block/iscsi.c

   1 /*
   2  * QEMU Block driver for iSCSI images
   3  *
   4  * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
   5  * Copyright (c) 2012-2013 Peter Lieven <pl@kamp.de>
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  23  * THE SOFTWARE.
  24  */
  25
  26 #include "config-host.h"
  27
  28 #include <poll.h>
  29 #include <arpa/inet.h>
  30 #include "qemu-common.h"
  31 #include "qemu/config-file.h"
  32 #include "qemu/error-report.h"
  33 #include "block/block_int.h"
  34 #include "trace.h"
  35 #include "block/scsi.h"
  36 #include "qemu/iov.h"
  37 #include "sysemu/sysemu.h"
  38 #include "qmp-commands.h"
  39
  40 #include <iscsi/iscsi.h>
  41 #include <iscsi/scsi-lowlevel.h>
  42
  43 #ifdef __linux__
  44 #include <scsi/sg.h>
  45 #include <block/scsi.h>
  46 #endif
  47
  48 typedef struct IscsiLun {
  49     struct iscsi_context *iscsi;
  50     int lun;
  51     enum scsi_inquiry_peripheral_device_type type;
  52     int block_size;
  53     uint64_t num_blocks;
  54     int events;
  55     QEMUTimer *nop_timer;
  56     uint8_t lbpme;
  57     uint8_t lbprz;
  58     struct scsi_inquiry_logical_block_provisioning lbp;
  59     struct scsi_inquiry_block_limits bl;
  60     unsigned char *zeroblock;
  61 } IscsiLun;
  62
  63 typedef struct IscsiTask {
  64     int status;
  65     int complete;
  66     int retries;
  67     int do_retry;
  68     struct scsi_task *task;
  69     Coroutine *co;
  70 } IscsiTask;
  71
  72 typedef struct IscsiAIOCB {
  73     BlockDriverAIOCB common;
  74     QEMUIOVector *qiov;
  75     QEMUBH *bh;
  76     IscsiLun *iscsilun;
  77     struct scsi_task *task;
  78     uint8_t *buf;
  79     int status;
  80     int canceled;
  81     int retries;
  82     int64_t sector_num;
  83     int nb_sectors;
  84 #ifdef __linux__
  85     sg_io_hdr_t *ioh;
  86 #endif
  87 } IscsiAIOCB;
  88
  89 #define NOP_INTERVAL 5000
  90 #define MAX_NOP_FAILURES 3
  91 #define ISCSI_CMD_RETRIES 5
  92
  93 static void
  94 iscsi_bh_cb(void *p)
  95 {
  96     IscsiAIOCB *acb = p;
  97
  98     qemu_bh_delete(acb->bh);
  99
 100     g_free(acb->buf);
 101     acb->buf = NULL;
 102
 103     if (acb->canceled == 0) {
 104         acb->common.cb(acb->common.opaque, acb->status);
 105     }
 106
 107     if (acb->task != NULL) {
 108         scsi_free_scsi_task(acb->task);
 109         acb->task = NULL;
 110     }
 111
 112     qemu_aio_release(acb);
 113 }
 114
 115 static void
 116 iscsi_schedule_bh(IscsiAIOCB *acb)
 117 {
 118     if (acb->bh) {
 119         return;
 120     }
 121     acb->bh = qemu_bh_new(iscsi_bh_cb, acb);
 122     qemu_bh_schedule(acb->bh);
 123 }
 124
 125 static void
 126 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
 127                         void *command_data, void *opaque)
 128 {
 129     struct IscsiTask *iTask = opaque;
 130     struct scsi_task *task = command_data;
 131
 132     iTask->complete = 1;
 133     iTask->status = status;
 134     iTask->do_retry = 0;
 135     iTask->task = task;
 136
 137     if (iTask->retries-- > 0 && status == SCSI_STATUS_CHECK_CONDITION
 138         && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
 139         iTask->do_retry = 1;
 140         goto out;
 141     }
 142
 143     if (status != SCSI_STATUS_GOOD) {
 144         error_report("iSCSI: Failure. %s", iscsi_get_error(iscsi));
 145     }
 146
 147 out:
 148     if (iTask->co) {
 149         qemu_coroutine_enter(iTask->co, NULL);
 150     }
 151 }
 152
 153 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
 154 {
 155     *iTask = (struct IscsiTask) {
 156         .co         = qemu_coroutine_self(),
 157         .retries    = ISCSI_CMD_RETRIES,
 158     };
 159 }
 160
 161 static void
 162 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
 163                     void *private_data)
 164 {
 165     IscsiAIOCB *acb = private_data;
 166
 167     acb->status = -ECANCELED;
 168     iscsi_schedule_bh(acb);
 169 }
 170
 171 static void
 172 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
 173 {
 174     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
 175     IscsiLun *iscsilun = acb->iscsilun;
 176
 177     if (acb->status != -EINPROGRESS) {
 178         return;
 179     }
 180
 181     acb->canceled = 1;
 182
 183     /* send a task mgmt call to the target to cancel the task on the target */
 184     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
 185                                      iscsi_abort_task_cb, acb);
 186
 187     while (acb->status == -EINPROGRESS) {
 188         qemu_aio_wait();
 189     }
 190 }
 191
 192 static const AIOCBInfo iscsi_aiocb_info = {
 193     .aiocb_size         = sizeof(IscsiAIOCB),
 194     .cancel             = iscsi_aio_cancel,
 195 };
 196
 197
 198 static void iscsi_process_read(void *arg);
 199 static void iscsi_process_write(void *arg);
 200
 201 static void
 202 iscsi_set_events(IscsiLun *iscsilun)
 203 {
 204     struct iscsi_context *iscsi = iscsilun->iscsi;
 205     int ev;
 206
 207     /* We always register a read handler.  */
 208     ev = POLLIN;
 209     ev |= iscsi_which_events(iscsi);
 210     if (ev != iscsilun->events) {
 211         qemu_aio_set_fd_handler(iscsi_get_fd(iscsi),
 212                       iscsi_process_read,
 213                       (ev & POLLOUT) ? iscsi_process_write : NULL,
 214                       iscsilun);
 215
 216     }
 217
 218     iscsilun->events = ev;
 219 }
 220
 221 static void
 222 iscsi_process_read(void *arg)
 223 {
 224     IscsiLun *iscsilun = arg;
 225     struct iscsi_context *iscsi = iscsilun->iscsi;
 226
 227     iscsi_service(iscsi, POLLIN);
 228     iscsi_set_events(iscsilun);
 229 }
 230
 231 static void
 232 iscsi_process_write(void *arg)
 233 {
 234     IscsiLun *iscsilun = arg;
 235     struct iscsi_context *iscsi = iscsilun->iscsi;
 236
 237     iscsi_service(iscsi, POLLOUT);
 238     iscsi_set_events(iscsilun);
 239 }
 240
 241 static int
 242 iscsi_aio_writev_acb(IscsiAIOCB *acb);
 243
 244 static void
 245 iscsi_aio_write16_cb(struct iscsi_context *iscsi, int status,
 246                      void *command_data, void *opaque)
 247 {
 248     IscsiAIOCB *acb = opaque;
 249
 250     trace_iscsi_aio_write16_cb(iscsi, status, acb, acb->canceled);
 251
 252     g_free(acb->buf);
 253     acb->buf = NULL;
 254
 255     if (acb->canceled != 0) {
 256         return;
 257     }
 258
 259     acb->status = 0;
 260     if (status != 0) {
 261         if (status == SCSI_STATUS_CHECK_CONDITION
 262             && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
 263             && acb->retries-- > 0) {
 264             scsi_free_scsi_task(acb->task);
 265             acb->task = NULL;
 266             if (iscsi_aio_writev_acb(acb) == 0) {
 267                 iscsi_set_events(acb->iscsilun);
 268                 return;
 269             }
 270         }
 271         error_report("Failed to write16 data to iSCSI lun. %s",
 272                      iscsi_get_error(iscsi));
 273         acb->status = -EIO;
 274     }
 275
 276     iscsi_schedule_bh(acb);
 277 }
 278
 279 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
 280 {
 281     return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
 282 }
 283
 284 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
 285 {
 286     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
 287 }
 288
 289 static bool is_request_lun_aligned(int64_t sector_num, int nb_sectors,
 290                                       IscsiLun *iscsilun)
 291 {
 292     if ((sector_num * BDRV_SECTOR_SIZE) % iscsilun->block_size ||
 293         (nb_sectors * BDRV_SECTOR_SIZE) % iscsilun->block_size) {
 294             error_report("iSCSI misaligned request: "
 295                          "iscsilun->block_size %u, sector_num %" PRIi64
 296                          ", nb_sectors %d",
 297                          iscsilun->block_size, sector_num, nb_sectors);
 298             return 0;
 299     }
 300     return 1;
 301 }
 302
 303 static int
 304 iscsi_aio_writev_acb(IscsiAIOCB *acb)
 305 {
 306     struct iscsi_context *iscsi = acb->iscsilun->iscsi;
 307     size_t size;
 308     uint32_t num_sectors;
 309     uint64_t lba;
 310 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 311     struct iscsi_data data;
 312 #endif
 313     int ret;
 314
 315     acb->canceled   = 0;
 316     acb->bh         = NULL;
 317     acb->status     = -EINPROGRESS;
 318     acb->buf        = NULL;
 319
 320     /* this will allow us to get rid of 'buf' completely */
 321     size = acb->nb_sectors * BDRV_SECTOR_SIZE;
 322
 323 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 324     data.size = MIN(size, acb->qiov->size);
 325
 326     /* if the iovec only contains one buffer we can pass it directly */
 327     if (acb->qiov->niov == 1) {
 328         data.data = acb->qiov->iov[0].iov_base;
 329     } else {
 330         acb->buf = g_malloc(data.size);
 331         qemu_iovec_to_buf(acb->qiov, 0, acb->buf, data.size);
 332         data.data = acb->buf;
 333     }
 334 #endif
 335
 336     acb->task = malloc(sizeof(struct scsi_task));
 337     if (acb->task == NULL) {
 338         error_report("iSCSI: Failed to allocate task for scsi WRITE16 "
 339                      "command. %s", iscsi_get_error(iscsi));
 340         return -1;
 341     }
 342     memset(acb->task, 0, sizeof(struct scsi_task));
 343
 344     acb->task->xfer_dir = SCSI_XFER_WRITE;
 345     acb->task->cdb_size = 16;
 346     acb->task->cdb[0] = 0x8a;
 347     lba = sector_qemu2lun(acb->sector_num, acb->iscsilun);
 348     *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 349     *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 350     num_sectors = sector_qemu2lun(acb->nb_sectors, acb->iscsilun);
 351     *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 352     acb->task->expxferlen = size;
 353
 354 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 355     ret = iscsi_scsi_command_async(iscsi, acb->iscsilun->lun, acb->task,
 356                                    iscsi_aio_write16_cb,
 357                                    NULL,
 358                                    acb);
 359 #else
 360     ret = iscsi_scsi_command_async(iscsi, acb->iscsilun->lun, acb->task,
 361                                    iscsi_aio_write16_cb,
 362                                    &data,
 363                                    acb);
 364 #endif
 365     if (ret != 0) {
 366         scsi_free_scsi_task(acb->task);
 367         g_free(acb->buf);
 368         return -1;
 369     }
 370
 371 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 372     scsi_task_set_iov_out(acb->task, (struct scsi_iovec*) acb->qiov->iov, acb->qiov->niov);
 373 #endif
 374
 375     return 0;
 376 }
 377
 378 static BlockDriverAIOCB *
 379 iscsi_aio_writev(BlockDriverState *bs, int64_t sector_num,
 380                  QEMUIOVector *qiov, int nb_sectors,
 381                  BlockDriverCompletionFunc *cb,
 382                  void *opaque)
 383 {
 384     IscsiLun *iscsilun = bs->opaque;
 385     IscsiAIOCB *acb;
 386
 387     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 388         return NULL;
 389     }
 390
 391     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 392     trace_iscsi_aio_writev(iscsilun->iscsi, sector_num, nb_sectors, opaque, acb);
 393
 394     acb->iscsilun    = iscsilun;
 395     acb->qiov        = qiov;
 396     acb->nb_sectors  = nb_sectors;
 397     acb->sector_num  = sector_num;
 398     acb->retries     = ISCSI_CMD_RETRIES;
 399
 400     if (iscsi_aio_writev_acb(acb) != 0) {
 401         qemu_aio_release(acb);
 402         return NULL;
 403     }
 404
 405     iscsi_set_events(iscsilun);
 406     return &acb->common;
 407 }
 408
 409 static int
 410 iscsi_aio_readv_acb(IscsiAIOCB *acb);
 411
 412 static void
 413 iscsi_aio_read16_cb(struct iscsi_context *iscsi, int status,
 414                     void *command_data, void *opaque)
 415 {
 416     IscsiAIOCB *acb = opaque;
 417
 418     trace_iscsi_aio_read16_cb(iscsi, status, acb, acb->canceled);
 419
 420     if (acb->canceled != 0) {
 421         return;
 422     }
 423
 424     acb->status = 0;
 425     if (status != 0) {
 426         if (status == SCSI_STATUS_CHECK_CONDITION
 427             && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
 428             && acb->retries-- > 0) {
 429             scsi_free_scsi_task(acb->task);
 430             acb->task = NULL;
 431             if (iscsi_aio_readv_acb(acb) == 0) {
 432                 iscsi_set_events(acb->iscsilun);
 433                 return;
 434             }
 435         }
 436         error_report("Failed to read16 data from iSCSI lun. %s",
 437                      iscsi_get_error(iscsi));
 438         acb->status = -EIO;
 439     }
 440
 441     iscsi_schedule_bh(acb);
 442 }
 443
 444 static int
 445 iscsi_aio_readv_acb(IscsiAIOCB *acb)
 446 {
 447     struct iscsi_context *iscsi = acb->iscsilun->iscsi;
 448     size_t size;
 449     uint64_t lba;
 450     uint32_t num_sectors;
 451     int ret;
 452 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 453     int i;
 454 #endif
 455
 456     acb->canceled    = 0;
 457     acb->bh          = NULL;
 458     acb->status      = -EINPROGRESS;
 459     acb->buf         = NULL;
 460
 461     size = acb->nb_sectors * BDRV_SECTOR_SIZE;
 462
 463     acb->task = malloc(sizeof(struct scsi_task));
 464     if (acb->task == NULL) {
 465         error_report("iSCSI: Failed to allocate task for scsi READ16 "
 466                      "command. %s", iscsi_get_error(iscsi));
 467         return -1;
 468     }
 469     memset(acb->task, 0, sizeof(struct scsi_task));
 470
 471     acb->task->xfer_dir = SCSI_XFER_READ;
 472     acb->task->expxferlen = size;
 473     lba = sector_qemu2lun(acb->sector_num, acb->iscsilun);
 474     num_sectors = sector_qemu2lun(acb->nb_sectors, acb->iscsilun);
 475
 476     switch (acb->iscsilun->type) {
 477     case TYPE_DISK:
 478         acb->task->cdb_size = 16;
 479         acb->task->cdb[0]  = 0x88;
 480         *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 481         *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 482         *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 483         break;
 484     default:
 485         acb->task->cdb_size = 10;
 486         acb->task->cdb[0]  = 0x28;
 487         *(uint32_t *)&acb->task->cdb[2] = htonl(lba);
 488         *(uint16_t *)&acb->task->cdb[7] = htons(num_sectors);
 489         break;
 490     }
 491
 492     ret = iscsi_scsi_command_async(iscsi, acb->iscsilun->lun, acb->task,
 493                                    iscsi_aio_read16_cb,
 494                                    NULL,
 495                                    acb);
 496     if (ret != 0) {
 497         scsi_free_scsi_task(acb->task);
 498         return -1;
 499     }
 500
 501 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 502     scsi_task_set_iov_in(acb->task, (struct scsi_iovec*) acb->qiov->iov, acb->qiov->niov);
 503 #else
 504     for (i = 0; i < acb->qiov->niov; i++) {
 505         scsi_task_add_data_in_buffer(acb->task,
 506                 acb->qiov->iov[i].iov_len,
 507                 acb->qiov->iov[i].iov_base);
 508     }
 509 #endif
 510     return 0;
 511 }
 512
 513 static BlockDriverAIOCB *
 514 iscsi_aio_readv(BlockDriverState *bs, int64_t sector_num,
 515                 QEMUIOVector *qiov, int nb_sectors,
 516                 BlockDriverCompletionFunc *cb,
 517                 void *opaque)
 518 {
 519     IscsiLun *iscsilun = bs->opaque;
 520     IscsiAIOCB *acb;
 521
 522     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 523         return NULL;
 524     }
 525
 526     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 527     trace_iscsi_aio_readv(iscsilun->iscsi, sector_num, nb_sectors, opaque, acb);
 528
 529     acb->nb_sectors  = nb_sectors;
 530     acb->sector_num  = sector_num;
 531     acb->iscsilun    = iscsilun;
 532     acb->qiov        = qiov;
 533     acb->retries     = ISCSI_CMD_RETRIES;
 534
 535     if (iscsi_aio_readv_acb(acb) != 0) {
 536         qemu_aio_release(acb);
 537         return NULL;
 538     }
 539
 540     iscsi_set_events(iscsilun);
 541     return &acb->common;
 542 }
 543
 544 static int
 545 iscsi_aio_flush_acb(IscsiAIOCB *acb);
 546
 547 static void
 548 iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
 549                      void *command_data, void *opaque)
 550 {
 551     IscsiAIOCB *acb = opaque;
 552
 553     if (acb->canceled != 0) {
 554         return;
 555     }
 556
 557     acb->status = 0;
 558     if (status != 0) {
 559         if (status == SCSI_STATUS_CHECK_CONDITION
 560             && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
 561             && acb->retries-- > 0) {
 562             scsi_free_scsi_task(acb->task);
 563             acb->task = NULL;
 564             if (iscsi_aio_flush_acb(acb) == 0) {
 565                 iscsi_set_events(acb->iscsilun);
 566                 return;
 567             }
 568         }
 569         error_report("Failed to sync10 data on iSCSI lun. %s",
 570                      iscsi_get_error(iscsi));
 571         acb->status = -EIO;
 572     }
 573
 574     iscsi_schedule_bh(acb);
 575 }
 576
 577 static int
 578 iscsi_aio_flush_acb(IscsiAIOCB *acb)
 579 {
 580     struct iscsi_context *iscsi = acb->iscsilun->iscsi;
 581
 582     acb->canceled   = 0;
 583     acb->bh         = NULL;
 584     acb->status     = -EINPROGRESS;
 585     acb->buf        = NULL;
 586
 587     acb->task = iscsi_synchronizecache10_task(iscsi, acb->iscsilun->lun,
 588                                          0, 0, 0, 0,
 589                                          iscsi_synccache10_cb,
 590                                          acb);
 591     if (acb->task == NULL) {
 592         error_report("iSCSI: Failed to send synchronizecache10 command. %s",
 593                      iscsi_get_error(iscsi));
 594         return -1;
 595     }
 596
 597     return 0;
 598 }
 599
 600 static BlockDriverAIOCB *
 601 iscsi_aio_flush(BlockDriverState *bs,
 602                 BlockDriverCompletionFunc *cb, void *opaque)
 603 {
 604     IscsiLun *iscsilun = bs->opaque;
 605
 606     IscsiAIOCB *acb;
 607
 608     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 609
 610     acb->iscsilun    = iscsilun;
 611     acb->retries     = ISCSI_CMD_RETRIES;
 612
 613     if (iscsi_aio_flush_acb(acb) != 0) {
 614         qemu_aio_release(acb);
 615         return NULL;
 616     }
 617
 618     iscsi_set_events(iscsilun);
 619
 620     return &acb->common;
 621 }
 622
 623 #ifdef __linux__
 624 static void
 625 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 626                      void *command_data, void *opaque)
 627 {
 628     IscsiAIOCB *acb = opaque;
 629
 630     g_free(acb->buf);
 631     acb->buf = NULL;
 632
 633     if (acb->canceled != 0) {
 634         return;
 635     }
 636
 637     acb->status = 0;
 638     if (status < 0) {
 639         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 640                      iscsi_get_error(iscsi));
 641         acb->status = -EIO;
 642     }
 643
 644     acb->ioh->driver_status = 0;
 645     acb->ioh->host_status   = 0;
 646     acb->ioh->resid         = 0;
 647
 648 #define SG_ERR_DRIVER_SENSE    0x08
 649
 650     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 651         int ss;
 652
 653         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 654
 655         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 656         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 657              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 658         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 659     }
 660
 661     iscsi_schedule_bh(acb);
 662 }
 663
 664 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 665         unsigned long int req, void *buf,
 666         BlockDriverCompletionFunc *cb, void *opaque)
 667 {
 668     IscsiLun *iscsilun = bs->opaque;
 669     struct iscsi_context *iscsi = iscsilun->iscsi;
 670     struct iscsi_data data;
 671     IscsiAIOCB *acb;
 672
 673     assert(req == SG_IO);
 674
 675     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 676
 677     acb->iscsilun = iscsilun;
 678     acb->canceled    = 0;
 679     acb->bh          = NULL;
 680     acb->status      = -EINPROGRESS;
 681     acb->buf         = NULL;
 682     acb->ioh         = buf;
 683
 684     acb->task = malloc(sizeof(struct scsi_task));
 685     if (acb->task == NULL) {
 686         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 687                      iscsi_get_error(iscsi));
 688         qemu_aio_release(acb);
 689         return NULL;
 690     }
 691     memset(acb->task, 0, sizeof(struct scsi_task));
 692
 693     switch (acb->ioh->dxfer_direction) {
 694     case SG_DXFER_TO_DEV:
 695         acb->task->xfer_dir = SCSI_XFER_WRITE;
 696         break;
 697     case SG_DXFER_FROM_DEV:
 698         acb->task->xfer_dir = SCSI_XFER_READ;
 699         break;
 700     default:
 701         acb->task->xfer_dir = SCSI_XFER_NONE;
 702         break;
 703     }
 704
 705     acb->task->cdb_size = acb->ioh->cmd_len;
 706     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 707     acb->task->expxferlen = acb->ioh->dxfer_len;
 708
 709     data.size = 0;
 710     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 711         if (acb->ioh->iovec_count == 0) {
 712             data.data = acb->ioh->dxferp;
 713             data.size = acb->ioh->dxfer_len;
 714         } else {
 715 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 716             scsi_task_set_iov_out(acb->task,
 717                                  (struct scsi_iovec *) acb->ioh->dxferp,
 718                                  acb->ioh->iovec_count);
 719 #else
 720             struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
 721
 722             acb->buf = g_malloc(acb->ioh->dxfer_len);
 723             data.data = acb->buf;
 724             data.size = iov_to_buf(iov, acb->ioh->iovec_count, 0,
 725                                    acb->buf, acb->ioh->dxfer_len);
 726 #endif
 727         }
 728     }
 729
 730     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 731                                  iscsi_aio_ioctl_cb,
 732                                  (data.size > 0) ? &data : NULL,
 733                                  acb) != 0) {
 734         scsi_free_scsi_task(acb->task);
 735         qemu_aio_release(acb);
 736         return NULL;
 737     }
 738
 739     /* tell libiscsi to read straight into the buffer we got from ioctl */
 740     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 741         if (acb->ioh->iovec_count == 0) {
 742             scsi_task_add_data_in_buffer(acb->task,
 743                                          acb->ioh->dxfer_len,
 744                                          acb->ioh->dxferp);
 745         } else {
 746 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 747             scsi_task_set_iov_in(acb->task,
 748                                  (struct scsi_iovec *) acb->ioh->dxferp,
 749                                  acb->ioh->iovec_count);
 750 #else
 751             int i;
 752             for (i = 0; i < acb->ioh->iovec_count; i++) {
 753                 struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
 754
 755                 scsi_task_add_data_in_buffer(acb->task,
 756                     iov[i].iov_len,
 757                     iov[i].iov_base);
 758             }
 759 #endif
 760         }
 761     }
 762
 763     iscsi_set_events(iscsilun);
 764
 765     return &acb->common;
 766 }
 767
 768
 769 static void ioctl_cb(void *opaque, int status)
 770 {
 771     int *p_status = opaque;
 772     *p_status = status;
 773 }
 774
 775 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 776 {
 777     IscsiLun *iscsilun = bs->opaque;
 778     int status;
 779
 780     switch (req) {
 781     case SG_GET_VERSION_NUM:
 782         *(int *)buf = 30000;
 783         break;
 784     case SG_GET_SCSI_ID:
 785         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 786         break;
 787     case SG_IO:
 788         status = -EINPROGRESS;
 789         iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status);
 790
 791         while (status == -EINPROGRESS) {
 792             qemu_aio_wait();
 793         }
 794
 795         return 0;
 796     default:
 797         return -1;
 798     }
 799     return 0;
 800 }
 801 #endif
 802
 803 static int64_t
 804 iscsi_getlength(BlockDriverState *bs)
 805 {
 806     IscsiLun *iscsilun = bs->opaque;
 807     int64_t len;
 808
 809     len  = iscsilun->num_blocks;
 810     len *= iscsilun->block_size;
 811
 812     return len;
 813 }
 814
 815 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 816
 817 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
 818                                                   int64_t sector_num,
 819                                                   int nb_sectors, int *pnum)
 820 {
 821     IscsiLun *iscsilun = bs->opaque;
 822     struct scsi_get_lba_status *lbas = NULL;
 823     struct scsi_lba_status_descriptor *lbasd = NULL;
 824     struct IscsiTask iTask;
 825     int64_t ret;
 826
 827     iscsi_co_init_iscsitask(iscsilun, &iTask);
 828
 829     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 830         ret = -EINVAL;
 831         goto out;
 832     }
 833
 834     /* default to all sectors allocated */
 835     ret = BDRV_BLOCK_DATA;
 836     ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
 837     *pnum = nb_sectors;
 838
 839     /* LUN does not support logical block provisioning */
 840     if (iscsilun->lbpme == 0) {
 841         goto out;
 842     }
 843
 844 retry:
 845     if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
 846                                   sector_qemu2lun(sector_num, iscsilun),
 847                                   8 + 16, iscsi_co_generic_cb,
 848                                   &iTask) == NULL) {
 849         ret = -EIO;
 850         goto out;
 851     }
 852
 853     while (!iTask.complete) {
 854         iscsi_set_events(iscsilun);
 855         qemu_coroutine_yield();
 856     }
 857
 858     if (iTask.do_retry) {
 859         if (iTask.task != NULL) {
 860             scsi_free_scsi_task(iTask.task);
 861             iTask.task = NULL;
 862         }
 863         goto retry;
 864     }
 865
 866     if (iTask.status != SCSI_STATUS_GOOD) {
 867         /* in case the get_lba_status_callout fails (i.e.
 868          * because the device is busy or the cmd is not
 869          * supported) we pretend all blocks are allocated
 870          * for backwards compatibility */
 871         goto out;
 872     }
 873
 874     lbas = scsi_datain_unmarshall(iTask.task);
 875     if (lbas == NULL) {
 876         ret = -EIO;
 877         goto out;
 878     }
 879
 880     lbasd = &lbas->descriptors[0];
 881
 882     if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
 883         ret = -EIO;
 884         goto out;
 885     }
 886
 887     *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
 888     if (*pnum > nb_sectors) {
 889         *pnum = nb_sectors;
 890     }
 891
 892     if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
 893         lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
 894         ret &= ~BDRV_BLOCK_DATA;
 895         if (iscsilun->lbprz) {
 896             ret |= BDRV_BLOCK_ZERO;
 897         }
 898     }
 899
 900 out:
 901     if (iTask.task != NULL) {
 902         scsi_free_scsi_task(iTask.task);
 903     }
 904     return ret;
 905 }
 906
 907 #endif /* LIBISCSI_FEATURE_IOVECTOR */
 908
 909 static int
 910 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num,
 911                                    int nb_sectors)
 912 {
 913     IscsiLun *iscsilun = bs->opaque;
 914     struct IscsiTask iTask;
 915     struct unmap_list list;
 916
 917     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 918         return -EINVAL;
 919     }
 920
 921     if (!iscsilun->lbp.lbpu) {
 922         /* UNMAP is not supported by the target */
 923         return 0;
 924     }
 925
 926     list.lba = sector_qemu2lun(sector_num, iscsilun);
 927     list.num = sector_qemu2lun(nb_sectors, iscsilun);
 928
 929     iscsi_co_init_iscsitask(iscsilun, &iTask);
 930 retry:
 931     if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
 932                      iscsi_co_generic_cb, &iTask) == NULL) {
 933         return -EIO;
 934     }
 935
 936     while (!iTask.complete) {
 937         iscsi_set_events(iscsilun);
 938         qemu_coroutine_yield();
 939     }
 940
 941     if (iTask.task != NULL) {
 942         scsi_free_scsi_task(iTask.task);
 943         iTask.task = NULL;
 944     }
 945
 946     if (iTask.do_retry) {
 947         goto retry;
 948     }
 949
 950     if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
 951         /* the target might fail with a check condition if it
 952            is not happy with the alignment of the UNMAP request
 953            we silently fail in this case */
 954         return 0;
 955     }
 956
 957     if (iTask.status != SCSI_STATUS_GOOD) {
 958         return -EIO;
 959     }
 960
 961     return 0;
 962 }
 963
 964 #if defined(SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED)
 965
 966 static int
 967 coroutine_fn iscsi_co_write_zeroes(BlockDriverState *bs, int64_t sector_num,
 968                                    int nb_sectors, BdrvRequestFlags flags)
 969 {
 970     IscsiLun *iscsilun = bs->opaque;
 971     struct IscsiTask iTask;
 972     uint64_t lba;
 973     uint32_t nb_blocks;
 974
 975     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 976         return -EINVAL;
 977     }
 978
 979     if (!iscsilun->lbp.lbpws) {
 980         /* WRITE SAME is not supported by the target */
 981         return -ENOTSUP;
 982     }
 983
 984     lba = sector_qemu2lun(sector_num, iscsilun);
 985     nb_blocks = sector_qemu2lun(nb_sectors, iscsilun);
 986
 987     if (iscsilun->zeroblock == NULL) {
 988         iscsilun->zeroblock = g_malloc0(iscsilun->block_size);
 989     }
 990
 991     iscsi_co_init_iscsitask(iscsilun, &iTask);
 992 retry:
 993     if (iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
 994                                iscsilun->zeroblock, iscsilun->block_size,
 995                                nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
 996                                0, 0, iscsi_co_generic_cb, &iTask) == NULL) {
 997         return -EIO;
 998     }
 999
1000     while (!iTask.complete) {
1001         iscsi_set_events(iscsilun);
1002         qemu_coroutine_yield();
1003     }
1004
1005     if (iTask.task != NULL) {
1006         scsi_free_scsi_task(iTask.task);
1007         iTask.task = NULL;
1008     }
1009
1010     if (iTask.do_retry) {
1011         goto retry;
1012     }
1013
1014     if (iTask.status != SCSI_STATUS_GOOD) {
1015         return -EIO;
1016     }
1017
1018     return 0;
1019 }
1020
1021 #endif /* SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED */
1022
1023 static int parse_chap(struct iscsi_context *iscsi, const char *target)
1024 {
1025     QemuOptsList *list;
1026     QemuOpts *opts;
1027     const char *user = NULL;
1028     const char *password = NULL;
1029
1030     list = qemu_find_opts("iscsi");
1031     if (!list) {
1032         return 0;
1033     }
1034
1035     opts = qemu_opts_find(list, target);
1036     if (opts == NULL) {
1037         opts = QTAILQ_FIRST(&list->head);
1038         if (!opts) {
1039             return 0;
1040         }
1041     }
1042
1043     user = qemu_opt_get(opts, "user");
1044     if (!user) {
1045         return 0;
1046     }
1047
1048     password = qemu_opt_get(opts, "password");
1049     if (!password) {
1050         error_report("CHAP username specified but no password was given");
1051         return -1;
1052     }
1053
1054     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1055         error_report("Failed to set initiator username and password");
1056         return -1;
1057     }
1058
1059     return 0;
1060 }
1061
1062 static void parse_header_digest(struct iscsi_context *iscsi, const char *target)
1063 {
1064     QemuOptsList *list;
1065     QemuOpts *opts;
1066     const char *digest = NULL;
1067
1068     list = qemu_find_opts("iscsi");
1069     if (!list) {
1070         return;
1071     }
1072
1073     opts = qemu_opts_find(list, target);
1074     if (opts == NULL) {
1075         opts = QTAILQ_FIRST(&list->head);
1076         if (!opts) {
1077             return;
1078         }
1079     }
1080
1081     digest = qemu_opt_get(opts, "header-digest");
1082     if (!digest) {
1083         return;
1084     }
1085
1086     if (!strcmp(digest, "CRC32C")) {
1087         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1088     } else if (!strcmp(digest, "NONE")) {
1089         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1090     } else if (!strcmp(digest, "CRC32C-NONE")) {
1091         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1092     } else if (!strcmp(digest, "NONE-CRC32C")) {
1093         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1094     } else {
1095         error_report("Invalid header-digest setting : %s", digest);
1096     }
1097 }
1098
1099 static char *parse_initiator_name(const char *target)
1100 {
1101     QemuOptsList *list;
1102     QemuOpts *opts;
1103     const char *name;
1104     char *iscsi_name;
1105     UuidInfo *uuid_info;
1106
1107     list = qemu_find_opts("iscsi");
1108     if (list) {
1109         opts = qemu_opts_find(list, target);
1110         if (!opts) {
1111             opts = QTAILQ_FIRST(&list->head);
1112         }
1113         if (opts) {
1114             name = qemu_opt_get(opts, "initiator-name");
1115             if (name) {
1116                 return g_strdup(name);
1117             }
1118         }
1119     }
1120
1121     uuid_info = qmp_query_uuid(NULL);
1122     if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1123         name = qemu_get_vm_name();
1124     } else {
1125         name = uuid_info->UUID;
1126     }
1127     iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1128                                  name ? ":" : "", name ? name : "");
1129     qapi_free_UuidInfo(uuid_info);
1130     return iscsi_name;
1131 }
1132
1133 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
1134 static void iscsi_nop_timed_event(void *opaque)
1135 {
1136     IscsiLun *iscsilun = opaque;
1137
1138     if (iscsi_get_nops_in_flight(iscsilun->iscsi) > MAX_NOP_FAILURES) {
1139         error_report("iSCSI: NOP timeout. Reconnecting...");
1140         iscsi_reconnect(iscsilun->iscsi);
1141     }
1142
1143     if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1144         error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1145         return;
1146     }
1147
1148     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1149     iscsi_set_events(iscsilun);
1150 }
1151 #endif
1152
1153 static int iscsi_readcapacity_sync(IscsiLun *iscsilun)
1154 {
1155     struct scsi_task *task = NULL;
1156     struct scsi_readcapacity10 *rc10 = NULL;
1157     struct scsi_readcapacity16 *rc16 = NULL;
1158     int ret = 0;
1159     int retries = ISCSI_CMD_RETRIES;
1160
1161     do {
1162         if (task != NULL) {
1163             scsi_free_scsi_task(task);
1164             task = NULL;
1165         }
1166
1167         switch (iscsilun->type) {
1168         case TYPE_DISK:
1169             task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1170             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1171                 rc16 = scsi_datain_unmarshall(task);
1172                 if (rc16 == NULL) {
1173                     error_report("iSCSI: Failed to unmarshall readcapacity16 data.");
1174                     ret = -EINVAL;
1175                 } else {
1176                     iscsilun->block_size = rc16->block_length;
1177                     iscsilun->num_blocks = rc16->returned_lba + 1;
1178                     iscsilun->lbpme = rc16->lbpme;
1179                     iscsilun->lbprz = rc16->lbprz;
1180                 }
1181             }
1182             break;
1183         case TYPE_ROM:
1184             task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1185             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1186                 rc10 = scsi_datain_unmarshall(task);
1187                 if (rc10 == NULL) {
1188                     error_report("iSCSI: Failed to unmarshall readcapacity10 data.");
1189                     ret = -EINVAL;
1190                 } else {
1191                     iscsilun->block_size = rc10->block_size;
1192                     if (rc10->lba == 0) {
1193                         /* blank disk loaded */
1194                         iscsilun->num_blocks = 0;
1195                     } else {
1196                         iscsilun->num_blocks = rc10->lba + 1;
1197                     }
1198                 }
1199             }
1200             break;
1201         default:
1202             return 0;
1203         }
1204     } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1205              && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1206              && retries-- > 0);
1207
1208     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1209         error_report("iSCSI: failed to send readcapacity10 command.");
1210         ret = -EINVAL;
1211     }
1212     if (task) {
1213         scsi_free_scsi_task(task);
1214     }
1215     return ret;
1216 }
1217
1218 /* TODO Convert to fine grained options */
1219 static QemuOptsList runtime_opts = {
1220     .name = "iscsi",
1221     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1222     .desc = {
1223         {
1224             .name = "filename",
1225             .type = QEMU_OPT_STRING,
1226             .help = "URL to the iscsi image",
1227         },
1228         { /* end of list */ }
1229     },
1230 };
1231
1232 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi,
1233                                           int lun, int evpd, int pc) {
1234         int full_size;
1235         struct scsi_task *task = NULL;
1236         task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1237         if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1238             goto fail;
1239         }
1240         full_size = scsi_datain_getfullsize(task);
1241         if (full_size > task->datain.size) {
1242             scsi_free_scsi_task(task);
1243
1244             /* we need more data for the full list */
1245             task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1246             if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1247                 goto fail;
1248             }
1249         }
1250
1251         return task;
1252
1253 fail:
1254         error_report("iSCSI: Inquiry command failed : %s",
1255                      iscsi_get_error(iscsi));
1256         if (task) {
1257             scsi_free_scsi_task(task);
1258             return NULL;
1259         }
1260         return NULL;
1261 }
1262
1263 /*
1264  * We support iscsi url's on the form
1265  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1266  */
1267 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1268                       Error **errp)
1269 {
1270     IscsiLun *iscsilun = bs->opaque;
1271     struct iscsi_context *iscsi = NULL;
1272     struct iscsi_url *iscsi_url = NULL;
1273     struct scsi_task *task = NULL;
1274     struct scsi_inquiry_standard *inq = NULL;
1275     char *initiator_name = NULL;
1276     QemuOpts *opts;
1277     Error *local_err = NULL;
1278     const char *filename;
1279     int ret;
1280
1281     if ((BDRV_SECTOR_SIZE % 512) != 0) {
1282         error_report("iSCSI: Invalid BDRV_SECTOR_SIZE. "
1283                      "BDRV_SECTOR_SIZE(%lld) is not a multiple "
1284                      "of 512", BDRV_SECTOR_SIZE);
1285         return -EINVAL;
1286     }
1287
1288     opts = qemu_opts_create_nofail(&runtime_opts);
1289     qemu_opts_absorb_qdict(opts, options, &local_err);
1290     if (error_is_set(&local_err)) {
1291         qerror_report_err(local_err);
1292         error_free(local_err);
1293         ret = -EINVAL;
1294         goto out;
1295     }
1296
1297     filename = qemu_opt_get(opts, "filename");
1298
1299
1300     iscsi_url = iscsi_parse_full_url(iscsi, filename);
1301     if (iscsi_url == NULL) {
1302         error_report("Failed to parse URL : %s", filename);
1303         ret = -EINVAL;
1304         goto out;
1305     }
1306
1307     memset(iscsilun, 0, sizeof(IscsiLun));
1308
1309     initiator_name = parse_initiator_name(iscsi_url->target);
1310
1311     iscsi = iscsi_create_context(initiator_name);
1312     if (iscsi == NULL) {
1313         error_report("iSCSI: Failed to create iSCSI context.");
1314         ret = -ENOMEM;
1315         goto out;
1316     }
1317
1318     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1319         error_report("iSCSI: Failed to set target name.");
1320         ret = -EINVAL;
1321         goto out;
1322     }
1323
1324     if (iscsi_url->user != NULL) {
1325         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1326                                               iscsi_url->passwd);
1327         if (ret != 0) {
1328             error_report("Failed to set initiator username and password");
1329             ret = -EINVAL;
1330             goto out;
1331         }
1332     }
1333
1334     /* check if we got CHAP username/password via the options */
1335     if (parse_chap(iscsi, iscsi_url->target) != 0) {
1336         error_report("iSCSI: Failed to set CHAP user/password");
1337         ret = -EINVAL;
1338         goto out;
1339     }
1340
1341     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1342         error_report("iSCSI: Failed to set session type to normal.");
1343         ret = -EINVAL;
1344         goto out;
1345     }
1346
1347     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1348
1349     /* check if we got HEADER_DIGEST via the options */
1350     parse_header_digest(iscsi, iscsi_url->target);
1351
1352     if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1353         error_report("iSCSI: Failed to connect to LUN : %s",
1354             iscsi_get_error(iscsi));
1355         ret = -EINVAL;
1356         goto out;
1357     }
1358
1359     iscsilun->iscsi = iscsi;
1360     iscsilun->lun   = iscsi_url->lun;
1361
1362     task = iscsi_inquiry_sync(iscsi, iscsilun->lun, 0, 0, 36);
1363
1364     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1365         error_report("iSCSI: failed to send inquiry command.");
1366         ret = -EINVAL;
1367         goto out;
1368     }
1369
1370     inq = scsi_datain_unmarshall(task);
1371     if (inq == NULL) {
1372         error_report("iSCSI: Failed to unmarshall inquiry data.");
1373         ret = -EINVAL;
1374         goto out;
1375     }
1376
1377     iscsilun->type = inq->periperal_device_type;
1378
1379     if ((ret = iscsi_readcapacity_sync(iscsilun)) != 0) {
1380         goto out;
1381     }
1382     bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1383
1384     /* Medium changer or tape. We dont have any emulation for this so this must
1385      * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
1386      * to read from the device to guess the image format.
1387      */
1388     if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
1389         iscsilun->type == TYPE_TAPE) {
1390         bs->sg = 1;
1391     }
1392
1393     if (iscsilun->lbpme) {
1394         struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1395         task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1396                                 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING);
1397         if (task == NULL) {
1398             ret = -EINVAL;
1399             goto out;
1400         }
1401         inq_lbp = scsi_datain_unmarshall(task);
1402         if (inq_lbp == NULL) {
1403             error_report("iSCSI: failed to unmarshall inquiry datain blob");
1404             ret = -EINVAL;
1405             goto out;
1406         }
1407         memcpy(&iscsilun->lbp, inq_lbp,
1408                sizeof(struct scsi_inquiry_logical_block_provisioning));
1409         scsi_free_scsi_task(task);
1410         task = NULL;
1411     }
1412
1413     if (iscsilun->lbp.lbpu || iscsilun->lbp.lbpws) {
1414         struct scsi_inquiry_block_limits *inq_bl;
1415         task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1416                                 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS);
1417         if (task == NULL) {
1418             ret = -EINVAL;
1419             goto out;
1420         }
1421         inq_bl = scsi_datain_unmarshall(task);
1422         if (inq_bl == NULL) {
1423             error_report("iSCSI: failed to unmarshall inquiry datain blob");
1424             ret = -EINVAL;
1425             goto out;
1426         }
1427         memcpy(&iscsilun->bl, inq_bl,
1428                sizeof(struct scsi_inquiry_block_limits));
1429         scsi_free_scsi_task(task);
1430         task = NULL;
1431
1432         if (iscsilun->bl.max_unmap < 0xffffffff) {
1433             bs->bl.max_discard = sector_lun2qemu(iscsilun->bl.max_unmap,
1434                                                  iscsilun);
1435         }
1436         bs->bl.discard_alignment = sector_lun2qemu(iscsilun->bl.opt_unmap_gran,
1437                                                    iscsilun);
1438
1439         if (iscsilun->bl.max_ws_len < 0xffffffff) {
1440             bs->bl.max_write_zeroes = sector_lun2qemu(iscsilun->bl.max_ws_len,
1441                                                       iscsilun);
1442         }
1443         bs->bl.write_zeroes_alignment = sector_lun2qemu(iscsilun->bl.opt_unmap_gran,
1444                                                         iscsilun);
1445     }
1446
1447 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
1448     /* Set up a timer for sending out iSCSI NOPs */
1449     iscsilun->nop_timer = timer_new_ms(QEMU_CLOCK_REALTIME, iscsi_nop_timed_event, iscsilun);
1450     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1451 #endif
1452
1453 out:
1454     qemu_opts_del(opts);
1455     if (initiator_name != NULL) {
1456         g_free(initiator_name);
1457     }
1458     if (iscsi_url != NULL) {
1459         iscsi_destroy_url(iscsi_url);
1460     }
1461     if (task != NULL) {
1462         scsi_free_scsi_task(task);
1463     }
1464
1465     if (ret) {
1466         if (iscsi != NULL) {
1467             iscsi_destroy_context(iscsi);
1468         }
1469         memset(iscsilun, 0, sizeof(IscsiLun));
1470     }
1471     return ret;
1472 }
1473
1474 static void iscsi_close(BlockDriverState *bs)
1475 {
1476     IscsiLun *iscsilun = bs->opaque;
1477     struct iscsi_context *iscsi = iscsilun->iscsi;
1478
1479     if (iscsilun->nop_timer) {
1480         timer_del(iscsilun->nop_timer);
1481         timer_free(iscsilun->nop_timer);
1482     }
1483     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL);
1484     iscsi_destroy_context(iscsi);
1485     g_free(iscsilun->zeroblock);
1486     memset(iscsilun, 0, sizeof(IscsiLun));
1487 }
1488
1489 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1490 {
1491     IscsiLun *iscsilun = bs->opaque;
1492     int ret = 0;
1493
1494     if (iscsilun->type != TYPE_DISK) {
1495         return -ENOTSUP;
1496     }
1497
1498     if ((ret = iscsi_readcapacity_sync(iscsilun)) != 0) {
1499         return ret;
1500     }
1501
1502     if (offset > iscsi_getlength(bs)) {
1503         return -EINVAL;
1504     }
1505
1506     return 0;
1507 }
1508
1509 static int iscsi_create(const char *filename, QEMUOptionParameter *options,
1510                         Error **errp)
1511 {
1512     int ret = 0;
1513     int64_t total_size = 0;
1514     BlockDriverState *bs;
1515     IscsiLun *iscsilun = NULL;
1516     QDict *bs_options;
1517
1518     bs = bdrv_new("");
1519
1520     /* Read out options */
1521     while (options && options->name) {
1522         if (!strcmp(options->name, "size")) {
1523             total_size = options->value.n / BDRV_SECTOR_SIZE;
1524         }
1525         options++;
1526     }
1527
1528     bs->opaque = g_malloc0(sizeof(struct IscsiLun));
1529     iscsilun = bs->opaque;
1530
1531     bs_options = qdict_new();
1532     qdict_put(bs_options, "filename", qstring_from_str(filename));
1533     ret = iscsi_open(bs, bs_options, 0, NULL);
1534     QDECREF(bs_options);
1535
1536     if (ret != 0) {
1537         goto out;
1538     }
1539     if (iscsilun->nop_timer) {
1540         timer_del(iscsilun->nop_timer);
1541         timer_free(iscsilun->nop_timer);
1542     }
1543     if (iscsilun->type != TYPE_DISK) {
1544         ret = -ENODEV;
1545         goto out;
1546     }
1547     if (bs->total_sectors < total_size) {
1548         ret = -ENOSPC;
1549         goto out;
1550     }
1551
1552     ret = 0;
1553 out:
1554     if (iscsilun->iscsi != NULL) {
1555         iscsi_destroy_context(iscsilun->iscsi);
1556     }
1557     g_free(bs->opaque);
1558     bs->opaque = NULL;
1559     bdrv_unref(bs);
1560     return ret;
1561 }
1562
1563 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1564 {
1565     IscsiLun *iscsilun = bs->opaque;
1566     bdi->unallocated_blocks_are_zero = !!iscsilun->lbprz;
1567     bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
1568     return 0;
1569 }
1570
1571 static QEMUOptionParameter iscsi_create_options[] = {
1572     {
1573         .name = BLOCK_OPT_SIZE,
1574         .type = OPT_SIZE,
1575         .help = "Virtual disk size"
1576     },
1577     { NULL }
1578 };
1579
1580 static BlockDriver bdrv_iscsi = {
1581     .format_name     = "iscsi",
1582     .protocol_name   = "iscsi",
1583
1584     .instance_size   = sizeof(IscsiLun),
1585     .bdrv_needs_filename = true,
1586     .bdrv_file_open  = iscsi_open,
1587     .bdrv_close      = iscsi_close,
1588     .bdrv_create     = iscsi_create,
1589     .create_options  = iscsi_create_options,
1590
1591     .bdrv_getlength  = iscsi_getlength,
1592     .bdrv_get_info   = iscsi_get_info,
1593     .bdrv_truncate   = iscsi_truncate,
1594
1595 #if defined(LIBISCSI_FEATURE_IOVECTOR)
1596     .bdrv_co_get_block_status = iscsi_co_get_block_status,
1597 #endif
1598     .bdrv_co_discard      = iscsi_co_discard,
1599 #if defined(SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED)
1600     .bdrv_co_write_zeroes = iscsi_co_write_zeroes,
1601 #endif
1602
1603     .bdrv_aio_readv  = iscsi_aio_readv,
1604     .bdrv_aio_writev = iscsi_aio_writev,
1605     .bdrv_aio_flush  = iscsi_aio_flush,
1606
1607 #ifdef __linux__
1608     .bdrv_ioctl       = iscsi_ioctl,
1609     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1610 #endif
1611 };
1612
1613 static QemuOptsList qemu_iscsi_opts = {
1614     .name = "iscsi",
1615     .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1616     .desc = {
1617         {
1618             .name = "user",
1619             .type = QEMU_OPT_STRING,
1620             .help = "username for CHAP authentication to target",
1621         },{
1622             .name = "password",
1623             .type = QEMU_OPT_STRING,
1624             .help = "password for CHAP authentication to target",
1625         },{
1626             .name = "header-digest",
1627             .type = QEMU_OPT_STRING,
1628             .help = "HeaderDigest setting. "
1629                     "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1630         },{
1631             .name = "initiator-name",
1632             .type = QEMU_OPT_STRING,
1633             .help = "Initiator iqn name to use when connecting",
1634         },
1635         { /* end of list */ }
1636     },
1637 };
1638
1639 static void iscsi_block_init(void)
1640 {
1641     bdrv_register(&bdrv_iscsi);
1642     qemu_add_opts(&qemu_iscsi_opts);
1643 }
1644
1645 block_init(iscsi_block_init);