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;
137 struct scsi_cmnd *sc = ctask->sc;
139 iser_ctask->command_sent = 0;
140 iser_ctask->iser_conn = iser_conn;
142 if (sc->sc_data_direction == DMA_TO_DEVICE) {
143 BUG_ON(ctask->total_length == 0);
144 /* bytes to be sent via RDMA operations */
145 iser_ctask->rdma_data_count = ctask->total_length -
149 debug_scsi("cmd [itt %x total %d imm %d unsol_data %d "
151 ctask->itt, ctask->total_length, ctask->imm_count,
152 ctask->unsol_count, iser_ctask->rdma_data_count);
154 /* bytes to be sent via RDMA operations */
155 iser_ctask->rdma_data_count = ctask->total_length;
157 iser_ctask_rdma_init(iser_ctask);
161 * iscsi_mtask_xmit - xmit management(immediate) task
162 * @conn: iscsi connection
163 * @mtask: task management task
166 * The function can return -EAGAIN in which case caller must
167 * call it again later, or recover. '0' return code means successful
172 iscsi_iser_mtask_xmit(struct iscsi_conn *conn,
173 struct iscsi_mgmt_task *mtask)
177 debug_scsi("mtask deq [cid %d itt 0x%x]\n", conn->id, mtask->itt);
179 error = iser_send_control(conn, mtask);
181 /* since iser xmits control with zero copy, mtasks can not be recycled
182 * right after sending them.
183 * The recycling scheme is based on whether a response is expected
184 * - if yes, the mtask is recycled at iscsi_complete_pdu
185 * - if no, the mtask is recycled at iser_snd_completion
187 if (error && error != -EAGAIN)
188 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
194 iscsi_iser_ctask_xmit_unsol_data(struct iscsi_conn *conn,
195 struct iscsi_cmd_task *ctask)
197 struct iscsi_data hdr;
199 struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data;
201 /* Send data-out PDUs while there's still unsolicited data to send */
202 while (ctask->unsol_count > 0) {
203 iscsi_prep_unsolicit_data_pdu(ctask, &hdr,
204 iser_ctask->rdma_data_count);
206 debug_scsi("Sending data-out: itt 0x%x, data count %d\n",
207 hdr.itt, ctask->data_count);
209 /* the buffer description has been passed with the command */
210 /* Send the command */
211 error = iser_send_data_out(conn, ctask, &hdr);
213 ctask->unsol_datasn--;
214 goto iscsi_iser_ctask_xmit_unsol_data_exit;
216 ctask->unsol_count -= ctask->data_count;
217 debug_scsi("Need to send %d more as data-out PDUs\n",
221 iscsi_iser_ctask_xmit_unsol_data_exit:
226 iscsi_iser_ctask_xmit(struct iscsi_conn *conn,
227 struct iscsi_cmd_task *ctask)
229 struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data;
232 debug_scsi("ctask deq [cid %d itt 0x%x]\n",
233 conn->id, ctask->itt);
236 * serialize with TMF AbortTask
241 /* Send the cmd PDU */
242 if (!iser_ctask->command_sent) {
243 error = iser_send_command(conn, ctask);
245 goto iscsi_iser_ctask_xmit_exit;
246 iser_ctask->command_sent = 1;
249 /* Send unsolicited data-out PDU(s) if necessary */
250 if (ctask->unsol_count)
251 error = iscsi_iser_ctask_xmit_unsol_data(conn, ctask);
253 iscsi_iser_ctask_xmit_exit:
254 if (error && error != -EAGAIN)
255 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
260 iscsi_iser_cleanup_ctask(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
262 struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data;
264 if (iser_ctask->status == ISER_TASK_STATUS_STARTED) {
265 iser_ctask->status = ISER_TASK_STATUS_COMPLETED;
266 iser_ctask_rdma_finalize(iser_ctask);
270 static struct iser_conn *
271 iscsi_iser_ib_conn_lookup(__u64 ep_handle)
273 struct iser_conn *ib_conn;
274 struct iser_conn *uib_conn = (struct iser_conn *)(unsigned long)ep_handle;
276 mutex_lock(&ig.connlist_mutex);
277 list_for_each_entry(ib_conn, &ig.connlist, conn_list) {
278 if (ib_conn == uib_conn) {
279 mutex_unlock(&ig.connlist_mutex);
283 mutex_unlock(&ig.connlist_mutex);
284 iser_err("no conn exists for eph %llx\n",(unsigned long long)ep_handle);
288 static struct iscsi_cls_conn *
289 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
291 struct iscsi_conn *conn;
292 struct iscsi_cls_conn *cls_conn;
293 struct iscsi_iser_conn *iser_conn;
295 cls_conn = iscsi_conn_setup(cls_session, conn_idx);
298 conn = cls_conn->dd_data;
301 * due to issues with the login code re iser sematics
302 * this not set in iscsi_conn_setup - FIXME
304 conn->max_recv_dlength = 128;
306 iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL);
308 goto conn_alloc_fail;
310 /* currently this is the only field which need to be initiated */
311 rwlock_init(&iser_conn->lock);
313 conn->dd_data = iser_conn;
314 iser_conn->iscsi_conn = conn;
319 iscsi_conn_teardown(cls_conn);
324 iscsi_iser_conn_destroy(struct iscsi_cls_conn *cls_conn)
326 struct iscsi_conn *conn = cls_conn->dd_data;
327 struct iscsi_iser_conn *iser_conn = conn->dd_data;
329 iscsi_conn_teardown(cls_conn);
334 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
335 struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
338 struct iscsi_conn *conn = cls_conn->dd_data;
339 struct iscsi_iser_conn *iser_conn;
340 struct iser_conn *ib_conn;
343 error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
347 /* the transport ep handle comes from user space so it must be
348 * verified against the global ib connections list */
349 ib_conn = iscsi_iser_ib_conn_lookup(transport_eph);
351 iser_err("can't bind eph %llx\n",
352 (unsigned long long)transport_eph);
355 /* binds the iSER connection retrieved from the previously
356 * connected ep_handle to the iSCSI layer connection. exchanges
357 * connection pointers */
358 iser_err("binding iscsi conn %p to iser_conn %p\n",conn,ib_conn);
359 iser_conn = conn->dd_data;
360 ib_conn->iser_conn = iser_conn;
361 iser_conn->ib_conn = ib_conn;
363 conn->recv_lock = &iser_conn->lock;
369 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
371 struct iscsi_conn *conn = cls_conn->dd_data;
374 err = iscsi_conn_start(cls_conn);
378 return iser_conn_set_full_featured_mode(conn);
382 iscsi_iser_conn_terminate(struct iscsi_conn *conn)
384 struct iscsi_iser_conn *iser_conn = conn->dd_data;
385 struct iser_conn *ib_conn = iser_conn->ib_conn;
388 /* starts conn teardown process, waits until all previously *
389 * posted buffers get flushed, deallocates all conn resources */
390 iser_conn_terminate(ib_conn);
391 iser_conn->ib_conn = NULL;
392 conn->recv_lock = NULL;
396 static struct iscsi_transport iscsi_iser_transport;
398 static struct iscsi_cls_session *
399 iscsi_iser_session_create(struct iscsi_transport *iscsit,
400 struct scsi_transport_template *scsit,
401 uint32_t initial_cmdsn, uint32_t *hostno)
403 struct iscsi_cls_session *cls_session;
404 struct iscsi_session *session;
407 struct iscsi_cmd_task *ctask;
408 struct iscsi_mgmt_task *mtask;
409 struct iscsi_iser_cmd_task *iser_ctask;
410 struct iser_desc *desc;
412 cls_session = iscsi_session_setup(iscsit, scsit,
413 sizeof(struct iscsi_iser_cmd_task),
414 sizeof(struct iser_desc),
420 session = class_to_transport_session(cls_session);
422 /* libiscsi setup itts, data and pool so just set desc fields */
423 for (i = 0; i < session->cmds_max; i++) {
424 ctask = session->cmds[i];
425 iser_ctask = ctask->dd_data;
426 ctask->hdr = (struct iscsi_cmd *)&iser_ctask->desc.iscsi_header;
429 for (i = 0; i < session->mgmtpool_max; i++) {
430 mtask = session->mgmt_cmds[i];
431 desc = mtask->dd_data;
432 mtask->hdr = &desc->iscsi_header;
433 desc->data = mtask->data;
440 iscsi_iser_conn_set_param(struct iscsi_cls_conn *cls_conn,
441 enum iscsi_param param, uint32_t value)
443 struct iscsi_conn *conn = cls_conn->dd_data;
444 struct iscsi_session *session = conn->session;
446 spin_lock_bh(&session->lock);
447 if (conn->c_stage != ISCSI_CONN_INITIAL_STAGE &&
448 conn->stop_stage != STOP_CONN_RECOVER) {
449 printk(KERN_ERR "iscsi_iser: can not change parameter [%d]\n",
451 spin_unlock_bh(&session->lock);
454 spin_unlock_bh(&session->lock);
457 case ISCSI_PARAM_MAX_RECV_DLENGTH:
460 case ISCSI_PARAM_MAX_XMIT_DLENGTH:
461 conn->max_xmit_dlength = value;
463 case ISCSI_PARAM_HDRDGST_EN:
465 printk(KERN_ERR "DataDigest wasn't negotiated to None");
469 case ISCSI_PARAM_DATADGST_EN:
471 printk(KERN_ERR "DataDigest wasn't negotiated to None");
475 case ISCSI_PARAM_INITIAL_R2T_EN:
476 session->initial_r2t_en = value;
478 case ISCSI_PARAM_IMM_DATA_EN:
479 session->imm_data_en = value;
481 case ISCSI_PARAM_FIRST_BURST:
482 session->first_burst = value;
484 case ISCSI_PARAM_MAX_BURST:
485 session->max_burst = value;
487 case ISCSI_PARAM_PDU_INORDER_EN:
488 session->pdu_inorder_en = value;
490 case ISCSI_PARAM_DATASEQ_INORDER_EN:
491 session->dataseq_inorder_en = value;
493 case ISCSI_PARAM_ERL:
494 session->erl = value;
496 case ISCSI_PARAM_IFMARKER_EN:
498 printk(KERN_ERR "IFMarker wasn't negotiated to No");
502 case ISCSI_PARAM_OFMARKER_EN:
504 printk(KERN_ERR "OFMarker wasn't negotiated to No");
516 iscsi_iser_session_get_param(struct iscsi_cls_session *cls_session,
517 enum iscsi_param param, uint32_t *value)
519 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
520 struct iscsi_session *session = iscsi_hostdata(shost->hostdata);
523 case ISCSI_PARAM_INITIAL_R2T_EN:
524 *value = session->initial_r2t_en;
526 case ISCSI_PARAM_MAX_R2T:
527 *value = session->max_r2t;
529 case ISCSI_PARAM_IMM_DATA_EN:
530 *value = session->imm_data_en;
532 case ISCSI_PARAM_FIRST_BURST:
533 *value = session->first_burst;
535 case ISCSI_PARAM_MAX_BURST:
536 *value = session->max_burst;
538 case ISCSI_PARAM_PDU_INORDER_EN:
539 *value = session->pdu_inorder_en;
541 case ISCSI_PARAM_DATASEQ_INORDER_EN:
542 *value = session->dataseq_inorder_en;
544 case ISCSI_PARAM_ERL:
545 *value = session->erl;
547 case ISCSI_PARAM_IFMARKER_EN:
550 case ISCSI_PARAM_OFMARKER_EN:
554 return ISCSI_ERR_PARAM_NOT_FOUND;
561 iscsi_iser_conn_get_param(struct iscsi_cls_conn *cls_conn,
562 enum iscsi_param param, uint32_t *value)
564 struct iscsi_conn *conn = cls_conn->dd_data;
567 case ISCSI_PARAM_MAX_RECV_DLENGTH:
568 *value = conn->max_recv_dlength;
570 case ISCSI_PARAM_MAX_XMIT_DLENGTH:
571 *value = conn->max_xmit_dlength;
573 case ISCSI_PARAM_HDRDGST_EN:
576 case ISCSI_PARAM_DATADGST_EN:
579 /*case ISCSI_PARAM_TARGET_RECV_DLENGTH:
580 *value = conn->target_recv_dlength;
582 case ISCSI_PARAM_INITIATOR_RECV_DLENGTH:
583 *value = conn->initiator_recv_dlength;
586 return ISCSI_ERR_PARAM_NOT_FOUND;
594 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
596 struct iscsi_conn *conn = cls_conn->dd_data;
598 stats->txdata_octets = conn->txdata_octets;
599 stats->rxdata_octets = conn->rxdata_octets;
600 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
601 stats->dataout_pdus = conn->dataout_pdus_cnt;
602 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
603 stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
604 stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
605 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
606 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
607 stats->custom_length = 3;
608 strcpy(stats->custom[0].desc, "qp_tx_queue_full");
609 stats->custom[0].value = 0; /* TB iser_conn->qp_tx_queue_full; */
610 strcpy(stats->custom[1].desc, "fmr_map_not_avail");
611 stats->custom[1].value = 0; /* TB iser_conn->fmr_map_not_avail */;
612 strcpy(stats->custom[2].desc, "eh_abort_cnt");
613 stats->custom[2].value = conn->eh_abort_cnt;
617 iscsi_iser_ep_connect(struct sockaddr *dst_addr, int non_blocking,
621 struct iser_conn *ib_conn;
623 err = iser_conn_init(&ib_conn);
627 err = iser_connect(ib_conn, NULL, (struct sockaddr_in *)dst_addr, non_blocking);
629 *ep_handle = (__u64)(unsigned long)ib_conn;
636 iscsi_iser_ep_poll(__u64 ep_handle, int timeout_ms)
638 struct iser_conn *ib_conn = iscsi_iser_ib_conn_lookup(ep_handle);
644 rc = wait_event_interruptible_timeout(ib_conn->wait,
645 ib_conn->state == ISER_CONN_UP,
646 msecs_to_jiffies(timeout_ms));
648 /* if conn establishment failed, return error code to iscsi */
650 (ib_conn->state == ISER_CONN_TERMINATING ||
651 ib_conn->state == ISER_CONN_DOWN))
654 iser_err("ib conn %p rc = %d\n", ib_conn, rc);
657 return 1; /* success, this is the equivalent of POLLOUT */
659 return 0; /* timeout */
661 return rc; /* signal */
665 iscsi_iser_ep_disconnect(__u64 ep_handle)
667 struct iser_conn *ib_conn = iscsi_iser_ib_conn_lookup(ep_handle);
672 iser_err("ib conn %p state %d\n",ib_conn, ib_conn->state);
674 iser_conn_terminate(ib_conn);
677 static struct scsi_host_template iscsi_iser_sht = {
678 .name = "iSCSI Initiator over iSER, v." DRV_VER,
679 .queuecommand = iscsi_queuecommand,
680 .can_queue = ISCSI_XMIT_CMDS_MAX - 1,
681 .sg_tablesize = ISCSI_ISER_SG_TABLESIZE,
682 .cmd_per_lun = ISCSI_MAX_CMD_PER_LUN,
683 .eh_abort_handler = iscsi_eh_abort,
684 .eh_host_reset_handler = iscsi_eh_host_reset,
685 .use_clustering = DISABLE_CLUSTERING,
686 .proc_name = "iscsi_iser",
690 static struct iscsi_transport iscsi_iser_transport = {
691 .owner = THIS_MODULE,
693 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T,
694 .param_mask = ISCSI_MAX_RECV_DLENGTH |
695 ISCSI_MAX_XMIT_DLENGTH |
698 ISCSI_INITIAL_R2T_EN |
703 ISCSI_PDU_INORDER_EN |
704 ISCSI_DATASEQ_INORDER_EN,
705 .host_template = &iscsi_iser_sht,
706 .conndata_size = sizeof(struct iscsi_conn),
707 .max_lun = ISCSI_ISER_MAX_LUN,
708 .max_cmd_len = ISCSI_ISER_MAX_CMD_LEN,
709 /* session management */
710 .create_session = iscsi_iser_session_create,
711 .destroy_session = iscsi_session_teardown,
712 /* connection management */
713 .create_conn = iscsi_iser_conn_create,
714 .bind_conn = iscsi_iser_conn_bind,
715 .destroy_conn = iscsi_iser_conn_destroy,
716 .set_param = iscsi_iser_conn_set_param,
717 .get_conn_param = iscsi_iser_conn_get_param,
718 .get_session_param = iscsi_iser_session_get_param,
719 .start_conn = iscsi_iser_conn_start,
720 .stop_conn = iscsi_conn_stop,
721 /* these are called as part of conn recovery */
722 .suspend_conn_recv = NULL, /* FIXME is/how this relvant to iser? */
723 .terminate_conn = iscsi_iser_conn_terminate,
725 .send_pdu = iscsi_conn_send_pdu,
726 .get_stats = iscsi_iser_conn_get_stats,
727 .init_cmd_task = iscsi_iser_cmd_init,
728 .xmit_cmd_task = iscsi_iser_ctask_xmit,
729 .xmit_mgmt_task = iscsi_iser_mtask_xmit,
730 .cleanup_cmd_task = iscsi_iser_cleanup_ctask,
732 .session_recovery_timedout = iscsi_session_recovery_timedout,
734 .ep_connect = iscsi_iser_ep_connect,
735 .ep_poll = iscsi_iser_ep_poll,
736 .ep_disconnect = iscsi_iser_ep_disconnect
739 static int __init iser_init(void)
743 iser_dbg("Starting iSER datamover...\n");
745 if (iscsi_max_lun < 1) {
746 printk(KERN_ERR "Invalid max_lun value of %u\n", iscsi_max_lun);
750 iscsi_iser_transport.max_lun = iscsi_max_lun;
752 memset(&ig, 0, sizeof(struct iser_global));
754 ig.desc_cache = kmem_cache_create("iser_descriptors",
755 sizeof (struct iser_desc),
756 0, SLAB_HWCACHE_ALIGN,
758 if (ig.desc_cache == NULL)
761 /* device init is called only after the first addr resolution */
762 mutex_init(&ig.device_list_mutex);
763 INIT_LIST_HEAD(&ig.device_list);
764 mutex_init(&ig.connlist_mutex);
765 INIT_LIST_HEAD(&ig.connlist);
767 if (!iscsi_register_transport(&iscsi_iser_transport)) {
768 iser_err("iscsi_register_transport failed\n");
770 goto register_transport_failure;
775 register_transport_failure:
776 kmem_cache_destroy(ig.desc_cache);
781 static void __exit iser_exit(void)
783 iser_dbg("Removing iSER datamover...\n");
784 iscsi_unregister_transport(&iscsi_iser_transport);
785 kmem_cache_destroy(ig.desc_cache);
788 module_init(iser_init);
789 module_exit(iser_exit);