2 * iSCSI Initiator over TCP/IP Data-Path
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
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.
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.
20 * See the file COPYING included with this distribution for more details.
29 #include <linux/types.h>
30 #include <linux/list.h>
31 #include <linux/inet.h>
32 #include <linux/file.h>
33 #include <linux/blkdev.h>
34 #include <linux/crypto.h>
35 #include <linux/delay.h>
36 #include <linux/kfifo.h>
37 #include <linux/scatterlist.h>
39 #include <scsi/scsi_cmnd.h>
40 #include <scsi/scsi_device.h>
41 #include <scsi/scsi_host.h>
42 #include <scsi/scsi.h>
43 #include <scsi/scsi_transport_iscsi.h>
45 #include "iscsi_tcp.h"
47 MODULE_AUTHOR("Dmitry Yusupov <dmitry_yus@yahoo.com>, "
48 "Alex Aizman <itn780@yahoo.com>");
49 MODULE_DESCRIPTION("iSCSI/TCP data-path");
50 MODULE_LICENSE("GPL");
55 #define debug_tcp(fmt...) printk(KERN_INFO "tcp: " fmt)
57 #define debug_tcp(fmt...)
60 static struct scsi_transport_template *iscsi_tcp_scsi_transport;
61 static struct scsi_host_template iscsi_sht;
62 static struct iscsi_transport iscsi_tcp_transport;
64 static unsigned int iscsi_max_lun = 512;
65 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
67 static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
68 struct iscsi_segment *segment);
71 * Scatterlist handling: inside the iscsi_segment, we
72 * remember an index into the scatterlist, and set data/size
73 * to the current scatterlist entry. For highmem pages, we
76 * Note that the page is unmapped when we return from
77 * TCP's data_ready handler, so we may end up mapping and
78 * unmapping the same page repeatedly. The whole reason
79 * for this is that we shouldn't keep the page mapped
80 * outside the softirq.
84 * iscsi_tcp_segment_init_sg - init indicated scatterlist entry
85 * @segment: the buffer object
87 * @offset: byte offset into that sg entry
89 * This function sets up the segment so that subsequent
90 * data is copied to the indicated sg entry, at the given
94 iscsi_tcp_segment_init_sg(struct iscsi_segment *segment,
95 struct scatterlist *sg, unsigned int offset)
98 segment->sg_offset = offset;
99 segment->size = min(sg->length - offset,
100 segment->total_size - segment->total_copied);
101 segment->data = NULL;
105 * iscsi_tcp_segment_map - map the current S/G page
106 * @segment: iscsi_segment
107 * @recv: 1 if called from recv path
109 * We only need to possibly kmap data if scatter lists are being used,
110 * because the iscsi passthrough and internal IO paths will never use high
114 iscsi_tcp_segment_map(struct iscsi_segment *segment, int recv)
116 struct scatterlist *sg;
118 if (segment->data != NULL || !segment->sg)
122 BUG_ON(segment->sg_mapped);
123 BUG_ON(sg->length == 0);
126 * If the page count is greater than one it is ok to send
127 * to the network layer's zero copy send path. If not we
128 * have to go the slow sendmsg path. We always map for the
131 if (page_count(sg_page(sg)) >= 1 && !recv)
134 debug_tcp("iscsi_tcp_segment_map %s %p\n", recv ? "recv" : "xmit",
136 segment->sg_mapped = kmap_atomic(sg_page(sg), KM_SOFTIRQ0);
137 segment->data = segment->sg_mapped + sg->offset + segment->sg_offset;
141 iscsi_tcp_segment_unmap(struct iscsi_segment *segment)
143 debug_tcp("iscsi_tcp_segment_unmap %p\n", segment);
145 if (segment->sg_mapped) {
146 debug_tcp("iscsi_tcp_segment_unmap valid\n");
147 kunmap_atomic(segment->sg_mapped, KM_SOFTIRQ0);
148 segment->sg_mapped = NULL;
149 segment->data = NULL;
154 * Splice the digest buffer into the buffer
157 iscsi_tcp_segment_splice_digest(struct iscsi_segment *segment, void *digest)
159 segment->data = digest;
160 segment->digest_len = ISCSI_DIGEST_SIZE;
161 segment->total_size += ISCSI_DIGEST_SIZE;
162 segment->size = ISCSI_DIGEST_SIZE;
165 segment->hash = NULL;
169 * iscsi_tcp_segment_done - check whether the segment is complete
170 * @segment: iscsi segment to check
171 * @recv: set to one of this is called from the recv path
172 * @copied: number of bytes copied
174 * Check if we're done receiving this segment. If the receive
175 * buffer is full but we expect more data, move on to the
176 * next entry in the scatterlist.
178 * If the amount of data we received isn't a multiple of 4,
179 * we will transparently receive the pad bytes, too.
181 * This function must be re-entrant.
184 iscsi_tcp_segment_done(struct iscsi_segment *segment, int recv, unsigned copied)
186 static unsigned char padbuf[ISCSI_PAD_LEN];
187 struct scatterlist sg;
190 debug_tcp("copied %u %u size %u %s\n", segment->copied, copied,
191 segment->size, recv ? "recv" : "xmit");
192 if (segment->hash && copied) {
194 * If a segment is kmapd we must unmap it before sending
195 * to the crypto layer since that will try to kmap it again.
197 iscsi_tcp_segment_unmap(segment);
199 if (!segment->data) {
200 sg_init_table(&sg, 1);
201 sg_set_page(&sg, sg_page(segment->sg), copied,
202 segment->copied + segment->sg_offset +
203 segment->sg->offset);
205 sg_init_one(&sg, segment->data + segment->copied,
207 crypto_hash_update(segment->hash, &sg, copied);
210 segment->copied += copied;
211 if (segment->copied < segment->size) {
212 iscsi_tcp_segment_map(segment, recv);
216 segment->total_copied += segment->copied;
220 /* Unmap the current scatterlist page, if there is one. */
221 iscsi_tcp_segment_unmap(segment);
223 /* Do we have more scatterlist entries? */
224 debug_tcp("total copied %u total size %u\n", segment->total_copied,
225 segment->total_size);
226 if (segment->total_copied < segment->total_size) {
227 /* Proceed to the next entry in the scatterlist. */
228 iscsi_tcp_segment_init_sg(segment, sg_next(segment->sg),
230 iscsi_tcp_segment_map(segment, recv);
231 BUG_ON(segment->size == 0);
235 /* Do we need to handle padding? */
236 pad = iscsi_padding(segment->total_copied);
238 debug_tcp("consume %d pad bytes\n", pad);
239 segment->total_size += pad;
241 segment->data = padbuf;
246 * Set us up for transferring the data digest. hdr digest
247 * is completely handled in hdr done function.
250 crypto_hash_final(segment->hash, segment->digest);
251 iscsi_tcp_segment_splice_digest(segment,
252 recv ? segment->recv_digest : segment->digest);
260 * iscsi_tcp_xmit_segment - transmit segment
261 * @tcp_conn: the iSCSI TCP connection
262 * @segment: the buffer to transmnit
264 * This function transmits as much of the buffer as
265 * the network layer will accept, and returns the number of
268 * If CRC hashing is enabled, the function will compute the
269 * hash as it goes. When the entire segment has been transmitted,
270 * it will retrieve the hash value and send it as well.
273 iscsi_tcp_xmit_segment(struct iscsi_tcp_conn *tcp_conn,
274 struct iscsi_segment *segment)
276 struct socket *sk = tcp_conn->sock;
277 unsigned int copied = 0;
280 while (!iscsi_tcp_segment_done(segment, 0, r)) {
281 struct scatterlist *sg;
282 unsigned int offset, copy;
286 offset = segment->copied;
287 copy = segment->size - offset;
289 if (segment->total_copied + segment->size < segment->total_size)
292 /* Use sendpage if we can; else fall back to sendmsg */
293 if (!segment->data) {
295 offset += segment->sg_offset + sg->offset;
296 r = tcp_conn->sendpage(sk, sg_page(sg), offset, copy,
299 struct msghdr msg = { .msg_flags = flags };
301 .iov_base = segment->data + offset,
305 r = kernel_sendmsg(sk, &msg, &iov, 1, copy);
309 iscsi_tcp_segment_unmap(segment);
310 if (copied || r == -EAGAIN)
320 * iscsi_tcp_segment_recv - copy data to segment
321 * @tcp_conn: the iSCSI TCP connection
322 * @segment: the buffer to copy to
324 * @len: amount of data available
326 * This function copies up to @len bytes to the
327 * given buffer, and returns the number of bytes
328 * consumed, which can actually be less than @len.
330 * If hash digest is enabled, the function will update the
331 * hash while copying.
332 * Combining these two operations doesn't buy us a lot (yet),
333 * but in the future we could implement combined copy+crc,
334 * just way we do for network layer checksums.
337 iscsi_tcp_segment_recv(struct iscsi_tcp_conn *tcp_conn,
338 struct iscsi_segment *segment, const void *ptr,
341 unsigned int copy = 0, copied = 0;
343 while (!iscsi_tcp_segment_done(segment, 1, copy)) {
345 debug_tcp("iscsi_tcp_segment_recv copied %d bytes\n",
350 copy = min(len - copied, segment->size - segment->copied);
351 debug_tcp("iscsi_tcp_segment_recv copying %d\n", copy);
352 memcpy(segment->data + segment->copied, ptr + copied, copy);
359 iscsi_tcp_dgst_header(struct hash_desc *hash, const void *hdr, size_t hdrlen,
360 unsigned char digest[ISCSI_DIGEST_SIZE])
362 struct scatterlist sg;
364 sg_init_one(&sg, hdr, hdrlen);
365 crypto_hash_digest(hash, &sg, hdrlen, digest);
369 iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
370 struct iscsi_segment *segment)
372 if (!segment->digest_len)
375 if (memcmp(segment->recv_digest, segment->digest,
376 segment->digest_len)) {
377 debug_scsi("digest mismatch\n");
385 * Helper function to set up segment buffer
388 __iscsi_segment_init(struct iscsi_segment *segment, size_t size,
389 iscsi_segment_done_fn_t *done, struct hash_desc *hash)
391 memset(segment, 0, sizeof(*segment));
392 segment->total_size = size;
393 segment->done = done;
396 segment->hash = hash;
397 crypto_hash_init(hash);
402 iscsi_segment_init_linear(struct iscsi_segment *segment, void *data,
403 size_t size, iscsi_segment_done_fn_t *done,
404 struct hash_desc *hash)
406 __iscsi_segment_init(segment, size, done, hash);
407 segment->data = data;
408 segment->size = size;
412 iscsi_segment_seek_sg(struct iscsi_segment *segment,
413 struct scatterlist *sg_list, unsigned int sg_count,
414 unsigned int offset, size_t size,
415 iscsi_segment_done_fn_t *done, struct hash_desc *hash)
417 struct scatterlist *sg;
420 debug_scsi("iscsi_segment_seek_sg offset %u size %llu\n",
422 __iscsi_segment_init(segment, size, done, hash);
423 for_each_sg(sg_list, sg, sg_count, i) {
424 debug_scsi("sg %d, len %u offset %u\n", i, sg->length,
426 if (offset < sg->length) {
427 iscsi_tcp_segment_init_sg(segment, sg, offset);
430 offset -= sg->length;
433 return ISCSI_ERR_DATA_OFFSET;
437 * iscsi_tcp_hdr_recv_prep - prep segment for hdr reception
438 * @tcp_conn: iscsi connection to prep for
440 * This function always passes NULL for the hash argument, because when this
441 * function is called we do not yet know the final size of the header and want
442 * to delay the digest processing until we know that.
445 iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
447 debug_tcp("iscsi_tcp_hdr_recv_prep(%p%s)\n", tcp_conn,
448 tcp_conn->iscsi_conn->hdrdgst_en ? ", digest enabled" : "");
449 iscsi_segment_init_linear(&tcp_conn->in.segment,
450 tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
451 iscsi_tcp_hdr_recv_done, NULL);
455 * Handle incoming reply to any other type of command
458 iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
459 struct iscsi_segment *segment)
461 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
464 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
465 return ISCSI_ERR_DATA_DGST;
467 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr,
468 conn->data, tcp_conn->in.datalen);
472 iscsi_tcp_hdr_recv_prep(tcp_conn);
477 iscsi_tcp_data_recv_prep(struct iscsi_tcp_conn *tcp_conn)
479 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
480 struct hash_desc *rx_hash = NULL;
482 if (conn->datadgst_en &
483 !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
484 rx_hash = &tcp_conn->rx_hash;
486 iscsi_segment_init_linear(&tcp_conn->in.segment,
487 conn->data, tcp_conn->in.datalen,
488 iscsi_tcp_data_recv_done, rx_hash);
492 * must be called with session lock
494 static void iscsi_tcp_cleanup_task(struct iscsi_task *task)
496 struct iscsi_tcp_task *tcp_task = task->dd_data;
497 struct iscsi_r2t_info *r2t;
499 /* nothing to do for mgmt or pending tasks */
500 if (!task->sc || task->state == ISCSI_TASK_PENDING)
503 /* flush task's r2t queues */
504 while (__kfifo_get(tcp_task->r2tqueue, (void*)&r2t, sizeof(void*))) {
505 __kfifo_put(tcp_task->r2tpool.queue, (void*)&r2t,
507 debug_scsi("iscsi_tcp_cleanup_task pending r2t dropped\n");
512 __kfifo_put(tcp_task->r2tpool.queue, (void*)&r2t,
514 tcp_task->r2t = NULL;
519 * iscsi_data_in - SCSI Data-In Response processing
520 * @conn: iscsi connection
521 * @task: scsi command task
524 iscsi_data_in(struct iscsi_conn *conn, struct iscsi_task *task)
526 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
527 struct iscsi_tcp_task *tcp_task = task->dd_data;
528 struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
529 int datasn = be32_to_cpu(rhdr->datasn);
530 unsigned total_in_length = scsi_in(task->sc)->length;
532 iscsi_update_cmdsn(conn->session, (struct iscsi_nopin*)rhdr);
533 if (tcp_conn->in.datalen == 0)
536 if (tcp_task->exp_datasn != datasn) {
537 debug_tcp("%s: task->exp_datasn(%d) != rhdr->datasn(%d)\n",
538 __func__, tcp_task->exp_datasn, datasn);
539 return ISCSI_ERR_DATASN;
542 tcp_task->exp_datasn++;
544 tcp_task->data_offset = be32_to_cpu(rhdr->offset);
545 if (tcp_task->data_offset + tcp_conn->in.datalen > total_in_length) {
546 debug_tcp("%s: data_offset(%d) + data_len(%d) > total_length_in(%d)\n",
547 __func__, tcp_task->data_offset,
548 tcp_conn->in.datalen, total_in_length);
549 return ISCSI_ERR_DATA_OFFSET;
552 conn->datain_pdus_cnt++;
557 * iscsi_solicit_data_init - initialize first Data-Out
558 * @conn: iscsi connection
559 * @task: scsi command task
563 * Initialize first Data-Out within this R2T sequence and finds
564 * proper data_offset within this SCSI command.
566 * This function is called with connection lock taken.
569 iscsi_solicit_data_init(struct iscsi_conn *conn, struct iscsi_task *task,
570 struct iscsi_r2t_info *r2t)
572 struct iscsi_data *hdr;
574 hdr = &r2t->dtask.hdr;
575 memset(hdr, 0, sizeof(struct iscsi_data));
577 hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
578 r2t->solicit_datasn++;
579 hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
580 memcpy(hdr->lun, task->hdr->lun, sizeof(hdr->lun));
581 hdr->itt = task->hdr->itt;
582 hdr->exp_statsn = r2t->exp_statsn;
583 hdr->offset = cpu_to_be32(r2t->data_offset);
584 if (r2t->data_length > conn->max_xmit_dlength) {
585 hton24(hdr->dlength, conn->max_xmit_dlength);
586 r2t->data_count = conn->max_xmit_dlength;
589 hton24(hdr->dlength, r2t->data_length);
590 r2t->data_count = r2t->data_length;
591 hdr->flags = ISCSI_FLAG_CMD_FINAL;
593 conn->dataout_pdus_cnt++;
599 * iscsi_r2t_rsp - iSCSI R2T Response processing
600 * @conn: iscsi connection
601 * @task: scsi command task
604 iscsi_r2t_rsp(struct iscsi_conn *conn, struct iscsi_task *task)
606 struct iscsi_session *session = conn->session;
607 struct iscsi_tcp_task *tcp_task = task->dd_data;
608 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
609 struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
610 struct iscsi_r2t_info *r2t;
611 int r2tsn = be32_to_cpu(rhdr->r2tsn);
614 if (tcp_conn->in.datalen) {
615 iscsi_conn_printk(KERN_ERR, conn,
616 "invalid R2t with datalen %d\n",
617 tcp_conn->in.datalen);
618 return ISCSI_ERR_DATALEN;
621 if (tcp_task->exp_datasn != r2tsn){
622 debug_tcp("%s: task->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
623 __func__, tcp_task->exp_datasn, r2tsn);
624 return ISCSI_ERR_R2TSN;
627 /* fill-in new R2T associated with the task */
628 iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
630 if (!task->sc || session->state != ISCSI_STATE_LOGGED_IN) {
631 iscsi_conn_printk(KERN_INFO, conn,
632 "dropping R2T itt %d in recovery.\n",
637 rc = __kfifo_get(tcp_task->r2tpool.queue, (void*)&r2t, sizeof(void*));
639 iscsi_conn_printk(KERN_ERR, conn, "Could not allocate R2T. "
640 "Target has sent more R2Ts than it "
641 "negotiated for or driver has has leaked.\n");
642 return ISCSI_ERR_PROTO;
645 r2t->exp_statsn = rhdr->statsn;
646 r2t->data_length = be32_to_cpu(rhdr->data_length);
647 if (r2t->data_length == 0) {
648 iscsi_conn_printk(KERN_ERR, conn,
649 "invalid R2T with zero data len\n");
650 __kfifo_put(tcp_task->r2tpool.queue, (void*)&r2t,
652 return ISCSI_ERR_DATALEN;
655 if (r2t->data_length > session->max_burst)
656 debug_scsi("invalid R2T with data len %u and max burst %u."
657 "Attempting to execute request.\n",
658 r2t->data_length, session->max_burst);
660 r2t->data_offset = be32_to_cpu(rhdr->data_offset);
661 if (r2t->data_offset + r2t->data_length > scsi_out(task->sc)->length) {
662 iscsi_conn_printk(KERN_ERR, conn,
663 "invalid R2T with data len %u at offset %u "
664 "and total length %d\n", r2t->data_length,
665 r2t->data_offset, scsi_out(task->sc)->length);
666 __kfifo_put(tcp_task->r2tpool.queue, (void*)&r2t,
668 return ISCSI_ERR_DATALEN;
671 r2t->ttt = rhdr->ttt; /* no flip */
675 tcp_task->exp_datasn = r2tsn + 1;
676 __kfifo_put(tcp_task->r2tqueue, (void*)&r2t, sizeof(void*));
677 conn->r2t_pdus_cnt++;
679 iscsi_requeue_task(task);
684 * Handle incoming reply to DataIn command
687 iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
688 struct iscsi_segment *segment)
690 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
691 struct iscsi_hdr *hdr = tcp_conn->in.hdr;
694 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
695 return ISCSI_ERR_DATA_DGST;
697 /* check for non-exceptional status */
698 if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
699 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
704 iscsi_tcp_hdr_recv_prep(tcp_conn);
709 * iscsi_tcp_hdr_dissect - process PDU header
710 * @conn: iSCSI connection
713 * This function analyzes the header of the PDU received,
714 * and performs several sanity checks. If the PDU is accompanied
715 * by data, the receive buffer is set up to copy the incoming data
716 * to the correct location.
719 iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
721 int rc = 0, opcode, ahslen;
722 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
723 struct iscsi_task *task;
725 /* verify PDU length */
726 tcp_conn->in.datalen = ntoh24(hdr->dlength);
727 if (tcp_conn->in.datalen > conn->max_recv_dlength) {
728 iscsi_conn_printk(KERN_ERR, conn,
729 "iscsi_tcp: datalen %d > %d\n",
730 tcp_conn->in.datalen, conn->max_recv_dlength);
731 return ISCSI_ERR_DATALEN;
734 /* Additional header segments. So far, we don't
735 * process additional headers.
737 ahslen = hdr->hlength << 2;
739 opcode = hdr->opcode & ISCSI_OPCODE_MASK;
740 /* verify itt (itt encoding: age+cid+itt) */
741 rc = iscsi_verify_itt(conn, hdr->itt);
745 debug_tcp("opcode 0x%x ahslen %d datalen %d\n",
746 opcode, ahslen, tcp_conn->in.datalen);
749 case ISCSI_OP_SCSI_DATA_IN:
750 spin_lock(&conn->session->lock);
751 task = iscsi_itt_to_ctask(conn, hdr->itt);
753 rc = ISCSI_ERR_BAD_ITT;
755 rc = iscsi_data_in(conn, task);
757 spin_unlock(&conn->session->lock);
761 if (tcp_conn->in.datalen) {
762 struct iscsi_tcp_task *tcp_task = task->dd_data;
763 struct hash_desc *rx_hash = NULL;
764 struct scsi_data_buffer *sdb = scsi_in(task->sc);
767 * Setup copy of Data-In into the Scsi_Cmnd
769 * We set up the iscsi_segment to point to the next
770 * scatterlist entry to copy to. As we go along,
771 * we move on to the next scatterlist entry and
772 * update the digest per-entry.
774 if (conn->datadgst_en &&
775 !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
776 rx_hash = &tcp_conn->rx_hash;
778 debug_tcp("iscsi_tcp_begin_data_in(%p, offset=%d, "
779 "datalen=%d)\n", tcp_conn,
780 tcp_task->data_offset,
781 tcp_conn->in.datalen);
782 rc = iscsi_segment_seek_sg(&tcp_conn->in.segment,
785 tcp_task->data_offset,
786 tcp_conn->in.datalen,
787 iscsi_tcp_process_data_in,
789 spin_unlock(&conn->session->lock);
792 rc = __iscsi_complete_pdu(conn, hdr, NULL, 0);
793 spin_unlock(&conn->session->lock);
795 case ISCSI_OP_SCSI_CMD_RSP:
796 if (tcp_conn->in.datalen) {
797 iscsi_tcp_data_recv_prep(tcp_conn);
800 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
803 spin_lock(&conn->session->lock);
804 task = iscsi_itt_to_ctask(conn, hdr->itt);
806 rc = ISCSI_ERR_BAD_ITT;
808 rc = ISCSI_ERR_AHSLEN;
809 else if (task->sc->sc_data_direction == DMA_TO_DEVICE)
810 rc = iscsi_r2t_rsp(conn, task);
812 rc = ISCSI_ERR_PROTO;
813 spin_unlock(&conn->session->lock);
815 case ISCSI_OP_LOGIN_RSP:
816 case ISCSI_OP_TEXT_RSP:
817 case ISCSI_OP_REJECT:
818 case ISCSI_OP_ASYNC_EVENT:
820 * It is possible that we could get a PDU with a buffer larger
821 * than 8K, but there are no targets that currently do this.
822 * For now we fail until we find a vendor that needs it
824 if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
825 iscsi_conn_printk(KERN_ERR, conn,
826 "iscsi_tcp: received buffer of "
827 "len %u but conn buffer is only %u "
829 tcp_conn->in.datalen,
830 ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
831 rc = ISCSI_ERR_PROTO;
835 /* If there's data coming in with the response,
836 * receive it to the connection's buffer.
838 if (tcp_conn->in.datalen) {
839 iscsi_tcp_data_recv_prep(tcp_conn);
843 case ISCSI_OP_LOGOUT_RSP:
844 case ISCSI_OP_NOOP_IN:
845 case ISCSI_OP_SCSI_TMFUNC_RSP:
846 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
849 rc = ISCSI_ERR_BAD_OPCODE;
854 /* Anything that comes with data should have
855 * been handled above. */
856 if (tcp_conn->in.datalen)
857 return ISCSI_ERR_PROTO;
858 iscsi_tcp_hdr_recv_prep(tcp_conn);
865 * iscsi_tcp_hdr_recv_done - process PDU header
867 * This is the callback invoked when the PDU header has
868 * been received. If the header is followed by additional
869 * header segments, we go back for more data.
872 iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
873 struct iscsi_segment *segment)
875 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
876 struct iscsi_hdr *hdr;
878 /* Check if there are additional header segments
879 * *prior* to computing the digest, because we
880 * may need to go back to the caller for more.
882 hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
883 if (segment->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
884 /* Bump the header length - the caller will
885 * just loop around and get the AHS for us, and
887 unsigned int ahslen = hdr->hlength << 2;
889 /* Make sure we don't overflow */
890 if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
891 return ISCSI_ERR_AHSLEN;
893 segment->total_size += ahslen;
894 segment->size += ahslen;
898 /* We're done processing the header. See if we're doing
899 * header digests; if so, set up the recv_digest buffer
900 * and go back for more. */
901 if (conn->hdrdgst_en) {
902 if (segment->digest_len == 0) {
904 * Even if we offload the digest processing we
905 * splice it in so we can increment the skb/segment
906 * counters in preparation for the data segment.
908 iscsi_tcp_segment_splice_digest(segment,
909 segment->recv_digest);
913 if (!(conn->session->tt->caps & CAP_DIGEST_OFFLOAD)) {
914 iscsi_tcp_dgst_header(&tcp_conn->rx_hash, hdr,
915 segment->total_copied - ISCSI_DIGEST_SIZE,
918 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
919 return ISCSI_ERR_HDR_DGST;
923 tcp_conn->in.hdr = hdr;
924 return iscsi_tcp_hdr_dissect(conn, hdr);
927 inline int iscsi_tcp_recv_segment_is_hdr(struct iscsi_tcp_conn *tcp_conn)
929 return tcp_conn->in.segment.done == iscsi_tcp_hdr_recv_done;
933 ISCSI_TCP_SEGMENT_DONE, /* curr seg has been processed */
934 ISCSI_TCP_SKB_DONE, /* skb is out of data */
935 ISCSI_TCP_CONN_ERR, /* iscsi layer has fired a conn err */
936 ISCSI_TCP_SUSPENDED, /* conn is suspended */
940 * iscsi_tcp_recv_skb - Process skb
941 * @conn: iscsi connection
942 * @skb: network buffer with header and/or data segment
943 * @offset: offset in skb
944 * @offload: bool indicating if transfer was offloaded
946 int iscsi_tcp_recv_skb(struct iscsi_conn *conn, struct sk_buff *skb,
947 unsigned int offset, bool offloaded, int *status)
949 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
950 struct iscsi_segment *segment = &tcp_conn->in.segment;
951 struct skb_seq_state seq;
952 unsigned int consumed = 0;
955 debug_tcp("in %d bytes\n", skb->len - offset);
957 if (unlikely(conn->suspend_rx)) {
958 debug_tcp("conn %d Rx suspended!\n", conn->id);
959 *status = ISCSI_TCP_SUSPENDED;
964 segment->total_copied = segment->total_size;
968 skb_prepare_seq_read(skb, offset, skb->len, &seq);
973 avail = skb_seq_read(consumed, &ptr, &seq);
975 debug_tcp("no more data avail. Consumed %d\n",
977 *status = ISCSI_TCP_SKB_DONE;
978 skb_abort_seq_read(&seq);
981 BUG_ON(segment->copied >= segment->size);
983 debug_tcp("skb %p ptr=%p avail=%u\n", skb, ptr, avail);
984 rc = iscsi_tcp_segment_recv(tcp_conn, segment, ptr, avail);
988 if (segment->total_copied >= segment->total_size) {
989 skb_abort_seq_read(&seq);
995 *status = ISCSI_TCP_SEGMENT_DONE;
996 debug_tcp("segment done\n");
997 rc = segment->done(tcp_conn, segment);
999 *status = ISCSI_TCP_CONN_ERR;
1000 debug_tcp("Error receiving PDU, errno=%d\n", rc);
1001 iscsi_conn_failure(conn, rc);
1004 /* The done() functions sets up the next segment. */
1007 conn->rxdata_octets += consumed;
1010 EXPORT_SYMBOL_GPL(iscsi_tcp_recv_skb);
1013 * iscsi_tcp_recv - TCP receive in sendfile fashion
1014 * @rd_desc: read descriptor
1015 * @skb: socket buffer
1016 * @offset: offset in skb
1017 * @len: skb->len - offset
1020 iscsi_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
1021 unsigned int offset, size_t len)
1023 struct iscsi_conn *conn = rd_desc->arg.data;
1024 unsigned int consumed, total_consumed = 0;
1027 debug_tcp("in %d bytes\n", skb->len - offset);
1031 consumed = iscsi_tcp_recv_skb(conn, skb, offset, 0, &status);
1033 total_consumed += consumed;
1034 } while (consumed != 0 && status != ISCSI_TCP_SKB_DONE);
1036 debug_tcp("read %d bytes status %d\n", skb->len - offset, status);
1037 return total_consumed;
1041 iscsi_tcp_data_ready(struct sock *sk, int flag)
1043 struct iscsi_conn *conn = sk->sk_user_data;
1044 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1045 read_descriptor_t rd_desc;
1047 read_lock(&sk->sk_callback_lock);
1050 * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
1051 * We set count to 1 because we want the network layer to
1052 * hand us all the skbs that are available. iscsi_tcp_recv
1053 * handled pdus that cross buffers or pdus that still need data.
1055 rd_desc.arg.data = conn;
1057 tcp_read_sock(sk, &rd_desc, iscsi_tcp_recv);
1059 read_unlock(&sk->sk_callback_lock);
1061 /* If we had to (atomically) map a highmem page,
1063 iscsi_tcp_segment_unmap(&tcp_conn->in.segment);
1067 iscsi_tcp_state_change(struct sock *sk)
1069 struct iscsi_tcp_conn *tcp_conn;
1070 struct iscsi_conn *conn;
1071 struct iscsi_session *session;
1072 void (*old_state_change)(struct sock *);
1074 read_lock(&sk->sk_callback_lock);
1076 conn = (struct iscsi_conn*)sk->sk_user_data;
1077 session = conn->session;
1079 if ((sk->sk_state == TCP_CLOSE_WAIT ||
1080 sk->sk_state == TCP_CLOSE) &&
1081 !atomic_read(&sk->sk_rmem_alloc)) {
1082 debug_tcp("iscsi_tcp_state_change: TCP_CLOSE|TCP_CLOSE_WAIT\n");
1083 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1086 tcp_conn = conn->dd_data;
1087 old_state_change = tcp_conn->old_state_change;
1089 read_unlock(&sk->sk_callback_lock);
1091 old_state_change(sk);
1095 * iscsi_write_space - Called when more output buffer space is available
1096 * @sk: socket space is available for
1099 iscsi_write_space(struct sock *sk)
1101 struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
1102 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1104 tcp_conn->old_write_space(sk);
1105 debug_tcp("iscsi_write_space: cid %d\n", conn->id);
1106 scsi_queue_work(conn->session->host, &conn->xmitwork);
1110 iscsi_conn_set_callbacks(struct iscsi_conn *conn)
1112 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1113 struct sock *sk = tcp_conn->sock->sk;
1115 /* assign new callbacks */
1116 write_lock_bh(&sk->sk_callback_lock);
1117 sk->sk_user_data = conn;
1118 tcp_conn->old_data_ready = sk->sk_data_ready;
1119 tcp_conn->old_state_change = sk->sk_state_change;
1120 tcp_conn->old_write_space = sk->sk_write_space;
1121 sk->sk_data_ready = iscsi_tcp_data_ready;
1122 sk->sk_state_change = iscsi_tcp_state_change;
1123 sk->sk_write_space = iscsi_write_space;
1124 write_unlock_bh(&sk->sk_callback_lock);
1128 iscsi_conn_restore_callbacks(struct iscsi_tcp_conn *tcp_conn)
1130 struct sock *sk = tcp_conn->sock->sk;
1132 /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
1133 write_lock_bh(&sk->sk_callback_lock);
1134 sk->sk_user_data = NULL;
1135 sk->sk_data_ready = tcp_conn->old_data_ready;
1136 sk->sk_state_change = tcp_conn->old_state_change;
1137 sk->sk_write_space = tcp_conn->old_write_space;
1138 sk->sk_no_check = 0;
1139 write_unlock_bh(&sk->sk_callback_lock);
1143 * iscsi_xmit - TCP transmit
1146 iscsi_xmit(struct iscsi_conn *conn)
1148 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1149 struct iscsi_segment *segment = &tcp_conn->out.segment;
1150 unsigned int consumed = 0;
1154 rc = iscsi_tcp_xmit_segment(tcp_conn, segment);
1156 rc = ISCSI_ERR_XMIT_FAILED;
1164 if (segment->total_copied >= segment->total_size) {
1165 if (segment->done != NULL) {
1166 rc = segment->done(tcp_conn, segment);
1173 debug_tcp("xmit %d bytes\n", consumed);
1175 conn->txdata_octets += consumed;
1179 /* Transmit error. We could initiate error recovery
1181 debug_tcp("Error sending PDU, errno=%d\n", rc);
1182 iscsi_conn_failure(conn, rc);
1187 * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
1190 iscsi_tcp_xmit_qlen(struct iscsi_conn *conn)
1192 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1193 struct iscsi_segment *segment = &tcp_conn->out.segment;
1195 return segment->total_copied - segment->total_size;
1198 static int iscsi_tcp_flush(struct iscsi_task *task)
1200 struct iscsi_conn *conn = task->conn;
1203 while (iscsi_tcp_xmit_qlen(conn)) {
1204 rc = iscsi_xmit(conn);
1215 * This is called when we're done sending the header.
1216 * Simply copy the data_segment to the send segment, and return.
1219 iscsi_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
1220 struct iscsi_segment *segment)
1222 tcp_conn->out.segment = tcp_conn->out.data_segment;
1223 debug_tcp("Header done. Next segment size %u total_size %u\n",
1224 tcp_conn->out.segment.size, tcp_conn->out.segment.total_size);
1229 iscsi_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr, size_t hdrlen)
1231 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1233 debug_tcp("%s(%p%s)\n", __func__, tcp_conn,
1234 conn->hdrdgst_en? ", digest enabled" : "");
1236 /* Clear the data segment - needs to be filled in by the
1237 * caller using iscsi_tcp_send_data_prep() */
1238 memset(&tcp_conn->out.data_segment, 0, sizeof(struct iscsi_segment));
1240 /* If header digest is enabled, compute the CRC and
1241 * place the digest into the same buffer. We make
1242 * sure that both iscsi_tcp_task and mtask have
1245 if (conn->hdrdgst_en) {
1246 iscsi_tcp_dgst_header(&tcp_conn->tx_hash, hdr, hdrlen,
1248 hdrlen += ISCSI_DIGEST_SIZE;
1251 /* Remember header pointer for later, when we need
1252 * to decide whether there's a payload to go along
1253 * with the header. */
1254 tcp_conn->out.hdr = hdr;
1256 iscsi_segment_init_linear(&tcp_conn->out.segment, hdr, hdrlen,
1257 iscsi_tcp_send_hdr_done, NULL);
1261 * Prepare the send buffer for the payload data.
1262 * Padding and checksumming will all be taken care
1263 * of by the iscsi_segment routines.
1266 iscsi_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
1267 unsigned int count, unsigned int offset,
1270 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1271 struct hash_desc *tx_hash = NULL;
1272 unsigned int hdr_spec_len;
1274 debug_tcp("%s(%p, offset=%d, datalen=%d%s)\n", __func__,
1275 tcp_conn, offset, len,
1276 conn->datadgst_en? ", digest enabled" : "");
1278 /* Make sure the datalen matches what the caller
1279 said he would send. */
1280 hdr_spec_len = ntoh24(tcp_conn->out.hdr->dlength);
1281 WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
1283 if (conn->datadgst_en)
1284 tx_hash = &tcp_conn->tx_hash;
1286 return iscsi_segment_seek_sg(&tcp_conn->out.data_segment,
1287 sg, count, offset, len,
1292 iscsi_tcp_send_linear_data_prepare(struct iscsi_conn *conn, void *data,
1295 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1296 struct hash_desc *tx_hash = NULL;
1297 unsigned int hdr_spec_len;
1299 debug_tcp("%s(%p, datalen=%d%s)\n", __func__, tcp_conn, len,
1300 conn->datadgst_en? ", digest enabled" : "");
1302 /* Make sure the datalen matches what the caller
1303 said he would send. */
1304 hdr_spec_len = ntoh24(tcp_conn->out.hdr->dlength);
1305 WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
1307 if (conn->datadgst_en)
1308 tx_hash = &tcp_conn->tx_hash;
1310 iscsi_segment_init_linear(&tcp_conn->out.data_segment,
1311 data, len, NULL, tx_hash);
1315 * iscsi_solicit_data_cont - initialize next Data-Out
1316 * @conn: iscsi connection
1317 * @task: scsi command task
1319 * @left: bytes left to transfer
1322 * Initialize next Data-Out within this R2T sequence and continue
1323 * to process next Scatter-Gather element(if any) of this SCSI command.
1325 * Called under connection lock.
1328 iscsi_solicit_data_cont(struct iscsi_conn *conn, struct iscsi_task *task,
1329 struct iscsi_r2t_info *r2t)
1331 struct iscsi_data *hdr;
1332 int new_offset, left;
1334 BUG_ON(r2t->data_length - r2t->sent < 0);
1335 left = r2t->data_length - r2t->sent;
1339 hdr = &r2t->dtask.hdr;
1340 memset(hdr, 0, sizeof(struct iscsi_data));
1341 hdr->ttt = r2t->ttt;
1342 hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
1343 r2t->solicit_datasn++;
1344 hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
1345 memcpy(hdr->lun, task->hdr->lun, sizeof(hdr->lun));
1346 hdr->itt = task->hdr->itt;
1347 hdr->exp_statsn = r2t->exp_statsn;
1348 new_offset = r2t->data_offset + r2t->sent;
1349 hdr->offset = cpu_to_be32(new_offset);
1350 if (left > conn->max_xmit_dlength) {
1351 hton24(hdr->dlength, conn->max_xmit_dlength);
1352 r2t->data_count = conn->max_xmit_dlength;
1354 hton24(hdr->dlength, left);
1355 r2t->data_count = left;
1356 hdr->flags = ISCSI_FLAG_CMD_FINAL;
1359 conn->dataout_pdus_cnt++;
1363 static int iscsi_tcp_pdu_init(struct iscsi_task *task,
1364 unsigned int offset, unsigned int count)
1366 struct iscsi_conn *conn = task->conn;
1369 iscsi_tcp_send_hdr_prep(conn, task->hdr, task->hdr_len);
1375 iscsi_tcp_send_linear_data_prepare(conn, task->data, count);
1377 struct scsi_data_buffer *sdb = scsi_out(task->sc);
1379 err = iscsi_tcp_send_data_prep(conn, sdb->table.sgl,
1380 sdb->table.nents, offset, count);
1384 iscsi_conn_failure(conn, err);
1390 static int iscsi_tcp_pdu_alloc(struct iscsi_task *task)
1392 struct iscsi_tcp_task *tcp_task = task->dd_data;
1394 task->hdr = &tcp_task->hdr.hdrbuf;
1395 task->hdr_max = sizeof(tcp_task->hdr) - ISCSI_DIGEST_SIZE;
1400 * iscsi_tcp_task - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
1401 * @conn: iscsi connection
1402 * @task: scsi command task
1405 static int iscsi_tcp_task_init(struct iscsi_task *task)
1407 struct iscsi_tcp_task *tcp_task = task->dd_data;
1408 struct iscsi_conn *conn = task->conn;
1409 struct scsi_cmnd *sc = task->sc;
1414 * mgmt tasks do not have a scatterlist since they come
1415 * in from the iscsi interface.
1417 debug_scsi("mtask deq [cid %d itt 0x%x]\n", conn->id,
1420 return conn->session->tt->init_pdu(task, 0, task->data_count);
1423 BUG_ON(__kfifo_len(tcp_task->r2tqueue));
1424 tcp_task->exp_datasn = 0;
1426 /* Prepare PDU, optionally w/ immediate data */
1427 debug_scsi("task deq [cid %d itt 0x%x imm %d unsol %d]\n",
1428 conn->id, task->itt, task->imm_count,
1429 task->unsol_r2t.data_length);
1431 err = conn->session->tt->init_pdu(task, 0, task->imm_count);
1434 task->imm_count = 0;
1438 static struct iscsi_r2t_info *iscsi_tcp_get_curr_r2t(struct iscsi_task *task)
1440 struct iscsi_session *session = task->conn->session;
1441 struct iscsi_tcp_task *tcp_task = task->dd_data;
1442 struct iscsi_r2t_info *r2t = NULL;
1444 if (iscsi_task_has_unsol_data(task))
1445 r2t = &task->unsol_r2t;
1447 spin_lock_bh(&session->lock);
1448 if (tcp_task->r2t) {
1449 r2t = tcp_task->r2t;
1450 /* Continue with this R2T? */
1451 if (r2t->data_length <= r2t->sent) {
1452 debug_scsi(" done with r2t %p\n", r2t);
1453 __kfifo_put(tcp_task->r2tpool.queue,
1454 (void *)&tcp_task->r2t,
1456 tcp_task->r2t = r2t = NULL;
1461 __kfifo_get(tcp_task->r2tqueue,
1462 (void *)&tcp_task->r2t, sizeof(void *));
1463 r2t = tcp_task->r2t;
1465 spin_unlock_bh(&session->lock);
1472 * iscsi_tcp_task_xmit - xmit normal PDU task
1473 * @task: iscsi command task
1475 * We're expected to return 0 when everything was transmitted succesfully,
1476 * -EAGAIN if there's still data in the queue, or != 0 for any other kind
1479 static int iscsi_tcp_task_xmit(struct iscsi_task *task)
1481 struct iscsi_conn *conn = task->conn;
1482 struct iscsi_session *session = conn->session;
1483 struct iscsi_r2t_info *r2t;
1487 /* Flush any pending data first. */
1488 rc = session->tt->xmit_pdu(task);
1494 if (task->hdr->itt == RESERVED_ITT)
1495 iscsi_put_task(task);
1499 /* Are we done already? */
1500 if (task->sc->sc_data_direction != DMA_TO_DEVICE)
1503 r2t = iscsi_tcp_get_curr_r2t(task);
1505 /* Waiting for more R2Ts to arrive. */
1506 debug_tcp("no R2Ts yet\n");
1510 rc = conn->session->tt->alloc_pdu(task);
1513 iscsi_prep_data_out_pdu(task, r2t, (struct iscsi_data *) task->hdr);
1515 debug_scsi("sol dout %p [dsn %d itt 0x%x doff %d dlen %d]\n",
1516 r2t, r2t->datasn - 1, task->hdr->itt,
1517 r2t->data_offset + r2t->sent, r2t->data_count);
1519 rc = conn->session->tt->init_pdu(task, r2t->data_offset + r2t->sent,
1523 r2t->sent += r2t->data_count;
1527 static struct iscsi_cls_conn *
1528 iscsi_tcp_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
1530 struct iscsi_conn *conn;
1531 struct iscsi_cls_conn *cls_conn;
1532 struct iscsi_tcp_conn *tcp_conn;
1534 cls_conn = iscsi_conn_setup(cls_session, sizeof(*tcp_conn), conn_idx);
1537 conn = cls_conn->dd_data;
1539 * due to strange issues with iser these are not set
1540 * in iscsi_conn_setup
1542 conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1544 tcp_conn = conn->dd_data;
1545 tcp_conn->iscsi_conn = conn;
1547 tcp_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1549 tcp_conn->tx_hash.flags = 0;
1550 if (IS_ERR(tcp_conn->tx_hash.tfm))
1553 tcp_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1555 tcp_conn->rx_hash.flags = 0;
1556 if (IS_ERR(tcp_conn->rx_hash.tfm))
1562 crypto_free_hash(tcp_conn->tx_hash.tfm);
1564 iscsi_conn_printk(KERN_ERR, conn,
1565 "Could not create connection due to crc32c "
1566 "loading error. Make sure the crc32c "
1567 "module is built as a module or into the "
1569 iscsi_conn_teardown(cls_conn);
1574 iscsi_tcp_release_conn(struct iscsi_conn *conn)
1576 struct iscsi_session *session = conn->session;
1577 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1578 struct socket *sock = tcp_conn->sock;
1583 sock_hold(sock->sk);
1584 iscsi_conn_restore_callbacks(tcp_conn);
1587 spin_lock_bh(&session->lock);
1588 tcp_conn->sock = NULL;
1589 spin_unlock_bh(&session->lock);
1594 iscsi_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
1596 struct iscsi_conn *conn = cls_conn->dd_data;
1597 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1599 iscsi_tcp_release_conn(conn);
1601 if (tcp_conn->tx_hash.tfm)
1602 crypto_free_hash(tcp_conn->tx_hash.tfm);
1603 if (tcp_conn->rx_hash.tfm)
1604 crypto_free_hash(tcp_conn->rx_hash.tfm);
1606 iscsi_conn_teardown(cls_conn);
1610 iscsi_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
1612 struct iscsi_conn *conn = cls_conn->dd_data;
1613 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1615 /* userspace may have goofed up and not bound us */
1616 if (!tcp_conn->sock)
1619 * Make sure our recv side is stopped.
1620 * Older tools called conn stop before ep_disconnect
1621 * so IO could still be coming in.
1623 write_lock_bh(&tcp_conn->sock->sk->sk_callback_lock);
1624 set_bit(ISCSI_SUSPEND_BIT, &conn->suspend_rx);
1625 write_unlock_bh(&tcp_conn->sock->sk->sk_callback_lock);
1627 iscsi_conn_stop(cls_conn, flag);
1628 iscsi_tcp_release_conn(conn);
1631 static int iscsi_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
1632 char *buf, int *port,
1633 int (*getname)(struct socket *, struct sockaddr *,
1636 struct sockaddr_storage *addr;
1637 struct sockaddr_in6 *sin6;
1638 struct sockaddr_in *sin;
1641 addr = kmalloc(sizeof(*addr), GFP_KERNEL);
1645 if (getname(sock, (struct sockaddr *) addr, &len)) {
1650 switch (addr->ss_family) {
1652 sin = (struct sockaddr_in *)addr;
1653 spin_lock_bh(&conn->session->lock);
1654 sprintf(buf, "%pI4", &sin->sin_addr.s_addr);
1655 *port = be16_to_cpu(sin->sin_port);
1656 spin_unlock_bh(&conn->session->lock);
1659 sin6 = (struct sockaddr_in6 *)addr;
1660 spin_lock_bh(&conn->session->lock);
1661 sprintf(buf, "%pI6", &sin6->sin6_addr);
1662 *port = be16_to_cpu(sin6->sin6_port);
1663 spin_unlock_bh(&conn->session->lock);
1672 iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
1673 struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
1676 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
1677 struct iscsi_host *ihost = shost_priv(shost);
1678 struct iscsi_conn *conn = cls_conn->dd_data;
1679 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1681 struct socket *sock;
1684 /* lookup for existing socket */
1685 sock = sockfd_lookup((int)transport_eph, &err);
1687 iscsi_conn_printk(KERN_ERR, conn,
1688 "sockfd_lookup failed %d\n", err);
1692 * copy these values now because if we drop the session
1693 * userspace may still want to query the values since we will
1694 * be using them for the reconnect
1696 err = iscsi_tcp_get_addr(conn, sock, conn->portal_address,
1697 &conn->portal_port, kernel_getpeername);
1701 err = iscsi_tcp_get_addr(conn, sock, ihost->local_address,
1702 &ihost->local_port, kernel_getsockname);
1706 err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
1710 /* bind iSCSI connection and socket */
1711 tcp_conn->sock = sock;
1713 /* setup Socket parameters */
1716 sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
1717 sk->sk_allocation = GFP_ATOMIC;
1719 iscsi_conn_set_callbacks(conn);
1720 tcp_conn->sendpage = tcp_conn->sock->ops->sendpage;
1722 * set receive state machine into initial state
1724 iscsi_tcp_hdr_recv_prep(tcp_conn);
1733 iscsi_r2tpool_alloc(struct iscsi_session *session)
1739 * initialize per-task: R2T pool and xmit queue
1741 for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1742 struct iscsi_task *task = session->cmds[cmd_i];
1743 struct iscsi_tcp_task *tcp_task = task->dd_data;
1746 * pre-allocated x2 as much r2ts to handle race when
1747 * target acks DataOut faster than we data_xmit() queues
1748 * could replenish r2tqueue.
1752 if (iscsi_pool_init(&tcp_task->r2tpool,
1753 session->max_r2t * 2, NULL,
1754 sizeof(struct iscsi_r2t_info))) {
1755 goto r2t_alloc_fail;
1758 /* R2T xmit queue */
1759 tcp_task->r2tqueue = kfifo_alloc(
1760 session->max_r2t * 4 * sizeof(void*), GFP_KERNEL, NULL);
1761 if (tcp_task->r2tqueue == ERR_PTR(-ENOMEM)) {
1762 iscsi_pool_free(&tcp_task->r2tpool);
1763 goto r2t_alloc_fail;
1770 for (i = 0; i < cmd_i; i++) {
1771 struct iscsi_task *task = session->cmds[i];
1772 struct iscsi_tcp_task *tcp_task = task->dd_data;
1774 kfifo_free(tcp_task->r2tqueue);
1775 iscsi_pool_free(&tcp_task->r2tpool);
1781 iscsi_r2tpool_free(struct iscsi_session *session)
1785 for (i = 0; i < session->cmds_max; i++) {
1786 struct iscsi_task *task = session->cmds[i];
1787 struct iscsi_tcp_task *tcp_task = task->dd_data;
1789 kfifo_free(tcp_task->r2tqueue);
1790 iscsi_pool_free(&tcp_task->r2tpool);
1795 iscsi_conn_set_param(struct iscsi_cls_conn *cls_conn, enum iscsi_param param,
1796 char *buf, int buflen)
1798 struct iscsi_conn *conn = cls_conn->dd_data;
1799 struct iscsi_session *session = conn->session;
1800 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1804 case ISCSI_PARAM_HDRDGST_EN:
1805 iscsi_set_param(cls_conn, param, buf, buflen);
1807 case ISCSI_PARAM_DATADGST_EN:
1808 iscsi_set_param(cls_conn, param, buf, buflen);
1809 tcp_conn->sendpage = conn->datadgst_en ?
1810 sock_no_sendpage : tcp_conn->sock->ops->sendpage;
1812 case ISCSI_PARAM_MAX_R2T:
1813 sscanf(buf, "%d", &value);
1814 if (value <= 0 || !is_power_of_2(value))
1816 if (session->max_r2t == value)
1818 iscsi_r2tpool_free(session);
1819 iscsi_set_param(cls_conn, param, buf, buflen);
1820 if (iscsi_r2tpool_alloc(session))
1824 return iscsi_set_param(cls_conn, param, buf, buflen);
1831 iscsi_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
1832 enum iscsi_param param, char *buf)
1834 struct iscsi_conn *conn = cls_conn->dd_data;
1838 case ISCSI_PARAM_CONN_PORT:
1839 spin_lock_bh(&conn->session->lock);
1840 len = sprintf(buf, "%hu\n", conn->portal_port);
1841 spin_unlock_bh(&conn->session->lock);
1843 case ISCSI_PARAM_CONN_ADDRESS:
1844 spin_lock_bh(&conn->session->lock);
1845 len = sprintf(buf, "%s\n", conn->portal_address);
1846 spin_unlock_bh(&conn->session->lock);
1849 return iscsi_conn_get_param(cls_conn, param, buf);
1856 iscsi_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
1858 struct iscsi_conn *conn = cls_conn->dd_data;
1859 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1861 stats->txdata_octets = conn->txdata_octets;
1862 stats->rxdata_octets = conn->rxdata_octets;
1863 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
1864 stats->dataout_pdus = conn->dataout_pdus_cnt;
1865 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
1866 stats->datain_pdus = conn->datain_pdus_cnt;
1867 stats->r2t_pdus = conn->r2t_pdus_cnt;
1868 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
1869 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
1870 stats->custom_length = 3;
1871 strcpy(stats->custom[0].desc, "tx_sendpage_failures");
1872 stats->custom[0].value = tcp_conn->sendpage_failures_cnt;
1873 strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
1874 stats->custom[1].value = tcp_conn->discontiguous_hdr_cnt;
1875 strcpy(stats->custom[2].desc, "eh_abort_cnt");
1876 stats->custom[2].value = conn->eh_abort_cnt;
1879 static struct iscsi_cls_session *
1880 iscsi_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,
1881 uint16_t qdepth, uint32_t initial_cmdsn,
1884 struct iscsi_cls_session *cls_session;
1885 struct iscsi_session *session;
1886 struct Scsi_Host *shost;
1889 printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep);
1893 shost = iscsi_host_alloc(&iscsi_sht, 0, qdepth);
1896 shost->transportt = iscsi_tcp_scsi_transport;
1897 shost->max_lun = iscsi_max_lun;
1899 shost->max_channel = 0;
1900 shost->max_cmd_len = SCSI_MAX_VARLEN_CDB_SIZE;
1902 if (iscsi_host_add(shost, NULL))
1904 *hostno = shost->host_no;
1906 cls_session = iscsi_session_setup(&iscsi_tcp_transport, shost, cmds_max,
1907 sizeof(struct iscsi_tcp_task),
1911 session = cls_session->dd_data;
1913 shost->can_queue = session->scsi_cmds_max;
1914 if (iscsi_r2tpool_alloc(session))
1915 goto remove_session;
1919 iscsi_session_teardown(cls_session);
1921 iscsi_host_remove(shost);
1923 iscsi_host_free(shost);
1927 static void iscsi_tcp_session_destroy(struct iscsi_cls_session *cls_session)
1929 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
1931 iscsi_r2tpool_free(cls_session->dd_data);
1932 iscsi_session_teardown(cls_session);
1934 iscsi_host_remove(shost);
1935 iscsi_host_free(shost);
1938 static int iscsi_tcp_slave_configure(struct scsi_device *sdev)
1940 blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
1941 blk_queue_dma_alignment(sdev->request_queue, 0);
1945 static struct scsi_host_template iscsi_sht = {
1946 .module = THIS_MODULE,
1947 .name = "iSCSI Initiator over TCP/IP",
1948 .queuecommand = iscsi_queuecommand,
1949 .change_queue_depth = iscsi_change_queue_depth,
1950 .can_queue = ISCSI_DEF_XMIT_CMDS_MAX - 1,
1951 .sg_tablesize = 4096,
1952 .max_sectors = 0xFFFF,
1953 .cmd_per_lun = ISCSI_DEF_CMD_PER_LUN,
1954 .eh_abort_handler = iscsi_eh_abort,
1955 .eh_device_reset_handler= iscsi_eh_device_reset,
1956 .eh_target_reset_handler= iscsi_eh_target_reset,
1957 .use_clustering = DISABLE_CLUSTERING,
1958 .slave_configure = iscsi_tcp_slave_configure,
1959 .proc_name = "iscsi_tcp",
1963 static struct iscsi_transport iscsi_tcp_transport = {
1964 .owner = THIS_MODULE,
1966 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
1968 .param_mask = ISCSI_MAX_RECV_DLENGTH |
1969 ISCSI_MAX_XMIT_DLENGTH |
1972 ISCSI_INITIAL_R2T_EN |
1977 ISCSI_PDU_INORDER_EN |
1978 ISCSI_DATASEQ_INORDER_EN |
1981 ISCSI_CONN_ADDRESS |
1983 ISCSI_PERSISTENT_PORT |
1984 ISCSI_PERSISTENT_ADDRESS |
1985 ISCSI_TARGET_NAME | ISCSI_TPGT |
1986 ISCSI_USERNAME | ISCSI_PASSWORD |
1987 ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
1988 ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
1989 ISCSI_LU_RESET_TMO |
1990 ISCSI_PING_TMO | ISCSI_RECV_TMO |
1991 ISCSI_IFACE_NAME | ISCSI_INITIATOR_NAME,
1992 .host_param_mask = ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
1993 ISCSI_HOST_INITIATOR_NAME |
1994 ISCSI_HOST_NETDEV_NAME,
1995 /* session management */
1996 .create_session = iscsi_tcp_session_create,
1997 .destroy_session = iscsi_tcp_session_destroy,
1998 /* connection management */
1999 .create_conn = iscsi_tcp_conn_create,
2000 .bind_conn = iscsi_tcp_conn_bind,
2001 .destroy_conn = iscsi_tcp_conn_destroy,
2002 .set_param = iscsi_conn_set_param,
2003 .get_conn_param = iscsi_tcp_conn_get_param,
2004 .get_session_param = iscsi_session_get_param,
2005 .start_conn = iscsi_conn_start,
2006 .stop_conn = iscsi_tcp_conn_stop,
2007 /* iscsi host params */
2008 .get_host_param = iscsi_host_get_param,
2009 .set_host_param = iscsi_host_set_param,
2011 .send_pdu = iscsi_conn_send_pdu,
2012 .get_stats = iscsi_conn_get_stats,
2013 /* iscsi task/cmd helpers */
2014 .init_task = iscsi_tcp_task_init,
2015 .xmit_task = iscsi_tcp_task_xmit,
2016 .cleanup_task = iscsi_tcp_cleanup_task,
2017 /* low level pdu helpers */
2018 .xmit_pdu = iscsi_tcp_flush,
2019 .init_pdu = iscsi_tcp_pdu_init,
2020 .alloc_pdu = iscsi_tcp_pdu_alloc,
2022 .session_recovery_timedout = iscsi_session_recovery_timedout,
2026 iscsi_tcp_init(void)
2028 if (iscsi_max_lun < 1) {
2029 printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
2034 iscsi_tcp_scsi_transport = iscsi_register_transport(
2035 &iscsi_tcp_transport);
2036 if (!iscsi_tcp_scsi_transport)
2043 iscsi_tcp_exit(void)
2045 iscsi_unregister_transport(&iscsi_tcp_transport);
2048 module_init(iscsi_tcp_init);
2049 module_exit(iscsi_tcp_exit);