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);
124 * iscsi_iser_task_init - Initialize task
127 * Initialize the task for the scsi command or mgmt command.
130 iscsi_iser_task_init(struct iscsi_task *task)
132 struct iscsi_iser_conn *iser_conn = task->conn->dd_data;
133 struct iscsi_iser_task *iser_task = task->dd_data;
137 iser_task->desc.data = task->data;
141 iser_task->command_sent = 0;
142 iser_task->iser_conn = iser_conn;
143 iser_task_rdma_init(iser_task);
148 * iscsi_iser_mtask_xmit - xmit management(immediate) task
149 * @conn: iscsi connection
150 * @task: task management task
153 * The function can return -EAGAIN in which case caller must
154 * call it again later, or recover. '0' return code means successful
159 iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
163 debug_scsi("task deq [cid %d itt 0x%x]\n", conn->id, task->itt);
165 error = iser_send_control(conn, task);
167 /* since iser xmits control with zero copy, tasks can not be recycled
168 * right after sending them.
169 * The recycling scheme is based on whether a response is expected
170 * - if yes, the task is recycled at iscsi_complete_pdu
171 * - if no, the task is recycled at iser_snd_completion
173 if (error && error != -ENOBUFS)
174 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
180 iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn,
181 struct iscsi_task *task)
183 struct iscsi_data hdr;
186 /* Send data-out PDUs while there's still unsolicited data to send */
187 while (task->unsol_count > 0) {
188 iscsi_prep_unsolicit_data_pdu(task, &hdr);
189 debug_scsi("Sending data-out: itt 0x%x, data count %d\n",
190 hdr.itt, task->data_count);
192 /* the buffer description has been passed with the command */
193 /* Send the command */
194 error = iser_send_data_out(conn, task, &hdr);
196 task->unsol_datasn--;
197 goto iscsi_iser_task_xmit_unsol_data_exit;
199 task->unsol_count -= task->data_count;
200 debug_scsi("Need to send %d more as data-out PDUs\n",
204 iscsi_iser_task_xmit_unsol_data_exit:
209 iscsi_iser_task_xmit(struct iscsi_task *task)
211 struct iscsi_conn *conn = task->conn;
212 struct iscsi_iser_task *iser_task = task->dd_data;
216 return iscsi_iser_mtask_xmit(conn, task);
218 if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
219 BUG_ON(scsi_bufflen(task->sc) == 0);
221 debug_scsi("cmd [itt %x total %d imm %d unsol_data %d\n",
222 task->itt, scsi_bufflen(task->sc),
223 task->imm_count, task->unsol_count);
226 debug_scsi("task deq [cid %d itt 0x%x]\n",
227 conn->id, task->itt);
229 /* Send the cmd PDU */
230 if (!iser_task->command_sent) {
231 error = iser_send_command(conn, task);
233 goto iscsi_iser_task_xmit_exit;
234 iser_task->command_sent = 1;
237 /* Send unsolicited data-out PDU(s) if necessary */
238 if (task->unsol_count)
239 error = iscsi_iser_task_xmit_unsol_data(conn, task);
241 iscsi_iser_task_xmit_exit:
242 if (error && error != -ENOBUFS)
243 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
248 iscsi_iser_cleanup_task(struct iscsi_conn *conn, struct iscsi_task *task)
250 struct iscsi_iser_task *iser_task = task->dd_data;
252 /* mgmt tasks do not need special cleanup */
256 if (iser_task->status == ISER_TASK_STATUS_STARTED) {
257 iser_task->status = ISER_TASK_STATUS_COMPLETED;
258 iser_task_rdma_finalize(iser_task);
262 static struct iscsi_cls_conn *
263 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
265 struct iscsi_conn *conn;
266 struct iscsi_cls_conn *cls_conn;
267 struct iscsi_iser_conn *iser_conn;
269 cls_conn = iscsi_conn_setup(cls_session, sizeof(*iser_conn), conn_idx);
272 conn = cls_conn->dd_data;
275 * due to issues with the login code re iser sematics
276 * this not set in iscsi_conn_setup - FIXME
278 conn->max_recv_dlength = 128;
280 iser_conn = conn->dd_data;
281 conn->dd_data = iser_conn;
282 iser_conn->iscsi_conn = conn;
288 iscsi_iser_conn_destroy(struct iscsi_cls_conn *cls_conn)
290 struct iscsi_conn *conn = cls_conn->dd_data;
291 struct iscsi_iser_conn *iser_conn = conn->dd_data;
292 struct iser_conn *ib_conn = iser_conn->ib_conn;
294 iscsi_conn_teardown(cls_conn);
296 * Userspace will normally call the stop callback and
297 * already have freed the ib_conn, but if it goofed up then
301 ib_conn->iser_conn = NULL;
302 iser_conn_put(ib_conn);
307 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
308 struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
311 struct iscsi_conn *conn = cls_conn->dd_data;
312 struct iscsi_iser_conn *iser_conn;
313 struct iser_conn *ib_conn;
314 struct iscsi_endpoint *ep;
317 error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
321 /* the transport ep handle comes from user space so it must be
322 * verified against the global ib connections list */
323 ep = iscsi_lookup_endpoint(transport_eph);
325 iser_err("can't bind eph %llx\n",
326 (unsigned long long)transport_eph);
329 ib_conn = ep->dd_data;
331 /* binds the iSER connection retrieved from the previously
332 * connected ep_handle to the iSCSI layer connection. exchanges
333 * connection pointers */
334 iser_err("binding iscsi conn %p to iser_conn %p\n",conn,ib_conn);
335 iser_conn = conn->dd_data;
336 ib_conn->iser_conn = iser_conn;
337 iser_conn->ib_conn = ib_conn;
338 iser_conn_get(ib_conn);
343 iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
345 struct iscsi_conn *conn = cls_conn->dd_data;
346 struct iscsi_iser_conn *iser_conn = conn->dd_data;
347 struct iser_conn *ib_conn = iser_conn->ib_conn;
350 * Userspace may have goofed up and not bound the connection or
351 * might have only partially setup the connection.
354 iscsi_conn_stop(cls_conn, flag);
356 * There is no unbind event so the stop callback
357 * must release the ref from the bind.
359 iser_conn_put(ib_conn);
361 iser_conn->ib_conn = NULL;
365 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
367 struct iscsi_conn *conn = cls_conn->dd_data;
370 err = iser_conn_set_full_featured_mode(conn);
374 return iscsi_conn_start(cls_conn);
377 static void iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session)
379 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
381 iscsi_session_teardown(cls_session);
382 iscsi_host_remove(shost);
383 iscsi_host_free(shost);
386 static struct iscsi_cls_session *
387 iscsi_iser_session_create(struct iscsi_endpoint *ep,
388 uint16_t cmds_max, uint16_t qdepth,
389 uint32_t initial_cmdsn, uint32_t *hostno)
391 struct iscsi_cls_session *cls_session;
392 struct iscsi_session *session;
393 struct Scsi_Host *shost;
395 struct iscsi_task *task;
396 struct iscsi_iser_task *iser_task;
397 struct iser_conn *ib_conn;
399 shost = iscsi_host_alloc(&iscsi_iser_sht, 0, ISCSI_MAX_CMD_PER_LUN);
402 shost->transportt = iscsi_iser_scsi_transport;
403 shost->max_lun = iscsi_max_lun;
405 shost->max_channel = 0;
406 shost->max_cmd_len = 16;
409 * older userspace tools (before 2.0-870) did not pass us
410 * the leading conn's ep so this will be NULL;
413 ib_conn = ep->dd_data;
415 if (iscsi_host_add(shost,
416 ep ? ib_conn->device->ib_device->dma_device : NULL))
418 *hostno = shost->host_no;
421 * we do not support setting can_queue cmd_per_lun from userspace yet
422 * because we preallocate so many resources
424 cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
425 ISCSI_DEF_XMIT_CMDS_MAX,
426 sizeof(struct iscsi_iser_task),
430 session = cls_session->dd_data;
432 shost->can_queue = session->scsi_cmds_max;
433 /* libiscsi setup itts, data and pool so just set desc fields */
434 for (i = 0; i < session->cmds_max; i++) {
435 task = session->cmds[i];
436 iser_task = task->dd_data;
437 task->hdr = (struct iscsi_cmd *)&iser_task->desc.iscsi_header;
438 task->hdr_max = sizeof(iser_task->desc.iscsi_header);
443 iscsi_host_remove(shost);
445 iscsi_host_free(shost);
450 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
451 enum iscsi_param param, char *buf, int buflen)
456 case ISCSI_PARAM_MAX_RECV_DLENGTH:
459 case ISCSI_PARAM_HDRDGST_EN:
460 sscanf(buf, "%d", &value);
462 printk(KERN_ERR "DataDigest wasn't negotiated to None");
466 case ISCSI_PARAM_DATADGST_EN:
467 sscanf(buf, "%d", &value);
469 printk(KERN_ERR "DataDigest wasn't negotiated to None");
473 case ISCSI_PARAM_IFMARKER_EN:
474 sscanf(buf, "%d", &value);
476 printk(KERN_ERR "IFMarker wasn't negotiated to No");
480 case ISCSI_PARAM_OFMARKER_EN:
481 sscanf(buf, "%d", &value);
483 printk(KERN_ERR "OFMarker wasn't negotiated to No");
488 return iscsi_set_param(cls_conn, param, buf, buflen);
495 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
497 struct iscsi_conn *conn = cls_conn->dd_data;
499 stats->txdata_octets = conn->txdata_octets;
500 stats->rxdata_octets = conn->rxdata_octets;
501 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
502 stats->dataout_pdus = conn->dataout_pdus_cnt;
503 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
504 stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
505 stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
506 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
507 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
508 stats->custom_length = 4;
509 strcpy(stats->custom[0].desc, "qp_tx_queue_full");
510 stats->custom[0].value = 0; /* TB iser_conn->qp_tx_queue_full; */
511 strcpy(stats->custom[1].desc, "fmr_map_not_avail");
512 stats->custom[1].value = 0; /* TB iser_conn->fmr_map_not_avail */;
513 strcpy(stats->custom[2].desc, "eh_abort_cnt");
514 stats->custom[2].value = conn->eh_abort_cnt;
515 strcpy(stats->custom[3].desc, "fmr_unalign_cnt");
516 stats->custom[3].value = conn->fmr_unalign_cnt;
519 static struct iscsi_endpoint *
520 iscsi_iser_ep_connect(struct sockaddr *dst_addr, int non_blocking)
523 struct iser_conn *ib_conn;
524 struct iscsi_endpoint *ep;
526 ep = iscsi_create_endpoint(sizeof(*ib_conn));
528 return ERR_PTR(-ENOMEM);
530 ib_conn = ep->dd_data;
532 iser_conn_init(ib_conn);
534 err = iser_connect(ib_conn, NULL, (struct sockaddr_in *)dst_addr,
537 iscsi_destroy_endpoint(ep);
544 iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
546 struct iser_conn *ib_conn;
549 ib_conn = ep->dd_data;
550 rc = wait_event_interruptible_timeout(ib_conn->wait,
551 ib_conn->state == ISER_CONN_UP,
552 msecs_to_jiffies(timeout_ms));
554 /* if conn establishment failed, return error code to iscsi */
556 (ib_conn->state == ISER_CONN_TERMINATING ||
557 ib_conn->state == ISER_CONN_DOWN))
560 iser_err("ib conn %p rc = %d\n", ib_conn, rc);
563 return 1; /* success, this is the equivalent of POLLOUT */
565 return 0; /* timeout */
567 return rc; /* signal */
571 iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep)
573 struct iser_conn *ib_conn;
575 ib_conn = ep->dd_data;
576 if (ib_conn->iser_conn)
578 * Must suspend xmit path if the ep is bound to the
579 * iscsi_conn, so we know we are not accessing the ib_conn
582 * This may not be bound if the ep poll failed.
584 iscsi_suspend_tx(ib_conn->iser_conn->iscsi_conn);
587 iser_err("ib conn %p state %d\n",ib_conn, ib_conn->state);
588 iser_conn_terminate(ib_conn);
591 static struct scsi_host_template iscsi_iser_sht = {
592 .module = THIS_MODULE,
593 .name = "iSCSI Initiator over iSER, v." DRV_VER,
594 .queuecommand = iscsi_queuecommand,
595 .change_queue_depth = iscsi_change_queue_depth,
596 .sg_tablesize = ISCSI_ISER_SG_TABLESIZE,
598 .cmd_per_lun = ISCSI_MAX_CMD_PER_LUN,
599 .eh_abort_handler = iscsi_eh_abort,
600 .eh_device_reset_handler= iscsi_eh_device_reset,
601 .eh_target_reset_handler= iscsi_eh_target_reset,
602 .use_clustering = DISABLE_CLUSTERING,
603 .proc_name = "iscsi_iser",
607 static struct iscsi_transport iscsi_iser_transport = {
608 .owner = THIS_MODULE,
610 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T,
611 .param_mask = ISCSI_MAX_RECV_DLENGTH |
612 ISCSI_MAX_XMIT_DLENGTH |
615 ISCSI_INITIAL_R2T_EN |
620 ISCSI_PDU_INORDER_EN |
621 ISCSI_DATASEQ_INORDER_EN |
623 ISCSI_PERSISTENT_PORT |
624 ISCSI_PERSISTENT_ADDRESS |
625 ISCSI_TARGET_NAME | ISCSI_TPGT |
626 ISCSI_USERNAME | ISCSI_PASSWORD |
627 ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
628 ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
629 ISCSI_PING_TMO | ISCSI_RECV_TMO |
630 ISCSI_IFACE_NAME | ISCSI_INITIATOR_NAME,
631 .host_param_mask = ISCSI_HOST_HWADDRESS |
632 ISCSI_HOST_NETDEV_NAME |
633 ISCSI_HOST_INITIATOR_NAME,
634 /* session management */
635 .create_session = iscsi_iser_session_create,
636 .destroy_session = iscsi_iser_session_destroy,
637 /* connection management */
638 .create_conn = iscsi_iser_conn_create,
639 .bind_conn = iscsi_iser_conn_bind,
640 .destroy_conn = iscsi_iser_conn_destroy,
641 .set_param = iscsi_iser_set_param,
642 .get_conn_param = iscsi_conn_get_param,
643 .get_session_param = iscsi_session_get_param,
644 .start_conn = iscsi_iser_conn_start,
645 .stop_conn = iscsi_iser_conn_stop,
646 /* iscsi host params */
647 .get_host_param = iscsi_host_get_param,
648 .set_host_param = iscsi_host_set_param,
650 .send_pdu = iscsi_conn_send_pdu,
651 .get_stats = iscsi_iser_conn_get_stats,
652 .init_task = iscsi_iser_task_init,
653 .xmit_task = iscsi_iser_task_xmit,
654 .cleanup_task = iscsi_iser_cleanup_task,
656 .session_recovery_timedout = iscsi_session_recovery_timedout,
658 .ep_connect = iscsi_iser_ep_connect,
659 .ep_poll = iscsi_iser_ep_poll,
660 .ep_disconnect = iscsi_iser_ep_disconnect
663 static int __init iser_init(void)
667 iser_dbg("Starting iSER datamover...\n");
669 if (iscsi_max_lun < 1) {
670 printk(KERN_ERR "Invalid max_lun value of %u\n", iscsi_max_lun);
674 memset(&ig, 0, sizeof(struct iser_global));
676 ig.desc_cache = kmem_cache_create("iser_descriptors",
677 sizeof (struct iser_desc),
678 0, SLAB_HWCACHE_ALIGN,
680 if (ig.desc_cache == NULL)
683 /* device init is called only after the first addr resolution */
684 mutex_init(&ig.device_list_mutex);
685 INIT_LIST_HEAD(&ig.device_list);
686 mutex_init(&ig.connlist_mutex);
687 INIT_LIST_HEAD(&ig.connlist);
689 iscsi_iser_scsi_transport = iscsi_register_transport(
690 &iscsi_iser_transport);
691 if (!iscsi_iser_scsi_transport) {
692 iser_err("iscsi_register_transport failed\n");
694 goto register_transport_failure;
699 register_transport_failure:
700 kmem_cache_destroy(ig.desc_cache);
705 static void __exit iser_exit(void)
707 iser_dbg("Removing iSER datamover...\n");
708 iscsi_unregister_transport(&iscsi_iser_transport);
709 kmem_cache_destroy(ig.desc_cache);
712 module_init(iser_init);
713 module_exit(iser_exit);