2 * iSCSI Initiator over iSER Data-Path
4 * Copyright (C) 2004 Dmitry Yusupov
5 * Copyright (C) 2004 Alex Aizman
6 * Copyright (C) 2005 Mike Christie
7 * Copyright (c) 2005, 2006 Voltaire, Inc. All rights reserved.
8 * maintained by openib-general@openib.org
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
47 * $Id: iscsi_iser.c 6965 2006-05-07 11:36:20Z ogerlitz $
50 #include <linux/types.h>
51 #include <linux/list.h>
52 #include <linux/hardirq.h>
53 #include <linux/kfifo.h>
54 #include <linux/blkdev.h>
55 #include <linux/init.h>
56 #include <linux/ioctl.h>
57 #include <linux/cdev.h>
59 #include <linux/net.h>
60 #include <linux/scatterlist.h>
61 #include <linux/delay.h>
65 #include <asm/uaccess.h>
67 #include <scsi/scsi_cmnd.h>
68 #include <scsi/scsi_device.h>
69 #include <scsi/scsi_eh.h>
70 #include <scsi/scsi_tcq.h>
71 #include <scsi/scsi_host.h>
72 #include <scsi/scsi.h>
73 #include <scsi/scsi_transport_iscsi.h>
75 #include "iscsi_iser.h"
77 static unsigned int iscsi_max_lun
= 512;
78 module_param_named(max_lun
, iscsi_max_lun
, uint
, S_IRUGO
);
80 int iser_debug_level
= 0;
82 MODULE_DESCRIPTION("iSER (iSCSI Extensions for RDMA) Datamover "
83 "v" DRV_VER
" (" DRV_DATE
")");
84 MODULE_LICENSE("Dual BSD/GPL");
85 MODULE_AUTHOR("Alex Nezhinsky, Dan Bar Dov, Or Gerlitz");
87 module_param_named(debug_level
, iser_debug_level
, int, 0644);
88 MODULE_PARM_DESC(debug_level
, "Enable debug tracing if > 0 (default:disabled)");
90 struct iser_global ig
;
93 iscsi_iser_recv(struct iscsi_conn
*conn
,
94 struct iscsi_hdr
*hdr
, char *rx_data
, int rx_data_len
)
101 /* verify PDU length */
102 datalen
= ntoh24(hdr
->dlength
);
103 if (datalen
!= rx_data_len
) {
104 printk(KERN_ERR
"iscsi_iser: datalen %d (hdr) != %d (IB) \n",
105 datalen
, rx_data_len
);
106 rc
= ISCSI_ERR_DATALEN
;
111 ahslen
= hdr
->hlength
* 4;
113 /* verify itt (itt encoding: age+cid+itt) */
114 rc
= iscsi_verify_itt(conn
, hdr
, &ret_itt
);
117 rc
= iscsi_complete_pdu(conn
, hdr
, rx_data
, rx_data_len
);
119 if (rc
&& rc
!= ISCSI_ERR_NO_SCSI_CMD
)
124 iscsi_conn_failure(conn
, rc
);
129 * iscsi_iser_cmd_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
133 iscsi_iser_cmd_init(struct iscsi_cmd_task
*ctask
)
135 struct iscsi_iser_conn
*iser_conn
= ctask
->conn
->dd_data
;
136 struct iscsi_iser_cmd_task
*iser_ctask
= ctask
->dd_data
;
138 iser_ctask
->command_sent
= 0;
139 iser_ctask
->iser_conn
= iser_conn
;
140 iser_ctask_rdma_init(iser_ctask
);
145 * iscsi_mtask_xmit - xmit management(immediate) task
146 * @conn: iscsi connection
147 * @mtask: task management task
150 * The function can return -EAGAIN in which case caller must
151 * call it again later, or recover. '0' return code means successful
156 iscsi_iser_mtask_xmit(struct iscsi_conn
*conn
,
157 struct iscsi_mgmt_task
*mtask
)
161 debug_scsi("mtask deq [cid %d itt 0x%x]\n", conn
->id
, mtask
->itt
);
163 error
= iser_send_control(conn
, mtask
);
165 /* since iser xmits control with zero copy, mtasks can not be recycled
166 * right after sending them.
167 * The recycling scheme is based on whether a response is expected
168 * - if yes, the mtask is recycled at iscsi_complete_pdu
169 * - if no, the mtask is recycled at iser_snd_completion
171 if (error
&& error
!= -ENOBUFS
)
172 iscsi_conn_failure(conn
, ISCSI_ERR_CONN_FAILED
);
178 iscsi_iser_ctask_xmit_unsol_data(struct iscsi_conn
*conn
,
179 struct iscsi_cmd_task
*ctask
)
181 struct iscsi_data hdr
;
184 /* Send data-out PDUs while there's still unsolicited data to send */
185 while (ctask
->unsol_count
> 0) {
186 iscsi_prep_unsolicit_data_pdu(ctask
, &hdr
);
187 debug_scsi("Sending data-out: itt 0x%x, data count %d\n",
188 hdr
.itt
, ctask
->data_count
);
190 /* the buffer description has been passed with the command */
191 /* Send the command */
192 error
= iser_send_data_out(conn
, ctask
, &hdr
);
194 ctask
->unsol_datasn
--;
195 goto iscsi_iser_ctask_xmit_unsol_data_exit
;
197 ctask
->unsol_count
-= ctask
->data_count
;
198 debug_scsi("Need to send %d more as data-out PDUs\n",
202 iscsi_iser_ctask_xmit_unsol_data_exit
:
207 iscsi_iser_ctask_xmit(struct iscsi_conn
*conn
,
208 struct iscsi_cmd_task
*ctask
)
210 struct iscsi_iser_cmd_task
*iser_ctask
= ctask
->dd_data
;
213 if (ctask
->sc
->sc_data_direction
== DMA_TO_DEVICE
) {
214 BUG_ON(scsi_bufflen(ctask
->sc
) == 0);
216 debug_scsi("cmd [itt %x total %d imm %d unsol_data %d\n",
217 ctask
->itt
, scsi_bufflen(ctask
->sc
),
218 ctask
->imm_count
, ctask
->unsol_count
);
221 debug_scsi("ctask deq [cid %d itt 0x%x]\n",
222 conn
->id
, ctask
->itt
);
224 /* Send the cmd PDU */
225 if (!iser_ctask
->command_sent
) {
226 error
= iser_send_command(conn
, ctask
);
228 goto iscsi_iser_ctask_xmit_exit
;
229 iser_ctask
->command_sent
= 1;
232 /* Send unsolicited data-out PDU(s) if necessary */
233 if (ctask
->unsol_count
)
234 error
= iscsi_iser_ctask_xmit_unsol_data(conn
, ctask
);
236 iscsi_iser_ctask_xmit_exit
:
237 if (error
&& error
!= -ENOBUFS
)
238 iscsi_conn_failure(conn
, ISCSI_ERR_CONN_FAILED
);
243 iscsi_iser_cleanup_ctask(struct iscsi_conn
*conn
, struct iscsi_cmd_task
*ctask
)
245 struct iscsi_iser_cmd_task
*iser_ctask
= ctask
->dd_data
;
247 if (iser_ctask
->status
== ISER_TASK_STATUS_STARTED
) {
248 iser_ctask
->status
= ISER_TASK_STATUS_COMPLETED
;
249 iser_ctask_rdma_finalize(iser_ctask
);
253 static struct iser_conn
*
254 iscsi_iser_ib_conn_lookup(__u64 ep_handle
)
256 struct iser_conn
*ib_conn
;
257 struct iser_conn
*uib_conn
= (struct iser_conn
*)(unsigned long)ep_handle
;
259 mutex_lock(&ig
.connlist_mutex
);
260 list_for_each_entry(ib_conn
, &ig
.connlist
, conn_list
) {
261 if (ib_conn
== uib_conn
) {
262 mutex_unlock(&ig
.connlist_mutex
);
266 mutex_unlock(&ig
.connlist_mutex
);
267 iser_err("no conn exists for eph %llx\n",(unsigned long long)ep_handle
);
271 static struct iscsi_cls_conn
*
272 iscsi_iser_conn_create(struct iscsi_cls_session
*cls_session
, uint32_t conn_idx
)
274 struct iscsi_conn
*conn
;
275 struct iscsi_cls_conn
*cls_conn
;
276 struct iscsi_iser_conn
*iser_conn
;
278 cls_conn
= iscsi_conn_setup(cls_session
, conn_idx
);
281 conn
= cls_conn
->dd_data
;
284 * due to issues with the login code re iser sematics
285 * this not set in iscsi_conn_setup - FIXME
287 conn
->max_recv_dlength
= 128;
289 iser_conn
= kzalloc(sizeof(*iser_conn
), GFP_KERNEL
);
291 goto conn_alloc_fail
;
293 /* currently this is the only field which need to be initiated */
294 rwlock_init(&iser_conn
->lock
);
296 conn
->dd_data
= iser_conn
;
297 iser_conn
->iscsi_conn
= conn
;
302 iscsi_conn_teardown(cls_conn
);
307 iscsi_iser_conn_destroy(struct iscsi_cls_conn
*cls_conn
)
309 struct iscsi_conn
*conn
= cls_conn
->dd_data
;
310 struct iscsi_iser_conn
*iser_conn
= conn
->dd_data
;
312 iscsi_conn_teardown(cls_conn
);
313 if (iser_conn
->ib_conn
)
314 iser_conn
->ib_conn
->iser_conn
= NULL
;
319 iscsi_iser_conn_bind(struct iscsi_cls_session
*cls_session
,
320 struct iscsi_cls_conn
*cls_conn
, uint64_t transport_eph
,
323 struct iscsi_conn
*conn
= cls_conn
->dd_data
;
324 struct iscsi_iser_conn
*iser_conn
;
325 struct iser_conn
*ib_conn
;
328 error
= iscsi_conn_bind(cls_session
, cls_conn
, is_leading
);
332 /* the transport ep handle comes from user space so it must be
333 * verified against the global ib connections list */
334 ib_conn
= iscsi_iser_ib_conn_lookup(transport_eph
);
336 iser_err("can't bind eph %llx\n",
337 (unsigned long long)transport_eph
);
340 /* binds the iSER connection retrieved from the previously
341 * connected ep_handle to the iSCSI layer connection. exchanges
342 * connection pointers */
343 iser_err("binding iscsi conn %p to iser_conn %p\n",conn
,ib_conn
);
344 iser_conn
= conn
->dd_data
;
345 ib_conn
->iser_conn
= iser_conn
;
346 iser_conn
->ib_conn
= ib_conn
;
348 conn
->recv_lock
= &iser_conn
->lock
;
354 iscsi_iser_conn_start(struct iscsi_cls_conn
*cls_conn
)
356 struct iscsi_conn
*conn
= cls_conn
->dd_data
;
359 err
= iser_conn_set_full_featured_mode(conn
);
363 return iscsi_conn_start(cls_conn
);
366 static struct iscsi_transport iscsi_iser_transport
;
368 static struct iscsi_cls_session
*
369 iscsi_iser_session_create(struct iscsi_transport
*iscsit
,
370 struct scsi_transport_template
*scsit
,
371 uint16_t cmds_max
, uint16_t qdepth
,
372 uint32_t initial_cmdsn
, uint32_t *hostno
)
374 struct iscsi_cls_session
*cls_session
;
375 struct iscsi_session
*session
;
378 struct iscsi_cmd_task
*ctask
;
379 struct iscsi_mgmt_task
*mtask
;
380 struct iscsi_iser_cmd_task
*iser_ctask
;
381 struct iser_desc
*desc
;
384 * we do not support setting can_queue cmd_per_lun from userspace yet
385 * because we preallocate so many resources
387 cls_session
= iscsi_session_setup(iscsit
, scsit
,
388 ISCSI_DEF_XMIT_CMDS_MAX
,
389 ISCSI_MAX_CMD_PER_LUN
,
390 sizeof(struct iscsi_iser_cmd_task
),
391 sizeof(struct iser_desc
),
397 session
= class_to_transport_session(cls_session
);
399 /* libiscsi setup itts, data and pool so just set desc fields */
400 for (i
= 0; i
< session
->cmds_max
; i
++) {
401 ctask
= session
->cmds
[i
];
402 iser_ctask
= ctask
->dd_data
;
403 ctask
->hdr
= (struct iscsi_cmd
*)&iser_ctask
->desc
.iscsi_header
;
404 ctask
->hdr_max
= sizeof(iser_ctask
->desc
.iscsi_header
);
407 for (i
= 0; i
< session
->mgmtpool_max
; i
++) {
408 mtask
= session
->mgmt_cmds
[i
];
409 desc
= mtask
->dd_data
;
410 mtask
->hdr
= &desc
->iscsi_header
;
411 desc
->data
= mtask
->data
;
418 iscsi_iser_set_param(struct iscsi_cls_conn
*cls_conn
,
419 enum iscsi_param param
, char *buf
, int buflen
)
424 case ISCSI_PARAM_MAX_RECV_DLENGTH
:
427 case ISCSI_PARAM_HDRDGST_EN
:
428 sscanf(buf
, "%d", &value
);
430 printk(KERN_ERR
"DataDigest wasn't negotiated to None");
434 case ISCSI_PARAM_DATADGST_EN
:
435 sscanf(buf
, "%d", &value
);
437 printk(KERN_ERR
"DataDigest wasn't negotiated to None");
441 case ISCSI_PARAM_IFMARKER_EN
:
442 sscanf(buf
, "%d", &value
);
444 printk(KERN_ERR
"IFMarker wasn't negotiated to No");
448 case ISCSI_PARAM_OFMARKER_EN
:
449 sscanf(buf
, "%d", &value
);
451 printk(KERN_ERR
"OFMarker wasn't negotiated to No");
456 return iscsi_set_param(cls_conn
, param
, buf
, buflen
);
463 iscsi_iser_conn_get_stats(struct iscsi_cls_conn
*cls_conn
, struct iscsi_stats
*stats
)
465 struct iscsi_conn
*conn
= cls_conn
->dd_data
;
467 stats
->txdata_octets
= conn
->txdata_octets
;
468 stats
->rxdata_octets
= conn
->rxdata_octets
;
469 stats
->scsicmd_pdus
= conn
->scsicmd_pdus_cnt
;
470 stats
->dataout_pdus
= conn
->dataout_pdus_cnt
;
471 stats
->scsirsp_pdus
= conn
->scsirsp_pdus_cnt
;
472 stats
->datain_pdus
= conn
->datain_pdus_cnt
; /* always 0 */
473 stats
->r2t_pdus
= conn
->r2t_pdus_cnt
; /* always 0 */
474 stats
->tmfcmd_pdus
= conn
->tmfcmd_pdus_cnt
;
475 stats
->tmfrsp_pdus
= conn
->tmfrsp_pdus_cnt
;
476 stats
->custom_length
= 3;
477 strcpy(stats
->custom
[0].desc
, "qp_tx_queue_full");
478 stats
->custom
[0].value
= 0; /* TB iser_conn->qp_tx_queue_full; */
479 strcpy(stats
->custom
[1].desc
, "fmr_map_not_avail");
480 stats
->custom
[1].value
= 0; /* TB iser_conn->fmr_map_not_avail */;
481 strcpy(stats
->custom
[2].desc
, "eh_abort_cnt");
482 stats
->custom
[2].value
= conn
->eh_abort_cnt
;
486 iscsi_iser_ep_connect(struct sockaddr
*dst_addr
, int non_blocking
,
490 struct iser_conn
*ib_conn
;
492 err
= iser_conn_init(&ib_conn
);
496 err
= iser_connect(ib_conn
, NULL
, (struct sockaddr_in
*)dst_addr
, non_blocking
);
498 *ep_handle
= (__u64
)(unsigned long)ib_conn
;
505 iscsi_iser_ep_poll(__u64 ep_handle
, int timeout_ms
)
507 struct iser_conn
*ib_conn
= iscsi_iser_ib_conn_lookup(ep_handle
);
513 rc
= wait_event_interruptible_timeout(ib_conn
->wait
,
514 ib_conn
->state
== ISER_CONN_UP
,
515 msecs_to_jiffies(timeout_ms
));
517 /* if conn establishment failed, return error code to iscsi */
519 (ib_conn
->state
== ISER_CONN_TERMINATING
||
520 ib_conn
->state
== ISER_CONN_DOWN
))
523 iser_err("ib conn %p rc = %d\n", ib_conn
, rc
);
526 return 1; /* success, this is the equivalent of POLLOUT */
528 return 0; /* timeout */
530 return rc
; /* signal */
534 iscsi_iser_ep_disconnect(__u64 ep_handle
)
536 struct iser_conn
*ib_conn
;
538 ib_conn
= iscsi_iser_ib_conn_lookup(ep_handle
);
542 iser_err("ib conn %p state %d\n",ib_conn
, ib_conn
->state
);
543 iser_conn_terminate(ib_conn
);
546 static struct scsi_host_template iscsi_iser_sht
= {
547 .module
= THIS_MODULE
,
548 .name
= "iSCSI Initiator over iSER, v." DRV_VER
,
549 .queuecommand
= iscsi_queuecommand
,
550 .change_queue_depth
= iscsi_change_queue_depth
,
551 .can_queue
= ISCSI_DEF_XMIT_CMDS_MAX
- 1,
552 .sg_tablesize
= ISCSI_ISER_SG_TABLESIZE
,
554 .cmd_per_lun
= ISCSI_MAX_CMD_PER_LUN
,
555 .eh_abort_handler
= iscsi_eh_abort
,
556 .eh_device_reset_handler
= iscsi_eh_device_reset
,
557 .eh_host_reset_handler
= iscsi_eh_host_reset
,
558 .use_clustering
= DISABLE_CLUSTERING
,
559 .proc_name
= "iscsi_iser",
563 static struct iscsi_transport iscsi_iser_transport
= {
564 .owner
= THIS_MODULE
,
566 .caps
= CAP_RECOVERY_L0
| CAP_MULTI_R2T
,
567 .param_mask
= ISCSI_MAX_RECV_DLENGTH
|
568 ISCSI_MAX_XMIT_DLENGTH
|
571 ISCSI_INITIAL_R2T_EN
|
576 ISCSI_PDU_INORDER_EN
|
577 ISCSI_DATASEQ_INORDER_EN
|
579 ISCSI_PERSISTENT_PORT
|
580 ISCSI_PERSISTENT_ADDRESS
|
581 ISCSI_TARGET_NAME
| ISCSI_TPGT
|
582 ISCSI_USERNAME
| ISCSI_PASSWORD
|
583 ISCSI_USERNAME_IN
| ISCSI_PASSWORD_IN
|
584 ISCSI_FAST_ABORT
| ISCSI_ABORT_TMO
|
585 ISCSI_PING_TMO
| ISCSI_RECV_TMO
,
586 .host_param_mask
= ISCSI_HOST_HWADDRESS
|
587 ISCSI_HOST_NETDEV_NAME
|
588 ISCSI_HOST_INITIATOR_NAME
,
589 .host_template
= &iscsi_iser_sht
,
590 .conndata_size
= sizeof(struct iscsi_conn
),
591 .max_lun
= ISCSI_ISER_MAX_LUN
,
592 .max_cmd_len
= ISCSI_ISER_MAX_CMD_LEN
,
593 /* session management */
594 .create_session
= iscsi_iser_session_create
,
595 .destroy_session
= iscsi_session_teardown
,
596 /* connection management */
597 .create_conn
= iscsi_iser_conn_create
,
598 .bind_conn
= iscsi_iser_conn_bind
,
599 .destroy_conn
= iscsi_iser_conn_destroy
,
600 .set_param
= iscsi_iser_set_param
,
601 .get_conn_param
= iscsi_conn_get_param
,
602 .get_session_param
= iscsi_session_get_param
,
603 .start_conn
= iscsi_iser_conn_start
,
604 .stop_conn
= iscsi_conn_stop
,
605 /* iscsi host params */
606 .get_host_param
= iscsi_host_get_param
,
607 .set_host_param
= iscsi_host_set_param
,
609 .send_pdu
= iscsi_conn_send_pdu
,
610 .get_stats
= iscsi_iser_conn_get_stats
,
611 .init_cmd_task
= iscsi_iser_cmd_init
,
612 .xmit_cmd_task
= iscsi_iser_ctask_xmit
,
613 .xmit_mgmt_task
= iscsi_iser_mtask_xmit
,
614 .cleanup_cmd_task
= iscsi_iser_cleanup_ctask
,
616 .session_recovery_timedout
= iscsi_session_recovery_timedout
,
618 .ep_connect
= iscsi_iser_ep_connect
,
619 .ep_poll
= iscsi_iser_ep_poll
,
620 .ep_disconnect
= iscsi_iser_ep_disconnect
623 static int __init
iser_init(void)
627 iser_dbg("Starting iSER datamover...\n");
629 if (iscsi_max_lun
< 1) {
630 printk(KERN_ERR
"Invalid max_lun value of %u\n", iscsi_max_lun
);
634 iscsi_iser_transport
.max_lun
= iscsi_max_lun
;
636 memset(&ig
, 0, sizeof(struct iser_global
));
638 ig
.desc_cache
= kmem_cache_create("iser_descriptors",
639 sizeof (struct iser_desc
),
640 0, SLAB_HWCACHE_ALIGN
,
642 if (ig
.desc_cache
== NULL
)
645 /* device init is called only after the first addr resolution */
646 mutex_init(&ig
.device_list_mutex
);
647 INIT_LIST_HEAD(&ig
.device_list
);
648 mutex_init(&ig
.connlist_mutex
);
649 INIT_LIST_HEAD(&ig
.connlist
);
651 if (!iscsi_register_transport(&iscsi_iser_transport
)) {
652 iser_err("iscsi_register_transport failed\n");
654 goto register_transport_failure
;
659 register_transport_failure
:
660 kmem_cache_destroy(ig
.desc_cache
);
665 static void __exit
iser_exit(void)
667 iser_dbg("Removing iSER datamover...\n");
668 iscsi_unregister_transport(&iscsi_iser_transport
);
669 kmem_cache_destroy(ig
.desc_cache
);
672 module_init(iser_init
);
673 module_exit(iser_exit
);