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     iscsi_co_init_iscsitask(iscsilun, &iTask);
 421
 422 retry:
 423     if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
 424                                       0, iscsi_co_generic_cb, &iTask) == NULL) {
 425         return -ENOMEM;
 426     }
 427
 428     while (!iTask.complete) {
 429         iscsi_set_events(iscsilun);
 430         qemu_coroutine_yield();
 431     }
 432
 433     if (iTask.task != NULL) {
 434         scsi_free_scsi_task(iTask.task);
 435         iTask.task = NULL;
 436     }
 437
 438     if (iTask.do_retry) {
 439         iTask.complete = 0;
 440         goto retry;
 441     }
 442
 443     if (iTask.status != SCSI_STATUS_GOOD) {
 444         return -EIO;
 445     }
 446
 447     return 0;
 448 }
 449
 450 #ifdef __linux__
 451 static void
 452 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 453                      void *command_data, void *opaque)
 454 {
 455     IscsiAIOCB *acb = opaque;
 456
 457     g_free(acb->buf);
 458     acb->buf = NULL;
 459
 460     if (acb->canceled != 0) {
 461         return;
 462     }
 463
 464     acb->status = 0;
 465     if (status < 0) {
 466         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 467                      iscsi_get_error(iscsi));
 468         acb->status = -EIO;
 469     }
 470
 471     acb->ioh->driver_status = 0;
 472     acb->ioh->host_status   = 0;
 473     acb->ioh->resid         = 0;
 474
 475 #define SG_ERR_DRIVER_SENSE    0x08
 476
 477     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 478         int ss;
 479
 480         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 481
 482         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 483         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 484              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 485         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 486     }
 487
 488     iscsi_schedule_bh(acb);
 489 }
 490
 491 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 492         unsigned long int req, void *buf,
 493         BlockDriverCompletionFunc *cb, void *opaque)
 494 {
 495     IscsiLun *iscsilun = bs->opaque;
 496     struct iscsi_context *iscsi = iscsilun->iscsi;
 497     struct iscsi_data data;
 498     IscsiAIOCB *acb;
 499
 500     assert(req == SG_IO);
 501
 502     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 503
 504     acb->iscsilun = iscsilun;
 505     acb->canceled    = 0;
 506     acb->bh          = NULL;
 507     acb->status      = -EINPROGRESS;
 508     acb->buf         = NULL;
 509     acb->ioh         = buf;
 510
 511     acb->task = malloc(sizeof(struct scsi_task));
 512     if (acb->task == NULL) {
 513         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 514                      iscsi_get_error(iscsi));
 515         qemu_aio_release(acb);
 516         return NULL;
 517     }
 518     memset(acb->task, 0, sizeof(struct scsi_task));
 519
 520     switch (acb->ioh->dxfer_direction) {
 521     case SG_DXFER_TO_DEV:
 522         acb->task->xfer_dir = SCSI_XFER_WRITE;
 523         break;
 524     case SG_DXFER_FROM_DEV:
 525         acb->task->xfer_dir = SCSI_XFER_READ;
 526         break;
 527     default:
 528         acb->task->xfer_dir = SCSI_XFER_NONE;
 529         break;
 530     }
 531
 532     acb->task->cdb_size = acb->ioh->cmd_len;
 533     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 534     acb->task->expxferlen = acb->ioh->dxfer_len;
 535
 536     data.size = 0;
 537     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 538         if (acb->ioh->iovec_count == 0) {
 539             data.data = acb->ioh->dxferp;
 540             data.size = acb->ioh->dxfer_len;
 541         } else {
 542 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 543             scsi_task_set_iov_out(acb->task,
 544                                  (struct scsi_iovec *) acb->ioh->dxferp,
 545                                  acb->ioh->iovec_count);
 546 #else
 547             struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
 548
 549             acb->buf = g_malloc(acb->ioh->dxfer_len);
 550             data.data = acb->buf;
 551             data.size = iov_to_buf(iov, acb->ioh->iovec_count, 0,
 552                                    acb->buf, acb->ioh->dxfer_len);
 553 #endif
 554         }
 555     }
 556
 557     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 558                                  iscsi_aio_ioctl_cb,
 559                                  (data.size > 0) ? &data : NULL,
 560                                  acb) != 0) {
 561         scsi_free_scsi_task(acb->task);
 562         qemu_aio_release(acb);
 563         return NULL;
 564     }
 565
 566     /* tell libiscsi to read straight into the buffer we got from ioctl */
 567     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 568         if (acb->ioh->iovec_count == 0) {
 569             scsi_task_add_data_in_buffer(acb->task,
 570                                          acb->ioh->dxfer_len,
 571                                          acb->ioh->dxferp);
 572         } else {
 573 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 574             scsi_task_set_iov_in(acb->task,
 575                                  (struct scsi_iovec *) acb->ioh->dxferp,
 576                                  acb->ioh->iovec_count);
 577 #else
 578             int i;
 579             for (i = 0; i < acb->ioh->iovec_count; i++) {
 580                 struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
 581
 582                 scsi_task_add_data_in_buffer(acb->task,
 583                     iov[i].iov_len,
 584                     iov[i].iov_base);
 585             }
 586 #endif
 587         }
 588     }
 589
 590     iscsi_set_events(iscsilun);
 591
 592     return &acb->common;
 593 }
 594
 595
 596 static void ioctl_cb(void *opaque, int status)
 597 {
 598     int *p_status = opaque;
 599     *p_status = status;
 600 }
 601
 602 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 603 {
 604     IscsiLun *iscsilun = bs->opaque;
 605     int status;
 606
 607     switch (req) {
 608     case SG_GET_VERSION_NUM:
 609         *(int *)buf = 30000;
 610         break;
 611     case SG_GET_SCSI_ID:
 612         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 613         break;
 614     case SG_IO:
 615         status = -EINPROGRESS;
 616         iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status);
 617
 618         while (status == -EINPROGRESS) {
 619             qemu_aio_wait();
 620         }
 621
 622         return 0;
 623     default:
 624         return -1;
 625     }
 626     return 0;
 627 }
 628 #endif
 629
 630 static int64_t
 631 iscsi_getlength(BlockDriverState *bs)
 632 {
 633     IscsiLun *iscsilun = bs->opaque;
 634     int64_t len;
 635
 636     len  = iscsilun->num_blocks;
 637     len *= iscsilun->block_size;
 638
 639     return len;
 640 }
 641
 642 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 643
 644 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
 645                                                   int64_t sector_num,
 646                                                   int nb_sectors, int *pnum)
 647 {
 648     IscsiLun *iscsilun = bs->opaque;
 649     struct scsi_get_lba_status *lbas = NULL;
 650     struct scsi_lba_status_descriptor *lbasd = NULL;
 651     struct IscsiTask iTask;
 652     int64_t ret;
 653
 654     iscsi_co_init_iscsitask(iscsilun, &iTask);
 655
 656     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 657         ret = -EINVAL;
 658         goto out;
 659     }
 660
 661     /* default to all sectors allocated */
 662     ret = BDRV_BLOCK_DATA;
 663     ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
 664     *pnum = nb_sectors;
 665
 666     /* LUN does not support logical block provisioning */
 667     if (iscsilun->lbpme == 0) {
 668         goto out;
 669     }
 670
 671 retry:
 672     if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
 673                                   sector_qemu2lun(sector_num, iscsilun),
 674                                   8 + 16, iscsi_co_generic_cb,
 675                                   &iTask) == NULL) {
 676         ret = -ENOMEM;
 677         goto out;
 678     }
 679
 680     while (!iTask.complete) {
 681         iscsi_set_events(iscsilun);
 682         qemu_coroutine_yield();
 683     }
 684
 685     if (iTask.do_retry) {
 686         if (iTask.task != NULL) {
 687             scsi_free_scsi_task(iTask.task);
 688             iTask.task = NULL;
 689         }
 690         iTask.complete = 0;
 691         goto retry;
 692     }
 693
 694     if (iTask.status != SCSI_STATUS_GOOD) {
 695         /* in case the get_lba_status_callout fails (i.e.
 696          * because the device is busy or the cmd is not
 697          * supported) we pretend all blocks are allocated
 698          * for backwards compatibility */
 699         goto out;
 700     }
 701
 702     lbas = scsi_datain_unmarshall(iTask.task);
 703     if (lbas == NULL) {
 704         ret = -EIO;
 705         goto out;
 706     }
 707
 708     lbasd = &lbas->descriptors[0];
 709
 710     if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
 711         ret = -EIO;
 712         goto out;
 713     }
 714
 715     *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
 716     if (*pnum > nb_sectors) {
 717         *pnum = nb_sectors;
 718     }
 719
 720     if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
 721         lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
 722         ret &= ~BDRV_BLOCK_DATA;
 723         if (iscsilun->lbprz) {
 724             ret |= BDRV_BLOCK_ZERO;
 725         }
 726     }
 727
 728 out:
 729     if (iTask.task != NULL) {
 730         scsi_free_scsi_task(iTask.task);
 731     }
 732     return ret;
 733 }
 734
 735 #endif /* LIBISCSI_FEATURE_IOVECTOR */
 736
 737 static int
 738 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num,
 739                                    int nb_sectors)
 740 {
 741     IscsiLun *iscsilun = bs->opaque;
 742     struct IscsiTask iTask;
 743     struct unmap_list list;
 744
 745     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 746         return -EINVAL;
 747     }
 748
 749     if (!iscsilun->lbp.lbpu) {
 750         /* UNMAP is not supported by the target */
 751         return 0;
 752     }
 753
 754     list.lba = sector_qemu2lun(sector_num, iscsilun);
 755     list.num = sector_qemu2lun(nb_sectors, iscsilun);
 756
 757     iscsi_co_init_iscsitask(iscsilun, &iTask);
 758 retry:
 759     if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
 760                      iscsi_co_generic_cb, &iTask) == NULL) {
 761         return -ENOMEM;
 762     }
 763
 764     while (!iTask.complete) {
 765         iscsi_set_events(iscsilun);
 766         qemu_coroutine_yield();
 767     }
 768
 769     if (iTask.task != NULL) {
 770         scsi_free_scsi_task(iTask.task);
 771         iTask.task = NULL;
 772     }
 773
 774     if (iTask.do_retry) {
 775         iTask.complete = 0;
 776         goto retry;
 777     }
 778
 779     if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
 780         /* the target might fail with a check condition if it
 781            is not happy with the alignment of the UNMAP request
 782            we silently fail in this case */
 783         return 0;
 784     }
 785
 786     if (iTask.status != SCSI_STATUS_GOOD) {
 787         return -EIO;
 788     }
 789
 790     return 0;
 791 }
 792
 793 #if defined(SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED)
 794
 795 static int
 796 coroutine_fn iscsi_co_write_zeroes(BlockDriverState *bs, int64_t sector_num,
 797                                    int nb_sectors, BdrvRequestFlags flags)
 798 {
 799     IscsiLun *iscsilun = bs->opaque;
 800     struct IscsiTask iTask;
 801     uint64_t lba;
 802     uint32_t nb_blocks;
 803
 804     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 805         return -EINVAL;
 806     }
 807
 808     if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
 809         /* WRITE SAME without UNMAP is not supported by the target */
 810         return -ENOTSUP;
 811     }
 812
 813     if ((flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->lbp.lbpws) {
 814         /* WRITE SAME with UNMAP is not supported by the target */
 815         return -ENOTSUP;
 816     }
 817
 818     lba = sector_qemu2lun(sector_num, iscsilun);
 819     nb_blocks = sector_qemu2lun(nb_sectors, iscsilun);
 820
 821     if (iscsilun->zeroblock == NULL) {
 822         iscsilun->zeroblock = g_malloc0(iscsilun->block_size);
 823     }
 824
 825     iscsi_co_init_iscsitask(iscsilun, &iTask);
 826 retry:
 827     if (iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
 828                                iscsilun->zeroblock, iscsilun->block_size,
 829                                nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
 830                                0, 0, iscsi_co_generic_cb, &iTask) == NULL) {
 831         return -ENOMEM;
 832     }
 833
 834     while (!iTask.complete) {
 835         iscsi_set_events(iscsilun);
 836         qemu_coroutine_yield();
 837     }
 838
 839     if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
 840         iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
 841         iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE) {
 842         /* WRITE SAME is not supported by the target */
 843         iscsilun->has_write_same = false;
 844         scsi_free_scsi_task(iTask.task);
 845         return -ENOTSUP;
 846     }
 847
 848     if (iTask.task != NULL) {
 849         scsi_free_scsi_task(iTask.task);
 850         iTask.task = NULL;
 851     }
 852
 853     if (iTask.do_retry) {
 854         iTask.complete = 0;
 855         goto retry;
 856     }
 857
 858     if (iTask.status != SCSI_STATUS_GOOD) {
 859         return -EIO;
 860     }
 861
 862     return 0;
 863 }
 864
 865 #endif /* SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED */
 866
 867 static void parse_chap(struct iscsi_context *iscsi, const char *target,
 868                        Error **errp)
 869 {
 870     QemuOptsList *list;
 871     QemuOpts *opts;
 872     const char *user = NULL;
 873     const char *password = NULL;
 874
 875     list = qemu_find_opts("iscsi");
 876     if (!list) {
 877         return;
 878     }
 879
 880     opts = qemu_opts_find(list, target);
 881     if (opts == NULL) {
 882         opts = QTAILQ_FIRST(&list->head);
 883         if (!opts) {
 884             return;
 885         }
 886     }
 887
 888     user = qemu_opt_get(opts, "user");
 889     if (!user) {
 890         return;
 891     }
 892
 893     password = qemu_opt_get(opts, "password");
 894     if (!password) {
 895         error_setg(errp, "CHAP username specified but no password was given");
 896         return;
 897     }
 898
 899     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
 900         error_setg(errp, "Failed to set initiator username and password");
 901     }
 902 }
 903
 904 static void parse_header_digest(struct iscsi_context *iscsi, const char *target,
 905                                 Error **errp)
 906 {
 907     QemuOptsList *list;
 908     QemuOpts *opts;
 909     const char *digest = NULL;
 910
 911     list = qemu_find_opts("iscsi");
 912     if (!list) {
 913         return;
 914     }
 915
 916     opts = qemu_opts_find(list, target);
 917     if (opts == NULL) {
 918         opts = QTAILQ_FIRST(&list->head);
 919         if (!opts) {
 920             return;
 921         }
 922     }
 923
 924     digest = qemu_opt_get(opts, "header-digest");
 925     if (!digest) {
 926         return;
 927     }
 928
 929     if (!strcmp(digest, "CRC32C")) {
 930         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
 931     } else if (!strcmp(digest, "NONE")) {
 932         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
 933     } else if (!strcmp(digest, "CRC32C-NONE")) {
 934         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
 935     } else if (!strcmp(digest, "NONE-CRC32C")) {
 936         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 937     } else {
 938         error_setg(errp, "Invalid header-digest setting : %s", digest);
 939     }
 940 }
 941
 942 static char *parse_initiator_name(const char *target)
 943 {
 944     QemuOptsList *list;
 945     QemuOpts *opts;
 946     const char *name;
 947     char *iscsi_name;
 948     UuidInfo *uuid_info;
 949
 950     list = qemu_find_opts("iscsi");
 951     if (list) {
 952         opts = qemu_opts_find(list, target);
 953         if (!opts) {
 954             opts = QTAILQ_FIRST(&list->head);
 955         }
 956         if (opts) {
 957             name = qemu_opt_get(opts, "initiator-name");
 958             if (name) {
 959                 return g_strdup(name);
 960             }
 961         }
 962     }
 963
 964     uuid_info = qmp_query_uuid(NULL);
 965     if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
 966         name = qemu_get_vm_name();
 967     } else {
 968         name = uuid_info->UUID;
 969     }
 970     iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
 971                                  name ? ":" : "", name ? name : "");
 972     qapi_free_UuidInfo(uuid_info);
 973     return iscsi_name;
 974 }
 975
 976 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
 977 static void iscsi_nop_timed_event(void *opaque)
 978 {
 979     IscsiLun *iscsilun = opaque;
 980
 981     if (iscsi_get_nops_in_flight(iscsilun->iscsi) > MAX_NOP_FAILURES) {
 982         error_report("iSCSI: NOP timeout. Reconnecting...");
 983         iscsi_reconnect(iscsilun->iscsi);
 984     }
 985
 986     if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
 987         error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
 988         return;
 989     }
 990
 991     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
 992     iscsi_set_events(iscsilun);
 993 }
 994 #endif
 995
 996 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
 997 {
 998     struct scsi_task *task = NULL;
 999     struct scsi_readcapacity10 *rc10 = NULL;
1000     struct scsi_readcapacity16 *rc16 = NULL;
1001     int retries = ISCSI_CMD_RETRIES;
1002
1003     do {
1004         if (task != NULL) {
1005             scsi_free_scsi_task(task);
1006             task = NULL;
1007         }
1008
1009         switch (iscsilun->type) {
1010         case TYPE_DISK:
1011             task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1012             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1013                 rc16 = scsi_datain_unmarshall(task);
1014                 if (rc16 == NULL) {
1015                     error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1016                 } else {
1017                     iscsilun->block_size = rc16->block_length;
1018                     iscsilun->num_blocks = rc16->returned_lba + 1;
1019                     iscsilun->lbpme = rc16->lbpme;
1020                     iscsilun->lbprz = rc16->lbprz;
1021                 }
1022             }
1023             break;
1024         case TYPE_ROM:
1025             task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1026             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1027                 rc10 = scsi_datain_unmarshall(task);
1028                 if (rc10 == NULL) {
1029                     error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1030                 } else {
1031                     iscsilun->block_size = rc10->block_size;
1032                     if (rc10->lba == 0) {
1033                         /* blank disk loaded */
1034                         iscsilun->num_blocks = 0;
1035                     } else {
1036                         iscsilun->num_blocks = rc10->lba + 1;
1037                     }
1038                 }
1039             }
1040             break;
1041         default:
1042             return;
1043         }
1044     } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1045              && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1046              && retries-- > 0);
1047
1048     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1049         error_setg(errp, "iSCSI: failed to send readcapacity10 command.");
1050     }
1051     if (task) {
1052         scsi_free_scsi_task(task);
1053     }
1054 }
1055
1056 /* TODO Convert to fine grained options */
1057 static QemuOptsList runtime_opts = {
1058     .name = "iscsi",
1059     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1060     .desc = {
1061         {
1062             .name = "filename",
1063             .type = QEMU_OPT_STRING,
1064             .help = "URL to the iscsi image",
1065         },
1066         { /* end of list */ }
1067     },
1068 };
1069
1070 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1071                                           int evpd, int pc, void **inq, Error **errp)
1072 {
1073     int full_size;
1074     struct scsi_task *task = NULL;
1075     task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1076     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1077         goto fail;
1078     }
1079     full_size = scsi_datain_getfullsize(task);
1080     if (full_size > task->datain.size) {
1081         scsi_free_scsi_task(task);
1082
1083         /* we need more data for the full list */
1084         task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1085         if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1086             goto fail;
1087         }
1088     }
1089
1090     *inq = scsi_datain_unmarshall(task);
1091     if (*inq == NULL) {
1092         error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1093         goto fail;
1094     }
1095
1096     return task;
1097
1098 fail:
1099     error_setg(errp, "iSCSI: Inquiry command failed : %s",
1100                iscsi_get_error(iscsi));
1101     if (task != NULL) {
1102         scsi_free_scsi_task(task);
1103     }
1104     return NULL;
1105 }
1106
1107 /*
1108  * We support iscsi url's on the form
1109  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1110  *
1111  * Note: flags are currently not used by iscsi_open.  If this function
1112  * is changed such that flags are used, please examine iscsi_reopen_prepare()
1113  * to see if needs to be changed as well.
1114  */
1115 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1116                       Error **errp)
1117 {
1118     IscsiLun *iscsilun = bs->opaque;
1119     struct iscsi_context *iscsi = NULL;
1120     struct iscsi_url *iscsi_url = NULL;
1121     struct scsi_task *task = NULL;
1122     struct scsi_inquiry_standard *inq = NULL;
1123     struct scsi_inquiry_supported_pages *inq_vpd;
1124     char *initiator_name = NULL;
1125     QemuOpts *opts;
1126     Error *local_err = NULL;
1127     const char *filename;
1128     int i, ret;
1129
1130     if ((BDRV_SECTOR_SIZE % 512) != 0) {
1131         error_setg(errp, "iSCSI: Invalid BDRV_SECTOR_SIZE. "
1132                    "BDRV_SECTOR_SIZE(%lld) is not a multiple "
1133                    "of 512", BDRV_SECTOR_SIZE);
1134         return -EINVAL;
1135     }
1136
1137     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1138     qemu_opts_absorb_qdict(opts, options, &local_err);
1139     if (local_err) {
1140         error_propagate(errp, local_err);
1141         ret = -EINVAL;
1142         goto out;
1143     }
1144
1145     filename = qemu_opt_get(opts, "filename");
1146
1147     iscsi_url = iscsi_parse_full_url(iscsi, filename);
1148     if (iscsi_url == NULL) {
1149         error_setg(errp, "Failed to parse URL : %s", filename);
1150         ret = -EINVAL;
1151         goto out;
1152     }
1153
1154     memset(iscsilun, 0, sizeof(IscsiLun));
1155
1156     initiator_name = parse_initiator_name(iscsi_url->target);
1157
1158     iscsi = iscsi_create_context(initiator_name);
1159     if (iscsi == NULL) {
1160         error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1161         ret = -ENOMEM;
1162         goto out;
1163     }
1164
1165     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1166         error_setg(errp, "iSCSI: Failed to set target name.");
1167         ret = -EINVAL;
1168         goto out;
1169     }
1170
1171     if (iscsi_url->user != NULL) {
1172         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1173                                               iscsi_url->passwd);
1174         if (ret != 0) {
1175             error_setg(errp, "Failed to set initiator username and password");
1176             ret = -EINVAL;
1177             goto out;
1178         }
1179     }
1180
1181     /* check if we got CHAP username/password via the options */
1182     parse_chap(iscsi, iscsi_url->target, &local_err);
1183     if (local_err != NULL) {
1184         error_propagate(errp, local_err);
1185         ret = -EINVAL;
1186         goto out;
1187     }
1188
1189     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1190         error_setg(errp, "iSCSI: Failed to set session type to normal.");
1191         ret = -EINVAL;
1192         goto out;
1193     }
1194
1195     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1196
1197     /* check if we got HEADER_DIGEST via the options */
1198     parse_header_digest(iscsi, iscsi_url->target, &local_err);
1199     if (local_err != NULL) {
1200         error_propagate(errp, local_err);
1201         ret = -EINVAL;
1202         goto out;
1203     }
1204
1205     if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1206         error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1207             iscsi_get_error(iscsi));
1208         ret = -EINVAL;
1209         goto out;
1210     }
1211
1212     iscsilun->iscsi = iscsi;
1213     iscsilun->lun   = iscsi_url->lun;
1214     iscsilun->has_write_same = true;
1215
1216     task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1217                             (void **) &inq, errp);
1218     if (task == NULL) {
1219         ret = -EINVAL;
1220         goto out;
1221     }
1222     iscsilun->type = inq->periperal_device_type;
1223     scsi_free_scsi_task(task);
1224     task = NULL;
1225
1226     iscsi_readcapacity_sync(iscsilun, &local_err);
1227     if (local_err != NULL) {
1228         error_propagate(errp, local_err);
1229         goto out;
1230     }
1231     bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1232     bs->request_alignment = iscsilun->block_size;
1233
1234     /* Medium changer or tape. We dont have any emulation for this so this must
1235      * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
1236      * to read from the device to guess the image format.
1237      */
1238     if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
1239         iscsilun->type == TYPE_TAPE) {
1240         bs->sg = 1;
1241     }
1242
1243     task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1244                             SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1245                             (void **) &inq_vpd, errp);
1246     if (task == NULL) {
1247         ret = -EINVAL;
1248         goto out;
1249     }
1250     for (i = 0; i < inq_vpd->num_pages; i++) {
1251         struct scsi_task *inq_task;
1252         struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1253         struct scsi_inquiry_block_limits *inq_bl;
1254         switch (inq_vpd->pages[i]) {
1255         case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1256             inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1257                                         SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1258                                         (void **) &inq_lbp, errp);
1259             if (inq_task == NULL) {
1260                 ret = -EINVAL;
1261                 goto out;
1262             }
1263             memcpy(&iscsilun->lbp, inq_lbp,
1264                    sizeof(struct scsi_inquiry_logical_block_provisioning));
1265             scsi_free_scsi_task(inq_task);
1266             break;
1267         case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1268             inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1269                                     SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1270                                     (void **) &inq_bl, errp);
1271             if (inq_task == NULL) {
1272                 ret = -EINVAL;
1273                 goto out;
1274             }
1275             memcpy(&iscsilun->bl, inq_bl,
1276                    sizeof(struct scsi_inquiry_block_limits));
1277             scsi_free_scsi_task(inq_task);
1278             break;
1279         default:
1280             break;
1281         }
1282     }
1283     scsi_free_scsi_task(task);
1284     task = NULL;
1285
1286 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
1287     /* Set up a timer for sending out iSCSI NOPs */
1288     iscsilun->nop_timer = timer_new_ms(QEMU_CLOCK_REALTIME, iscsi_nop_timed_event, iscsilun);
1289     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1290 #endif
1291
1292 out:
1293     qemu_opts_del(opts);
1294     if (initiator_name != NULL) {
1295         g_free(initiator_name);
1296     }
1297     if (iscsi_url != NULL) {
1298         iscsi_destroy_url(iscsi_url);
1299     }
1300     if (task != NULL) {
1301         scsi_free_scsi_task(task);
1302     }
1303
1304     if (ret) {
1305         if (iscsi != NULL) {
1306             iscsi_destroy_context(iscsi);
1307         }
1308         memset(iscsilun, 0, sizeof(IscsiLun));
1309     }
1310     return ret;
1311 }
1312
1313 static void iscsi_close(BlockDriverState *bs)
1314 {
1315     IscsiLun *iscsilun = bs->opaque;
1316     struct iscsi_context *iscsi = iscsilun->iscsi;
1317
1318     if (iscsilun->nop_timer) {
1319         timer_del(iscsilun->nop_timer);
1320         timer_free(iscsilun->nop_timer);
1321     }
1322     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL);
1323     iscsi_destroy_context(iscsi);
1324     g_free(iscsilun->zeroblock);
1325     memset(iscsilun, 0, sizeof(IscsiLun));
1326 }
1327
1328 static int iscsi_refresh_limits(BlockDriverState *bs)
1329 {
1330     IscsiLun *iscsilun = bs->opaque;
1331
1332     /* We don't actually refresh here, but just return data queried in
1333      * iscsi_open(): iscsi targets don't change their limits. */
1334     if (iscsilun->lbp.lbpu || iscsilun->lbp.lbpws) {
1335         if (iscsilun->bl.max_unmap < 0xffffffff) {
1336             bs->bl.max_discard = sector_lun2qemu(iscsilun->bl.max_unmap,
1337                                                  iscsilun);
1338         }
1339         bs->bl.discard_alignment = sector_lun2qemu(iscsilun->bl.opt_unmap_gran,
1340                                                    iscsilun);
1341
1342         if (iscsilun->bl.max_ws_len < 0xffffffff) {
1343             bs->bl.max_write_zeroes = sector_lun2qemu(iscsilun->bl.max_ws_len,
1344                                                       iscsilun);
1345         }
1346         bs->bl.write_zeroes_alignment = sector_lun2qemu(iscsilun->bl.opt_unmap_gran,
1347                                                         iscsilun);
1348     }
1349     bs->bl.opt_transfer_length = sector_lun2qemu(iscsilun->bl.opt_xfer_len,
1350                                                  iscsilun);
1351     return 0;
1352 }
1353
1354 /* Since iscsi_open() ignores bdrv_flags, there is nothing to do here in
1355  * prepare.  Note that this will not re-establish a connection with an iSCSI
1356  * target - it is effectively a NOP.  */
1357 static int iscsi_reopen_prepare(BDRVReopenState *state,
1358                                 BlockReopenQueue *queue, Error **errp)
1359 {
1360     /* NOP */
1361     return 0;
1362 }
1363
1364 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1365 {
1366     IscsiLun *iscsilun = bs->opaque;
1367     Error *local_err = NULL;
1368
1369     if (iscsilun->type != TYPE_DISK) {
1370         return -ENOTSUP;
1371     }
1372
1373     iscsi_readcapacity_sync(iscsilun, &local_err);
1374     if (local_err != NULL) {
1375         error_free(local_err);
1376         return -EIO;
1377     }
1378
1379     if (offset > iscsi_getlength(bs)) {
1380         return -EINVAL;
1381     }
1382
1383     return 0;
1384 }
1385
1386 static int iscsi_create(const char *filename, QEMUOptionParameter *options,
1387                         Error **errp)
1388 {
1389     int ret = 0;
1390     int64_t total_size = 0;
1391     BlockDriverState *bs;
1392     IscsiLun *iscsilun = NULL;
1393     QDict *bs_options;
1394
1395     bs = bdrv_new("");
1396
1397     /* Read out options */
1398     while (options && options->name) {
1399         if (!strcmp(options->name, "size")) {
1400             total_size = options->value.n / BDRV_SECTOR_SIZE;
1401         }
1402         options++;
1403     }
1404
1405     bs->opaque = g_malloc0(sizeof(struct IscsiLun));
1406     iscsilun = bs->opaque;
1407
1408     bs_options = qdict_new();
1409     qdict_put(bs_options, "filename", qstring_from_str(filename));
1410     ret = iscsi_open(bs, bs_options, 0, NULL);
1411     QDECREF(bs_options);
1412
1413     if (ret != 0) {
1414         goto out;
1415     }
1416     if (iscsilun->nop_timer) {
1417         timer_del(iscsilun->nop_timer);
1418         timer_free(iscsilun->nop_timer);
1419     }
1420     if (iscsilun->type != TYPE_DISK) {
1421         ret = -ENODEV;
1422         goto out;
1423     }
1424     if (bs->total_sectors < total_size) {
1425         ret = -ENOSPC;
1426         goto out;
1427     }
1428
1429     ret = 0;
1430 out:
1431     if (iscsilun->iscsi != NULL) {
1432         iscsi_destroy_context(iscsilun->iscsi);
1433     }
1434     g_free(bs->opaque);
1435     bs->opaque = NULL;
1436     bdrv_unref(bs);
1437     return ret;
1438 }
1439
1440 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1441 {
1442     IscsiLun *iscsilun = bs->opaque;
1443     bdi->unallocated_blocks_are_zero = !!iscsilun->lbprz;
1444     bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
1445     /* Guess the internal cluster (page) size of the iscsi target by the means
1446      * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1447      * reasonable size for bdi->cluster_size */
1448     if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 64 * 1024 &&
1449         iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1450         bdi->cluster_size = iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
1451     }
1452     return 0;
1453 }
1454
1455 static QEMUOptionParameter iscsi_create_options[] = {
1456     {
1457         .name = BLOCK_OPT_SIZE,
1458         .type = OPT_SIZE,
1459         .help = "Virtual disk size"
1460     },
1461     { NULL }
1462 };
1463
1464 static BlockDriver bdrv_iscsi = {
1465     .format_name     = "iscsi",
1466     .protocol_name   = "iscsi",
1467
1468     .instance_size   = sizeof(IscsiLun),
1469     .bdrv_needs_filename = true,
1470     .bdrv_file_open  = iscsi_open,
1471     .bdrv_close      = iscsi_close,
1472     .bdrv_create     = iscsi_create,
1473     .create_options  = iscsi_create_options,
1474     .bdrv_reopen_prepare  = iscsi_reopen_prepare,
1475
1476     .bdrv_getlength  = iscsi_getlength,
1477     .bdrv_get_info   = iscsi_get_info,
1478     .bdrv_truncate   = iscsi_truncate,
1479     .bdrv_refresh_limits = iscsi_refresh_limits,
1480
1481 #if defined(LIBISCSI_FEATURE_IOVECTOR)
1482     .bdrv_co_get_block_status = iscsi_co_get_block_status,
1483 #endif
1484     .bdrv_co_discard      = iscsi_co_discard,
1485 #if defined(SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED)
1486     .bdrv_co_write_zeroes = iscsi_co_write_zeroes,
1487 #endif
1488     .bdrv_co_readv         = iscsi_co_readv,
1489     .bdrv_co_writev        = iscsi_co_writev,
1490     .bdrv_co_flush_to_disk = iscsi_co_flush,
1491
1492 #ifdef __linux__
1493     .bdrv_ioctl       = iscsi_ioctl,
1494     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1495 #endif
1496 };
1497
1498 static QemuOptsList qemu_iscsi_opts = {
1499     .name = "iscsi",
1500     .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1501     .desc = {
1502         {
1503             .name = "user",
1504             .type = QEMU_OPT_STRING,
1505             .help = "username for CHAP authentication to target",
1506         },{
1507             .name = "password",
1508             .type = QEMU_OPT_STRING,
1509             .help = "password for CHAP authentication to target",
1510         },{
1511             .name = "header-digest",
1512             .type = QEMU_OPT_STRING,
1513             .help = "HeaderDigest setting. "
1514                     "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1515         },{
1516             .name = "initiator-name",
1517             .type = QEMU_OPT_STRING,
1518             .help = "Initiator iqn name to use when connecting",
1519         },
1520         { /* end of list */ }
1521     },
1522 };
1523
1524 static void iscsi_block_init(void)
1525 {
1526     bdrv_register(&bdrv_iscsi);
1527     qemu_add_opts(&qemu_iscsi_opts);
1528 }
1529
1530 block_init(iscsi_block_init);