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         error_report("iSCSI CheckCondition: %s", iscsi_get_error(iscsi));
 149         iTask->do_retry = 1;
 150         goto out;
 151     }
 152
 153     if (status != SCSI_STATUS_GOOD) {
 154         error_report("iSCSI Failure: %s", iscsi_get_error(iscsi));
 155     }
 156
 157 out:
 158     if (iTask->co) {
 159         iTask->bh = qemu_bh_new(iscsi_co_generic_bh_cb, iTask);
 160         qemu_bh_schedule(iTask->bh);
 161     }
 162 }
 163
 164 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
 165 {
 166     *iTask = (struct IscsiTask) {
 167         .co         = qemu_coroutine_self(),
 168         .retries    = ISCSI_CMD_RETRIES,
 169     };
 170 }
 171
 172 static void
 173 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
 174                     void *private_data)
 175 {
 176     IscsiAIOCB *acb = private_data;
 177
 178     acb->status = -ECANCELED;
 179     iscsi_schedule_bh(acb);
 180 }
 181
 182 static void
 183 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
 184 {
 185     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
 186     IscsiLun *iscsilun = acb->iscsilun;
 187
 188     if (acb->status != -EINPROGRESS) {
 189         return;
 190     }
 191
 192     acb->canceled = 1;
 193
 194     /* send a task mgmt call to the target to cancel the task on the target */
 195     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
 196                                      iscsi_abort_task_cb, acb);
 197
 198     while (acb->status == -EINPROGRESS) {
 199         qemu_aio_wait();
 200     }
 201 }
 202
 203 static const AIOCBInfo iscsi_aiocb_info = {
 204     .aiocb_size         = sizeof(IscsiAIOCB),
 205     .cancel             = iscsi_aio_cancel,
 206 };
 207
 208
 209 static void iscsi_process_read(void *arg);
 210 static void iscsi_process_write(void *arg);
 211
 212 static void
 213 iscsi_set_events(IscsiLun *iscsilun)
 214 {
 215     struct iscsi_context *iscsi = iscsilun->iscsi;
 216     int ev;
 217
 218     /* We always register a read handler.  */
 219     ev = POLLIN;
 220     ev |= iscsi_which_events(iscsi);
 221     if (ev != iscsilun->events) {
 222         qemu_aio_set_fd_handler(iscsi_get_fd(iscsi),
 223                       iscsi_process_read,
 224                       (ev & POLLOUT) ? iscsi_process_write : NULL,
 225                       iscsilun);
 226
 227     }
 228
 229     iscsilun->events = ev;
 230 }
 231
 232 static void
 233 iscsi_process_read(void *arg)
 234 {
 235     IscsiLun *iscsilun = arg;
 236     struct iscsi_context *iscsi = iscsilun->iscsi;
 237
 238     iscsi_service(iscsi, POLLIN);
 239     iscsi_set_events(iscsilun);
 240 }
 241
 242 static void
 243 iscsi_process_write(void *arg)
 244 {
 245     IscsiLun *iscsilun = arg;
 246     struct iscsi_context *iscsi = iscsilun->iscsi;
 247
 248     iscsi_service(iscsi, POLLOUT);
 249     iscsi_set_events(iscsilun);
 250 }
 251
 252 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
 253 {
 254     return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
 255 }
 256
 257 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
 258 {
 259     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
 260 }
 261
 262 static bool is_request_lun_aligned(int64_t sector_num, int nb_sectors,
 263                                       IscsiLun *iscsilun)
 264 {
 265     if ((sector_num * BDRV_SECTOR_SIZE) % iscsilun->block_size ||
 266         (nb_sectors * BDRV_SECTOR_SIZE) % iscsilun->block_size) {
 267             error_report("iSCSI misaligned request: "
 268                          "iscsilun->block_size %u, sector_num %" PRIi64
 269                          ", nb_sectors %d",
 270                          iscsilun->block_size, sector_num, nb_sectors);
 271             return 0;
 272     }
 273     return 1;
 274 }
 275
 276 static int coroutine_fn iscsi_co_writev(BlockDriverState *bs,
 277                                         int64_t sector_num, int nb_sectors,
 278                                         QEMUIOVector *iov)
 279 {
 280     IscsiLun *iscsilun = bs->opaque;
 281     struct IscsiTask iTask;
 282     uint64_t lba;
 283     uint32_t num_sectors;
 284     uint8_t *data = NULL;
 285     uint8_t *buf = NULL;
 286
 287     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 288         return -EINVAL;
 289     }
 290
 291     lba = sector_qemu2lun(sector_num, iscsilun);
 292     num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
 293 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 294     /* if the iovec only contains one buffer we can pass it directly */
 295     if (iov->niov == 1) {
 296         data = iov->iov[0].iov_base;
 297     } else {
 298         size_t size = MIN(nb_sectors * BDRV_SECTOR_SIZE, iov->size);
 299         buf = g_malloc(size);
 300         qemu_iovec_to_buf(iov, 0, buf, size);
 301         data = buf;
 302     }
 303 #endif
 304     iscsi_co_init_iscsitask(iscsilun, &iTask);
 305 retry:
 306     iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
 307                                     data, num_sectors * iscsilun->block_size,
 308                                     iscsilun->block_size, 0, 0, 0, 0, 0,
 309                                     iscsi_co_generic_cb, &iTask);
 310     if (iTask.task == NULL) {
 311         g_free(buf);
 312         return -ENOMEM;
 313     }
 314 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 315     scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
 316                           iov->niov);
 317 #endif
 318     while (!iTask.complete) {
 319         iscsi_set_events(iscsilun);
 320         qemu_coroutine_yield();
 321     }
 322
 323     if (iTask.task != NULL) {
 324         scsi_free_scsi_task(iTask.task);
 325         iTask.task = NULL;
 326     }
 327
 328     if (iTask.do_retry) {
 329         iTask.complete = 0;
 330         goto retry;
 331     }
 332
 333     g_free(buf);
 334
 335     if (iTask.status != SCSI_STATUS_GOOD) {
 336         return -EIO;
 337     }
 338
 339     return 0;
 340 }
 341
 342 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
 343                                        int64_t sector_num, int nb_sectors,
 344                                        QEMUIOVector *iov)
 345 {
 346     IscsiLun *iscsilun = bs->opaque;
 347     struct IscsiTask iTask;
 348     uint64_t lba;
 349     uint32_t num_sectors;
 350 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 351     int i;
 352 #endif
 353
 354     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 355         return -EINVAL;
 356     }
 357
 358     lba = sector_qemu2lun(sector_num, iscsilun);
 359     num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
 360
 361     iscsi_co_init_iscsitask(iscsilun, &iTask);
 362 retry:
 363     switch (iscsilun->type) {
 364     case TYPE_DISK:
 365         iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
 366                                        num_sectors * iscsilun->block_size,
 367                                        iscsilun->block_size, 0, 0, 0, 0, 0,
 368                                        iscsi_co_generic_cb, &iTask);
 369         break;
 370     default:
 371         iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
 372                                        num_sectors * iscsilun->block_size,
 373                                        iscsilun->block_size,
 374 #if !defined(CONFIG_LIBISCSI_1_4) /* API change from 1.4.0 to 1.5.0 */
 375                                        0, 0, 0, 0, 0,
 376 #endif
 377                                        iscsi_co_generic_cb, &iTask);
 378         break;
 379     }
 380     if (iTask.task == NULL) {
 381         return -ENOMEM;
 382     }
 383 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 384     scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
 385 #else
 386     for (i = 0; i < iov->niov; i++) {
 387         scsi_task_add_data_in_buffer(iTask.task,
 388                                      iov->iov[i].iov_len,
 389                                      iov->iov[i].iov_base);
 390     }
 391 #endif
 392
 393     while (!iTask.complete) {
 394         iscsi_set_events(iscsilun);
 395         qemu_coroutine_yield();
 396     }
 397
 398     if (iTask.task != NULL) {
 399         scsi_free_scsi_task(iTask.task);
 400         iTask.task = NULL;
 401     }
 402
 403     if (iTask.do_retry) {
 404         iTask.complete = 0;
 405         goto retry;
 406     }
 407
 408     if (iTask.status != SCSI_STATUS_GOOD) {
 409         return -EIO;
 410     }
 411
 412     return 0;
 413 }
 414
 415 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
 416 {
 417     IscsiLun *iscsilun = bs->opaque;
 418     struct IscsiTask iTask;
 419
 420     if (bs->sg) {
 421         return 0;
 422     }
 423
 424     iscsi_co_init_iscsitask(iscsilun, &iTask);
 425
 426 retry:
 427     if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
 428                                       0, iscsi_co_generic_cb, &iTask) == NULL) {
 429         return -ENOMEM;
 430     }
 431
 432     while (!iTask.complete) {
 433         iscsi_set_events(iscsilun);
 434         qemu_coroutine_yield();
 435     }
 436
 437     if (iTask.task != NULL) {
 438         scsi_free_scsi_task(iTask.task);
 439         iTask.task = NULL;
 440     }
 441
 442     if (iTask.do_retry) {
 443         iTask.complete = 0;
 444         goto retry;
 445     }
 446
 447     if (iTask.status != SCSI_STATUS_GOOD) {
 448         return -EIO;
 449     }
 450
 451     return 0;
 452 }
 453
 454 #ifdef __linux__
 455 static void
 456 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 457                      void *command_data, void *opaque)
 458 {
 459     IscsiAIOCB *acb = opaque;
 460
 461     g_free(acb->buf);
 462     acb->buf = NULL;
 463
 464     if (acb->canceled != 0) {
 465         return;
 466     }
 467
 468     acb->status = 0;
 469     if (status < 0) {
 470         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 471                      iscsi_get_error(iscsi));
 472         acb->status = -EIO;
 473     }
 474
 475     acb->ioh->driver_status = 0;
 476     acb->ioh->host_status   = 0;
 477     acb->ioh->resid         = 0;
 478
 479 #define SG_ERR_DRIVER_SENSE    0x08
 480
 481     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 482         int ss;
 483
 484         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 485
 486         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 487         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 488              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 489         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 490     }
 491
 492     iscsi_schedule_bh(acb);
 493 }
 494
 495 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 496         unsigned long int req, void *buf,
 497         BlockDriverCompletionFunc *cb, void *opaque)
 498 {
 499     IscsiLun *iscsilun = bs->opaque;
 500     struct iscsi_context *iscsi = iscsilun->iscsi;
 501     struct iscsi_data data;
 502     IscsiAIOCB *acb;
 503
 504     assert(req == SG_IO);
 505
 506     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 507
 508     acb->iscsilun = iscsilun;
 509     acb->canceled    = 0;
 510     acb->bh          = NULL;
 511     acb->status      = -EINPROGRESS;
 512     acb->buf         = NULL;
 513     acb->ioh         = buf;
 514
 515     acb->task = malloc(sizeof(struct scsi_task));
 516     if (acb->task == NULL) {
 517         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 518                      iscsi_get_error(iscsi));
 519         qemu_aio_release(acb);
 520         return NULL;
 521     }
 522     memset(acb->task, 0, sizeof(struct scsi_task));
 523
 524     switch (acb->ioh->dxfer_direction) {
 525     case SG_DXFER_TO_DEV:
 526         acb->task->xfer_dir = SCSI_XFER_WRITE;
 527         break;
 528     case SG_DXFER_FROM_DEV:
 529         acb->task->xfer_dir = SCSI_XFER_READ;
 530         break;
 531     default:
 532         acb->task->xfer_dir = SCSI_XFER_NONE;
 533         break;
 534     }
 535
 536     acb->task->cdb_size = acb->ioh->cmd_len;
 537     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 538     acb->task->expxferlen = acb->ioh->dxfer_len;
 539
 540     data.size = 0;
 541     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 542         if (acb->ioh->iovec_count == 0) {
 543             data.data = acb->ioh->dxferp;
 544             data.size = acb->ioh->dxfer_len;
 545         } else {
 546 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 547             scsi_task_set_iov_out(acb->task,
 548                                  (struct scsi_iovec *) acb->ioh->dxferp,
 549                                  acb->ioh->iovec_count);
 550 #else
 551             struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
 552
 553             acb->buf = g_malloc(acb->ioh->dxfer_len);
 554             data.data = acb->buf;
 555             data.size = iov_to_buf(iov, acb->ioh->iovec_count, 0,
 556                                    acb->buf, acb->ioh->dxfer_len);
 557 #endif
 558         }
 559     }
 560
 561     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 562                                  iscsi_aio_ioctl_cb,
 563                                  (data.size > 0) ? &data : NULL,
 564                                  acb) != 0) {
 565         scsi_free_scsi_task(acb->task);
 566         qemu_aio_release(acb);
 567         return NULL;
 568     }
 569
 570     /* tell libiscsi to read straight into the buffer we got from ioctl */
 571     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 572         if (acb->ioh->iovec_count == 0) {
 573             scsi_task_add_data_in_buffer(acb->task,
 574                                          acb->ioh->dxfer_len,
 575                                          acb->ioh->dxferp);
 576         } else {
 577 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 578             scsi_task_set_iov_in(acb->task,
 579                                  (struct scsi_iovec *) acb->ioh->dxferp,
 580                                  acb->ioh->iovec_count);
 581 #else
 582             int i;
 583             for (i = 0; i < acb->ioh->iovec_count; i++) {
 584                 struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
 585
 586                 scsi_task_add_data_in_buffer(acb->task,
 587                     iov[i].iov_len,
 588                     iov[i].iov_base);
 589             }
 590 #endif
 591         }
 592     }
 593
 594     iscsi_set_events(iscsilun);
 595
 596     return &acb->common;
 597 }
 598
 599
 600 static void ioctl_cb(void *opaque, int status)
 601 {
 602     int *p_status = opaque;
 603     *p_status = status;
 604 }
 605
 606 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 607 {
 608     IscsiLun *iscsilun = bs->opaque;
 609     int status;
 610
 611     switch (req) {
 612     case SG_GET_VERSION_NUM:
 613         *(int *)buf = 30000;
 614         break;
 615     case SG_GET_SCSI_ID:
 616         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 617         break;
 618     case SG_IO:
 619         status = -EINPROGRESS;
 620         iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status);
 621
 622         while (status == -EINPROGRESS) {
 623             qemu_aio_wait();
 624         }
 625
 626         return 0;
 627     default:
 628         return -1;
 629     }
 630     return 0;
 631 }
 632 #endif
 633
 634 static int64_t
 635 iscsi_getlength(BlockDriverState *bs)
 636 {
 637     IscsiLun *iscsilun = bs->opaque;
 638     int64_t len;
 639
 640     len  = iscsilun->num_blocks;
 641     len *= iscsilun->block_size;
 642
 643     return len;
 644 }
 645
 646 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 647
 648 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
 649                                                   int64_t sector_num,
 650                                                   int nb_sectors, int *pnum)
 651 {
 652     IscsiLun *iscsilun = bs->opaque;
 653     struct scsi_get_lba_status *lbas = NULL;
 654     struct scsi_lba_status_descriptor *lbasd = NULL;
 655     struct IscsiTask iTask;
 656     int64_t ret;
 657
 658     iscsi_co_init_iscsitask(iscsilun, &iTask);
 659
 660     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 661         ret = -EINVAL;
 662         goto out;
 663     }
 664
 665     /* default to all sectors allocated */
 666     ret = BDRV_BLOCK_DATA;
 667     ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
 668     *pnum = nb_sectors;
 669
 670     /* LUN does not support logical block provisioning */
 671     if (iscsilun->lbpme == 0) {
 672         goto out;
 673     }
 674
 675 retry:
 676     if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
 677                                   sector_qemu2lun(sector_num, iscsilun),
 678                                   8 + 16, iscsi_co_generic_cb,
 679                                   &iTask) == NULL) {
 680         ret = -ENOMEM;
 681         goto out;
 682     }
 683
 684     while (!iTask.complete) {
 685         iscsi_set_events(iscsilun);
 686         qemu_coroutine_yield();
 687     }
 688
 689     if (iTask.do_retry) {
 690         if (iTask.task != NULL) {
 691             scsi_free_scsi_task(iTask.task);
 692             iTask.task = NULL;
 693         }
 694         iTask.complete = 0;
 695         goto retry;
 696     }
 697
 698     if (iTask.status != SCSI_STATUS_GOOD) {
 699         /* in case the get_lba_status_callout fails (i.e.
 700          * because the device is busy or the cmd is not
 701          * supported) we pretend all blocks are allocated
 702          * for backwards compatibility */
 703         goto out;
 704     }
 705
 706     lbas = scsi_datain_unmarshall(iTask.task);
 707     if (lbas == NULL) {
 708         ret = -EIO;
 709         goto out;
 710     }
 711
 712     lbasd = &lbas->descriptors[0];
 713
 714     if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
 715         ret = -EIO;
 716         goto out;
 717     }
 718
 719     *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
 720     if (*pnum > nb_sectors) {
 721         *pnum = nb_sectors;
 722     }
 723
 724     if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
 725         lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
 726         ret &= ~BDRV_BLOCK_DATA;
 727         if (iscsilun->lbprz) {
 728             ret |= BDRV_BLOCK_ZERO;
 729         }
 730     }
 731
 732 out:
 733     if (iTask.task != NULL) {
 734         scsi_free_scsi_task(iTask.task);
 735     }
 736     return ret;
 737 }
 738
 739 #endif /* LIBISCSI_FEATURE_IOVECTOR */
 740
 741 static int
 742 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num,
 743                                    int nb_sectors)
 744 {
 745     IscsiLun *iscsilun = bs->opaque;
 746     struct IscsiTask iTask;
 747     struct unmap_list list;
 748
 749     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 750         return -EINVAL;
 751     }
 752
 753     if (!iscsilun->lbp.lbpu) {
 754         /* UNMAP is not supported by the target */
 755         return 0;
 756     }
 757
 758     list.lba = sector_qemu2lun(sector_num, iscsilun);
 759     list.num = sector_qemu2lun(nb_sectors, iscsilun);
 760
 761     iscsi_co_init_iscsitask(iscsilun, &iTask);
 762 retry:
 763     if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
 764                      iscsi_co_generic_cb, &iTask) == NULL) {
 765         return -ENOMEM;
 766     }
 767
 768     while (!iTask.complete) {
 769         iscsi_set_events(iscsilun);
 770         qemu_coroutine_yield();
 771     }
 772
 773     if (iTask.task != NULL) {
 774         scsi_free_scsi_task(iTask.task);
 775         iTask.task = NULL;
 776     }
 777
 778     if (iTask.do_retry) {
 779         iTask.complete = 0;
 780         goto retry;
 781     }
 782
 783     if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
 784         /* the target might fail with a check condition if it
 785            is not happy with the alignment of the UNMAP request
 786            we silently fail in this case */
 787         return 0;
 788     }
 789
 790     if (iTask.status != SCSI_STATUS_GOOD) {
 791         return -EIO;
 792     }
 793
 794     return 0;
 795 }
 796
 797 #if defined(SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED)
 798
 799 static int
 800 coroutine_fn iscsi_co_write_zeroes(BlockDriverState *bs, int64_t sector_num,
 801                                    int nb_sectors, BdrvRequestFlags flags)
 802 {
 803     IscsiLun *iscsilun = bs->opaque;
 804     struct IscsiTask iTask;
 805     uint64_t lba;
 806     uint32_t nb_blocks;
 807
 808     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 809         return -EINVAL;
 810     }
 811
 812     if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
 813         /* WRITE SAME without UNMAP is not supported by the target */
 814         return -ENOTSUP;
 815     }
 816
 817     if ((flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->lbp.lbpws) {
 818         /* WRITE SAME with UNMAP is not supported by the target */
 819         return -ENOTSUP;
 820     }
 821
 822     lba = sector_qemu2lun(sector_num, iscsilun);
 823     nb_blocks = sector_qemu2lun(nb_sectors, iscsilun);
 824
 825     if (iscsilun->zeroblock == NULL) {
 826         iscsilun->zeroblock = g_malloc0(iscsilun->block_size);
 827     }
 828
 829     iscsi_co_init_iscsitask(iscsilun, &iTask);
 830 retry:
 831     if (iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
 832                                iscsilun->zeroblock, iscsilun->block_size,
 833                                nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
 834                                0, 0, iscsi_co_generic_cb, &iTask) == NULL) {
 835         return -ENOMEM;
 836     }
 837
 838     while (!iTask.complete) {
 839         iscsi_set_events(iscsilun);
 840         qemu_coroutine_yield();
 841     }
 842
 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          iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
 847         /* WRITE SAME is not supported by the target */
 848         iscsilun->has_write_same = false;
 849         scsi_free_scsi_task(iTask.task);
 850         return -ENOTSUP;
 851     }
 852
 853     if (iTask.task != NULL) {
 854         scsi_free_scsi_task(iTask.task);
 855         iTask.task = NULL;
 856     }
 857
 858     if (iTask.do_retry) {
 859         iTask.complete = 0;
 860         goto retry;
 861     }
 862
 863     if (iTask.status != SCSI_STATUS_GOOD) {
 864         return -EIO;
 865     }
 866
 867     return 0;
 868 }
 869
 870 #endif /* SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED */
 871
 872 static void parse_chap(struct iscsi_context *iscsi, const char *target,
 873                        Error **errp)
 874 {
 875     QemuOptsList *list;
 876     QemuOpts *opts;
 877     const char *user = NULL;
 878     const char *password = NULL;
 879
 880     list = qemu_find_opts("iscsi");
 881     if (!list) {
 882         return;
 883     }
 884
 885     opts = qemu_opts_find(list, target);
 886     if (opts == NULL) {
 887         opts = QTAILQ_FIRST(&list->head);
 888         if (!opts) {
 889             return;
 890         }
 891     }
 892
 893     user = qemu_opt_get(opts, "user");
 894     if (!user) {
 895         return;
 896     }
 897
 898     password = qemu_opt_get(opts, "password");
 899     if (!password) {
 900         error_setg(errp, "CHAP username specified but no password was given");
 901         return;
 902     }
 903
 904     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
 905         error_setg(errp, "Failed to set initiator username and password");
 906     }
 907 }
 908
 909 static void parse_header_digest(struct iscsi_context *iscsi, const char *target,
 910                                 Error **errp)
 911 {
 912     QemuOptsList *list;
 913     QemuOpts *opts;
 914     const char *digest = NULL;
 915
 916     list = qemu_find_opts("iscsi");
 917     if (!list) {
 918         return;
 919     }
 920
 921     opts = qemu_opts_find(list, target);
 922     if (opts == NULL) {
 923         opts = QTAILQ_FIRST(&list->head);
 924         if (!opts) {
 925             return;
 926         }
 927     }
 928
 929     digest = qemu_opt_get(opts, "header-digest");
 930     if (!digest) {
 931         return;
 932     }
 933
 934     if (!strcmp(digest, "CRC32C")) {
 935         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
 936     } else if (!strcmp(digest, "NONE")) {
 937         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
 938     } else if (!strcmp(digest, "CRC32C-NONE")) {
 939         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
 940     } else if (!strcmp(digest, "NONE-CRC32C")) {
 941         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 942     } else {
 943         error_setg(errp, "Invalid header-digest setting : %s", digest);
 944     }
 945 }
 946
 947 static char *parse_initiator_name(const char *target)
 948 {
 949     QemuOptsList *list;
 950     QemuOpts *opts;
 951     const char *name;
 952     char *iscsi_name;
 953     UuidInfo *uuid_info;
 954
 955     list = qemu_find_opts("iscsi");
 956     if (list) {
 957         opts = qemu_opts_find(list, target);
 958         if (!opts) {
 959             opts = QTAILQ_FIRST(&list->head);
 960         }
 961         if (opts) {
 962             name = qemu_opt_get(opts, "initiator-name");
 963             if (name) {
 964                 return g_strdup(name);
 965             }
 966         }
 967     }
 968
 969     uuid_info = qmp_query_uuid(NULL);
 970     if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
 971         name = qemu_get_vm_name();
 972     } else {
 973         name = uuid_info->UUID;
 974     }
 975     iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
 976                                  name ? ":" : "", name ? name : "");
 977     qapi_free_UuidInfo(uuid_info);
 978     return iscsi_name;
 979 }
 980
 981 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
 982 static void iscsi_nop_timed_event(void *opaque)
 983 {
 984     IscsiLun *iscsilun = opaque;
 985
 986     if (iscsi_get_nops_in_flight(iscsilun->iscsi) > MAX_NOP_FAILURES) {
 987         error_report("iSCSI: NOP timeout. Reconnecting...");
 988         iscsi_reconnect(iscsilun->iscsi);
 989     }
 990
 991     if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
 992         error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
 993         return;
 994     }
 995
 996     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
 997     iscsi_set_events(iscsilun);
 998 }
 999 #endif
1000
1001 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1002 {
1003     struct scsi_task *task = NULL;
1004     struct scsi_readcapacity10 *rc10 = NULL;
1005     struct scsi_readcapacity16 *rc16 = NULL;
1006     int retries = ISCSI_CMD_RETRIES;
1007
1008     do {
1009         if (task != NULL) {
1010             scsi_free_scsi_task(task);
1011             task = NULL;
1012         }
1013
1014         switch (iscsilun->type) {
1015         case TYPE_DISK:
1016             task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1017             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1018                 rc16 = scsi_datain_unmarshall(task);
1019                 if (rc16 == NULL) {
1020                     error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1021                 } else {
1022                     iscsilun->block_size = rc16->block_length;
1023                     iscsilun->num_blocks = rc16->returned_lba + 1;
1024                     iscsilun->lbpme = rc16->lbpme;
1025                     iscsilun->lbprz = rc16->lbprz;
1026                 }
1027             }
1028             break;
1029         case TYPE_ROM:
1030             task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1031             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1032                 rc10 = scsi_datain_unmarshall(task);
1033                 if (rc10 == NULL) {
1034                     error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1035                 } else {
1036                     iscsilun->block_size = rc10->block_size;
1037                     if (rc10->lba == 0) {
1038                         /* blank disk loaded */
1039                         iscsilun->num_blocks = 0;
1040                     } else {
1041                         iscsilun->num_blocks = rc10->lba + 1;
1042                     }
1043                 }
1044             }
1045             break;
1046         default:
1047             return;
1048         }
1049     } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1050              && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1051              && retries-- > 0);
1052
1053     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1054         error_setg(errp, "iSCSI: failed to send readcapacity10 command.");
1055     }
1056     if (task) {
1057         scsi_free_scsi_task(task);
1058     }
1059 }
1060
1061 /* TODO Convert to fine grained options */
1062 static QemuOptsList runtime_opts = {
1063     .name = "iscsi",
1064     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1065     .desc = {
1066         {
1067             .name = "filename",
1068             .type = QEMU_OPT_STRING,
1069             .help = "URL to the iscsi image",
1070         },
1071         { /* end of list */ }
1072     },
1073 };
1074
1075 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1076                                           int evpd, int pc, void **inq, Error **errp)
1077 {
1078     int full_size;
1079     struct scsi_task *task = NULL;
1080     task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1081     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1082         goto fail;
1083     }
1084     full_size = scsi_datain_getfullsize(task);
1085     if (full_size > task->datain.size) {
1086         scsi_free_scsi_task(task);
1087
1088         /* we need more data for the full list */
1089         task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1090         if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1091             goto fail;
1092         }
1093     }
1094
1095     *inq = scsi_datain_unmarshall(task);
1096     if (*inq == NULL) {
1097         error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1098         goto fail;
1099     }
1100
1101     return task;
1102
1103 fail:
1104     if (!error_is_set(errp)) {
1105         error_setg(errp, "iSCSI: Inquiry command failed : %s",
1106                    iscsi_get_error(iscsi));
1107     }
1108     if (task != NULL) {
1109         scsi_free_scsi_task(task);
1110     }
1111     return NULL;
1112 }
1113
1114 /*
1115  * We support iscsi url's on the form
1116  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1117  *
1118  * Note: flags are currently not used by iscsi_open.  If this function
1119  * is changed such that flags are used, please examine iscsi_reopen_prepare()
1120  * to see if needs to be changed as well.
1121  */
1122 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1123                       Error **errp)
1124 {
1125     IscsiLun *iscsilun = bs->opaque;
1126     struct iscsi_context *iscsi = NULL;
1127     struct iscsi_url *iscsi_url = NULL;
1128     struct scsi_task *task = NULL;
1129     struct scsi_inquiry_standard *inq = NULL;
1130     struct scsi_inquiry_supported_pages *inq_vpd;
1131     char *initiator_name = NULL;
1132     QemuOpts *opts;
1133     Error *local_err = NULL;
1134     const char *filename;
1135     int i, ret;
1136
1137     if ((BDRV_SECTOR_SIZE % 512) != 0) {
1138         error_setg(errp, "iSCSI: Invalid BDRV_SECTOR_SIZE. "
1139                    "BDRV_SECTOR_SIZE(%lld) is not a multiple "
1140                    "of 512", BDRV_SECTOR_SIZE);
1141         return -EINVAL;
1142     }
1143
1144     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1145     qemu_opts_absorb_qdict(opts, options, &local_err);
1146     if (local_err) {
1147         error_propagate(errp, local_err);
1148         ret = -EINVAL;
1149         goto out;
1150     }
1151
1152     filename = qemu_opt_get(opts, "filename");
1153
1154     iscsi_url = iscsi_parse_full_url(iscsi, filename);
1155     if (iscsi_url == NULL) {
1156         error_setg(errp, "Failed to parse URL : %s", filename);
1157         ret = -EINVAL;
1158         goto out;
1159     }
1160
1161     memset(iscsilun, 0, sizeof(IscsiLun));
1162
1163     initiator_name = parse_initiator_name(iscsi_url->target);
1164
1165     iscsi = iscsi_create_context(initiator_name);
1166     if (iscsi == NULL) {
1167         error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1168         ret = -ENOMEM;
1169         goto out;
1170     }
1171
1172     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1173         error_setg(errp, "iSCSI: Failed to set target name.");
1174         ret = -EINVAL;
1175         goto out;
1176     }
1177
1178     if (iscsi_url->user != NULL) {
1179         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1180                                               iscsi_url->passwd);
1181         if (ret != 0) {
1182             error_setg(errp, "Failed to set initiator username and password");
1183             ret = -EINVAL;
1184             goto out;
1185         }
1186     }
1187
1188     /* check if we got CHAP username/password via the options */
1189     parse_chap(iscsi, iscsi_url->target, &local_err);
1190     if (local_err != NULL) {
1191         error_propagate(errp, local_err);
1192         ret = -EINVAL;
1193         goto out;
1194     }
1195
1196     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1197         error_setg(errp, "iSCSI: Failed to set session type to normal.");
1198         ret = -EINVAL;
1199         goto out;
1200     }
1201
1202     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1203
1204     /* check if we got HEADER_DIGEST via the options */
1205     parse_header_digest(iscsi, iscsi_url->target, &local_err);
1206     if (local_err != NULL) {
1207         error_propagate(errp, local_err);
1208         ret = -EINVAL;
1209         goto out;
1210     }
1211
1212     if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1213         error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1214             iscsi_get_error(iscsi));
1215         ret = -EINVAL;
1216         goto out;
1217     }
1218
1219     iscsilun->iscsi = iscsi;
1220     iscsilun->lun   = iscsi_url->lun;
1221     iscsilun->has_write_same = true;
1222
1223     task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1224                             (void **) &inq, errp);
1225     if (task == NULL) {
1226         ret = -EINVAL;
1227         goto out;
1228     }
1229     iscsilun->type = inq->periperal_device_type;
1230     scsi_free_scsi_task(task);
1231     task = NULL;
1232
1233     iscsi_readcapacity_sync(iscsilun, &local_err);
1234     if (local_err != NULL) {
1235         error_propagate(errp, local_err);
1236         ret = -EINVAL;
1237         goto out;
1238     }
1239     bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1240     bs->request_alignment = iscsilun->block_size;
1241
1242     /* We don't have any emulation for devices other than disks and CD-ROMs, so
1243      * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1244      * will try to read from the device to guess the image format.
1245      */
1246     if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1247         bs->sg = 1;
1248     }
1249
1250     task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1251                             SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1252                             (void **) &inq_vpd, errp);
1253     if (task == NULL) {
1254         ret = -EINVAL;
1255         goto out;
1256     }
1257     for (i = 0; i < inq_vpd->num_pages; i++) {
1258         struct scsi_task *inq_task;
1259         struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1260         struct scsi_inquiry_block_limits *inq_bl;
1261         switch (inq_vpd->pages[i]) {
1262         case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1263             inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1264                                         SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1265                                         (void **) &inq_lbp, errp);
1266             if (inq_task == NULL) {
1267                 ret = -EINVAL;
1268                 goto out;
1269             }
1270             memcpy(&iscsilun->lbp, inq_lbp,
1271                    sizeof(struct scsi_inquiry_logical_block_provisioning));
1272             scsi_free_scsi_task(inq_task);
1273             break;
1274         case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1275             inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1276                                     SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1277                                     (void **) &inq_bl, errp);
1278             if (inq_task == NULL) {
1279                 ret = -EINVAL;
1280                 goto out;
1281             }
1282             memcpy(&iscsilun->bl, inq_bl,
1283                    sizeof(struct scsi_inquiry_block_limits));
1284             scsi_free_scsi_task(inq_task);
1285             break;
1286         default:
1287             break;
1288         }
1289     }
1290     scsi_free_scsi_task(task);
1291     task = NULL;
1292
1293 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
1294     /* Set up a timer for sending out iSCSI NOPs */
1295     iscsilun->nop_timer = timer_new_ms(QEMU_CLOCK_REALTIME, iscsi_nop_timed_event, iscsilun);
1296     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1297 #endif
1298
1299 out:
1300     qemu_opts_del(opts);
1301     if (initiator_name != NULL) {
1302         g_free(initiator_name);
1303     }
1304     if (iscsi_url != NULL) {
1305         iscsi_destroy_url(iscsi_url);
1306     }
1307     if (task != NULL) {
1308         scsi_free_scsi_task(task);
1309     }
1310
1311     if (ret) {
1312         if (iscsi != NULL) {
1313             iscsi_destroy_context(iscsi);
1314         }
1315         memset(iscsilun, 0, sizeof(IscsiLun));
1316     }
1317     return ret;
1318 }
1319
1320 static void iscsi_close(BlockDriverState *bs)
1321 {
1322     IscsiLun *iscsilun = bs->opaque;
1323     struct iscsi_context *iscsi = iscsilun->iscsi;
1324
1325     if (iscsilun->nop_timer) {
1326         timer_del(iscsilun->nop_timer);
1327         timer_free(iscsilun->nop_timer);
1328     }
1329     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL);
1330     iscsi_destroy_context(iscsi);
1331     g_free(iscsilun->zeroblock);
1332     memset(iscsilun, 0, sizeof(IscsiLun));
1333 }
1334
1335 static int iscsi_refresh_limits(BlockDriverState *bs)
1336 {
1337     IscsiLun *iscsilun = bs->opaque;
1338
1339     /* We don't actually refresh here, but just return data queried in
1340      * iscsi_open(): iscsi targets don't change their limits. */
1341     if (iscsilun->lbp.lbpu) {
1342         if (iscsilun->bl.max_unmap < 0xffffffff) {
1343             bs->bl.max_discard = sector_lun2qemu(iscsilun->bl.max_unmap,
1344                                                  iscsilun);
1345         }
1346         bs->bl.discard_alignment = sector_lun2qemu(iscsilun->bl.opt_unmap_gran,
1347                                                    iscsilun);
1348     }
1349
1350     if (iscsilun->bl.max_ws_len < 0xffffffff) {
1351         bs->bl.max_write_zeroes = sector_lun2qemu(iscsilun->bl.max_ws_len,
1352                                                   iscsilun);
1353     }
1354     if (iscsilun->lbp.lbpws) {
1355         bs->bl.write_zeroes_alignment = sector_lun2qemu(iscsilun->bl.opt_unmap_gran,
1356                                                         iscsilun);
1357     }
1358     bs->bl.opt_transfer_length = sector_lun2qemu(iscsilun->bl.opt_xfer_len,
1359                                                  iscsilun);
1360     return 0;
1361 }
1362
1363 /* Since iscsi_open() ignores bdrv_flags, there is nothing to do here in
1364  * prepare.  Note that this will not re-establish a connection with an iSCSI
1365  * target - it is effectively a NOP.  */
1366 static int iscsi_reopen_prepare(BDRVReopenState *state,
1367                                 BlockReopenQueue *queue, Error **errp)
1368 {
1369     /* NOP */
1370     return 0;
1371 }
1372
1373 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1374 {
1375     IscsiLun *iscsilun = bs->opaque;
1376     Error *local_err = NULL;
1377
1378     if (iscsilun->type != TYPE_DISK) {
1379         return -ENOTSUP;
1380     }
1381
1382     iscsi_readcapacity_sync(iscsilun, &local_err);
1383     if (local_err != NULL) {
1384         error_free(local_err);
1385         return -EIO;
1386     }
1387
1388     if (offset > iscsi_getlength(bs)) {
1389         return -EINVAL;
1390     }
1391
1392     return 0;
1393 }
1394
1395 static int iscsi_create(const char *filename, QEMUOptionParameter *options,
1396                         Error **errp)
1397 {
1398     int ret = 0;
1399     int64_t total_size = 0;
1400     BlockDriverState *bs;
1401     IscsiLun *iscsilun = NULL;
1402     QDict *bs_options;
1403
1404     bs = bdrv_new("");
1405
1406     /* Read out options */
1407     while (options && options->name) {
1408         if (!strcmp(options->name, "size")) {
1409             total_size = options->value.n / BDRV_SECTOR_SIZE;
1410         }
1411         options++;
1412     }
1413
1414     bs->opaque = g_malloc0(sizeof(struct IscsiLun));
1415     iscsilun = bs->opaque;
1416
1417     bs_options = qdict_new();
1418     qdict_put(bs_options, "filename", qstring_from_str(filename));
1419     ret = iscsi_open(bs, bs_options, 0, NULL);
1420     QDECREF(bs_options);
1421
1422     if (ret != 0) {
1423         goto out;
1424     }
1425     if (iscsilun->nop_timer) {
1426         timer_del(iscsilun->nop_timer);
1427         timer_free(iscsilun->nop_timer);
1428     }
1429     if (iscsilun->type != TYPE_DISK) {
1430         ret = -ENODEV;
1431         goto out;
1432     }
1433     if (bs->total_sectors < total_size) {
1434         ret = -ENOSPC;
1435         goto out;
1436     }
1437
1438     ret = 0;
1439 out:
1440     if (iscsilun->iscsi != NULL) {
1441         iscsi_destroy_context(iscsilun->iscsi);
1442     }
1443     g_free(bs->opaque);
1444     bs->opaque = NULL;
1445     bdrv_unref(bs);
1446     return ret;
1447 }
1448
1449 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1450 {
1451     IscsiLun *iscsilun = bs->opaque;
1452     bdi->unallocated_blocks_are_zero = !!iscsilun->lbprz;
1453     bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
1454     /* Guess the internal cluster (page) size of the iscsi target by the means
1455      * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1456      * reasonable size for bdi->cluster_size */
1457     if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 64 * 1024 &&
1458         iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1459         bdi->cluster_size = iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
1460     }
1461     return 0;
1462 }
1463
1464 static QEMUOptionParameter iscsi_create_options[] = {
1465     {
1466         .name = BLOCK_OPT_SIZE,
1467         .type = OPT_SIZE,
1468         .help = "Virtual disk size"
1469     },
1470     { NULL }
1471 };
1472
1473 static BlockDriver bdrv_iscsi = {
1474     .format_name     = "iscsi",
1475     .protocol_name   = "iscsi",
1476
1477     .instance_size   = sizeof(IscsiLun),
1478     .bdrv_needs_filename = true,
1479     .bdrv_file_open  = iscsi_open,
1480     .bdrv_close      = iscsi_close,
1481     .bdrv_create     = iscsi_create,
1482     .create_options  = iscsi_create_options,
1483     .bdrv_reopen_prepare  = iscsi_reopen_prepare,
1484
1485     .bdrv_getlength  = iscsi_getlength,
1486     .bdrv_get_info   = iscsi_get_info,
1487     .bdrv_truncate   = iscsi_truncate,
1488     .bdrv_refresh_limits = iscsi_refresh_limits,
1489
1490 #if defined(LIBISCSI_FEATURE_IOVECTOR)
1491     .bdrv_co_get_block_status = iscsi_co_get_block_status,
1492 #endif
1493     .bdrv_co_discard      = iscsi_co_discard,
1494 #if defined(SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED)
1495     .bdrv_co_write_zeroes = iscsi_co_write_zeroes,
1496 #endif
1497     .bdrv_co_readv         = iscsi_co_readv,
1498     .bdrv_co_writev        = iscsi_co_writev,
1499     .bdrv_co_flush_to_disk = iscsi_co_flush,
1500
1501 #ifdef __linux__
1502     .bdrv_ioctl       = iscsi_ioctl,
1503     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1504 #endif
1505 };
1506
1507 static QemuOptsList qemu_iscsi_opts = {
1508     .name = "iscsi",
1509     .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1510     .desc = {
1511         {
1512             .name = "user",
1513             .type = QEMU_OPT_STRING,
1514             .help = "username for CHAP authentication to target",
1515         },{
1516             .name = "password",
1517             .type = QEMU_OPT_STRING,
1518             .help = "password for CHAP authentication to target",
1519         },{
1520             .name = "header-digest",
1521             .type = QEMU_OPT_STRING,
1522             .help = "HeaderDigest setting. "
1523                     "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1524         },{
1525             .name = "initiator-name",
1526             .type = QEMU_OPT_STRING,
1527             .help = "Initiator iqn name to use when connecting",
1528         },
1529         { /* end of list */ }
1530     },
1531 };
1532
1533 static void iscsi_block_init(void)
1534 {
1535     bdrv_register(&bdrv_iscsi);
1536     qemu_add_opts(&qemu_iscsi_opts);
1537 }
1538
1539 block_init(iscsi_block_init);