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 unsigned int iscsi_max_lun = 512;
75 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
77 int iser_debug_level = 0;
79 MODULE_DESCRIPTION("iSER (iSCSI Extensions for RDMA) Datamover "
80 "v" DRV_VER " (" DRV_DATE ")");
81 MODULE_LICENSE("Dual BSD/GPL");
82 MODULE_AUTHOR("Alex Nezhinsky, Dan Bar Dov, Or Gerlitz");
84 module_param_named(debug_level, iser_debug_level, int, 0644);
85 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:disabled)");
87 struct iser_global ig;
90 iscsi_iser_recv(struct iscsi_conn *conn,
91 struct iscsi_hdr *hdr, char *rx_data, int rx_data_len)
98 /* verify PDU length */
99 datalen = ntoh24(hdr->dlength);
100 if (datalen != rx_data_len) {
101 printk(KERN_ERR "iscsi_iser: datalen %d (hdr) != %d (IB) \n",
102 datalen, rx_data_len);
103 rc = ISCSI_ERR_DATALEN;
108 ahslen = hdr->hlength * 4;
110 /* verify itt (itt encoding: age+cid+itt) */
111 rc = iscsi_verify_itt(conn, hdr, &ret_itt);
114 rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len);
116 if (rc && rc != ISCSI_ERR_NO_SCSI_CMD)
121 iscsi_conn_failure(conn, rc);
126 * iscsi_iser_cmd_init - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
130 iscsi_iser_cmd_init(struct iscsi_cmd_task *ctask)
132 struct iscsi_iser_conn *iser_conn = ctask->conn->dd_data;
133 struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data;
135 iser_ctask->command_sent = 0;
136 iser_ctask->iser_conn = iser_conn;
137 iser_ctask_rdma_init(iser_ctask);
142 * iscsi_mtask_xmit - xmit management(immediate) task
143 * @conn: iscsi connection
144 * @mtask: task management task
147 * The function can return -EAGAIN in which case caller must
148 * call it again later, or recover. '0' return code means successful
153 iscsi_iser_mtask_xmit(struct iscsi_conn *conn,
154 struct iscsi_mgmt_task *mtask)
158 debug_scsi("mtask deq [cid %d itt 0x%x]\n", conn->id, mtask->itt);
160 error = iser_send_control(conn, mtask);
162 /* since iser xmits control with zero copy, mtasks can not be recycled
163 * right after sending them.
164 * The recycling scheme is based on whether a response is expected
165 * - if yes, the mtask is recycled at iscsi_complete_pdu
166 * - if no, the mtask is recycled at iser_snd_completion
168 if (error && error != -ENOBUFS)
169 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
175 iscsi_iser_ctask_xmit_unsol_data(struct iscsi_conn *conn,
176 struct iscsi_cmd_task *ctask)
178 struct iscsi_data hdr;
181 /* Send data-out PDUs while there's still unsolicited data to send */
182 while (ctask->unsol_count > 0) {
183 iscsi_prep_unsolicit_data_pdu(ctask, &hdr);
184 debug_scsi("Sending data-out: itt 0x%x, data count %d\n",
185 hdr.itt, ctask->data_count);
187 /* the buffer description has been passed with the command */
188 /* Send the command */
189 error = iser_send_data_out(conn, ctask, &hdr);
191 ctask->unsol_datasn--;
192 goto iscsi_iser_ctask_xmit_unsol_data_exit;
194 ctask->unsol_count -= ctask->data_count;
195 debug_scsi("Need to send %d more as data-out PDUs\n",
199 iscsi_iser_ctask_xmit_unsol_data_exit:
204 iscsi_iser_ctask_xmit(struct iscsi_conn *conn,
205 struct iscsi_cmd_task *ctask)
207 struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data;
210 if (ctask->sc->sc_data_direction == DMA_TO_DEVICE) {
211 BUG_ON(scsi_bufflen(ctask->sc) == 0);
213 debug_scsi("cmd [itt %x total %d imm %d unsol_data %d\n",
214 ctask->itt, scsi_bufflen(ctask->sc),
215 ctask->imm_count, ctask->unsol_count);
218 debug_scsi("ctask deq [cid %d itt 0x%x]\n",
219 conn->id, ctask->itt);
221 /* Send the cmd PDU */
222 if (!iser_ctask->command_sent) {
223 error = iser_send_command(conn, ctask);
225 goto iscsi_iser_ctask_xmit_exit;
226 iser_ctask->command_sent = 1;
229 /* Send unsolicited data-out PDU(s) if necessary */
230 if (ctask->unsol_count)
231 error = iscsi_iser_ctask_xmit_unsol_data(conn, ctask);
233 iscsi_iser_ctask_xmit_exit:
234 if (error && error != -ENOBUFS)
235 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
240 iscsi_iser_cleanup_ctask(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
242 struct iscsi_iser_cmd_task *iser_ctask = ctask->dd_data;
244 if (iser_ctask->status == ISER_TASK_STATUS_STARTED) {
245 iser_ctask->status = ISER_TASK_STATUS_COMPLETED;
246 iser_ctask_rdma_finalize(iser_ctask);
250 static struct iser_conn *
251 iscsi_iser_ib_conn_lookup(__u64 ep_handle)
253 struct iser_conn *ib_conn;
254 struct iser_conn *uib_conn = (struct iser_conn *)(unsigned long)ep_handle;
256 mutex_lock(&ig.connlist_mutex);
257 list_for_each_entry(ib_conn, &ig.connlist, conn_list) {
258 if (ib_conn == uib_conn) {
259 mutex_unlock(&ig.connlist_mutex);
263 mutex_unlock(&ig.connlist_mutex);
264 iser_err("no conn exists for eph %llx\n",(unsigned long long)ep_handle);
268 static struct iscsi_cls_conn *
269 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
271 struct iscsi_conn *conn;
272 struct iscsi_cls_conn *cls_conn;
273 struct iscsi_iser_conn *iser_conn;
275 cls_conn = iscsi_conn_setup(cls_session, conn_idx);
278 conn = cls_conn->dd_data;
281 * due to issues with the login code re iser sematics
282 * this not set in iscsi_conn_setup - FIXME
284 conn->max_recv_dlength = 128;
286 iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL);
288 goto conn_alloc_fail;
290 /* currently this is the only field which need to be initiated */
291 rwlock_init(&iser_conn->lock);
293 conn->dd_data = iser_conn;
294 iser_conn->iscsi_conn = conn;
299 iscsi_conn_teardown(cls_conn);
304 iscsi_iser_conn_destroy(struct iscsi_cls_conn *cls_conn)
306 struct iscsi_conn *conn = cls_conn->dd_data;
307 struct iscsi_iser_conn *iser_conn = conn->dd_data;
309 iscsi_conn_teardown(cls_conn);
310 if (iser_conn->ib_conn)
311 iser_conn->ib_conn->iser_conn = NULL;
316 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
317 struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
320 struct iscsi_conn *conn = cls_conn->dd_data;
321 struct iscsi_iser_conn *iser_conn;
322 struct iser_conn *ib_conn;
325 error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
329 /* the transport ep handle comes from user space so it must be
330 * verified against the global ib connections list */
331 ib_conn = iscsi_iser_ib_conn_lookup(transport_eph);
333 iser_err("can't bind eph %llx\n",
334 (unsigned long long)transport_eph);
337 /* binds the iSER connection retrieved from the previously
338 * connected ep_handle to the iSCSI layer connection. exchanges
339 * connection pointers */
340 iser_err("binding iscsi conn %p to iser_conn %p\n",conn,ib_conn);
341 iser_conn = conn->dd_data;
342 ib_conn->iser_conn = iser_conn;
343 iser_conn->ib_conn = ib_conn;
345 conn->recv_lock = &iser_conn->lock;
351 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
353 struct iscsi_conn *conn = cls_conn->dd_data;
356 err = iser_conn_set_full_featured_mode(conn);
360 return iscsi_conn_start(cls_conn);
363 static struct iscsi_transport iscsi_iser_transport;
365 static struct iscsi_cls_session *
366 iscsi_iser_session_create(struct iscsi_transport *iscsit,
367 struct scsi_transport_template *scsit,
368 uint16_t cmds_max, uint16_t qdepth,
369 uint32_t initial_cmdsn, uint32_t *hostno)
371 struct iscsi_cls_session *cls_session;
372 struct iscsi_session *session;
375 struct iscsi_cmd_task *ctask;
376 struct iscsi_mgmt_task *mtask;
377 struct iscsi_iser_cmd_task *iser_ctask;
378 struct iser_desc *desc;
381 * we do not support setting can_queue cmd_per_lun from userspace yet
382 * because we preallocate so many resources
384 cls_session = iscsi_session_setup(iscsit, scsit,
385 ISCSI_DEF_XMIT_CMDS_MAX,
386 ISCSI_MAX_CMD_PER_LUN,
387 sizeof(struct iscsi_iser_cmd_task),
388 sizeof(struct iser_desc),
394 session = class_to_transport_session(cls_session);
396 /* libiscsi setup itts, data and pool so just set desc fields */
397 for (i = 0; i < session->cmds_max; i++) {
398 ctask = session->cmds[i];
399 iser_ctask = ctask->dd_data;
400 ctask->hdr = (struct iscsi_cmd *)&iser_ctask->desc.iscsi_header;
401 ctask->hdr_max = sizeof(iser_ctask->desc.iscsi_header);
404 for (i = 0; i < session->mgmtpool_max; i++) {
405 mtask = session->mgmt_cmds[i];
406 desc = mtask->dd_data;
407 mtask->hdr = &desc->iscsi_header;
408 desc->data = mtask->data;
415 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
416 enum iscsi_param param, char *buf, int buflen)
421 case ISCSI_PARAM_MAX_RECV_DLENGTH:
424 case ISCSI_PARAM_HDRDGST_EN:
425 sscanf(buf, "%d", &value);
427 printk(KERN_ERR "DataDigest wasn't negotiated to None");
431 case ISCSI_PARAM_DATADGST_EN:
432 sscanf(buf, "%d", &value);
434 printk(KERN_ERR "DataDigest wasn't negotiated to None");
438 case ISCSI_PARAM_IFMARKER_EN:
439 sscanf(buf, "%d", &value);
441 printk(KERN_ERR "IFMarker wasn't negotiated to No");
445 case ISCSI_PARAM_OFMARKER_EN:
446 sscanf(buf, "%d", &value);
448 printk(KERN_ERR "OFMarker wasn't negotiated to No");
453 return iscsi_set_param(cls_conn, param, buf, buflen);
460 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
462 struct iscsi_conn *conn = cls_conn->dd_data;
464 stats->txdata_octets = conn->txdata_octets;
465 stats->rxdata_octets = conn->rxdata_octets;
466 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
467 stats->dataout_pdus = conn->dataout_pdus_cnt;
468 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
469 stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
470 stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
471 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
472 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
473 stats->custom_length = 4;
474 strcpy(stats->custom[0].desc, "qp_tx_queue_full");
475 stats->custom[0].value = 0; /* TB iser_conn->qp_tx_queue_full; */
476 strcpy(stats->custom[1].desc, "fmr_map_not_avail");
477 stats->custom[1].value = 0; /* TB iser_conn->fmr_map_not_avail */;
478 strcpy(stats->custom[2].desc, "eh_abort_cnt");
479 stats->custom[2].value = conn->eh_abort_cnt;
480 strcpy(stats->custom[3].desc, "fmr_unalign_cnt");
481 stats->custom[3].value = conn->fmr_unalign_cnt;
485 iscsi_iser_ep_connect(struct sockaddr *dst_addr, int non_blocking,
489 struct iser_conn *ib_conn;
491 err = iser_conn_init(&ib_conn);
495 err = iser_connect(ib_conn, NULL, (struct sockaddr_in *)dst_addr, non_blocking);
497 *ep_handle = (__u64)(unsigned long)ib_conn;
504 iscsi_iser_ep_poll(__u64 ep_handle, int timeout_ms)
506 struct iser_conn *ib_conn = iscsi_iser_ib_conn_lookup(ep_handle);
512 rc = wait_event_interruptible_timeout(ib_conn->wait,
513 ib_conn->state == ISER_CONN_UP,
514 msecs_to_jiffies(timeout_ms));
516 /* if conn establishment failed, return error code to iscsi */
518 (ib_conn->state == ISER_CONN_TERMINATING ||
519 ib_conn->state == ISER_CONN_DOWN))
522 iser_err("ib conn %p rc = %d\n", ib_conn, rc);
525 return 1; /* success, this is the equivalent of POLLOUT */
527 return 0; /* timeout */
529 return rc; /* signal */
533 iscsi_iser_ep_disconnect(__u64 ep_handle)
535 struct iser_conn *ib_conn;
537 ib_conn = iscsi_iser_ib_conn_lookup(ep_handle);
541 iser_err("ib conn %p state %d\n",ib_conn, ib_conn->state);
542 iser_conn_terminate(ib_conn);
545 static struct scsi_host_template iscsi_iser_sht = {
546 .module = THIS_MODULE,
547 .name = "iSCSI Initiator over iSER, v." DRV_VER,
548 .queuecommand = iscsi_queuecommand,
549 .change_queue_depth = iscsi_change_queue_depth,
550 .can_queue = ISCSI_DEF_XMIT_CMDS_MAX - 1,
551 .sg_tablesize = ISCSI_ISER_SG_TABLESIZE,
553 .cmd_per_lun = ISCSI_MAX_CMD_PER_LUN,
554 .eh_abort_handler = iscsi_eh_abort,
555 .eh_device_reset_handler= iscsi_eh_device_reset,
556 .eh_host_reset_handler = iscsi_eh_host_reset,
557 .use_clustering = DISABLE_CLUSTERING,
558 .proc_name = "iscsi_iser",
562 static struct iscsi_transport iscsi_iser_transport = {
563 .owner = THIS_MODULE,
565 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T,
566 .param_mask = ISCSI_MAX_RECV_DLENGTH |
567 ISCSI_MAX_XMIT_DLENGTH |
570 ISCSI_INITIAL_R2T_EN |
575 ISCSI_PDU_INORDER_EN |
576 ISCSI_DATASEQ_INORDER_EN |
578 ISCSI_PERSISTENT_PORT |
579 ISCSI_PERSISTENT_ADDRESS |
580 ISCSI_TARGET_NAME | ISCSI_TPGT |
581 ISCSI_USERNAME | ISCSI_PASSWORD |
582 ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
583 ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
584 ISCSI_PING_TMO | ISCSI_RECV_TMO,
585 .host_param_mask = ISCSI_HOST_HWADDRESS |
586 ISCSI_HOST_NETDEV_NAME |
587 ISCSI_HOST_INITIATOR_NAME,
588 .host_template = &iscsi_iser_sht,
589 .conndata_size = sizeof(struct iscsi_conn),
590 .max_lun = ISCSI_ISER_MAX_LUN,
591 .max_cmd_len = ISCSI_ISER_MAX_CMD_LEN,
592 /* session management */
593 .create_session = iscsi_iser_session_create,
594 .destroy_session = iscsi_session_teardown,
595 /* connection management */
596 .create_conn = iscsi_iser_conn_create,
597 .bind_conn = iscsi_iser_conn_bind,
598 .destroy_conn = iscsi_iser_conn_destroy,
599 .set_param = iscsi_iser_set_param,
600 .get_conn_param = iscsi_conn_get_param,
601 .get_session_param = iscsi_session_get_param,
602 .start_conn = iscsi_iser_conn_start,
603 .stop_conn = iscsi_conn_stop,
604 /* iscsi host params */
605 .get_host_param = iscsi_host_get_param,
606 .set_host_param = iscsi_host_set_param,
608 .send_pdu = iscsi_conn_send_pdu,
609 .get_stats = iscsi_iser_conn_get_stats,
610 .init_cmd_task = iscsi_iser_cmd_init,
611 .xmit_cmd_task = iscsi_iser_ctask_xmit,
612 .xmit_mgmt_task = iscsi_iser_mtask_xmit,
613 .cleanup_cmd_task = iscsi_iser_cleanup_ctask,
615 .session_recovery_timedout = iscsi_session_recovery_timedout,
617 .ep_connect = iscsi_iser_ep_connect,
618 .ep_poll = iscsi_iser_ep_poll,
619 .ep_disconnect = iscsi_iser_ep_disconnect
622 static int __init iser_init(void)
626 iser_dbg("Starting iSER datamover...\n");
628 if (iscsi_max_lun < 1) {
629 printk(KERN_ERR "Invalid max_lun value of %u\n", iscsi_max_lun);
633 iscsi_iser_transport.max_lun = iscsi_max_lun;
635 memset(&ig, 0, sizeof(struct iser_global));
637 ig.desc_cache = kmem_cache_create("iser_descriptors",
638 sizeof (struct iser_desc),
639 0, SLAB_HWCACHE_ALIGN,
641 if (ig.desc_cache == NULL)
644 /* device init is called only after the first addr resolution */
645 mutex_init(&ig.device_list_mutex);
646 INIT_LIST_HEAD(&ig.device_list);
647 mutex_init(&ig.connlist_mutex);
648 INIT_LIST_HEAD(&ig.connlist);
650 if (!iscsi_register_transport(&iscsi_iser_transport)) {
651 iser_err("iscsi_register_transport failed\n");
653 goto register_transport_failure;
658 register_transport_failure:
659 kmem_cache_destroy(ig.desc_cache);
664 static void __exit iser_exit(void)
666 iser_dbg("Removing iSER datamover...\n");
667 iscsi_unregister_transport(&iscsi_iser_transport);
668 kmem_cache_destroy(ig.desc_cache);
671 module_init(iser_init);
672 module_exit(iser_exit);