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 struct scsi_host_template iscsi_iser_sht;
78 static struct iscsi_transport iscsi_iser_transport;
79 static struct scsi_transport_template *iscsi_iser_scsi_transport;
81 static unsigned int iscsi_max_lun = 512;
82 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
84 int iser_debug_level = 0;
86 MODULE_DESCRIPTION("iSER (iSCSI Extensions for RDMA) Datamover "
87 "v" DRV_VER " (" DRV_DATE ")");
88 MODULE_LICENSE("Dual BSD/GPL");
89 MODULE_AUTHOR("Alex Nezhinsky, Dan Bar Dov, Or Gerlitz");
91 module_param_named(debug_level, iser_debug_level, int, 0644);
92 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:disabled)");
94 struct iser_global ig;
97 iscsi_iser_recv(struct iscsi_conn *conn,
98 struct iscsi_hdr *hdr, char *rx_data, int rx_data_len)
104 /* verify PDU length */
105 datalen = ntoh24(hdr->dlength);
106 if (datalen != rx_data_len) {
107 printk(KERN_ERR "iscsi_iser: datalen %d (hdr) != %d (IB) \n",
108 datalen, rx_data_len);
109 rc = ISCSI_ERR_DATALEN;
114 ahslen = hdr->hlength * 4;
116 rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len);
117 if (rc && rc != ISCSI_ERR_NO_SCSI_CMD)
122 iscsi_conn_failure(conn, rc);
127 * iscsi_iser_task_init - Initialize ctask
128 * @ctask: iscsi ctask
130 * Initialize the ctask for the scsi command or mgmt command.
133 iscsi_iser_task_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;
140 iser_ctask->desc.data = ctask->data;
144 iser_ctask->command_sent = 0;
145 iser_ctask->iser_conn = iser_conn;
146 iser_ctask_rdma_init(iser_ctask);
151 * iscsi_iser_mtask_xmit - xmit management(immediate) ctask
152 * @conn: iscsi connection
153 * @ctask: ctask management ctask
156 * The function can return -EAGAIN in which case caller must
157 * call it again later, or recover. '0' return code means successful
162 iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
166 debug_scsi("ctask deq [cid %d itt 0x%x]\n", conn->id, ctask->itt);
168 error = iser_send_control(conn, ctask);
170 /* since iser xmits control with zero copy, ctasks can not be recycled
171 * right after sending them.
172 * The recycling scheme is based on whether a response is expected
173 * - if yes, the ctask is recycled at iscsi_complete_pdu
174 * - if no, the ctask is recycled at iser_snd_completion
176 if (error && error != -ENOBUFS)
177 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
183 iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn,
184 struct iscsi_cmd_task *ctask)
186 struct iscsi_data hdr;
189 /* Send data-out PDUs while there's still unsolicited data to send */
190 while (ctask->unsol_count > 0) {
191 iscsi_prep_unsolicit_data_pdu(ctask, &hdr);
192 debug_scsi("Sending data-out: itt 0x%x, data count %d\n",
193 hdr.itt, ctask->data_count);
195 /* the buffer description has been passed with the command */
196 /* Send the command */
197 error = iser_send_data_out(conn, ctask, &hdr);
199 ctask->unsol_datasn--;
200 goto iscsi_iser_task_xmit_unsol_data_exit;
202 ctask->unsol_count -= ctask->data_count;
203 debug_scsi("Need to send %d more as data-out PDUs\n",
207 iscsi_iser_task_xmit_unsol_data_exit:
212 iscsi_iser_task_xmit(struct iscsi_cmd_task *ctask)
214 struct iscsi_conn *conn = ctask->conn;
215 struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data;
219 return iscsi_iser_mtask_xmit(conn, ctask);
221 if (ctask->sc->sc_data_direction == DMA_TO_DEVICE) {
222 BUG_ON(scsi_bufflen(ctask->sc) == 0);
224 debug_scsi("cmd [itt %x total %d imm %d unsol_data %d\n",
225 ctask->itt, scsi_bufflen(ctask->sc),
226 ctask->imm_count, ctask->unsol_count);
229 debug_scsi("ctask deq [cid %d itt 0x%x]\n",
230 conn->id, ctask->itt);
232 /* Send the cmd PDU */
233 if (!iser_ctask->command_sent) {
234 error = iser_send_command(conn, ctask);
236 goto iscsi_iser_task_xmit_exit;
237 iser_ctask->command_sent = 1;
240 /* Send unsolicited data-out PDU(s) if necessary */
241 if (ctask->unsol_count)
242 error = iscsi_iser_task_xmit_unsol_data(conn, ctask);
244 iscsi_iser_task_xmit_exit:
245 if (error && error != -ENOBUFS)
246 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
251 iscsi_iser_cleanup_task(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
253 struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data;
255 /* mgmt tasks do not need special cleanup */
259 if (iser_ctask->status == ISER_TASK_STATUS_STARTED) {
260 iser_ctask->status = ISER_TASK_STATUS_COMPLETED;
261 iser_ctask_rdma_finalize(iser_ctask);
265 static struct iser_conn *
266 iscsi_iser_ib_conn_lookup(__u64 ep_handle)
268 struct iser_conn *ib_conn;
269 struct iser_conn *uib_conn = (struct iser_conn *)(unsigned long)ep_handle;
271 mutex_lock(&ig.connlist_mutex);
272 list_for_each_entry(ib_conn, &ig.connlist, conn_list) {
273 if (ib_conn == uib_conn) {
274 mutex_unlock(&ig.connlist_mutex);
278 mutex_unlock(&ig.connlist_mutex);
279 iser_err("no conn exists for eph %llx\n",(unsigned long long)ep_handle);
283 static struct iscsi_cls_conn *
284 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
286 struct iscsi_conn *conn;
287 struct iscsi_cls_conn *cls_conn;
288 struct iscsi_iser_conn *iser_conn;
290 cls_conn = iscsi_conn_setup(cls_session, sizeof(*iser_conn), conn_idx);
293 conn = cls_conn->dd_data;
296 * due to issues with the login code re iser sematics
297 * this not set in iscsi_conn_setup - FIXME
299 conn->max_recv_dlength = 128;
301 iser_conn = conn->dd_data;
302 /* currently this is the only field which need to be initiated */
303 rwlock_init(&iser_conn->lock);
305 conn->dd_data = iser_conn;
306 iser_conn->iscsi_conn = conn;
312 iscsi_iser_conn_destroy(struct iscsi_cls_conn *cls_conn)
314 struct iscsi_conn *conn = cls_conn->dd_data;
315 struct iscsi_iser_conn *iser_conn = conn->dd_data;
316 struct iser_conn *ib_conn = iser_conn->ib_conn;
318 iscsi_conn_teardown(cls_conn);
320 * Userspace will normally call the stop callback and
321 * already have freed the ib_conn, but if it goofed up then
325 ib_conn->iser_conn = NULL;
326 iser_conn_put(ib_conn);
331 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
332 struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
335 struct iscsi_conn *conn = cls_conn->dd_data;
336 struct iscsi_iser_conn *iser_conn;
337 struct iser_conn *ib_conn;
340 error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
344 /* the transport ep handle comes from user space so it must be
345 * verified against the global ib connections list */
346 ib_conn = iscsi_iser_ib_conn_lookup(transport_eph);
348 iser_err("can't bind eph %llx\n",
349 (unsigned long long)transport_eph);
352 /* binds the iSER connection retrieved from the previously
353 * connected ep_handle to the iSCSI layer connection. exchanges
354 * connection pointers */
355 iser_err("binding iscsi conn %p to iser_conn %p\n",conn,ib_conn);
356 iser_conn = conn->dd_data;
357 ib_conn->iser_conn = iser_conn;
358 iser_conn->ib_conn = ib_conn;
359 iser_conn_get(ib_conn);
361 conn->recv_lock = &iser_conn->lock;
367 iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
369 struct iscsi_conn *conn = cls_conn->dd_data;
370 struct iscsi_iser_conn *iser_conn = conn->dd_data;
371 struct iser_conn *ib_conn = iser_conn->ib_conn;
373 iscsi_conn_stop(cls_conn, flag);
375 * There is no unbind event so the stop callback
376 * must release the ref from the bind.
378 iser_conn_put(ib_conn);
379 iser_conn->ib_conn = NULL;
383 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
385 struct iscsi_conn *conn = cls_conn->dd_data;
388 err = iser_conn_set_full_featured_mode(conn);
392 return iscsi_conn_start(cls_conn);
395 static void iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session)
397 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
399 iscsi_host_remove(shost);
400 iscsi_host_free(shost);
403 static struct iscsi_cls_session *
404 iscsi_iser_session_create(struct Scsi_Host *shost,
405 uint16_t cmds_max, uint16_t qdepth,
406 uint32_t initial_cmdsn, uint32_t *hostno)
408 struct iscsi_cls_session *cls_session;
409 struct iscsi_session *session;
411 struct iscsi_cmd_task *ctask;
412 struct iscsi_iser_cmd_task *iser_ctask;
415 printk(KERN_ERR "iscsi_tcp: invalid shost %d.\n",
420 shost = iscsi_host_alloc(&iscsi_iser_sht, 0, ISCSI_MAX_CMD_PER_LUN);
423 shost->transportt = iscsi_iser_scsi_transport;
424 shost->max_lun = iscsi_max_lun;
426 shost->max_channel = 0;
427 shost->max_cmd_len = 16;
429 if (iscsi_host_add(shost, NULL))
431 *hostno = shost->host_no;
434 * we do not support setting can_queue cmd_per_lun from userspace yet
435 * because we preallocate so many resources
437 cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
438 ISCSI_DEF_XMIT_CMDS_MAX,
439 sizeof(struct iscsi_iser_cmd_task),
443 session = cls_session->dd_data;
445 shost->can_queue = session->scsi_cmds_max;
446 /* libiscsi setup itts, data and pool so just set desc fields */
447 for (i = 0; i < session->cmds_max; i++) {
448 ctask = session->cmds[i];
449 iser_ctask = ctask->dd_data;
450 ctask->hdr = (struct iscsi_cmd *)&iser_ctask->desc.iscsi_header;
451 ctask->hdr_max = sizeof(iser_ctask->desc.iscsi_header);
456 iscsi_host_remove(shost);
458 iscsi_host_free(shost);
463 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
464 enum iscsi_param param, char *buf, int buflen)
469 case ISCSI_PARAM_MAX_RECV_DLENGTH:
472 case ISCSI_PARAM_HDRDGST_EN:
473 sscanf(buf, "%d", &value);
475 printk(KERN_ERR "DataDigest wasn't negotiated to None");
479 case ISCSI_PARAM_DATADGST_EN:
480 sscanf(buf, "%d", &value);
482 printk(KERN_ERR "DataDigest wasn't negotiated to None");
486 case ISCSI_PARAM_IFMARKER_EN:
487 sscanf(buf, "%d", &value);
489 printk(KERN_ERR "IFMarker wasn't negotiated to No");
493 case ISCSI_PARAM_OFMARKER_EN:
494 sscanf(buf, "%d", &value);
496 printk(KERN_ERR "OFMarker wasn't negotiated to No");
501 return iscsi_set_param(cls_conn, param, buf, buflen);
508 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
510 struct iscsi_conn *conn = cls_conn->dd_data;
512 stats->txdata_octets = conn->txdata_octets;
513 stats->rxdata_octets = conn->rxdata_octets;
514 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
515 stats->dataout_pdus = conn->dataout_pdus_cnt;
516 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
517 stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
518 stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
519 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
520 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
521 stats->custom_length = 4;
522 strcpy(stats->custom[0].desc, "qp_tx_queue_full");
523 stats->custom[0].value = 0; /* TB iser_conn->qp_tx_queue_full; */
524 strcpy(stats->custom[1].desc, "fmr_map_not_avail");
525 stats->custom[1].value = 0; /* TB iser_conn->fmr_map_not_avail */;
526 strcpy(stats->custom[2].desc, "eh_abort_cnt");
527 stats->custom[2].value = conn->eh_abort_cnt;
528 strcpy(stats->custom[3].desc, "fmr_unalign_cnt");
529 stats->custom[3].value = conn->fmr_unalign_cnt;
533 iscsi_iser_ep_connect(struct sockaddr *dst_addr, int non_blocking,
537 struct iser_conn *ib_conn;
539 err = iser_conn_init(&ib_conn);
543 err = iser_connect(ib_conn, NULL, (struct sockaddr_in *)dst_addr, non_blocking);
545 *ep_handle = (__u64)(unsigned long)ib_conn;
552 iscsi_iser_ep_poll(__u64 ep_handle, int timeout_ms)
554 struct iser_conn *ib_conn = iscsi_iser_ib_conn_lookup(ep_handle);
560 rc = wait_event_interruptible_timeout(ib_conn->wait,
561 ib_conn->state == ISER_CONN_UP,
562 msecs_to_jiffies(timeout_ms));
564 /* if conn establishment failed, return error code to iscsi */
566 (ib_conn->state == ISER_CONN_TERMINATING ||
567 ib_conn->state == ISER_CONN_DOWN))
570 iser_err("ib conn %p rc = %d\n", ib_conn, rc);
573 return 1; /* success, this is the equivalent of POLLOUT */
575 return 0; /* timeout */
577 return rc; /* signal */
581 iscsi_iser_ep_disconnect(__u64 ep_handle)
583 struct iser_conn *ib_conn;
585 ib_conn = iscsi_iser_ib_conn_lookup(ep_handle);
589 if (ib_conn->iser_conn)
591 * Must suspend xmit path if the ep is bound to the
592 * iscsi_conn, so we know we are not accessing the ib_conn
595 * This may not be bound if the ep poll failed.
597 iscsi_suspend_tx(ib_conn->iser_conn->iscsi_conn);
600 iser_err("ib conn %p state %d\n",ib_conn, ib_conn->state);
601 iser_conn_terminate(ib_conn);
604 static struct scsi_host_template iscsi_iser_sht = {
605 .module = THIS_MODULE,
606 .name = "iSCSI Initiator over iSER, v." DRV_VER,
607 .queuecommand = iscsi_queuecommand,
608 .change_queue_depth = iscsi_change_queue_depth,
609 .can_queue = ISCSI_DEF_XMIT_CMDS_MAX - 1,
610 .sg_tablesize = ISCSI_ISER_SG_TABLESIZE,
612 .cmd_per_lun = ISCSI_MAX_CMD_PER_LUN,
613 .eh_abort_handler = iscsi_eh_abort,
614 .eh_device_reset_handler= iscsi_eh_device_reset,
615 .eh_host_reset_handler = iscsi_eh_host_reset,
616 .use_clustering = DISABLE_CLUSTERING,
617 .proc_name = "iscsi_iser",
621 static struct iscsi_transport iscsi_iser_transport = {
622 .owner = THIS_MODULE,
624 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T,
625 .param_mask = ISCSI_MAX_RECV_DLENGTH |
626 ISCSI_MAX_XMIT_DLENGTH |
629 ISCSI_INITIAL_R2T_EN |
634 ISCSI_PDU_INORDER_EN |
635 ISCSI_DATASEQ_INORDER_EN |
637 ISCSI_PERSISTENT_PORT |
638 ISCSI_PERSISTENT_ADDRESS |
639 ISCSI_TARGET_NAME | ISCSI_TPGT |
640 ISCSI_USERNAME | ISCSI_PASSWORD |
641 ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
642 ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
643 ISCSI_PING_TMO | ISCSI_RECV_TMO,
644 .host_param_mask = ISCSI_HOST_HWADDRESS |
645 ISCSI_HOST_NETDEV_NAME |
646 ISCSI_HOST_INITIATOR_NAME,
647 /* session management */
648 .create_session = iscsi_iser_session_create,
649 .destroy_session = iscsi_iser_session_destroy,
650 /* connection management */
651 .create_conn = iscsi_iser_conn_create,
652 .bind_conn = iscsi_iser_conn_bind,
653 .destroy_conn = iscsi_iser_conn_destroy,
654 .set_param = iscsi_iser_set_param,
655 .get_conn_param = iscsi_conn_get_param,
656 .get_session_param = iscsi_session_get_param,
657 .start_conn = iscsi_iser_conn_start,
658 .stop_conn = iscsi_iser_conn_stop,
659 /* iscsi host params */
660 .get_host_param = iscsi_host_get_param,
661 .set_host_param = iscsi_host_set_param,
663 .send_pdu = iscsi_conn_send_pdu,
664 .get_stats = iscsi_iser_conn_get_stats,
665 .init_task = iscsi_iser_task_init,
666 .xmit_task = iscsi_iser_task_xmit,
667 .cleanup_task = iscsi_iser_cleanup_task,
669 .session_recovery_timedout = iscsi_session_recovery_timedout,
671 .ep_connect = iscsi_iser_ep_connect,
672 .ep_poll = iscsi_iser_ep_poll,
673 .ep_disconnect = iscsi_iser_ep_disconnect
676 static int __init iser_init(void)
680 iser_dbg("Starting iSER datamover...\n");
682 if (iscsi_max_lun < 1) {
683 printk(KERN_ERR "Invalid max_lun value of %u\n", iscsi_max_lun);
687 memset(&ig, 0, sizeof(struct iser_global));
689 ig.desc_cache = kmem_cache_create("iser_descriptors",
690 sizeof (struct iser_desc),
691 0, SLAB_HWCACHE_ALIGN,
693 if (ig.desc_cache == NULL)
696 /* device init is called only after the first addr resolution */
697 mutex_init(&ig.device_list_mutex);
698 INIT_LIST_HEAD(&ig.device_list);
699 mutex_init(&ig.connlist_mutex);
700 INIT_LIST_HEAD(&ig.connlist);
702 iscsi_iser_scsi_transport = iscsi_register_transport(
703 &iscsi_iser_transport);
704 if (!iscsi_iser_scsi_transport) {
705 iser_err("iscsi_register_transport failed\n");
707 goto register_transport_failure;
712 register_transport_failure:
713 kmem_cache_destroy(ig.desc_cache);
718 static void __exit iser_exit(void)
720 iser_dbg("Removing iSER datamover...\n");
721 iscsi_unregister_transport(&iscsi_iser_transport);
722 kmem_cache_destroy(ig.desc_cache);
725 module_init(iser_init);
726 module_exit(iser_exit);