Merge branch 'next' of git://git.monstr.eu/linux-2.6-microblaze
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / scsi / libiscsi_tcp.c
blob5715a3d0a3d3e7db2b8a1b4c4de6ac983c4f6f48
1 /*
2 * iSCSI over TCP/IP Data-Path lib
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.
22 * Credits:
23 * Christoph Hellwig
24 * FUJITA Tomonori
25 * Arne Redlich
26 * Zhenyu Wang
29 #include <linux/types.h>
30 #include <linux/list.h>
31 #include <linux/inet.h>
32 #include <linux/slab.h>
33 #include <linux/file.h>
34 #include <linux/blkdev.h>
35 #include <linux/crypto.h>
36 #include <linux/delay.h>
37 #include <linux/kfifo.h>
38 #include <linux/scatterlist.h>
39 #include <linux/module.h>
40 #include <net/tcp.h>
41 #include <scsi/scsi_cmnd.h>
42 #include <scsi/scsi_device.h>
43 #include <scsi/scsi_host.h>
44 #include <scsi/scsi.h>
45 #include <scsi/scsi_transport_iscsi.h>
47 #include "iscsi_tcp.h"
49 MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
50 "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
51 "Alex Aizman <itn780@yahoo.com>");
52 MODULE_DESCRIPTION("iSCSI/TCP data-path");
53 MODULE_LICENSE("GPL");
55 static int iscsi_dbg_libtcp;
56 module_param_named(debug_libiscsi_tcp, iscsi_dbg_libtcp, int,
57 S_IRUGO | S_IWUSR);
58 MODULE_PARM_DESC(debug_libiscsi_tcp, "Turn on debugging for libiscsi_tcp "
59 "module. Set to 1 to turn on, and zero to turn off. Default "
60 "is off.");
62 #define ISCSI_DBG_TCP(_conn, dbg_fmt, arg...) \
63 do { \
64 if (iscsi_dbg_libtcp) \
65 iscsi_conn_printk(KERN_INFO, _conn, \
66 "%s " dbg_fmt, \
67 __func__, ##arg); \
68 } while (0);
70 static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
71 struct iscsi_segment *segment);
74 * Scatterlist handling: inside the iscsi_segment, we
75 * remember an index into the scatterlist, and set data/size
76 * to the current scatterlist entry. For highmem pages, we
77 * kmap as needed.
79 * Note that the page is unmapped when we return from
80 * TCP's data_ready handler, so we may end up mapping and
81 * unmapping the same page repeatedly. The whole reason
82 * for this is that we shouldn't keep the page mapped
83 * outside the softirq.
86 /**
87 * iscsi_tcp_segment_init_sg - init indicated scatterlist entry
88 * @segment: the buffer object
89 * @sg: scatterlist
90 * @offset: byte offset into that sg entry
92 * This function sets up the segment so that subsequent
93 * data is copied to the indicated sg entry, at the given
94 * offset.
96 static inline void
97 iscsi_tcp_segment_init_sg(struct iscsi_segment *segment,
98 struct scatterlist *sg, unsigned int offset)
100 segment->sg = sg;
101 segment->sg_offset = offset;
102 segment->size = min(sg->length - offset,
103 segment->total_size - segment->total_copied);
104 segment->data = NULL;
108 * iscsi_tcp_segment_map - map the current S/G page
109 * @segment: iscsi_segment
110 * @recv: 1 if called from recv path
112 * We only need to possibly kmap data if scatter lists are being used,
113 * because the iscsi passthrough and internal IO paths will never use high
114 * mem pages.
116 static void iscsi_tcp_segment_map(struct iscsi_segment *segment, int recv)
118 struct scatterlist *sg;
120 if (segment->data != NULL || !segment->sg)
121 return;
123 sg = segment->sg;
124 BUG_ON(segment->sg_mapped);
125 BUG_ON(sg->length == 0);
128 * If the page count is greater than one it is ok to send
129 * to the network layer's zero copy send path. If not we
130 * have to go the slow sendmsg path. We always map for the
131 * recv path.
133 if (page_count(sg_page(sg)) >= 1 && !recv)
134 return;
136 if (recv) {
137 segment->atomic_mapped = true;
138 segment->sg_mapped = kmap_atomic(sg_page(sg), KM_SOFTIRQ0);
139 } else {
140 segment->atomic_mapped = false;
141 /* the xmit path can sleep with the page mapped so use kmap */
142 segment->sg_mapped = kmap(sg_page(sg));
145 segment->data = segment->sg_mapped + sg->offset + segment->sg_offset;
148 void iscsi_tcp_segment_unmap(struct iscsi_segment *segment)
150 if (segment->sg_mapped) {
151 if (segment->atomic_mapped)
152 kunmap_atomic(segment->sg_mapped, KM_SOFTIRQ0);
153 else
154 kunmap(sg_page(segment->sg));
155 segment->sg_mapped = NULL;
156 segment->data = NULL;
159 EXPORT_SYMBOL_GPL(iscsi_tcp_segment_unmap);
162 * Splice the digest buffer into the buffer
164 static inline void
165 iscsi_tcp_segment_splice_digest(struct iscsi_segment *segment, void *digest)
167 segment->data = digest;
168 segment->digest_len = ISCSI_DIGEST_SIZE;
169 segment->total_size += ISCSI_DIGEST_SIZE;
170 segment->size = ISCSI_DIGEST_SIZE;
171 segment->copied = 0;
172 segment->sg = NULL;
173 segment->hash = NULL;
177 * iscsi_tcp_segment_done - check whether the segment is complete
178 * @tcp_conn: iscsi tcp connection
179 * @segment: iscsi segment to check
180 * @recv: set to one of this is called from the recv path
181 * @copied: number of bytes copied
183 * Check if we're done receiving this segment. If the receive
184 * buffer is full but we expect more data, move on to the
185 * next entry in the scatterlist.
187 * If the amount of data we received isn't a multiple of 4,
188 * we will transparently receive the pad bytes, too.
190 * This function must be re-entrant.
192 int iscsi_tcp_segment_done(struct iscsi_tcp_conn *tcp_conn,
193 struct iscsi_segment *segment, int recv,
194 unsigned copied)
196 struct scatterlist sg;
197 unsigned int pad;
199 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copied %u %u size %u %s\n",
200 segment->copied, copied, segment->size,
201 recv ? "recv" : "xmit");
202 if (segment->hash && copied) {
204 * If a segment is kmapd we must unmap it before sending
205 * to the crypto layer since that will try to kmap it again.
207 iscsi_tcp_segment_unmap(segment);
209 if (!segment->data) {
210 sg_init_table(&sg, 1);
211 sg_set_page(&sg, sg_page(segment->sg), copied,
212 segment->copied + segment->sg_offset +
213 segment->sg->offset);
214 } else
215 sg_init_one(&sg, segment->data + segment->copied,
216 copied);
217 crypto_hash_update(segment->hash, &sg, copied);
220 segment->copied += copied;
221 if (segment->copied < segment->size) {
222 iscsi_tcp_segment_map(segment, recv);
223 return 0;
226 segment->total_copied += segment->copied;
227 segment->copied = 0;
228 segment->size = 0;
230 /* Unmap the current scatterlist page, if there is one. */
231 iscsi_tcp_segment_unmap(segment);
233 /* Do we have more scatterlist entries? */
234 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "total copied %u total size %u\n",
235 segment->total_copied, segment->total_size);
236 if (segment->total_copied < segment->total_size) {
237 /* Proceed to the next entry in the scatterlist. */
238 iscsi_tcp_segment_init_sg(segment, sg_next(segment->sg),
240 iscsi_tcp_segment_map(segment, recv);
241 BUG_ON(segment->size == 0);
242 return 0;
245 /* Do we need to handle padding? */
246 if (!(tcp_conn->iscsi_conn->session->tt->caps & CAP_PADDING_OFFLOAD)) {
247 pad = iscsi_padding(segment->total_copied);
248 if (pad != 0) {
249 ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
250 "consume %d pad bytes\n", pad);
251 segment->total_size += pad;
252 segment->size = pad;
253 segment->data = segment->padbuf;
254 return 0;
259 * Set us up for transferring the data digest. hdr digest
260 * is completely handled in hdr done function.
262 if (segment->hash) {
263 crypto_hash_final(segment->hash, segment->digest);
264 iscsi_tcp_segment_splice_digest(segment,
265 recv ? segment->recv_digest : segment->digest);
266 return 0;
269 return 1;
271 EXPORT_SYMBOL_GPL(iscsi_tcp_segment_done);
274 * iscsi_tcp_segment_recv - copy data to segment
275 * @tcp_conn: the iSCSI TCP connection
276 * @segment: the buffer to copy to
277 * @ptr: data pointer
278 * @len: amount of data available
280 * This function copies up to @len bytes to the
281 * given buffer, and returns the number of bytes
282 * consumed, which can actually be less than @len.
284 * If hash digest is enabled, the function will update the
285 * hash while copying.
286 * Combining these two operations doesn't buy us a lot (yet),
287 * but in the future we could implement combined copy+crc,
288 * just way we do for network layer checksums.
290 static int
291 iscsi_tcp_segment_recv(struct iscsi_tcp_conn *tcp_conn,
292 struct iscsi_segment *segment, const void *ptr,
293 unsigned int len)
295 unsigned int copy = 0, copied = 0;
297 while (!iscsi_tcp_segment_done(tcp_conn, segment, 1, copy)) {
298 if (copied == len) {
299 ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
300 "copied %d bytes\n", len);
301 break;
304 copy = min(len - copied, segment->size - segment->copied);
305 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "copying %d\n", copy);
306 memcpy(segment->data + segment->copied, ptr + copied, copy);
307 copied += copy;
309 return copied;
312 inline void
313 iscsi_tcp_dgst_header(struct hash_desc *hash, const void *hdr, size_t hdrlen,
314 unsigned char digest[ISCSI_DIGEST_SIZE])
316 struct scatterlist sg;
318 sg_init_one(&sg, hdr, hdrlen);
319 crypto_hash_digest(hash, &sg, hdrlen, digest);
321 EXPORT_SYMBOL_GPL(iscsi_tcp_dgst_header);
323 static inline int
324 iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
325 struct iscsi_segment *segment)
327 if (!segment->digest_len)
328 return 1;
330 if (memcmp(segment->recv_digest, segment->digest,
331 segment->digest_len)) {
332 ISCSI_DBG_TCP(tcp_conn->iscsi_conn, "digest mismatch\n");
333 return 0;
336 return 1;
340 * Helper function to set up segment buffer
342 static inline void
343 __iscsi_segment_init(struct iscsi_segment *segment, size_t size,
344 iscsi_segment_done_fn_t *done, struct hash_desc *hash)
346 memset(segment, 0, sizeof(*segment));
347 segment->total_size = size;
348 segment->done = done;
350 if (hash) {
351 segment->hash = hash;
352 crypto_hash_init(hash);
356 inline void
357 iscsi_segment_init_linear(struct iscsi_segment *segment, void *data,
358 size_t size, iscsi_segment_done_fn_t *done,
359 struct hash_desc *hash)
361 __iscsi_segment_init(segment, size, done, hash);
362 segment->data = data;
363 segment->size = size;
365 EXPORT_SYMBOL_GPL(iscsi_segment_init_linear);
367 inline int
368 iscsi_segment_seek_sg(struct iscsi_segment *segment,
369 struct scatterlist *sg_list, unsigned int sg_count,
370 unsigned int offset, size_t size,
371 iscsi_segment_done_fn_t *done, struct hash_desc *hash)
373 struct scatterlist *sg;
374 unsigned int i;
376 __iscsi_segment_init(segment, size, done, hash);
377 for_each_sg(sg_list, sg, sg_count, i) {
378 if (offset < sg->length) {
379 iscsi_tcp_segment_init_sg(segment, sg, offset);
380 return 0;
382 offset -= sg->length;
385 return ISCSI_ERR_DATA_OFFSET;
387 EXPORT_SYMBOL_GPL(iscsi_segment_seek_sg);
390 * iscsi_tcp_hdr_recv_prep - prep segment for hdr reception
391 * @tcp_conn: iscsi connection to prep for
393 * This function always passes NULL for the hash argument, because when this
394 * function is called we do not yet know the final size of the header and want
395 * to delay the digest processing until we know that.
397 void iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
399 ISCSI_DBG_TCP(tcp_conn->iscsi_conn,
400 "(%s)\n", tcp_conn->iscsi_conn->hdrdgst_en ?
401 "digest enabled" : "digest disabled");
402 iscsi_segment_init_linear(&tcp_conn->in.segment,
403 tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
404 iscsi_tcp_hdr_recv_done, NULL);
406 EXPORT_SYMBOL_GPL(iscsi_tcp_hdr_recv_prep);
409 * Handle incoming reply to any other type of command
411 static int
412 iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
413 struct iscsi_segment *segment)
415 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
416 int rc = 0;
418 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
419 return ISCSI_ERR_DATA_DGST;
421 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr,
422 conn->data, tcp_conn->in.datalen);
423 if (rc)
424 return rc;
426 iscsi_tcp_hdr_recv_prep(tcp_conn);
427 return 0;
430 static void
431 iscsi_tcp_data_recv_prep(struct iscsi_tcp_conn *tcp_conn)
433 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
434 struct hash_desc *rx_hash = NULL;
436 if (conn->datadgst_en &&
437 !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
438 rx_hash = tcp_conn->rx_hash;
440 iscsi_segment_init_linear(&tcp_conn->in.segment,
441 conn->data, tcp_conn->in.datalen,
442 iscsi_tcp_data_recv_done, rx_hash);
446 * iscsi_tcp_cleanup_task - free tcp_task resources
447 * @task: iscsi task
449 * must be called with session lock
451 void iscsi_tcp_cleanup_task(struct iscsi_task *task)
453 struct iscsi_tcp_task *tcp_task = task->dd_data;
454 struct iscsi_r2t_info *r2t;
456 /* nothing to do for mgmt */
457 if (!task->sc)
458 return;
460 /* flush task's r2t queues */
461 while (kfifo_out(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*))) {
462 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
463 sizeof(void*));
464 ISCSI_DBG_TCP(task->conn, "pending r2t dropped\n");
467 r2t = tcp_task->r2t;
468 if (r2t != NULL) {
469 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
470 sizeof(void*));
471 tcp_task->r2t = NULL;
474 EXPORT_SYMBOL_GPL(iscsi_tcp_cleanup_task);
477 * iscsi_tcp_data_in - SCSI Data-In Response processing
478 * @conn: iscsi connection
479 * @task: scsi command task
481 static int iscsi_tcp_data_in(struct iscsi_conn *conn, struct iscsi_task *task)
483 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
484 struct iscsi_tcp_task *tcp_task = task->dd_data;
485 struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
486 int datasn = be32_to_cpu(rhdr->datasn);
487 unsigned total_in_length = scsi_in(task->sc)->length;
490 * lib iscsi will update this in the completion handling if there
491 * is status.
493 if (!(rhdr->flags & ISCSI_FLAG_DATA_STATUS))
494 iscsi_update_cmdsn(conn->session, (struct iscsi_nopin*)rhdr);
496 if (tcp_conn->in.datalen == 0)
497 return 0;
499 if (tcp_task->exp_datasn != datasn) {
500 ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->datasn(%d)"
501 "\n", tcp_task->exp_datasn, datasn);
502 return ISCSI_ERR_DATASN;
505 tcp_task->exp_datasn++;
507 tcp_task->data_offset = be32_to_cpu(rhdr->offset);
508 if (tcp_task->data_offset + tcp_conn->in.datalen > total_in_length) {
509 ISCSI_DBG_TCP(conn, "data_offset(%d) + data_len(%d) > "
510 "total_length_in(%d)\n", tcp_task->data_offset,
511 tcp_conn->in.datalen, total_in_length);
512 return ISCSI_ERR_DATA_OFFSET;
515 conn->datain_pdus_cnt++;
516 return 0;
520 * iscsi_tcp_r2t_rsp - iSCSI R2T Response processing
521 * @conn: iscsi connection
522 * @task: scsi command task
524 static int iscsi_tcp_r2t_rsp(struct iscsi_conn *conn, struct iscsi_task *task)
526 struct iscsi_session *session = conn->session;
527 struct iscsi_tcp_task *tcp_task = task->dd_data;
528 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
529 struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
530 struct iscsi_r2t_info *r2t;
531 int r2tsn = be32_to_cpu(rhdr->r2tsn);
532 int rc;
534 if (tcp_conn->in.datalen) {
535 iscsi_conn_printk(KERN_ERR, conn,
536 "invalid R2t with datalen %d\n",
537 tcp_conn->in.datalen);
538 return ISCSI_ERR_DATALEN;
541 if (tcp_task->exp_datasn != r2tsn){
542 ISCSI_DBG_TCP(conn, "task->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
543 tcp_task->exp_datasn, r2tsn);
544 return ISCSI_ERR_R2TSN;
547 /* fill-in new R2T associated with the task */
548 iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
550 if (!task->sc || session->state != ISCSI_STATE_LOGGED_IN) {
551 iscsi_conn_printk(KERN_INFO, conn,
552 "dropping R2T itt %d in recovery.\n",
553 task->itt);
554 return 0;
557 rc = kfifo_out(&tcp_task->r2tpool.queue, (void*)&r2t, sizeof(void*));
558 if (!rc) {
559 iscsi_conn_printk(KERN_ERR, conn, "Could not allocate R2T. "
560 "Target has sent more R2Ts than it "
561 "negotiated for or driver has has leaked.\n");
562 return ISCSI_ERR_PROTO;
565 r2t->exp_statsn = rhdr->statsn;
566 r2t->data_length = be32_to_cpu(rhdr->data_length);
567 if (r2t->data_length == 0) {
568 iscsi_conn_printk(KERN_ERR, conn,
569 "invalid R2T with zero data len\n");
570 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
571 sizeof(void*));
572 return ISCSI_ERR_DATALEN;
575 if (r2t->data_length > session->max_burst)
576 ISCSI_DBG_TCP(conn, "invalid R2T with data len %u and max "
577 "burst %u. Attempting to execute request.\n",
578 r2t->data_length, session->max_burst);
580 r2t->data_offset = be32_to_cpu(rhdr->data_offset);
581 if (r2t->data_offset + r2t->data_length > scsi_out(task->sc)->length) {
582 iscsi_conn_printk(KERN_ERR, conn,
583 "invalid R2T with data len %u at offset %u "
584 "and total length %d\n", r2t->data_length,
585 r2t->data_offset, scsi_out(task->sc)->length);
586 kfifo_in(&tcp_task->r2tpool.queue, (void*)&r2t,
587 sizeof(void*));
588 return ISCSI_ERR_DATALEN;
591 r2t->ttt = rhdr->ttt; /* no flip */
592 r2t->datasn = 0;
593 r2t->sent = 0;
595 tcp_task->exp_datasn = r2tsn + 1;
596 kfifo_in(&tcp_task->r2tqueue, (void*)&r2t, sizeof(void*));
597 conn->r2t_pdus_cnt++;
599 iscsi_requeue_task(task);
600 return 0;
604 * Handle incoming reply to DataIn command
606 static int
607 iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
608 struct iscsi_segment *segment)
610 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
611 struct iscsi_hdr *hdr = tcp_conn->in.hdr;
612 int rc;
614 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
615 return ISCSI_ERR_DATA_DGST;
617 /* check for non-exceptional status */
618 if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
619 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
620 if (rc)
621 return rc;
624 iscsi_tcp_hdr_recv_prep(tcp_conn);
625 return 0;
629 * iscsi_tcp_hdr_dissect - process PDU header
630 * @conn: iSCSI connection
631 * @hdr: PDU header
633 * This function analyzes the header of the PDU received,
634 * and performs several sanity checks. If the PDU is accompanied
635 * by data, the receive buffer is set up to copy the incoming data
636 * to the correct location.
638 static int
639 iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
641 int rc = 0, opcode, ahslen;
642 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
643 struct iscsi_task *task;
645 /* verify PDU length */
646 tcp_conn->in.datalen = ntoh24(hdr->dlength);
647 if (tcp_conn->in.datalen > conn->max_recv_dlength) {
648 iscsi_conn_printk(KERN_ERR, conn,
649 "iscsi_tcp: datalen %d > %d\n",
650 tcp_conn->in.datalen, conn->max_recv_dlength);
651 return ISCSI_ERR_DATALEN;
654 /* Additional header segments. So far, we don't
655 * process additional headers.
657 ahslen = hdr->hlength << 2;
659 opcode = hdr->opcode & ISCSI_OPCODE_MASK;
660 /* verify itt (itt encoding: age+cid+itt) */
661 rc = iscsi_verify_itt(conn, hdr->itt);
662 if (rc)
663 return rc;
665 ISCSI_DBG_TCP(conn, "opcode 0x%x ahslen %d datalen %d\n",
666 opcode, ahslen, tcp_conn->in.datalen);
668 switch(opcode) {
669 case ISCSI_OP_SCSI_DATA_IN:
670 spin_lock(&conn->session->lock);
671 task = iscsi_itt_to_ctask(conn, hdr->itt);
672 if (!task)
673 rc = ISCSI_ERR_BAD_ITT;
674 else
675 rc = iscsi_tcp_data_in(conn, task);
676 if (rc) {
677 spin_unlock(&conn->session->lock);
678 break;
681 if (tcp_conn->in.datalen) {
682 struct iscsi_tcp_task *tcp_task = task->dd_data;
683 struct hash_desc *rx_hash = NULL;
684 struct scsi_data_buffer *sdb = scsi_in(task->sc);
687 * Setup copy of Data-In into the Scsi_Cmnd
688 * Scatterlist case:
689 * We set up the iscsi_segment to point to the next
690 * scatterlist entry to copy to. As we go along,
691 * we move on to the next scatterlist entry and
692 * update the digest per-entry.
694 if (conn->datadgst_en &&
695 !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD))
696 rx_hash = tcp_conn->rx_hash;
698 ISCSI_DBG_TCP(conn, "iscsi_tcp_begin_data_in( "
699 "offset=%d, datalen=%d)\n",
700 tcp_task->data_offset,
701 tcp_conn->in.datalen);
702 task->last_xfer = jiffies;
703 rc = iscsi_segment_seek_sg(&tcp_conn->in.segment,
704 sdb->table.sgl,
705 sdb->table.nents,
706 tcp_task->data_offset,
707 tcp_conn->in.datalen,
708 iscsi_tcp_process_data_in,
709 rx_hash);
710 spin_unlock(&conn->session->lock);
711 return rc;
713 rc = __iscsi_complete_pdu(conn, hdr, NULL, 0);
714 spin_unlock(&conn->session->lock);
715 break;
716 case ISCSI_OP_SCSI_CMD_RSP:
717 if (tcp_conn->in.datalen) {
718 iscsi_tcp_data_recv_prep(tcp_conn);
719 return 0;
721 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
722 break;
723 case ISCSI_OP_R2T:
724 spin_lock(&conn->session->lock);
725 task = iscsi_itt_to_ctask(conn, hdr->itt);
726 if (!task)
727 rc = ISCSI_ERR_BAD_ITT;
728 else if (ahslen)
729 rc = ISCSI_ERR_AHSLEN;
730 else if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
731 task->last_xfer = jiffies;
732 rc = iscsi_tcp_r2t_rsp(conn, task);
733 } else
734 rc = ISCSI_ERR_PROTO;
735 spin_unlock(&conn->session->lock);
736 break;
737 case ISCSI_OP_LOGIN_RSP:
738 case ISCSI_OP_TEXT_RSP:
739 case ISCSI_OP_REJECT:
740 case ISCSI_OP_ASYNC_EVENT:
742 * It is possible that we could get a PDU with a buffer larger
743 * than 8K, but there are no targets that currently do this.
744 * For now we fail until we find a vendor that needs it
746 if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
747 iscsi_conn_printk(KERN_ERR, conn,
748 "iscsi_tcp: received buffer of "
749 "len %u but conn buffer is only %u "
750 "(opcode %0x)\n",
751 tcp_conn->in.datalen,
752 ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
753 rc = ISCSI_ERR_PROTO;
754 break;
757 /* If there's data coming in with the response,
758 * receive it to the connection's buffer.
760 if (tcp_conn->in.datalen) {
761 iscsi_tcp_data_recv_prep(tcp_conn);
762 return 0;
764 /* fall through */
765 case ISCSI_OP_LOGOUT_RSP:
766 case ISCSI_OP_NOOP_IN:
767 case ISCSI_OP_SCSI_TMFUNC_RSP:
768 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
769 break;
770 default:
771 rc = ISCSI_ERR_BAD_OPCODE;
772 break;
775 if (rc == 0) {
776 /* Anything that comes with data should have
777 * been handled above. */
778 if (tcp_conn->in.datalen)
779 return ISCSI_ERR_PROTO;
780 iscsi_tcp_hdr_recv_prep(tcp_conn);
783 return rc;
787 * iscsi_tcp_hdr_recv_done - process PDU header
789 * This is the callback invoked when the PDU header has
790 * been received. If the header is followed by additional
791 * header segments, we go back for more data.
793 static int
794 iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
795 struct iscsi_segment *segment)
797 struct iscsi_conn *conn = tcp_conn->iscsi_conn;
798 struct iscsi_hdr *hdr;
800 /* Check if there are additional header segments
801 * *prior* to computing the digest, because we
802 * may need to go back to the caller for more.
804 hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
805 if (segment->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
806 /* Bump the header length - the caller will
807 * just loop around and get the AHS for us, and
808 * call again. */
809 unsigned int ahslen = hdr->hlength << 2;
811 /* Make sure we don't overflow */
812 if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
813 return ISCSI_ERR_AHSLEN;
815 segment->total_size += ahslen;
816 segment->size += ahslen;
817 return 0;
820 /* We're done processing the header. See if we're doing
821 * header digests; if so, set up the recv_digest buffer
822 * and go back for more. */
823 if (conn->hdrdgst_en &&
824 !(conn->session->tt->caps & CAP_DIGEST_OFFLOAD)) {
825 if (segment->digest_len == 0) {
827 * Even if we offload the digest processing we
828 * splice it in so we can increment the skb/segment
829 * counters in preparation for the data segment.
831 iscsi_tcp_segment_splice_digest(segment,
832 segment->recv_digest);
833 return 0;
836 iscsi_tcp_dgst_header(tcp_conn->rx_hash, hdr,
837 segment->total_copied - ISCSI_DIGEST_SIZE,
838 segment->digest);
840 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
841 return ISCSI_ERR_HDR_DGST;
844 tcp_conn->in.hdr = hdr;
845 return iscsi_tcp_hdr_dissect(conn, hdr);
849 * iscsi_tcp_recv_segment_is_hdr - tests if we are reading in a header
850 * @tcp_conn: iscsi tcp conn
852 * returns non zero if we are currently processing or setup to process
853 * a header.
855 inline int iscsi_tcp_recv_segment_is_hdr(struct iscsi_tcp_conn *tcp_conn)
857 return tcp_conn->in.segment.done == iscsi_tcp_hdr_recv_done;
859 EXPORT_SYMBOL_GPL(iscsi_tcp_recv_segment_is_hdr);
862 * iscsi_tcp_recv_skb - Process skb
863 * @conn: iscsi connection
864 * @skb: network buffer with header and/or data segment
865 * @offset: offset in skb
866 * @offload: bool indicating if transfer was offloaded
868 * Will return status of transfer in status. And will return
869 * number of bytes copied.
871 int iscsi_tcp_recv_skb(struct iscsi_conn *conn, struct sk_buff *skb,
872 unsigned int offset, bool offloaded, int *status)
874 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
875 struct iscsi_segment *segment = &tcp_conn->in.segment;
876 struct skb_seq_state seq;
877 unsigned int consumed = 0;
878 int rc = 0;
880 ISCSI_DBG_TCP(conn, "in %d bytes\n", skb->len - offset);
882 * Update for each skb instead of pdu, because over slow networks a
883 * data_in's data could take a while to read in. We also want to
884 * account for r2ts.
886 conn->last_recv = jiffies;
888 if (unlikely(conn->suspend_rx)) {
889 ISCSI_DBG_TCP(conn, "Rx suspended!\n");
890 *status = ISCSI_TCP_SUSPENDED;
891 return 0;
894 if (offloaded) {
895 segment->total_copied = segment->total_size;
896 goto segment_done;
899 skb_prepare_seq_read(skb, offset, skb->len, &seq);
900 while (1) {
901 unsigned int avail;
902 const u8 *ptr;
904 avail = skb_seq_read(consumed, &ptr, &seq);
905 if (avail == 0) {
906 ISCSI_DBG_TCP(conn, "no more data avail. Consumed %d\n",
907 consumed);
908 *status = ISCSI_TCP_SKB_DONE;
909 skb_abort_seq_read(&seq);
910 goto skb_done;
912 BUG_ON(segment->copied >= segment->size);
914 ISCSI_DBG_TCP(conn, "skb %p ptr=%p avail=%u\n", skb, ptr,
915 avail);
916 rc = iscsi_tcp_segment_recv(tcp_conn, segment, ptr, avail);
917 BUG_ON(rc == 0);
918 consumed += rc;
920 if (segment->total_copied >= segment->total_size) {
921 skb_abort_seq_read(&seq);
922 goto segment_done;
926 segment_done:
927 *status = ISCSI_TCP_SEGMENT_DONE;
928 ISCSI_DBG_TCP(conn, "segment done\n");
929 rc = segment->done(tcp_conn, segment);
930 if (rc != 0) {
931 *status = ISCSI_TCP_CONN_ERR;
932 ISCSI_DBG_TCP(conn, "Error receiving PDU, errno=%d\n", rc);
933 iscsi_conn_failure(conn, rc);
934 return 0;
936 /* The done() functions sets up the next segment. */
938 skb_done:
939 conn->rxdata_octets += consumed;
940 return consumed;
942 EXPORT_SYMBOL_GPL(iscsi_tcp_recv_skb);
945 * iscsi_tcp_task_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
946 * @conn: iscsi connection
947 * @task: scsi command task
948 * @sc: scsi command
950 int iscsi_tcp_task_init(struct iscsi_task *task)
952 struct iscsi_tcp_task *tcp_task = task->dd_data;
953 struct iscsi_conn *conn = task->conn;
954 struct scsi_cmnd *sc = task->sc;
955 int err;
957 if (!sc) {
959 * mgmt tasks do not have a scatterlist since they come
960 * in from the iscsi interface.
962 ISCSI_DBG_TCP(conn, "mtask deq [itt 0x%x]\n", task->itt);
964 return conn->session->tt->init_pdu(task, 0, task->data_count);
967 BUG_ON(kfifo_len(&tcp_task->r2tqueue));
968 tcp_task->exp_datasn = 0;
970 /* Prepare PDU, optionally w/ immediate data */
971 ISCSI_DBG_TCP(conn, "task deq [itt 0x%x imm %d unsol %d]\n",
972 task->itt, task->imm_count, task->unsol_r2t.data_length);
974 err = conn->session->tt->init_pdu(task, 0, task->imm_count);
975 if (err)
976 return err;
977 task->imm_count = 0;
978 return 0;
980 EXPORT_SYMBOL_GPL(iscsi_tcp_task_init);
982 static struct iscsi_r2t_info *iscsi_tcp_get_curr_r2t(struct iscsi_task *task)
984 struct iscsi_session *session = task->conn->session;
985 struct iscsi_tcp_task *tcp_task = task->dd_data;
986 struct iscsi_r2t_info *r2t = NULL;
988 if (iscsi_task_has_unsol_data(task))
989 r2t = &task->unsol_r2t;
990 else {
991 spin_lock_bh(&session->lock);
992 if (tcp_task->r2t) {
993 r2t = tcp_task->r2t;
994 /* Continue with this R2T? */
995 if (r2t->data_length <= r2t->sent) {
996 ISCSI_DBG_TCP(task->conn,
997 " done with r2t %p\n", r2t);
998 kfifo_in(&tcp_task->r2tpool.queue,
999 (void *)&tcp_task->r2t,
1000 sizeof(void *));
1001 tcp_task->r2t = r2t = NULL;
1005 if (r2t == NULL) {
1006 if (kfifo_out(&tcp_task->r2tqueue,
1007 (void *)&tcp_task->r2t, sizeof(void *)) !=
1008 sizeof(void *))
1009 r2t = NULL;
1010 else
1011 r2t = tcp_task->r2t;
1013 spin_unlock_bh(&session->lock);
1016 return r2t;
1020 * iscsi_tcp_task_xmit - xmit normal PDU task
1021 * @task: iscsi command task
1023 * We're expected to return 0 when everything was transmitted successfully,
1024 * -EAGAIN if there's still data in the queue, or != 0 for any other kind
1025 * of error.
1027 int iscsi_tcp_task_xmit(struct iscsi_task *task)
1029 struct iscsi_conn *conn = task->conn;
1030 struct iscsi_session *session = conn->session;
1031 struct iscsi_r2t_info *r2t;
1032 int rc = 0;
1034 flush:
1035 /* Flush any pending data first. */
1036 rc = session->tt->xmit_pdu(task);
1037 if (rc < 0)
1038 return rc;
1040 /* mgmt command */
1041 if (!task->sc) {
1042 if (task->hdr->itt == RESERVED_ITT)
1043 iscsi_put_task(task);
1044 return 0;
1047 /* Are we done already? */
1048 if (task->sc->sc_data_direction != DMA_TO_DEVICE)
1049 return 0;
1051 r2t = iscsi_tcp_get_curr_r2t(task);
1052 if (r2t == NULL) {
1053 /* Waiting for more R2Ts to arrive. */
1054 ISCSI_DBG_TCP(conn, "no R2Ts yet\n");
1055 return 0;
1058 rc = conn->session->tt->alloc_pdu(task, ISCSI_OP_SCSI_DATA_OUT);
1059 if (rc)
1060 return rc;
1061 iscsi_prep_data_out_pdu(task, r2t, (struct iscsi_data *) task->hdr);
1063 ISCSI_DBG_TCP(conn, "sol dout %p [dsn %d itt 0x%x doff %d dlen %d]\n",
1064 r2t, r2t->datasn - 1, task->hdr->itt,
1065 r2t->data_offset + r2t->sent, r2t->data_count);
1067 rc = conn->session->tt->init_pdu(task, r2t->data_offset + r2t->sent,
1068 r2t->data_count);
1069 if (rc) {
1070 iscsi_conn_failure(conn, ISCSI_ERR_XMIT_FAILED);
1071 return rc;
1074 r2t->sent += r2t->data_count;
1075 goto flush;
1077 EXPORT_SYMBOL_GPL(iscsi_tcp_task_xmit);
1079 struct iscsi_cls_conn *
1080 iscsi_tcp_conn_setup(struct iscsi_cls_session *cls_session, int dd_data_size,
1081 uint32_t conn_idx)
1084 struct iscsi_conn *conn;
1085 struct iscsi_cls_conn *cls_conn;
1086 struct iscsi_tcp_conn *tcp_conn;
1088 cls_conn = iscsi_conn_setup(cls_session,
1089 sizeof(*tcp_conn) + dd_data_size, conn_idx);
1090 if (!cls_conn)
1091 return NULL;
1092 conn = cls_conn->dd_data;
1094 * due to strange issues with iser these are not set
1095 * in iscsi_conn_setup
1097 conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1099 tcp_conn = conn->dd_data;
1100 tcp_conn->iscsi_conn = conn;
1101 tcp_conn->dd_data = conn->dd_data + sizeof(*tcp_conn);
1102 return cls_conn;
1104 EXPORT_SYMBOL_GPL(iscsi_tcp_conn_setup);
1106 void iscsi_tcp_conn_teardown(struct iscsi_cls_conn *cls_conn)
1108 iscsi_conn_teardown(cls_conn);
1110 EXPORT_SYMBOL_GPL(iscsi_tcp_conn_teardown);
1112 int iscsi_tcp_r2tpool_alloc(struct iscsi_session *session)
1114 int i;
1115 int cmd_i;
1118 * initialize per-task: R2T pool and xmit queue
1120 for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1121 struct iscsi_task *task = session->cmds[cmd_i];
1122 struct iscsi_tcp_task *tcp_task = task->dd_data;
1125 * pre-allocated x2 as much r2ts to handle race when
1126 * target acks DataOut faster than we data_xmit() queues
1127 * could replenish r2tqueue.
1130 /* R2T pool */
1131 if (iscsi_pool_init(&tcp_task->r2tpool,
1132 session->max_r2t * 2, NULL,
1133 sizeof(struct iscsi_r2t_info))) {
1134 goto r2t_alloc_fail;
1137 /* R2T xmit queue */
1138 if (kfifo_alloc(&tcp_task->r2tqueue,
1139 session->max_r2t * 4 * sizeof(void*), GFP_KERNEL)) {
1140 iscsi_pool_free(&tcp_task->r2tpool);
1141 goto r2t_alloc_fail;
1145 return 0;
1147 r2t_alloc_fail:
1148 for (i = 0; i < cmd_i; i++) {
1149 struct iscsi_task *task = session->cmds[i];
1150 struct iscsi_tcp_task *tcp_task = task->dd_data;
1152 kfifo_free(&tcp_task->r2tqueue);
1153 iscsi_pool_free(&tcp_task->r2tpool);
1155 return -ENOMEM;
1157 EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_alloc);
1159 void iscsi_tcp_r2tpool_free(struct iscsi_session *session)
1161 int i;
1163 for (i = 0; i < session->cmds_max; i++) {
1164 struct iscsi_task *task = session->cmds[i];
1165 struct iscsi_tcp_task *tcp_task = task->dd_data;
1167 kfifo_free(&tcp_task->r2tqueue);
1168 iscsi_pool_free(&tcp_task->r2tpool);
1171 EXPORT_SYMBOL_GPL(iscsi_tcp_r2tpool_free);
1173 void iscsi_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
1174 struct iscsi_stats *stats)
1176 struct iscsi_conn *conn = cls_conn->dd_data;
1178 stats->txdata_octets = conn->txdata_octets;
1179 stats->rxdata_octets = conn->rxdata_octets;
1180 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
1181 stats->dataout_pdus = conn->dataout_pdus_cnt;
1182 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
1183 stats->datain_pdus = conn->datain_pdus_cnt;
1184 stats->r2t_pdus = conn->r2t_pdus_cnt;
1185 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
1186 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
1188 EXPORT_SYMBOL_GPL(iscsi_tcp_conn_get_stats);