block/iscsi.c

   1 /*
   2  * QEMU Block driver for iSCSI images
   3  *
   4  * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25 #include "config-host.h"
  26
  27 #include <poll.h>
  28 #include <arpa/inet.h>
  29 #include "qemu-common.h"
  30 #include "qemu/config-file.h"
  31 #include "qemu/error-report.h"
  32 #include "block/block_int.h"
  33 #include "trace.h"
  34 #include "hw/scsi-defs.h"
  35
  36 #include <iscsi/iscsi.h>
  37 #include <iscsi/scsi-lowlevel.h>
  38
  39 #ifdef __linux__
  40 #include <scsi/sg.h>
  41 #include <hw/scsi-defs.h>
  42 #endif
  43
  44 typedef struct IscsiLun {
  45     struct iscsi_context *iscsi;
  46     int lun;
  47     enum scsi_inquiry_peripheral_device_type type;
  48     int block_size;
  49     uint64_t num_blocks;
  50     int events;
  51     QEMUTimer *nop_timer;
  52 } IscsiLun;
  53
  54 typedef struct IscsiAIOCB {
  55     BlockDriverAIOCB common;
  56     QEMUIOVector *qiov;
  57     QEMUBH *bh;
  58     IscsiLun *iscsilun;
  59     struct scsi_task *task;
  60     uint8_t *buf;
  61     int status;
  62     int canceled;
  63     size_t read_size;
  64     size_t read_offset;
  65 #ifdef __linux__
  66     sg_io_hdr_t *ioh;
  67 #endif
  68 } IscsiAIOCB;
  69
  70 #define NOP_INTERVAL 5000
  71 #define MAX_NOP_FAILURES 3
  72
  73 static void
  74 iscsi_bh_cb(void *p)
  75 {
  76     IscsiAIOCB *acb = p;
  77
  78     qemu_bh_delete(acb->bh);
  79
  80     if (acb->canceled == 0) {
  81         acb->common.cb(acb->common.opaque, acb->status);
  82     }
  83
  84     if (acb->task != NULL) {
  85         scsi_free_scsi_task(acb->task);
  86         acb->task = NULL;
  87     }
  88
  89     qemu_aio_release(acb);
  90 }
  91
  92 static void
  93 iscsi_schedule_bh(IscsiAIOCB *acb)
  94 {
  95     if (acb->bh) {
  96         return;
  97     }
  98     acb->bh = qemu_bh_new(iscsi_bh_cb, acb);
  99     qemu_bh_schedule(acb->bh);
 100 }
 101
 102
 103 static void
 104 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
 105                     void *private_data)
 106 {
 107     IscsiAIOCB *acb = private_data;
 108
 109     acb->status = -ECANCELED;
 110     iscsi_schedule_bh(acb);
 111 }
 112
 113 static void
 114 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
 115 {
 116     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
 117     IscsiLun *iscsilun = acb->iscsilun;
 118
 119     if (acb->status != -EINPROGRESS) {
 120         return;
 121     }
 122
 123     acb->canceled = 1;
 124
 125     /* send a task mgmt call to the target to cancel the task on the target */
 126     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
 127                                      iscsi_abort_task_cb, acb);
 128
 129     while (acb->status == -EINPROGRESS) {
 130         qemu_aio_wait();
 131     }
 132 }
 133
 134 static const AIOCBInfo iscsi_aiocb_info = {
 135     .aiocb_size         = sizeof(IscsiAIOCB),
 136     .cancel             = iscsi_aio_cancel,
 137 };
 138
 139
 140 static void iscsi_process_read(void *arg);
 141 static void iscsi_process_write(void *arg);
 142
 143 static int iscsi_process_flush(void *arg)
 144 {
 145     IscsiLun *iscsilun = arg;
 146
 147     return iscsi_queue_length(iscsilun->iscsi) > 0;
 148 }
 149
 150 static void
 151 iscsi_set_events(IscsiLun *iscsilun)
 152 {
 153     struct iscsi_context *iscsi = iscsilun->iscsi;
 154     int ev;
 155
 156     /* We always register a read handler.  */
 157     ev = POLLIN;
 158     ev |= iscsi_which_events(iscsi);
 159     if (ev != iscsilun->events) {
 160         qemu_aio_set_fd_handler(iscsi_get_fd(iscsi),
 161                       iscsi_process_read,
 162                       (ev & POLLOUT) ? iscsi_process_write : NULL,
 163                       iscsi_process_flush,
 164                       iscsilun);
 165
 166     }
 167
 168     iscsilun->events = ev;
 169 }
 170
 171 static void
 172 iscsi_process_read(void *arg)
 173 {
 174     IscsiLun *iscsilun = arg;
 175     struct iscsi_context *iscsi = iscsilun->iscsi;
 176
 177     iscsi_service(iscsi, POLLIN);
 178     iscsi_set_events(iscsilun);
 179 }
 180
 181 static void
 182 iscsi_process_write(void *arg)
 183 {
 184     IscsiLun *iscsilun = arg;
 185     struct iscsi_context *iscsi = iscsilun->iscsi;
 186
 187     iscsi_service(iscsi, POLLOUT);
 188     iscsi_set_events(iscsilun);
 189 }
 190
 191
 192 static void
 193 iscsi_aio_write16_cb(struct iscsi_context *iscsi, int status,
 194                      void *command_data, void *opaque)
 195 {
 196     IscsiAIOCB *acb = opaque;
 197
 198     trace_iscsi_aio_write16_cb(iscsi, status, acb, acb->canceled);
 199
 200     g_free(acb->buf);
 201
 202     if (acb->canceled != 0) {
 203         return;
 204     }
 205
 206     acb->status = 0;
 207     if (status < 0) {
 208         error_report("Failed to write16 data to iSCSI lun. %s",
 209                      iscsi_get_error(iscsi));
 210         acb->status = -EIO;
 211     }
 212
 213     iscsi_schedule_bh(acb);
 214 }
 215
 216 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
 217 {
 218     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
 219 }
 220
 221 static BlockDriverAIOCB *
 222 iscsi_aio_writev(BlockDriverState *bs, int64_t sector_num,
 223                  QEMUIOVector *qiov, int nb_sectors,
 224                  BlockDriverCompletionFunc *cb,
 225                  void *opaque)
 226 {
 227     IscsiLun *iscsilun = bs->opaque;
 228     struct iscsi_context *iscsi = iscsilun->iscsi;
 229     IscsiAIOCB *acb;
 230     size_t size;
 231     uint32_t num_sectors;
 232     uint64_t lba;
 233     struct iscsi_data data;
 234
 235     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 236     trace_iscsi_aio_writev(iscsi, sector_num, nb_sectors, opaque, acb);
 237
 238     acb->iscsilun = iscsilun;
 239     acb->qiov     = qiov;
 240
 241     acb->canceled   = 0;
 242     acb->bh         = NULL;
 243     acb->status     = -EINPROGRESS;
 244
 245     /* XXX we should pass the iovec to write16 to avoid the extra copy */
 246     /* this will allow us to get rid of 'buf' completely */
 247     size = nb_sectors * BDRV_SECTOR_SIZE;
 248     data.size = MIN(size, acb->qiov->size);
 249
 250     /* if the iovec only contains one buffer we can pass it directly */
 251     if (acb->qiov->niov == 1) {
 252         acb->buf = NULL;
 253         data.data = acb->qiov->iov[0].iov_base;
 254     } else {
 255         acb->buf = g_malloc(data.size);
 256         qemu_iovec_to_buf(acb->qiov, 0, acb->buf, data.size);
 257         data.data = acb->buf;
 258     }
 259
 260     acb->task = malloc(sizeof(struct scsi_task));
 261     if (acb->task == NULL) {
 262         error_report("iSCSI: Failed to allocate task for scsi WRITE16 "
 263                      "command. %s", iscsi_get_error(iscsi));
 264         qemu_aio_release(acb);
 265         return NULL;
 266     }
 267     memset(acb->task, 0, sizeof(struct scsi_task));
 268
 269     acb->task->xfer_dir = SCSI_XFER_WRITE;
 270     acb->task->cdb_size = 16;
 271     acb->task->cdb[0] = 0x8a;
 272     lba = sector_qemu2lun(sector_num, iscsilun);
 273     *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 274     *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 275     num_sectors = size / iscsilun->block_size;
 276     *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 277     acb->task->expxferlen = size;
 278
 279     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 280                                  iscsi_aio_write16_cb,
 281                                  &data,
 282                                  acb) != 0) {
 283         scsi_free_scsi_task(acb->task);
 284         g_free(acb->buf);
 285         qemu_aio_release(acb);
 286         return NULL;
 287     }
 288
 289     iscsi_set_events(iscsilun);
 290
 291     return &acb->common;
 292 }
 293
 294 static void
 295 iscsi_aio_read16_cb(struct iscsi_context *iscsi, int status,
 296                     void *command_data, void *opaque)
 297 {
 298     IscsiAIOCB *acb = opaque;
 299
 300     trace_iscsi_aio_read16_cb(iscsi, status, acb, acb->canceled);
 301
 302     if (acb->canceled != 0) {
 303         return;
 304     }
 305
 306     acb->status = 0;
 307     if (status != 0) {
 308         error_report("Failed to read16 data from iSCSI lun. %s",
 309                      iscsi_get_error(iscsi));
 310         acb->status = -EIO;
 311     }
 312
 313     iscsi_schedule_bh(acb);
 314 }
 315
 316 static BlockDriverAIOCB *
 317 iscsi_aio_readv(BlockDriverState *bs, int64_t sector_num,
 318                 QEMUIOVector *qiov, int nb_sectors,
 319                 BlockDriverCompletionFunc *cb,
 320                 void *opaque)
 321 {
 322     IscsiLun *iscsilun = bs->opaque;
 323     struct iscsi_context *iscsi = iscsilun->iscsi;
 324     IscsiAIOCB *acb;
 325     size_t qemu_read_size;
 326     int i;
 327     uint64_t lba;
 328     uint32_t num_sectors;
 329
 330     qemu_read_size = BDRV_SECTOR_SIZE * (size_t)nb_sectors;
 331
 332     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 333     trace_iscsi_aio_readv(iscsi, sector_num, nb_sectors, opaque, acb);
 334
 335     acb->iscsilun = iscsilun;
 336     acb->qiov     = qiov;
 337
 338     acb->canceled    = 0;
 339     acb->bh          = NULL;
 340     acb->status      = -EINPROGRESS;
 341     acb->read_size   = qemu_read_size;
 342     acb->buf         = NULL;
 343
 344     /* If LUN blocksize is bigger than BDRV_BLOCK_SIZE a read from QEMU
 345      * may be misaligned to the LUN, so we may need to read some extra
 346      * data.
 347      */
 348     acb->read_offset = 0;
 349     if (iscsilun->block_size > BDRV_SECTOR_SIZE) {
 350         uint64_t bdrv_offset = BDRV_SECTOR_SIZE * sector_num;
 351
 352         acb->read_offset  = bdrv_offset % iscsilun->block_size;
 353     }
 354
 355     num_sectors  = (qemu_read_size + iscsilun->block_size
 356                     + acb->read_offset - 1)
 357                     / iscsilun->block_size;
 358
 359     acb->task = malloc(sizeof(struct scsi_task));
 360     if (acb->task == NULL) {
 361         error_report("iSCSI: Failed to allocate task for scsi READ16 "
 362                      "command. %s", iscsi_get_error(iscsi));
 363         qemu_aio_release(acb);
 364         return NULL;
 365     }
 366     memset(acb->task, 0, sizeof(struct scsi_task));
 367
 368     acb->task->xfer_dir = SCSI_XFER_READ;
 369     lba = sector_qemu2lun(sector_num, iscsilun);
 370     acb->task->expxferlen = qemu_read_size;
 371
 372     switch (iscsilun->type) {
 373     case TYPE_DISK:
 374         acb->task->cdb_size = 16;
 375         acb->task->cdb[0]  = 0x88;
 376         *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 377         *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 378         *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 379         break;
 380     default:
 381         acb->task->cdb_size = 10;
 382         acb->task->cdb[0]  = 0x28;
 383         *(uint32_t *)&acb->task->cdb[2] = htonl(lba);
 384         *(uint16_t *)&acb->task->cdb[7] = htons(num_sectors);
 385         break;
 386     }
 387
 388     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 389                                  iscsi_aio_read16_cb,
 390                                  NULL,
 391                                  acb) != 0) {
 392         scsi_free_scsi_task(acb->task);
 393         qemu_aio_release(acb);
 394         return NULL;
 395     }
 396
 397     for (i = 0; i < acb->qiov->niov; i++) {
 398         scsi_task_add_data_in_buffer(acb->task,
 399                 acb->qiov->iov[i].iov_len,
 400                 acb->qiov->iov[i].iov_base);
 401     }
 402
 403     iscsi_set_events(iscsilun);
 404
 405     return &acb->common;
 406 }
 407
 408
 409 static void
 410 iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
 411                      void *command_data, void *opaque)
 412 {
 413     IscsiAIOCB *acb = opaque;
 414
 415     if (acb->canceled != 0) {
 416         return;
 417     }
 418
 419     acb->status = 0;
 420     if (status < 0) {
 421         error_report("Failed to sync10 data on iSCSI lun. %s",
 422                      iscsi_get_error(iscsi));
 423         acb->status = -EIO;
 424     }
 425
 426     iscsi_schedule_bh(acb);
 427 }
 428
 429 static BlockDriverAIOCB *
 430 iscsi_aio_flush(BlockDriverState *bs,
 431                 BlockDriverCompletionFunc *cb, void *opaque)
 432 {
 433     IscsiLun *iscsilun = bs->opaque;
 434     struct iscsi_context *iscsi = iscsilun->iscsi;
 435     IscsiAIOCB *acb;
 436
 437     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 438
 439     acb->iscsilun = iscsilun;
 440     acb->canceled   = 0;
 441     acb->bh         = NULL;
 442     acb->status     = -EINPROGRESS;
 443
 444     acb->task = iscsi_synchronizecache10_task(iscsi, iscsilun->lun,
 445                                          0, 0, 0, 0,
 446                                          iscsi_synccache10_cb,
 447                                          acb);
 448     if (acb->task == NULL) {
 449         error_report("iSCSI: Failed to send synchronizecache10 command. %s",
 450                      iscsi_get_error(iscsi));
 451         qemu_aio_release(acb);
 452         return NULL;
 453     }
 454
 455     iscsi_set_events(iscsilun);
 456
 457     return &acb->common;
 458 }
 459
 460 static void
 461 iscsi_unmap_cb(struct iscsi_context *iscsi, int status,
 462                      void *command_data, void *opaque)
 463 {
 464     IscsiAIOCB *acb = opaque;
 465
 466     if (acb->canceled != 0) {
 467         return;
 468     }
 469
 470     acb->status = 0;
 471     if (status < 0) {
 472         error_report("Failed to unmap data on iSCSI lun. %s",
 473                      iscsi_get_error(iscsi));
 474         acb->status = -EIO;
 475     }
 476
 477     iscsi_schedule_bh(acb);
 478 }
 479
 480 static BlockDriverAIOCB *
 481 iscsi_aio_discard(BlockDriverState *bs,
 482                   int64_t sector_num, int nb_sectors,
 483                   BlockDriverCompletionFunc *cb, void *opaque)
 484 {
 485     IscsiLun *iscsilun = bs->opaque;
 486     struct iscsi_context *iscsi = iscsilun->iscsi;
 487     IscsiAIOCB *acb;
 488     struct unmap_list list[1];
 489
 490     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 491
 492     acb->iscsilun = iscsilun;
 493     acb->canceled   = 0;
 494     acb->bh         = NULL;
 495     acb->status     = -EINPROGRESS;
 496
 497     list[0].lba = sector_qemu2lun(sector_num, iscsilun);
 498     list[0].num = nb_sectors * BDRV_SECTOR_SIZE / iscsilun->block_size;
 499
 500     acb->task = iscsi_unmap_task(iscsi, iscsilun->lun,
 501                                  0, 0, &list[0], 1,
 502                                  iscsi_unmap_cb,
 503                                  acb);
 504     if (acb->task == NULL) {
 505         error_report("iSCSI: Failed to send unmap command. %s",
 506                      iscsi_get_error(iscsi));
 507         qemu_aio_release(acb);
 508         return NULL;
 509     }
 510
 511     iscsi_set_events(iscsilun);
 512
 513     return &acb->common;
 514 }
 515
 516 #ifdef __linux__
 517 static void
 518 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 519                      void *command_data, void *opaque)
 520 {
 521     IscsiAIOCB *acb = opaque;
 522
 523     if (acb->canceled != 0) {
 524         return;
 525     }
 526
 527     acb->status = 0;
 528     if (status < 0) {
 529         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 530                      iscsi_get_error(iscsi));
 531         acb->status = -EIO;
 532     }
 533
 534     acb->ioh->driver_status = 0;
 535     acb->ioh->host_status   = 0;
 536     acb->ioh->resid         = 0;
 537
 538 #define SG_ERR_DRIVER_SENSE    0x08
 539
 540     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 541         int ss;
 542
 543         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 544
 545         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 546         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 547              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 548         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 549     }
 550
 551     iscsi_schedule_bh(acb);
 552 }
 553
 554 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 555         unsigned long int req, void *buf,
 556         BlockDriverCompletionFunc *cb, void *opaque)
 557 {
 558     IscsiLun *iscsilun = bs->opaque;
 559     struct iscsi_context *iscsi = iscsilun->iscsi;
 560     struct iscsi_data data;
 561     IscsiAIOCB *acb;
 562
 563     assert(req == SG_IO);
 564
 565     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 566
 567     acb->iscsilun = iscsilun;
 568     acb->canceled    = 0;
 569     acb->bh          = NULL;
 570     acb->status      = -EINPROGRESS;
 571     acb->buf         = NULL;
 572     acb->ioh         = buf;
 573
 574     acb->task = malloc(sizeof(struct scsi_task));
 575     if (acb->task == NULL) {
 576         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 577                      iscsi_get_error(iscsi));
 578         qemu_aio_release(acb);
 579         return NULL;
 580     }
 581     memset(acb->task, 0, sizeof(struct scsi_task));
 582
 583     switch (acb->ioh->dxfer_direction) {
 584     case SG_DXFER_TO_DEV:
 585         acb->task->xfer_dir = SCSI_XFER_WRITE;
 586         break;
 587     case SG_DXFER_FROM_DEV:
 588         acb->task->xfer_dir = SCSI_XFER_READ;
 589         break;
 590     default:
 591         acb->task->xfer_dir = SCSI_XFER_NONE;
 592         break;
 593     }
 594
 595     acb->task->cdb_size = acb->ioh->cmd_len;
 596     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 597     acb->task->expxferlen = acb->ioh->dxfer_len;
 598
 599     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 600         data.data = acb->ioh->dxferp;
 601         data.size = acb->ioh->dxfer_len;
 602     }
 603     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 604                                  iscsi_aio_ioctl_cb,
 605                                  (acb->task->xfer_dir == SCSI_XFER_WRITE) ?
 606                                      &data : NULL,
 607                                  acb) != 0) {
 608         scsi_free_scsi_task(acb->task);
 609         qemu_aio_release(acb);
 610         return NULL;
 611     }
 612
 613     /* tell libiscsi to read straight into the buffer we got from ioctl */
 614     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 615         scsi_task_add_data_in_buffer(acb->task,
 616                                      acb->ioh->dxfer_len,
 617                                      acb->ioh->dxferp);
 618     }
 619
 620     iscsi_set_events(iscsilun);
 621
 622     return &acb->common;
 623 }
 624
 625
 626 static void ioctl_cb(void *opaque, int status)
 627 {
 628     int *p_status = opaque;
 629     *p_status = status;
 630 }
 631
 632 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 633 {
 634     IscsiLun *iscsilun = bs->opaque;
 635     int status;
 636
 637     switch (req) {
 638     case SG_GET_VERSION_NUM:
 639         *(int *)buf = 30000;
 640         break;
 641     case SG_GET_SCSI_ID:
 642         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 643         break;
 644     case SG_IO:
 645         status = -EINPROGRESS;
 646         iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status);
 647
 648         while (status == -EINPROGRESS) {
 649             qemu_aio_wait();
 650         }
 651
 652         return 0;
 653     default:
 654         return -1;
 655     }
 656     return 0;
 657 }
 658 #endif
 659
 660 static int64_t
 661 iscsi_getlength(BlockDriverState *bs)
 662 {
 663     IscsiLun *iscsilun = bs->opaque;
 664     int64_t len;
 665
 666     len  = iscsilun->num_blocks;
 667     len *= iscsilun->block_size;
 668
 669     return len;
 670 }
 671
 672 static int parse_chap(struct iscsi_context *iscsi, const char *target)
 673 {
 674     QemuOptsList *list;
 675     QemuOpts *opts;
 676     const char *user = NULL;
 677     const char *password = NULL;
 678
 679     list = qemu_find_opts("iscsi");
 680     if (!list) {
 681         return 0;
 682     }
 683
 684     opts = qemu_opts_find(list, target);
 685     if (opts == NULL) {
 686         opts = QTAILQ_FIRST(&list->head);
 687         if (!opts) {
 688             return 0;
 689         }
 690     }
 691
 692     user = qemu_opt_get(opts, "user");
 693     if (!user) {
 694         return 0;
 695     }
 696
 697     password = qemu_opt_get(opts, "password");
 698     if (!password) {
 699         error_report("CHAP username specified but no password was given");
 700         return -1;
 701     }
 702
 703     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
 704         error_report("Failed to set initiator username and password");
 705         return -1;
 706     }
 707
 708     return 0;
 709 }
 710
 711 static void parse_header_digest(struct iscsi_context *iscsi, const char *target)
 712 {
 713     QemuOptsList *list;
 714     QemuOpts *opts;
 715     const char *digest = NULL;
 716
 717     list = qemu_find_opts("iscsi");
 718     if (!list) {
 719         return;
 720     }
 721
 722     opts = qemu_opts_find(list, target);
 723     if (opts == NULL) {
 724         opts = QTAILQ_FIRST(&list->head);
 725         if (!opts) {
 726             return;
 727         }
 728     }
 729
 730     digest = qemu_opt_get(opts, "header-digest");
 731     if (!digest) {
 732         return;
 733     }
 734
 735     if (!strcmp(digest, "CRC32C")) {
 736         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
 737     } else if (!strcmp(digest, "NONE")) {
 738         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
 739     } else if (!strcmp(digest, "CRC32C-NONE")) {
 740         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
 741     } else if (!strcmp(digest, "NONE-CRC32C")) {
 742         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 743     } else {
 744         error_report("Invalid header-digest setting : %s", digest);
 745     }
 746 }
 747
 748 static char *parse_initiator_name(const char *target)
 749 {
 750     QemuOptsList *list;
 751     QemuOpts *opts;
 752     const char *name = NULL;
 753     const char *iscsi_name = qemu_get_vm_name();
 754
 755     list = qemu_find_opts("iscsi");
 756     if (list) {
 757         opts = qemu_opts_find(list, target);
 758         if (!opts) {
 759             opts = QTAILQ_FIRST(&list->head);
 760         }
 761         if (opts) {
 762             name = qemu_opt_get(opts, "initiator-name");
 763         }
 764     }
 765
 766     if (name) {
 767         return g_strdup(name);
 768     } else {
 769         return g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
 770                                iscsi_name ? ":" : "",
 771                                iscsi_name ? iscsi_name : "");
 772     }
 773 }
 774
 775 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
 776 static void iscsi_nop_timed_event(void *opaque)
 777 {
 778     IscsiLun *iscsilun = opaque;
 779
 780     if (iscsi_get_nops_in_flight(iscsilun->iscsi) > MAX_NOP_FAILURES) {
 781         error_report("iSCSI: NOP timeout. Reconnecting...");
 782         iscsi_reconnect(iscsilun->iscsi);
 783     }
 784
 785     if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
 786         error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
 787         return;
 788     }
 789
 790     qemu_mod_timer(iscsilun->nop_timer, qemu_get_clock_ms(rt_clock) + NOP_INTERVAL);
 791     iscsi_set_events(iscsilun);
 792 }
 793 #endif
 794
 795 /*
 796  * We support iscsi url's on the form
 797  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
 798  */
 799 static int iscsi_open(BlockDriverState *bs, const char *filename, int flags)
 800 {
 801     IscsiLun *iscsilun = bs->opaque;
 802     struct iscsi_context *iscsi = NULL;
 803     struct iscsi_url *iscsi_url = NULL;
 804     struct scsi_task *task = NULL;
 805     struct scsi_inquiry_standard *inq = NULL;
 806     struct scsi_readcapacity10 *rc10 = NULL;
 807     struct scsi_readcapacity16 *rc16 = NULL;
 808     char *initiator_name = NULL;
 809     int ret;
 810
 811     if ((BDRV_SECTOR_SIZE % 512) != 0) {
 812         error_report("iSCSI: Invalid BDRV_SECTOR_SIZE. "
 813                      "BDRV_SECTOR_SIZE(%lld) is not a multiple "
 814                      "of 512", BDRV_SECTOR_SIZE);
 815         return -EINVAL;
 816     }
 817
 818     iscsi_url = iscsi_parse_full_url(iscsi, filename);
 819     if (iscsi_url == NULL) {
 820         error_report("Failed to parse URL : %s", filename);
 821         ret = -EINVAL;
 822         goto out;
 823     }
 824
 825     memset(iscsilun, 0, sizeof(IscsiLun));
 826
 827     initiator_name = parse_initiator_name(iscsi_url->target);
 828
 829     iscsi = iscsi_create_context(initiator_name);
 830     if (iscsi == NULL) {
 831         error_report("iSCSI: Failed to create iSCSI context.");
 832         ret = -ENOMEM;
 833         goto out;
 834     }
 835
 836     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
 837         error_report("iSCSI: Failed to set target name.");
 838         ret = -EINVAL;
 839         goto out;
 840     }
 841
 842     if (iscsi_url->user != NULL) {
 843         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
 844                                               iscsi_url->passwd);
 845         if (ret != 0) {
 846             error_report("Failed to set initiator username and password");
 847             ret = -EINVAL;
 848             goto out;
 849         }
 850     }
 851
 852     /* check if we got CHAP username/password via the options */
 853     if (parse_chap(iscsi, iscsi_url->target) != 0) {
 854         error_report("iSCSI: Failed to set CHAP user/password");
 855         ret = -EINVAL;
 856         goto out;
 857     }
 858
 859     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
 860         error_report("iSCSI: Failed to set session type to normal.");
 861         ret = -EINVAL;
 862         goto out;
 863     }
 864
 865     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 866
 867     /* check if we got HEADER_DIGEST via the options */
 868     parse_header_digest(iscsi, iscsi_url->target);
 869
 870     if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
 871         error_report("iSCSI: Failed to connect to LUN : %s",
 872             iscsi_get_error(iscsi));
 873         ret = -EINVAL;
 874         goto out;
 875     }
 876
 877     iscsilun->iscsi = iscsi;
 878     iscsilun->lun   = iscsi_url->lun;
 879
 880     task = iscsi_inquiry_sync(iscsi, iscsilun->lun, 0, 0, 36);
 881
 882     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
 883         error_report("iSCSI: failed to send inquiry command.");
 884         ret = -EINVAL;
 885         goto out;
 886     }
 887
 888     inq = scsi_datain_unmarshall(task);
 889     if (inq == NULL) {
 890         error_report("iSCSI: Failed to unmarshall inquiry data.");
 891         ret = -EINVAL;
 892         goto out;
 893     }
 894
 895     iscsilun->type = inq->periperal_device_type;
 896
 897     scsi_free_scsi_task(task);
 898
 899     switch (iscsilun->type) {
 900     case TYPE_DISK:
 901         task = iscsi_readcapacity16_sync(iscsi, iscsilun->lun);
 902         if (task == NULL || task->status != SCSI_STATUS_GOOD) {
 903             error_report("iSCSI: failed to send readcapacity16 command.");
 904             ret = -EINVAL;
 905             goto out;
 906         }
 907         rc16 = scsi_datain_unmarshall(task);
 908         if (rc16 == NULL) {
 909             error_report("iSCSI: Failed to unmarshall readcapacity16 data.");
 910             ret = -EINVAL;
 911             goto out;
 912         }
 913         iscsilun->block_size = rc16->block_length;
 914         iscsilun->num_blocks = rc16->returned_lba + 1;
 915         break;
 916     case TYPE_ROM:
 917         task = iscsi_readcapacity10_sync(iscsi, iscsilun->lun, 0, 0);
 918         if (task == NULL || task->status != SCSI_STATUS_GOOD) {
 919             error_report("iSCSI: failed to send readcapacity10 command.");
 920             ret = -EINVAL;
 921             goto out;
 922         }
 923         rc10 = scsi_datain_unmarshall(task);
 924         if (rc10 == NULL) {
 925             error_report("iSCSI: Failed to unmarshall readcapacity10 data.");
 926             ret = -EINVAL;
 927             goto out;
 928         }
 929         iscsilun->block_size = rc10->block_size;
 930         if (rc10->lba == 0) {
 931             /* blank disk loaded */
 932             iscsilun->num_blocks = 0;
 933         } else {
 934             iscsilun->num_blocks = rc10->lba + 1;
 935         }
 936         break;
 937     default:
 938         break;
 939     }
 940
 941     bs->total_sectors    = iscsilun->num_blocks *
 942                            iscsilun->block_size / BDRV_SECTOR_SIZE ;
 943
 944     /* Medium changer or tape. We dont have any emulation for this so this must
 945      * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
 946      * to read from the device to guess the image format.
 947      */
 948     if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
 949         iscsilun->type == TYPE_TAPE) {
 950         bs->sg = 1;
 951     }
 952
 953     ret = 0;
 954
 955 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
 956     /* Set up a timer for sending out iSCSI NOPs */
 957     iscsilun->nop_timer = qemu_new_timer_ms(rt_clock, iscsi_nop_timed_event, iscsilun);
 958     qemu_mod_timer(iscsilun->nop_timer, qemu_get_clock_ms(rt_clock) + NOP_INTERVAL);
 959 #endif
 960
 961 out:
 962     if (initiator_name != NULL) {
 963         g_free(initiator_name);
 964     }
 965     if (iscsi_url != NULL) {
 966         iscsi_destroy_url(iscsi_url);
 967     }
 968     if (task != NULL) {
 969         scsi_free_scsi_task(task);
 970     }
 971
 972     if (ret) {
 973         if (iscsi != NULL) {
 974             iscsi_destroy_context(iscsi);
 975         }
 976         memset(iscsilun, 0, sizeof(IscsiLun));
 977     }
 978     return ret;
 979 }
 980
 981 static void iscsi_close(BlockDriverState *bs)
 982 {
 983     IscsiLun *iscsilun = bs->opaque;
 984     struct iscsi_context *iscsi = iscsilun->iscsi;
 985
 986     if (iscsilun->nop_timer) {
 987         qemu_del_timer(iscsilun->nop_timer);
 988         qemu_free_timer(iscsilun->nop_timer);
 989     }
 990     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL, NULL);
 991     iscsi_destroy_context(iscsi);
 992     memset(iscsilun, 0, sizeof(IscsiLun));
 993 }
 994
 995 static int iscsi_has_zero_init(BlockDriverState *bs)
 996 {
 997     return 0;
 998 }
 999
1000 static int iscsi_create(const char *filename, QEMUOptionParameter *options)
1001 {
1002     int ret = 0;
1003     int64_t total_size = 0;
1004     BlockDriverState bs;
1005     IscsiLun *iscsilun = NULL;
1006
1007     memset(&bs, 0, sizeof(BlockDriverState));
1008
1009     /* Read out options */
1010     while (options && options->name) {
1011         if (!strcmp(options->name, "size")) {
1012             total_size = options->value.n / BDRV_SECTOR_SIZE;
1013         }
1014         options++;
1015     }
1016
1017     bs.opaque = g_malloc0(sizeof(struct IscsiLun));
1018     iscsilun = bs.opaque;
1019
1020     ret = iscsi_open(&bs, filename, 0);
1021     if (ret != 0) {
1022         goto out;
1023     }
1024     if (iscsilun->nop_timer) {
1025         qemu_del_timer(iscsilun->nop_timer);
1026         qemu_free_timer(iscsilun->nop_timer);
1027     }
1028     if (iscsilun->type != TYPE_DISK) {
1029         ret = -ENODEV;
1030         goto out;
1031     }
1032     if (bs.total_sectors < total_size) {
1033         ret = -ENOSPC;
1034     }
1035
1036     ret = 0;
1037 out:
1038     if (iscsilun->iscsi != NULL) {
1039         iscsi_destroy_context(iscsilun->iscsi);
1040     }
1041     g_free(bs.opaque);
1042     return ret;
1043 }
1044
1045 static QEMUOptionParameter iscsi_create_options[] = {
1046     {
1047         .name = BLOCK_OPT_SIZE,
1048         .type = OPT_SIZE,
1049         .help = "Virtual disk size"
1050     },
1051     { NULL }
1052 };
1053
1054 static BlockDriver bdrv_iscsi = {
1055     .format_name     = "iscsi",
1056     .protocol_name   = "iscsi",
1057
1058     .instance_size   = sizeof(IscsiLun),
1059     .bdrv_file_open  = iscsi_open,
1060     .bdrv_close      = iscsi_close,
1061     .bdrv_create     = iscsi_create,
1062     .create_options  = iscsi_create_options,
1063
1064     .bdrv_getlength  = iscsi_getlength,
1065
1066     .bdrv_aio_readv  = iscsi_aio_readv,
1067     .bdrv_aio_writev = iscsi_aio_writev,
1068     .bdrv_aio_flush  = iscsi_aio_flush,
1069
1070     .bdrv_aio_discard = iscsi_aio_discard,
1071     .bdrv_has_zero_init = iscsi_has_zero_init,
1072
1073 #ifdef __linux__
1074     .bdrv_ioctl       = iscsi_ioctl,
1075     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1076 #endif
1077 };
1078
1079 static void iscsi_block_init(void)
1080 {
1081     bdrv_register(&bdrv_iscsi);
1082 }
1083
1084 block_init(iscsi_block_init);