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_bh_cb(void *p)
  77 {
  78     IscsiAIOCB *acb = p;
  79
  80     qemu_bh_delete(acb->bh);
  81
  82     if (acb->canceled == 0) {
  83         acb->common.cb(acb->common.opaque, acb->status);
  84     }
  85
  86     if (acb->task != NULL) {
  87         scsi_free_scsi_task(acb->task);
  88         acb->task = NULL;
  89     }
  90
  91     qemu_aio_release(acb);
  92 }
  93
  94 static void
  95 iscsi_schedule_bh(IscsiAIOCB *acb)
  96 {
  97     if (acb->bh) {
  98         return;
  99     }
 100     acb->bh = qemu_bh_new(iscsi_bh_cb, acb);
 101     qemu_bh_schedule(acb->bh);
 102 }
 103
 104
 105 static void
 106 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
 107                     void *private_data)
 108 {
 109     IscsiAIOCB *acb = private_data;
 110
 111     acb->status = -ECANCELED;
 112     iscsi_schedule_bh(acb);
 113 }
 114
 115 static void
 116 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
 117 {
 118     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
 119     IscsiLun *iscsilun = acb->iscsilun;
 120
 121     if (acb->status != -EINPROGRESS) {
 122         return;
 123     }
 124
 125     acb->canceled = 1;
 126
 127     /* send a task mgmt call to the target to cancel the task on the target */
 128     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
 129                                      iscsi_abort_task_cb, acb);
 130
 131     while (acb->status == -EINPROGRESS) {
 132         qemu_aio_wait();
 133     }
 134 }
 135
 136 static AIOPool iscsi_aio_pool = {
 137     .aiocb_size         = sizeof(IscsiAIOCB),
 138     .cancel             = iscsi_aio_cancel,
 139 };
 140
 141
 142 static void iscsi_process_read(void *arg);
 143 static void iscsi_process_write(void *arg);
 144
 145 static int iscsi_process_flush(void *arg)
 146 {
 147     IscsiLun *iscsilun = arg;
 148
 149     return iscsi_queue_length(iscsilun->iscsi) > 0;
 150 }
 151
 152 static void
 153 iscsi_set_events(IscsiLun *iscsilun)
 154 {
 155     struct iscsi_context *iscsi = iscsilun->iscsi;
 156     int ev;
 157
 158     /* We always register a read handler.  */
 159     ev = POLLIN;
 160     ev |= iscsi_which_events(iscsi);
 161     if (ev != iscsilun->events) {
 162         qemu_aio_set_fd_handler(iscsi_get_fd(iscsi),
 163                       iscsi_process_read,
 164                       (ev & POLLOUT) ? iscsi_process_write : NULL,
 165                       iscsi_process_flush,
 166                       iscsilun);
 167
 168     }
 169
 170     /* If we just added an event, the callback might be delayed
 171      * unless we call qemu_notify_event().
 172      */
 173     if (ev & ~iscsilun->events) {
 174         qemu_notify_event();
 175     }
 176     iscsilun->events = ev;
 177 }
 178
 179 static void
 180 iscsi_process_read(void *arg)
 181 {
 182     IscsiLun *iscsilun = arg;
 183     struct iscsi_context *iscsi = iscsilun->iscsi;
 184
 185     iscsi_service(iscsi, POLLIN);
 186     iscsi_set_events(iscsilun);
 187 }
 188
 189 static void
 190 iscsi_process_write(void *arg)
 191 {
 192     IscsiLun *iscsilun = arg;
 193     struct iscsi_context *iscsi = iscsilun->iscsi;
 194
 195     iscsi_service(iscsi, POLLOUT);
 196     iscsi_set_events(iscsilun);
 197 }
 198
 199
 200 static void
 201 iscsi_aio_write16_cb(struct iscsi_context *iscsi, int status,
 202                      void *command_data, void *opaque)
 203 {
 204     IscsiAIOCB *acb = opaque;
 205
 206     trace_iscsi_aio_write16_cb(iscsi, status, acb, acb->canceled);
 207
 208     g_free(acb->buf);
 209
 210     if (acb->canceled != 0) {
 211         return;
 212     }
 213
 214     acb->status = 0;
 215     if (status < 0) {
 216         error_report("Failed to write16 data to iSCSI lun. %s",
 217                      iscsi_get_error(iscsi));
 218         acb->status = -EIO;
 219     }
 220
 221     iscsi_schedule_bh(acb);
 222 }
 223
 224 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
 225 {
 226     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
 227 }
 228
 229 static BlockDriverAIOCB *
 230 iscsi_aio_writev(BlockDriverState *bs, int64_t sector_num,
 231                  QEMUIOVector *qiov, int nb_sectors,
 232                  BlockDriverCompletionFunc *cb,
 233                  void *opaque)
 234 {
 235     IscsiLun *iscsilun = bs->opaque;
 236     struct iscsi_context *iscsi = iscsilun->iscsi;
 237     IscsiAIOCB *acb;
 238     size_t size;
 239     uint32_t num_sectors;
 240     uint64_t lba;
 241     struct iscsi_data data;
 242
 243     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 244     trace_iscsi_aio_writev(iscsi, sector_num, nb_sectors, opaque, acb);
 245
 246     acb->iscsilun = iscsilun;
 247     acb->qiov     = qiov;
 248
 249     acb->canceled   = 0;
 250     acb->bh         = NULL;
 251     acb->status     = -EINPROGRESS;
 252
 253     /* XXX we should pass the iovec to write16 to avoid the extra copy */
 254     /* this will allow us to get rid of 'buf' completely */
 255     size = nb_sectors * BDRV_SECTOR_SIZE;
 256     acb->buf = g_malloc(size);
 257     qemu_iovec_to_buf(acb->qiov, 0, acb->buf, size);
 258
 259     acb->task = malloc(sizeof(struct scsi_task));
 260     if (acb->task == NULL) {
 261         error_report("iSCSI: Failed to allocate task for scsi WRITE16 "
 262                      "command. %s", iscsi_get_error(iscsi));
 263         qemu_aio_release(acb);
 264         return NULL;
 265     }
 266     memset(acb->task, 0, sizeof(struct scsi_task));
 267
 268     acb->task->xfer_dir = SCSI_XFER_WRITE;
 269     acb->task->cdb_size = 16;
 270     acb->task->cdb[0] = 0x8a;
 271     if (!(bs->open_flags & BDRV_O_CACHE_WB)) {
 272         /* set FUA on writes when cache mode is write through */
 273         acb->task->cdb[1] |= 0x04;
 274     }
 275     lba = sector_qemu2lun(sector_num, iscsilun);
 276     *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 277     *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 278     num_sectors = size / iscsilun->block_size;
 279     *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 280     acb->task->expxferlen = size;
 281
 282     data.data = acb->buf;
 283     data.size = size;
 284
 285     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 286                                  iscsi_aio_write16_cb,
 287                                  &data,
 288                                  acb) != 0) {
 289         scsi_free_scsi_task(acb->task);
 290         g_free(acb->buf);
 291         qemu_aio_release(acb);
 292         return NULL;
 293     }
 294
 295     iscsi_set_events(iscsilun);
 296
 297     return &acb->common;
 298 }
 299
 300 static void
 301 iscsi_aio_read16_cb(struct iscsi_context *iscsi, int status,
 302                     void *command_data, void *opaque)
 303 {
 304     IscsiAIOCB *acb = opaque;
 305
 306     trace_iscsi_aio_read16_cb(iscsi, status, acb, acb->canceled);
 307
 308     if (acb->canceled != 0) {
 309         return;
 310     }
 311
 312     acb->status = 0;
 313     if (status != 0) {
 314         error_report("Failed to read16 data from iSCSI lun. %s",
 315                      iscsi_get_error(iscsi));
 316         acb->status = -EIO;
 317     }
 318
 319     iscsi_schedule_bh(acb);
 320 }
 321
 322 static BlockDriverAIOCB *
 323 iscsi_aio_readv(BlockDriverState *bs, int64_t sector_num,
 324                 QEMUIOVector *qiov, int nb_sectors,
 325                 BlockDriverCompletionFunc *cb,
 326                 void *opaque)
 327 {
 328     IscsiLun *iscsilun = bs->opaque;
 329     struct iscsi_context *iscsi = iscsilun->iscsi;
 330     IscsiAIOCB *acb;
 331     size_t qemu_read_size;
 332     int i;
 333     uint64_t lba;
 334     uint32_t num_sectors;
 335
 336     qemu_read_size = BDRV_SECTOR_SIZE * (size_t)nb_sectors;
 337
 338     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 339     trace_iscsi_aio_readv(iscsi, sector_num, nb_sectors, opaque, acb);
 340
 341     acb->iscsilun = iscsilun;
 342     acb->qiov     = qiov;
 343
 344     acb->canceled    = 0;
 345     acb->bh          = NULL;
 346     acb->status      = -EINPROGRESS;
 347     acb->read_size   = qemu_read_size;
 348     acb->buf         = NULL;
 349
 350     /* If LUN blocksize is bigger than BDRV_BLOCK_SIZE a read from QEMU
 351      * may be misaligned to the LUN, so we may need to read some extra
 352      * data.
 353      */
 354     acb->read_offset = 0;
 355     if (iscsilun->block_size > BDRV_SECTOR_SIZE) {
 356         uint64_t bdrv_offset = BDRV_SECTOR_SIZE * sector_num;
 357
 358         acb->read_offset  = bdrv_offset % iscsilun->block_size;
 359     }
 360
 361     num_sectors  = (qemu_read_size + iscsilun->block_size
 362                     + acb->read_offset - 1)
 363                     / iscsilun->block_size;
 364
 365     acb->task = malloc(sizeof(struct scsi_task));
 366     if (acb->task == NULL) {
 367         error_report("iSCSI: Failed to allocate task for scsi READ16 "
 368                      "command. %s", iscsi_get_error(iscsi));
 369         qemu_aio_release(acb);
 370         return NULL;
 371     }
 372     memset(acb->task, 0, sizeof(struct scsi_task));
 373
 374     acb->task->xfer_dir = SCSI_XFER_READ;
 375     lba = sector_qemu2lun(sector_num, iscsilun);
 376     acb->task->expxferlen = qemu_read_size;
 377
 378     switch (iscsilun->type) {
 379     case TYPE_DISK:
 380         acb->task->cdb_size = 16;
 381         acb->task->cdb[0]  = 0x88;
 382         *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 383         *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 384         *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 385         break;
 386     default:
 387         acb->task->cdb_size = 10;
 388         acb->task->cdb[0]  = 0x28;
 389         *(uint32_t *)&acb->task->cdb[2] = htonl(lba);
 390         *(uint16_t *)&acb->task->cdb[7] = htons(num_sectors);
 391         break;
 392     }
 393
 394     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 395                                  iscsi_aio_read16_cb,
 396                                  NULL,
 397                                  acb) != 0) {
 398         scsi_free_scsi_task(acb->task);
 399         qemu_aio_release(acb);
 400         return NULL;
 401     }
 402
 403     for (i = 0; i < acb->qiov->niov; i++) {
 404         scsi_task_add_data_in_buffer(acb->task,
 405                 acb->qiov->iov[i].iov_len,
 406                 acb->qiov->iov[i].iov_base);
 407     }
 408
 409     iscsi_set_events(iscsilun);
 410
 411     return &acb->common;
 412 }
 413
 414
 415 static void
 416 iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
 417                      void *command_data, void *opaque)
 418 {
 419     IscsiAIOCB *acb = opaque;
 420
 421     if (acb->canceled != 0) {
 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(acb);
 433 }
 434
 435 static BlockDriverAIOCB *
 436 iscsi_aio_flush(BlockDriverState *bs,
 437                 BlockDriverCompletionFunc *cb, void *opaque)
 438 {
 439     IscsiLun *iscsilun = bs->opaque;
 440     struct iscsi_context *iscsi = iscsilun->iscsi;
 441     IscsiAIOCB *acb;
 442
 443     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 444
 445     acb->iscsilun = iscsilun;
 446     acb->canceled   = 0;
 447     acb->bh         = NULL;
 448     acb->status     = -EINPROGRESS;
 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         return;
 474     }
 475
 476     acb->status = 0;
 477     if (status < 0) {
 478         error_report("Failed to unmap data on iSCSI lun. %s",
 479                      iscsi_get_error(iscsi));
 480         acb->status = -EIO;
 481     }
 482
 483     iscsi_schedule_bh(acb);
 484 }
 485
 486 static BlockDriverAIOCB *
 487 iscsi_aio_discard(BlockDriverState *bs,
 488                   int64_t sector_num, int nb_sectors,
 489                   BlockDriverCompletionFunc *cb, void *opaque)
 490 {
 491     IscsiLun *iscsilun = bs->opaque;
 492     struct iscsi_context *iscsi = iscsilun->iscsi;
 493     IscsiAIOCB *acb;
 494     struct unmap_list list[1];
 495
 496     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 497
 498     acb->iscsilun = iscsilun;
 499     acb->canceled   = 0;
 500     acb->bh         = NULL;
 501     acb->status     = -EINPROGRESS;
 502
 503     list[0].lba = sector_qemu2lun(sector_num, iscsilun);
 504     list[0].num = nb_sectors * BDRV_SECTOR_SIZE / iscsilun->block_size;
 505
 506     acb->task = iscsi_unmap_task(iscsi, iscsilun->lun,
 507                                  0, 0, &list[0], 1,
 508                                  iscsi_unmap_cb,
 509                                  acb);
 510     if (acb->task == NULL) {
 511         error_report("iSCSI: Failed to send unmap command. %s",
 512                      iscsi_get_error(iscsi));
 513         qemu_aio_release(acb);
 514         return NULL;
 515     }
 516
 517     iscsi_set_events(iscsilun);
 518
 519     return &acb->common;
 520 }
 521
 522 #ifdef __linux__
 523 static void
 524 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 525                      void *command_data, void *opaque)
 526 {
 527     IscsiAIOCB *acb = opaque;
 528
 529     if (acb->canceled != 0) {
 530         return;
 531     }
 532
 533     acb->status = 0;
 534     if (status < 0) {
 535         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 536                      iscsi_get_error(iscsi));
 537         acb->status = -EIO;
 538     }
 539
 540     acb->ioh->driver_status = 0;
 541     acb->ioh->host_status   = 0;
 542     acb->ioh->resid         = 0;
 543
 544 #define SG_ERR_DRIVER_SENSE    0x08
 545
 546     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 547         int ss;
 548
 549         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 550
 551         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 552         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 553              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 554         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 555     }
 556
 557     iscsi_schedule_bh(acb);
 558 }
 559
 560 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 561         unsigned long int req, void *buf,
 562         BlockDriverCompletionFunc *cb, void *opaque)
 563 {
 564     IscsiLun *iscsilun = bs->opaque;
 565     struct iscsi_context *iscsi = iscsilun->iscsi;
 566     struct iscsi_data data;
 567     IscsiAIOCB *acb;
 568
 569     assert(req == SG_IO);
 570
 571     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 572
 573     acb->iscsilun = iscsilun;
 574     acb->canceled    = 0;
 575     acb->bh          = NULL;
 576     acb->status      = -EINPROGRESS;
 577     acb->buf         = NULL;
 578     acb->ioh         = buf;
 579
 580     acb->task = malloc(sizeof(struct scsi_task));
 581     if (acb->task == NULL) {
 582         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 583                      iscsi_get_error(iscsi));
 584         qemu_aio_release(acb);
 585         return NULL;
 586     }
 587     memset(acb->task, 0, sizeof(struct scsi_task));
 588
 589     switch (acb->ioh->dxfer_direction) {
 590     case SG_DXFER_TO_DEV:
 591         acb->task->xfer_dir = SCSI_XFER_WRITE;
 592         break;
 593     case SG_DXFER_FROM_DEV:
 594         acb->task->xfer_dir = SCSI_XFER_READ;
 595         break;
 596     default:
 597         acb->task->xfer_dir = SCSI_XFER_NONE;
 598         break;
 599     }
 600
 601     acb->task->cdb_size = acb->ioh->cmd_len;
 602     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 603     acb->task->expxferlen = acb->ioh->dxfer_len;
 604
 605     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 606         data.data = acb->ioh->dxferp;
 607         data.size = acb->ioh->dxfer_len;
 608     }
 609     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 610                                  iscsi_aio_ioctl_cb,
 611                                  (acb->task->xfer_dir == SCSI_XFER_WRITE) ?
 612                                      &data : NULL,
 613                                  acb) != 0) {
 614         scsi_free_scsi_task(acb->task);
 615         qemu_aio_release(acb);
 616         return NULL;
 617     }
 618
 619     /* tell libiscsi to read straight into the buffer we got from ioctl */
 620     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 621         scsi_task_add_data_in_buffer(acb->task,
 622                                      acb->ioh->dxfer_len,
 623                                      acb->ioh->dxferp);
 624     }
 625
 626     iscsi_set_events(iscsilun);
 627
 628     return &acb->common;
 629 }
 630
 631 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 632 {
 633     IscsiLun *iscsilun = bs->opaque;
 634
 635     switch (req) {
 636     case SG_GET_VERSION_NUM:
 637         *(int *)buf = 30000;
 638         break;
 639     case SG_GET_SCSI_ID:
 640         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 641         break;
 642     default:
 643         return -1;
 644     }
 645     return 0;
 646 }
 647 #endif
 648
 649 static int64_t
 650 iscsi_getlength(BlockDriverState *bs)
 651 {
 652     IscsiLun *iscsilun = bs->opaque;
 653     int64_t len;
 654
 655     len  = iscsilun->num_blocks;
 656     len *= iscsilun->block_size;
 657
 658     return len;
 659 }
 660
 661 static void
 662 iscsi_readcapacity16_cb(struct iscsi_context *iscsi, int status,
 663                         void *command_data, void *opaque)
 664 {
 665     struct IscsiTask *itask = opaque;
 666     struct scsi_readcapacity16 *rc16;
 667     struct scsi_task *task = command_data;
 668
 669     if (status != 0) {
 670         error_report("iSCSI: Failed to read capacity of iSCSI lun. %s",
 671                      iscsi_get_error(iscsi));
 672         itask->status   = 1;
 673         itask->complete = 1;
 674         scsi_free_scsi_task(task);
 675         return;
 676     }
 677
 678     rc16 = scsi_datain_unmarshall(task);
 679     if (rc16 == NULL) {
 680         error_report("iSCSI: Failed to unmarshall readcapacity16 data.");
 681         itask->status   = 1;
 682         itask->complete = 1;
 683         scsi_free_scsi_task(task);
 684         return;
 685     }
 686
 687     itask->iscsilun->block_size = rc16->block_length;
 688     itask->iscsilun->num_blocks = rc16->returned_lba + 1;
 689     itask->bs->total_sectors    = itask->iscsilun->num_blocks *
 690                                itask->iscsilun->block_size / BDRV_SECTOR_SIZE ;
 691
 692     itask->status   = 0;
 693     itask->complete = 1;
 694     scsi_free_scsi_task(task);
 695 }
 696
 697 static void
 698 iscsi_readcapacity10_cb(struct iscsi_context *iscsi, int status,
 699                         void *command_data, void *opaque)
 700 {
 701     struct IscsiTask *itask = opaque;
 702     struct scsi_readcapacity10 *rc10;
 703     struct scsi_task *task = command_data;
 704
 705     if (status != 0) {
 706         error_report("iSCSI: Failed to read capacity of iSCSI lun. %s",
 707                      iscsi_get_error(iscsi));
 708         itask->status   = 1;
 709         itask->complete = 1;
 710         scsi_free_scsi_task(task);
 711         return;
 712     }
 713
 714     rc10 = scsi_datain_unmarshall(task);
 715     if (rc10 == NULL) {
 716         error_report("iSCSI: Failed to unmarshall readcapacity10 data.");
 717         itask->status   = 1;
 718         itask->complete = 1;
 719         scsi_free_scsi_task(task);
 720         return;
 721     }
 722
 723     itask->iscsilun->block_size = rc10->block_size;
 724     if (rc10->lba == 0) {
 725         /* blank disk loaded */
 726         itask->iscsilun->num_blocks = 0;
 727     } else {
 728         itask->iscsilun->num_blocks = rc10->lba + 1;
 729     }
 730     itask->bs->total_sectors    = itask->iscsilun->num_blocks *
 731                                itask->iscsilun->block_size / BDRV_SECTOR_SIZE ;
 732
 733     itask->status   = 0;
 734     itask->complete = 1;
 735     scsi_free_scsi_task(task);
 736 }
 737
 738 static void
 739 iscsi_inquiry_cb(struct iscsi_context *iscsi, int status, void *command_data,
 740                  void *opaque)
 741 {
 742     struct IscsiTask *itask = opaque;
 743     struct scsi_task *task = command_data;
 744     struct scsi_inquiry_standard *inq;
 745
 746     if (status != 0) {
 747         itask->status   = 1;
 748         itask->complete = 1;
 749         scsi_free_scsi_task(task);
 750         return;
 751     }
 752
 753     inq = scsi_datain_unmarshall(task);
 754     if (inq == NULL) {
 755         error_report("iSCSI: Failed to unmarshall inquiry data.");
 756         itask->status   = 1;
 757         itask->complete = 1;
 758         scsi_free_scsi_task(task);
 759         return;
 760     }
 761
 762     itask->iscsilun->type = inq->periperal_device_type;
 763
 764     scsi_free_scsi_task(task);
 765
 766     switch (itask->iscsilun->type) {
 767     case TYPE_DISK:
 768         task = iscsi_readcapacity16_task(iscsi, itask->iscsilun->lun,
 769                                    iscsi_readcapacity16_cb, opaque);
 770         if (task == NULL) {
 771             error_report("iSCSI: failed to send readcapacity16 command.");
 772             itask->status   = 1;
 773             itask->complete = 1;
 774             return;
 775         }
 776         break;
 777     case TYPE_ROM:
 778         task = iscsi_readcapacity10_task(iscsi, itask->iscsilun->lun,
 779                                    0, 0,
 780                                    iscsi_readcapacity10_cb, opaque);
 781         if (task == NULL) {
 782             error_report("iSCSI: failed to send readcapacity16 command.");
 783             itask->status   = 1;
 784             itask->complete = 1;
 785             return;
 786         }
 787         break;
 788     default:
 789         itask->status   = 0;
 790         itask->complete = 1;
 791     }
 792 }
 793
 794 static void
 795 iscsi_connect_cb(struct iscsi_context *iscsi, int status, void *command_data,
 796                  void *opaque)
 797 {
 798     struct IscsiTask *itask = opaque;
 799     struct scsi_task *task;
 800
 801     if (status != 0) {
 802         itask->status   = 1;
 803         itask->complete = 1;
 804         return;
 805     }
 806
 807     task = iscsi_inquiry_task(iscsi, itask->iscsilun->lun,
 808                               0, 0, 36,
 809                               iscsi_inquiry_cb, opaque);
 810     if (task == NULL) {
 811         error_report("iSCSI: failed to send inquiry command.");
 812         itask->status   = 1;
 813         itask->complete = 1;
 814         return;
 815     }
 816 }
 817
 818 static int parse_chap(struct iscsi_context *iscsi, const char *target)
 819 {
 820     QemuOptsList *list;
 821     QemuOpts *opts;
 822     const char *user = NULL;
 823     const char *password = NULL;
 824
 825     list = qemu_find_opts("iscsi");
 826     if (!list) {
 827         return 0;
 828     }
 829
 830     opts = qemu_opts_find(list, target);
 831     if (opts == NULL) {
 832         opts = QTAILQ_FIRST(&list->head);
 833         if (!opts) {
 834             return 0;
 835         }
 836     }
 837
 838     user = qemu_opt_get(opts, "user");
 839     if (!user) {
 840         return 0;
 841     }
 842
 843     password = qemu_opt_get(opts, "password");
 844     if (!password) {
 845         error_report("CHAP username specified but no password was given");
 846         return -1;
 847     }
 848
 849     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
 850         error_report("Failed to set initiator username and password");
 851         return -1;
 852     }
 853
 854     return 0;
 855 }
 856
 857 static void parse_header_digest(struct iscsi_context *iscsi, const char *target)
 858 {
 859     QemuOptsList *list;
 860     QemuOpts *opts;
 861     const char *digest = NULL;
 862
 863     list = qemu_find_opts("iscsi");
 864     if (!list) {
 865         return;
 866     }
 867
 868     opts = qemu_opts_find(list, target);
 869     if (opts == NULL) {
 870         opts = QTAILQ_FIRST(&list->head);
 871         if (!opts) {
 872             return;
 873         }
 874     }
 875
 876     digest = qemu_opt_get(opts, "header-digest");
 877     if (!digest) {
 878         return;
 879     }
 880
 881     if (!strcmp(digest, "CRC32C")) {
 882         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
 883     } else if (!strcmp(digest, "NONE")) {
 884         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
 885     } else if (!strcmp(digest, "CRC32C-NONE")) {
 886         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
 887     } else if (!strcmp(digest, "NONE-CRC32C")) {
 888         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 889     } else {
 890         error_report("Invalid header-digest setting : %s", digest);
 891     }
 892 }
 893
 894 static char *parse_initiator_name(const char *target)
 895 {
 896     QemuOptsList *list;
 897     QemuOpts *opts;
 898     const char *name = NULL;
 899     const char *iscsi_name = qemu_get_vm_name();
 900
 901     list = qemu_find_opts("iscsi");
 902     if (list) {
 903         opts = qemu_opts_find(list, target);
 904         if (!opts) {
 905             opts = QTAILQ_FIRST(&list->head);
 906         }
 907         if (opts) {
 908             name = qemu_opt_get(opts, "initiator-name");
 909         }
 910     }
 911
 912     if (name) {
 913         return g_strdup(name);
 914     } else {
 915         return g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
 916                                iscsi_name ? ":" : "",
 917                                iscsi_name ? iscsi_name : "");
 918     }
 919 }
 920
 921 /*
 922  * We support iscsi url's on the form
 923  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
 924  */
 925 static int iscsi_open(BlockDriverState *bs, const char *filename, int flags)
 926 {
 927     IscsiLun *iscsilun = bs->opaque;
 928     struct iscsi_context *iscsi = NULL;
 929     struct iscsi_url *iscsi_url = NULL;
 930     struct IscsiTask task;
 931     char *initiator_name = NULL;
 932     int ret;
 933
 934     if ((BDRV_SECTOR_SIZE % 512) != 0) {
 935         error_report("iSCSI: Invalid BDRV_SECTOR_SIZE. "
 936                      "BDRV_SECTOR_SIZE(%lld) is not a multiple "
 937                      "of 512", BDRV_SECTOR_SIZE);
 938         return -EINVAL;
 939     }
 940
 941     iscsi_url = iscsi_parse_full_url(iscsi, filename);
 942     if (iscsi_url == NULL) {
 943         error_report("Failed to parse URL : %s %s", filename,
 944                      iscsi_get_error(iscsi));
 945         ret = -EINVAL;
 946         goto out;
 947     }
 948
 949     memset(iscsilun, 0, sizeof(IscsiLun));
 950
 951     initiator_name = parse_initiator_name(iscsi_url->target);
 952
 953     iscsi = iscsi_create_context(initiator_name);
 954     if (iscsi == NULL) {
 955         error_report("iSCSI: Failed to create iSCSI context.");
 956         ret = -ENOMEM;
 957         goto out;
 958     }
 959
 960     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
 961         error_report("iSCSI: Failed to set target name.");
 962         ret = -EINVAL;
 963         goto out;
 964     }
 965
 966     if (iscsi_url->user != NULL) {
 967         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
 968                                               iscsi_url->passwd);
 969         if (ret != 0) {
 970             error_report("Failed to set initiator username and password");
 971             ret = -EINVAL;
 972             goto out;
 973         }
 974     }
 975
 976     /* check if we got CHAP username/password via the options */
 977     if (parse_chap(iscsi, iscsi_url->target) != 0) {
 978         error_report("iSCSI: Failed to set CHAP user/password");
 979         ret = -EINVAL;
 980         goto out;
 981     }
 982
 983     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
 984         error_report("iSCSI: Failed to set session type to normal.");
 985         ret = -EINVAL;
 986         goto out;
 987     }
 988
 989     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 990
 991     /* check if we got HEADER_DIGEST via the options */
 992     parse_header_digest(iscsi, iscsi_url->target);
 993
 994     task.iscsilun = iscsilun;
 995     task.status = 0;
 996     task.complete = 0;
 997     task.bs = bs;
 998
 999     iscsilun->iscsi = iscsi;
1000     iscsilun->lun   = iscsi_url->lun;
1001
1002     if (iscsi_full_connect_async(iscsi, iscsi_url->portal, iscsi_url->lun,
1003                                  iscsi_connect_cb, &task)
1004         != 0) {
1005         error_report("iSCSI: Failed to start async connect.");
1006         ret = -EINVAL;
1007         goto out;
1008     }
1009
1010     while (!task.complete) {
1011         iscsi_set_events(iscsilun);
1012         qemu_aio_wait();
1013     }
1014     if (task.status != 0) {
1015         error_report("iSCSI: Failed to connect to LUN : %s",
1016                      iscsi_get_error(iscsi));
1017         ret = -EINVAL;
1018         goto out;
1019     }
1020
1021     /* Medium changer or tape. We dont have any emulation for this so this must
1022      * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
1023      * to read from the device to guess the image format.
1024      */
1025     if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
1026         iscsilun->type == TYPE_TAPE) {
1027         bs->sg = 1;
1028     }
1029
1030     ret = 0;
1031
1032 out:
1033     if (initiator_name != NULL) {
1034         g_free(initiator_name);
1035     }
1036     if (iscsi_url != NULL) {
1037         iscsi_destroy_url(iscsi_url);
1038     }
1039
1040     if (ret) {
1041         if (iscsi != NULL) {
1042             iscsi_destroy_context(iscsi);
1043         }
1044         memset(iscsilun, 0, sizeof(IscsiLun));
1045     }
1046     return ret;
1047 }
1048
1049 static void iscsi_close(BlockDriverState *bs)
1050 {
1051     IscsiLun *iscsilun = bs->opaque;
1052     struct iscsi_context *iscsi = iscsilun->iscsi;
1053
1054     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL, NULL);
1055     iscsi_destroy_context(iscsi);
1056     memset(iscsilun, 0, sizeof(IscsiLun));
1057 }
1058
1059 static BlockDriver bdrv_iscsi = {
1060     .format_name     = "iscsi",
1061     .protocol_name   = "iscsi",
1062
1063     .instance_size   = sizeof(IscsiLun),
1064     .bdrv_file_open  = iscsi_open,
1065     .bdrv_close      = iscsi_close,
1066
1067     .bdrv_getlength  = iscsi_getlength,
1068
1069     .bdrv_aio_readv  = iscsi_aio_readv,
1070     .bdrv_aio_writev = iscsi_aio_writev,
1071     .bdrv_aio_flush  = iscsi_aio_flush,
1072
1073     .bdrv_aio_discard = iscsi_aio_discard,
1074
1075 #ifdef __linux__
1076     .bdrv_ioctl       = iscsi_ioctl,
1077     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1078 #endif
1079 };
1080
1081 static void iscsi_block_init(void)
1082 {
1083     bdrv_register(&bdrv_iscsi);
1084 }
1085
1086 block_init(iscsi_block_init);