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->inv_wr.next = NULL;
 209        tx_desc->reg_wr.wr.next = NULL;
 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;
 215
 216        iser_task->iser_conn = iser_conn;
 217out:
 218        if (unlikely(mgmt_task))
 219                mutex_unlock(&iser_conn->state_mutex);
 220
 221        return ret;
 222}
 223
 224/**
 225 * iscsi_iser_task_init() - Initialize iscsi-iser task
 226 * @task: iscsi task
 227 *
 228 * Initialize the task for the scsi command or mgmt command.
 229 *
 230 * Return: Returns zero on success or -ENOMEM when failing
 231 *         to init task headers (dma mapping error).
 232 */
 233static int
 234iscsi_iser_task_init(struct iscsi_task *task)
 235{
 236        struct iscsi_iser_task *iser_task = task->dd_data;
 237        int ret;
 238
 239        ret = iser_initialize_task_headers(task, &iser_task->desc);
 240        if (ret) {
 241                iser_err("Failed to init task %p, err = %d\n",
 242                         iser_task, ret);
 243                return ret;
 244        }
 245
 246        /* mgmt task */
 247        if (!task->sc)
 248                return 0;
 249
 250        iser_task->command_sent = 0;
 251        iser_task_rdma_init(iser_task);
 252        iser_task->sc = task->sc;
 253
 254        return 0;
 255}
 256
 257/**
 258 * iscsi_iser_mtask_xmit() - xmit management (immediate) task
 259 * @conn: iscsi connection
 260 * @task: task management task
 261 *
 262 * Notes:
 263 *      The function can return -EAGAIN in which case caller must
 264 *      call it again later, or recover. '0' return code means successful
 265 *      xmit.
 266 *
 267 **/
 268static int
 269iscsi_iser_mtask_xmit(struct iscsi_conn *conn, struct iscsi_task *task)
 270{
 271        int error = 0;
 272
 273        iser_dbg("mtask xmit [cid %d itt 0x%x]\n", conn->id, task->itt);
 274
 275        error = iser_send_control(conn, task);
 276
 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
 282         */
 283        return error;
 284}
 285
 286static int
 287iscsi_iser_task_xmit_unsol_data(struct iscsi_conn *conn,
 288                                 struct iscsi_task *task)
 289{
 290        struct iscsi_r2t_info *r2t = &task->unsol_r2t;
 291        struct iscsi_data hdr;
 292        int error = 0;
 293
 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);
 299
 300                /* the buffer description has been passed with the command */
 301                /* Send the command */
 302                error = iser_send_data_out(conn, task, &hdr);
 303                if (error) {
 304                        r2t->datasn--;
 305                        goto iscsi_iser_task_xmit_unsol_data_exit;
 306                }
 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);
 310        }
 311
 312iscsi_iser_task_xmit_unsol_data_exit:
 313        return error;
 314}
 315
 316/**
 317 * iscsi_iser_task_xmit() - xmit iscsi-iser task
 318 * @task: iscsi task
 319 *
 320 * Return: zero on success or escalates $error on failure.
 321 */
 322static int
 323iscsi_iser_task_xmit(struct iscsi_task *task)
 324{
 325        struct iscsi_conn *conn = task->conn;
 326        struct iscsi_iser_task *iser_task = task->dd_data;
 327        int error = 0;
 328
 329        if (!task->sc)
 330                return iscsi_iser_mtask_xmit(conn, task);
 331
 332        if (task->sc->sc_data_direction == DMA_TO_DEVICE) {
 333                BUG_ON(scsi_bufflen(task->sc) == 0);
 334
 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);
 338        }
 339
 340        iser_dbg("ctask xmit [cid %d itt 0x%x]\n",
 341                   conn->id, task->itt);
 342
 343        /* Send the cmd PDU */
 344        if (!iser_task->command_sent) {
 345                error = iser_send_command(conn, task);
 346                if (error)
 347                        goto iscsi_iser_task_xmit_exit;
 348                iser_task->command_sent = 1;
 349        }
 350
 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);
 354
 355 iscsi_iser_task_xmit_exit:
 356        return error;
 357}
 358
 359/**
 360 * iscsi_iser_cleanup_task() - cleanup an iscsi-iser task
 361 * @task: iscsi task
 362 *
 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.
 366 */
 367static void iscsi_iser_cleanup_task(struct iscsi_task *task)
 368{
 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;
 373
 374        /* DEVICE_REMOVAL event might have already released the device */
 375        if (!device)
 376                return;
 377
 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;
 382        }
 383
 384        /* mgmt tasks do not need special cleanup */
 385        if (!task->sc)
 386                return;
 387
 388        if (iser_task->status == ISER_TASK_STATUS_STARTED) {
 389                iser_task->status = ISER_TASK_STATUS_COMPLETED;
 390                iser_task_rdma_finalize(iser_task);
 391        }
 392}
 393
 394/**
 395 * iscsi_iser_check_protection() - check protection information status of task.
 396 * @task:     iscsi task
 397 * @sector:   error sector if exsists (output)
 398 *
 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
 403 *
 404 *         In addition the error sector is marked.
 405 */
 406static u8
 407iscsi_iser_check_protection(struct iscsi_task *task, sector_t *sector)
 408{
 409        struct iscsi_iser_task *iser_task = task->dd_data;
 410        enum iser_data_dir dir = iser_task->dir[ISER_DIR_IN] ?
 411                                        ISER_DIR_IN : ISER_DIR_OUT;
 412
 413        return iser_check_task_pi_status(iser_task, dir, sector);
 414}
 415
 416/**
 417 * iscsi_iser_conn_create() - create a new iscsi-iser connection
 418 * @cls_session: iscsi class connection
 419 * @conn_idx:    connection index within the session (for MCS)
 420 *
 421 * Return: iscsi_cls_conn when iscsi_conn_setup succeeds or NULL
 422 *         otherwise.
 423 */
 424static struct iscsi_cls_conn *
 425iscsi_iser_conn_create(struct iscsi_cls_session *cls_session,
 426                       uint32_t conn_idx)
 427{
 428        struct iscsi_conn *conn;
 429        struct iscsi_cls_conn *cls_conn;
 430
 431        cls_conn = iscsi_conn_setup(cls_session, 0, conn_idx);
 432        if (!cls_conn)
 433                return NULL;
 434        conn = cls_conn->dd_data;
 435
 436        /*
 437         * due to issues with the login code re iser sematics
 438         * this not set in iscsi_conn_setup - FIXME
 439         */
 440        conn->max_recv_dlength = ISER_RECV_DATA_SEG_LEN;
 441
 442        return cls_conn;
 443}
 444
 445/**
 446 * iscsi_iser_conn_bind() - bind iscsi and iser connection structures
 447 * @cls_session:     iscsi class session
 448 * @cls_conn:        iscsi class connection
 449 * @transport_eph:   transport end-point handle
 450 * @is_leading:      indicate if this is the session leading connection (MCS)
 451 *
 452 * Return: zero on success, $error if iscsi_conn_bind fails and
 453 *         -EINVAL in case end-point doesn't exsits anymore or iser connection
 454 *         state is not UP (teardown already started).
 455 */
 456static int
 457iscsi_iser_conn_bind(struct iscsi_cls_session *cls_session,
 458                     struct iscsi_cls_conn *cls_conn,
 459                     uint64_t transport_eph,
 460                     int is_leading)
 461{
 462        struct iscsi_conn *conn = cls_conn->dd_data;
 463        struct iser_conn *iser_conn;
 464        struct iscsi_endpoint *ep;
 465        int error;
 466
 467        error = iscsi_conn_bind(cls_session, cls_conn, is_leading);
 468        if (error)
 469                return error;
 470
 471        /* the transport ep handle comes from user space so it must be
 472         * verified against the global ib connections list */
 473        ep = iscsi_lookup_endpoint(transport_eph);
 474        if (!ep) {
 475                iser_err("can't bind eph %llx\n",
 476                         (unsigned long long)transport_eph);
 477                return -EINVAL;
 478        }
 479        iser_conn = ep->dd_data;
 480
 481        mutex_lock(&iser_conn->state_mutex);
 482        if (iser_conn->state != ISER_CONN_UP) {
 483                error = -EINVAL;
 484                iser_err("iser_conn %p state is %d, teardown started\n",
 485                         iser_conn, iser_conn->state);
 486                goto out;
 487        }
 488
 489        error = iser_alloc_rx_descriptors(iser_conn, conn->session);
 490        if (error)
 491                goto out;
 492
 493        /* binds the iSER connection retrieved from the previously
 494         * connected ep_handle to the iSCSI layer connection. exchanges
 495         * connection pointers */
 496        iser_info("binding iscsi conn %p to iser_conn %p\n", conn, iser_conn);
 497
 498        conn->dd_data = iser_conn;
 499        iser_conn->iscsi_conn = conn;
 500
 501out:
 502        mutex_unlock(&iser_conn->state_mutex);
 503        return error;
 504}
 505
 506/**
 507 * iscsi_iser_conn_start() - start iscsi-iser connection
 508 * @cls_conn: iscsi class connection
 509 *
 510 * Notes: Here iser intialize (or re-initialize) stop_completion as
 511 *        from this point iscsi must call conn_stop in session/connection
 512 *        teardown so iser transport must wait for it.
 513 */
 514static int
 515iscsi_iser_conn_start(struct iscsi_cls_conn *cls_conn)
 516{
 517        struct iscsi_conn *iscsi_conn;
 518        struct iser_conn *iser_conn;
 519
 520        iscsi_conn = cls_conn->dd_data;
 521        iser_conn = iscsi_conn->dd_data;
 522        reinit_completion(&iser_conn->stop_completion);
 523
 524        return iscsi_conn_start(cls_conn);
 525}
 526
 527/**
 528 * iscsi_iser_conn_stop() - stop iscsi-iser connection
 529 * @cls_conn:  iscsi class connection
 530 * @flag:      indicate if recover or terminate (passed as is)
 531 *
 532 * Notes: Calling iscsi_conn_stop might theoretically race with
 533 *        DEVICE_REMOVAL event and dereference a previously freed RDMA device
 534 *        handle, so we call it under iser the state lock to protect against
 535 *        this kind of race.
 536 */
 537static void
 538iscsi_iser_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
 539{
 540        struct iscsi_conn *conn = cls_conn->dd_data;
 541        struct iser_conn *iser_conn = conn->dd_data;
 542
 543        iser_info("stopping iscsi_conn: %p, iser_conn: %p\n", conn, iser_conn);
 544
 545        /*
 546         * Userspace may have goofed up and not bound the connection or
 547         * might have only partially setup the connection.
 548         */
 549        if (iser_conn) {
 550                mutex_lock(&iser_conn->state_mutex);
 551                mutex_lock(&unbind_iser_conn_mutex);
 552                iser_conn_terminate(iser_conn);
 553                iscsi_conn_stop(cls_conn, flag);
 554
 555                /* unbind */
 556                iser_conn->iscsi_conn = NULL;
 557                conn->dd_data = NULL;
 558                mutex_unlock(&unbind_iser_conn_mutex);
 559
 560                complete(&iser_conn->stop_completion);
 561                mutex_unlock(&iser_conn->state_mutex);
 562        } else {
 563                iscsi_conn_stop(cls_conn, flag);
 564        }
 565}
 566
 567/**
 568 * iscsi_iser_session_destroy() - destroy iscsi-iser session
 569 * @cls_session: iscsi class session
 570 *
 571 * Removes and free iscsi host.
 572 */
 573static void
 574iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session)
 575{
 576        struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
 577
 578        iscsi_session_teardown(cls_session);
 579        iscsi_host_remove(shost);
 580        iscsi_host_free(shost);
 581}
 582
 583static inline unsigned int
 584iser_dif_prot_caps(int prot_caps)
 585{
 586        return ((prot_caps & IB_PROT_T10DIF_TYPE_1) ?
 587                SHOST_DIF_TYPE1_PROTECTION | SHOST_DIX_TYPE0_PROTECTION |
 588                SHOST_DIX_TYPE1_PROTECTION : 0) |
 589               ((prot_caps & IB_PROT_T10DIF_TYPE_2) ?
 590                SHOST_DIF_TYPE2_PROTECTION | SHOST_DIX_TYPE2_PROTECTION : 0) |
 591               ((prot_caps & IB_PROT_T10DIF_TYPE_3) ?
 592                SHOST_DIF_TYPE3_PROTECTION | SHOST_DIX_TYPE3_PROTECTION : 0);
 593}
 594
 595/**
 596 * iscsi_iser_session_create() - create an iscsi-iser session
 597 * @ep:             iscsi end-point handle
 598 * @cmds_max:       maximum commands in this session
 599 * @qdepth:         session command queue depth
 600 * @initial_cmdsn:  initiator command sequnce number
 601 *
 602 * Allocates and adds a scsi host, expose DIF supprot if
 603 * exists, and sets up an iscsi session.
 604 */
 605static struct iscsi_cls_session *
 606iscsi_iser_session_create(struct iscsi_endpoint *ep,
 607                          uint16_t cmds_max, uint16_t qdepth,
 608                          uint32_t initial_cmdsn)
 609{
 610        struct iscsi_cls_session *cls_session;
 611        struct Scsi_Host *shost;
 612        struct iser_conn *iser_conn = NULL;
 613        struct ib_conn *ib_conn;
 614        struct ib_device *ib_dev;
 615        u32 max_fr_sectors;
 616
 617        shost = iscsi_host_alloc(&iscsi_iser_sht, 0, 0);
 618        if (!shost)
 619                return NULL;
 620        shost->transportt = iscsi_iser_scsi_transport;
 621        shost->cmd_per_lun = qdepth;
 622        shost->max_lun = iscsi_max_lun;
 623        shost->max_id = 0;
 624        shost->max_channel = 0;
 625        shost->max_cmd_len = 16;
 626
 627        /*
 628         * older userspace tools (before 2.0-870) did not pass us
 629         * the leading conn's ep so this will be NULL;
 630         */
 631        if (ep) {
 632                iser_conn = ep->dd_data;
 633                shost->sg_tablesize = iser_conn->scsi_sg_tablesize;
 634                shost->can_queue = min_t(u16, cmds_max, iser_conn->max_cmds);
 635
 636                mutex_lock(&iser_conn->state_mutex);
 637                if (iser_conn->state != ISER_CONN_UP) {
 638                        iser_err("iser conn %p already started teardown\n",
 639                                 iser_conn);
 640                        mutex_unlock(&iser_conn->state_mutex);
 641                        goto free_host;
 642                }
 643
 644                ib_conn = &iser_conn->ib_conn;
 645                ib_dev = ib_conn->device->ib_device;
 646                if (ib_conn->pi_support) {
 647                        u32 sig_caps = ib_dev->attrs.sig_prot_cap;
 648
 649                        shost->sg_prot_tablesize = shost->sg_tablesize;
 650                        scsi_host_set_prot(shost, iser_dif_prot_caps(sig_caps));
 651                        scsi_host_set_guard(shost, SHOST_DIX_GUARD_IP |
 652                                                   SHOST_DIX_GUARD_CRC);
 653                }
 654
 655                if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_SG_GAPS_REG))
 656                        shost->virt_boundary_mask = SZ_4K - 1;
 657
 658                if (iscsi_host_add(shost, ib_dev->dev.parent)) {
 659                        mutex_unlock(&iser_conn->state_mutex);
 660                        goto free_host;
 661                }
 662                mutex_unlock(&iser_conn->state_mutex);
 663        } else {
 664                shost->can_queue = min_t(u16, cmds_max, ISER_DEF_XMIT_CMDS_MAX);
 665                if (iscsi_host_add(shost, NULL))
 666                        goto free_host;
 667        }
 668
 669        max_fr_sectors = (shost->sg_tablesize * PAGE_SIZE) >> 9;
 670        shost->max_sectors = min(iser_max_sectors, max_fr_sectors);
 671
 672        iser_dbg("iser_conn %p, sg_tablesize %u, max_sectors %u\n",
 673                 iser_conn, shost->sg_tablesize,
 674                 shost->max_sectors);
 675
 676        if (shost->max_sectors < iser_max_sectors)
 677                iser_warn("max_sectors was reduced from %u to %u\n",
 678                          iser_max_sectors, shost->max_sectors);
 679
 680        cls_session = iscsi_session_setup(&iscsi_iser_transport, shost,
 681                                          shost->can_queue, 0,
 682                                          sizeof(struct iscsi_iser_task),
 683                                          initial_cmdsn, 0);
 684        if (!cls_session)
 685                goto remove_host;
 686
 687        return cls_session;
 688
 689remove_host:
 690        iscsi_host_remove(shost);
 691free_host:
 692        iscsi_host_free(shost);
 693        return NULL;
 694}
 695
 696static int
 697iscsi_iser_set_param(struct iscsi_cls_conn *cls_conn,
 698                     enum iscsi_param param, char *buf, int buflen)
 699{
 700        int value;
 701
 702        switch (param) {
 703        case ISCSI_PARAM_MAX_RECV_DLENGTH:
 704                /* TBD */
 705                break;
 706        case ISCSI_PARAM_HDRDGST_EN:
 707                sscanf(buf, "%d", &value);
 708                if (value) {
 709                        iser_err("DataDigest wasn't negotiated to None\n");
 710                        return -EPROTO;
 711                }
 712                break;
 713        case ISCSI_PARAM_DATADGST_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_IFMARKER_EN:
 721                sscanf(buf, "%d", &value);
 722                if (value) {
 723                        iser_err("IFMarker wasn't negotiated to No\n");
 724                        return -EPROTO;
 725                }
 726                break;
 727        case ISCSI_PARAM_OFMARKER_EN:
 728                sscanf(buf, "%d", &value);
 729                if (value) {
 730                        iser_err("OFMarker wasn't negotiated to No\n");
 731                        return -EPROTO;
 732                }
 733                break;
 734        default:
 735                return iscsi_set_param(cls_conn, param, buf, buflen);
 736        }
 737
 738        return 0;
 739}
 740
 741/**
 742 * iscsi_iser_set_param() - set class connection parameter
 743 * @cls_conn:    iscsi class connection
 744 * @stats:       iscsi stats to output
 745 *
 746 * Output connection statistics.
 747 */
 748static void
 749iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
 750{
 751        struct iscsi_conn *conn = cls_conn->dd_data;
 752
 753        stats->txdata_octets = conn->txdata_octets;
 754        stats->rxdata_octets = conn->rxdata_octets;
 755        stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
 756        stats->dataout_pdus = conn->dataout_pdus_cnt;
 757        stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
 758        stats->datain_pdus = conn->datain_pdus_cnt; /* always 0 */
 759        stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
 760        stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
 761        stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
 762        stats->custom_length = 0;
 763}
 764
 765static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep,
 766                                   enum iscsi_param param, char *buf)
 767{
 768        struct iser_conn *iser_conn = ep->dd_data;
 769
 770        switch (param) {
 771        case ISCSI_PARAM_CONN_PORT:
 772        case ISCSI_PARAM_CONN_ADDRESS:
 773                if (!iser_conn || !iser_conn->ib_conn.cma_id)
 774                        return -ENOTCONN;
 775
 776                return iscsi_conn_get_addr_param((struct sockaddr_storage *)
 777                                &iser_conn->ib_conn.cma_id->route.addr.dst_addr,
 778                                param, buf);
 779        default:
 780                break;
 781        }
 782        return -ENOSYS;
 783}
 784
 785/**
 786 * iscsi_iser_ep_connect() - Initiate iSER connection establishment
 787 * @shost:          scsi_host
 788 * @dst_addr:       destination address
 789 * @non_blocking:   indicate if routine can block
 790 *
 791 * Allocate an iscsi endpoint, an iser_conn structure and bind them.
 792 * After that start RDMA connection establishment via rdma_cm. We
 793 * don't allocate iser_conn embedded in iscsi_endpoint since in teardown
 794 * the endpoint will be destroyed at ep_disconnect while iser_conn will
 795 * cleanup its resources asynchronuously.
 796 *
 797 * Return: iscsi_endpoint created by iscsi layer or ERR_PTR(error)
 798 *         if fails.
 799 */
 800static struct iscsi_endpoint *
 801iscsi_iser_ep_connect(struct Scsi_Host *shost, struct sockaddr *dst_addr,
 802                      int non_blocking)
 803{
 804        int err;
 805        struct iser_conn *iser_conn;
 806        struct iscsi_endpoint *ep;
 807
 808        ep = iscsi_create_endpoint(0);
 809        if (!ep)
 810                return ERR_PTR(-ENOMEM);
 811
 812        iser_conn = kzalloc(sizeof(*iser_conn), GFP_KERNEL);
 813        if (!iser_conn) {
 814                err = -ENOMEM;
 815                goto failure;
 816        }
 817
 818        ep->dd_data = iser_conn;
 819        iser_conn->ep = ep;
 820        iser_conn_init(iser_conn);
 821
 822        err = iser_connect(iser_conn, NULL, dst_addr, non_blocking);
 823        if (err)
 824                goto failure;
 825
 826        return ep;
 827failure:
 828        iscsi_destroy_endpoint(ep);
 829        return ERR_PTR(err);
 830}
 831
 832/**
 833 * iscsi_iser_ep_poll() - poll for iser connection establishment to complete
 834 * @ep:            iscsi endpoint (created at ep_connect)
 835 * @timeout_ms:    polling timeout allowed in ms.
 836 *
 837 * This routine boils down to waiting for up_completion signaling
 838 * that cma_id got CONNECTED event.
 839 *
 840 * Return: 1 if succeeded in connection establishment, 0 if timeout expired
 841 *         (libiscsi will retry will kick in) or -1 if interrupted by signal
 842 *         or more likely iser connection state transitioned to TEMINATING or
 843 *         DOWN during the wait period.
 844 */
 845static int
 846iscsi_iser_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
 847{
 848        struct iser_conn *iser_conn = ep->dd_data;
 849        int rc;
 850
 851        rc = wait_for_completion_interruptible_timeout(&iser_conn->up_completion,
 852                                                       msecs_to_jiffies(timeout_ms));
 853        /* if conn establishment failed, return error code to iscsi */
 854        if (rc == 0) {
 855                mutex_lock(&iser_conn->state_mutex);
 856                if (iser_conn->state == ISER_CONN_TERMINATING ||
 857                    iser_conn->state == ISER_CONN_DOWN)
 858                        rc = -1;
 859                mutex_unlock(&iser_conn->state_mutex);
 860        }
 861
 862        iser_info("iser conn %p rc = %d\n", iser_conn, rc);
 863
 864        if (rc > 0)
 865                return 1; /* success, this is the equivalent of EPOLLOUT */
 866        else if (!rc)
 867                return 0; /* timeout */
 868        else
 869                return rc; /* signal */
 870}
 871
 872/**
 873 * iscsi_iser_ep_disconnect() - Initiate connection teardown process
 874 * @ep:    iscsi endpoint handle
 875 *
 876 * This routine is not blocked by iser and RDMA termination process
 877 * completion as we queue a deffered work for iser/RDMA destruction
 878 * and cleanup or actually call it immediately in case we didn't pass
 879 * iscsi conn bind/start stage, thus it is safe.
 880 */
 881static void
 882iscsi_iser_ep_disconnect(struct iscsi_endpoint *ep)
 883{
 884        struct iser_conn *iser_conn = ep->dd_data;
 885
 886        iser_info("ep %p iser conn %p\n", ep, iser_conn);
 887
 888        mutex_lock(&iser_conn->state_mutex);
 889        iser_conn_terminate(iser_conn);
 890
 891        /*
 892         * if iser_conn and iscsi_conn are bound, we must wait for
 893         * iscsi_conn_stop and flush errors completion before freeing
 894         * the iser resources. Otherwise we are safe to free resources
 895         * immediately.
 896         */
 897        if (iser_conn->iscsi_conn) {
 898                INIT_WORK(&iser_conn->release_work, iser_release_work);
 899                queue_work(release_wq, &iser_conn->release_work);
 900                mutex_unlock(&iser_conn->state_mutex);
 901        } else {
 902                iser_conn->state = ISER_CONN_DOWN;
 903                mutex_unlock(&iser_conn->state_mutex);
 904                iser_conn_release(iser_conn);
 905        }
 906
 907        iscsi_destroy_endpoint(ep);
 908}
 909
 910static umode_t iser_attr_is_visible(int param_type, int param)
 911{
 912        switch (param_type) {
 913        case ISCSI_HOST_PARAM:
 914                switch (param) {
 915                case ISCSI_HOST_PARAM_NETDEV_NAME:
 916                case ISCSI_HOST_PARAM_HWADDRESS:
 917                case ISCSI_HOST_PARAM_INITIATOR_NAME:
 918                        return S_IRUGO;
 919                default:
 920                        return 0;
 921                }
 922        case ISCSI_PARAM:
 923                switch (param) {
 924                case ISCSI_PARAM_MAX_RECV_DLENGTH:
 925                case ISCSI_PARAM_MAX_XMIT_DLENGTH:
 926                case ISCSI_PARAM_HDRDGST_EN:
 927                case ISCSI_PARAM_DATADGST_EN:
 928                case ISCSI_PARAM_CONN_ADDRESS:
 929                case ISCSI_PARAM_CONN_PORT:
 930                case ISCSI_PARAM_EXP_STATSN:
 931                case ISCSI_PARAM_PERSISTENT_ADDRESS:
 932                case ISCSI_PARAM_PERSISTENT_PORT:
 933                case ISCSI_PARAM_PING_TMO:
 934                case ISCSI_PARAM_RECV_TMO:
 935                case ISCSI_PARAM_INITIAL_R2T_EN:
 936                case ISCSI_PARAM_MAX_R2T:
 937                case ISCSI_PARAM_IMM_DATA_EN:
 938                case ISCSI_PARAM_FIRST_BURST:
 939                case ISCSI_PARAM_MAX_BURST:
 940                case ISCSI_PARAM_PDU_INORDER_EN:
 941                case ISCSI_PARAM_DATASEQ_INORDER_EN:
 942                case ISCSI_PARAM_TARGET_NAME:
 943                case ISCSI_PARAM_TPGT:
 944                case ISCSI_PARAM_USERNAME:
 945                case ISCSI_PARAM_PASSWORD:
 946                case ISCSI_PARAM_USERNAME_IN:
 947                case ISCSI_PARAM_PASSWORD_IN:
 948                case ISCSI_PARAM_FAST_ABORT:
 949                case ISCSI_PARAM_ABORT_TMO:
 950                case ISCSI_PARAM_LU_RESET_TMO:
 951                case ISCSI_PARAM_TGT_RESET_TMO:
 952                case ISCSI_PARAM_IFACE_NAME:
 953                case ISCSI_PARAM_INITIATOR_NAME:
 954                case ISCSI_PARAM_DISCOVERY_SESS:
 955                        return S_IRUGO;
 956                default:
 957                        return 0;
 958                }
 959        }
 960
 961        return 0;
 962}
 963
 964static struct scsi_host_template iscsi_iser_sht = {
 965        .module                 = THIS_MODULE,
 966        .name                   = "iSCSI Initiator over iSER",
 967        .queuecommand           = iscsi_queuecommand,
 968        .change_queue_depth     = scsi_change_queue_depth,
 969        .sg_tablesize           = ISCSI_ISER_DEF_SG_TABLESIZE,
 970        .cmd_per_lun            = ISER_DEF_CMD_PER_LUN,
 971        .eh_timed_out           = iscsi_eh_cmd_timed_out,
 972        .eh_abort_handler       = iscsi_eh_abort,
 973        .eh_device_reset_handler= iscsi_eh_device_reset,
 974        .eh_target_reset_handler = iscsi_eh_recover_target,
 975        .target_alloc           = iscsi_target_alloc,
 976        .proc_name              = "iscsi_iser",
 977        .this_id                = -1,
 978        .track_queue_depth      = 1,
 979};
 980
 981static struct iscsi_transport iscsi_iser_transport = {
 982        .owner                  = THIS_MODULE,
 983        .name                   = "iser",
 984        .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_TEXT_NEGO,
 985        /* session management */
 986        .create_session         = iscsi_iser_session_create,
 987        .destroy_session        = iscsi_iser_session_destroy,
 988        /* connection management */
 989        .create_conn            = iscsi_iser_conn_create,
 990        .bind_conn              = iscsi_iser_conn_bind,
 991        .destroy_conn           = iscsi_conn_teardown,
 992        .attr_is_visible        = iser_attr_is_visible,
 993        .set_param              = iscsi_iser_set_param,
 994        .get_conn_param         = iscsi_conn_get_param,
 995        .get_ep_param           = iscsi_iser_get_ep_param,
 996        .get_session_param      = iscsi_session_get_param,
 997        .start_conn             = iscsi_iser_conn_start,
 998        .stop_conn              = iscsi_iser_conn_stop,
 999        /* iscsi host params */
1000        .get_host_param         = iscsi_host_get_param,
1001        .set_host_param         = iscsi_host_set_param,
1002        /* IO */
1003        .send_pdu               = iscsi_conn_send_pdu,
1004        .get_stats              = iscsi_iser_conn_get_stats,
1005        .init_task              = iscsi_iser_task_init,
1006        .xmit_task              = iscsi_iser_task_xmit,
1007        .cleanup_task           = iscsi_iser_cleanup_task,
1008        .alloc_pdu              = iscsi_iser_pdu_alloc,
1009        .check_protection       = iscsi_iser_check_protection,
1010        /* recovery */
1011        .session_recovery_timedout = iscsi_session_recovery_timedout,
1012
1013        .ep_connect             = iscsi_iser_ep_connect,
1014        .ep_poll                = iscsi_iser_ep_poll,
1015        .ep_disconnect          = iscsi_iser_ep_disconnect
1016};
1017
1018static int __init iser_init(void)
1019{
1020        int err;
1021
1022        iser_dbg("Starting iSER datamover...\n");
1023
1024        if (iscsi_max_lun < 1) {
1025                iser_err("Invalid max_lun value of %u\n", iscsi_max_lun);
1026                return -EINVAL;
1027        }
1028
1029        memset(&ig, 0, sizeof(struct iser_global));
1030
1031        ig.desc_cache = kmem_cache_create("iser_descriptors",
1032                                          sizeof(struct iser_tx_desc),
1033                                          0, SLAB_HWCACHE_ALIGN,
1034                                          NULL);
1035        if (ig.desc_cache == NULL)
1036                return -ENOMEM;
1037
1038        /* device init is called only after the first addr resolution */
1039        mutex_init(&ig.device_list_mutex);
1040        INIT_LIST_HEAD(&ig.device_list);
1041        mutex_init(&ig.connlist_mutex);
1042        INIT_LIST_HEAD(&ig.connlist);
1043
1044        release_wq = alloc_workqueue("release workqueue", 0, 0);
1045        if (!release_wq) {
1046                iser_err("failed to allocate release workqueue\n");
1047                err = -ENOMEM;
1048                goto err_alloc_wq;
1049        }
1050
1051        iscsi_iser_scsi_transport = iscsi_register_transport(
1052                                                        &iscsi_iser_transport);
1053        if (!iscsi_iser_scsi_transport) {
1054                iser_err("iscsi_register_transport failed\n");
1055                err = -EINVAL;
1056                goto err_reg;
1057        }
1058
1059        return 0;
1060
1061err_reg:
1062        destroy_workqueue(release_wq);
1063err_alloc_wq:
1064        kmem_cache_destroy(ig.desc_cache);
1065
1066        return err;
1067}
1068
1069static void __exit iser_exit(void)
1070{
1071        struct iser_conn *iser_conn, *n;
1072        int connlist_empty;
1073
1074        iser_dbg("Removing iSER datamover...\n");
1075        destroy_workqueue(release_wq);
1076
1077        mutex_lock(&ig.connlist_mutex);
1078        connlist_empty = list_empty(&ig.connlist);
1079        mutex_unlock(&ig.connlist_mutex);
1080
1081        if (!connlist_empty) {
1082                iser_err("Error cleanup stage completed but we still have iser "
1083                         "connections, destroying them anyway\n");
1084                list_for_each_entry_safe(iser_conn, n, &ig.connlist,
1085                                         conn_list) {
1086                        iser_conn_release(iser_conn);
1087                }
1088        }
1089
1090        iscsi_unregister_transport(&iscsi_iser_transport);
1091        kmem_cache_destroy(ig.desc_cache);
1092}
1093
1094module_init(iser_init);
1095module_exit(iser_exit);
1096