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 #include <linux/types.h>
48 #include <linux/list.h>
49 #include <linux/hardirq.h>
50 #include <linux/kfifo.h>
51 #include <linux/blkdev.h>
52 #include <linux/init.h>
53 #include <linux/ioctl.h>
54 #include <linux/cdev.h>
56 #include <linux/net.h>
57 #include <linux/scatterlist.h>
58 #include <linux/delay.h>
62 #include <asm/uaccess.h>
64 #include <scsi/scsi_cmnd.h>
65 #include <scsi/scsi_device.h>
66 #include <scsi/scsi_eh.h>
67 #include <scsi/scsi_tcq.h>
68 #include <scsi/scsi_host.h>
69 #include <scsi/scsi.h>
70 #include <scsi/scsi_transport_iscsi.h>
72 #include "iscsi_iser.h"
74 static struct scsi_host_template iscsi_iser_sht;
75 static struct iscsi_transport iscsi_iser_transport;
76 static struct scsi_transport_template *iscsi_iser_scsi_transport;
78 static unsigned int iscsi_max_lun = 512;
79 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
81 int iser_debug_level = 0;
83 MODULE_DESCRIPTION("iSER (iSCSI Extensions for RDMA) Datamover "
84 "v" DRV_VER " (" DRV_DATE ")");
85 MODULE_LICENSE("Dual BSD/GPL");
86 MODULE_AUTHOR("Alex Nezhinsky, Dan Bar Dov, Or Gerlitz");
88 module_param_named(debug_level, iser_debug_level, int, 0644);
89 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:disabled)");
91 struct iser_global ig;
94 iscsi_iser_recv(struct iscsi_conn *conn,
95 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 rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len);
114 if (rc && rc != ISCSI_ERR_NO_SCSI_CMD)
119 iscsi_conn_failure(conn, rc);
122 static int iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
124 struct iscsi_iser_task *iser_task = task->dd_data;
126 task->hdr = (struct iscsi_hdr *)&iser_task->desc.iscsi_header;
127 task->hdr_max = sizeof(iser_task->desc.iscsi_header);
131 int iser_initialize_task_headers(struct iscsi_task *task,
132 struct iser_tx_desc *tx_desc)
134 struct iscsi_iser_conn *iser_conn = task->conn->dd_data;
135 struct iser_device *device = iser_conn->ib_conn->device;
136 struct iscsi_iser_task *iser_task = task->dd_data;
139 dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc,
140 ISER_HEADERS_LEN, DMA_TO_DEVICE);
141 if (ib_dma_mapping_error(device->ib_device, dma_addr))
144 tx_desc->dma_addr = dma_addr;
145 tx_desc->tx_sg[0].addr = tx_desc->dma_addr;
146 tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
147 tx_desc->tx_sg[0].lkey = device->mr->lkey;
149 iser_task->headers_initialized = 1;
150 iser_task->iser_conn = iser_conn;
154 * iscsi_iser_task_init - Initialize task
157 * Initialize the task for the scsi command or mgmt command.
160 iscsi_iser_task_init(struct iscsi_task *task)
162 struct iscsi_iser_task *iser_task = task->dd_data;
164 if (!iser_task->headers_initialized)
165 if (iser_initialize_task_headers(task, &iser_task->desc))
172 iser_task->command_sent = 0;
173 iser_task_rdma_init(iser_task);
178 * iscsi_iser_mtask_xmit - xmit management(immediate) task
179 * @conn: iscsi connection
180 * @task: task management task
183 * The function can return -EAGAIN in which case caller must
184 * call it again later, or recover. '0' return code means successful
189 iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
193 iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt);
195 error = iser_send_control(conn, task);
197 /* since iser xmits control with zero copy, tasks can not be recycled
198 * right after sending them.
199 * The recycling scheme is based on whether a response is expected
200 * - if yes, the task is recycled at iscsi_complete_pdu
201 * - if no, the task is recycled at iser_snd_completion
207 iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn,
208 struct iscsi_task *task)
210 struct iscsi_r2t_info *r2t = &task->unsol_r2t;
211 struct iscsi_data hdr;
214 /* Send data-out PDUs while there's still unsolicited data to send */
215 while (iscsi_task_has_unsol_data(task)) {
216 iscsi_prep_data_out_pdu(task, r2t, &hdr);
217 iser_dbg("Sending data-out: itt 0x%x, data count %d\n",
218 hdr.itt, r2t->data_count);
220 /* the buffer description has been passed with the command */
221 /* Send the command */
222 error = iser_send_data_out(conn, task, &hdr);
225 goto iscsi_iser_task_xmit_unsol_data_exit;
227 r2t->sent += r2t->data_count;
228 iser_dbg("Need to send %d more as data-out PDUs\n",
229 r2t->data_length - r2t->sent);
232 iscsi_iser_task_xmit_unsol_data_exit:
237 iscsi_iser_task_xmit(struct iscsi_task *task)
239 struct iscsi_conn *conn = task->conn;
240 struct iscsi_iser_task *iser_task = task->dd_data;
244 return iscsi_iser_mtask_xmit(conn, task);
246 if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
247 BUG_ON(scsi_bufflen(task->sc) == 0);
249 iser_dbg("cmd [itt %x total %d imm %d unsol_data %d\n",
250 task->itt, scsi_bufflen(task->sc),
251 task->imm_count, task->unsol_r2t.data_length);
254 iser_dbg("ctask xmit [cid %d itt 0x%x]\n",
255 conn->id, task->itt);
257 /* Send the cmd PDU */
258 if (!iser_task->command_sent) {
259 error = iser_send_command(conn, task);
261 goto iscsi_iser_task_xmit_exit;
262 iser_task->command_sent = 1;
265 /* Send unsolicited data-out PDU(s) if necessary */
266 if (iscsi_task_has_unsol_data(task))
267 error = iscsi_iser_task_xmit_unsol_data(conn, task);
269 iscsi_iser_task_xmit_exit:
273 static void iscsi_iser_cleanup_task(struct iscsi_task *task)
275 struct iscsi_iser_task *iser_task = task->dd_data;
277 /* mgmt tasks do not need special cleanup */
281 if (iser_task->status == ISER_TASK_STATUS_STARTED) {
282 iser_task->status = ISER_TASK_STATUS_COMPLETED;
283 iser_task_rdma_finalize(iser_task);
287 static struct iscsi_cls_conn *
288 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
290 struct iscsi_conn *conn;
291 struct iscsi_cls_conn *cls_conn;
292 struct iscsi_iser_conn *iser_conn;
294 cls_conn = iscsi_conn_setup(cls_session, sizeof(*iser_conn), conn_idx);
297 conn = cls_conn->dd_data;
300 * due to issues with the login code re iser sematics
301 * this not set in iscsi_conn_setup - FIXME
303 conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN;
305 iser_conn = conn->dd_data;
306 conn->dd_data = iser_conn;
307 iser_conn->iscsi_conn = conn;
313 iscsi_iser_conn_destroy(struct iscsi_cls_conn *cls_conn)
315 struct iscsi_conn *conn = cls_conn->dd_data;
316 struct iscsi_iser_conn *iser_conn = conn->dd_data;
317 struct iser_conn *ib_conn = iser_conn->ib_conn;
319 iscsi_conn_teardown(cls_conn);
321 * Userspace will normally call the stop callback and
322 * already have freed the ib_conn, but if it goofed up then
326 ib_conn->iser_conn = NULL;
327 iser_conn_put(ib_conn);
332 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
333 struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
336 struct iscsi_conn *conn = cls_conn->dd_data;
337 struct iscsi_iser_conn *iser_conn;
338 struct iser_conn *ib_conn;
339 struct iscsi_endpoint *ep;
342 error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
346 /* the transport ep handle comes from user space so it must be
347 * verified against the global ib connections list */
348 ep = iscsi_lookup_endpoint(transport_eph);
350 iser_err("can't bind eph %llx\n",
351 (unsigned long long)transport_eph);
354 ib_conn = ep->dd_data;
356 /* binds the iSER connection retrieved from the previously
357 * connected ep_handle to the iSCSI layer connection. exchanges
358 * connection pointers */
359 iser_err("binding iscsi conn %p to iser_conn %p\n",conn,ib_conn);
360 iser_conn = conn->dd_data;
361 ib_conn->iser_conn = iser_conn;
362 iser_conn->ib_conn = ib_conn;
363 iser_conn_get(ib_conn);
368 iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
370 struct iscsi_conn *conn = cls_conn->dd_data;
371 struct iscsi_iser_conn *iser_conn = conn->dd_data;
372 struct iser_conn *ib_conn = iser_conn->ib_conn;
375 * Userspace may have goofed up and not bound the connection or
376 * might have only partially setup the connection.
379 iscsi_conn_stop(cls_conn, flag);
381 * There is no unbind event so the stop callback
382 * must release the ref from the bind.
384 iser_conn_put(ib_conn);
386 iser_conn->ib_conn = NULL;
390 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
392 struct iscsi_conn *conn = cls_conn->dd_data;
395 err = iser_conn_set_full_featured_mode(conn);
399 return iscsi_conn_start(cls_conn);
402 static void iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session)
404 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
406 iscsi_session_teardown(cls_session);
407 iscsi_host_remove(shost);
408 iscsi_host_free(shost);
411 static struct iscsi_cls_session *
412 iscsi_iser_session_create(struct iscsi_endpoint *ep,
413 uint16_t cmds_max, uint16_t qdepth,
414 uint32_t initial_cmdsn)
416 struct iscsi_cls_session *cls_session;
417 struct iscsi_session *session;
418 struct Scsi_Host *shost;
419 struct iser_conn *ib_conn;
421 shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0);
424 shost->transportt = iscsi_iser_scsi_transport;
425 shost->max_lun = iscsi_max_lun;
427 shost->max_channel = 0;
428 shost->max_cmd_len = 16;
431 * older userspace tools (before 2.0-870) did not pass us
432 * the leading conn's ep so this will be NULL;
435 ib_conn = ep->dd_data;
437 if (iscsi_host_add(shost,
438 ep ? ib_conn->device->ib_device->dma_device : NULL))
442 * we do not support setting can_queue cmd_per_lun from userspace yet
443 * because we preallocate so many resources
445 cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
446 ISCSI_DEF_XMIT_CMDS_MAX, 0,
447 sizeof(struct iscsi_iser_task),
451 session = cls_session->dd_data;
453 shost->can_queue = session->scsi_cmds_max;
457 iscsi_host_remove(shost);
459 iscsi_host_free(shost);
464 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
465 enum iscsi_param param, char *buf, int buflen)
470 case ISCSI_PARAM_MAX_RECV_DLENGTH:
473 case ISCSI_PARAM_HDRDGST_EN:
474 sscanf(buf, "%d", &value);
476 printk(KERN_ERR "DataDigest wasn't negotiated to None");
480 case ISCSI_PARAM_DATADGST_EN:
481 sscanf(buf, "%d", &value);
483 printk(KERN_ERR "DataDigest wasn't negotiated to None");
487 case ISCSI_PARAM_IFMARKER_EN:
488 sscanf(buf, "%d", &value);
490 printk(KERN_ERR "IFMarker wasn't negotiated to No");
494 case ISCSI_PARAM_OFMARKER_EN:
495 sscanf(buf, "%d", &value);
497 printk(KERN_ERR "OFMarker wasn't negotiated to No");
502 return iscsi_set_param(cls_conn, param, buf, buflen);
509 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
511 struct iscsi_conn *conn = cls_conn->dd_data;
513 stats->txdata_octets = conn->txdata_octets;
514 stats->rxdata_octets = conn->rxdata_octets;
515 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
516 stats->dataout_pdus = conn->dataout_pdus_cnt;
517 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
518 stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
519 stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
520 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
521 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
522 stats->custom_length = 4;
523 strcpy(stats->custom[0].desc, "qp_tx_queue_full");
524 stats->custom[0].value = 0; /* TB iser_conn->qp_tx_queue_full; */
525 strcpy(stats->custom[1].desc, "fmr_map_not_avail");
526 stats->custom[1].value = 0; /* TB iser_conn->fmr_map_not_avail */;
527 strcpy(stats->custom[2].desc, "eh_abort_cnt");
528 stats->custom[2].value = conn->eh_abort_cnt;
529 strcpy(stats->custom[3].desc, "fmr_unalign_cnt");
530 stats->custom[3].value = conn->fmr_unalign_cnt;
533 static struct iscsi_endpoint *
534 iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr,
538 struct iser_conn *ib_conn;
539 struct iscsi_endpoint *ep;
541 ep = iscsi_create_endpoint(sizeof(*ib_conn));
543 return ERR_PTR(-ENOMEM);
545 ib_conn = ep->dd_data;
547 iser_conn_init(ib_conn);
549 err = iser_connect(ib_conn, NULL, (struct sockaddr_in *)dst_addr,
552 iscsi_destroy_endpoint(ep);
559 iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
561 struct iser_conn *ib_conn;
564 ib_conn = ep->dd_data;
565 rc = wait_event_interruptible_timeout(ib_conn->wait,
566 ib_conn->state == ISER_CONN_UP,
567 msecs_to_jiffies(timeout_ms));
569 /* if conn establishment failed, return error code to iscsi */
571 (ib_conn->state == ISER_CONN_TERMINATING ||
572 ib_conn->state == ISER_CONN_DOWN))
575 iser_err("ib conn %p rc = %d\n", ib_conn, rc);
578 return 1; /* success, this is the equivalent of POLLOUT */
580 return 0; /* timeout */
582 return rc; /* signal */
586 iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep)
588 struct iser_conn *ib_conn;
590 ib_conn = ep->dd_data;
591 if (ib_conn->iser_conn)
593 * Must suspend xmit path if the ep is bound to the
594 * iscsi_conn, so we know we are not accessing the ib_conn
597 * This may not be bound if the ep poll failed.
599 iscsi_suspend_tx(ib_conn->iser_conn->iscsi_conn);
602 iser_err("ib conn %p state %d\n",ib_conn, ib_conn->state);
603 iser_conn_terminate(ib_conn);
606 static struct scsi_host_template iscsi_iser_sht = {
607 .module = THIS_MODULE,
608 .name = "iSCSI Initiator over iSER, v." DRV_VER,
609 .queuecommand = iscsi_queuecommand,
610 .change_queue_depth = iscsi_change_queue_depth,
611 .sg_tablesize = ISCSI_ISER_SG_TABLESIZE,
613 .cmd_per_lun = ISER_DEF_CMD_PER_LUN,
614 .eh_abort_handler = iscsi_eh_abort,
615 .eh_device_reset_handler= iscsi_eh_device_reset,
616 .eh_target_reset_handler= iscsi_eh_target_reset,
617 .target_alloc = iscsi_target_alloc,
618 .use_clustering = DISABLE_CLUSTERING,
619 .proc_name = "iscsi_iser",
623 static struct iscsi_transport iscsi_iser_transport = {
624 .owner = THIS_MODULE,
626 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T,
627 .param_mask = ISCSI_MAX_RECV_DLENGTH |
628 ISCSI_MAX_XMIT_DLENGTH |
631 ISCSI_INITIAL_R2T_EN |
636 ISCSI_PDU_INORDER_EN |
637 ISCSI_DATASEQ_INORDER_EN |
639 ISCSI_PERSISTENT_PORT |
640 ISCSI_PERSISTENT_ADDRESS |
641 ISCSI_TARGET_NAME | ISCSI_TPGT |
642 ISCSI_USERNAME | ISCSI_PASSWORD |
643 ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
644 ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
645 ISCSI_LU_RESET_TMO | ISCSI_TGT_RESET_TMO |
646 ISCSI_PING_TMO | ISCSI_RECV_TMO |
647 ISCSI_IFACE_NAME | ISCSI_INITIATOR_NAME,
648 .host_param_mask = ISCSI_HOST_HWADDRESS |
649 ISCSI_HOST_NETDEV_NAME |
650 ISCSI_HOST_INITIATOR_NAME,
651 /* session management */
652 .create_session = iscsi_iser_session_create,
653 .destroy_session = iscsi_iser_session_destroy,
654 /* connection management */
655 .create_conn = iscsi_iser_conn_create,
656 .bind_conn = iscsi_iser_conn_bind,
657 .destroy_conn = iscsi_iser_conn_destroy,
658 .set_param = iscsi_iser_set_param,
659 .get_conn_param = iscsi_conn_get_param,
660 .get_session_param = iscsi_session_get_param,
661 .start_conn = iscsi_iser_conn_start,
662 .stop_conn = iscsi_iser_conn_stop,
663 /* iscsi host params */
664 .get_host_param = iscsi_host_get_param,
665 .set_host_param = iscsi_host_set_param,
667 .send_pdu = iscsi_conn_send_pdu,
668 .get_stats = iscsi_iser_conn_get_stats,
669 .init_task = iscsi_iser_task_init,
670 .xmit_task = iscsi_iser_task_xmit,
671 .cleanup_task = iscsi_iser_cleanup_task,
672 .alloc_pdu = iscsi_iser_pdu_alloc,
674 .session_recovery_timedout = iscsi_session_recovery_timedout,
676 .ep_connect = iscsi_iser_ep_connect,
677 .ep_poll = iscsi_iser_ep_poll,
678 .ep_disconnect = iscsi_iser_ep_disconnect
681 static int __init iser_init(void)
685 iser_dbg("Starting iSER datamover...\n");
687 if (iscsi_max_lun < 1) {
688 printk(KERN_ERR "Invalid max_lun value of %u\n", iscsi_max_lun);
692 memset(&ig, 0, sizeof(struct iser_global));
694 ig.desc_cache = kmem_cache_create("iser_descriptors",
695 sizeof(struct iser_tx_desc),
696 0, SLAB_HWCACHE_ALIGN,
698 if (ig.desc_cache == NULL)
701 /* device init is called only after the first addr resolution */
702 mutex_init(&ig.device_list_mutex);
703 INIT_LIST_HEAD(&ig.device_list);
704 mutex_init(&ig.connlist_mutex);
705 INIT_LIST_HEAD(&ig.connlist);
707 iscsi_iser_scsi_transport = iscsi_register_transport(
708 &iscsi_iser_transport);
709 if (!iscsi_iser_scsi_transport) {
710 iser_err("iscsi_register_transport failed\n");
712 goto register_transport_failure;
717 register_transport_failure:
718 kmem_cache_destroy(ig.desc_cache);
723 static void __exit iser_exit(void)
725 iser_dbg("Removing iSER datamover...\n");
726 iscsi_unregister_transport(&iscsi_iser_transport);
727 kmem_cache_destroy(ig.desc_cache);
730 module_init(iser_init);
731 module_exit(iser_exit);