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-error.h"
  31 #include "block_int.h"
  32 #include "trace.h"
  33 #include "hw/scsi-defs.h"
  34
  35 #include <iscsi/iscsi.h>
  36 #include <iscsi/scsi-lowlevel.h>
  37
  38 #ifdef __linux__
  39 #include <scsi/sg.h>
  40 #include <hw/scsi-defs.h>
  41 #endif
  42
  43 typedef struct IscsiLun {
  44     struct iscsi_context *iscsi;
  45     int lun;
  46     enum scsi_inquiry_peripheral_device_type type;
  47     int block_size;
  48     uint64_t num_blocks;
  49     int events;
  50 } IscsiLun;
  51
  52 typedef struct IscsiAIOCB {
  53     BlockDriverAIOCB common;
  54     QEMUIOVector *qiov;
  55     QEMUBH *bh;
  56     IscsiLun *iscsilun;
  57     struct scsi_task *task;
  58     uint8_t *buf;
  59     int status;
  60     int canceled;
  61     size_t read_size;
  62     size_t read_offset;
  63 #ifdef __linux__
  64     sg_io_hdr_t *ioh;
  65 #endif
  66 } IscsiAIOCB;
  67
  68 struct IscsiTask {
  69     IscsiLun *iscsilun;
  70     BlockDriverState *bs;
  71     int status;
  72     int complete;
  73 };
  74
  75 static void
  76 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
  77                     void *private_data)
  78 {
  79 }
  80
  81 static void
  82 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
  83 {
  84     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
  85     IscsiLun *iscsilun = acb->iscsilun;
  86
  87     acb->common.cb(acb->common.opaque, -ECANCELED);
  88     acb->canceled = 1;
  89
  90     /* send a task mgmt call to the target to cancel the task on the target */
  91     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
  92                                      iscsi_abort_task_cb, NULL);
  93
  94     /* then also cancel the task locally in libiscsi */
  95     iscsi_scsi_task_cancel(iscsilun->iscsi, acb->task);
  96 }
  97
  98 static AIOPool iscsi_aio_pool = {
  99     .aiocb_size         = sizeof(IscsiAIOCB),
 100     .cancel             = iscsi_aio_cancel,
 101 };
 102
 103
 104 static void iscsi_process_read(void *arg);
 105 static void iscsi_process_write(void *arg);
 106
 107 static int iscsi_process_flush(void *arg)
 108 {
 109     IscsiLun *iscsilun = arg;
 110
 111     return iscsi_queue_length(iscsilun->iscsi) > 0;
 112 }
 113
 114 static void
 115 iscsi_set_events(IscsiLun *iscsilun)
 116 {
 117     struct iscsi_context *iscsi = iscsilun->iscsi;
 118     int ev;
 119
 120     /* We always register a read handler.  */
 121     ev = POLLIN;
 122     ev |= iscsi_which_events(iscsi);
 123     if (ev != iscsilun->events) {
 124         qemu_aio_set_fd_handler(iscsi_get_fd(iscsi),
 125                       iscsi_process_read,
 126                       (ev & POLLOUT) ? iscsi_process_write : NULL,
 127                       iscsi_process_flush,
 128                       iscsilun);
 129
 130     }
 131
 132     /* If we just added an event, the callback might be delayed
 133      * unless we call qemu_notify_event().
 134      */
 135     if (ev & ~iscsilun->events) {
 136         qemu_notify_event();
 137     }
 138     iscsilun->events = ev;
 139 }
 140
 141 static void
 142 iscsi_process_read(void *arg)
 143 {
 144     IscsiLun *iscsilun = arg;
 145     struct iscsi_context *iscsi = iscsilun->iscsi;
 146
 147     iscsi_service(iscsi, POLLIN);
 148     iscsi_set_events(iscsilun);
 149 }
 150
 151 static void
 152 iscsi_process_write(void *arg)
 153 {
 154     IscsiLun *iscsilun = arg;
 155     struct iscsi_context *iscsi = iscsilun->iscsi;
 156
 157     iscsi_service(iscsi, POLLOUT);
 158     iscsi_set_events(iscsilun);
 159 }
 160
 161
 162 static int
 163 iscsi_schedule_bh(QEMUBHFunc *cb, IscsiAIOCB *acb)
 164 {
 165     acb->bh = qemu_bh_new(cb, acb);
 166     if (!acb->bh) {
 167         error_report("oom: could not create iscsi bh");
 168         return -EIO;
 169     }
 170
 171     qemu_bh_schedule(acb->bh);
 172     return 0;
 173 }
 174
 175 static void
 176 iscsi_readv_writev_bh_cb(void *p)
 177 {
 178     IscsiAIOCB *acb = p;
 179
 180     qemu_bh_delete(acb->bh);
 181
 182     if (acb->canceled == 0) {
 183         acb->common.cb(acb->common.opaque, acb->status);
 184     }
 185
 186     qemu_aio_release(acb);
 187 }
 188
 189
 190 static void
 191 iscsi_aio_write16_cb(struct iscsi_context *iscsi, int status,
 192                      void *command_data, void *opaque)
 193 {
 194     IscsiAIOCB *acb = opaque;
 195
 196     trace_iscsi_aio_write16_cb(iscsi, status, acb, acb->canceled);
 197
 198     g_free(acb->buf);
 199
 200     if (acb->canceled != 0) {
 201         qemu_aio_release(acb);
 202         scsi_free_scsi_task(acb->task);
 203         acb->task = NULL;
 204         return;
 205     }
 206
 207     acb->status = 0;
 208     if (status < 0) {
 209         error_report("Failed to write16 data to iSCSI lun. %s",
 210                      iscsi_get_error(iscsi));
 211         acb->status = -EIO;
 212     }
 213
 214     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 215     scsi_free_scsi_task(acb->task);
 216     acb->task = NULL;
 217 }
 218
 219 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
 220 {
 221     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
 222 }
 223
 224 static BlockDriverAIOCB *
 225 iscsi_aio_writev(BlockDriverState *bs, int64_t sector_num,
 226                  QEMUIOVector *qiov, int nb_sectors,
 227                  BlockDriverCompletionFunc *cb,
 228                  void *opaque)
 229 {
 230     IscsiLun *iscsilun = bs->opaque;
 231     struct iscsi_context *iscsi = iscsilun->iscsi;
 232     IscsiAIOCB *acb;
 233     size_t size;
 234     uint32_t num_sectors;
 235     uint64_t lba;
 236     struct iscsi_data data;
 237
 238     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 239     trace_iscsi_aio_writev(iscsi, sector_num, nb_sectors, opaque, acb);
 240
 241     acb->iscsilun = iscsilun;
 242     acb->qiov     = qiov;
 243
 244     acb->canceled   = 0;
 245
 246     /* XXX we should pass the iovec to write16 to avoid the extra copy */
 247     /* this will allow us to get rid of 'buf' completely */
 248     size = nb_sectors * BDRV_SECTOR_SIZE;
 249     acb->buf = g_malloc(size);
 250     qemu_iovec_to_buffer(acb->qiov, acb->buf);
 251
 252
 253     acb->task = malloc(sizeof(struct scsi_task));
 254     if (acb->task == NULL) {
 255         error_report("iSCSI: Failed to allocate task for scsi WRITE16 "
 256                      "command. %s", iscsi_get_error(iscsi));
 257         qemu_aio_release(acb);
 258         return NULL;
 259     }
 260     memset(acb->task, 0, sizeof(struct scsi_task));
 261
 262     acb->task->xfer_dir = SCSI_XFER_WRITE;
 263     acb->task->cdb_size = 16;
 264     acb->task->cdb[0] = 0x8a;
 265     if (!(bs->open_flags & BDRV_O_CACHE_WB)) {
 266         /* set FUA on writes when cache mode is write through */
 267         acb->task->cdb[1] |= 0x04;
 268     }
 269     lba = sector_qemu2lun(sector_num, iscsilun);
 270     *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 271     *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 272     num_sectors = size / iscsilun->block_size;
 273     *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 274     acb->task->expxferlen = size;
 275
 276     data.data = acb->buf;
 277     data.size = 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         qemu_aio_release(acb);
 304         scsi_free_scsi_task(acb->task);
 305         acb->task = NULL;
 306         return;
 307     }
 308
 309     acb->status = 0;
 310     if (status != 0) {
 311         error_report("Failed to read16 data from iSCSI lun. %s",
 312                      iscsi_get_error(iscsi));
 313         acb->status = -EIO;
 314     }
 315
 316     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 317     scsi_free_scsi_task(acb->task);
 318     acb->task = NULL;
 319 }
 320
 321 static BlockDriverAIOCB *
 322 iscsi_aio_readv(BlockDriverState *bs, int64_t sector_num,
 323                 QEMUIOVector *qiov, int nb_sectors,
 324                 BlockDriverCompletionFunc *cb,
 325                 void *opaque)
 326 {
 327     IscsiLun *iscsilun = bs->opaque;
 328     struct iscsi_context *iscsi = iscsilun->iscsi;
 329     IscsiAIOCB *acb;
 330     size_t qemu_read_size;
 331     int i;
 332     uint64_t lba;
 333     uint32_t num_sectors;
 334
 335     qemu_read_size = BDRV_SECTOR_SIZE * (size_t)nb_sectors;
 336
 337     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 338     trace_iscsi_aio_readv(iscsi, sector_num, nb_sectors, opaque, acb);
 339
 340     acb->iscsilun = iscsilun;
 341     acb->qiov     = qiov;
 342
 343     acb->canceled    = 0;
 344     acb->read_size   = qemu_read_size;
 345     acb->buf         = NULL;
 346
 347     /* If LUN blocksize is bigger than BDRV_BLOCK_SIZE a read from QEMU
 348      * may be misaligned to the LUN, so we may need to read some extra
 349      * data.
 350      */
 351     acb->read_offset = 0;
 352     if (iscsilun->block_size > BDRV_SECTOR_SIZE) {
 353         uint64_t bdrv_offset = BDRV_SECTOR_SIZE * sector_num;
 354
 355         acb->read_offset  = bdrv_offset % iscsilun->block_size;
 356     }
 357
 358     num_sectors  = (qemu_read_size + iscsilun->block_size
 359                     + acb->read_offset - 1)
 360                     / iscsilun->block_size;
 361
 362     acb->task = malloc(sizeof(struct scsi_task));
 363     if (acb->task == NULL) {
 364         error_report("iSCSI: Failed to allocate task for scsi READ16 "
 365                      "command. %s", iscsi_get_error(iscsi));
 366         qemu_aio_release(acb);
 367         return NULL;
 368     }
 369     memset(acb->task, 0, sizeof(struct scsi_task));
 370
 371     acb->task->xfer_dir = SCSI_XFER_READ;
 372     lba = sector_qemu2lun(sector_num, iscsilun);
 373     acb->task->expxferlen = qemu_read_size;
 374
 375     switch (iscsilun->type) {
 376     case TYPE_DISK:
 377         acb->task->cdb_size = 16;
 378         acb->task->cdb[0]  = 0x88;
 379         *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 380         *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 381         *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 382         break;
 383     default:
 384         acb->task->cdb_size = 10;
 385         acb->task->cdb[0]  = 0x28;
 386         *(uint32_t *)&acb->task->cdb[2] = htonl(lba);
 387         *(uint16_t *)&acb->task->cdb[7] = htons(num_sectors);
 388         break;
 389     }
 390
 391     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 392                                  iscsi_aio_read16_cb,
 393                                  NULL,
 394                                  acb) != 0) {
 395         scsi_free_scsi_task(acb->task);
 396         qemu_aio_release(acb);
 397         return NULL;
 398     }
 399
 400     for (i = 0; i < acb->qiov->niov; i++) {
 401         scsi_task_add_data_in_buffer(acb->task,
 402                 acb->qiov->iov[i].iov_len,
 403                 acb->qiov->iov[i].iov_base);
 404     }
 405
 406     iscsi_set_events(iscsilun);
 407
 408     return &acb->common;
 409 }
 410
 411
 412 static void
 413 iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
 414                      void *command_data, void *opaque)
 415 {
 416     IscsiAIOCB *acb = opaque;
 417
 418     if (acb->canceled != 0) {
 419         qemu_aio_release(acb);
 420         scsi_free_scsi_task(acb->task);
 421         acb->task = NULL;
 422         return;
 423     }
 424
 425     acb->status = 0;
 426     if (status < 0) {
 427         error_report("Failed to sync10 data on iSCSI lun. %s",
 428                      iscsi_get_error(iscsi));
 429         acb->status = -EIO;
 430     }
 431
 432     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 433     scsi_free_scsi_task(acb->task);
 434     acb->task = NULL;
 435 }
 436
 437 static BlockDriverAIOCB *
 438 iscsi_aio_flush(BlockDriverState *bs,
 439                 BlockDriverCompletionFunc *cb, void *opaque)
 440 {
 441     IscsiLun *iscsilun = bs->opaque;
 442     struct iscsi_context *iscsi = iscsilun->iscsi;
 443     IscsiAIOCB *acb;
 444
 445     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 446
 447     acb->iscsilun = iscsilun;
 448     acb->canceled   = 0;
 449
 450     acb->task = iscsi_synchronizecache10_task(iscsi, iscsilun->lun,
 451                                          0, 0, 0, 0,
 452                                          iscsi_synccache10_cb,
 453                                          acb);
 454     if (acb->task == NULL) {
 455         error_report("iSCSI: Failed to send synchronizecache10 command. %s",
 456                      iscsi_get_error(iscsi));
 457         qemu_aio_release(acb);
 458         return NULL;
 459     }
 460
 461     iscsi_set_events(iscsilun);
 462
 463     return &acb->common;
 464 }
 465
 466 static void
 467 iscsi_unmap_cb(struct iscsi_context *iscsi, int status,
 468                      void *command_data, void *opaque)
 469 {
 470     IscsiAIOCB *acb = opaque;
 471
 472     if (acb->canceled != 0) {
 473         qemu_aio_release(acb);
 474         scsi_free_scsi_task(acb->task);
 475         acb->task = NULL;
 476         return;
 477     }
 478
 479     acb->status = 0;
 480     if (status < 0) {
 481         error_report("Failed to unmap data on iSCSI lun. %s",
 482                      iscsi_get_error(iscsi));
 483         acb->status = -EIO;
 484     }
 485
 486     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 487     scsi_free_scsi_task(acb->task);
 488     acb->task = NULL;
 489 }
 490
 491 static BlockDriverAIOCB *
 492 iscsi_aio_discard(BlockDriverState *bs,
 493                   int64_t sector_num, int nb_sectors,
 494                   BlockDriverCompletionFunc *cb, void *opaque)
 495 {
 496     IscsiLun *iscsilun = bs->opaque;
 497     struct iscsi_context *iscsi = iscsilun->iscsi;
 498     IscsiAIOCB *acb;
 499     struct unmap_list list[1];
 500
 501     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 502
 503     acb->iscsilun = iscsilun;
 504     acb->canceled   = 0;
 505
 506     list[0].lba = sector_qemu2lun(sector_num, iscsilun);
 507     list[0].num = nb_sectors * BDRV_SECTOR_SIZE / iscsilun->block_size;
 508
 509     acb->task = iscsi_unmap_task(iscsi, iscsilun->lun,
 510                                  0, 0, &list[0], 1,
 511                                  iscsi_unmap_cb,
 512                                  acb);
 513     if (acb->task == NULL) {
 514         error_report("iSCSI: Failed to send unmap command. %s",
 515                      iscsi_get_error(iscsi));
 516         qemu_aio_release(acb);
 517         return NULL;
 518     }
 519
 520     iscsi_set_events(iscsilun);
 521
 522     return &acb->common;
 523 }
 524
 525 #ifdef __linux__
 526 static void
 527 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 528                      void *command_data, void *opaque)
 529 {
 530     IscsiAIOCB *acb = opaque;
 531
 532     if (acb->canceled != 0) {
 533         qemu_aio_release(acb);
 534         scsi_free_scsi_task(acb->task);
 535         acb->task = NULL;
 536         return;
 537     }
 538
 539     acb->status = 0;
 540     if (status < 0) {
 541         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 542                      iscsi_get_error(iscsi));
 543         acb->status = -EIO;
 544     }
 545
 546     acb->ioh->driver_status = 0;
 547     acb->ioh->host_status   = 0;
 548     acb->ioh->resid         = 0;
 549
 550 #define SG_ERR_DRIVER_SENSE    0x08
 551
 552     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 553         int ss;
 554
 555         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 556
 557         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 558         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 559              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 560         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 561     }
 562
 563     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 564     scsi_free_scsi_task(acb->task);
 565     acb->task = NULL;
 566 }
 567
 568 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 569         unsigned long int req, void *buf,
 570         BlockDriverCompletionFunc *cb, void *opaque)
 571 {
 572     IscsiLun *iscsilun = bs->opaque;
 573     struct iscsi_context *iscsi = iscsilun->iscsi;
 574     struct iscsi_data data;
 575     IscsiAIOCB *acb;
 576
 577     assert(req == SG_IO);
 578
 579     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 580
 581     acb->iscsilun = iscsilun;
 582     acb->canceled    = 0;
 583     acb->buf         = NULL;
 584     acb->ioh         = buf;
 585
 586     acb->task = malloc(sizeof(struct scsi_task));
 587     if (acb->task == NULL) {
 588         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 589                      iscsi_get_error(iscsi));
 590         qemu_aio_release(acb);
 591         return NULL;
 592     }
 593     memset(acb->task, 0, sizeof(struct scsi_task));
 594
 595     switch (acb->ioh->dxfer_direction) {
 596     case SG_DXFER_TO_DEV:
 597         acb->task->xfer_dir = SCSI_XFER_WRITE;
 598         break;
 599     case SG_DXFER_FROM_DEV:
 600         acb->task->xfer_dir = SCSI_XFER_READ;
 601         break;
 602     default:
 603         acb->task->xfer_dir = SCSI_XFER_NONE;
 604         break;
 605     }
 606
 607     acb->task->cdb_size = acb->ioh->cmd_len;
 608     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 609     acb->task->expxferlen = acb->ioh->dxfer_len;
 610
 611     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 612         data.data = acb->ioh->dxferp;
 613         data.size = acb->ioh->dxfer_len;
 614     }
 615     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 616                                  iscsi_aio_ioctl_cb,
 617                                  (acb->task->xfer_dir == SCSI_XFER_WRITE) ?
 618                                      &data : NULL,
 619                                  acb) != 0) {
 620         scsi_free_scsi_task(acb->task);
 621         qemu_aio_release(acb);
 622         return NULL;
 623     }
 624
 625     /* tell libiscsi to read straight into the buffer we got from ioctl */
 626     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 627         scsi_task_add_data_in_buffer(acb->task,
 628                                      acb->ioh->dxfer_len,
 629                                      acb->ioh->dxferp);
 630     }
 631
 632     iscsi_set_events(iscsilun);
 633
 634     return &acb->common;
 635 }
 636
 637 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 638 {
 639     IscsiLun *iscsilun = bs->opaque;
 640
 641     switch (req) {
 642     case SG_GET_VERSION_NUM:
 643         *(int *)buf = 30000;
 644         break;
 645     case SG_GET_SCSI_ID:
 646         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 647         break;
 648     default:
 649         return -1;
 650     }
 651     return 0;
 652 }
 653 #endif
 654
 655 static int64_t
 656 iscsi_getlength(BlockDriverState *bs)
 657 {
 658     IscsiLun *iscsilun = bs->opaque;
 659     int64_t len;
 660
 661     len  = iscsilun->num_blocks;
 662     len *= iscsilun->block_size;
 663
 664     return len;
 665 }
 666
 667 static void
 668 iscsi_readcapacity16_cb(struct iscsi_context *iscsi, int status,
 669                         void *command_data, void *opaque)
 670 {
 671     struct IscsiTask *itask = opaque;
 672     struct scsi_readcapacity16 *rc16;
 673     struct scsi_task *task = command_data;
 674
 675     if (status != 0) {
 676         error_report("iSCSI: Failed to read capacity of iSCSI lun. %s",
 677                      iscsi_get_error(iscsi));
 678         itask->status   = 1;
 679         itask->complete = 1;
 680         scsi_free_scsi_task(task);
 681         return;
 682     }
 683
 684     rc16 = scsi_datain_unmarshall(task);
 685     if (rc16 == NULL) {
 686         error_report("iSCSI: Failed to unmarshall readcapacity16 data.");
 687         itask->status   = 1;
 688         itask->complete = 1;
 689         scsi_free_scsi_task(task);
 690         return;
 691     }
 692
 693     itask->iscsilun->block_size = rc16->block_length;
 694     itask->iscsilun->num_blocks = rc16->returned_lba + 1;
 695     itask->bs->total_sectors    = itask->iscsilun->num_blocks *
 696                                itask->iscsilun->block_size / BDRV_SECTOR_SIZE ;
 697
 698     itask->status   = 0;
 699     itask->complete = 1;
 700     scsi_free_scsi_task(task);
 701 }
 702
 703 static void
 704 iscsi_readcapacity10_cb(struct iscsi_context *iscsi, int status,
 705                         void *command_data, void *opaque)
 706 {
 707     struct IscsiTask *itask = opaque;
 708     struct scsi_readcapacity10 *rc10;
 709     struct scsi_task *task = command_data;
 710
 711     if (status != 0) {
 712         error_report("iSCSI: Failed to read capacity of iSCSI lun. %s",
 713                      iscsi_get_error(iscsi));
 714         itask->status   = 1;
 715         itask->complete = 1;
 716         scsi_free_scsi_task(task);
 717         return;
 718     }
 719
 720     rc10 = scsi_datain_unmarshall(task);
 721     if (rc10 == NULL) {
 722         error_report("iSCSI: Failed to unmarshall readcapacity10 data.");
 723         itask->status   = 1;
 724         itask->complete = 1;
 725         scsi_free_scsi_task(task);
 726         return;
 727     }
 728
 729     itask->iscsilun->block_size = rc10->block_size;
 730     itask->iscsilun->num_blocks = rc10->lba + 1;
 731     itask->bs->total_sectors    = itask->iscsilun->num_blocks *
 732                                itask->iscsilun->block_size / BDRV_SECTOR_SIZE ;
 733
 734     itask->status   = 0;
 735     itask->complete = 1;
 736     scsi_free_scsi_task(task);
 737 }
 738
 739 static void
 740 iscsi_inquiry_cb(struct iscsi_context *iscsi, int status, void *command_data,
 741                  void *opaque)
 742 {
 743     struct IscsiTask *itask = opaque;
 744     struct scsi_task *task = command_data;
 745     struct scsi_inquiry_standard *inq;
 746
 747     if (status != 0) {
 748         itask->status   = 1;
 749         itask->complete = 1;
 750         scsi_free_scsi_task(task);
 751         return;
 752     }
 753
 754     inq = scsi_datain_unmarshall(task);
 755     if (inq == NULL) {
 756         error_report("iSCSI: Failed to unmarshall inquiry data.");
 757         itask->status   = 1;
 758         itask->complete = 1;
 759         scsi_free_scsi_task(task);
 760         return;
 761     }
 762
 763     itask->iscsilun->type = inq->periperal_device_type;
 764
 765     scsi_free_scsi_task(task);
 766
 767     switch (itask->iscsilun->type) {
 768     case TYPE_DISK:
 769         task = iscsi_readcapacity16_task(iscsi, itask->iscsilun->lun,
 770                                    iscsi_readcapacity16_cb, opaque);
 771         if (task == NULL) {
 772             error_report("iSCSI: failed to send readcapacity16 command.");
 773             itask->status   = 1;
 774             itask->complete = 1;
 775             return;
 776         }
 777         break;
 778     case TYPE_ROM:
 779         task = iscsi_readcapacity10_task(iscsi, itask->iscsilun->lun,
 780                                    0, 0,
 781                                    iscsi_readcapacity10_cb, opaque);
 782         if (task == NULL) {
 783             error_report("iSCSI: failed to send readcapacity16 command.");
 784             itask->status   = 1;
 785             itask->complete = 1;
 786             return;
 787         }
 788         break;
 789     default:
 790         itask->status   = 0;
 791         itask->complete = 1;
 792     }
 793 }
 794
 795 static void
 796 iscsi_connect_cb(struct iscsi_context *iscsi, int status, void *command_data,
 797                  void *opaque)
 798 {
 799     struct IscsiTask *itask = opaque;
 800     struct scsi_task *task;
 801
 802     if (status != 0) {
 803         itask->status   = 1;
 804         itask->complete = 1;
 805         return;
 806     }
 807
 808     task = iscsi_inquiry_task(iscsi, itask->iscsilun->lun,
 809                               0, 0, 36,
 810                               iscsi_inquiry_cb, opaque);
 811     if (task == NULL) {
 812         error_report("iSCSI: failed to send inquiry command.");
 813         itask->status   = 1;
 814         itask->complete = 1;
 815         return;
 816     }
 817 }
 818
 819 static int parse_chap(struct iscsi_context *iscsi, const char *target)
 820 {
 821     QemuOptsList *list;
 822     QemuOpts *opts;
 823     const char *user = NULL;
 824     const char *password = NULL;
 825
 826     list = qemu_find_opts("iscsi");
 827     if (!list) {
 828         return 0;
 829     }
 830
 831     opts = qemu_opts_find(list, target);
 832     if (opts == NULL) {
 833         opts = QTAILQ_FIRST(&list->head);
 834         if (!opts) {
 835             return 0;
 836         }
 837     }
 838
 839     user = qemu_opt_get(opts, "user");
 840     if (!user) {
 841         return 0;
 842     }
 843
 844     password = qemu_opt_get(opts, "password");
 845     if (!password) {
 846         error_report("CHAP username specified but no password was given");
 847         return -1;
 848     }
 849
 850     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
 851         error_report("Failed to set initiator username and password");
 852         return -1;
 853     }
 854
 855     return 0;
 856 }
 857
 858 static void parse_header_digest(struct iscsi_context *iscsi, const char *target)
 859 {
 860     QemuOptsList *list;
 861     QemuOpts *opts;
 862     const char *digest = NULL;
 863
 864     list = qemu_find_opts("iscsi");
 865     if (!list) {
 866         return;
 867     }
 868
 869     opts = qemu_opts_find(list, target);
 870     if (opts == NULL) {
 871         opts = QTAILQ_FIRST(&list->head);
 872         if (!opts) {
 873             return;
 874         }
 875     }
 876
 877     digest = qemu_opt_get(opts, "header-digest");
 878     if (!digest) {
 879         return;
 880     }
 881
 882     if (!strcmp(digest, "CRC32C")) {
 883         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
 884     } else if (!strcmp(digest, "NONE")) {
 885         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
 886     } else if (!strcmp(digest, "CRC32C-NONE")) {
 887         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
 888     } else if (!strcmp(digest, "NONE-CRC32C")) {
 889         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 890     } else {
 891         error_report("Invalid header-digest setting : %s", digest);
 892     }
 893 }
 894
 895 static char *parse_initiator_name(const char *target)
 896 {
 897     QemuOptsList *list;
 898     QemuOpts *opts;
 899     const char *name = NULL;
 900
 901     list = qemu_find_opts("iscsi");
 902     if (!list) {
 903         return g_strdup("iqn.2008-11.org.linux-kvm");
 904     }
 905
 906     opts = qemu_opts_find(list, target);
 907     if (opts == NULL) {
 908         opts = QTAILQ_FIRST(&list->head);
 909         if (!opts) {
 910             return g_strdup("iqn.2008-11.org.linux-kvm");
 911         }
 912     }
 913
 914     name = qemu_opt_get(opts, "initiator-name");
 915     if (!name) {
 916         return g_strdup("iqn.2008-11.org.linux-kvm");
 917     }
 918
 919     return g_strdup(name);
 920 }
 921
 922 /*
 923  * We support iscsi url's on the form
 924  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
 925  */
 926 static int iscsi_open(BlockDriverState *bs, const char *filename, int flags)
 927 {
 928     IscsiLun *iscsilun = bs->opaque;
 929     struct iscsi_context *iscsi = NULL;
 930     struct iscsi_url *iscsi_url = NULL;
 931     struct IscsiTask task;
 932     char *initiator_name = NULL;
 933     int ret;
 934
 935     if ((BDRV_SECTOR_SIZE % 512) != 0) {
 936         error_report("iSCSI: Invalid BDRV_SECTOR_SIZE. "
 937                      "BDRV_SECTOR_SIZE(%lld) is not a multiple "
 938                      "of 512", BDRV_SECTOR_SIZE);
 939         return -EINVAL;
 940     }
 941
 942     iscsi_url = iscsi_parse_full_url(iscsi, filename);
 943     if (iscsi_url == NULL) {
 944         error_report("Failed to parse URL : %s %s", filename,
 945                      iscsi_get_error(iscsi));
 946         ret = -EINVAL;
 947         goto failed;
 948     }
 949
 950     memset(iscsilun, 0, sizeof(IscsiLun));
 951
 952     initiator_name = parse_initiator_name(iscsi_url->target);
 953
 954     iscsi = iscsi_create_context(initiator_name);
 955     if (iscsi == NULL) {
 956         error_report("iSCSI: Failed to create iSCSI context.");
 957         ret = -ENOMEM;
 958         goto failed;
 959     }
 960
 961     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
 962         error_report("iSCSI: Failed to set target name.");
 963         ret = -EINVAL;
 964         goto failed;
 965     }
 966
 967     if (iscsi_url->user != NULL) {
 968         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
 969                                               iscsi_url->passwd);
 970         if (ret != 0) {
 971             error_report("Failed to set initiator username and password");
 972             ret = -EINVAL;
 973             goto failed;
 974         }
 975     }
 976
 977     /* check if we got CHAP username/password via the options */
 978     if (parse_chap(iscsi, iscsi_url->target) != 0) {
 979         error_report("iSCSI: Failed to set CHAP user/password");
 980         ret = -EINVAL;
 981         goto failed;
 982     }
 983
 984     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
 985         error_report("iSCSI: Failed to set session type to normal.");
 986         ret = -EINVAL;
 987         goto failed;
 988     }
 989
 990     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 991
 992     /* check if we got HEADER_DIGEST via the options */
 993     parse_header_digest(iscsi, iscsi_url->target);
 994
 995     task.iscsilun = iscsilun;
 996     task.status = 0;
 997     task.complete = 0;
 998     task.bs = bs;
 999
1000     iscsilun->iscsi = iscsi;
1001     iscsilun->lun   = iscsi_url->lun;
1002
1003     if (iscsi_full_connect_async(iscsi, iscsi_url->portal, iscsi_url->lun,
1004                                  iscsi_connect_cb, &task)
1005         != 0) {
1006         error_report("iSCSI: Failed to start async connect.");
1007         ret = -EINVAL;
1008         goto failed;
1009     }
1010
1011     while (!task.complete) {
1012         iscsi_set_events(iscsilun);
1013         qemu_aio_wait();
1014     }
1015     if (task.status != 0) {
1016         error_report("iSCSI: Failed to connect to LUN : %s",
1017                      iscsi_get_error(iscsi));
1018         ret = -EINVAL;
1019         goto failed;
1020     }
1021
1022     if (iscsi_url != NULL) {
1023         iscsi_destroy_url(iscsi_url);
1024     }
1025
1026     /* Medium changer or tape. We dont have any emulation for this so this must
1027      * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
1028      * to read from the device to guess the image format.
1029      */
1030     if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
1031         iscsilun->type == TYPE_TAPE) {
1032         bs->sg = 1;
1033     }
1034
1035     return 0;
1036
1037 failed:
1038     if (initiator_name != NULL) {
1039         g_free(initiator_name);
1040     }
1041     if (iscsi_url != NULL) {
1042         iscsi_destroy_url(iscsi_url);
1043     }
1044     if (iscsi != NULL) {
1045         iscsi_destroy_context(iscsi);
1046     }
1047     memset(iscsilun, 0, sizeof(IscsiLun));
1048     return ret;
1049 }
1050
1051 static void iscsi_close(BlockDriverState *bs)
1052 {
1053     IscsiLun *iscsilun = bs->opaque;
1054     struct iscsi_context *iscsi = iscsilun->iscsi;
1055
1056     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL, NULL);
1057     iscsi_destroy_context(iscsi);
1058     memset(iscsilun, 0, sizeof(IscsiLun));
1059 }
1060
1061 static BlockDriver bdrv_iscsi = {
1062     .format_name     = "iscsi",
1063     .protocol_name   = "iscsi",
1064
1065     .instance_size   = sizeof(IscsiLun),
1066     .bdrv_file_open  = iscsi_open,
1067     .bdrv_close      = iscsi_close,
1068
1069     .bdrv_getlength  = iscsi_getlength,
1070
1071     .bdrv_aio_readv  = iscsi_aio_readv,
1072     .bdrv_aio_writev = iscsi_aio_writev,
1073     .bdrv_aio_flush  = iscsi_aio_flush,
1074
1075     .bdrv_aio_discard = iscsi_aio_discard,
1076
1077 #ifdef __linux__
1078     .bdrv_ioctl       = iscsi_ioctl,
1079     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1080 #endif
1081 };
1082
1083 static void iscsi_block_init(void)
1084 {
1085     bdrv_register(&bdrv_iscsi);
1086 }
1087
1088 block_init(iscsi_block_init);