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 * Copyright (c) 2013-2014 Mellanox Technologies. All rights reserved.
9 * maintained by openib-general@openib.org
11 * This software is available to you under a choice of one of two
12 * licenses. You may choose to be licensed under the terms of the GNU
13 * General Public License (GPL) Version 2, available from the file
14 * COPYING in the main directory of this source tree, or the
15 * OpenIB.org BSD license below:
17 * Redistribution and use in source and binary forms, with or
18 * without modification, are permitted provided that the following
21 * - Redistributions of source code must retain the above
22 * copyright notice, this list of conditions and the following
25 * - Redistributions in binary form must reproduce the above
26 * copyright notice, this list of conditions and the following
27 * disclaimer in the documentation and/or other materials
28 * provided with the distribution.
30 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
31 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
32 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
33 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
34 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
35 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
36 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
48 #include <linux/types.h>
49 #include <linux/list.h>
50 #include <linux/hardirq.h>
51 #include <linux/kfifo.h>
52 #include <linux/blkdev.h>
53 #include <linux/init.h>
54 #include <linux/ioctl.h>
55 #include <linux/cdev.h>
57 #include <linux/net.h>
58 #include <linux/scatterlist.h>
59 #include <linux/delay.h>
60 #include <linux/slab.h>
61 #include <linux/module.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 MODULE_DESCRIPTION("iSER (iSCSI Extensions for RDMA) Datamover");
78 MODULE_LICENSE("Dual BSD/GPL");
79 MODULE_AUTHOR("Alex Nezhinsky, Dan Bar Dov, Or Gerlitz");
80 MODULE_VERSION(DRV_VER);
82 static struct scsi_host_template iscsi_iser_sht;
83 static struct iscsi_transport iscsi_iser_transport;
84 static struct scsi_transport_template *iscsi_iser_scsi_transport;
85 static struct workqueue_struct *release_wq;
86 static DEFINE_MUTEX(unbind_iser_conn_mutex);
87 struct iser_global ig;
89 int iser_debug_level = 0;
90 module_param_named(debug_level, iser_debug_level, int, S_IRUGO | S_IWUSR);
91 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:disabled)");
93 static unsigned int iscsi_max_lun = 512;
94 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
95 MODULE_PARM_DESC(max_lun, "Max LUNs to allow per session (default:512");
97 unsigned int iser_max_sectors = ISER_DEF_MAX_SECTORS;
98 module_param_named(max_sectors, iser_max_sectors, uint, S_IRUGO | S_IWUSR);
99 MODULE_PARM_DESC(max_sectors, "Max number of sectors in a single scsi command (default:1024");
101 bool iser_always_reg = true;
102 module_param_named(always_register, iser_always_reg, bool, S_IRUGO);
103 MODULE_PARM_DESC(always_register,
104 "Always register memory, even for continuous memory regions (default:true)");
106 bool iser_pi_enable = false;
107 module_param_named(pi_enable, iser_pi_enable, bool, S_IRUGO);
108 MODULE_PARM_DESC(pi_enable, "Enable T10-PI offload support (default:disabled)");
111 module_param_named(pi_guard, iser_pi_guard, int, S_IRUGO);
112 MODULE_PARM_DESC(pi_guard, "T10-PI guard_type [deprecated]");
115 * iscsi_iser_recv() - Process a successful recv completion
116 * @conn: iscsi connection
118 * @rx_data: buffer containing receive data payload
119 * @rx_data_len: length of rx_data
121 * Notes: In case of data length errors or iscsi PDU completion failures
122 * this routine will signal iscsi layer of connection failure.
125 iscsi_iser_recv(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
126 char *rx_data, int rx_data_len)
131 /* verify PDU length */
132 datalen = ntoh24(hdr->dlength);
133 if (datalen > rx_data_len || (datalen + 4) < rx_data_len) {
134 iser_err("wrong datalen %d (hdr), %d (IB)\n",
135 datalen, rx_data_len);
136 rc = ISCSI_ERR_DATALEN;
140 if (datalen != rx_data_len)
141 iser_dbg("aligned datalen (%d) hdr, %d (IB)\n",
142 datalen, rx_data_len);
144 rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len);
145 if (rc && rc != ISCSI_ERR_NO_SCSI_CMD)
150 iscsi_conn_failure(conn, rc);
154 * iscsi_iser_pdu_alloc() - allocate an iscsi-iser PDU
156 * @opcode: iscsi command opcode
158 * Netes: This routine can't fail, just assign iscsi task
159 * hdr and max hdr size.
162 iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
164 struct iscsi_iser_task *iser_task = task->dd_data;
166 task->hdr = (struct iscsi_hdr *)&iser_task->desc.iscsi_header;
167 task->hdr_max = sizeof(iser_task->desc.iscsi_header);
173 * iser_initialize_task_headers() - Initialize task headers
175 * @tx_desc: iser tx descriptor
178 * This routine may race with iser teardown flow for scsi
179 * error handling TMFs. So for TMF we should acquire the
180 * state mutex to avoid dereferencing the IB device which
181 * may have already been terminated.
184 iser_initialize_task_headers(struct iscsi_task *task,
185 struct iser_tx_desc *tx_desc)
187 struct iser_conn *iser_conn = task->conn->dd_data;
188 struct iser_device *device = iser_conn->ib_conn.device;
189 struct iscsi_iser_task *iser_task = task->dd_data;
191 const bool mgmt_task = !task->sc && !in_interrupt();
194 if (unlikely(mgmt_task))
195 mutex_lock(&iser_conn->state_mutex);
197 if (unlikely(iser_conn->state != ISER_CONN_UP)) {
202 dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc,
203 ISER_HEADERS_LEN, DMA_TO_DEVICE);
204 if (ib_dma_mapping_error(device->ib_device, dma_addr)) {
210 tx_desc->mapped = true;
211 tx_desc->dma_addr = dma_addr;
212 tx_desc->tx_sg[0].addr = tx_desc->dma_addr;
213 tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
214 tx_desc->tx_sg[0].lkey = device->pd->local_dma_lkey;
216 iser_task->iser_conn = iser_conn;
218 if (unlikely(mgmt_task))
219 mutex_unlock(&iser_conn->state_mutex);
225 * iscsi_iser_task_init() - Initialize iscsi-iser task
228 * Initialize the task for the scsi command or mgmt command.
230 * Return: Returns zero on success or -ENOMEM when failing
231 * to init task headers (dma mapping error).
234 iscsi_iser_task_init(struct iscsi_task *task)
236 struct iscsi_iser_task *iser_task = task->dd_data;
239 ret = iser_initialize_task_headers(task, &iser_task->desc);
241 iser_err("Failed to init task %p, err = %d\n",
250 iser_task->command_sent = 0;
251 iser_task_rdma_init(iser_task);
252 iser_task->sc = task->sc;
258 * iscsi_iser_mtask_xmit() - xmit management (immediate) task
259 * @conn: iscsi connection
260 * @task: task management task
263 * The function can return -EAGAIN in which case caller must
264 * call it again later, or recover. '0' return code means successful
269 iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
273 iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt);
275 error = iser_send_control(conn, task);
277 /* since iser xmits control with zero copy, tasks can not be recycled
278 * right after sending them.
279 * The recycling scheme is based on whether a response is expected
280 * - if yes, the task is recycled at iscsi_complete_pdu
281 * - if no, the task is recycled at iser_snd_completion
287 iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn,
288 struct iscsi_task *task)
290 struct iscsi_r2t_info *r2t = &task->unsol_r2t;
291 struct iscsi_data hdr;
294 /* Send data-out PDUs while there's still unsolicited data to send */
295 while (iscsi_task_has_unsol_data(task)) {
296 iscsi_prep_data_out_pdu(task, r2t, &hdr);
297 iser_dbg("Sending data-out: itt 0x%x, data count %d\n",
298 hdr.itt, r2t->data_count);
300 /* the buffer description has been passed with the command */
301 /* Send the command */
302 error = iser_send_data_out(conn, task, &hdr);
305 goto iscsi_iser_task_xmit_unsol_data_exit;
307 r2t->sent += r2t->data_count;
308 iser_dbg("Need to send %d more as data-out PDUs\n",
309 r2t->data_length - r2t->sent);
312 iscsi_iser_task_xmit_unsol_data_exit:
317 * iscsi_iser_task_xmit() - xmit iscsi-iser task
320 * Return: zero on success or escalates $error on failure.
323 iscsi_iser_task_xmit(struct iscsi_task *task)
325 struct iscsi_conn *conn = task->conn;
326 struct iscsi_iser_task *iser_task = task->dd_data;
330 return iscsi_iser_mtask_xmit(conn, task);
332 if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
333 BUG_ON(scsi_bufflen(task->sc) == 0);
335 iser_dbg("cmd [itt %x total %d imm %d unsol_data %d\n",
336 task->itt, scsi_bufflen(task->sc),
337 task->imm_count, task->unsol_r2t.data_length);
340 iser_dbg("ctask xmit [cid %d itt 0x%x]\n",
341 conn->id, task->itt);
343 /* Send the cmd PDU */
344 if (!iser_task->command_sent) {
345 error = iser_send_command(conn, task);
347 goto iscsi_iser_task_xmit_exit;
348 iser_task->command_sent = 1;
351 /* Send unsolicited data-out PDU(s) if necessary */
352 if (iscsi_task_has_unsol_data(task))
353 error = iscsi_iser_task_xmit_unsol_data(conn, task);
355 iscsi_iser_task_xmit_exit:
360 * iscsi_iser_cleanup_task() - cleanup an iscsi-iser task
363 * Notes: In case the RDMA device is already NULL (might have
364 * been removed in DEVICE_REMOVAL CM event it will bail-out
365 * without doing dma unmapping.
367 static void iscsi_iser_cleanup_task(struct iscsi_task *task)
369 struct iscsi_iser_task *iser_task = task->dd_data;
370 struct iser_tx_desc *tx_desc = &iser_task->desc;
371 struct iser_conn *iser_conn = task->conn->dd_data;
372 struct iser_device *device = iser_conn->ib_conn.device;
374 /* DEVICE_REMOVAL event might have already released the device */
378 if (likely(tx_desc->mapped)) {
379 ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr,
380 ISER_HEADERS_LEN, DMA_TO_DEVICE);
381 tx_desc->mapped = false;
384 /* mgmt tasks do not need special cleanup */
388 if (iser_task->status == ISER_TASK_STATUS_STARTED) {
389 iser_task->status = ISER_TASK_STATUS_COMPLETED;
390 iser_task_rdma_finalize(iser_task);
395 * iscsi_iser_check_protection() - check protection information status of task.
397 * @sector: error sector if exsists (output)
399 * Return: zero if no data-integrity errors have occured
400 * 0x1: data-integrity error occured in the guard-block
401 * 0x2: data-integrity error occured in the reference tag
402 * 0x3: data-integrity error occured in the application tag
404 * In addition the error sector is marked.
407 iscsi_iser_check_protection(struct iscsi_task *task, sector_t *sector)
409 struct iscsi_iser_task *iser_task = task->dd_data;
411 if (iser_task->dir[ISER_DIR_IN])
412 return iser_check_task_pi_status(iser_task, ISER_DIR_IN,
415 return iser_check_task_pi_status(iser_task, ISER_DIR_OUT,
420 * iscsi_iser_conn_create() - create a new iscsi-iser connection
421 * @cls_session: iscsi class connection
422 * @conn_idx: connection index within the session (for MCS)
424 * Return: iscsi_cls_conn when iscsi_conn_setup succeeds or NULL
427 static struct iscsi_cls_conn *
428 iscsi_iser_conn_create(struct iscsi_cls_session *cls_session,
431 struct iscsi_conn *conn;
432 struct iscsi_cls_conn *cls_conn;
434 cls_conn = iscsi_conn_setup(cls_session, 0, conn_idx);
437 conn = cls_conn->dd_data;
440 * due to issues with the login code re iser sematics
441 * this not set in iscsi_conn_setup - FIXME
443 conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN;
449 * iscsi_iser_conn_bind() - bind iscsi and iser connection structures
450 * @cls_session: iscsi class session
451 * @cls_conn: iscsi class connection
452 * @transport_eph: transport end-point handle
453 * @is_leading: indicate if this is the session leading connection (MCS)
455 * Return: zero on success, $error if iscsi_conn_bind fails and
456 * -EINVAL in case end-point doesn't exsits anymore or iser connection
457 * state is not UP (teardown already started).
460 iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
461 struct iscsi_cls_conn *cls_conn,
462 uint64_t transport_eph,
465 struct iscsi_conn *conn = cls_conn->dd_data;
466 struct iser_conn *iser_conn;
467 struct iscsi_endpoint *ep;
470 error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
474 /* the transport ep handle comes from user space so it must be
475 * verified against the global ib connections list */
476 ep = iscsi_lookup_endpoint(transport_eph);
478 iser_err("can't bind eph %llx\n",
479 (unsigned long long)transport_eph);
482 iser_conn = ep->dd_data;
484 mutex_lock(&iser_conn->state_mutex);
485 if (iser_conn->state != ISER_CONN_UP) {
487 iser_err("iser_conn %p state is %d, teardown started\n",
488 iser_conn, iser_conn->state);
492 error = iser_alloc_rx_descriptors(iser_conn, conn->session);
496 /* binds the iSER connection retrieved from the previously
497 * connected ep_handle to the iSCSI layer connection. exchanges
498 * connection pointers */
499 iser_info("binding iscsi conn %p to iser_conn %p\n", conn, iser_conn);
501 conn->dd_data = iser_conn;
502 iser_conn->iscsi_conn = conn;
505 mutex_unlock(&iser_conn->state_mutex);
510 * iscsi_iser_conn_start() - start iscsi-iser connection
511 * @cls_conn: iscsi class connection
513 * Notes: Here iser intialize (or re-initialize) stop_completion as
514 * from this point iscsi must call conn_stop in session/connection
515 * teardown so iser transport must wait for it.
518 iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
520 struct iscsi_conn *iscsi_conn;
521 struct iser_conn *iser_conn;
523 iscsi_conn = cls_conn->dd_data;
524 iser_conn = iscsi_conn->dd_data;
525 reinit_completion(&iser_conn->stop_completion);
527 return iscsi_conn_start(cls_conn);
531 * iscsi_iser_conn_stop() - stop iscsi-iser connection
532 * @cls_conn: iscsi class connection
533 * @flag: indicate if recover or terminate (passed as is)
535 * Notes: Calling iscsi_conn_stop might theoretically race with
536 * DEVICE_REMOVAL event and dereference a previously freed RDMA device
537 * handle, so we call it under iser the state lock to protect against
541 iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
543 struct iscsi_conn *conn = cls_conn->dd_data;
544 struct iser_conn *iser_conn = conn->dd_data;
546 iser_info("stopping iscsi_conn: %p, iser_conn: %p\n", conn, iser_conn);
549 * Userspace may have goofed up and not bound the connection or
550 * might have only partially setup the connection.
553 mutex_lock(&iser_conn->state_mutex);
554 mutex_lock(&unbind_iser_conn_mutex);
555 iser_conn_terminate(iser_conn);
556 iscsi_conn_stop(cls_conn, flag);
559 iser_conn->iscsi_conn = NULL;
560 conn->dd_data = NULL;
561 mutex_unlock(&unbind_iser_conn_mutex);
563 complete(&iser_conn->stop_completion);
564 mutex_unlock(&iser_conn->state_mutex);
566 iscsi_conn_stop(cls_conn, flag);
571 * iscsi_iser_session_destroy() - destroy iscsi-iser session
572 * @cls_session: iscsi class session
574 * Removes and free iscsi host.
577 iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session)
579 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
581 iscsi_session_teardown(cls_session);
582 iscsi_host_remove(shost);
583 iscsi_host_free(shost);
586 static inline unsigned int
587 iser_dif_prot_caps(int prot_caps)
589 return ((prot_caps & IB_PROT_T10DIF_TYPE_1) ?
590 SHOST_DIF_TYPE1_PROTECTION | SHOST_DIX_TYPE0_PROTECTION |
591 SHOST_DIX_TYPE1_PROTECTION : 0) |
592 ((prot_caps & IB_PROT_T10DIF_TYPE_2) ?
593 SHOST_DIF_TYPE2_PROTECTION | SHOST_DIX_TYPE2_PROTECTION : 0) |
594 ((prot_caps & IB_PROT_T10DIF_TYPE_3) ?
595 SHOST_DIF_TYPE3_PROTECTION | SHOST_DIX_TYPE3_PROTECTION : 0);
599 * iscsi_iser_session_create() - create an iscsi-iser session
600 * @ep: iscsi end-point handle
601 * @cmds_max: maximum commands in this session
602 * @qdepth: session command queue depth
603 * @initial_cmdsn: initiator command sequnce number
605 * Allocates and adds a scsi host, expose DIF supprot if
606 * exists, and sets up an iscsi session.
608 static struct iscsi_cls_session *
609 iscsi_iser_session_create(struct iscsi_endpoint *ep,
610 uint16_t cmds_max, uint16_t qdepth,
611 uint32_t initial_cmdsn)
613 struct iscsi_cls_session *cls_session;
614 struct iscsi_session *session;
615 struct Scsi_Host *shost;
616 struct iser_conn *iser_conn = NULL;
617 struct ib_conn *ib_conn;
621 shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0);
624 shost->transportt = iscsi_iser_scsi_transport;
625 shost->cmd_per_lun = qdepth;
626 shost->max_lun = iscsi_max_lun;
628 shost->max_channel = 0;
629 shost->max_cmd_len = 16;
632 * older userspace tools (before 2.0-870) did not pass us
633 * the leading conn's ep so this will be NULL;
636 iser_conn = ep->dd_data;
637 max_cmds = iser_conn->max_cmds;
638 shost->sg_tablesize = iser_conn->scsi_sg_tablesize;
640 mutex_lock(&iser_conn->state_mutex);
641 if (iser_conn->state != ISER_CONN_UP) {
642 iser_err("iser conn %p already started teardown\n",
644 mutex_unlock(&iser_conn->state_mutex);
648 ib_conn = &iser_conn->ib_conn;
649 if (ib_conn->pi_support) {
650 u32 sig_caps = ib_conn->device->ib_device->attrs.sig_prot_cap;
652 shost->sg_prot_tablesize = shost->sg_tablesize;
653 scsi_host_set_prot(shost, iser_dif_prot_caps(sig_caps));
654 scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP |
655 SHOST_DIX_GUARD_CRC);
658 if (iscsi_host_add(shost,
659 ib_conn->device->ib_device->dma_device)) {
660 mutex_unlock(&iser_conn->state_mutex);
663 mutex_unlock(&iser_conn->state_mutex);
665 max_cmds = ISER_DEF_XMIT_CMDS_MAX;
666 if (iscsi_host_add(shost, NULL))
671 * FRs or FMRs can only map up to a (device) page per entry, but if the
672 * first entry is misaligned we'll end up using using two entries
673 * (head and tail) for a single page worth data, so we have to drop
674 * one segment from the calculation.
676 max_fr_sectors = ((shost->sg_tablesize - 1) * PAGE_SIZE) >> 9;
677 shost->max_sectors = min(iser_max_sectors, max_fr_sectors);
679 if (cmds_max > max_cmds) {
680 iser_info("cmds_max changed from %u to %u\n",
685 cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
687 sizeof(struct iscsi_iser_task),
691 session = cls_session->dd_data;
693 shost->can_queue = session->scsi_cmds_max;
697 iscsi_host_remove(shost);
699 iscsi_host_free(shost);
704 iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
705 enum iscsi_param param, char *buf, int buflen)
710 case ISCSI_PARAM_MAX_RECV_DLENGTH:
713 case ISCSI_PARAM_HDRDGST_EN:
714 sscanf(buf, "%d", &value);
716 iser_err("DataDigest wasn't negotiated to None\n");
720 case ISCSI_PARAM_DATADGST_EN:
721 sscanf(buf, "%d", &value);
723 iser_err("DataDigest wasn't negotiated to None\n");
727 case ISCSI_PARAM_IFMARKER_EN:
728 sscanf(buf, "%d", &value);
730 iser_err("IFMarker wasn't negotiated to No\n");
734 case ISCSI_PARAM_OFMARKER_EN:
735 sscanf(buf, "%d", &value);
737 iser_err("OFMarker wasn't negotiated to No\n");
742 return iscsi_set_param(cls_conn, param, buf, buflen);
749 * iscsi_iser_set_param() - set class connection parameter
750 * @cls_conn: iscsi class connection
751 * @stats: iscsi stats to output
753 * Output connection statistics.
756 iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
758 struct iscsi_conn *conn = cls_conn->dd_data;
760 stats->txdata_octets = conn->txdata_octets;
761 stats->rxdata_octets = conn->rxdata_octets;
762 stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
763 stats->dataout_pdus = conn->dataout_pdus_cnt;
764 stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
765 stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
766 stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
767 stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
768 stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
769 stats->custom_length = 0;
772 static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep,
773 enum iscsi_param param, char *buf)
775 struct iser_conn *iser_conn = ep->dd_data;
779 case ISCSI_PARAM_CONN_PORT:
780 case ISCSI_PARAM_CONN_ADDRESS:
781 if (!iser_conn || !iser_conn->ib_conn.cma_id)
784 return iscsi_conn_get_addr_param((struct sockaddr_storage *)
785 &iser_conn->ib_conn.cma_id->route.addr.dst_addr,
796 * iscsi_iser_ep_connect() - Initiate iSER connection establishment
798 * @dst_addr: destination address
799 * @non-blocking: indicate if routine can block
801 * Allocate an iscsi endpoint, an iser_conn structure and bind them.
802 * After that start RDMA connection establishment via rdma_cm. We
803 * don't allocate iser_conn embedded in iscsi_endpoint since in teardown
804 * the endpoint will be destroyed at ep_disconnect while iser_conn will
805 * cleanup its resources asynchronuously.
807 * Return: iscsi_endpoint created by iscsi layer or ERR_PTR(error)
810 static struct iscsi_endpoint *
811 iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr,
815 struct iser_conn *iser_conn;
816 struct iscsi_endpoint *ep;
818 ep = iscsi_create_endpoint(0);
820 return ERR_PTR(-ENOMEM);
822 iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL);
828 ep->dd_data = iser_conn;
830 iser_conn_init(iser_conn);
832 err = iser_connect(iser_conn, NULL, dst_addr, non_blocking);
838 iscsi_destroy_endpoint(ep);
843 * iscsi_iser_ep_poll() - poll for iser connection establishment to complete
844 * @ep: iscsi endpoint (created at ep_connect)
845 * @timeout_ms: polling timeout allowed in ms.
847 * This routine boils down to waiting for up_completion signaling
848 * that cma_id got CONNECTED event.
850 * Return: 1 if succeeded in connection establishment, 0 if timeout expired
851 * (libiscsi will retry will kick in) or -1 if interrupted by signal
852 * or more likely iser connection state transitioned to TEMINATING or
853 * DOWN during the wait period.
856 iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
858 struct iser_conn *iser_conn = ep->dd_data;
861 rc = wait_for_completion_interruptible_timeout(&iser_conn->up_completion,
862 msecs_to_jiffies(timeout_ms));
863 /* if conn establishment failed, return error code to iscsi */
865 mutex_lock(&iser_conn->state_mutex);
866 if (iser_conn->state == ISER_CONN_TERMINATING ||
867 iser_conn->state == ISER_CONN_DOWN)
869 mutex_unlock(&iser_conn->state_mutex);
872 iser_info("iser conn %p rc = %d\n", iser_conn, rc);
875 return 1; /* success, this is the equivalent of POLLOUT */
877 return 0; /* timeout */
879 return rc; /* signal */
883 * iscsi_iser_ep_disconnect() - Initiate connection teardown process
884 * @ep: iscsi endpoint handle
886 * This routine is not blocked by iser and RDMA termination process
887 * completion as we queue a deffered work for iser/RDMA destruction
888 * and cleanup or actually call it immediately in case we didn't pass
889 * iscsi conn bind/start stage, thus it is safe.
892 iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep)
894 struct iser_conn *iser_conn = ep->dd_data;
896 iser_info("ep %p iser conn %p\n", ep, iser_conn);
898 mutex_lock(&iser_conn->state_mutex);
899 iser_conn_terminate(iser_conn);
902 * if iser_conn and iscsi_conn are bound, we must wait for
903 * iscsi_conn_stop and flush errors completion before freeing
904 * the iser resources. Otherwise we are safe to free resources
907 if (iser_conn->iscsi_conn) {
908 INIT_WORK(&iser_conn->release_work, iser_release_work);
909 queue_work(release_wq, &iser_conn->release_work);
910 mutex_unlock(&iser_conn->state_mutex);
912 iser_conn->state = ISER_CONN_DOWN;
913 mutex_unlock(&iser_conn->state_mutex);
914 iser_conn_release(iser_conn);
917 iscsi_destroy_endpoint(ep);
920 static umode_t iser_attr_is_visible(int param_type, int param)
922 switch (param_type) {
923 case ISCSI_HOST_PARAM:
925 case ISCSI_HOST_PARAM_NETDEV_NAME:
926 case ISCSI_HOST_PARAM_HWADDRESS:
927 case ISCSI_HOST_PARAM_INITIATOR_NAME:
934 case ISCSI_PARAM_MAX_RECV_DLENGTH:
935 case ISCSI_PARAM_MAX_XMIT_DLENGTH:
936 case ISCSI_PARAM_HDRDGST_EN:
937 case ISCSI_PARAM_DATADGST_EN:
938 case ISCSI_PARAM_CONN_ADDRESS:
939 case ISCSI_PARAM_CONN_PORT:
940 case ISCSI_PARAM_EXP_STATSN:
941 case ISCSI_PARAM_PERSISTENT_ADDRESS:
942 case ISCSI_PARAM_PERSISTENT_PORT:
943 case ISCSI_PARAM_PING_TMO:
944 case ISCSI_PARAM_RECV_TMO:
945 case ISCSI_PARAM_INITIAL_R2T_EN:
946 case ISCSI_PARAM_MAX_R2T:
947 case ISCSI_PARAM_IMM_DATA_EN:
948 case ISCSI_PARAM_FIRST_BURST:
949 case ISCSI_PARAM_MAX_BURST:
950 case ISCSI_PARAM_PDU_INORDER_EN:
951 case ISCSI_PARAM_DATASEQ_INORDER_EN:
952 case ISCSI_PARAM_TARGET_NAME:
953 case ISCSI_PARAM_TPGT:
954 case ISCSI_PARAM_USERNAME:
955 case ISCSI_PARAM_PASSWORD:
956 case ISCSI_PARAM_USERNAME_IN:
957 case ISCSI_PARAM_PASSWORD_IN:
958 case ISCSI_PARAM_FAST_ABORT:
959 case ISCSI_PARAM_ABORT_TMO:
960 case ISCSI_PARAM_LU_RESET_TMO:
961 case ISCSI_PARAM_TGT_RESET_TMO:
962 case ISCSI_PARAM_IFACE_NAME:
963 case ISCSI_PARAM_INITIATOR_NAME:
964 case ISCSI_PARAM_DISCOVERY_SESS:
974 static int iscsi_iser_slave_alloc(struct scsi_device *sdev)
976 struct iscsi_session *session;
977 struct iser_conn *iser_conn;
978 struct ib_device *ib_dev;
980 mutex_lock(&unbind_iser_conn_mutex);
982 session = starget_to_session(scsi_target(sdev))->dd_data;
983 iser_conn = session->leadconn->dd_data;
985 mutex_unlock(&unbind_iser_conn_mutex);
988 ib_dev = iser_conn->ib_conn.device->ib_device;
990 if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG))
991 blk_queue_virt_boundary(sdev->request_queue, ~MASK_4K);
993 mutex_unlock(&unbind_iser_conn_mutex);
998 static struct scsi_host_template iscsi_iser_sht = {
999 .module = THIS_MODULE,
1000 .name = "iSCSI Initiator over iSER",
1001 .queuecommand = iscsi_queuecommand,
1002 .change_queue_depth = scsi_change_queue_depth,
1003 .sg_tablesize = ISCSI_ISER_DEF_SG_TABLESIZE,
1004 .cmd_per_lun = ISER_DEF_CMD_PER_LUN,
1005 .eh_abort_handler = iscsi_eh_abort,
1006 .eh_device_reset_handler= iscsi_eh_device_reset,
1007 .eh_target_reset_handler = iscsi_eh_recover_target,
1008 .target_alloc = iscsi_target_alloc,
1009 .use_clustering = ENABLE_CLUSTERING,
1010 .slave_alloc = iscsi_iser_slave_alloc,
1011 .proc_name = "iscsi_iser",
1013 .track_queue_depth = 1,
1016 static struct iscsi_transport iscsi_iser_transport = {
1017 .owner = THIS_MODULE,
1019 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO,
1020 /* session management */
1021 .create_session = iscsi_iser_session_create,
1022 .destroy_session = iscsi_iser_session_destroy,
1023 /* connection management */
1024 .create_conn = iscsi_iser_conn_create,
1025 .bind_conn = iscsi_iser_conn_bind,
1026 .destroy_conn = iscsi_conn_teardown,
1027 .attr_is_visible = iser_attr_is_visible,
1028 .set_param = iscsi_iser_set_param,
1029 .get_conn_param = iscsi_conn_get_param,
1030 .get_ep_param = iscsi_iser_get_ep_param,
1031 .get_session_param = iscsi_session_get_param,
1032 .start_conn = iscsi_iser_conn_start,
1033 .stop_conn = iscsi_iser_conn_stop,
1034 /* iscsi host params */
1035 .get_host_param = iscsi_host_get_param,
1036 .set_host_param = iscsi_host_set_param,
1038 .send_pdu = iscsi_conn_send_pdu,
1039 .get_stats = iscsi_iser_conn_get_stats,
1040 .init_task = iscsi_iser_task_init,
1041 .xmit_task = iscsi_iser_task_xmit,
1042 .cleanup_task = iscsi_iser_cleanup_task,
1043 .alloc_pdu = iscsi_iser_pdu_alloc,
1044 .check_protection = iscsi_iser_check_protection,
1046 .session_recovery_timedout = iscsi_session_recovery_timedout,
1048 .ep_connect = iscsi_iser_ep_connect,
1049 .ep_poll = iscsi_iser_ep_poll,
1050 .ep_disconnect = iscsi_iser_ep_disconnect
1053 static int __init iser_init(void)
1057 iser_dbg("Starting iSER datamover...\n");
1059 if (iscsi_max_lun < 1) {
1060 iser_err("Invalid max_lun value of %u\n", iscsi_max_lun);
1064 memset(&ig, 0, sizeof(struct iser_global));
1066 ig.desc_cache = kmem_cache_create("iser_descriptors",
1067 sizeof(struct iser_tx_desc),
1068 0, SLAB_HWCACHE_ALIGN,
1070 if (ig.desc_cache == NULL)
1073 /* device init is called only after the first addr resolution */
1074 mutex_init(&ig.device_list_mutex);
1075 INIT_LIST_HEAD(&ig.device_list);
1076 mutex_init(&ig.connlist_mutex);
1077 INIT_LIST_HEAD(&ig.connlist);
1079 release_wq = alloc_workqueue("release workqueue", 0, 0);
1081 iser_err("failed to allocate release workqueue\n");
1086 iscsi_iser_scsi_transport = iscsi_register_transport(
1087 &iscsi_iser_transport);
1088 if (!iscsi_iser_scsi_transport) {
1089 iser_err("iscsi_register_transport failed\n");
1097 destroy_workqueue(release_wq);
1099 kmem_cache_destroy(ig.desc_cache);
1104 static void __exit iser_exit(void)
1106 struct iser_conn *iser_conn, *n;
1109 iser_dbg("Removing iSER datamover...\n");
1110 destroy_workqueue(release_wq);
1112 mutex_lock(&ig.connlist_mutex);
1113 connlist_empty = list_empty(&ig.connlist);
1114 mutex_unlock(&ig.connlist_mutex);
1116 if (!connlist_empty) {
1117 iser_err("Error cleanup stage completed but we still have iser "
1118 "connections, destroying them anyway\n");
1119 list_for_each_entry_safe(iser_conn, n, &ig.connlist,
1121 iser_conn_release(iser_conn);
1125 iscsi_unregister_transport(&iscsi_iser_transport);
1126 kmem_cache_destroy(ig.desc_cache);
1129 module_init(iser_init);
1130 module_exit(iser_exit);