linux/drivers/scsi/iscsi_tcp.c
<<
>>
Prefs
   1/*
   2 * iSCSI Initiator over TCP/IP Data-Path
   3 *
   4 * Copyright (C) 2004 Dmitry Yusupov
   5 * Copyright (C) 2004 Alex Aizman
   6 * Copyright (C) 2005 - 2006 Mike Christie
   7 * Copyright (C) 2006 Red Hat, Inc.  All rights reserved.
   8 * maintained by open-iscsi@googlegroups.com
   9 *
  10 * This program is free software; you can redistribute it and/or modify
  11 * it under the terms of the GNU General Public License as published
  12 * by the Free Software Foundation; either version 2 of the License, or
  13 * (at your option) any later version.
  14 *
  15 * This program is distributed in the hope that it will be useful, but
  16 * WITHOUT ANY WARRANTY; without even the implied warranty of
  17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  18 * General Public License for more details.
  19 *
  20 * See the file COPYING included with this distribution for more details.
  21 *
  22 * Credits:
  23 *      Christoph Hellwig
  24 *      FUJITA Tomonori
  25 *      Arne Redlich
  26 *      Zhenyu Wang
  27 */
  28
  29#include <linux/types.h>
  30#include <linux/inet.h>
  31#include <linux/file.h>
  32#include <linux/blkdev.h>
  33#include <linux/crypto.h>
  34#include <linux/delay.h>
  35#include <linux/kfifo.h>
  36#include <linux/scatterlist.h>
  37#include <net/tcp.h>
  38#include <scsi/scsi_cmnd.h>
  39#include <scsi/scsi_device.h>
  40#include <scsi/scsi_host.h>
  41#include <scsi/scsi.h>
  42#include <scsi/scsi_transport_iscsi.h>
  43
  44#include "iscsi_tcp.h"
  45
  46MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
  47              "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
  48              "Alex Aizman <itn780@yahoo.com>");
  49MODULE_DESCRIPTION("iSCSI/TCP data-path");
  50MODULE_LICENSE("GPL");
  51
  52static struct scsi_transport_template *iscsi_sw_tcp_scsi_transport;
  53static struct scsi_host_template iscsi_sw_tcp_sht;
  54static struct iscsi_transport iscsi_sw_tcp_transport;
  55
  56static unsigned int iscsi_max_lun = 512;
  57module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
  58
  59static int iscsi_sw_tcp_dbg;
  60module_param_named(debug_iscsi_tcp, iscsi_sw_tcp_dbg, int,
  61                   S_IRUGO | S_IWUSR);
  62MODULE_PARM_DESC(debug_iscsi_tcp, "Turn on debugging for iscsi_tcp module "
  63                 "Set to 1 to turn on, and zero to turn off. Default is off.");
  64
  65#define ISCSI_SW_TCP_DBG(_conn, dbg_fmt, arg...)                \
  66        do {                                                    \
  67                if (iscsi_sw_tcp_dbg)                           \
  68                        iscsi_conn_printk(KERN_INFO, _conn,     \
  69                                             "%s " dbg_fmt,     \
  70                                             __func__, ##arg);  \
  71        } while (0);
  72
  73
  74/**
  75 * iscsi_sw_tcp_recv - TCP receive in sendfile fashion
  76 * @rd_desc: read descriptor
  77 * @skb: socket buffer
  78 * @offset: offset in skb
  79 * @len: skb->len - offset
  80 */
  81static int iscsi_sw_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
  82                             unsigned int offset, size_t len)
  83{
  84        struct iscsi_conn *conn = rd_desc->arg.data;
  85        unsigned int consumed, total_consumed = 0;
  86        int status;
  87
  88        ISCSI_SW_TCP_DBG(conn, "in %d bytes\n", skb->len - offset);
  89
  90        do {
  91                status = 0;
  92                consumed = iscsi_tcp_recv_skb(conn, skb, offset, 0, &status);
  93                offset += consumed;
  94                total_consumed += consumed;
  95        } while (consumed != 0 && status != ISCSI_TCP_SKB_DONE);
  96
  97        ISCSI_SW_TCP_DBG(conn, "read %d bytes status %d\n",
  98                         skb->len - offset, status);
  99        return total_consumed;
 100}
 101
 102/**
 103 * iscsi_sw_sk_state_check - check socket state
 104 * @sk: socket
 105 *
 106 * If the socket is in CLOSE or CLOSE_WAIT we should
 107 * not close the connection if there is still some
 108 * data pending.
 109 */
 110static inline int iscsi_sw_sk_state_check(struct sock *sk)
 111{
 112        struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
 113
 114        if ((sk->sk_state == TCP_CLOSE_WAIT || sk->sk_state == TCP_CLOSE) &&
 115            !atomic_read(&sk->sk_rmem_alloc)) {
 116                ISCSI_SW_TCP_DBG(conn, "TCP_CLOSE|TCP_CLOSE_WAIT\n");
 117                iscsi_conn_failure(conn, ISCSI_ERR_TCP_CONN_CLOSE);
 118                return -ECONNRESET;
 119        }
 120        return 0;
 121}
 122
 123static void iscsi_sw_tcp_data_ready(struct sock *sk, int flag)
 124{
 125        struct iscsi_conn *conn = sk->sk_user_data;
 126        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 127        read_descriptor_t rd_desc;
 128
 129        read_lock(&sk->sk_callback_lock);
 130
 131        /*
 132         * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
 133         * We set count to 1 because we want the network layer to
 134         * hand us all the skbs that are available. iscsi_tcp_recv
 135         * handled pdus that cross buffers or pdus that still need data.
 136         */
 137        rd_desc.arg.data = conn;
 138        rd_desc.count = 1;
 139        tcp_read_sock(sk, &rd_desc, iscsi_sw_tcp_recv);
 140
 141        iscsi_sw_sk_state_check(sk);
 142
 143        read_unlock(&sk->sk_callback_lock);
 144
 145        /* If we had to (atomically) map a highmem page,
 146         * unmap it now. */
 147        iscsi_tcp_segment_unmap(&tcp_conn->in.segment);
 148}
 149
 150static void iscsi_sw_tcp_state_change(struct sock *sk)
 151{
 152        struct iscsi_tcp_conn *tcp_conn;
 153        struct iscsi_sw_tcp_conn *tcp_sw_conn;
 154        struct iscsi_conn *conn;
 155        struct iscsi_session *session;
 156        void (*old_state_change)(struct sock *);
 157
 158        read_lock(&sk->sk_callback_lock);
 159
 160        conn = (struct iscsi_conn*)sk->sk_user_data;
 161        session = conn->session;
 162
 163        iscsi_sw_sk_state_check(sk);
 164
 165        tcp_conn = conn->dd_data;
 166        tcp_sw_conn = tcp_conn->dd_data;
 167        old_state_change = tcp_sw_conn->old_state_change;
 168
 169        read_unlock(&sk->sk_callback_lock);
 170
 171        old_state_change(sk);
 172}
 173
 174/**
 175 * iscsi_write_space - Called when more output buffer space is available
 176 * @sk: socket space is available for
 177 **/
 178static void iscsi_sw_tcp_write_space(struct sock *sk)
 179{
 180        struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
 181        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 182        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 183
 184        tcp_sw_conn->old_write_space(sk);
 185        ISCSI_SW_TCP_DBG(conn, "iscsi_write_space\n");
 186        iscsi_conn_queue_work(conn);
 187}
 188
 189static void iscsi_sw_tcp_conn_set_callbacks(struct iscsi_conn *conn)
 190{
 191        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 192        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 193        struct sock *sk = tcp_sw_conn->sock->sk;
 194
 195        /* assign new callbacks */
 196        write_lock_bh(&sk->sk_callback_lock);
 197        sk->sk_user_data = conn;
 198        tcp_sw_conn->old_data_ready = sk->sk_data_ready;
 199        tcp_sw_conn->old_state_change = sk->sk_state_change;
 200        tcp_sw_conn->old_write_space = sk->sk_write_space;
 201        sk->sk_data_ready = iscsi_sw_tcp_data_ready;
 202        sk->sk_state_change = iscsi_sw_tcp_state_change;
 203        sk->sk_write_space = iscsi_sw_tcp_write_space;
 204        write_unlock_bh(&sk->sk_callback_lock);
 205}
 206
 207static void
 208iscsi_sw_tcp_conn_restore_callbacks(struct iscsi_sw_tcp_conn *tcp_sw_conn)
 209{
 210        struct sock *sk = tcp_sw_conn->sock->sk;
 211
 212        /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
 213        write_lock_bh(&sk->sk_callback_lock);
 214        sk->sk_user_data    = NULL;
 215        sk->sk_data_ready   = tcp_sw_conn->old_data_ready;
 216        sk->sk_state_change = tcp_sw_conn->old_state_change;
 217        sk->sk_write_space  = tcp_sw_conn->old_write_space;
 218        sk->sk_no_check  = 0;
 219        write_unlock_bh(&sk->sk_callback_lock);
 220}
 221
 222/**
 223 * iscsi_sw_tcp_xmit_segment - transmit segment
 224 * @tcp_conn: the iSCSI TCP connection
 225 * @segment: the buffer to transmnit
 226 *
 227 * This function transmits as much of the buffer as
 228 * the network layer will accept, and returns the number of
 229 * bytes transmitted.
 230 *
 231 * If CRC hashing is enabled, the function will compute the
 232 * hash as it goes. When the entire segment has been transmitted,
 233 * it will retrieve the hash value and send it as well.
 234 */
 235static int iscsi_sw_tcp_xmit_segment(struct iscsi_tcp_conn *tcp_conn,
 236                                     struct iscsi_segment *segment)
 237{
 238        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 239        struct socket *sk = tcp_sw_conn->sock;
 240        unsigned int copied = 0;
 241        int r = 0;
 242
 243        while (!iscsi_tcp_segment_done(tcp_conn, segment, 0, r)) {
 244                struct scatterlist *sg;
 245                unsigned int offset, copy;
 246                int flags = 0;
 247
 248                r = 0;
 249                offset = segment->copied;
 250                copy = segment->size - offset;
 251
 252                if (segment->total_copied + segment->size < segment->total_size)
 253                        flags |= MSG_MORE;
 254
 255                /* Use sendpage if we can; else fall back to sendmsg */
 256                if (!segment->data) {
 257                        sg = segment->sg;
 258                        offset += segment->sg_offset + sg->offset;
 259                        r = tcp_sw_conn->sendpage(sk, sg_page(sg), offset,
 260                                                  copy, flags);
 261                } else {
 262                        struct msghdr msg = { .msg_flags = flags };
 263                        struct kvec iov = {
 264                                .iov_base = segment->data + offset,
 265                                .iov_len = copy
 266                        };
 267
 268                        r = kernel_sendmsg(sk, &msg, &iov, 1, copy);
 269                }
 270
 271                if (r < 0) {
 272                        iscsi_tcp_segment_unmap(segment);
 273                        return r;
 274                }
 275                copied += r;
 276        }
 277        return copied;
 278}
 279
 280/**
 281 * iscsi_sw_tcp_xmit - TCP transmit
 282 **/
 283static int iscsi_sw_tcp_xmit(struct iscsi_conn *conn)
 284{
 285        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 286        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 287        struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
 288        unsigned int consumed = 0;
 289        int rc = 0;
 290
 291        while (1) {
 292                rc = iscsi_sw_tcp_xmit_segment(tcp_conn, segment);
 293                /*
 294                 * We may not have been able to send data because the conn
 295                 * is getting stopped. libiscsi will know so propogate err
 296                 * for it to do the right thing.
 297                 */
 298                if (rc == -EAGAIN)
 299                        return rc;
 300                else if (rc < 0) {
 301                        rc = ISCSI_ERR_XMIT_FAILED;
 302                        goto error;
 303                } else if (rc == 0)
 304                        break;
 305
 306                consumed += rc;
 307
 308                if (segment->total_copied >= segment->total_size) {
 309                        if (segment->done != NULL) {
 310                                rc = segment->done(tcp_conn, segment);
 311                                if (rc != 0)
 312                                        goto error;
 313                        }
 314                }
 315        }
 316
 317        ISCSI_SW_TCP_DBG(conn, "xmit %d bytes\n", consumed);
 318
 319        conn->txdata_octets += consumed;
 320        return consumed;
 321
 322error:
 323        /* Transmit error. We could initiate error recovery
 324         * here. */
 325        ISCSI_SW_TCP_DBG(conn, "Error sending PDU, errno=%d\n", rc);
 326        iscsi_conn_failure(conn, rc);
 327        return -EIO;
 328}
 329
 330/**
 331 * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
 332 */
 333static inline int iscsi_sw_tcp_xmit_qlen(struct iscsi_conn *conn)
 334{
 335        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 336        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 337        struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
 338
 339        return segment->total_copied - segment->total_size;
 340}
 341
 342static int iscsi_sw_tcp_pdu_xmit(struct iscsi_task *task)
 343{
 344        struct iscsi_conn *conn = task->conn;
 345        int rc;
 346
 347        while (iscsi_sw_tcp_xmit_qlen(conn)) {
 348                rc = iscsi_sw_tcp_xmit(conn);
 349                if (rc == 0)
 350                        return -EAGAIN;
 351                if (rc < 0)
 352                        return rc;
 353        }
 354
 355        return 0;
 356}
 357
 358/*
 359 * This is called when we're done sending the header.
 360 * Simply copy the data_segment to the send segment, and return.
 361 */
 362static int iscsi_sw_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
 363                                      struct iscsi_segment *segment)
 364{
 365        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 366
 367        tcp_sw_conn->out.segment = tcp_sw_conn->out.data_segment;
 368        ISCSI_SW_TCP_DBG(tcp_conn->iscsi_conn,
 369                         "Header done. Next segment size %u total_size %u\n",
 370                         tcp_sw_conn->out.segment.size,
 371                         tcp_sw_conn->out.segment.total_size);
 372        return 0;
 373}
 374
 375static void iscsi_sw_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr,
 376                                       size_t hdrlen)
 377{
 378        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 379        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 380
 381        ISCSI_SW_TCP_DBG(conn, "%s\n", conn->hdrdgst_en ?
 382                         "digest enabled" : "digest disabled");
 383
 384        /* Clear the data segment - needs to be filled in by the
 385         * caller using iscsi_tcp_send_data_prep() */
 386        memset(&tcp_sw_conn->out.data_segment, 0,
 387               sizeof(struct iscsi_segment));
 388
 389        /* If header digest is enabled, compute the CRC and
 390         * place the digest into the same buffer. We make
 391         * sure that both iscsi_tcp_task and mtask have
 392         * sufficient room.
 393         */
 394        if (conn->hdrdgst_en) {
 395                iscsi_tcp_dgst_header(&tcp_sw_conn->tx_hash, hdr, hdrlen,
 396                                      hdr + hdrlen);
 397                hdrlen += ISCSI_DIGEST_SIZE;
 398        }
 399
 400        /* Remember header pointer for later, when we need
 401         * to decide whether there's a payload to go along
 402         * with the header. */
 403        tcp_sw_conn->out.hdr = hdr;
 404
 405        iscsi_segment_init_linear(&tcp_sw_conn->out.segment, hdr, hdrlen,
 406                                  iscsi_sw_tcp_send_hdr_done, NULL);
 407}
 408
 409/*
 410 * Prepare the send buffer for the payload data.
 411 * Padding and checksumming will all be taken care
 412 * of by the iscsi_segment routines.
 413 */
 414static int
 415iscsi_sw_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
 416                            unsigned int count, unsigned int offset,
 417                            unsigned int len)
 418{
 419        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 420        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 421        struct hash_desc *tx_hash = NULL;
 422        unsigned int hdr_spec_len;
 423
 424        ISCSI_SW_TCP_DBG(conn, "offset=%d, datalen=%d %s\n", offset, len,
 425                         conn->datadgst_en ?
 426                         "digest enabled" : "digest disabled");
 427
 428        /* Make sure the datalen matches what the caller
 429           said he would send. */
 430        hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
 431        WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
 432
 433        if (conn->datadgst_en)
 434                tx_hash = &tcp_sw_conn->tx_hash;
 435
 436        return iscsi_segment_seek_sg(&tcp_sw_conn->out.data_segment,
 437                                     sg, count, offset, len,
 438                                     NULL, tx_hash);
 439}
 440
 441static void
 442iscsi_sw_tcp_send_linear_data_prep(struct iscsi_conn *conn, void *data,
 443                                   size_t len)
 444{
 445        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 446        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 447        struct hash_desc *tx_hash = NULL;
 448        unsigned int hdr_spec_len;
 449
 450        ISCSI_SW_TCP_DBG(conn, "datalen=%zd %s\n", len, conn->datadgst_en ?
 451                         "digest enabled" : "digest disabled");
 452
 453        /* Make sure the datalen matches what the caller
 454           said he would send. */
 455        hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
 456        WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
 457
 458        if (conn->datadgst_en)
 459                tx_hash = &tcp_sw_conn->tx_hash;
 460
 461        iscsi_segment_init_linear(&tcp_sw_conn->out.data_segment,
 462                                data, len, NULL, tx_hash);
 463}
 464
 465static int iscsi_sw_tcp_pdu_init(struct iscsi_task *task,
 466                                 unsigned int offset, unsigned int count)
 467{
 468        struct iscsi_conn *conn = task->conn;
 469        int err = 0;
 470
 471        iscsi_sw_tcp_send_hdr_prep(conn, task->hdr, task->hdr_len);
 472
 473        if (!count)
 474                return 0;
 475
 476        if (!task->sc)
 477                iscsi_sw_tcp_send_linear_data_prep(conn, task->data, count);
 478        else {
 479                struct scsi_data_buffer *sdb = scsi_out(task->sc);
 480
 481                err = iscsi_sw_tcp_send_data_prep(conn, sdb->table.sgl,
 482                                                  sdb->table.nents, offset,
 483                                                  count);
 484        }
 485
 486        if (err) {
 487                /* got invalid offset/len */
 488                return -EIO;
 489        }
 490        return 0;
 491}
 492
 493static int iscsi_sw_tcp_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
 494{
 495        struct iscsi_tcp_task *tcp_task = task->dd_data;
 496
 497        task->hdr = task->dd_data + sizeof(*tcp_task);
 498        task->hdr_max = sizeof(struct iscsi_sw_tcp_hdrbuf) - ISCSI_DIGEST_SIZE;
 499        return 0;
 500}
 501
 502static struct iscsi_cls_conn *
 503iscsi_sw_tcp_conn_create(struct iscsi_cls_session *cls_session,
 504                         uint32_t conn_idx)
 505{
 506        struct iscsi_conn *conn;
 507        struct iscsi_cls_conn *cls_conn;
 508        struct iscsi_tcp_conn *tcp_conn;
 509        struct iscsi_sw_tcp_conn *tcp_sw_conn;
 510
 511        cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*tcp_sw_conn),
 512                                        conn_idx);
 513        if (!cls_conn)
 514                return NULL;
 515        conn = cls_conn->dd_data;
 516        tcp_conn = conn->dd_data;
 517        tcp_sw_conn = tcp_conn->dd_data;
 518
 519        tcp_sw_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
 520                                                     CRYPTO_ALG_ASYNC);
 521        tcp_sw_conn->tx_hash.flags = 0;
 522        if (IS_ERR(tcp_sw_conn->tx_hash.tfm))
 523                goto free_conn;
 524
 525        tcp_sw_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
 526                                                     CRYPTO_ALG_ASYNC);
 527        tcp_sw_conn->rx_hash.flags = 0;
 528        if (IS_ERR(tcp_sw_conn->rx_hash.tfm))
 529                goto free_tx_tfm;
 530        tcp_conn->rx_hash = &tcp_sw_conn->rx_hash;
 531
 532        return cls_conn;
 533
 534free_tx_tfm:
 535        crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
 536free_conn:
 537        iscsi_conn_printk(KERN_ERR, conn,
 538                          "Could not create connection due to crc32c "
 539                          "loading error. Make sure the crc32c "
 540                          "module is built as a module or into the "
 541                          "kernel\n");
 542        iscsi_tcp_conn_teardown(cls_conn);
 543        return NULL;
 544}
 545
 546static void iscsi_sw_tcp_release_conn(struct iscsi_conn *conn)
 547{
 548        struct iscsi_session *session = conn->session;
 549        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 550        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 551        struct socket *sock = tcp_sw_conn->sock;
 552
 553        if (!sock)
 554                return;
 555
 556        sock_hold(sock->sk);
 557        iscsi_sw_tcp_conn_restore_callbacks(tcp_sw_conn);
 558        sock_put(sock->sk);
 559
 560        spin_lock_bh(&session->lock);
 561        tcp_sw_conn->sock = NULL;
 562        spin_unlock_bh(&session->lock);
 563        sockfd_put(sock);
 564}
 565
 566static void iscsi_sw_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
 567{
 568        struct iscsi_conn *conn = cls_conn->dd_data;
 569        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 570        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 571
 572        iscsi_sw_tcp_release_conn(conn);
 573
 574        if (tcp_sw_conn->tx_hash.tfm)
 575                crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
 576        if (tcp_sw_conn->rx_hash.tfm)
 577                crypto_free_hash(tcp_sw_conn->rx_hash.tfm);
 578
 579        iscsi_tcp_conn_teardown(cls_conn);
 580}
 581
 582static void iscsi_sw_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
 583{
 584        struct iscsi_conn *conn = cls_conn->dd_data;
 585        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 586        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 587
 588        /* userspace may have goofed up and not bound us */
 589        if (!tcp_sw_conn->sock)
 590                return;
 591        /*
 592         * Make sure our recv side is stopped.
 593         * Older tools called conn stop before ep_disconnect
 594         * so IO could still be coming in.
 595         */
 596        write_lock_bh(&tcp_sw_conn->sock->sk->sk_callback_lock);
 597        set_bit(ISCSI_SUSPEND_BIT, &conn->suspend_rx);
 598        write_unlock_bh(&tcp_sw_conn->sock->sk->sk_callback_lock);
 599
 600        iscsi_conn_stop(cls_conn, flag);
 601        iscsi_sw_tcp_release_conn(conn);
 602}
 603
 604static int iscsi_sw_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
 605                                 char *buf, int *port,
 606                                 int (*getname)(struct socket *,
 607                                                struct sockaddr *,
 608                                                int *addrlen))
 609{
 610        struct sockaddr_storage *addr;
 611        struct sockaddr_in6 *sin6;
 612        struct sockaddr_in *sin;
 613        int rc = 0, len;
 614
 615        addr = kmalloc(sizeof(*addr), GFP_KERNEL);
 616        if (!addr)
 617                return -ENOMEM;
 618
 619        if (getname(sock, (struct sockaddr *) addr, &len)) {
 620                rc = -ENODEV;
 621                goto free_addr;
 622        }
 623
 624        switch (addr->ss_family) {
 625        case AF_INET:
 626                sin = (struct sockaddr_in *)addr;
 627                spin_lock_bh(&conn->session->lock);
 628                sprintf(buf, "%pI4", &sin->sin_addr.s_addr);
 629                *port = be16_to_cpu(sin->sin_port);
 630                spin_unlock_bh(&conn->session->lock);
 631                break;
 632        case AF_INET6:
 633                sin6 = (struct sockaddr_in6 *)addr;
 634                spin_lock_bh(&conn->session->lock);
 635                sprintf(buf, "%pI6", &sin6->sin6_addr);
 636                *port = be16_to_cpu(sin6->sin6_port);
 637                spin_unlock_bh(&conn->session->lock);
 638                break;
 639        }
 640free_addr:
 641        kfree(addr);
 642        return rc;
 643}
 644
 645static int
 646iscsi_sw_tcp_conn_bind(struct iscsi_cls_session *cls_session,
 647                       struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
 648                       int is_leading)
 649{
 650        struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
 651        struct iscsi_host *ihost = shost_priv(shost);
 652        struct iscsi_conn *conn = cls_conn->dd_data;
 653        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 654        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 655        struct sock *sk;
 656        struct socket *sock;
 657        int err;
 658
 659        /* lookup for existing socket */
 660        sock = sockfd_lookup((int)transport_eph, &err);
 661        if (!sock) {
 662                iscsi_conn_printk(KERN_ERR, conn,
 663                                  "sockfd_lookup failed %d\n", err);
 664                return -EEXIST;
 665        }
 666        /*
 667         * copy these values now because if we drop the session
 668         * userspace may still want to query the values since we will
 669         * be using them for the reconnect
 670         */
 671        err = iscsi_sw_tcp_get_addr(conn, sock, conn->portal_address,
 672                                    &conn->portal_port, kernel_getpeername);
 673        if (err)
 674                goto free_socket;
 675
 676        err = iscsi_sw_tcp_get_addr(conn, sock, ihost->local_address,
 677                                    &ihost->local_port, kernel_getsockname);
 678        if (err)
 679                goto free_socket;
 680
 681        err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
 682        if (err)
 683                goto free_socket;
 684
 685        /* bind iSCSI connection and socket */
 686        tcp_sw_conn->sock = sock;
 687
 688        /* setup Socket parameters */
 689        sk = sock->sk;
 690        sk->sk_reuse = 1;
 691        sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
 692        sk->sk_allocation = GFP_ATOMIC;
 693
 694        iscsi_sw_tcp_conn_set_callbacks(conn);
 695        tcp_sw_conn->sendpage = tcp_sw_conn->sock->ops->sendpage;
 696        /*
 697         * set receive state machine into initial state
 698         */
 699        iscsi_tcp_hdr_recv_prep(tcp_conn);
 700        return 0;
 701
 702free_socket:
 703        sockfd_put(sock);
 704        return err;
 705}
 706
 707static int iscsi_sw_tcp_conn_set_param(struct iscsi_cls_conn *cls_conn,
 708                                       enum iscsi_param param, char *buf,
 709                                       int buflen)
 710{
 711        struct iscsi_conn *conn = cls_conn->dd_data;
 712        struct iscsi_session *session = conn->session;
 713        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 714        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 715        int value;
 716
 717        switch(param) {
 718        case ISCSI_PARAM_HDRDGST_EN:
 719                iscsi_set_param(cls_conn, param, buf, buflen);
 720                break;
 721        case ISCSI_PARAM_DATADGST_EN:
 722                iscsi_set_param(cls_conn, param, buf, buflen);
 723                tcp_sw_conn->sendpage = conn->datadgst_en ?
 724                        sock_no_sendpage : tcp_sw_conn->sock->ops->sendpage;
 725                break;
 726        case ISCSI_PARAM_MAX_R2T:
 727                sscanf(buf, "%d", &value);
 728                if (value <= 0 || !is_power_of_2(value))
 729                        return -EINVAL;
 730                if (session->max_r2t == value)
 731                        break;
 732                iscsi_tcp_r2tpool_free(session);
 733                iscsi_set_param(cls_conn, param, buf, buflen);
 734                if (iscsi_tcp_r2tpool_alloc(session))
 735                        return -ENOMEM;
 736                break;
 737        default:
 738                return iscsi_set_param(cls_conn, param, buf, buflen);
 739        }
 740
 741        return 0;
 742}
 743
 744static int iscsi_sw_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
 745                                       enum iscsi_param param, char *buf)
 746{
 747        struct iscsi_conn *conn = cls_conn->dd_data;
 748        int len;
 749
 750        switch(param) {
 751        case ISCSI_PARAM_CONN_PORT:
 752                spin_lock_bh(&conn->session->lock);
 753                len = sprintf(buf, "%hu\n", conn->portal_port);
 754                spin_unlock_bh(&conn->session->lock);
 755                break;
 756        case ISCSI_PARAM_CONN_ADDRESS:
 757                spin_lock_bh(&conn->session->lock);
 758                len = sprintf(buf, "%s\n", conn->portal_address);
 759                spin_unlock_bh(&conn->session->lock);
 760                break;
 761        default:
 762                return iscsi_conn_get_param(cls_conn, param, buf);
 763        }
 764
 765        return len;
 766}
 767
 768static void
 769iscsi_sw_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
 770                            struct iscsi_stats *stats)
 771{
 772        struct iscsi_conn *conn = cls_conn->dd_data;
 773        struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 774        struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 775
 776        stats->custom_length = 3;
 777        strcpy(stats->custom[0].desc, "tx_sendpage_failures");
 778        stats->custom[0].value = tcp_sw_conn->sendpage_failures_cnt;
 779        strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
 780        stats->custom[1].value = tcp_sw_conn->discontiguous_hdr_cnt;
 781        strcpy(stats->custom[2].desc, "eh_abort_cnt");
 782        stats->custom[2].value = conn->eh_abort_cnt;
 783
 784        iscsi_tcp_conn_get_stats(cls_conn, stats);
 785}
 786
 787static struct iscsi_cls_session *
 788iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,
 789                            uint16_t qdepth, uint32_t initial_cmdsn)
 790{
 791        struct iscsi_cls_session *cls_session;
 792        struct iscsi_session *session;
 793        struct Scsi_Host *shost;
 794
 795        if (ep) {
 796                printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep);
 797                return NULL;
 798        }
 799
 800        shost = iscsi_host_alloc(&iscsi_sw_tcp_sht, 0, 1);
 801        if (!shost)
 802                return NULL;
 803        shost->transportt = iscsi_sw_tcp_scsi_transport;
 804        shost->cmd_per_lun = qdepth;
 805        shost->max_lun = iscsi_max_lun;
 806        shost->max_id = 0;
 807        shost->max_channel = 0;
 808        shost->max_cmd_len = SCSI_MAX_VARLEN_CDB_SIZE;
 809
 810        if (iscsi_host_add(shost, NULL))
 811                goto free_host;
 812
 813        cls_session = iscsi_session_setup(&iscsi_sw_tcp_transport, shost,
 814                                          cmds_max, 0,
 815                                          sizeof(struct iscsi_tcp_task) +
 816                                          sizeof(struct iscsi_sw_tcp_hdrbuf),
 817                                          initial_cmdsn, 0);
 818        if (!cls_session)
 819                goto remove_host;
 820        session = cls_session->dd_data;
 821
 822        shost->can_queue = session->scsi_cmds_max;
 823        if (iscsi_tcp_r2tpool_alloc(session))
 824                goto remove_session;
 825        return cls_session;
 826
 827remove_session:
 828        iscsi_session_teardown(cls_session);
 829remove_host:
 830        iscsi_host_remove(shost);
 831free_host:
 832        iscsi_host_free(shost);
 833        return NULL;
 834}
 835
 836static void iscsi_sw_tcp_session_destroy(struct iscsi_cls_session *cls_session)
 837{
 838        struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
 839
 840        iscsi_tcp_r2tpool_free(cls_session->dd_data);
 841        iscsi_session_teardown(cls_session);
 842
 843        iscsi_host_remove(shost);
 844        iscsi_host_free(shost);
 845}
 846
 847static int iscsi_sw_tcp_slave_alloc(struct scsi_device *sdev)
 848{
 849        set_bit(QUEUE_FLAG_BIDI, &sdev->request_queue->queue_flags);
 850        return 0;
 851}
 852
 853static int iscsi_sw_tcp_slave_configure(struct scsi_device *sdev)
 854{
 855        blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
 856        blk_queue_dma_alignment(sdev->request_queue, 0);
 857        return 0;
 858}
 859
 860static struct scsi_host_template iscsi_sw_tcp_sht = {
 861        .module                 = THIS_MODULE,
 862        .name                   = "iSCSI Initiator over TCP/IP",
 863        .queuecommand           = iscsi_queuecommand,
 864        .change_queue_depth     = iscsi_change_queue_depth,
 865        .can_queue              = ISCSI_DEF_XMIT_CMDS_MAX - 1,
 866        .sg_tablesize           = 4096,
 867        .max_sectors            = 0xFFFF,
 868        .cmd_per_lun            = ISCSI_DEF_CMD_PER_LUN,
 869        .eh_abort_handler       = iscsi_eh_abort,
 870        .eh_device_reset_handler= iscsi_eh_device_reset,
 871        .eh_target_reset_handler= iscsi_eh_target_reset,
 872        .use_clustering         = DISABLE_CLUSTERING,
 873        .slave_alloc            = iscsi_sw_tcp_slave_alloc,
 874        .slave_configure        = iscsi_sw_tcp_slave_configure,
 875        .target_alloc           = iscsi_target_alloc,
 876        .proc_name              = "iscsi_tcp",
 877        .this_id                = -1,
 878};
 879
 880static struct iscsi_transport iscsi_sw_tcp_transport = {
 881        .owner                  = THIS_MODULE,
 882        .name                   = "tcp",
 883        .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
 884                                  | CAP_DATADGST,
 885        .param_mask             = ISCSI_MAX_RECV_DLENGTH |
 886                                  ISCSI_MAX_XMIT_DLENGTH |
 887                                  ISCSI_HDRDGST_EN |
 888                                  ISCSI_DATADGST_EN |
 889                                  ISCSI_INITIAL_R2T_EN |
 890                                  ISCSI_MAX_R2T |
 891                                  ISCSI_IMM_DATA_EN |
 892                                  ISCSI_FIRST_BURST |
 893                                  ISCSI_MAX_BURST |
 894                                  ISCSI_PDU_INORDER_EN |
 895                                  ISCSI_DATASEQ_INORDER_EN |
 896                                  ISCSI_ERL |
 897                                  ISCSI_CONN_PORT |
 898                                  ISCSI_CONN_ADDRESS |
 899                                  ISCSI_EXP_STATSN |
 900                                  ISCSI_PERSISTENT_PORT |
 901                                  ISCSI_PERSISTENT_ADDRESS |
 902                                  ISCSI_TARGET_NAME | ISCSI_TPGT |
 903                                  ISCSI_USERNAME | ISCSI_PASSWORD |
 904                                  ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
 905                                  ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
 906                                  ISCSI_LU_RESET_TMO |
 907                                  ISCSI_PING_TMO | ISCSI_RECV_TMO |
 908                                  ISCSI_IFACE_NAME | ISCSI_INITIATOR_NAME,
 909        .host_param_mask        = ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
 910                                  ISCSI_HOST_INITIATOR_NAME |
 911                                  ISCSI_HOST_NETDEV_NAME,
 912        /* session management */
 913        .create_session         = iscsi_sw_tcp_session_create,
 914        .destroy_session        = iscsi_sw_tcp_session_destroy,
 915        /* connection management */
 916        .create_conn            = iscsi_sw_tcp_conn_create,
 917        .bind_conn              = iscsi_sw_tcp_conn_bind,
 918        .destroy_conn           = iscsi_sw_tcp_conn_destroy,
 919        .set_param              = iscsi_sw_tcp_conn_set_param,
 920        .get_conn_param         = iscsi_sw_tcp_conn_get_param,
 921        .get_session_param      = iscsi_session_get_param,
 922        .start_conn             = iscsi_conn_start,
 923        .stop_conn              = iscsi_sw_tcp_conn_stop,
 924        /* iscsi host params */
 925        .get_host_param         = iscsi_host_get_param,
 926        .set_host_param         = iscsi_host_set_param,
 927        /* IO */
 928        .send_pdu               = iscsi_conn_send_pdu,
 929        .get_stats              = iscsi_sw_tcp_conn_get_stats,
 930        /* iscsi task/cmd helpers */
 931        .init_task              = iscsi_tcp_task_init,
 932        .xmit_task              = iscsi_tcp_task_xmit,
 933        .cleanup_task           = iscsi_tcp_cleanup_task,
 934        /* low level pdu helpers */
 935        .xmit_pdu               = iscsi_sw_tcp_pdu_xmit,
 936        .init_pdu               = iscsi_sw_tcp_pdu_init,
 937        .alloc_pdu              = iscsi_sw_tcp_pdu_alloc,
 938        /* recovery */
 939        .session_recovery_timedout = iscsi_session_recovery_timedout,
 940};
 941
 942static int __init iscsi_sw_tcp_init(void)
 943{
 944        if (iscsi_max_lun < 1) {
 945                printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
 946                       iscsi_max_lun);
 947                return -EINVAL;
 948        }
 949
 950        iscsi_sw_tcp_scsi_transport = iscsi_register_transport(
 951                                                &iscsi_sw_tcp_transport);
 952        if (!iscsi_sw_tcp_scsi_transport)
 953                return -ENODEV;
 954
 955        return 0;
 956}
 957
 958static void __exit iscsi_sw_tcp_exit(void)
 959{
 960        iscsi_unregister_transport(&iscsi_sw_tcp_transport);
 961}
 962
 963module_init(iscsi_sw_tcp_init);
 964module_exit(iscsi_sw_tcp_exit);
 965