linux/drivers/infiniband/ulp/iser/iscsi_iser.c
<<
>>
Prefs
   1/*
   2 * iSCSI Initiator over iSER Data-Path
   3 *
   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
  10 *
  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:
  16 *
  17 *     Redistribution and use in source and binary forms, with or
  18 *     without modification, are permitted provided that the following
  19 *     conditions are met:
  20 *
  21 *      - Redistributions of source code must retain the above
  22 *        copyright notice, this list of conditions and the following
  23 *        disclaimer.
  24 *
  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.
  29 *
  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
  37 * SOFTWARE.
  38 *
  39 * Credits:
  40 *      Christoph Hellwig
  41 *      FUJITA Tomonori
  42 *      Arne Redlich
  43 *      Zhenyu Wang
  44 * Modified by:
  45 *      Erez Zilber
  46 */
  47
  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>
  56#include <linux/in.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>
  62
  63#include <net/sock.h>
  64
  65#include <linux/uaccess.h>
  66
  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>
  74
  75#include "iscsi_iser.h"
  76
  77MODULE_DESCRIPTION("iSER (iSCSI Extensions for RDMA) Datamover");
  78MODULE_LICENSE("Dual BSD/GPL");
  79MODULE_AUTHOR("Alex Nezhinsky, Dan Bar Dov, Or Gerlitz");
  80
  81static struct scsi_host_template iscsi_iser_sht;
  82static struct iscsi_transport iscsi_iser_transport;
  83static struct scsi_transport_template *iscsi_iser_scsi_transport;
  84static struct workqueue_struct *release_wq;
  85static DEFINE_MUTEX(unbind_iser_conn_mutex);
  86struct iser_global ig;
  87
  88int iser_debug_level = 0;
  89module_param_named(debug_level, iser_debug_level, int, S_IRUGO | S_IWUSR);
  90MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0 (default:disabled)");
  91
  92static unsigned int iscsi_max_lun = 512;
  93module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
  94MODULE_PARM_DESC(max_lun, "Max LUNs to allow per session (default:512");
  95
  96unsigned int iser_max_sectors = ISER_DEF_MAX_SECTORS;
  97module_param_named(max_sectors, iser_max_sectors, uint, S_IRUGO | S_IWUSR);
  98MODULE_PARM_DESC(max_sectors, "Max number of sectors in a single scsi command (default:1024");
  99
 100bool iser_always_reg = true;
 101module_param_named(always_register, iser_always_reg, bool, S_IRUGO);
 102MODULE_PARM_DESC(always_register,
 103                 "Always register memory, even for continuous memory regions (default:true)");
 104
 105bool iser_pi_enable = false;
 106module_param_named(pi_enable, iser_pi_enable, bool, S_IRUGO);
 107MODULE_PARM_DESC(pi_enable, "Enable T10-PI offload support (default:disabled)");
 108
 109int iser_pi_guard;
 110module_param_named(pi_guard, iser_pi_guard, int, S_IRUGO);
 111MODULE_PARM_DESC(pi_guard, "T10-PI guard_type [deprecated]");
 112
 113/*
 114 * iscsi_iser_recv() - Process a successful recv completion
 115 * @conn:         iscsi connection
 116 * @hdr:          iscsi header
 117 * @rx_data:      buffer containing receive data payload
 118 * @rx_data_len:  length of rx_data
 119 *
 120 * Notes: In case of data length errors or iscsi PDU completion failures
 121 *        this routine will signal iscsi layer of connection failure.
 122 */
 123void
 124iscsi_iser_recv(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
 125                char *rx_data, int rx_data_len)
 126{
 127        int rc = 0;
 128        int datalen;
 129
 130        /* verify PDU length */
 131        datalen = ntoh24(hdr->dlength);
 132        if (datalen > rx_data_len || (datalen + 4) < rx_data_len) {
 133                iser_err("wrong datalen %d (hdr), %d (IB)\n",
 134                        datalen, rx_data_len);
 135                rc = ISCSI_ERR_DATALEN;
 136                goto error;
 137        }
 138
 139        if (datalen != rx_data_len)
 140                iser_dbg("aligned datalen (%d) hdr, %d (IB)\n",
 141                        datalen, rx_data_len);
 142
 143        rc = iscsi_complete_pdu(conn, hdr, rx_data, rx_data_len);
 144        if (rc && rc != ISCSI_ERR_NO_SCSI_CMD)
 145                goto error;
 146
 147        return;
 148error:
 149        iscsi_conn_failure(conn, rc);
 150}
 151
 152/**
 153 * iscsi_iser_pdu_alloc() - allocate an iscsi-iser PDU
 154 * @task:     iscsi task
 155 * @opcode:   iscsi command opcode
 156 *
 157 * Netes: This routine can't fail, just assign iscsi task
 158 *        hdr and max hdr size.
 159 */
 160static int
 161iscsi_iser_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
 162{
 163        struct iscsi_iser_task *iser_task = task->dd_data;
 164
 165        task->hdr = (struct iscsi_hdr *)&iser_task->desc.iscsi_header;
 166        task->hdr_max = sizeof(iser_task->desc.iscsi_header);
 167
 168        return 0;
 169}
 170
 171/**
 172 * iser_initialize_task_headers() - Initialize task headers
 173 * @task:       iscsi task
 174 * @tx_desc:    iser tx descriptor
 175 *
 176 * Notes:
 177 * This routine may race with iser teardown flow for scsi
 178 * error handling TMFs. So for TMF we should acquire the
 179 * state mutex to avoid dereferencing the IB device which
 180 * may have already been terminated.
 181 */
 182int
 183iser_initialize_task_headers(struct iscsi_task *task,
 184                             struct iser_tx_desc *tx_desc)
 185{
 186        struct iser_conn *iser_conn = task->conn->dd_data;
 187        struct iser_device *device = iser_conn->ib_conn.device;
 188        struct iscsi_iser_task *iser_task = task->dd_data;
 189        u64 dma_addr;
 190        const bool mgmt_task = !task->sc && !in_interrupt();
 191        int ret = 0;
 192
 193        if (unlikely(mgmt_task))
 194                mutex_lock(&iser_conn->state_mutex);
 195
 196        if (unlikely(iser_conn->state != ISER_CONN_UP)) {
 197                ret = -ENODEV;
 198                goto out;
 199        }
 200
 201        dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc,
 202                                ISER_HEADERS_LEN, DMA_TO_DEVICE);
 203        if (ib_dma_mapping_error(device->ib_device, dma_addr)) {
 204                ret = -ENOMEM;
 205                goto out;
 206        }
 207
 208        tx_desc->wr_idx = 0;
 209        tx_desc->mapped = true;
 210        tx_desc->dma_addr = dma_addr;
 211        tx_desc->tx_sg[0].addr   = tx_desc->dma_addr;
 212        tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
 213        tx_desc->tx_sg[0].lkey   = device->pd->local_dma_lkey;
 214
 215        iser_task->iser_conn = iser_conn;
 216out:
 217        if (unlikely(mgmt_task))
 218                mutex_unlock(&iser_conn->state_mutex);
 219
 220        return ret;
 221}
 222
 223/**
 224 * iscsi_iser_task_init() - Initialize iscsi-iser task
 225 * @task: iscsi task
 226 *
 227 * Initialize the task for the scsi command or mgmt command.
 228 *
 229 * Return: Returns zero on success or -ENOMEM when failing
 230 *         to init task headers (dma mapping error).
 231 */
 232static int
 233iscsi_iser_task_init(struct iscsi_task *task)
 234{
 235        struct iscsi_iser_task *iser_task = task->dd_data;
 236        int ret;
 237
 238        ret = iser_initialize_task_headers(task, &iser_task->desc);
 239        if (ret) {
 240                iser_err("Failed to init task %p, err = %d\n",
 241                         iser_task, ret);
 242                return ret;
 243        }
 244
 245        /* mgmt task */
 246        if (!task->sc)
 247                return 0;
 248
 249        iser_task->command_sent = 0;
 250        iser_task_rdma_init(iser_task);
 251        iser_task->sc = task->sc;
 252
 253        return 0;
 254}
 255
 256/**
 257 * iscsi_iser_mtask_xmit() - xmit management (immediate) task
 258 * @conn: iscsi connection
 259 * @task: task management task
 260 *
 261 * Notes:
 262 *      The function can return -EAGAIN in which case caller must
 263 *      call it again later, or recover. '0' return code means successful
 264 *      xmit.
 265 *
 266 **/
 267static int
 268iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
 269{
 270        int error = 0;
 271
 272        iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt);
 273
 274        error = iser_send_control(conn, task);
 275
 276        /* since iser xmits control with zero copy, tasks can not be recycled
 277         * right after sending them.
 278         * The recycling scheme is based on whether a response is expected
 279         * - if yes, the task is recycled at iscsi_complete_pdu
 280         * - if no,  the task is recycled at iser_snd_completion
 281         */
 282        return error;
 283}
 284
 285static int
 286iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn,
 287                                 struct iscsi_task *task)
 288{
 289        struct iscsi_r2t_info *r2t = &task->unsol_r2t;
 290        struct iscsi_data hdr;
 291        int error = 0;
 292
 293        /* Send data-out PDUs while there's still unsolicited data to send */
 294        while (iscsi_task_has_unsol_data(task)) {
 295                iscsi_prep_data_out_pdu(task, r2t, &hdr);
 296                iser_dbg("Sending data-out: itt 0x%x, data count %d\n",
 297                           hdr.itt, r2t->data_count);
 298
 299                /* the buffer description has been passed with the command */
 300                /* Send the command */
 301                error = iser_send_data_out(conn, task, &hdr);
 302                if (error) {
 303                        r2t->datasn--;
 304                        goto iscsi_iser_task_xmit_unsol_data_exit;
 305                }
 306                r2t->sent += r2t->data_count;
 307                iser_dbg("Need to send %d more as data-out PDUs\n",
 308                           r2t->data_length - r2t->sent);
 309        }
 310
 311iscsi_iser_task_xmit_unsol_data_exit:
 312        return error;
 313}
 314
 315/**
 316 * iscsi_iser_task_xmit() - xmit iscsi-iser task
 317 * @task: iscsi task
 318 *
 319 * Return: zero on success or escalates $error on failure.
 320 */
 321static int
 322iscsi_iser_task_xmit(struct iscsi_task *task)
 323{
 324        struct iscsi_conn *conn = task->conn;
 325        struct iscsi_iser_task *iser_task = task->dd_data;
 326        int error = 0;
 327
 328        if (!task->sc)
 329                return iscsi_iser_mtask_xmit(conn, task);
 330
 331        if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
 332                BUG_ON(scsi_bufflen(task->sc) == 0);
 333
 334                iser_dbg("cmd [itt %x total %d imm %d unsol_data %d\n",
 335                           task->itt, scsi_bufflen(task->sc),
 336                           task->imm_count, task->unsol_r2t.data_length);
 337        }
 338
 339        iser_dbg("ctask xmit [cid %d itt 0x%x]\n",
 340                   conn->id, task->itt);
 341
 342        /* Send the cmd PDU */
 343        if (!iser_task->command_sent) {
 344                error = iser_send_command(conn, task);
 345                if (error)
 346                        goto iscsi_iser_task_xmit_exit;
 347                iser_task->command_sent = 1;
 348        }
 349
 350        /* Send unsolicited data-out PDU(s) if necessary */
 351        if (iscsi_task_has_unsol_data(task))
 352                error = iscsi_iser_task_xmit_unsol_data(conn, task);
 353
 354 iscsi_iser_task_xmit_exit:
 355        return error;
 356}
 357
 358/**
 359 * iscsi_iser_cleanup_task() - cleanup an iscsi-iser task
 360 * @task: iscsi task
 361 *
 362 * Notes: In case the RDMA device is already NULL (might have
 363 *        been removed in DEVICE_REMOVAL CM event it will bail-out
 364 *        without doing dma unmapping.
 365 */
 366static void iscsi_iser_cleanup_task(struct iscsi_task *task)
 367{
 368        struct iscsi_iser_task *iser_task = task->dd_data;
 369        struct iser_tx_desc *tx_desc = &iser_task->desc;
 370        struct iser_conn *iser_conn = task->conn->dd_data;
 371        struct iser_device *device = iser_conn->ib_conn.device;
 372
 373        /* DEVICE_REMOVAL event might have already released the device */
 374        if (!device)
 375                return;
 376
 377        if (likely(tx_desc->mapped)) {
 378                ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr,
 379                                    ISER_HEADERS_LEN, DMA_TO_DEVICE);
 380                tx_desc->mapped = false;
 381        }
 382
 383        /* mgmt tasks do not need special cleanup */
 384        if (!task->sc)
 385                return;
 386
 387        if (iser_task->status == ISER_TASK_STATUS_STARTED) {
 388                iser_task->status = ISER_TASK_STATUS_COMPLETED;
 389                iser_task_rdma_finalize(iser_task);
 390        }
 391}
 392
 393/**
 394 * iscsi_iser_check_protection() - check protection information status of task.
 395 * @task:     iscsi task
 396 * @sector:   error sector if exsists (output)
 397 *
 398 * Return: zero if no data-integrity errors have occured
 399 *         0x1: data-integrity error occured in the guard-block
 400 *         0x2: data-integrity error occured in the reference tag
 401 *         0x3: data-integrity error occured in the application tag
 402 *
 403 *         In addition the error sector is marked.
 404 */
 405static u8
 406iscsi_iser_check_protection(struct iscsi_task *task, sector_t *sector)
 407{
 408        struct iscsi_iser_task *iser_task = task->dd_data;
 409
 410        if (iser_task->dir[ISER_DIR_IN])
 411                return iser_check_task_pi_status(iser_task, ISER_DIR_IN,
 412                                                 sector);
 413        else
 414                return iser_check_task_pi_status(iser_task, ISER_DIR_OUT,
 415                                                 sector);
 416}
 417
 418/**
 419 * iscsi_iser_conn_create() - create a new iscsi-iser connection
 420 * @cls_session: iscsi class connection
 421 * @conn_idx:    connection index within the session (for MCS)
 422 *
 423 * Return: iscsi_cls_conn when iscsi_conn_setup succeeds or NULL
 424 *         otherwise.
 425 */
 426static struct iscsi_cls_conn *
 427iscsi_iser_conn_create(struct iscsi_cls_session *cls_session,
 428                       uint32_t conn_idx)
 429{
 430        struct iscsi_conn *conn;
 431        struct iscsi_cls_conn *cls_conn;
 432
 433        cls_conn = iscsi_conn_setup(cls_session, 0, conn_idx);
 434        if (!cls_conn)
 435                return NULL;
 436        conn = cls_conn->dd_data;
 437
 438        /*
 439         * due to issues with the login code re iser sematics
 440         * this not set in iscsi_conn_setup - FIXME
 441         */
 442        conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN;
 443
 444        return cls_conn;
 445}
 446
 447/**
 448 * iscsi_iser_conn_bind() - bind iscsi and iser connection structures
 449 * @cls_session:     iscsi class session
 450 * @cls_conn:        iscsi class connection
 451 * @transport_eph:   transport end-point handle
 452 * @is_leading:      indicate if this is the session leading connection (MCS)
 453 *
 454 * Return: zero on success, $error if iscsi_conn_bind fails and
 455 *         -EINVAL in case end-point doesn't exsits anymore or iser connection
 456 *         state is not UP (teardown already started).
 457 */
 458static int
 459iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
 460                     struct iscsi_cls_conn *cls_conn,
 461                     uint64_t transport_eph,
 462                     int is_leading)
 463{
 464        struct iscsi_conn *conn = cls_conn->dd_data;
 465        struct iser_conn *iser_conn;
 466        struct iscsi_endpoint *ep;
 467        int error;
 468
 469        error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
 470        if (error)
 471                return error;
 472
 473        /* the transport ep handle comes from user space so it must be
 474         * verified against the global ib connections list */
 475        ep = iscsi_lookup_endpoint(transport_eph);
 476        if (!ep) {
 477                iser_err("can't bind eph %llx\n",
 478                         (unsigned long long)transport_eph);
 479                return -EINVAL;
 480        }
 481        iser_conn = ep->dd_data;
 482
 483        mutex_lock(&iser_conn->state_mutex);
 484        if (iser_conn->state != ISER_CONN_UP) {
 485                error = -EINVAL;
 486                iser_err("iser_conn %p state is %d, teardown started\n",
 487                         iser_conn, iser_conn->state);
 488                goto out;
 489        }
 490
 491        error = iser_alloc_rx_descriptors(iser_conn, conn->session);
 492        if (error)
 493                goto out;
 494
 495        /* binds the iSER connection retrieved from the previously
 496         * connected ep_handle to the iSCSI layer connection. exchanges
 497         * connection pointers */
 498        iser_info("binding iscsi conn %p to iser_conn %p\n", conn, iser_conn);
 499
 500        conn->dd_data = iser_conn;
 501        iser_conn->iscsi_conn = conn;
 502
 503out:
 504        mutex_unlock(&iser_conn->state_mutex);
 505        return error;
 506}
 507
 508/**
 509 * iscsi_iser_conn_start() - start iscsi-iser connection
 510 * @cls_conn: iscsi class connection
 511 *
 512 * Notes: Here iser intialize (or re-initialize) stop_completion as
 513 *        from this point iscsi must call conn_stop in session/connection
 514 *        teardown so iser transport must wait for it.
 515 */
 516static int
 517iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
 518{
 519        struct iscsi_conn *iscsi_conn;
 520        struct iser_conn *iser_conn;
 521
 522        iscsi_conn = cls_conn->dd_data;
 523        iser_conn = iscsi_conn->dd_data;
 524        reinit_completion(&iser_conn->stop_completion);
 525
 526        return iscsi_conn_start(cls_conn);
 527}
 528
 529/**
 530 * iscsi_iser_conn_stop() - stop iscsi-iser connection
 531 * @cls_conn:  iscsi class connection
 532 * @flag:      indicate if recover or terminate (passed as is)
 533 *
 534 * Notes: Calling iscsi_conn_stop might theoretically race with
 535 *        DEVICE_REMOVAL event and dereference a previously freed RDMA device
 536 *        handle, so we call it under iser the state lock to protect against
 537 *        this kind of race.
 538 */
 539static void
 540iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
 541{
 542        struct iscsi_conn *conn = cls_conn->dd_data;
 543        struct iser_conn *iser_conn = conn->dd_data;
 544
 545        iser_info("stopping iscsi_conn: %p, iser_conn: %p\n", conn, iser_conn);
 546
 547        /*
 548         * Userspace may have goofed up and not bound the connection or
 549         * might have only partially setup the connection.
 550         */
 551        if (iser_conn) {
 552                mutex_lock(&iser_conn->state_mutex);
 553                mutex_lock(&unbind_iser_conn_mutex);
 554                iser_conn_terminate(iser_conn);
 555                iscsi_conn_stop(cls_conn, flag);
 556
 557                /* unbind */
 558                iser_conn->iscsi_conn = NULL;
 559                conn->dd_data = NULL;
 560                mutex_unlock(&unbind_iser_conn_mutex);
 561
 562                complete(&iser_conn->stop_completion);
 563                mutex_unlock(&iser_conn->state_mutex);
 564        } else {
 565                iscsi_conn_stop(cls_conn, flag);
 566        }
 567}
 568
 569/**
 570 * iscsi_iser_session_destroy() - destroy iscsi-iser session
 571 * @cls_session: iscsi class session
 572 *
 573 * Removes and free iscsi host.
 574 */
 575static void
 576iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session)
 577{
 578        struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
 579
 580        iscsi_session_teardown(cls_session);
 581        iscsi_host_remove(shost);
 582        iscsi_host_free(shost);
 583}
 584
 585static inline unsigned int
 586iser_dif_prot_caps(int prot_caps)
 587{
 588        return ((prot_caps & IB_PROT_T10DIF_TYPE_1) ?
 589                SHOST_DIF_TYPE1_PROTECTION | SHOST_DIX_TYPE0_PROTECTION |
 590                SHOST_DIX_TYPE1_PROTECTION : 0) |
 591               ((prot_caps & IB_PROT_T10DIF_TYPE_2) ?
 592                SHOST_DIF_TYPE2_PROTECTION | SHOST_DIX_TYPE2_PROTECTION : 0) |
 593               ((prot_caps & IB_PROT_T10DIF_TYPE_3) ?
 594                SHOST_DIF_TYPE3_PROTECTION | SHOST_DIX_TYPE3_PROTECTION : 0);
 595}
 596
 597/**
 598 * iscsi_iser_session_create() - create an iscsi-iser session
 599 * @ep:             iscsi end-point handle
 600 * @cmds_max:       maximum commands in this session
 601 * @qdepth:         session command queue depth
 602 * @initial_cmdsn:  initiator command sequnce number
 603 *
 604 * Allocates and adds a scsi host, expose DIF supprot if
 605 * exists, and sets up an iscsi session.
 606 */
 607static struct iscsi_cls_session *
 608iscsi_iser_session_create(struct iscsi_endpoint *ep,
 609                          uint16_t cmds_max, uint16_t qdepth,
 610                          uint32_t initial_cmdsn)
 611{
 612        struct iscsi_cls_session *cls_session;
 613        struct iscsi_session *session;
 614        struct Scsi_Host *shost;
 615        struct iser_conn *iser_conn = NULL;
 616        struct ib_conn *ib_conn;
 617        u32 max_fr_sectors;
 618        u16 max_cmds;
 619
 620        shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0);
 621        if (!shost)
 622                return NULL;
 623        shost->transportt = iscsi_iser_scsi_transport;
 624        shost->cmd_per_lun = qdepth;
 625        shost->max_lun = iscsi_max_lun;
 626        shost->max_id = 0;
 627        shost->max_channel = 0;
 628        shost->max_cmd_len = 16;
 629
 630        /*
 631         * older userspace tools (before 2.0-870) did not pass us
 632         * the leading conn's ep so this will be NULL;
 633         */
 634        if (ep) {
 635                iser_conn = ep->dd_data;
 636                max_cmds = iser_conn->max_cmds;
 637                shost->sg_tablesize = iser_conn->scsi_sg_tablesize;
 638
 639                mutex_lock(&iser_conn->state_mutex);
 640                if (iser_conn->state != ISER_CONN_UP) {
 641                        iser_err("iser conn %p already started teardown\n",
 642                                 iser_conn);
 643                        mutex_unlock(&iser_conn->state_mutex);
 644                        goto free_host;
 645                }
 646
 647                ib_conn = &iser_conn->ib_conn;
 648                if (ib_conn->pi_support) {
 649                        u32 sig_caps = ib_conn->device->ib_device->attrs.sig_prot_cap;
 650
 651                        scsi_host_set_prot(shost, iser_dif_prot_caps(sig_caps));
 652                        scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP |
 653                                                   SHOST_DIX_GUARD_CRC);
 654                }
 655
 656                if (iscsi_host_add(shost,
 657                                   ib_conn->device->ib_device->dev.parent)) {
 658                        mutex_unlock(&iser_conn->state_mutex);
 659                        goto free_host;
 660                }
 661                mutex_unlock(&iser_conn->state_mutex);
 662        } else {
 663                max_cmds = ISER_DEF_XMIT_CMDS_MAX;
 664                if (iscsi_host_add(shost, NULL))
 665                        goto free_host;
 666        }
 667
 668        max_fr_sectors = (shost->sg_tablesize * PAGE_SIZE) >> 9;
 669        shost->max_sectors = min(iser_max_sectors, max_fr_sectors);
 670
 671        iser_dbg("iser_conn %p, sg_tablesize %u, max_sectors %u\n",
 672                 iser_conn, shost->sg_tablesize,
 673                 shost->max_sectors);
 674
 675        if (shost->max_sectors < iser_max_sectors)
 676                iser_warn("max_sectors was reduced from %u to %u\n",
 677                          iser_max_sectors, shost->max_sectors);
 678
 679        if (cmds_max > max_cmds) {
 680                iser_info("cmds_max changed from %u to %u\n",
 681                          cmds_max, max_cmds);
 682                cmds_max = max_cmds;
 683        }
 684
 685        cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
 686                                          cmds_max, 0,
 687                                          sizeof(struct iscsi_iser_task),
 688                                          initial_cmdsn, 0);
 689        if (!cls_session)
 690                goto remove_host;
 691        session = cls_session->dd_data;
 692
 693        shost->can_queue = session->scsi_cmds_max;
 694        return cls_session;
 695
 696remove_host:
 697        iscsi_host_remove(shost);
 698free_host:
 699        iscsi_host_free(shost);
 700        return NULL;
 701}
 702
 703static int
 704iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
 705                     enum iscsi_param param, char *buf, int buflen)
 706{
 707        int value;
 708
 709        switch (param) {
 710        case ISCSI_PARAM_MAX_RECV_DLENGTH:
 711                /* TBD */
 712                break;
 713        case ISCSI_PARAM_HDRDGST_EN:
 714                sscanf(buf, "%d", &value);
 715                if (value) {
 716                        iser_err("DataDigest wasn't negotiated to None\n");
 717                        return -EPROTO;
 718                }
 719                break;
 720        case ISCSI_PARAM_DATADGST_EN:
 721                sscanf(buf, "%d", &value);
 722                if (value) {
 723                        iser_err("DataDigest wasn't negotiated to None\n");
 724                        return -EPROTO;
 725                }
 726                break;
 727        case ISCSI_PARAM_IFMARKER_EN:
 728                sscanf(buf, "%d", &value);
 729                if (value) {
 730                        iser_err("IFMarker wasn't negotiated to No\n");
 731                        return -EPROTO;
 732                }
 733                break;
 734        case ISCSI_PARAM_OFMARKER_EN:
 735                sscanf(buf, "%d", &value);
 736                if (value) {
 737                        iser_err("OFMarker wasn't negotiated to No\n");
 738                        return -EPROTO;
 739                }
 740                break;
 741        default:
 742                return iscsi_set_param(cls_conn, param, buf, buflen);
 743        }
 744
 745        return 0;
 746}
 747
 748/**
 749 * iscsi_iser_set_param() - set class connection parameter
 750 * @cls_conn:    iscsi class connection
 751 * @stats:       iscsi stats to output
 752 *
 753 * Output connection statistics.
 754 */
 755static void
 756iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
 757{
 758        struct iscsi_conn *conn = cls_conn->dd_data;
 759
 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;
 770}
 771
 772static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep,
 773                                   enum iscsi_param param, char *buf)
 774{
 775        struct iser_conn *iser_conn = ep->dd_data;
 776        int len;
 777
 778        switch (param) {
 779        case ISCSI_PARAM_CONN_PORT:
 780        case ISCSI_PARAM_CONN_ADDRESS:
 781                if (!iser_conn || !iser_conn->ib_conn.cma_id)
 782                        return -ENOTCONN;
 783
 784                return iscsi_conn_get_addr_param((struct sockaddr_storage *)
 785                                &iser_conn->ib_conn.cma_id->route.addr.dst_addr,
 786                                param, buf);
 787                break;
 788        default:
 789                return -ENOSYS;
 790        }
 791
 792        return len;
 793}
 794
 795/**
 796 * iscsi_iser_ep_connect() - Initiate iSER connection establishment
 797 * @shost:          scsi_host
 798 * @dst_addr:       destination address
 799 * @non-blocking:   indicate if routine can block
 800 *
 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.
 806 *
 807 * Return: iscsi_endpoint created by iscsi layer or ERR_PTR(error)
 808 *         if fails.
 809 */
 810static struct iscsi_endpoint *
 811iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr,
 812                      int non_blocking)
 813{
 814        int err;
 815        struct iser_conn *iser_conn;
 816        struct iscsi_endpoint *ep;
 817
 818        ep = iscsi_create_endpoint(0);
 819        if (!ep)
 820                return ERR_PTR(-ENOMEM);
 821
 822        iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL);
 823        if (!iser_conn) {
 824                err = -ENOMEM;
 825                goto failure;
 826        }
 827
 828        ep->dd_data = iser_conn;
 829        iser_conn->ep = ep;
 830        iser_conn_init(iser_conn);
 831
 832        err = iser_connect(iser_conn, NULL, dst_addr, non_blocking);
 833        if (err)
 834                goto failure;
 835
 836        return ep;
 837failure:
 838        iscsi_destroy_endpoint(ep);
 839        return ERR_PTR(err);
 840}
 841
 842/**
 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.
 846 *
 847 * This routine boils down to waiting for up_completion signaling
 848 * that cma_id got CONNECTED event.
 849 *
 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.
 854 */
 855static int
 856iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
 857{
 858        struct iser_conn *iser_conn = ep->dd_data;
 859        int rc;
 860
 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 */
 864        if (rc == 0) {
 865                mutex_lock(&iser_conn->state_mutex);
 866                if (iser_conn->state == ISER_CONN_TERMINATING ||
 867                    iser_conn->state == ISER_CONN_DOWN)
 868                        rc = -1;
 869                mutex_unlock(&iser_conn->state_mutex);
 870        }
 871
 872        iser_info("iser conn %p rc = %d\n", iser_conn, rc);
 873
 874        if (rc > 0)
 875                return 1; /* success, this is the equivalent of EPOLLOUT */
 876        else if (!rc)
 877                return 0; /* timeout */
 878        else
 879                return rc; /* signal */
 880}
 881
 882/**
 883 * iscsi_iser_ep_disconnect() - Initiate connection teardown process
 884 * @ep:    iscsi endpoint handle
 885 *
 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.
 890 */
 891static void
 892iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep)
 893{
 894        struct iser_conn *iser_conn = ep->dd_data;
 895
 896        iser_info("ep %p iser conn %p\n", ep, iser_conn);
 897
 898        mutex_lock(&iser_conn->state_mutex);
 899        iser_conn_terminate(iser_conn);
 900
 901        /*
 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
 905         * immediately.
 906         */
 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);
 911        } else {
 912                iser_conn->state = ISER_CONN_DOWN;
 913                mutex_unlock(&iser_conn->state_mutex);
 914                iser_conn_release(iser_conn);
 915        }
 916
 917        iscsi_destroy_endpoint(ep);
 918}
 919
 920static umode_t iser_attr_is_visible(int param_type, int param)
 921{
 922        switch (param_type) {
 923        case ISCSI_HOST_PARAM:
 924                switch (param) {
 925                case ISCSI_HOST_PARAM_NETDEV_NAME:
 926                case ISCSI_HOST_PARAM_HWADDRESS:
 927                case ISCSI_HOST_PARAM_INITIATOR_NAME:
 928                        return S_IRUGO;
 929                default:
 930                        return 0;
 931                }
 932        case ISCSI_PARAM:
 933                switch (param) {
 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:
 965                        return S_IRUGO;
 966                default:
 967                        return 0;
 968                }
 969        }
 970
 971        return 0;
 972}
 973
 974static int iscsi_iser_slave_alloc(struct scsi_device *sdev)
 975{
 976        struct iscsi_session *session;
 977        struct iser_conn *iser_conn;
 978        struct ib_device *ib_dev;
 979
 980        mutex_lock(&unbind_iser_conn_mutex);
 981
 982        session = starget_to_session(scsi_target(sdev))->dd_data;
 983        iser_conn = session->leadconn->dd_data;
 984        if (!iser_conn) {
 985                mutex_unlock(&unbind_iser_conn_mutex);
 986                return -ENOTCONN;
 987        }
 988        ib_dev = iser_conn->ib_conn.device->ib_device;
 989
 990        if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG))
 991                blk_queue_virt_boundary(sdev->request_queue, ~MASK_4K);
 992
 993        mutex_unlock(&unbind_iser_conn_mutex);
 994
 995        return 0;
 996}
 997
 998static 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_timed_out           = iscsi_eh_cmd_timed_out,
1006        .eh_abort_handler       = iscsi_eh_abort,
1007        .eh_device_reset_handler= iscsi_eh_device_reset,
1008        .eh_target_reset_handler = iscsi_eh_recover_target,
1009        .target_alloc           = iscsi_target_alloc,
1010        .use_clustering         = ENABLE_CLUSTERING,
1011        .slave_alloc            = iscsi_iser_slave_alloc,
1012        .proc_name              = "iscsi_iser",
1013        .this_id                = -1,
1014        .track_queue_depth      = 1,
1015};
1016
1017static struct iscsi_transport iscsi_iser_transport = {
1018        .owner                  = THIS_MODULE,
1019        .name                   = "iser",
1020        .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO,
1021        /* session management */
1022        .create_session         = iscsi_iser_session_create,
1023        .destroy_session        = iscsi_iser_session_destroy,
1024        /* connection management */
1025        .create_conn            = iscsi_iser_conn_create,
1026        .bind_conn              = iscsi_iser_conn_bind,
1027        .destroy_conn           = iscsi_conn_teardown,
1028        .attr_is_visible        = iser_attr_is_visible,
1029        .set_param              = iscsi_iser_set_param,
1030        .get_conn_param         = iscsi_conn_get_param,
1031        .get_ep_param           = iscsi_iser_get_ep_param,
1032        .get_session_param      = iscsi_session_get_param,
1033        .start_conn             = iscsi_iser_conn_start,
1034        .stop_conn              = iscsi_iser_conn_stop,
1035        /* iscsi host params */
1036        .get_host_param         = iscsi_host_get_param,
1037        .set_host_param         = iscsi_host_set_param,
1038        /* IO */
1039        .send_pdu               = iscsi_conn_send_pdu,
1040        .get_stats              = iscsi_iser_conn_get_stats,
1041        .init_task              = iscsi_iser_task_init,
1042        .xmit_task              = iscsi_iser_task_xmit,
1043        .cleanup_task           = iscsi_iser_cleanup_task,
1044        .alloc_pdu              = iscsi_iser_pdu_alloc,
1045        .check_protection       = iscsi_iser_check_protection,
1046        /* recovery */
1047        .session_recovery_timedout = iscsi_session_recovery_timedout,
1048
1049        .ep_connect             = iscsi_iser_ep_connect,
1050        .ep_poll                = iscsi_iser_ep_poll,
1051        .ep_disconnect          = iscsi_iser_ep_disconnect
1052};
1053
1054static int __init iser_init(void)
1055{
1056        int err;
1057
1058        iser_dbg("Starting iSER datamover...\n");
1059
1060        if (iscsi_max_lun < 1) {
1061                iser_err("Invalid max_lun value of %u\n", iscsi_max_lun);
1062                return -EINVAL;
1063        }
1064
1065        memset(&ig, 0, sizeof(struct iser_global));
1066
1067        ig.desc_cache = kmem_cache_create("iser_descriptors",
1068                                          sizeof(struct iser_tx_desc),
1069                                          0, SLAB_HWCACHE_ALIGN,
1070                                          NULL);
1071        if (ig.desc_cache == NULL)
1072                return -ENOMEM;
1073
1074        /* device init is called only after the first addr resolution */
1075        mutex_init(&ig.device_list_mutex);
1076        INIT_LIST_HEAD(&ig.device_list);
1077        mutex_init(&ig.connlist_mutex);
1078        INIT_LIST_HEAD(&ig.connlist);
1079
1080        release_wq = alloc_workqueue("release workqueue", 0, 0);
1081        if (!release_wq) {
1082                iser_err("failed to allocate release workqueue\n");
1083                err = -ENOMEM;
1084                goto err_alloc_wq;
1085        }
1086
1087        iscsi_iser_scsi_transport = iscsi_register_transport(
1088                                                        &iscsi_iser_transport);
1089        if (!iscsi_iser_scsi_transport) {
1090                iser_err("iscsi_register_transport failed\n");
1091                err = -EINVAL;
1092                goto err_reg;
1093        }
1094
1095        return 0;
1096
1097err_reg:
1098        destroy_workqueue(release_wq);
1099err_alloc_wq:
1100        kmem_cache_destroy(ig.desc_cache);
1101
1102        return err;
1103}
1104
1105static void __exit iser_exit(void)
1106{
1107        struct iser_conn *iser_conn, *n;
1108        int connlist_empty;
1109
1110        iser_dbg("Removing iSER datamover...\n");
1111        destroy_workqueue(release_wq);
1112
1113        mutex_lock(&ig.connlist_mutex);
1114        connlist_empty = list_empty(&ig.connlist);
1115        mutex_unlock(&ig.connlist_mutex);
1116
1117        if (!connlist_empty) {
1118                iser_err("Error cleanup stage completed but we still have iser "
1119                         "connections, destroying them anyway\n");
1120                list_for_each_entry_safe(iser_conn, n, &ig.connlist,
1121                                         conn_list) {
1122                        iser_conn_release(iser_conn);
1123                }
1124        }
1125
1126        iscsi_unregister_transport(&iscsi_iser_transport);
1127        kmem_cache_destroy(ig.desc_cache);
1128}
1129
1130module_init(iser_init);
1131module_exit(iser_exit);
1132