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     uint8_t has_write_same;
  59     struct scsi_inquiry_logical_block_provisioning lbp;
  60     struct scsi_inquiry_block_limits bl;
  61     unsigned char *zeroblock;
  62 } IscsiLun;
  63
  64 typedef struct IscsiTask {
  65     int status;
  66     int complete;
  67     int retries;
  68     int do_retry;
  69     struct scsi_task *task;
  70     Coroutine *co;
  71     QEMUBH *bh;
  72 } IscsiTask;
  73
  74 typedef struct IscsiAIOCB {
  75     BlockDriverAIOCB common;
  76     QEMUIOVector *qiov;
  77     QEMUBH *bh;
  78     IscsiLun *iscsilun;
  79     struct scsi_task *task;
  80     uint8_t *buf;
  81     int status;
  82     int canceled;
  83     int retries;
  84     int64_t sector_num;
  85     int nb_sectors;
  86 #ifdef __linux__
  87     sg_io_hdr_t *ioh;
  88 #endif
  89 } IscsiAIOCB;
  90
  91 #define NOP_INTERVAL 5000
  92 #define MAX_NOP_FAILURES 3
  93 #define ISCSI_CMD_RETRIES 5
  94
  95 static void
  96 iscsi_bh_cb(void *p)
  97 {
  98     IscsiAIOCB *acb = p;
  99
 100     qemu_bh_delete(acb->bh);
 101
 102     g_free(acb->buf);
 103     acb->buf = NULL;
 104
 105     if (acb->canceled == 0) {
 106         acb->common.cb(acb->common.opaque, acb->status);
 107     }
 108
 109     if (acb->task != NULL) {
 110         scsi_free_scsi_task(acb->task);
 111         acb->task = NULL;
 112     }
 113
 114     qemu_aio_release(acb);
 115 }
 116
 117 static void
 118 iscsi_schedule_bh(IscsiAIOCB *acb)
 119 {
 120     if (acb->bh) {
 121         return;
 122     }
 123     acb->bh = qemu_bh_new(iscsi_bh_cb, acb);
 124     qemu_bh_schedule(acb->bh);
 125 }
 126
 127 static void iscsi_co_generic_bh_cb(void *opaque)
 128 {
 129     struct IscsiTask *iTask = opaque;
 130     qemu_bh_delete(iTask->bh);
 131     qemu_coroutine_enter(iTask->co, NULL);
 132 }
 133
 134 static void
 135 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
 136                         void *command_data, void *opaque)
 137 {
 138     struct IscsiTask *iTask = opaque;
 139     struct scsi_task *task = command_data;
 140
 141     iTask->complete = 1;
 142     iTask->status = status;
 143     iTask->do_retry = 0;
 144     iTask->task = task;
 145
 146     if (iTask->retries-- > 0 && status == SCSI_STATUS_CHECK_CONDITION
 147         && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
 148         iTask->do_retry = 1;
 149         goto out;
 150     }
 151
 152     if (status != SCSI_STATUS_GOOD) {
 153         error_report("iSCSI: Failure. %s", iscsi_get_error(iscsi));
 154     }
 155
 156 out:
 157     if (iTask->co) {
 158         iTask->bh = qemu_bh_new(iscsi_co_generic_bh_cb, iTask);
 159         qemu_bh_schedule(iTask->bh);
 160     }
 161 }
 162
 163 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
 164 {
 165     *iTask = (struct IscsiTask) {
 166         .co         = qemu_coroutine_self(),
 167         .retries    = ISCSI_CMD_RETRIES,
 168     };
 169 }
 170
 171 static void
 172 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
 173                     void *private_data)
 174 {
 175     IscsiAIOCB *acb = private_data;
 176
 177     acb->status = -ECANCELED;
 178     iscsi_schedule_bh(acb);
 179 }
 180
 181 static void
 182 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
 183 {
 184     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
 185     IscsiLun *iscsilun = acb->iscsilun;
 186
 187     if (acb->status != -EINPROGRESS) {
 188         return;
 189     }
 190
 191     acb->canceled = 1;
 192
 193     /* send a task mgmt call to the target to cancel the task on the target */
 194     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
 195                                      iscsi_abort_task_cb, acb);
 196
 197     while (acb->status == -EINPROGRESS) {
 198         qemu_aio_wait();
 199     }
 200 }
 201
 202 static const AIOCBInfo iscsi_aiocb_info = {
 203     .aiocb_size         = sizeof(IscsiAIOCB),
 204     .cancel             = iscsi_aio_cancel,
 205 };
 206
 207
 208 static void iscsi_process_read(void *arg);
 209 static void iscsi_process_write(void *arg);
 210
 211 static void
 212 iscsi_set_events(IscsiLun *iscsilun)
 213 {
 214     struct iscsi_context *iscsi = iscsilun->iscsi;
 215     int ev;
 216
 217     /* We always register a read handler.  */
 218     ev = POLLIN;
 219     ev |= iscsi_which_events(iscsi);
 220     if (ev != iscsilun->events) {
 221         qemu_aio_set_fd_handler(iscsi_get_fd(iscsi),
 222                       iscsi_process_read,
 223                       (ev & POLLOUT) ? iscsi_process_write : NULL,
 224                       iscsilun);
 225
 226     }
 227
 228     iscsilun->events = ev;
 229 }
 230
 231 static void
 232 iscsi_process_read(void *arg)
 233 {
 234     IscsiLun *iscsilun = arg;
 235     struct iscsi_context *iscsi = iscsilun->iscsi;
 236
 237     iscsi_service(iscsi, POLLIN);
 238     iscsi_set_events(iscsilun);
 239 }
 240
 241 static void
 242 iscsi_process_write(void *arg)
 243 {
 244     IscsiLun *iscsilun = arg;
 245     struct iscsi_context *iscsi = iscsilun->iscsi;
 246
 247     iscsi_service(iscsi, POLLOUT);
 248     iscsi_set_events(iscsilun);
 249 }
 250
 251 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
 252 {
 253     return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
 254 }
 255
 256 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
 257 {
 258     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
 259 }
 260
 261 static bool is_request_lun_aligned(int64_t sector_num, int nb_sectors,
 262                                       IscsiLun *iscsilun)
 263 {
 264     if ((sector_num * BDRV_SECTOR_SIZE) % iscsilun->block_size ||
 265         (nb_sectors * BDRV_SECTOR_SIZE) % iscsilun->block_size) {
 266             error_report("iSCSI misaligned request: "
 267                          "iscsilun->block_size %u, sector_num %" PRIi64
 268                          ", nb_sectors %d",
 269                          iscsilun->block_size, sector_num, nb_sectors);
 270             return 0;
 271     }
 272     return 1;
 273 }
 274
 275 static int coroutine_fn iscsi_co_writev(BlockDriverState *bs,
 276                                         int64_t sector_num, int nb_sectors,
 277                                         QEMUIOVector *iov)
 278 {
 279     IscsiLun *iscsilun = bs->opaque;
 280     struct IscsiTask iTask;
 281     uint64_t lba;
 282     uint32_t num_sectors;
 283     uint8_t *data = NULL;
 284     uint8_t *buf = NULL;
 285
 286     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 287         return -EINVAL;
 288     }
 289
 290     lba = sector_qemu2lun(sector_num, iscsilun);
 291     num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
 292 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 293     /* if the iovec only contains one buffer we can pass it directly */
 294     if (iov->niov == 1) {
 295         data = iov->iov[0].iov_base;
 296     } else {
 297         size_t size = MIN(nb_sectors * BDRV_SECTOR_SIZE, iov->size);
 298         buf = g_malloc(size);
 299         qemu_iovec_to_buf(iov, 0, buf, size);
 300         data = buf;
 301     }
 302 #endif
 303     iscsi_co_init_iscsitask(iscsilun, &iTask);
 304 retry:
 305     iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
 306                                     data, num_sectors * iscsilun->block_size,
 307                                     iscsilun->block_size, 0, 0, 0, 0, 0,
 308                                     iscsi_co_generic_cb, &iTask);
 309     if (iTask.task == NULL) {
 310         g_free(buf);
 311         return -ENOMEM;
 312     }
 313 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 314     scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
 315                           iov->niov);
 316 #endif
 317     while (!iTask.complete) {
 318         iscsi_set_events(iscsilun);
 319         qemu_coroutine_yield();
 320     }
 321
 322     if (iTask.task != NULL) {
 323         scsi_free_scsi_task(iTask.task);
 324         iTask.task = NULL;
 325     }
 326
 327     if (iTask.do_retry) {
 328         goto retry;
 329     }
 330
 331     g_free(buf);
 332
 333     if (iTask.status != SCSI_STATUS_GOOD) {
 334         return -EIO;
 335     }
 336
 337     return 0;
 338 }
 339
 340 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
 341                                        int64_t sector_num, int nb_sectors,
 342                                        QEMUIOVector *iov)
 343 {
 344     IscsiLun *iscsilun = bs->opaque;
 345     struct IscsiTask iTask;
 346     uint64_t lba;
 347     uint32_t num_sectors;
 348 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 349     int i;
 350 #endif
 351
 352     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 353         return -EINVAL;
 354     }
 355
 356     lba = sector_qemu2lun(sector_num, iscsilun);
 357     num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
 358
 359     iscsi_co_init_iscsitask(iscsilun, &iTask);
 360 retry:
 361     switch (iscsilun->type) {
 362     case TYPE_DISK:
 363         iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
 364                                        num_sectors * iscsilun->block_size,
 365                                        iscsilun->block_size, 0, 0, 0, 0, 0,
 366                                        iscsi_co_generic_cb, &iTask);
 367         break;
 368     default:
 369         iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
 370                                        num_sectors * iscsilun->block_size,
 371                                        iscsilun->block_size,
 372 #if !defined(CONFIG_LIBISCSI_1_4) /* API change from 1.4.0 to 1.5.0 */
 373                                        0, 0, 0, 0, 0,
 374 #endif
 375                                        iscsi_co_generic_cb, &iTask);
 376         break;
 377     }
 378     if (iTask.task == NULL) {
 379         return -ENOMEM;
 380     }
 381 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 382     scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
 383 #else
 384     for (i = 0; i < iov->niov; i++) {
 385         scsi_task_add_data_in_buffer(iTask.task,
 386                                      iov->iov[i].iov_len,
 387                                      iov->iov[i].iov_base);
 388     }
 389 #endif
 390
 391     while (!iTask.complete) {
 392         iscsi_set_events(iscsilun);
 393         qemu_coroutine_yield();
 394     }
 395
 396     if (iTask.task != NULL) {
 397         scsi_free_scsi_task(iTask.task);
 398         iTask.task = NULL;
 399     }
 400
 401     if (iTask.do_retry) {
 402         goto retry;
 403     }
 404
 405     if (iTask.status != SCSI_STATUS_GOOD) {
 406         return -EIO;
 407     }
 408
 409     return 0;
 410 }
 411
 412 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
 413 {
 414     IscsiLun *iscsilun = bs->opaque;
 415     struct IscsiTask iTask;
 416
 417     iscsi_co_init_iscsitask(iscsilun, &iTask);
 418
 419 retry:
 420     if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
 421                                       0, iscsi_co_generic_cb, &iTask) == NULL) {
 422         return -ENOMEM;
 423     }
 424
 425     while (!iTask.complete) {
 426         iscsi_set_events(iscsilun);
 427         qemu_coroutine_yield();
 428     }
 429
 430     if (iTask.task != NULL) {
 431         scsi_free_scsi_task(iTask.task);
 432         iTask.task = NULL;
 433     }
 434
 435     if (iTask.do_retry) {
 436         goto retry;
 437     }
 438
 439     if (iTask.status != SCSI_STATUS_GOOD) {
 440         return -EIO;
 441     }
 442
 443     return 0;
 444 }
 445
 446 #ifdef __linux__
 447 static void
 448 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 449                      void *command_data, void *opaque)
 450 {
 451     IscsiAIOCB *acb = opaque;
 452
 453     g_free(acb->buf);
 454     acb->buf = NULL;
 455
 456     if (acb->canceled != 0) {
 457         return;
 458     }
 459
 460     acb->status = 0;
 461     if (status < 0) {
 462         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 463                      iscsi_get_error(iscsi));
 464         acb->status = -EIO;
 465     }
 466
 467     acb->ioh->driver_status = 0;
 468     acb->ioh->host_status   = 0;
 469     acb->ioh->resid         = 0;
 470
 471 #define SG_ERR_DRIVER_SENSE    0x08
 472
 473     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 474         int ss;
 475
 476         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 477
 478         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 479         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 480              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 481         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 482     }
 483
 484     iscsi_schedule_bh(acb);
 485 }
 486
 487 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 488         unsigned long int req, void *buf,
 489         BlockDriverCompletionFunc *cb, void *opaque)
 490 {
 491     IscsiLun *iscsilun = bs->opaque;
 492     struct iscsi_context *iscsi = iscsilun->iscsi;
 493     struct iscsi_data data;
 494     IscsiAIOCB *acb;
 495
 496     assert(req == SG_IO);
 497
 498     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 499
 500     acb->iscsilun = iscsilun;
 501     acb->canceled    = 0;
 502     acb->bh          = NULL;
 503     acb->status      = -EINPROGRESS;
 504     acb->buf         = NULL;
 505     acb->ioh         = buf;
 506
 507     acb->task = malloc(sizeof(struct scsi_task));
 508     if (acb->task == NULL) {
 509         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 510                      iscsi_get_error(iscsi));
 511         qemu_aio_release(acb);
 512         return NULL;
 513     }
 514     memset(acb->task, 0, sizeof(struct scsi_task));
 515
 516     switch (acb->ioh->dxfer_direction) {
 517     case SG_DXFER_TO_DEV:
 518         acb->task->xfer_dir = SCSI_XFER_WRITE;
 519         break;
 520     case SG_DXFER_FROM_DEV:
 521         acb->task->xfer_dir = SCSI_XFER_READ;
 522         break;
 523     default:
 524         acb->task->xfer_dir = SCSI_XFER_NONE;
 525         break;
 526     }
 527
 528     acb->task->cdb_size = acb->ioh->cmd_len;
 529     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 530     acb->task->expxferlen = acb->ioh->dxfer_len;
 531
 532     data.size = 0;
 533     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 534         if (acb->ioh->iovec_count == 0) {
 535             data.data = acb->ioh->dxferp;
 536             data.size = acb->ioh->dxfer_len;
 537         } else {
 538 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 539             scsi_task_set_iov_out(acb->task,
 540                                  (struct scsi_iovec *) acb->ioh->dxferp,
 541                                  acb->ioh->iovec_count);
 542 #else
 543             struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
 544
 545             acb->buf = g_malloc(acb->ioh->dxfer_len);
 546             data.data = acb->buf;
 547             data.size = iov_to_buf(iov, acb->ioh->iovec_count, 0,
 548                                    acb->buf, acb->ioh->dxfer_len);
 549 #endif
 550         }
 551     }
 552
 553     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 554                                  iscsi_aio_ioctl_cb,
 555                                  (data.size > 0) ? &data : NULL,
 556                                  acb) != 0) {
 557         scsi_free_scsi_task(acb->task);
 558         qemu_aio_release(acb);
 559         return NULL;
 560     }
 561
 562     /* tell libiscsi to read straight into the buffer we got from ioctl */
 563     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 564         if (acb->ioh->iovec_count == 0) {
 565             scsi_task_add_data_in_buffer(acb->task,
 566                                          acb->ioh->dxfer_len,
 567                                          acb->ioh->dxferp);
 568         } else {
 569 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 570             scsi_task_set_iov_in(acb->task,
 571                                  (struct scsi_iovec *) acb->ioh->dxferp,
 572                                  acb->ioh->iovec_count);
 573 #else
 574             int i;
 575             for (i = 0; i < acb->ioh->iovec_count; i++) {
 576                 struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
 577
 578                 scsi_task_add_data_in_buffer(acb->task,
 579                     iov[i].iov_len,
 580                     iov[i].iov_base);
 581             }
 582 #endif
 583         }
 584     }
 585
 586     iscsi_set_events(iscsilun);
 587
 588     return &acb->common;
 589 }
 590
 591
 592 static void ioctl_cb(void *opaque, int status)
 593 {
 594     int *p_status = opaque;
 595     *p_status = status;
 596 }
 597
 598 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 599 {
 600     IscsiLun *iscsilun = bs->opaque;
 601     int status;
 602
 603     switch (req) {
 604     case SG_GET_VERSION_NUM:
 605         *(int *)buf = 30000;
 606         break;
 607     case SG_GET_SCSI_ID:
 608         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 609         break;
 610     case SG_IO:
 611         status = -EINPROGRESS;
 612         iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status);
 613
 614         while (status == -EINPROGRESS) {
 615             qemu_aio_wait();
 616         }
 617
 618         return 0;
 619     default:
 620         return -1;
 621     }
 622     return 0;
 623 }
 624 #endif
 625
 626 static int64_t
 627 iscsi_getlength(BlockDriverState *bs)
 628 {
 629     IscsiLun *iscsilun = bs->opaque;
 630     int64_t len;
 631
 632     len  = iscsilun->num_blocks;
 633     len *= iscsilun->block_size;
 634
 635     return len;
 636 }
 637
 638 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 639
 640 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
 641                                                   int64_t sector_num,
 642                                                   int nb_sectors, int *pnum)
 643 {
 644     IscsiLun *iscsilun = bs->opaque;
 645     struct scsi_get_lba_status *lbas = NULL;
 646     struct scsi_lba_status_descriptor *lbasd = NULL;
 647     struct IscsiTask iTask;
 648     int64_t ret;
 649
 650     iscsi_co_init_iscsitask(iscsilun, &iTask);
 651
 652     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 653         ret = -EINVAL;
 654         goto out;
 655     }
 656
 657     /* default to all sectors allocated */
 658     ret = BDRV_BLOCK_DATA;
 659     ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
 660     *pnum = nb_sectors;
 661
 662     /* LUN does not support logical block provisioning */
 663     if (iscsilun->lbpme == 0) {
 664         goto out;
 665     }
 666
 667 retry:
 668     if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
 669                                   sector_qemu2lun(sector_num, iscsilun),
 670                                   8 + 16, iscsi_co_generic_cb,
 671                                   &iTask) == NULL) {
 672         ret = -ENOMEM;
 673         goto out;
 674     }
 675
 676     while (!iTask.complete) {
 677         iscsi_set_events(iscsilun);
 678         qemu_coroutine_yield();
 679     }
 680
 681     if (iTask.do_retry) {
 682         if (iTask.task != NULL) {
 683             scsi_free_scsi_task(iTask.task);
 684             iTask.task = NULL;
 685         }
 686         goto retry;
 687     }
 688
 689     if (iTask.status != SCSI_STATUS_GOOD) {
 690         /* in case the get_lba_status_callout fails (i.e.
 691          * because the device is busy or the cmd is not
 692          * supported) we pretend all blocks are allocated
 693          * for backwards compatibility */
 694         goto out;
 695     }
 696
 697     lbas = scsi_datain_unmarshall(iTask.task);
 698     if (lbas == NULL) {
 699         ret = -EIO;
 700         goto out;
 701     }
 702
 703     lbasd = &lbas->descriptors[0];
 704
 705     if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
 706         ret = -EIO;
 707         goto out;
 708     }
 709
 710     *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
 711     if (*pnum > nb_sectors) {
 712         *pnum = nb_sectors;
 713     }
 714
 715     if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
 716         lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
 717         ret &= ~BDRV_BLOCK_DATA;
 718         if (iscsilun->lbprz) {
 719             ret |= BDRV_BLOCK_ZERO;
 720         }
 721     }
 722
 723 out:
 724     if (iTask.task != NULL) {
 725         scsi_free_scsi_task(iTask.task);
 726     }
 727     return ret;
 728 }
 729
 730 #endif /* LIBISCSI_FEATURE_IOVECTOR */
 731
 732 static int
 733 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num,
 734                                    int nb_sectors)
 735 {
 736     IscsiLun *iscsilun = bs->opaque;
 737     struct IscsiTask iTask;
 738     struct unmap_list list;
 739
 740     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 741         return -EINVAL;
 742     }
 743
 744     if (!iscsilun->lbp.lbpu) {
 745         /* UNMAP is not supported by the target */
 746         return 0;
 747     }
 748
 749     list.lba = sector_qemu2lun(sector_num, iscsilun);
 750     list.num = sector_qemu2lun(nb_sectors, iscsilun);
 751
 752     iscsi_co_init_iscsitask(iscsilun, &iTask);
 753 retry:
 754     if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
 755                      iscsi_co_generic_cb, &iTask) == NULL) {
 756         return -ENOMEM;
 757     }
 758
 759     while (!iTask.complete) {
 760         iscsi_set_events(iscsilun);
 761         qemu_coroutine_yield();
 762     }
 763
 764     if (iTask.task != NULL) {
 765         scsi_free_scsi_task(iTask.task);
 766         iTask.task = NULL;
 767     }
 768
 769     if (iTask.do_retry) {
 770         goto retry;
 771     }
 772
 773     if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
 774         /* the target might fail with a check condition if it
 775            is not happy with the alignment of the UNMAP request
 776            we silently fail in this case */
 777         return 0;
 778     }
 779
 780     if (iTask.status != SCSI_STATUS_GOOD) {
 781         return -EIO;
 782     }
 783
 784     return 0;
 785 }
 786
 787 #if defined(SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED)
 788
 789 static int
 790 coroutine_fn iscsi_co_write_zeroes(BlockDriverState *bs, int64_t sector_num,
 791                                    int nb_sectors, BdrvRequestFlags flags)
 792 {
 793     IscsiLun *iscsilun = bs->opaque;
 794     struct IscsiTask iTask;
 795     uint64_t lba;
 796     uint32_t nb_blocks;
 797
 798     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 799         return -EINVAL;
 800     }
 801
 802     if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
 803         /* WRITE SAME without UNMAP is not supported by the target */
 804         return -ENOTSUP;
 805     }
 806
 807     if ((flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->lbp.lbpws) {
 808         /* WRITE SAME with UNMAP is not supported by the target */
 809         return -ENOTSUP;
 810     }
 811
 812     lba = sector_qemu2lun(sector_num, iscsilun);
 813     nb_blocks = sector_qemu2lun(nb_sectors, iscsilun);
 814
 815     if (iscsilun->zeroblock == NULL) {
 816         iscsilun->zeroblock = g_malloc0(iscsilun->block_size);
 817     }
 818
 819     iscsi_co_init_iscsitask(iscsilun, &iTask);
 820 retry:
 821     if (iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
 822                                iscsilun->zeroblock, iscsilun->block_size,
 823                                nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
 824                                0, 0, iscsi_co_generic_cb, &iTask) == NULL) {
 825         return -ENOMEM;
 826     }
 827
 828     while (!iTask.complete) {
 829         iscsi_set_events(iscsilun);
 830         qemu_coroutine_yield();
 831     }
 832
 833     if (iTask.task != NULL) {
 834         scsi_free_scsi_task(iTask.task);
 835         iTask.task = NULL;
 836     }
 837
 838     if (iTask.do_retry) {
 839         goto retry;
 840     }
 841
 842     if (iTask.status != SCSI_STATUS_GOOD) {
 843         if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
 844             iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
 845             iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE) {
 846             /* WRITE SAME is not supported by the target */
 847             iscsilun->has_write_same = false;
 848             return -ENOTSUP;
 849         }
 850
 851         return -EIO;
 852     }
 853
 854     return 0;
 855 }
 856
 857 #endif /* SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED */
 858
 859 static int parse_chap(struct iscsi_context *iscsi, const char *target)
 860 {
 861     QemuOptsList *list;
 862     QemuOpts *opts;
 863     const char *user = NULL;
 864     const char *password = NULL;
 865
 866     list = qemu_find_opts("iscsi");
 867     if (!list) {
 868         return 0;
 869     }
 870
 871     opts = qemu_opts_find(list, target);
 872     if (opts == NULL) {
 873         opts = QTAILQ_FIRST(&list->head);
 874         if (!opts) {
 875             return 0;
 876         }
 877     }
 878
 879     user = qemu_opt_get(opts, "user");
 880     if (!user) {
 881         return 0;
 882     }
 883
 884     password = qemu_opt_get(opts, "password");
 885     if (!password) {
 886         error_report("CHAP username specified but no password was given");
 887         return -1;
 888     }
 889
 890     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
 891         error_report("Failed to set initiator username and password");
 892         return -1;
 893     }
 894
 895     return 0;
 896 }
 897
 898 static void parse_header_digest(struct iscsi_context *iscsi, const char *target)
 899 {
 900     QemuOptsList *list;
 901     QemuOpts *opts;
 902     const char *digest = NULL;
 903
 904     list = qemu_find_opts("iscsi");
 905     if (!list) {
 906         return;
 907     }
 908
 909     opts = qemu_opts_find(list, target);
 910     if (opts == NULL) {
 911         opts = QTAILQ_FIRST(&list->head);
 912         if (!opts) {
 913             return;
 914         }
 915     }
 916
 917     digest = qemu_opt_get(opts, "header-digest");
 918     if (!digest) {
 919         return;
 920     }
 921
 922     if (!strcmp(digest, "CRC32C")) {
 923         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
 924     } else if (!strcmp(digest, "NONE")) {
 925         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
 926     } else if (!strcmp(digest, "CRC32C-NONE")) {
 927         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
 928     } else if (!strcmp(digest, "NONE-CRC32C")) {
 929         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 930     } else {
 931         error_report("Invalid header-digest setting : %s", digest);
 932     }
 933 }
 934
 935 static char *parse_initiator_name(const char *target)
 936 {
 937     QemuOptsList *list;
 938     QemuOpts *opts;
 939     const char *name;
 940     char *iscsi_name;
 941     UuidInfo *uuid_info;
 942
 943     list = qemu_find_opts("iscsi");
 944     if (list) {
 945         opts = qemu_opts_find(list, target);
 946         if (!opts) {
 947             opts = QTAILQ_FIRST(&list->head);
 948         }
 949         if (opts) {
 950             name = qemu_opt_get(opts, "initiator-name");
 951             if (name) {
 952                 return g_strdup(name);
 953             }
 954         }
 955     }
 956
 957     uuid_info = qmp_query_uuid(NULL);
 958     if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
 959         name = qemu_get_vm_name();
 960     } else {
 961         name = uuid_info->UUID;
 962     }
 963     iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
 964                                  name ? ":" : "", name ? name : "");
 965     qapi_free_UuidInfo(uuid_info);
 966     return iscsi_name;
 967 }
 968
 969 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
 970 static void iscsi_nop_timed_event(void *opaque)
 971 {
 972     IscsiLun *iscsilun = opaque;
 973
 974     if (iscsi_get_nops_in_flight(iscsilun->iscsi) > MAX_NOP_FAILURES) {
 975         error_report("iSCSI: NOP timeout. Reconnecting...");
 976         iscsi_reconnect(iscsilun->iscsi);
 977     }
 978
 979     if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
 980         error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
 981         return;
 982     }
 983
 984     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
 985     iscsi_set_events(iscsilun);
 986 }
 987 #endif
 988
 989 static int iscsi_readcapacity_sync(IscsiLun *iscsilun)
 990 {
 991     struct scsi_task *task = NULL;
 992     struct scsi_readcapacity10 *rc10 = NULL;
 993     struct scsi_readcapacity16 *rc16 = NULL;
 994     int ret = 0;
 995     int retries = ISCSI_CMD_RETRIES;
 996
 997     do {
 998         if (task != NULL) {
 999             scsi_free_scsi_task(task);
1000             task = NULL;
1001         }
1002
1003         switch (iscsilun->type) {
1004         case TYPE_DISK:
1005             task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1006             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1007                 rc16 = scsi_datain_unmarshall(task);
1008                 if (rc16 == NULL) {
1009                     error_report("iSCSI: Failed to unmarshall readcapacity16 data.");
1010                     ret = -EINVAL;
1011                 } else {
1012                     iscsilun->block_size = rc16->block_length;
1013                     iscsilun->num_blocks = rc16->returned_lba + 1;
1014                     iscsilun->lbpme = rc16->lbpme;
1015                     iscsilun->lbprz = rc16->lbprz;
1016                 }
1017             }
1018             break;
1019         case TYPE_ROM:
1020             task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1021             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1022                 rc10 = scsi_datain_unmarshall(task);
1023                 if (rc10 == NULL) {
1024                     error_report("iSCSI: Failed to unmarshall readcapacity10 data.");
1025                     ret = -EINVAL;
1026                 } else {
1027                     iscsilun->block_size = rc10->block_size;
1028                     if (rc10->lba == 0) {
1029                         /* blank disk loaded */
1030                         iscsilun->num_blocks = 0;
1031                     } else {
1032                         iscsilun->num_blocks = rc10->lba + 1;
1033                     }
1034                 }
1035             }
1036             break;
1037         default:
1038             return 0;
1039         }
1040     } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1041              && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1042              && retries-- > 0);
1043
1044     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1045         error_report("iSCSI: failed to send readcapacity10 command.");
1046         ret = -EINVAL;
1047     }
1048     if (task) {
1049         scsi_free_scsi_task(task);
1050     }
1051     return ret;
1052 }
1053
1054 /* TODO Convert to fine grained options */
1055 static QemuOptsList runtime_opts = {
1056     .name = "iscsi",
1057     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1058     .desc = {
1059         {
1060             .name = "filename",
1061             .type = QEMU_OPT_STRING,
1062             .help = "URL to the iscsi image",
1063         },
1064         { /* end of list */ }
1065     },
1066 };
1067
1068 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi,
1069                                           int lun, int evpd, int pc) {
1070         int full_size;
1071         struct scsi_task *task = NULL;
1072         task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1073         if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1074             goto fail;
1075         }
1076         full_size = scsi_datain_getfullsize(task);
1077         if (full_size > task->datain.size) {
1078             scsi_free_scsi_task(task);
1079
1080             /* we need more data for the full list */
1081             task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1082             if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1083                 goto fail;
1084             }
1085         }
1086
1087         return task;
1088
1089 fail:
1090         error_report("iSCSI: Inquiry command failed : %s",
1091                      iscsi_get_error(iscsi));
1092         if (task) {
1093             scsi_free_scsi_task(task);
1094             return NULL;
1095         }
1096         return NULL;
1097 }
1098
1099 /*
1100  * We support iscsi url's on the form
1101  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1102  *
1103  * Note: flags are currently not used by iscsi_open.  If this function
1104  * is changed such that flags are used, please examine iscsi_reopen_prepare()
1105  * to see if needs to be changed as well.
1106  */
1107 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1108                       Error **errp)
1109 {
1110     IscsiLun *iscsilun = bs->opaque;
1111     struct iscsi_context *iscsi = NULL;
1112     struct iscsi_url *iscsi_url = NULL;
1113     struct scsi_task *task = NULL;
1114     struct scsi_inquiry_standard *inq = NULL;
1115     char *initiator_name = NULL;
1116     QemuOpts *opts;
1117     Error *local_err = NULL;
1118     const char *filename;
1119     int ret;
1120
1121     if ((BDRV_SECTOR_SIZE % 512) != 0) {
1122         error_report("iSCSI: Invalid BDRV_SECTOR_SIZE. "
1123                      "BDRV_SECTOR_SIZE(%lld) is not a multiple "
1124                      "of 512", BDRV_SECTOR_SIZE);
1125         return -EINVAL;
1126     }
1127
1128     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1129     qemu_opts_absorb_qdict(opts, options, &local_err);
1130     if (error_is_set(&local_err)) {
1131         qerror_report_err(local_err);
1132         error_free(local_err);
1133         ret = -EINVAL;
1134         goto out;
1135     }
1136
1137     filename = qemu_opt_get(opts, "filename");
1138
1139
1140     iscsi_url = iscsi_parse_full_url(iscsi, filename);
1141     if (iscsi_url == NULL) {
1142         error_report("Failed to parse URL : %s", filename);
1143         ret = -EINVAL;
1144         goto out;
1145     }
1146
1147     memset(iscsilun, 0, sizeof(IscsiLun));
1148
1149     initiator_name = parse_initiator_name(iscsi_url->target);
1150
1151     iscsi = iscsi_create_context(initiator_name);
1152     if (iscsi == NULL) {
1153         error_report("iSCSI: Failed to create iSCSI context.");
1154         ret = -ENOMEM;
1155         goto out;
1156     }
1157
1158     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1159         error_report("iSCSI: Failed to set target name.");
1160         ret = -EINVAL;
1161         goto out;
1162     }
1163
1164     if (iscsi_url->user != NULL) {
1165         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1166                                               iscsi_url->passwd);
1167         if (ret != 0) {
1168             error_report("Failed to set initiator username and password");
1169             ret = -EINVAL;
1170             goto out;
1171         }
1172     }
1173
1174     /* check if we got CHAP username/password via the options */
1175     if (parse_chap(iscsi, iscsi_url->target) != 0) {
1176         error_report("iSCSI: Failed to set CHAP user/password");
1177         ret = -EINVAL;
1178         goto out;
1179     }
1180
1181     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1182         error_report("iSCSI: Failed to set session type to normal.");
1183         ret = -EINVAL;
1184         goto out;
1185     }
1186
1187     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1188
1189     /* check if we got HEADER_DIGEST via the options */
1190     parse_header_digest(iscsi, iscsi_url->target);
1191
1192     if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1193         error_report("iSCSI: Failed to connect to LUN : %s",
1194             iscsi_get_error(iscsi));
1195         ret = -EINVAL;
1196         goto out;
1197     }
1198
1199     iscsilun->iscsi = iscsi;
1200     iscsilun->lun   = iscsi_url->lun;
1201
1202     task = iscsi_inquiry_sync(iscsi, iscsilun->lun, 0, 0, 36);
1203
1204     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1205         error_report("iSCSI: failed to send inquiry command.");
1206         ret = -EINVAL;
1207         goto out;
1208     }
1209
1210     inq = scsi_datain_unmarshall(task);
1211     if (inq == NULL) {
1212         error_report("iSCSI: Failed to unmarshall inquiry data.");
1213         ret = -EINVAL;
1214         goto out;
1215     }
1216
1217     iscsilun->type = inq->periperal_device_type;
1218     iscsilun->has_write_same = true;
1219
1220     if ((ret = iscsi_readcapacity_sync(iscsilun)) != 0) {
1221         goto out;
1222     }
1223     bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1224     bs->request_alignment = iscsilun->block_size;
1225
1226     /* Medium changer or tape. We dont have any emulation for this so this must
1227      * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
1228      * to read from the device to guess the image format.
1229      */
1230     if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
1231         iscsilun->type == TYPE_TAPE) {
1232         bs->sg = 1;
1233     }
1234
1235     if (iscsilun->lbpme) {
1236         struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1237         task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1238                                 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING);
1239         if (task == NULL) {
1240             ret = -EINVAL;
1241             goto out;
1242         }
1243         inq_lbp = scsi_datain_unmarshall(task);
1244         if (inq_lbp == NULL) {
1245             error_report("iSCSI: failed to unmarshall inquiry datain blob");
1246             ret = -EINVAL;
1247             goto out;
1248         }
1249         memcpy(&iscsilun->lbp, inq_lbp,
1250                sizeof(struct scsi_inquiry_logical_block_provisioning));
1251         scsi_free_scsi_task(task);
1252         task = NULL;
1253     }
1254
1255     if (iscsilun->lbp.lbpu || iscsilun->lbp.lbpws) {
1256         struct scsi_inquiry_block_limits *inq_bl;
1257         task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1258                                 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS);
1259         if (task == NULL) {
1260             ret = -EINVAL;
1261             goto out;
1262         }
1263         inq_bl = scsi_datain_unmarshall(task);
1264         if (inq_bl == NULL) {
1265             error_report("iSCSI: failed to unmarshall inquiry datain blob");
1266             ret = -EINVAL;
1267             goto out;
1268         }
1269         memcpy(&iscsilun->bl, inq_bl,
1270                sizeof(struct scsi_inquiry_block_limits));
1271         scsi_free_scsi_task(task);
1272         task = NULL;
1273     }
1274
1275 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
1276     /* Set up a timer for sending out iSCSI NOPs */
1277     iscsilun->nop_timer = timer_new_ms(QEMU_CLOCK_REALTIME, iscsi_nop_timed_event, iscsilun);
1278     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1279 #endif
1280
1281 out:
1282     qemu_opts_del(opts);
1283     if (initiator_name != NULL) {
1284         g_free(initiator_name);
1285     }
1286     if (iscsi_url != NULL) {
1287         iscsi_destroy_url(iscsi_url);
1288     }
1289     if (task != NULL) {
1290         scsi_free_scsi_task(task);
1291     }
1292
1293     if (ret) {
1294         if (iscsi != NULL) {
1295             iscsi_destroy_context(iscsi);
1296         }
1297         memset(iscsilun, 0, sizeof(IscsiLun));
1298     }
1299     return ret;
1300 }
1301
1302 static void iscsi_close(BlockDriverState *bs)
1303 {
1304     IscsiLun *iscsilun = bs->opaque;
1305     struct iscsi_context *iscsi = iscsilun->iscsi;
1306
1307     if (iscsilun->nop_timer) {
1308         timer_del(iscsilun->nop_timer);
1309         timer_free(iscsilun->nop_timer);
1310     }
1311     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL);
1312     iscsi_destroy_context(iscsi);
1313     g_free(iscsilun->zeroblock);
1314     memset(iscsilun, 0, sizeof(IscsiLun));
1315 }
1316
1317 static int iscsi_refresh_limits(BlockDriverState *bs)
1318 {
1319     IscsiLun *iscsilun = bs->opaque;
1320
1321     /* We don't actually refresh here, but just return data queried in
1322      * iscsi_open(): iscsi targets don't change their limits. */
1323     if (iscsilun->lbp.lbpu || iscsilun->lbp.lbpws) {
1324         if (iscsilun->bl.max_unmap < 0xffffffff) {
1325             bs->bl.max_discard = sector_lun2qemu(iscsilun->bl.max_unmap,
1326                                                  iscsilun);
1327         }
1328         bs->bl.discard_alignment = sector_lun2qemu(iscsilun->bl.opt_unmap_gran,
1329                                                    iscsilun);
1330
1331         if (iscsilun->bl.max_ws_len < 0xffffffff) {
1332             bs->bl.max_write_zeroes = sector_lun2qemu(iscsilun->bl.max_ws_len,
1333                                                       iscsilun);
1334         }
1335         bs->bl.write_zeroes_alignment = sector_lun2qemu(iscsilun->bl.opt_unmap_gran,
1336                                                         iscsilun);
1337     }
1338     bs->bl.opt_transfer_length = sector_lun2qemu(iscsilun->bl.opt_xfer_len,
1339                                                  iscsilun);
1340     return 0;
1341 }
1342
1343 /* Since iscsi_open() ignores bdrv_flags, there is nothing to do here in
1344  * prepare.  Note that this will not re-establish a connection with an iSCSI
1345  * target - it is effectively a NOP.  */
1346 static int iscsi_reopen_prepare(BDRVReopenState *state,
1347                                 BlockReopenQueue *queue, Error **errp)
1348 {
1349     /* NOP */
1350     return 0;
1351 }
1352
1353 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1354 {
1355     IscsiLun *iscsilun = bs->opaque;
1356     int ret = 0;
1357
1358     if (iscsilun->type != TYPE_DISK) {
1359         return -ENOTSUP;
1360     }
1361
1362     if ((ret = iscsi_readcapacity_sync(iscsilun)) != 0) {
1363         return ret;
1364     }
1365
1366     if (offset > iscsi_getlength(bs)) {
1367         return -EINVAL;
1368     }
1369
1370     return 0;
1371 }
1372
1373 static int iscsi_create(const char *filename, QEMUOptionParameter *options,
1374                         Error **errp)
1375 {
1376     int ret = 0;
1377     int64_t total_size = 0;
1378     BlockDriverState *bs;
1379     IscsiLun *iscsilun = NULL;
1380     QDict *bs_options;
1381
1382     bs = bdrv_new("");
1383
1384     /* Read out options */
1385     while (options && options->name) {
1386         if (!strcmp(options->name, "size")) {
1387             total_size = options->value.n / BDRV_SECTOR_SIZE;
1388         }
1389         options++;
1390     }
1391
1392     bs->opaque = g_malloc0(sizeof(struct IscsiLun));
1393     iscsilun = bs->opaque;
1394
1395     bs_options = qdict_new();
1396     qdict_put(bs_options, "filename", qstring_from_str(filename));
1397     ret = iscsi_open(bs, bs_options, 0, NULL);
1398     QDECREF(bs_options);
1399
1400     if (ret != 0) {
1401         goto out;
1402     }
1403     if (iscsilun->nop_timer) {
1404         timer_del(iscsilun->nop_timer);
1405         timer_free(iscsilun->nop_timer);
1406     }
1407     if (iscsilun->type != TYPE_DISK) {
1408         ret = -ENODEV;
1409         goto out;
1410     }
1411     if (bs->total_sectors < total_size) {
1412         ret = -ENOSPC;
1413         goto out;
1414     }
1415
1416     ret = 0;
1417 out:
1418     if (iscsilun->iscsi != NULL) {
1419         iscsi_destroy_context(iscsilun->iscsi);
1420     }
1421     g_free(bs->opaque);
1422     bs->opaque = NULL;
1423     bdrv_unref(bs);
1424     return ret;
1425 }
1426
1427 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1428 {
1429     IscsiLun *iscsilun = bs->opaque;
1430     bdi->unallocated_blocks_are_zero = !!iscsilun->lbprz;
1431     bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
1432     /* Guess the internal cluster (page) size of the iscsi target by the means
1433      * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1434      * reasonable size for bdi->cluster_size */
1435     if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 64 * 1024 &&
1436         iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1437         bdi->cluster_size = iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
1438     }
1439     return 0;
1440 }
1441
1442 static QEMUOptionParameter iscsi_create_options[] = {
1443     {
1444         .name = BLOCK_OPT_SIZE,
1445         .type = OPT_SIZE,
1446         .help = "Virtual disk size"
1447     },
1448     { NULL }
1449 };
1450
1451 static BlockDriver bdrv_iscsi = {
1452     .format_name     = "iscsi",
1453     .protocol_name   = "iscsi",
1454
1455     .instance_size   = sizeof(IscsiLun),
1456     .bdrv_needs_filename = true,
1457     .bdrv_file_open  = iscsi_open,
1458     .bdrv_close      = iscsi_close,
1459     .bdrv_create     = iscsi_create,
1460     .create_options  = iscsi_create_options,
1461     .bdrv_reopen_prepare  = iscsi_reopen_prepare,
1462
1463     .bdrv_getlength  = iscsi_getlength,
1464     .bdrv_get_info   = iscsi_get_info,
1465     .bdrv_truncate   = iscsi_truncate,
1466     .bdrv_refresh_limits = iscsi_refresh_limits,
1467
1468 #if defined(LIBISCSI_FEATURE_IOVECTOR)
1469     .bdrv_co_get_block_status = iscsi_co_get_block_status,
1470 #endif
1471     .bdrv_co_discard      = iscsi_co_discard,
1472 #if defined(SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED)
1473     .bdrv_co_write_zeroes = iscsi_co_write_zeroes,
1474 #endif
1475     .bdrv_co_readv         = iscsi_co_readv,
1476     .bdrv_co_writev        = iscsi_co_writev,
1477     .bdrv_co_flush_to_disk = iscsi_co_flush,
1478
1479 #ifdef __linux__
1480     .bdrv_ioctl       = iscsi_ioctl,
1481     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1482 #endif
1483 };
1484
1485 static QemuOptsList qemu_iscsi_opts = {
1486     .name = "iscsi",
1487     .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1488     .desc = {
1489         {
1490             .name = "user",
1491             .type = QEMU_OPT_STRING,
1492             .help = "username for CHAP authentication to target",
1493         },{
1494             .name = "password",
1495             .type = QEMU_OPT_STRING,
1496             .help = "password for CHAP authentication to target",
1497         },{
1498             .name = "header-digest",
1499             .type = QEMU_OPT_STRING,
1500             .help = "HeaderDigest setting. "
1501                     "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1502         },{
1503             .name = "initiator-name",
1504             .type = QEMU_OPT_STRING,
1505             .help = "Initiator iqn name to use when connecting",
1506         },
1507         { /* end of list */ }
1508     },
1509 };
1510
1511 static void iscsi_block_init(void)
1512 {
1513     bdrv_register(&bdrv_iscsi);
1514     qemu_add_opts(&qemu_iscsi_opts);
1515 }
1516
1517 block_init(iscsi_block_init);