * @conn: iSCSI connection
630 * @hdr: PDU header
631 *
632 * This function analyzes the header of the PDU received,
633 * and performs several sanity checks. If the PDU is accompanied
634 * by data, the receive buffer is set up to copy the incoming data
635 * to the correct location.
636 */
637static int
638iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
639{
640 int rc = 0, opcode, ahslen;
641 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
642 struct iscsi_task *task;
643
644 /* verify PDU length */
645 tcp_conn->in.datalen = ntoh24(hdr->dlength);
646 if (tcp_conn->in.datalen > conn->max_recv_dlength) {
647 iscsi_conn_printk(KERN_ERR, conn,
648 "iscsi_tcp: datalen %d > %d\n",
649 tcp_conn->in.datalen, conn->max_recv_dlength);
650 return ISCSI_ERR_DATALEN;
651 }
652
653 /* Additional header segments. So far, we don't
654 * process additional headers.
655 */
656 ahslen = hdr->hlength << 2;
657
658 opcode = hdr->opcode & ISCSI_OPCODE_MASK;
659 /* verify itt (itt encoding: age+cid+itt) */
660 rc = iscsi_verify_itt(conn, hdr->itt);
661 if (rc)
662 return rc;
663
664 ISCSI_DBG_TCP(conn, "opcode 0x%x ahslen %d datalen %d\n",
665 opcode, ahslen, tcp_conn->in.datalen);
666
667 switch(opcode) {
668 case ISCSI_OP_SCSI_DATA_IN:
669 spin_lock(&conn->session->lock);
670 task = iscsi_itt_to_ctask(conn, hdr->itt);
671 if (!task)
672 rc = ISCSI_ERR_BAD_ITT;
673 else
674 rc = iscsi_tcp_data_in(conn, task);
675 if (rc) {
676 spin_unlock(&conn->session->lock);
677 break;
678 }
679
680 if (tcp_conn->in.datalen) {
681 struct iscsi_tcp_task *tcp_task = task->dd_data;
682 struct hash_desc *rx_hash = NULL;
683 struct scsi_data_buffer *sdb = scsi_in(task->sc);
684
685 /*
686 * Setup copy of Data-In into the Scsi_Cmnd
687 * Scatterlist case:
688 * We set up the iscsi_segment to point to the next
689 * scatterlist entry to copy to. As we go along,
690 * we move on to the next scatterlist entry and
691 * update the digest per-entry.
692 */
693 if (conn->datadgst_en &&
694 !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
695 rx_hash = tcp_conn->rx_hash;
696
697 ISCSI_DBG_TCP(conn, "iscsi_tcp_begin_data_in( "
698 "offset=%d, datalen=%d)\n",
699 tcp_task->data_offset,
700 tcp_conn->in.datalen);
701 task->last_xfer = jiffies;
702 rc = iscsi_segment_seek_sg(&tcp_conn->in.segment,
703 sdb->table.sgl,
704 sdb->table.nents,
705 tcp_task->data_offset,
706 tcp_conn->in.datalen,
707 iscsi_tcp_process_data_in,
708 rx_hash);
709 spin_unlock(&conn->session->lock);
710 return rc;
711 }
712 rc = __iscsi_complete_pdu(conn, hdr, NULL, 0);
713 spin_unlock(&conn->session->lock);
714 break;
715 case ISCSI_OP_SCSI_CMD_RSP:
716 if (tcp_conn->in.datalen) {
717 iscsi_tcp_data_recv_prep(tcp_conn);
718 return 0;
719 }
720 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
721 break;
722 case ISCSI_OP_R2T:
723 spin_lock(&conn->session->lock);
724 task = iscsi_itt_to_ctask(conn, hdr->itt);
725 if (!task)
726 rc = ISCSI_ERR_BAD_ITT;
727 else if (ahslen)
728 rc = ISCSI_ERR_AHSLEN;
729 else if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
730 task->last_xfer = jiffies;
731 rc = iscsi_tcp_r2t_rsp(conn, task);
732 } else
733 rc = ISCSI_ERR_PROTO;
734 spin_unlock(&conn->session->lock);
735 break;
736 case ISCSI_OP_LOGIN_RSP:
737 case ISCSI_OP_TEXT_RSP:
738 case ISCSI_OP_REJECT:
739 case ISCSI_OP_ASYNC_EVENT:
740 /*
741 * It is possible that we could get a PDU with a buffer larger
742 * than 8K, but there are no targets that currently do this.
743 * For now we fail until we find a vendor that needs it
744 */
745 if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
746 iscsi_conn_printk(KERN_ERR, conn,
747 "iscsi_tcp: received buffer of "
748 "len %u but conn buffer is only %u "
749 "(opcode %0x)\n",
750 tcp_conn->in.datalen,
751 ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
752 rc = ISCSI_ERR_PROTO;
753 break;
754 }
755
756 /* If there's data coming in with the response,
757 * receive it to the connection's buffer.
758 */
759 if (tcp_conn->in.datalen) {
760 iscsi_tcp_data_recv_prep(tcp_conn);
761 return 0;
762 }
763 /* fall through */
764 case ISCSI_OP_LOGOUT_RSP:
765 case ISCSI_OP_NOOP_IN:
766 case ISCSI_OP_SCSI_TMFUNC_RSP:
767 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
768 break;
769 default:
770 rc = ISCSI_ERR_BAD_OPCODE;
771 break;
772 }
773
774 if (rc == 0) {
775 /* Anything that comes with data should have
776 * been handled above. */
777 if (tcp_conn->in.datalen)
778 return ISCSI_ERR_PROTO;
779 iscsi_tcp_hdr_recv_prep(tcp_conn);
780 }
781
782 return rc;
783}
784
785/**
786 * iscsi_tcp_hdr_recv_done - process PDU header
787 *
788 * This is the callback invoked when the PDU header has
789 * been received. If the header is followed by additional
790 * header segments, we go back for more data.
791 */
792static int
793iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
794 struct iscsi_segment *segment)
795{
796 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
797 struct iscsi_hdr *hdr;
798
799 /* Check if there are additional header segments
800 * *prior* to computing the digest, because we
801 * may need to go back to the caller for more.
802 */
803 hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
804 if (segment->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
805 /* Bump the header length - the caller will
806 * just loop around and get the AHS for us, and
807 * call again. */
808 unsigned int ahslen = hdr->hlength << 2;
809
810 /* Make sure we don't overflow */
811 if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
812 return ISCSI_ERR_AHSLEN;
813
814 segment->total_size += ahslen;
815 segment->size += ahslen;
816 return 0;
817 }
818
819 /* We're done processing the header. See if we're doing
820 * header digests; if so, set up the recv_digest buffer
821 * and go back for more. */
822 if (conn->hdrdgst_en &&
823 !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD)) {
824 if (segment->digest_len == 0) {
825 /*
826 * Even if we offload the digest processing we
827 * splice it in so we can increment the skb/segment
828 * counters in preparation for the data segment.
829 */
830 iscsi_tcp_segment_splice_digest(segment,
831 segment->recv_digest);
832 return 0;
833 }
834
835 iscsi_tcp_dgst_header(tcp_conn->rx_hash, hdr,
836 segment->total_copied - ISCSI_DIGEST_SIZE,
837 segment->digest);
838
839 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
840 return ISCSI_ERR_HDR_DGST;
841 }
842
843 tcp_conn->in.hdr = hdr;
844 return iscsi_tcp_hdr_dissect(conn, hdr);
845}
846
847/**
848 * iscsi_tcp_recv_segment_is_hdr - tests if we are reading in a header
849 * @tcp_conn: iscsi tcp conn
850 *
851 * returns non zero if we are currently processing or setup to process
852 * a header.
853 */
854inline int iscsi_tcp_recv_segment_is_hdr(struct iscsi_tcp_conn *tcp_conn)
855{
856 return tcp_conn->in.segment.done == iscsi_tcp_hdr_recv_done;
857}
858EXPORT_SYMBOL_GPL(iscsi_tcp_recv_segment_is_hdr);
859
860/**
861 * iscsi_tcp_recv_skb - Process skb
862 * @conn: iscsi connection
863 * @skb: network buffer with header and/or data segment
864 * @offset: offset in skb
865 * @offload: bool indicating if transfer was offloaded
866 *
867 * Will return status of transfer in status. And will return
868 * number of bytes copied.
869 */
870int iscsi_tcp_recv_skb(struct iscsi_conn *conn, struct sk_buff *skb,
871 unsigned int offset, bool offloaded, int *status)
872{
873 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
874 struct iscsi_segment *segment = &tcp_conn->in.segment;
875 struct skb_seq_state seq;
876 unsigned int consumed = 0;
877 int rc = 0;
878
879 ISCSI_DBG_TCP(conn, "in %d bytes\n", skb->len - offset);
880 /*
881 * Update for each skb instead of pdu, because over slow networks a
882 * data_in's data could take a while to read in. We also want to
883 * account for r2ts.
884 */
885 conn->last_recv = jiffies;
886
887 if (unlikely(conn->suspend_rx)) {
888 ISCSI_DBG_TCP(conn, "Rx suspended!\n");
889 *status = ISCSI_TCP_SUSPENDED;
890 return 0;
891 }
892
893 if (offloaded) {
894 segment->total_copied = segment->total_size;
895 goto segment_done;
896 }
897
898 skb_prepare_seq_read(skb, offset, skb->len, &seq);
899 while (1) {
900 unsigned int avail;
901 const u8 *ptr;
902
903 avail = skb_seq_read(consumed, &ptr, &seq);
904 if (avail == 0) {
905 ISCSI_DBG_TCP(conn, "no more data avail. Consumed %d\n",
906 consumed);
907 *status = ISCSI_TCP_SKB_DONE;
908 skb_abort_seq_read(&seq);
909 goto skb_done;
910 }
911 BUG_ON(segment->copied >= segment->size);
912
913 ISCSI_DBG_TCP(conn, "skb %p ptr=%p avail=%u\n", skb, ptr,
914 avail);
915 rc = iscsi_tcp_segment_recv(tcp_conn, segment, ptr, avail);
916 BUG_ON(rc == 0);
917 consumed += rc;
918
919 if (segment->total_copied >= segment->total_size) {
920 skb_abort_seq_read(&seq);
921 goto segment_done;
922 }
923 }
924
925segment_done:
926 *status = ISCSI_TCP_SEGMENT_DONE;
927 ISCSI_DBG_TCP(conn, "segment done\n");
928 rc = segment->done(tcp_conn, segment);
929 if (rc != 0) {
930 *status = ISCSI_TCP_CONN_ERR;
931 ISCSI_DBG_TCP(conn, "Error receiving PDU, errno=%d\n", rc);
932 iscsi_conn_failure(conn, rc);
933 return 0;
934 }
935 /* The done() functions sets up the next segment. */
936
937skb_done:
938 conn->rxdata_octets += consumed;
939 return consumed;
940}
941EXPORT_SYMBOL_GPL(iscsi_tcp_recv_skb);
942
943/**
944 * iscsi_tcp_task_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
945 * @conn: iscsi connection
946 * @task: scsi command task
947 * @sc: scsi command
948 */
949int iscsi_tcp_task_init(struct iscsi_task *task)
950{
951 struct iscsi_tcp_task *tcp_task = task->dd_data;
952 struct iscsi_conn *conn = task->conn;
953 struct scsi_cmnd *sc = task->sc;
954 int err;
955
956 if (!sc) {
957 /*
958 * mgmt tasks do not have a scatterlist since they come
959 * in from the iscsi interface.
960 */
961 ISCSI_DBG_TCP(conn, "mtask deq [itt 0x%x]\n", task->itt);
962
963 return conn->session->tt->init_pdu(task, 0, task->data_count);
964 }
965
966 BUG_ON(kfifo_len(&tcp_task->r2tqueue));
967 tcp_task->exp_datasn = 0;
968
969 /* Prepare PDU, optionally w/ immediate data */
970 ISCSI_DBG_TCP(conn, "task deq [itt 0x%x imm %d unsol %d]\n",
971 task->itt, task->imm_count, task->unsol_r2t.data_length);
972
973 err = conn->session->tt->init_pdu(task, 0, task->imm_count);
974 if (err)
975 return err;
976 task->imm_count = 0;
977 return 0;
978}
979EXPORT_SYMBOL_GPL(iscsi_tcp_task_init);
980
981static struct iscsi_r2t_info *iscsi_tcp_get_curr_r2t(struct iscsi_task *task)
982{
983 struct iscsi_session *session = task->conn->session;
984 struct iscsi_tcp_task *tcp_task = task->dd_data;
985 struct iscsi_r2t_info *r2t = NULL;
986
987 if (iscsi_task_has_unsol_data(task))
988 r2t = &task->unsol_r2t;
989 else {
990 spin_lock_bh(&session->lock);
991 if (tcp_task->r2t) {
992 r2t = tcp_task->r2t;
993 /* Continue with this R2T? */
994 if (r2t->data_length <= r2t->sent) {
995 ISCSI_DBG_TCP(task->conn,
996 " done with r2t %p\n", r2t);
997 kfifo_in(&tcp_task->r2tpool.queue,
998 (void *)&tcp_task->r2t,
999 sizeof(void *));
1000 tcp_task->r2t = r2t = NULL;
1001 }
1002 }
1003
1004 if (r2t == NULL) {
1005 if (kfifo_out(&tcp_task->r2tqueue,
1006 (void *)&tcp_task->r2t, sizeof(void *)) !=
1007 sizeof(void *))
1008 r2t = NULL;
1009 else
1010 r2t = tcp_task->r2t;
1011 }
1012 spin_unlock_bh(&session->lock);
1013 }
1014
1015 return r2t;
1016}
1017
1018/**
1019 * iscsi_tcp_task_xmit - xmit normal PDU task
1020 * @task: iscsi command task
1021 *
1022 * We're expected to return 0 when everything was transmitted successfully,
1023 * -EAGAIN if there's still data in the queue, or != 0 for any other kind
1024 * of error.
1025 */
1026int iscsi_tcp_task_xmit(struct iscsi_task *task)
1027{
1028 struct iscsi_conn *conn = task->conn;
1029 struct iscsi_session *session = conn->session;
1030 struct iscsi_r2t_info *r2t;
1031 int rc = 0;
1032
1033flush:
1034 /* Flush any pending data first. */
1035 rc = session->tt->xmit_pdu(task);
1036 if (rc < 0)
1037 return rc;
1038
1039 /* mgmt command */
1040 if (!task->sc) {
1041 if (task->hdr->itt == RESERVED_ITT)
1042 iscsi_put_task(task);
1043 return 0;
1044 }
1045
1046 /* Are we done already? */
1047 if (task->sc->sc_data_direction != DMA_TO_DEVICE)
1048 return 0;
1049
1050 r2t = iscsi_tcp_get_curr_r2t(task);
1051 if (r2t == NULL) {
1052 /* Waiting for more R2Ts to arrive. */
1053 ISCSI_DBG_TCP(conn, "no R2Ts yet\n");
1054 return 0;
1055 }
1056
1057 rc = conn->session->tt->alloc_pdu(task, ISCSI_OP_SCSI_DATA_OUT);
1058 if (rc)
1059 return rc;
1060 iscsi_prep_data_out_pdu(task, r2t, (struct iscsi_data *) task->hdr);
1061
1062 ISCSI_DBG_TCP(conn, "sol dout %p [dsn %d itt 0x%x doff %d dlen %d]\n",
1063 r2t, r2t->datasn - 1, task->hdr->itt,
1064 r2t->data_offset + r2t->sent, r2t->data_count);
1065
1066 rc = conn->session->tt->init_pdu(task, r2t->data_offset + r2t->sent,
1067 r2t->data_count);
1068 if (rc) {
1069 iscsi_conn_failure(conn, ISCSI_ERR_XMIT_FAILED);
1070 return rc;
1071 }
1072
1073 r2t->sent += r2t->data_count;
1074 goto flush;
1075}
1076EXPORT_SYMBOL_GPL(iscsi_tcp_task_xmit);
1077
1078struct iscsi_cls_conn *
1079iscsi_tcp_conn_setup(struct iscsi_cls_session *cls_session, int dd_data_size,
1080 uint32_t conn_idx)
1081
1082{
1083 struct iscsi_conn *conn;
1084 struct iscsi_cls_conn *cls_conn;
1085 struct iscsi_tcp_conn *tcp_conn;
1086
1087 cls_conn = iscsi_conn_setup(cls_session,
1088 sizeof(*tcp_conn) + dd_data_size, conn_idx);
1089 if (!cls_conn)
1090 return NULL;
1091 conn = cls_conn->dd_data;
1092 /*
1093 * due to strange issues with iser these are not set
1094 * in iscsi_conn_setup
1095 */
1096 conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1097
1098 tcp_conn = conn->dd_data;
1099 tcp_conn->iscsi_conn = conn;
1100 tcp_conn->dd_data = conn->dd_data + sizeof(*tcp_conn);
1101 return cls_conn;
1102}
1103EXPORT_SYMBOL_GPL(iscsi_tcp_conn_setup);
1104
1105void iscsi_tcp_conn_teardown(struct iscsi_cls_conn *cls_conn)
1106{
1107 iscsi_conn_teardown(cls_conn);
1108}
1109EXPORT_SYMBOL_GPL(iscsi_tcp_conn_teardown);
1110
1111int iscsi_tcp_r2tpool_alloc(struct iscsi_session *session)
1112{
1113 int i;
1114 int cmd_i;
1115
1116 /*
1117 * initialize per-task: R2T pool and xmit queue
1118 */
1119 for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1120 struct iscsi_task *task = session->cmds[cmd_i];
1121 struct iscsi_tcp_task *tcp_task = task->dd_data;
1122
1123 /*
1124 * pre-allocated x2 as much r2ts to handle race when
1125 * target acks DataOut faster than we data_xmit() queues
1126 * could replenish r2tqueue.
1127 */
1128
1129 /* R2T pool */
1130 if (iscsi_pool_init(&tcp_task->r2tpool,
1131 session->max_r2t * 2, NULL,
1132 sizeof(struct iscsi_r2t_info))) {
1133 goto r2t_alloc_fail;
1134 }
1135
1136 /* R2T xmit queue */
1137 if (kfifo_alloc(&tcp_task->r2tqueue,
1138 session->max_r2t * 4 * sizeof(void*), GFP_KERNEL)) {
1139 iscsi_pool_free(&tcp_task->r2tpool);
1140 goto r2t_alloc_fail;
1141 }
1142 }
1143
1144 return 0;
1145
1146r2t_alloc_fail:
1147 for (i = 0; i < cmd_i; i++) {
1148 struct iscsi_task *task = session->cmds[i];
1149 struct iscsi_tcp_task *tcp_task = task->dd_data;
1150
1151 kfifo_free(&tcp_task->r2tqueue);
1152 iscsi_pool_free(&tcp_task->r2tpool);
1153 }
1154 return -ENOMEM;
1155}
1156EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_alloc);
1157
1158void iscsi_tcp_r2tpool_free(struct iscsi_session *session)
1159{
1160 int i;
1161
1162 for (i = 0; i < session->cmds_max; i++) {
1163 struct iscsi_task *task = session->cmds[i];
1164 struct iscsi_tcp_task *tcp_task = task->dd_data;
1165
1166 kfifo_free(&tcp_task->r2tqueue);
1167 iscsi_pool_free(&tcp_task->r2tpool);
1168 }
1169}
1170EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_free);
1171
1172void iscsi_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
1173 struct iscsi_stats *stats)
1174{
1175 struct iscsi_conn *conn = cls_conn->dd_data;
1176
1177 stats->txdata_octets = conn->txdata_octets;
1178 stats->rxdata_octets = conn->rxdata_octets;
1179 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
1180 stats->dataout_pdus = conn->dataout_pdus_cnt;
1181 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
1182 stats->datain_pdus = conn->datain_pdus_cnt;
1183 stats->r2t_pdus = conn->r2t_pdus_cnt;
1184 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
1185 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
1186}
1187EXPORT_SYMBOL_GPL(iscsi_tcp_conn_get_stats);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * iSCSI over TCP/IP Data-Path lib
4 *
5 * Copyright (C) 2004 Dmitry Yusupov
6 * Copyright (C) 2004 Alex Aizman
7 * Copyright (C) 2005 - 2006 Mike Christie
8 * Copyright (C) 2006 Red Hat, Inc. All rights reserved.
9 * maintained by open-iscsi@googlegroups.com
10 *
11 * Credits:
12 * Christoph Hellwig
13 * FUJITA Tomonori
14 * Arne Redlich
15 * Zhenyu Wang
16 */
17
18#include <crypto/hash.h>
19#include <linux/types.h>
20#include <linux/list.h>
21#include <linux/inet.h>
22#include <linux/slab.h>
23#include <linux/file.h>
24#include <linux/blkdev.h>
25#include <linux/delay.h>
26#include <linux/kfifo.h>
27#include <linux/scatterlist.h>
28#include <linux/module.h>
29#include <net/tcp.h>
30#include <scsi/scsi_cmnd.h>
31#include <scsi/scsi_device.h>
32#include <scsi/scsi_host.h>
33#include <scsi/scsi.h>
34#include <scsi/scsi_transport_iscsi.h>
35#include <trace/events/iscsi.h>
36
37#include "iscsi_tcp.h"
38
39MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
40 "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
41 "Alex Aizman <itn780@yahoo.com>");
42MODULE_DESCRIPTION("iSCSI/TCP data-path");
43MODULE_LICENSE("GPL");
44
45static int iscsi_dbg_libtcp;
46module_param_named(debug_libiscsi_tcp, iscsi_dbg_libtcp, int,
47 S_IRUGO | S_IWUSR);
48MODULE_PARM_DESC(debug_libiscsi_tcp, "Turn on debugging for libiscsi_tcp "
49 "module. Set to 1 to turn on, and zero to turn off. Default "
50 "is off.");
51
52#define ISCSI_DBG_TCP(_conn, dbg_fmt, arg...) \
53 do { \
54 if (iscsi_dbg_libtcp) \
55 iscsi_conn_printk(KERN_INFO, _conn, \
56 "%s " dbg_fmt, \
57 __func__, ##arg); \
58 iscsi_dbg_trace(trace_iscsi_dbg_tcp, \
59 &(_conn)->cls_conn->dev, \
60 "%s " dbg_fmt, __func__, ##arg);\
61 } while (0);
62
63static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
64 struct iscsi_segment *segment);
65
66/*
67 * Scatterlist handling: inside the iscsi_segment, we
68 * remember an index into the scatterlist, and set data/size
69 * to the current scatterlist entry. For highmem pages, we
70 * kmap as needed.
71 *
72 * Note that the page is unmapped when we return from
73 * TCP's data_ready handler, so we may end up mapping and
74 * unmapping the same page repeatedly. The whole reason
75 * for this is that we shouldn't keep the page mapped
76 * outside the softirq.
77 */
78
79/**
80 * iscsi_tcp_segment_init_sg - init indicated scatterlist entry
81 * @segment: the buffer object
82 * @sg: scatterlist
83 * @offset: byte offset into that sg entry
84 *
85 * This function sets up the segment so that subsequent
86 * data is copied to the indicated sg entry, at the given
87 * offset.
88 */
89static inline void
90iscsi_tcp_segment_init_sg(struct iscsi_segment *segment,
91 struct scatterlist *sg, unsigned int offset)
92{
93 segment->sg = sg;
94 segment->sg_offset = offset;
95 segment->size = min(sg->length - offset,
96 segment->total_size - segment->total_copied);
97 segment->data = NULL;
98}
99
100/**
101 * iscsi_tcp_segment_map - map the current S/G page
102 * @segment: iscsi_segment
103 * @recv: 1 if called from recv path
104 *
105 * We only need to possibly kmap data if scatter lists are being used,
106 * because the iscsi passthrough and internal IO paths will never use high
107 * mem pages.
108 */
109static void iscsi_tcp_segment_map(struct iscsi_segment *segment, int recv)
110{
111 struct scatterlist *sg;
112
113 if (segment->data != NULL || !segment->sg)
114 return;
115
116 sg = segment->sg;
117 BUG_ON(segment->sg_mapped);
118 BUG_ON(sg->length == 0);
119
120 /*
121 * We always map for the recv path.
122 *
123 * If the page count is greater than one it is ok to send
124 * to the network layer's zero copy send path. If not we
125 * have to go the slow sendmsg path.
126 *
127 * Same goes for slab pages: skb_can_coalesce() allows
128 * coalescing neighboring slab objects into a single frag which
129 * triggers one of hardened usercopy checks.
130 */
131 if (!recv && sendpage_ok(sg_page(sg)))
132 return;
133
134 if (recv) {
135 segment->atomic_mapped = true;
136 segment->sg_mapped = kmap_atomic(sg_page(sg));
137 } else {
138 segment->atomic_mapped = false;
139 /* the xmit path can sleep with the page mapped so use kmap */
140 segment->sg_mapped = kmap(sg_page(sg));
141 }
142
143 segment->data = segment->sg_mapped + sg->offset + segment->sg_offset;
144}
145
146void iscsi_tcp_segment_unmap(struct iscsi_segment *segment)
147{
148 if (segment->sg_mapped) {
149 if (segment->atomic_mapped)
150 kunmap_atomic(segment->sg_mapped);
151 else
152 kunmap(sg_page(segment->sg));
153 segment->sg_mapped = NULL;
154 segment->data = NULL;
155 }
156}
157EXPORT_SYMBOL_GPL(iscsi_tcp_segment_unmap);
158
159/*
160 * Splice the digest buffer into the buffer
161 */
162static inline void
163iscsi_tcp_segment_splice_digest(struct iscsi_segment *segment, void *digest)
164{
165 segment->data = digest;
166 segment->digest_len = ISCSI_DIGEST_SIZE;
167 segment->total_size += ISCSI_DIGEST_SIZE;
168 segment->size = ISCSI_DIGEST_SIZE;
169 segment->copied = 0;
170 segment->sg = NULL;
171 segment->hash = NULL;
172}
173
174/**
175 * iscsi_tcp_segment_done - check whether the segment is complete
176 * @tcp_conn: iscsi tcp connection
177 * @segment: iscsi segment to check
178 * @recv: set to one of this is called from the recv path
179 * @copied: number of bytes copied
180 *
181 * Check if we're done receiving this segment. If the receive
182 * buffer is full but we expect more data, move on to the
183 * next entry in the scatterlist.
184 *
185 * If the amount of data we received isn't a multiple of 4,
186 * we will transparently receive the pad bytes, too.
187 *
188 * This function must be re-entrant.
189 */
190int iscsi_tcp_segment_done(struct iscsi_tcp_conn *tcp_conn,
191 struct iscsi_segment *segment, int recv,
192 unsigned copied)
193{
194 struct scatterlist sg;
195 unsigned int pad;
196
197 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copied %u %u size %u %s\n",
198 segment->copied, copied, segment->size,
199 recv ? "recv" : "xmit");
200 if (segment->hash && copied) {
201 /*
202 * If a segment is kmapd we must unmap it before sending
203 * to the crypto layer since that will try to kmap it again.
204 */
205 iscsi_tcp_segment_unmap(segment);
206
207 if (!segment->data) {
208 sg_init_table(&sg, 1);
209 sg_set_page(&sg, sg_page(segment->sg), copied,
210 segment->copied + segment->sg_offset +
211 segment->sg->offset);
212 } else
213 sg_init_one(&sg, segment->data + segment->copied,
214 copied);
215 ahash_request_set_crypt(segment->hash, &sg, NULL, copied);
216 crypto_ahash_update(segment->hash);
217 }
218
219 segment->copied += copied;
220 if (segment->copied < segment->size) {
221 iscsi_tcp_segment_map(segment, recv);
222 return 0;
223 }
224
225 segment->total_copied += segment->copied;
226 segment->copied = 0;
227 segment->size = 0;
228
229 /* Unmap the current scatterlist page, if there is one. */
230 iscsi_tcp_segment_unmap(segment);
231
232 /* Do we have more scatterlist entries? */
233 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "total copied %u total size %u\n",
234 segment->total_copied, segment->total_size);
235 if (segment->total_copied < segment->total_size) {
236 /* Proceed to the next entry in the scatterlist. */
237 iscsi_tcp_segment_init_sg(segment, sg_next(segment->sg),
238 0);
239 iscsi_tcp_segment_map(segment, recv);
240 BUG_ON(segment->size == 0);
241 return 0;
242 }
243
244 /* Do we need to handle padding? */
245 if (!(tcp_conn->iscsi_conn->session->tt->caps & CAP_PADDING_OFFLOAD)) {
246 pad = iscsi_padding(segment->total_copied);
247 if (pad != 0) {
248 ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
249 "consume %d pad bytes\n", pad);
250 segment->total_size += pad;
251 segment->size = pad;
252 segment->data = segment->padbuf;
253 return 0;
254 }
255 }
256
257 /*
258 * Set us up for transferring the data digest. hdr digest
259 * is completely handled in hdr done function.
260 */
261 if (segment->hash) {
262 ahash_request_set_crypt(segment->hash, NULL,
263 segment->digest, 0);
264 crypto_ahash_final(segment->hash);
265 iscsi_tcp_segment_splice_digest(segment,
266 recv ? segment->recv_digest : segment->digest);
267 return 0;
268 }
269
270 return 1;
271}
272EXPORT_SYMBOL_GPL(iscsi_tcp_segment_done);
273
274/**
275 * iscsi_tcp_segment_recv - copy data to segment
276 * @tcp_conn: the iSCSI TCP connection
277 * @segment: the buffer to copy to
278 * @ptr: data pointer
279 * @len: amount of data available
280 *
281 * This function copies up to @len bytes to the
282 * given buffer, and returns the number of bytes
283 * consumed, which can actually be less than @len.
284 *
285 * If hash digest is enabled, the function will update the
286 * hash while copying.
287 * Combining these two operations doesn't buy us a lot (yet),
288 * but in the future we could implement combined copy+crc,
289 * just way we do for network layer checksums.
290 */
291static int
292iscsi_tcp_segment_recv(struct iscsi_tcp_conn *tcp_conn,
293 struct iscsi_segment *segment, const void *ptr,
294 unsigned int len)
295{
296 unsigned int copy = 0, copied = 0;
297
298 while (!iscsi_tcp_segment_done(tcp_conn, segment, 1, copy)) {
299 if (copied == len) {
300 ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
301 "copied %d bytes\n", len);
302 break;
303 }
304
305 copy = min(len - copied, segment->size - segment->copied);
306 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copying %d\n", copy);
307 memcpy(segment->data + segment->copied, ptr + copied, copy);
308 copied += copy;
309 }
310 return copied;
311}
312
313inline void
314iscsi_tcp_dgst_header(struct ahash_request *hash, const void *hdr,
315 size_t hdrlen, unsigned char digest[ISCSI_DIGEST_SIZE])
316{
317 struct scatterlist sg;
318
319 sg_init_one(&sg, hdr, hdrlen);
320 ahash_request_set_crypt(hash, &sg, digest, hdrlen);
321 crypto_ahash_digest(hash);
322}
323EXPORT_SYMBOL_GPL(iscsi_tcp_dgst_header);
324
325static inline int
326iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
327 struct iscsi_segment *segment)
328{
329 if (!segment->digest_len)
330 return 1;
331
332 if (memcmp(segment->recv_digest, segment->digest,
333 segment->digest_len)) {
334 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "digest mismatch\n");
335 return 0;
336 }
337
338 return 1;
339}
340
341/*
342 * Helper function to set up segment buffer
343 */
344static inline void
345__iscsi_segment_init(struct iscsi_segment *segment, size_t size,
346 iscsi_segment_done_fn_t *done, struct ahash_request *hash)
347{
348 memset(segment, 0, sizeof(*segment));
349 segment->total_size = size;
350 segment->done = done;
351
352 if (hash) {
353 segment->hash = hash;
354 crypto_ahash_init(hash);
355 }
356}
357
358inline void
359iscsi_segment_init_linear(struct iscsi_segment *segment, void *data,
360 size_t size, iscsi_segment_done_fn_t *done,
361 struct ahash_request *hash)
362{
363 __iscsi_segment_init(segment, size, done, hash);
364 segment->data = data;
365 segment->size = size;
366}
367EXPORT_SYMBOL_GPL(iscsi_segment_init_linear);
368
369inline int
370iscsi_segment_seek_sg(struct iscsi_segment *segment,
371 struct scatterlist *sg_list, unsigned int sg_count,
372 unsigned int offset, size_t size,
373 iscsi_segment_done_fn_t *done,
374 struct ahash_request *hash)
375{
376 struct scatterlist *sg;
377 unsigned int i;
378
379 __iscsi_segment_init(segment, size, done, hash);
380 for_each_sg(sg_list, sg, sg_count, i) {
381 if (offset < sg->length) {
382 iscsi_tcp_segment_init_sg(segment, sg, offset);
383 return 0;
384 }
385 offset -= sg->length;
386 }
387
388 return ISCSI_ERR_DATA_OFFSET;
389}
390EXPORT_SYMBOL_GPL(iscsi_segment_seek_sg);
391
392/**
393 * iscsi_tcp_hdr_recv_prep - prep segment for hdr reception
394 * @tcp_conn: iscsi connection to prep for
395 *
396 * This function always passes NULL for the hash argument, because when this
397 * function is called we do not yet know the final size of the header and want
398 * to delay the digest processing until we know that.
399 */
400void iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
401{
402 ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
403 "(%s)\n", tcp_conn->iscsi_conn->hdrdgst_en ?
404 "digest enabled" : "digest disabled");
405 iscsi_segment_init_linear(&tcp_conn->in.segment,
406 tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
407 iscsi_tcp_hdr_recv_done, NULL);
408}
409EXPORT_SYMBOL_GPL(iscsi_tcp_hdr_recv_prep);
410
411/*
412 * Handle incoming reply to any other type of command
413 */
414static int
415iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
416 struct iscsi_segment *segment)
417{
418 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
419 int rc = 0;
420
421 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
422 return ISCSI_ERR_DATA_DGST;
423
424 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr,
425 conn->data, tcp_conn->in.datalen);
426 if (rc)
427 return rc;
428
429 iscsi_tcp_hdr_recv_prep(tcp_conn);
430 return 0;
431}
432
433static void
434iscsi_tcp_data_recv_prep(struct iscsi_tcp_conn *tcp_conn)
435{
436 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
437 struct ahash_request *rx_hash = NULL;
438
439 if (conn->datadgst_en &&
440 !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
441 rx_hash = tcp_conn->rx_hash;
442
443 iscsi_segment_init_linear(&tcp_conn->in.segment,
444 conn->data, tcp_conn->in.datalen,
445 iscsi_tcp_data_recv_done, rx_hash);
446}
447
448/**
449 * iscsi_tcp_cleanup_task - free tcp_task resources
450 * @task: iscsi task
451 *
452 * must be called with session back_lock
453 */
454void iscsi_tcp_cleanup_task(struct iscsi_task *task)
455{
456 struct iscsi_tcp_task *tcp_task = task->dd_data;
457 struct iscsi_r2t_info *r2t;
458
459 /* nothing to do for mgmt */
460 if (!task->sc)
461 return;
462
463 spin_lock_bh(&tcp_task->queue2pool);
464 /* flush task's r2t queues */
465 while (kfifo_out(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*))) {
466 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
467 sizeof(void*));
468 ISCSI_DBG_TCP(task->conn, "pending r2t dropped\n");
469 }
470
471 r2t = tcp_task->r2t;
472 if (r2t != NULL) {
473 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
474 sizeof(void*));
475 tcp_task->r2t = NULL;
476 }
477 spin_unlock_bh(&tcp_task->queue2pool);
478}
479EXPORT_SYMBOL_GPL(iscsi_tcp_cleanup_task);
480
481/**
482 * iscsi_tcp_data_in - SCSI Data-In Response processing
483 * @conn: iscsi connection
484 * @task: scsi command task
485 */
486static int iscsi_tcp_data_in(struct iscsi_conn *conn, struct iscsi_task *task)
487{
488 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
489 struct iscsi_tcp_task *tcp_task = task->dd_data;
490 struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
491 int datasn = be32_to_cpu(rhdr->datasn);
492 unsigned total_in_length = task->sc->sdb.length;
493
494 /*
495 * lib iscsi will update this in the completion handling if there
496 * is status.
497 */
498 if (!(rhdr->flags & ISCSI_FLAG_DATA_STATUS))
499 iscsi_update_cmdsn(conn->session, (struct iscsi_nopin*)rhdr);
500
501 if (tcp_conn->in.datalen == 0)
502 return 0;
503
504 if (tcp_task->exp_datasn != datasn) {
505 ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->datasn(%d)"
506 "\n", tcp_task->exp_datasn, datasn);
507 return ISCSI_ERR_DATASN;
508 }
509
510 tcp_task->exp_datasn++;
511
512 tcp_task->data_offset = be32_to_cpu(rhdr->offset);
513 if (tcp_task->data_offset + tcp_conn->in.datalen > total_in_length) {
514 ISCSI_DBG_TCP(conn, "data_offset(%d) + data_len(%d) > "
515 "total_length_in(%d)\n", tcp_task->data_offset,
516 tcp_conn->in.datalen, total_in_length);
517 return ISCSI_ERR_DATA_OFFSET;
518 }
519
520 conn->datain_pdus_cnt++;
521 return 0;
522}
523
524/**
525 * iscsi_tcp_r2t_rsp - iSCSI R2T Response processing
526 * @conn: iscsi connection
527 * @hdr: PDU header
528 */
529static int iscsi_tcp_r2t_rsp(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
530{
531 struct iscsi_session *session = conn->session;
532 struct iscsi_tcp_task *tcp_task;
533 struct iscsi_tcp_conn *tcp_conn;
534 struct iscsi_r2t_rsp *rhdr;
535 struct iscsi_r2t_info *r2t;
536 struct iscsi_task *task;
537 u32 data_length;
538 u32 data_offset;
539 int r2tsn;
540 int rc;
541
542 spin_lock(&session->back_lock);
543 task = iscsi_itt_to_ctask(conn, hdr->itt);
544 if (!task) {
545 spin_unlock(&session->back_lock);
546 return ISCSI_ERR_BAD_ITT;
547 } else if (task->sc->sc_data_direction != DMA_TO_DEVICE) {
548 spin_unlock(&session->back_lock);
549 return ISCSI_ERR_PROTO;
550 }
551 /*
552 * A bad target might complete the cmd before we have handled R2Ts
553 * so get a ref to the task that will be dropped in the xmit path.
554 */
555 if (task->state != ISCSI_TASK_RUNNING) {
556 spin_unlock(&session->back_lock);
557 /* Let the path that got the early rsp complete it */
558 return 0;
559 }
560 task->last_xfer = jiffies;
561 __iscsi_get_task(task);
562
563 tcp_conn = conn->dd_data;
564 rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
565 /* fill-in new R2T associated with the task */
566 iscsi_update_cmdsn(session, (struct iscsi_nopin *)rhdr);
567 spin_unlock(&session->back_lock);
568
569 if (tcp_conn->in.datalen) {
570 iscsi_conn_printk(KERN_ERR, conn,
571 "invalid R2t with datalen %d\n",
572 tcp_conn->in.datalen);
573 rc = ISCSI_ERR_DATALEN;
574 goto put_task;
575 }
576
577 tcp_task = task->dd_data;
578 r2tsn = be32_to_cpu(rhdr->r2tsn);
579 if (tcp_task->exp_datasn != r2tsn){
580 ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
581 tcp_task->exp_datasn, r2tsn);
582 rc = ISCSI_ERR_R2TSN;
583 goto put_task;
584 }
585
586 if (session->state != ISCSI_STATE_LOGGED_IN) {
587 iscsi_conn_printk(KERN_INFO, conn,
588 "dropping R2T itt %d in recovery.\n",
589 task->itt);
590 rc = 0;
591 goto put_task;
592 }
593
594 data_length = be32_to_cpu(rhdr->data_length);
595 if (data_length == 0) {
596 iscsi_conn_printk(KERN_ERR, conn,
597 "invalid R2T with zero data len\n");
598 rc = ISCSI_ERR_DATALEN;
599 goto put_task;
600 }
601
602 if (data_length > session->max_burst)
603 ISCSI_DBG_TCP(conn, "invalid R2T with data len %u and max "
604 "burst %u. Attempting to execute request.\n",
605 data_length, session->max_burst);
606
607 data_offset = be32_to_cpu(rhdr->data_offset);
608 if (data_offset + data_length > task->sc->sdb.length) {
609 iscsi_conn_printk(KERN_ERR, conn,
610 "invalid R2T with data len %u at offset %u "
611 "and total length %d\n", data_length,
612 data_offset, task->sc->sdb.length);
613 rc = ISCSI_ERR_DATALEN;
614 goto put_task;
615 }
616
617 spin_lock(&tcp_task->pool2queue);
618 rc = kfifo_out(&tcp_task->r2tpool.queue, (void *)&r2t, sizeof(void *));
619 if (!rc) {
620 iscsi_conn_printk(KERN_ERR, conn, "Could not allocate R2T. "
621 "Target has sent more R2Ts than it "
622 "negotiated for or driver has leaked.\n");
623 spin_unlock(&tcp_task->pool2queue);
624 rc = ISCSI_ERR_PROTO;
625 goto put_task;
626 }
627
628 r2t->exp_statsn = rhdr->statsn;
629 r2t->data_length = data_length;
630 r2t->data_offset = data_offset;
631
632 r2t->ttt = rhdr->ttt; /* no flip */
633 r2t->datasn = 0;
634 r2t->sent = 0;
635
636 tcp_task->exp_datasn = r2tsn + 1;
637 kfifo_in(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*));
638 conn->r2t_pdus_cnt++;
639 spin_unlock(&tcp_task->pool2queue);
640
641 iscsi_requeue_task(task);
642 return 0;
643
644put_task:
645 iscsi_put_task(task);
646 return rc;
647}
648
649/*
650 * Handle incoming reply to DataIn command
651 */
652static int
653iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
654 struct iscsi_segment *segment)
655{
656 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
657 struct iscsi_hdr *hdr = tcp_conn->in.hdr;
658 int rc;
659
660 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
661 return ISCSI_ERR_DATA_DGST;
662
663 /* check for non-exceptional status */
664 if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
665 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
666 if (rc)
667 return rc;
668 }
669
670 iscsi_tcp_hdr_recv_prep(tcp_conn);
671 return 0;
672}
673
674/**
675 * iscsi_tcp_hdr_dissect - process PDU header
676 * @conn: iSCSI connection
677 * @hdr: PDU header
678 *
679 * This function analyzes the header of the PDU received,
680 * and performs several sanity checks. If the PDU is accompanied
681 * by data, the receive buffer is set up to copy the incoming data
682 * to the correct location.
683 */
684static int
685iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
686{
687 int rc = 0, opcode, ahslen;
688 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
689 struct iscsi_task *task;
690
691 /* verify PDU length */
692 tcp_conn->in.datalen = ntoh24(hdr->dlength);
693 if (tcp_conn->in.datalen > conn->max_recv_dlength) {
694 iscsi_conn_printk(KERN_ERR, conn,
695 "iscsi_tcp: datalen %d > %d\n",
696 tcp_conn->in.datalen, conn->max_recv_dlength);
697 return ISCSI_ERR_DATALEN;
698 }
699
700 /* Additional header segments. So far, we don't
701 * process additional headers.
702 */
703 ahslen = hdr->hlength << 2;
704
705 opcode = hdr->opcode & ISCSI_OPCODE_MASK;
706 /* verify itt (itt encoding: age+cid+itt) */
707 rc = iscsi_verify_itt(conn, hdr->itt);
708 if (rc)
709 return rc;
710
711 ISCSI_DBG_TCP(conn, "opcode 0x%x ahslen %d datalen %d\n",
712 opcode, ahslen, tcp_conn->in.datalen);
713
714 switch(opcode) {
715 case ISCSI_OP_SCSI_DATA_IN:
716 spin_lock(&conn->session->back_lock);
717 task = iscsi_itt_to_ctask(conn, hdr->itt);
718 if (!task)
719 rc = ISCSI_ERR_BAD_ITT;
720 else
721 rc = iscsi_tcp_data_in(conn, task);
722 if (rc) {
723 spin_unlock(&conn->session->back_lock);
724 break;
725 }
726
727 if (tcp_conn->in.datalen) {
728 struct iscsi_tcp_task *tcp_task = task->dd_data;
729 struct ahash_request *rx_hash = NULL;
730 struct scsi_data_buffer *sdb = &task->sc->sdb;
731
732 /*
733 * Setup copy of Data-In into the struct scsi_cmnd
734 * Scatterlist case:
735 * We set up the iscsi_segment to point to the next
736 * scatterlist entry to copy to. As we go along,
737 * we move on to the next scatterlist entry and
738 * update the digest per-entry.
739 */
740 if (conn->datadgst_en &&
741 !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
742 rx_hash = tcp_conn->rx_hash;
743
744 ISCSI_DBG_TCP(conn, "iscsi_tcp_begin_data_in( "
745 "offset=%d, datalen=%d)\n",
746 tcp_task->data_offset,
747 tcp_conn->in.datalen);
748 task->last_xfer = jiffies;
749 rc = iscsi_segment_seek_sg(&tcp_conn->in.segment,
750 sdb->table.sgl,
751 sdb->table.nents,
752 tcp_task->data_offset,
753 tcp_conn->in.datalen,
754 iscsi_tcp_process_data_in,
755 rx_hash);
756 spin_unlock(&conn->session->back_lock);
757 return rc;
758 }
759 rc = __iscsi_complete_pdu(conn, hdr, NULL, 0);
760 spin_unlock(&conn->session->back_lock);
761 break;
762 case ISCSI_OP_SCSI_CMD_RSP:
763 if (tcp_conn->in.datalen) {
764 iscsi_tcp_data_recv_prep(tcp_conn);
765 return 0;
766 }
767 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
768 break;
769 case ISCSI_OP_R2T:
770 if (ahslen) {
771 rc = ISCSI_ERR_AHSLEN;
772 break;
773 }
774 rc = iscsi_tcp_r2t_rsp(conn, hdr);
775 break;
776 case ISCSI_OP_LOGIN_RSP:
777 case ISCSI_OP_TEXT_RSP:
778 case ISCSI_OP_REJECT:
779 case ISCSI_OP_ASYNC_EVENT:
780 /*
781 * It is possible that we could get a PDU with a buffer larger
782 * than 8K, but there are no targets that currently do this.
783 * For now we fail until we find a vendor that needs it
784 */
785 if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
786 iscsi_conn_printk(KERN_ERR, conn,
787 "iscsi_tcp: received buffer of "
788 "len %u but conn buffer is only %u "
789 "(opcode %0x)\n",
790 tcp_conn->in.datalen,
791 ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
792 rc = ISCSI_ERR_PROTO;
793 break;
794 }
795
796 /* If there's data coming in with the response,
797 * receive it to the connection's buffer.
798 */
799 if (tcp_conn->in.datalen) {
800 iscsi_tcp_data_recv_prep(tcp_conn);
801 return 0;
802 }
803 fallthrough;
804 case ISCSI_OP_LOGOUT_RSP:
805 case ISCSI_OP_NOOP_IN:
806 case ISCSI_OP_SCSI_TMFUNC_RSP:
807 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
808 break;
809 default:
810 rc = ISCSI_ERR_BAD_OPCODE;
811 break;
812 }
813
814 if (rc == 0) {
815 /* Anything that comes with data should have
816 * been handled above. */
817 if (tcp_conn->in.datalen)
818 return ISCSI_ERR_PROTO;
819 iscsi_tcp_hdr_recv_prep(tcp_conn);
820 }
821
822 return rc;
823}
824
825/**
826 * iscsi_tcp_hdr_recv_done - process PDU header
827 * @tcp_conn: iSCSI TCP connection
828 * @segment: the buffer segment being processed
829 *
830 * This is the callback invoked when the PDU header has
831 * been received. If the header is followed by additional
832 * header segments, we go back for more data.
833 */
834static int
835iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
836 struct iscsi_segment *segment)
837{
838 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
839 struct iscsi_hdr *hdr;
840
841 /* Check if there are additional header segments
842 * *prior* to computing the digest, because we
843 * may need to go back to the caller for more.
844 */
845 hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
846 if (segment->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
847 /* Bump the header length - the caller will
848 * just loop around and get the AHS for us, and
849 * call again. */
850 unsigned int ahslen = hdr->hlength << 2;
851
852 /* Make sure we don't overflow */
853 if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
854 return ISCSI_ERR_AHSLEN;
855
856 segment->total_size += ahslen;
857 segment->size += ahslen;
858 return 0;
859 }
860
861 /* We're done processing the header. See if we're doing
862 * header digests; if so, set up the recv_digest buffer
863 * and go back for more. */
864 if (conn->hdrdgst_en &&
865 !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD)) {
866 if (segment->digest_len == 0) {
867 /*
868 * Even if we offload the digest processing we
869 * splice it in so we can increment the skb/segment
870 * counters in preparation for the data segment.
871 */
872 iscsi_tcp_segment_splice_digest(segment,
873 segment->recv_digest);
874 return 0;
875 }
876
877 iscsi_tcp_dgst_header(tcp_conn->rx_hash, hdr,
878 segment->total_copied - ISCSI_DIGEST_SIZE,
879 segment->digest);
880
881 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
882 return ISCSI_ERR_HDR_DGST;
883 }
884
885 tcp_conn->in.hdr = hdr;
886 return iscsi_tcp_hdr_dissect(conn, hdr);
887}
888
889/**
890 * iscsi_tcp_recv_segment_is_hdr - tests if we are reading in a header
891 * @tcp_conn: iscsi tcp conn
892 *
893 * returns non zero if we are currently processing or setup to process
894 * a header.
895 */
896inline int iscsi_tcp_recv_segment_is_hdr(struct iscsi_tcp_conn *tcp_conn)
897{
898 return tcp_conn->in.segment.done == iscsi_tcp_hdr_recv_done;
899}
900EXPORT_SYMBOL_GPL(iscsi_tcp_recv_segment_is_hdr);
901
902/**
903 * iscsi_tcp_recv_skb - Process skb
904 * @conn: iscsi connection
905 * @skb: network buffer with header and/or data segment
906 * @offset: offset in skb
907 * @offloaded: bool indicating if transfer was offloaded
908 * @status: iscsi TCP status result
909 *
910 * Will return status of transfer in @status. And will return
911 * number of bytes copied.
912 */
913int iscsi_tcp_recv_skb(struct iscsi_conn *conn, struct sk_buff *skb,
914 unsigned int offset, bool offloaded, int *status)
915{
916 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
917 struct iscsi_segment *segment = &tcp_conn->in.segment;
918 struct skb_seq_state seq;
919 unsigned int consumed = 0;
920 int rc = 0;
921
922 ISCSI_DBG_TCP(conn, "in %d bytes\n", skb->len - offset);
923 /*
924 * Update for each skb instead of pdu, because over slow networks a
925 * data_in's data could take a while to read in. We also want to
926 * account for r2ts.
927 */
928 conn->last_recv = jiffies;
929
930 if (unlikely(conn->suspend_rx)) {
931 ISCSI_DBG_TCP(conn, "Rx suspended!\n");
932 *status = ISCSI_TCP_SUSPENDED;
933 return 0;
934 }
935
936 if (offloaded) {
937 segment->total_copied = segment->total_size;
938 goto segment_done;
939 }
940
941 skb_prepare_seq_read(skb, offset, skb->len, &seq);
942 while (1) {
943 unsigned int avail;
944 const u8 *ptr;
945
946 avail = skb_seq_read(consumed, &ptr, &seq);
947 if (avail == 0) {
948 ISCSI_DBG_TCP(conn, "no more data avail. Consumed %d\n",
949 consumed);
950 *status = ISCSI_TCP_SKB_DONE;
951 goto skb_done;
952 }
953 BUG_ON(segment->copied >= segment->size);
954
955 ISCSI_DBG_TCP(conn, "skb %p ptr=%p avail=%u\n", skb, ptr,
956 avail);
957 rc = iscsi_tcp_segment_recv(tcp_conn, segment, ptr, avail);
958 BUG_ON(rc == 0);
959 consumed += rc;
960
961 if (segment->total_copied >= segment->total_size) {
962 skb_abort_seq_read(&seq);
963 goto segment_done;
964 }
965 }
966
967segment_done:
968 *status = ISCSI_TCP_SEGMENT_DONE;
969 ISCSI_DBG_TCP(conn, "segment done\n");
970 rc = segment->done(tcp_conn, segment);
971 if (rc != 0) {
972 *status = ISCSI_TCP_CONN_ERR;
973 ISCSI_DBG_TCP(conn, "Error receiving PDU, errno=%d\n", rc);
974 iscsi_conn_failure(conn, rc);
975 return 0;
976 }
977 /* The done() functions sets up the next segment. */
978
979skb_done:
980 conn->rxdata_octets += consumed;
981 return consumed;
982}
983EXPORT_SYMBOL_GPL(iscsi_tcp_recv_skb);
984
985/**
986 * iscsi_tcp_task_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
987 * @task: scsi command task
988 */
989int iscsi_tcp_task_init(struct iscsi_task *task)
990{
991 struct iscsi_tcp_task *tcp_task = task->dd_data;
992 struct iscsi_conn *conn = task->conn;
993 struct scsi_cmnd *sc = task->sc;
994 int err;
995
996 if (!sc) {
997 /*
998 * mgmt tasks do not have a scatterlist since they come
999 * in from the iscsi interface.
1000 */
1001 ISCSI_DBG_TCP(conn, "mtask deq [itt 0x%x]\n", task->itt);
1002
1003 return conn->session->tt->init_pdu(task, 0, task->data_count);
1004 }
1005
1006 BUG_ON(kfifo_len(&tcp_task->r2tqueue));
1007 tcp_task->exp_datasn = 0;
1008
1009 /* Prepare PDU, optionally w/ immediate data */
1010 ISCSI_DBG_TCP(conn, "task deq [itt 0x%x imm %d unsol %d]\n",
1011 task->itt, task->imm_count, task->unsol_r2t.data_length);
1012
1013 err = conn->session->tt->init_pdu(task, 0, task->imm_count);
1014 if (err)
1015 return err;
1016 task->imm_count = 0;
1017 return 0;
1018}
1019EXPORT_SYMBOL_GPL(iscsi_tcp_task_init);
1020
1021static struct iscsi_r2t_info *iscsi_tcp_get_curr_r2t(struct iscsi_task *task)
1022{
1023 struct iscsi_tcp_task *tcp_task = task->dd_data;
1024 struct iscsi_r2t_info *r2t = NULL;
1025
1026 if (iscsi_task_has_unsol_data(task))
1027 r2t = &task->unsol_r2t;
1028 else {
1029 spin_lock_bh(&tcp_task->queue2pool);
1030 if (tcp_task->r2t) {
1031 r2t = tcp_task->r2t;
1032 /* Continue with this R2T? */
1033 if (r2t->data_length <= r2t->sent) {
1034 ISCSI_DBG_TCP(task->conn,
1035 " done with r2t %p\n", r2t);
1036 kfifo_in(&tcp_task->r2tpool.queue,
1037 (void *)&tcp_task->r2t,
1038 sizeof(void *));
1039 tcp_task->r2t = r2t = NULL;
1040 }
1041 }
1042
1043 if (r2t == NULL) {
1044 if (kfifo_out(&tcp_task->r2tqueue,
1045 (void *)&tcp_task->r2t, sizeof(void *)) !=
1046 sizeof(void *))
1047 r2t = NULL;
1048 else
1049 r2t = tcp_task->r2t;
1050 }
1051 spin_unlock_bh(&tcp_task->queue2pool);
1052 }
1053
1054 return r2t;
1055}
1056
1057/**
1058 * iscsi_tcp_task_xmit - xmit normal PDU task
1059 * @task: iscsi command task
1060 *
1061 * We're expected to return 0 when everything was transmitted successfully,
1062 * -EAGAIN if there's still data in the queue, or != 0 for any other kind
1063 * of error.
1064 */
1065int iscsi_tcp_task_xmit(struct iscsi_task *task)
1066{
1067 struct iscsi_conn *conn = task->conn;
1068 struct iscsi_session *session = conn->session;
1069 struct iscsi_r2t_info *r2t;
1070 int rc = 0;
1071
1072flush:
1073 /* Flush any pending data first. */
1074 rc = session->tt->xmit_pdu(task);
1075 if (rc < 0)
1076 return rc;
1077
1078 /* mgmt command */
1079 if (!task->sc) {
1080 if (task->hdr->itt == RESERVED_ITT)
1081 iscsi_put_task(task);
1082 return 0;
1083 }
1084
1085 /* Are we done already? */
1086 if (task->sc->sc_data_direction != DMA_TO_DEVICE)
1087 return 0;
1088
1089 r2t = iscsi_tcp_get_curr_r2t(task);
1090 if (r2t == NULL) {
1091 /* Waiting for more R2Ts to arrive. */
1092 ISCSI_DBG_TCP(conn, "no R2Ts yet\n");
1093 return 0;
1094 }
1095
1096 rc = conn->session->tt->alloc_pdu(task, ISCSI_OP_SCSI_DATA_OUT);
1097 if (rc)
1098 return rc;
1099 iscsi_prep_data_out_pdu(task, r2t, (struct iscsi_data *) task->hdr);
1100
1101 ISCSI_DBG_TCP(conn, "sol dout %p [dsn %d itt 0x%x doff %d dlen %d]\n",
1102 r2t, r2t->datasn - 1, task->hdr->itt,
1103 r2t->data_offset + r2t->sent, r2t->data_count);
1104
1105 rc = conn->session->tt->init_pdu(task, r2t->data_offset + r2t->sent,
1106 r2t->data_count);
1107 if (rc) {
1108 iscsi_conn_failure(conn, ISCSI_ERR_XMIT_FAILED);
1109 return rc;
1110 }
1111
1112 r2t->sent += r2t->data_count;
1113 goto flush;
1114}
1115EXPORT_SYMBOL_GPL(iscsi_tcp_task_xmit);
1116
1117struct iscsi_cls_conn *
1118iscsi_tcp_conn_setup(struct iscsi_cls_session *cls_session, int dd_data_size,
1119 uint32_t conn_idx)
1120
1121{
1122 struct iscsi_conn *conn;
1123 struct iscsi_cls_conn *cls_conn;
1124 struct iscsi_tcp_conn *tcp_conn;
1125
1126 cls_conn = iscsi_conn_setup(cls_session,
1127 sizeof(*tcp_conn) + dd_data_size, conn_idx);
1128 if (!cls_conn)
1129 return NULL;
1130 conn = cls_conn->dd_data;
1131 /*
1132 * due to strange issues with iser these are not set
1133 * in iscsi_conn_setup
1134 */
1135 conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1136
1137 tcp_conn = conn->dd_data;
1138 tcp_conn->iscsi_conn = conn;
1139 tcp_conn->dd_data = conn->dd_data + sizeof(*tcp_conn);
1140 return cls_conn;
1141}
1142EXPORT_SYMBOL_GPL(iscsi_tcp_conn_setup);
1143
1144void iscsi_tcp_conn_teardown(struct iscsi_cls_conn *cls_conn)
1145{
1146 iscsi_conn_teardown(cls_conn);
1147}
1148EXPORT_SYMBOL_GPL(iscsi_tcp_conn_teardown);
1149
1150int iscsi_tcp_r2tpool_alloc(struct iscsi_session *session)
1151{
1152 int i;
1153 int cmd_i;
1154
1155 /*
1156 * initialize per-task: R2T pool and xmit queue
1157 */
1158 for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1159 struct iscsi_task *task = session->cmds[cmd_i];
1160 struct iscsi_tcp_task *tcp_task = task->dd_data;
1161
1162 /*
1163 * pre-allocated x2 as much r2ts to handle race when
1164 * target acks DataOut faster than we data_xmit() queues
1165 * could replenish r2tqueue.
1166 */
1167
1168 /* R2T pool */
1169 if (iscsi_pool_init(&tcp_task->r2tpool,
1170 session->max_r2t * 2, NULL,
1171 sizeof(struct iscsi_r2t_info))) {
1172 goto r2t_alloc_fail;
1173 }
1174
1175 /* R2T xmit queue */
1176 if (kfifo_alloc(&tcp_task->r2tqueue,
1177 session->max_r2t * 4 * sizeof(void*), GFP_KERNEL)) {
1178 iscsi_pool_free(&tcp_task->r2tpool);
1179 goto r2t_alloc_fail;
1180 }
1181 spin_lock_init(&tcp_task->pool2queue);
1182 spin_lock_init(&tcp_task->queue2pool);
1183 }
1184
1185 return 0;
1186
1187r2t_alloc_fail:
1188 for (i = 0; i < cmd_i; i++) {
1189 struct iscsi_task *task = session->cmds[i];
1190 struct iscsi_tcp_task *tcp_task = task->dd_data;
1191
1192 kfifo_free(&tcp_task->r2tqueue);
1193 iscsi_pool_free(&tcp_task->r2tpool);
1194 }
1195 return -ENOMEM;
1196}
1197EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_alloc);
1198
1199void iscsi_tcp_r2tpool_free(struct iscsi_session *session)
1200{
1201 int i;
1202
1203 for (i = 0; i < session->cmds_max; i++) {
1204 struct iscsi_task *task = session->cmds[i];
1205 struct iscsi_tcp_task *tcp_task = task->dd_data;
1206
1207 kfifo_free(&tcp_task->r2tqueue);
1208 iscsi_pool_free(&tcp_task->r2tpool);
1209 }
1210}
1211EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_free);
1212
1213int iscsi_tcp_set_max_r2t(struct iscsi_conn *conn, char *buf)
1214{
1215 struct iscsi_session *session = conn->session;
1216 unsigned short r2ts = 0;
1217
1218 sscanf(buf, "%hu", &r2ts);
1219 if (session->max_r2t == r2ts)
1220 return 0;
1221
1222 if (!r2ts || !is_power_of_2(r2ts))
1223 return -EINVAL;
1224
1225 session->max_r2t = r2ts;
1226 iscsi_tcp_r2tpool_free(session);
1227 return iscsi_tcp_r2tpool_alloc(session);
1228}
1229EXPORT_SYMBOL_GPL(iscsi_tcp_set_max_r2t);
1230
1231void iscsi_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
1232 struct iscsi_stats *stats)
1233{
1234 struct iscsi_conn *conn = cls_conn->dd_data;
1235
1236 stats->txdata_octets = conn->txdata_octets;
1237 stats->rxdata_octets = conn->rxdata_octets;
1238 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
1239 stats->dataout_pdus = conn->dataout_pdus_cnt;
1240 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
1241 stats->datain_pdus = conn->datain_pdus_cnt;
1242 stats->r2t_pdus = conn->r2t_pdus_cnt;
1243 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
1244 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
1245}
1246EXPORT_SYMBOL_GPL(iscsi_tcp_conn_get_stats);