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);
132 * iscsi_iser_task_init - Initialize task
135 * Initialize the task for the scsi command or mgmt command.
138 iscsi_iser_task_init(struct iscsi_task *task)
140 struct iscsi_iser_conn *iser_conn = task->conn->dd_data;
141 struct iscsi_iser_task *iser_task = task->dd_data;
145 iser_task->desc.data = task->data;
149 iser_task->command_sent = 0;
150 iser_task->iser_conn = iser_conn;
151 iser_task_rdma_init(iser_task);
156 * iscsi_iser_mtask_xmit - xmit management(immediate) task
157 * @conn: iscsi connection
158 * @task: task management task
161 * The function can return -EAGAIN in which case caller must
162 * call it again later, or recover. '0' return code means successful
167 iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
171 debug_scsi("task deq [cid %d itt 0x%x]\n", conn->id, task->itt);
173 error = iser_send_control(conn, task);
175 /* since iser xmits control with zero copy, tasks can not be recycled
176 * right after sending them.
177 * The recycling scheme is based on whether a response is expected
178 * - if yes, the task is recycled at iscsi_complete_pdu
179 * - if no, the task is recycled at iser_snd_completion
181 if (error && error != -ENOBUFS)
182 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
188 iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn,
189 struct iscsi_task *task)
191 struct iscsi_r2t_info *r2t = &task->unsol_r2t;
192 struct iscsi_data hdr;
195 /* Send data-out PDUs while there's still unsolicited data to send */
196 while (iscsi_task_has_unsol_data(task)) {
197 iscsi_prep_data_out_pdu(task, r2t, &hdr);
198 debug_scsi("Sending data-out: itt 0x%x, data count %d\n",
199 hdr.itt, r2t->data_count);
201 /* the buffer description has been passed with the command */
202 /* Send the command */
203 error = iser_send_data_out(conn, task, &hdr);
206 goto iscsi_iser_task_xmit_unsol_data_exit;
208 r2t->sent += r2t->data_count;
209 debug_scsi("Need to send %d more as data-out PDUs\n",
210 r2t->data_length - r2t->sent);
213 iscsi_iser_task_xmit_unsol_data_exit:
218 iscsi_iser_task_xmit(struct iscsi_task *task)
220 struct iscsi_conn *conn = task->conn;
221 struct iscsi_iser_task *iser_task = task->dd_data;
225 return iscsi_iser_mtask_xmit(conn, task);
227 if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
228 BUG_ON(scsi_bufflen(task->sc) == 0);
230 debug_scsi("cmd [itt %x total %d imm %d unsol_data %d\n",
231 task->itt, scsi_bufflen(task->sc),
232 task->imm_count, task->unsol_r2t.data_length);
235 debug_scsi("task deq [cid %d itt 0x%x]\n",
236 conn->id, task->itt);
238 /* Send the cmd PDU */
239 if (!iser_task->command_sent) {
240 error = iser_send_command(conn, task);
242 goto iscsi_iser_task_xmit_exit;
243 iser_task->command_sent = 1;
246 /* Send unsolicited data-out PDU(s) if necessary */
247 if (iscsi_task_has_unsol_data(task))
248 error = iscsi_iser_task_xmit_unsol_data(conn, task);
250 iscsi_iser_task_xmit_exit:
251 if (error && error != -ENOBUFS)
252 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
256 static void iscsi_iser_cleanup_task(struct iscsi_task *task)
258 struct iscsi_iser_task *iser_task = task->dd_data;
261 * mgmt tasks do not need special cleanup and we do not
262 * allocate anything in the init task callout
264 if (!task->sc || task->state == ISCSI_TASK_PENDING)
267 if (iser_task->status == ISER_TASK_STATUS_STARTED) {
268 iser_task->status = ISER_TASK_STATUS_COMPLETED;
269 iser_task_rdma_finalize(iser_task);
273 static struct iscsi_cls_conn *
274 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
276 struct iscsi_conn *conn;
277 struct iscsi_cls_conn *cls_conn;
278 struct iscsi_iser_conn *iser_conn;
280 cls_conn = iscsi_conn_setup(cls_session, sizeof(*iser_conn), conn_idx);
283 conn = cls_conn->dd_data;
286 * due to issues with the login code re iser sematics
287 * this not set in iscsi_conn_setup - FIXME
289 conn->max_recv_dlength = 128;
291 iser_conn = conn->dd_data;
292 conn->dd_data = iser_conn;
293 iser_conn->iscsi_conn = conn;
299 iscsi_iser_conn_destroy(struct iscsi_cls_conn *cls_conn)
301 struct iscsi_conn *conn = cls_conn->dd_data;
302 struct iscsi_iser_conn *iser_conn = conn->dd_data;
303 struct iser_conn *ib_conn = iser_conn->ib_conn;
305 iscsi_conn_teardown(cls_conn);
307 * Userspace will normally call the stop callback and
308 * already have freed the ib_conn, but if it goofed up then
312 ib_conn->iser_conn = NULL;
313 iser_conn_put(ib_conn);
318 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
319 struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
322 struct iscsi_conn *conn = cls_conn->dd_data;
323 struct iscsi_iser_conn *iser_conn;
324 struct iser_conn *ib_conn;
325 struct iscsi_endpoint *ep;
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 ep = iscsi_lookup_endpoint(transport_eph);
336 iser_err("can't bind eph %llx\n",
337 (unsigned long long)transport_eph);
340 ib_conn = ep->dd_data;
342 /* binds the iSER connection retrieved from the previously
343 * connected ep_handle to the iSCSI layer connection. exchanges
344 * connection pointers */
345 iser_err("binding iscsi conn %p to iser_conn %p\n",conn,ib_conn);
346 iser_conn = conn->dd_data;
347 ib_conn->iser_conn = iser_conn;
348 iser_conn->ib_conn = ib_conn;
349 iser_conn_get(ib_conn);
354 iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
356 struct iscsi_conn *conn = cls_conn->dd_data;
357 struct iscsi_iser_conn *iser_conn = conn->dd_data;
358 struct iser_conn *ib_conn = iser_conn->ib_conn;
361 * Userspace may have goofed up and not bound the connection or
362 * might have only partially setup the connection.
365 iscsi_conn_stop(cls_conn, flag);
367 * There is no unbind event so the stop callback
368 * must release the ref from the bind.
370 iser_conn_put(ib_conn);
372 iser_conn->ib_conn = NULL;
376 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
378 struct iscsi_conn *conn = cls_conn->dd_data;
381 err = iser_conn_set_full_featured_mode(conn);
385 return iscsi_conn_start(cls_conn);
388 static void iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session)
390 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
392 iscsi_session_teardown(cls_session);
393 iscsi_host_remove(shost);
394 iscsi_host_free(shost);
397 static struct iscsi_cls_session *
398 iscsi_iser_session_create(struct iscsi_endpoint *ep,
399 uint16_t cmds_max, uint16_t qdepth,
400 uint32_t initial_cmdsn, uint32_t *hostno)
402 struct iscsi_cls_session *cls_session;
403 struct iscsi_session *session;
404 struct Scsi_Host *shost;
405 struct iser_conn *ib_conn;
407 shost = iscsi_host_alloc(&iscsi_iser_sht, 0, ISCSI_MAX_CMD_PER_LUN);
410 shost->transportt = iscsi_iser_scsi_transport;
411 shost->max_lun = iscsi_max_lun;
413 shost->max_channel = 0;
414 shost->max_cmd_len = 16;
417 * older userspace tools (before 2.0-870) did not pass us
418 * the leading conn's ep so this will be NULL;
421 ib_conn = ep->dd_data;
423 if (iscsi_host_add(shost,
424 ep ? ib_conn->device->ib_device->dma_device : NULL))
426 *hostno = shost->host_no;
429 * we do not support setting can_queue cmd_per_lun from userspace yet
430 * because we preallocate so many resources
432 cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
433 ISCSI_DEF_XMIT_CMDS_MAX,
434 sizeof(struct iscsi_iser_task),
438 session = cls_session->dd_data;
440 shost->can_queue = session->scsi_cmds_max;
444 iscsi_host_remove(shost);
446 iscsi_host_free(shost);
451 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
452 enum iscsi_param param, char *buf, int buflen)
457 case ISCSI_PARAM_MAX_RECV_DLENGTH:
460 case ISCSI_PARAM_HDRDGST_EN:
461 sscanf(buf, "%d", &value);
463 printk(KERN_ERR "DataDigest wasn't negotiated to None");
467 case ISCSI_PARAM_DATADGST_EN:
468 sscanf(buf, "%d", &value);
470 printk(KERN_ERR "DataDigest wasn't negotiated to None");
474 case ISCSI_PARAM_IFMARKER_EN:
475 sscanf(buf, "%d", &value);
477 printk(KERN_ERR "IFMarker wasn't negotiated to No");
481 case ISCSI_PARAM_OFMARKER_EN:
482 sscanf(buf, "%d", &value);
484 printk(KERN_ERR "OFMarker wasn't negotiated to No");
489 return iscsi_set_param(cls_conn, param, buf, buflen);
496 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
498 struct iscsi_conn *conn = cls_conn->dd_data;
500 stats->txdata_octets = conn->txdata_octets;
501 stats->rxdata_octets = conn->rxdata_octets;
502 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
503 stats->dataout_pdus = conn->dataout_pdus_cnt;
504 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
505 stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
506 stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
507 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
508 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
509 stats->custom_length = 4;
510 strcpy(stats->custom[0].desc, "qp_tx_queue_full");
511 stats->custom[0].value = 0; /* TB iser_conn->qp_tx_queue_full; */
512 strcpy(stats->custom[1].desc, "fmr_map_not_avail");
513 stats->custom[1].value = 0; /* TB iser_conn->fmr_map_not_avail */;
514 strcpy(stats->custom[2].desc, "eh_abort_cnt");
515 stats->custom[2].value = conn->eh_abort_cnt;
516 strcpy(stats->custom[3].desc, "fmr_unalign_cnt");
517 stats->custom[3].value = conn->fmr_unalign_cnt;
520 static struct iscsi_endpoint *
521 iscsi_iser_ep_connect(struct sockaddr *dst_addr, int non_blocking)
524 struct iser_conn *ib_conn;
525 struct iscsi_endpoint *ep;
527 ep = iscsi_create_endpoint(sizeof(*ib_conn));
529 return ERR_PTR(-ENOMEM);
531 ib_conn = ep->dd_data;
533 iser_conn_init(ib_conn);
535 err = iser_connect(ib_conn, NULL, (struct sockaddr_in *)dst_addr,
538 iscsi_destroy_endpoint(ep);
545 iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
547 struct iser_conn *ib_conn;
550 ib_conn = ep->dd_data;
551 rc = wait_event_interruptible_timeout(ib_conn->wait,
552 ib_conn->state == ISER_CONN_UP,
553 msecs_to_jiffies(timeout_ms));
555 /* if conn establishment failed, return error code to iscsi */
557 (ib_conn->state == ISER_CONN_TERMINATING ||
558 ib_conn->state == ISER_CONN_DOWN))
561 iser_err("ib conn %p rc = %d\n", ib_conn, rc);
564 return 1; /* success, this is the equivalent of POLLOUT */
566 return 0; /* timeout */
568 return rc; /* signal */
572 iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep)
574 struct iser_conn *ib_conn;
576 ib_conn = ep->dd_data;
577 if (ib_conn->iser_conn)
579 * Must suspend xmit path if the ep is bound to the
580 * iscsi_conn, so we know we are not accessing the ib_conn
583 * This may not be bound if the ep poll failed.
585 iscsi_suspend_tx(ib_conn->iser_conn->iscsi_conn);
588 iser_err("ib conn %p state %d\n",ib_conn, ib_conn->state);
589 iser_conn_terminate(ib_conn);
592 static struct scsi_host_template iscsi_iser_sht = {
593 .module = THIS_MODULE,
594 .name = "iSCSI Initiator over iSER, v." DRV_VER,
595 .queuecommand = iscsi_queuecommand,
596 .change_queue_depth = iscsi_change_queue_depth,
597 .sg_tablesize = ISCSI_ISER_SG_TABLESIZE,
599 .cmd_per_lun = ISCSI_MAX_CMD_PER_LUN,
600 .eh_abort_handler = iscsi_eh_abort,
601 .eh_device_reset_handler= iscsi_eh_device_reset,
602 .eh_target_reset_handler= iscsi_eh_target_reset,
603 .use_clustering = DISABLE_CLUSTERING,
604 .proc_name = "iscsi_iser",
608 static struct iscsi_transport iscsi_iser_transport = {
609 .owner = THIS_MODULE,
611 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T,
612 .param_mask = ISCSI_MAX_RECV_DLENGTH |
613 ISCSI_MAX_XMIT_DLENGTH |
616 ISCSI_INITIAL_R2T_EN |
621 ISCSI_PDU_INORDER_EN |
622 ISCSI_DATASEQ_INORDER_EN |
624 ISCSI_PERSISTENT_PORT |
625 ISCSI_PERSISTENT_ADDRESS |
626 ISCSI_TARGET_NAME | ISCSI_TPGT |
627 ISCSI_USERNAME | ISCSI_PASSWORD |
628 ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
629 ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
630 ISCSI_PING_TMO | ISCSI_RECV_TMO |
631 ISCSI_IFACE_NAME | ISCSI_INITIATOR_NAME,
632 .host_param_mask = ISCSI_HOST_HWADDRESS |
633 ISCSI_HOST_NETDEV_NAME |
634 ISCSI_HOST_INITIATOR_NAME,
635 /* session management */
636 .create_session = iscsi_iser_session_create,
637 .destroy_session = iscsi_iser_session_destroy,
638 /* connection management */
639 .create_conn = iscsi_iser_conn_create,
640 .bind_conn = iscsi_iser_conn_bind,
641 .destroy_conn = iscsi_iser_conn_destroy,
642 .set_param = iscsi_iser_set_param,
643 .get_conn_param = iscsi_conn_get_param,
644 .get_session_param = iscsi_session_get_param,
645 .start_conn = iscsi_iser_conn_start,
646 .stop_conn = iscsi_iser_conn_stop,
647 /* iscsi host params */
648 .get_host_param = iscsi_host_get_param,
649 .set_host_param = iscsi_host_set_param,
651 .send_pdu = iscsi_conn_send_pdu,
652 .get_stats = iscsi_iser_conn_get_stats,
653 .init_task = iscsi_iser_task_init,
654 .xmit_task = iscsi_iser_task_xmit,
655 .cleanup_task = iscsi_iser_cleanup_task,
656 .alloc_pdu = iscsi_iser_pdu_alloc,
658 .session_recovery_timedout = iscsi_session_recovery_timedout,
660 .ep_connect = iscsi_iser_ep_connect,
661 .ep_poll = iscsi_iser_ep_poll,
662 .ep_disconnect = iscsi_iser_ep_disconnect
665 static int __init iser_init(void)
669 iser_dbg("Starting iSER datamover...\n");
671 if (iscsi_max_lun < 1) {
672 printk(KERN_ERR "Invalid max_lun value of %u\n", iscsi_max_lun);
676 memset(&ig, 0, sizeof(struct iser_global));
678 ig.desc_cache = kmem_cache_create("iser_descriptors",
679 sizeof (struct iser_desc),
680 0, SLAB_HWCACHE_ALIGN,
682 if (ig.desc_cache == NULL)
685 /* device init is called only after the first addr resolution */
686 mutex_init(&ig.device_list_mutex);
687 INIT_LIST_HEAD(&ig.device_list);
688 mutex_init(&ig.connlist_mutex);
689 INIT_LIST_HEAD(&ig.connlist);
691 iscsi_iser_scsi_transport = iscsi_register_transport(
692 &iscsi_iser_transport);
693 if (!iscsi_iser_scsi_transport) {
694 iser_err("iscsi_register_transport failed\n");
696 goto register_transport_failure;
701 register_transport_failure:
702 kmem_cache_destroy(ig.desc_cache);
707 static void __exit iser_exit(void)
709 iser_dbg("Removing iSER datamover...\n");
710 iscsi_unregister_transport(&iscsi_iser_transport);
711 kmem_cache_destroy(ig.desc_cache);
714 module_init(iser_init);
715 module_exit(iser_exit);