block/iscsi.c

   1 /*
   2  * QEMU Block driver for iSCSI images
   3  *
   4  * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25 #include "config-host.h"
  26
  27 #include <poll.h>
  28 #include <arpa/inet.h>
  29 #include "qemu-common.h"
  30 #include "qemu/config-file.h"
  31 #include "qemu/error-report.h"
  32 #include "block/block_int.h"
  33 #include "trace.h"
  34 #include "block/scsi.h"
  35 #include "qemu/iov.h"
  36 #include "sysemu/sysemu.h"
  37 #include "qmp-commands.h"
  38
  39 #include <iscsi/iscsi.h>
  40 #include <iscsi/scsi-lowlevel.h>
  41
  42 #ifdef __linux__
  43 #include <scsi/sg.h>
  44 #include <block/scsi.h>
  45 #endif
  46
  47 typedef struct IscsiLun {
  48     struct iscsi_context *iscsi;
  49     int lun;
  50     enum scsi_inquiry_peripheral_device_type type;
  51     int block_size;
  52     uint64_t num_blocks;
  53     int events;
  54     QEMUTimer *nop_timer;
  55     uint8_t lbpme;
  56     uint8_t lbprz;
  57     struct scsi_inquiry_logical_block_provisioning lbp;
  58     struct scsi_inquiry_block_limits bl;
  59 } IscsiLun;
  60
  61 typedef struct IscsiTask {
  62     int status;
  63     int complete;
  64     int retries;
  65     int do_retry;
  66     struct scsi_task *task;
  67     Coroutine *co;
  68 } IscsiTask;
  69
  70 typedef struct IscsiAIOCB {
  71     BlockDriverAIOCB common;
  72     QEMUIOVector *qiov;
  73     QEMUBH *bh;
  74     IscsiLun *iscsilun;
  75     struct scsi_task *task;
  76     uint8_t *buf;
  77     int status;
  78     int canceled;
  79     int retries;
  80     int64_t sector_num;
  81     int nb_sectors;
  82 #ifdef __linux__
  83     sg_io_hdr_t *ioh;
  84 #endif
  85 } IscsiAIOCB;
  86
  87 #define NOP_INTERVAL 5000
  88 #define MAX_NOP_FAILURES 3
  89 #define ISCSI_CMD_RETRIES 5
  90 #define ISCSI_MAX_UNMAP 131072
  91
  92 static void
  93 iscsi_bh_cb(void *p)
  94 {
  95     IscsiAIOCB *acb = p;
  96
  97     qemu_bh_delete(acb->bh);
  98
  99     g_free(acb->buf);
 100     acb->buf = NULL;
 101
 102     if (acb->canceled == 0) {
 103         acb->common.cb(acb->common.opaque, acb->status);
 104     }
 105
 106     if (acb->task != NULL) {
 107         scsi_free_scsi_task(acb->task);
 108         acb->task = NULL;
 109     }
 110
 111     qemu_aio_release(acb);
 112 }
 113
 114 static void
 115 iscsi_schedule_bh(IscsiAIOCB *acb)
 116 {
 117     if (acb->bh) {
 118         return;
 119     }
 120     acb->bh = qemu_bh_new(iscsi_bh_cb, acb);
 121     qemu_bh_schedule(acb->bh);
 122 }
 123
 124 static void
 125 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
 126                         void *command_data, void *opaque)
 127 {
 128     struct IscsiTask *iTask = opaque;
 129     struct scsi_task *task = command_data;
 130
 131     iTask->complete = 1;
 132     iTask->status = status;
 133     iTask->do_retry = 0;
 134     iTask->task = task;
 135
 136     if (iTask->retries-- > 0 && status == SCSI_STATUS_CHECK_CONDITION
 137         && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
 138         iTask->do_retry = 1;
 139         goto out;
 140     }
 141
 142     if (status != SCSI_STATUS_GOOD) {
 143         error_report("iSCSI: Failure. %s", iscsi_get_error(iscsi));
 144     }
 145
 146 out:
 147     if (iTask->co) {
 148         qemu_coroutine_enter(iTask->co, NULL);
 149     }
 150 }
 151
 152 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
 153 {
 154     *iTask = (struct IscsiTask) {
 155         .co         = qemu_coroutine_self(),
 156         .retries    = ISCSI_CMD_RETRIES,
 157     };
 158 }
 159
 160 static void
 161 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
 162                     void *private_data)
 163 {
 164     IscsiAIOCB *acb = private_data;
 165
 166     acb->status = -ECANCELED;
 167     iscsi_schedule_bh(acb);
 168 }
 169
 170 static void
 171 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
 172 {
 173     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
 174     IscsiLun *iscsilun = acb->iscsilun;
 175
 176     if (acb->status != -EINPROGRESS) {
 177         return;
 178     }
 179
 180     acb->canceled = 1;
 181
 182     /* send a task mgmt call to the target to cancel the task on the target */
 183     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
 184                                      iscsi_abort_task_cb, acb);
 185
 186     while (acb->status == -EINPROGRESS) {
 187         qemu_aio_wait();
 188     }
 189 }
 190
 191 static const AIOCBInfo iscsi_aiocb_info = {
 192     .aiocb_size         = sizeof(IscsiAIOCB),
 193     .cancel             = iscsi_aio_cancel,
 194 };
 195
 196
 197 static void iscsi_process_read(void *arg);
 198 static void iscsi_process_write(void *arg);
 199
 200 static void
 201 iscsi_set_events(IscsiLun *iscsilun)
 202 {
 203     struct iscsi_context *iscsi = iscsilun->iscsi;
 204     int ev;
 205
 206     /* We always register a read handler.  */
 207     ev = POLLIN;
 208     ev |= iscsi_which_events(iscsi);
 209     if (ev != iscsilun->events) {
 210         qemu_aio_set_fd_handler(iscsi_get_fd(iscsi),
 211                       iscsi_process_read,
 212                       (ev & POLLOUT) ? iscsi_process_write : NULL,
 213                       iscsilun);
 214
 215     }
 216
 217     iscsilun->events = ev;
 218 }
 219
 220 static void
 221 iscsi_process_read(void *arg)
 222 {
 223     IscsiLun *iscsilun = arg;
 224     struct iscsi_context *iscsi = iscsilun->iscsi;
 225
 226     iscsi_service(iscsi, POLLIN);
 227     iscsi_set_events(iscsilun);
 228 }
 229
 230 static void
 231 iscsi_process_write(void *arg)
 232 {
 233     IscsiLun *iscsilun = arg;
 234     struct iscsi_context *iscsi = iscsilun->iscsi;
 235
 236     iscsi_service(iscsi, POLLOUT);
 237     iscsi_set_events(iscsilun);
 238 }
 239
 240 static int
 241 iscsi_aio_writev_acb(IscsiAIOCB *acb);
 242
 243 static void
 244 iscsi_aio_write16_cb(struct iscsi_context *iscsi, int status,
 245                      void *command_data, void *opaque)
 246 {
 247     IscsiAIOCB *acb = opaque;
 248
 249     trace_iscsi_aio_write16_cb(iscsi, status, acb, acb->canceled);
 250
 251     g_free(acb->buf);
 252     acb->buf = NULL;
 253
 254     if (acb->canceled != 0) {
 255         return;
 256     }
 257
 258     acb->status = 0;
 259     if (status != 0) {
 260         if (status == SCSI_STATUS_CHECK_CONDITION
 261             && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
 262             && acb->retries-- > 0) {
 263             scsi_free_scsi_task(acb->task);
 264             acb->task = NULL;
 265             if (iscsi_aio_writev_acb(acb) == 0) {
 266                 iscsi_set_events(acb->iscsilun);
 267                 return;
 268             }
 269         }
 270         error_report("Failed to write16 data to iSCSI lun. %s",
 271                      iscsi_get_error(iscsi));
 272         acb->status = -EIO;
 273     }
 274
 275     iscsi_schedule_bh(acb);
 276 }
 277
 278 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
 279 {
 280     return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
 281 }
 282
 283 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
 284 {
 285     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
 286 }
 287
 288 static bool is_request_lun_aligned(int64_t sector_num, int nb_sectors,
 289                                       IscsiLun *iscsilun)
 290 {
 291     if ((sector_num * BDRV_SECTOR_SIZE) % iscsilun->block_size ||
 292         (nb_sectors * BDRV_SECTOR_SIZE) % iscsilun->block_size) {
 293             error_report("iSCSI misaligned request: "
 294                          "iscsilun->block_size %u, sector_num %" PRIi64
 295                          ", nb_sectors %d",
 296                          iscsilun->block_size, sector_num, nb_sectors);
 297             return 0;
 298     }
 299     return 1;
 300 }
 301
 302 static int
 303 iscsi_aio_writev_acb(IscsiAIOCB *acb)
 304 {
 305     struct iscsi_context *iscsi = acb->iscsilun->iscsi;
 306     size_t size;
 307     uint32_t num_sectors;
 308     uint64_t lba;
 309 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 310     struct iscsi_data data;
 311 #endif
 312     int ret;
 313
 314     acb->canceled   = 0;
 315     acb->bh         = NULL;
 316     acb->status     = -EINPROGRESS;
 317     acb->buf        = NULL;
 318
 319     /* this will allow us to get rid of 'buf' completely */
 320     size = acb->nb_sectors * BDRV_SECTOR_SIZE;
 321
 322 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 323     data.size = MIN(size, acb->qiov->size);
 324
 325     /* if the iovec only contains one buffer we can pass it directly */
 326     if (acb->qiov->niov == 1) {
 327         data.data = acb->qiov->iov[0].iov_base;
 328     } else {
 329         acb->buf = g_malloc(data.size);
 330         qemu_iovec_to_buf(acb->qiov, 0, acb->buf, data.size);
 331         data.data = acb->buf;
 332     }
 333 #endif
 334
 335     acb->task = malloc(sizeof(struct scsi_task));
 336     if (acb->task == NULL) {
 337         error_report("iSCSI: Failed to allocate task for scsi WRITE16 "
 338                      "command. %s", iscsi_get_error(iscsi));
 339         return -1;
 340     }
 341     memset(acb->task, 0, sizeof(struct scsi_task));
 342
 343     acb->task->xfer_dir = SCSI_XFER_WRITE;
 344     acb->task->cdb_size = 16;
 345     acb->task->cdb[0] = 0x8a;
 346     lba = sector_qemu2lun(acb->sector_num, acb->iscsilun);
 347     *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 348     *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 349     num_sectors = sector_qemu2lun(acb->nb_sectors, acb->iscsilun);
 350     *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 351     acb->task->expxferlen = size;
 352
 353 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 354     ret = iscsi_scsi_command_async(iscsi, acb->iscsilun->lun, acb->task,
 355                                    iscsi_aio_write16_cb,
 356                                    NULL,
 357                                    acb);
 358 #else
 359     ret = iscsi_scsi_command_async(iscsi, acb->iscsilun->lun, acb->task,
 360                                    iscsi_aio_write16_cb,
 361                                    &data,
 362                                    acb);
 363 #endif
 364     if (ret != 0) {
 365         scsi_free_scsi_task(acb->task);
 366         g_free(acb->buf);
 367         return -1;
 368     }
 369
 370 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 371     scsi_task_set_iov_out(acb->task, (struct scsi_iovec*) acb->qiov->iov, acb->qiov->niov);
 372 #endif
 373
 374     return 0;
 375 }
 376
 377 static BlockDriverAIOCB *
 378 iscsi_aio_writev(BlockDriverState *bs, int64_t sector_num,
 379                  QEMUIOVector *qiov, int nb_sectors,
 380                  BlockDriverCompletionFunc *cb,
 381                  void *opaque)
 382 {
 383     IscsiLun *iscsilun = bs->opaque;
 384     IscsiAIOCB *acb;
 385
 386     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 387         return NULL;
 388     }
 389
 390     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 391     trace_iscsi_aio_writev(iscsilun->iscsi, sector_num, nb_sectors, opaque, acb);
 392
 393     acb->iscsilun    = iscsilun;
 394     acb->qiov        = qiov;
 395     acb->nb_sectors  = nb_sectors;
 396     acb->sector_num  = sector_num;
 397     acb->retries     = ISCSI_CMD_RETRIES;
 398
 399     if (iscsi_aio_writev_acb(acb) != 0) {
 400         qemu_aio_release(acb);
 401         return NULL;
 402     }
 403
 404     iscsi_set_events(iscsilun);
 405     return &acb->common;
 406 }
 407
 408 static int
 409 iscsi_aio_readv_acb(IscsiAIOCB *acb);
 410
 411 static void
 412 iscsi_aio_read16_cb(struct iscsi_context *iscsi, int status,
 413                     void *command_data, void *opaque)
 414 {
 415     IscsiAIOCB *acb = opaque;
 416
 417     trace_iscsi_aio_read16_cb(iscsi, status, acb, acb->canceled);
 418
 419     if (acb->canceled != 0) {
 420         return;
 421     }
 422
 423     acb->status = 0;
 424     if (status != 0) {
 425         if (status == SCSI_STATUS_CHECK_CONDITION
 426             && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
 427             && acb->retries-- > 0) {
 428             scsi_free_scsi_task(acb->task);
 429             acb->task = NULL;
 430             if (iscsi_aio_readv_acb(acb) == 0) {
 431                 iscsi_set_events(acb->iscsilun);
 432                 return;
 433             }
 434         }
 435         error_report("Failed to read16 data from iSCSI lun. %s",
 436                      iscsi_get_error(iscsi));
 437         acb->status = -EIO;
 438     }
 439
 440     iscsi_schedule_bh(acb);
 441 }
 442
 443 static int
 444 iscsi_aio_readv_acb(IscsiAIOCB *acb)
 445 {
 446     struct iscsi_context *iscsi = acb->iscsilun->iscsi;
 447     size_t size;
 448     uint64_t lba;
 449     uint32_t num_sectors;
 450     int ret;
 451 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 452     int i;
 453 #endif
 454
 455     acb->canceled    = 0;
 456     acb->bh          = NULL;
 457     acb->status      = -EINPROGRESS;
 458     acb->buf         = NULL;
 459
 460     size = acb->nb_sectors * BDRV_SECTOR_SIZE;
 461
 462     acb->task = malloc(sizeof(struct scsi_task));
 463     if (acb->task == NULL) {
 464         error_report("iSCSI: Failed to allocate task for scsi READ16 "
 465                      "command. %s", iscsi_get_error(iscsi));
 466         return -1;
 467     }
 468     memset(acb->task, 0, sizeof(struct scsi_task));
 469
 470     acb->task->xfer_dir = SCSI_XFER_READ;
 471     acb->task->expxferlen = size;
 472     lba = sector_qemu2lun(acb->sector_num, acb->iscsilun);
 473     num_sectors = sector_qemu2lun(acb->nb_sectors, acb->iscsilun);
 474
 475     switch (acb->iscsilun->type) {
 476     case TYPE_DISK:
 477         acb->task->cdb_size = 16;
 478         acb->task->cdb[0]  = 0x88;
 479         *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 480         *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 481         *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 482         break;
 483     default:
 484         acb->task->cdb_size = 10;
 485         acb->task->cdb[0]  = 0x28;
 486         *(uint32_t *)&acb->task->cdb[2] = htonl(lba);
 487         *(uint16_t *)&acb->task->cdb[7] = htons(num_sectors);
 488         break;
 489     }
 490
 491     ret = iscsi_scsi_command_async(iscsi, acb->iscsilun->lun, acb->task,
 492                                    iscsi_aio_read16_cb,
 493                                    NULL,
 494                                    acb);
 495     if (ret != 0) {
 496         scsi_free_scsi_task(acb->task);
 497         return -1;
 498     }
 499
 500 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 501     scsi_task_set_iov_in(acb->task, (struct scsi_iovec*) acb->qiov->iov, acb->qiov->niov);
 502 #else
 503     for (i = 0; i < acb->qiov->niov; i++) {
 504         scsi_task_add_data_in_buffer(acb->task,
 505                 acb->qiov->iov[i].iov_len,
 506                 acb->qiov->iov[i].iov_base);
 507     }
 508 #endif
 509     return 0;
 510 }
 511
 512 static BlockDriverAIOCB *
 513 iscsi_aio_readv(BlockDriverState *bs, int64_t sector_num,
 514                 QEMUIOVector *qiov, int nb_sectors,
 515                 BlockDriverCompletionFunc *cb,
 516                 void *opaque)
 517 {
 518     IscsiLun *iscsilun = bs->opaque;
 519     IscsiAIOCB *acb;
 520
 521     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 522         return NULL;
 523     }
 524
 525     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 526     trace_iscsi_aio_readv(iscsilun->iscsi, sector_num, nb_sectors, opaque, acb);
 527
 528     acb->nb_sectors  = nb_sectors;
 529     acb->sector_num  = sector_num;
 530     acb->iscsilun    = iscsilun;
 531     acb->qiov        = qiov;
 532     acb->retries     = ISCSI_CMD_RETRIES;
 533
 534     if (iscsi_aio_readv_acb(acb) != 0) {
 535         qemu_aio_release(acb);
 536         return NULL;
 537     }
 538
 539     iscsi_set_events(iscsilun);
 540     return &acb->common;
 541 }
 542
 543 static int
 544 iscsi_aio_flush_acb(IscsiAIOCB *acb);
 545
 546 static void
 547 iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
 548                      void *command_data, void *opaque)
 549 {
 550     IscsiAIOCB *acb = opaque;
 551
 552     if (acb->canceled != 0) {
 553         return;
 554     }
 555
 556     acb->status = 0;
 557     if (status != 0) {
 558         if (status == SCSI_STATUS_CHECK_CONDITION
 559             && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
 560             && acb->retries-- > 0) {
 561             scsi_free_scsi_task(acb->task);
 562             acb->task = NULL;
 563             if (iscsi_aio_flush_acb(acb) == 0) {
 564                 iscsi_set_events(acb->iscsilun);
 565                 return;
 566             }
 567         }
 568         error_report("Failed to sync10 data on iSCSI lun. %s",
 569                      iscsi_get_error(iscsi));
 570         acb->status = -EIO;
 571     }
 572
 573     iscsi_schedule_bh(acb);
 574 }
 575
 576 static int
 577 iscsi_aio_flush_acb(IscsiAIOCB *acb)
 578 {
 579     struct iscsi_context *iscsi = acb->iscsilun->iscsi;
 580
 581     acb->canceled   = 0;
 582     acb->bh         = NULL;
 583     acb->status     = -EINPROGRESS;
 584     acb->buf        = NULL;
 585
 586     acb->task = iscsi_synchronizecache10_task(iscsi, acb->iscsilun->lun,
 587                                          0, 0, 0, 0,
 588                                          iscsi_synccache10_cb,
 589                                          acb);
 590     if (acb->task == NULL) {
 591         error_report("iSCSI: Failed to send synchronizecache10 command. %s",
 592                      iscsi_get_error(iscsi));
 593         return -1;
 594     }
 595
 596     return 0;
 597 }
 598
 599 static BlockDriverAIOCB *
 600 iscsi_aio_flush(BlockDriverState *bs,
 601                 BlockDriverCompletionFunc *cb, void *opaque)
 602 {
 603     IscsiLun *iscsilun = bs->opaque;
 604
 605     IscsiAIOCB *acb;
 606
 607     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 608
 609     acb->iscsilun    = iscsilun;
 610     acb->retries     = ISCSI_CMD_RETRIES;
 611
 612     if (iscsi_aio_flush_acb(acb) != 0) {
 613         qemu_aio_release(acb);
 614         return NULL;
 615     }
 616
 617     iscsi_set_events(iscsilun);
 618
 619     return &acb->common;
 620 }
 621
 622 #ifdef __linux__
 623 static void
 624 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 625                      void *command_data, void *opaque)
 626 {
 627     IscsiAIOCB *acb = opaque;
 628
 629     g_free(acb->buf);
 630     acb->buf = NULL;
 631
 632     if (acb->canceled != 0) {
 633         return;
 634     }
 635
 636     acb->status = 0;
 637     if (status < 0) {
 638         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 639                      iscsi_get_error(iscsi));
 640         acb->status = -EIO;
 641     }
 642
 643     acb->ioh->driver_status = 0;
 644     acb->ioh->host_status   = 0;
 645     acb->ioh->resid         = 0;
 646
 647 #define SG_ERR_DRIVER_SENSE    0x08
 648
 649     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 650         int ss;
 651
 652         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 653
 654         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 655         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 656              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 657         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 658     }
 659
 660     iscsi_schedule_bh(acb);
 661 }
 662
 663 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 664         unsigned long int req, void *buf,
 665         BlockDriverCompletionFunc *cb, void *opaque)
 666 {
 667     IscsiLun *iscsilun = bs->opaque;
 668     struct iscsi_context *iscsi = iscsilun->iscsi;
 669     struct iscsi_data data;
 670     IscsiAIOCB *acb;
 671
 672     assert(req == SG_IO);
 673
 674     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 675
 676     acb->iscsilun = iscsilun;
 677     acb->canceled    = 0;
 678     acb->bh          = NULL;
 679     acb->status      = -EINPROGRESS;
 680     acb->buf         = NULL;
 681     acb->ioh         = buf;
 682
 683     acb->task = malloc(sizeof(struct scsi_task));
 684     if (acb->task == NULL) {
 685         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 686                      iscsi_get_error(iscsi));
 687         qemu_aio_release(acb);
 688         return NULL;
 689     }
 690     memset(acb->task, 0, sizeof(struct scsi_task));
 691
 692     switch (acb->ioh->dxfer_direction) {
 693     case SG_DXFER_TO_DEV:
 694         acb->task->xfer_dir = SCSI_XFER_WRITE;
 695         break;
 696     case SG_DXFER_FROM_DEV:
 697         acb->task->xfer_dir = SCSI_XFER_READ;
 698         break;
 699     default:
 700         acb->task->xfer_dir = SCSI_XFER_NONE;
 701         break;
 702     }
 703
 704     acb->task->cdb_size = acb->ioh->cmd_len;
 705     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 706     acb->task->expxferlen = acb->ioh->dxfer_len;
 707
 708     data.size = 0;
 709     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 710         if (acb->ioh->iovec_count == 0) {
 711             data.data = acb->ioh->dxferp;
 712             data.size = acb->ioh->dxfer_len;
 713         } else {
 714 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 715             scsi_task_set_iov_out(acb->task,
 716                                  (struct scsi_iovec *) acb->ioh->dxferp,
 717                                  acb->ioh->iovec_count);
 718 #else
 719             struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
 720
 721             acb->buf = g_malloc(acb->ioh->dxfer_len);
 722             data.data = acb->buf;
 723             data.size = iov_to_buf(iov, acb->ioh->iovec_count, 0,
 724                                    acb->buf, acb->ioh->dxfer_len);
 725 #endif
 726         }
 727     }
 728
 729     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 730                                  iscsi_aio_ioctl_cb,
 731                                  (data.size > 0) ? &data : NULL,
 732                                  acb) != 0) {
 733         scsi_free_scsi_task(acb->task);
 734         qemu_aio_release(acb);
 735         return NULL;
 736     }
 737
 738     /* tell libiscsi to read straight into the buffer we got from ioctl */
 739     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 740         if (acb->ioh->iovec_count == 0) {
 741             scsi_task_add_data_in_buffer(acb->task,
 742                                          acb->ioh->dxfer_len,
 743                                          acb->ioh->dxferp);
 744         } else {
 745 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 746             scsi_task_set_iov_in(acb->task,
 747                                  (struct scsi_iovec *) acb->ioh->dxferp,
 748                                  acb->ioh->iovec_count);
 749 #else
 750             int i;
 751             for (i = 0; i < acb->ioh->iovec_count; i++) {
 752                 struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
 753
 754                 scsi_task_add_data_in_buffer(acb->task,
 755                     iov[i].iov_len,
 756                     iov[i].iov_base);
 757             }
 758 #endif
 759         }
 760     }
 761
 762     iscsi_set_events(iscsilun);
 763
 764     return &acb->common;
 765 }
 766
 767
 768 static void ioctl_cb(void *opaque, int status)
 769 {
 770     int *p_status = opaque;
 771     *p_status = status;
 772 }
 773
 774 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 775 {
 776     IscsiLun *iscsilun = bs->opaque;
 777     int status;
 778
 779     switch (req) {
 780     case SG_GET_VERSION_NUM:
 781         *(int *)buf = 30000;
 782         break;
 783     case SG_GET_SCSI_ID:
 784         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 785         break;
 786     case SG_IO:
 787         status = -EINPROGRESS;
 788         iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status);
 789
 790         while (status == -EINPROGRESS) {
 791             qemu_aio_wait();
 792         }
 793
 794         return 0;
 795     default:
 796         return -1;
 797     }
 798     return 0;
 799 }
 800 #endif
 801
 802 static int64_t
 803 iscsi_getlength(BlockDriverState *bs)
 804 {
 805     IscsiLun *iscsilun = bs->opaque;
 806     int64_t len;
 807
 808     len  = iscsilun->num_blocks;
 809     len *= iscsilun->block_size;
 810
 811     return len;
 812 }
 813
 814 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 815
 816 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
 817                                                   int64_t sector_num,
 818                                                   int nb_sectors, int *pnum)
 819 {
 820     IscsiLun *iscsilun = bs->opaque;
 821     struct scsi_get_lba_status *lbas = NULL;
 822     struct scsi_lba_status_descriptor *lbasd = NULL;
 823     struct IscsiTask iTask;
 824     int64_t ret;
 825
 826     iscsi_co_init_iscsitask(iscsilun, &iTask);
 827
 828     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 829         ret = -EINVAL;
 830         goto out;
 831     }
 832
 833     /* default to all sectors allocated */
 834     ret = BDRV_BLOCK_DATA;
 835     ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
 836     *pnum = nb_sectors;
 837
 838     /* LUN does not support logical block provisioning */
 839     if (iscsilun->lbpme == 0) {
 840         goto out;
 841     }
 842
 843 retry:
 844     if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
 845                                   sector_qemu2lun(sector_num, iscsilun),
 846                                   8 + 16, iscsi_co_generic_cb,
 847                                   &iTask) == NULL) {
 848         ret = -EIO;
 849         goto out;
 850     }
 851
 852     while (!iTask.complete) {
 853         iscsi_set_events(iscsilun);
 854         qemu_coroutine_yield();
 855     }
 856
 857     if (iTask.do_retry) {
 858         if (iTask.task != NULL) {
 859             scsi_free_scsi_task(iTask.task);
 860             iTask.task = NULL;
 861         }
 862         goto retry;
 863     }
 864
 865     if (iTask.status != SCSI_STATUS_GOOD) {
 866         /* in case the get_lba_status_callout fails (i.e.
 867          * because the device is busy or the cmd is not
 868          * supported) we pretend all blocks are allocated
 869          * for backwards compatibility */
 870         goto out;
 871     }
 872
 873     lbas = scsi_datain_unmarshall(iTask.task);
 874     if (lbas == NULL) {
 875         ret = -EIO;
 876         goto out;
 877     }
 878
 879     lbasd = &lbas->descriptors[0];
 880
 881     if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
 882         ret = -EIO;
 883         goto out;
 884     }
 885
 886     *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
 887     if (*pnum > nb_sectors) {
 888         *pnum = nb_sectors;
 889     }
 890
 891     if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
 892         lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
 893         ret &= ~BDRV_BLOCK_DATA;
 894         if (iscsilun->lbprz) {
 895             ret |= BDRV_BLOCK_ZERO;
 896         }
 897     }
 898
 899 out:
 900     if (iTask.task != NULL) {
 901         scsi_free_scsi_task(iTask.task);
 902     }
 903     return ret;
 904 }
 905
 906 #endif /* LIBISCSI_FEATURE_IOVECTOR */
 907
 908 static int
 909 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num,
 910                                    int nb_sectors)
 911 {
 912     IscsiLun *iscsilun = bs->opaque;
 913     struct IscsiTask iTask;
 914     struct unmap_list list;
 915     uint32_t nb_blocks;
 916     uint32_t max_unmap;
 917
 918     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 919         return -EINVAL;
 920     }
 921
 922     if (!iscsilun->lbp.lbpu) {
 923         /* UNMAP is not supported by the target */
 924         return 0;
 925     }
 926
 927     list.lba = sector_qemu2lun(sector_num, iscsilun);
 928     nb_blocks = sector_qemu2lun(nb_sectors, iscsilun);
 929
 930     max_unmap = iscsilun->bl.max_unmap;
 931     if (max_unmap == 0xffffffff) {
 932         max_unmap = ISCSI_MAX_UNMAP;
 933     }
 934
 935     while (nb_blocks > 0) {
 936         iscsi_co_init_iscsitask(iscsilun, &iTask);
 937         list.num = nb_blocks;
 938         if (list.num > max_unmap) {
 939             list.num = max_unmap;
 940         }
 941 retry:
 942         if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
 943                          iscsi_co_generic_cb, &iTask) == NULL) {
 944             return -EIO;
 945         }
 946
 947         while (!iTask.complete) {
 948             iscsi_set_events(iscsilun);
 949             qemu_coroutine_yield();
 950         }
 951
 952         if (iTask.task != NULL) {
 953             scsi_free_scsi_task(iTask.task);
 954             iTask.task = NULL;
 955         }
 956
 957         if (iTask.do_retry) {
 958             goto retry;
 959         }
 960
 961         if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
 962             /* the target might fail with a check condition if it
 963                is not happy with the alignment of the UNMAP request
 964                we silently fail in this case */
 965             return 0;
 966         }
 967
 968         if (iTask.status != SCSI_STATUS_GOOD) {
 969             return -EIO;
 970         }
 971
 972         list.lba += list.num;
 973         nb_blocks -= list.num;
 974     }
 975
 976     return 0;
 977 }
 978
 979 static int parse_chap(struct iscsi_context *iscsi, const char *target)
 980 {
 981     QemuOptsList *list;
 982     QemuOpts *opts;
 983     const char *user = NULL;
 984     const char *password = NULL;
 985
 986     list = qemu_find_opts("iscsi");
 987     if (!list) {
 988         return 0;
 989     }
 990
 991     opts = qemu_opts_find(list, target);
 992     if (opts == NULL) {
 993         opts = QTAILQ_FIRST(&list->head);
 994         if (!opts) {
 995             return 0;
 996         }
 997     }
 998
 999     user = qemu_opt_get(opts, "user");
1000     if (!user) {
1001         return 0;
1002     }
1003
1004     password = qemu_opt_get(opts, "password");
1005     if (!password) {
1006         error_report("CHAP username specified but no password was given");
1007         return -1;
1008     }
1009
1010     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1011         error_report("Failed to set initiator username and password");
1012         return -1;
1013     }
1014
1015     return 0;
1016 }
1017
1018 static void parse_header_digest(struct iscsi_context *iscsi, const char *target)
1019 {
1020     QemuOptsList *list;
1021     QemuOpts *opts;
1022     const char *digest = NULL;
1023
1024     list = qemu_find_opts("iscsi");
1025     if (!list) {
1026         return;
1027     }
1028
1029     opts = qemu_opts_find(list, target);
1030     if (opts == NULL) {
1031         opts = QTAILQ_FIRST(&list->head);
1032         if (!opts) {
1033             return;
1034         }
1035     }
1036
1037     digest = qemu_opt_get(opts, "header-digest");
1038     if (!digest) {
1039         return;
1040     }
1041
1042     if (!strcmp(digest, "CRC32C")) {
1043         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1044     } else if (!strcmp(digest, "NONE")) {
1045         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1046     } else if (!strcmp(digest, "CRC32C-NONE")) {
1047         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1048     } else if (!strcmp(digest, "NONE-CRC32C")) {
1049         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1050     } else {
1051         error_report("Invalid header-digest setting : %s", digest);
1052     }
1053 }
1054
1055 static char *parse_initiator_name(const char *target)
1056 {
1057     QemuOptsList *list;
1058     QemuOpts *opts;
1059     const char *name;
1060     char *iscsi_name;
1061     UuidInfo *uuid_info;
1062
1063     list = qemu_find_opts("iscsi");
1064     if (list) {
1065         opts = qemu_opts_find(list, target);
1066         if (!opts) {
1067             opts = QTAILQ_FIRST(&list->head);
1068         }
1069         if (opts) {
1070             name = qemu_opt_get(opts, "initiator-name");
1071             if (name) {
1072                 return g_strdup(name);
1073             }
1074         }
1075     }
1076
1077     uuid_info = qmp_query_uuid(NULL);
1078     if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1079         name = qemu_get_vm_name();
1080     } else {
1081         name = uuid_info->UUID;
1082     }
1083     iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1084                                  name ? ":" : "", name ? name : "");
1085     qapi_free_UuidInfo(uuid_info);
1086     return iscsi_name;
1087 }
1088
1089 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
1090 static void iscsi_nop_timed_event(void *opaque)
1091 {
1092     IscsiLun *iscsilun = opaque;
1093
1094     if (iscsi_get_nops_in_flight(iscsilun->iscsi) > MAX_NOP_FAILURES) {
1095         error_report("iSCSI: NOP timeout. Reconnecting...");
1096         iscsi_reconnect(iscsilun->iscsi);
1097     }
1098
1099     if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1100         error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1101         return;
1102     }
1103
1104     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1105     iscsi_set_events(iscsilun);
1106 }
1107 #endif
1108
1109 static int iscsi_readcapacity_sync(IscsiLun *iscsilun)
1110 {
1111     struct scsi_task *task = NULL;
1112     struct scsi_readcapacity10 *rc10 = NULL;
1113     struct scsi_readcapacity16 *rc16 = NULL;
1114     int ret = 0;
1115     int retries = ISCSI_CMD_RETRIES;
1116
1117     do {
1118         if (task != NULL) {
1119             scsi_free_scsi_task(task);
1120             task = NULL;
1121         }
1122
1123         switch (iscsilun->type) {
1124         case TYPE_DISK:
1125             task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1126             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1127                 rc16 = scsi_datain_unmarshall(task);
1128                 if (rc16 == NULL) {
1129                     error_report("iSCSI: Failed to unmarshall readcapacity16 data.");
1130                     ret = -EINVAL;
1131                 } else {
1132                     iscsilun->block_size = rc16->block_length;
1133                     iscsilun->num_blocks = rc16->returned_lba + 1;
1134                     iscsilun->lbpme = rc16->lbpme;
1135                     iscsilun->lbprz = rc16->lbprz;
1136                 }
1137             }
1138             break;
1139         case TYPE_ROM:
1140             task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1141             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1142                 rc10 = scsi_datain_unmarshall(task);
1143                 if (rc10 == NULL) {
1144                     error_report("iSCSI: Failed to unmarshall readcapacity10 data.");
1145                     ret = -EINVAL;
1146                 } else {
1147                     iscsilun->block_size = rc10->block_size;
1148                     if (rc10->lba == 0) {
1149                         /* blank disk loaded */
1150                         iscsilun->num_blocks = 0;
1151                     } else {
1152                         iscsilun->num_blocks = rc10->lba + 1;
1153                     }
1154                 }
1155             }
1156             break;
1157         default:
1158             return 0;
1159         }
1160     } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1161              && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1162              && retries-- > 0);
1163
1164     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1165         error_report("iSCSI: failed to send readcapacity10 command.");
1166         ret = -EINVAL;
1167     }
1168     if (task) {
1169         scsi_free_scsi_task(task);
1170     }
1171     return ret;
1172 }
1173
1174 /* TODO Convert to fine grained options */
1175 static QemuOptsList runtime_opts = {
1176     .name = "iscsi",
1177     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1178     .desc = {
1179         {
1180             .name = "filename",
1181             .type = QEMU_OPT_STRING,
1182             .help = "URL to the iscsi image",
1183         },
1184         { /* end of list */ }
1185     },
1186 };
1187
1188 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi,
1189                                           int lun, int evpd, int pc) {
1190         int full_size;
1191         struct scsi_task *task = NULL;
1192         task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1193         if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1194             goto fail;
1195         }
1196         full_size = scsi_datain_getfullsize(task);
1197         if (full_size > task->datain.size) {
1198             scsi_free_scsi_task(task);
1199
1200             /* we need more data for the full list */
1201             task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1202             if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1203                 goto fail;
1204             }
1205         }
1206
1207         return task;
1208
1209 fail:
1210         error_report("iSCSI: Inquiry command failed : %s",
1211                      iscsi_get_error(iscsi));
1212         if (task) {
1213             scsi_free_scsi_task(task);
1214             return NULL;
1215         }
1216         return NULL;
1217 }
1218
1219 /*
1220  * We support iscsi url's on the form
1221  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1222  */
1223 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1224                       Error **errp)
1225 {
1226     IscsiLun *iscsilun = bs->opaque;
1227     struct iscsi_context *iscsi = NULL;
1228     struct iscsi_url *iscsi_url = NULL;
1229     struct scsi_task *task = NULL;
1230     struct scsi_inquiry_standard *inq = NULL;
1231     char *initiator_name = NULL;
1232     QemuOpts *opts;
1233     Error *local_err = NULL;
1234     const char *filename;
1235     int ret;
1236
1237     if ((BDRV_SECTOR_SIZE % 512) != 0) {
1238         error_report("iSCSI: Invalid BDRV_SECTOR_SIZE. "
1239                      "BDRV_SECTOR_SIZE(%lld) is not a multiple "
1240                      "of 512", BDRV_SECTOR_SIZE);
1241         return -EINVAL;
1242     }
1243
1244     opts = qemu_opts_create_nofail(&runtime_opts);
1245     qemu_opts_absorb_qdict(opts, options, &local_err);
1246     if (error_is_set(&local_err)) {
1247         qerror_report_err(local_err);
1248         error_free(local_err);
1249         ret = -EINVAL;
1250         goto out;
1251     }
1252
1253     filename = qemu_opt_get(opts, "filename");
1254
1255
1256     iscsi_url = iscsi_parse_full_url(iscsi, filename);
1257     if (iscsi_url == NULL) {
1258         error_report("Failed to parse URL : %s", filename);
1259         ret = -EINVAL;
1260         goto out;
1261     }
1262
1263     memset(iscsilun, 0, sizeof(IscsiLun));
1264
1265     initiator_name = parse_initiator_name(iscsi_url->target);
1266
1267     iscsi = iscsi_create_context(initiator_name);
1268     if (iscsi == NULL) {
1269         error_report("iSCSI: Failed to create iSCSI context.");
1270         ret = -ENOMEM;
1271         goto out;
1272     }
1273
1274     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1275         error_report("iSCSI: Failed to set target name.");
1276         ret = -EINVAL;
1277         goto out;
1278     }
1279
1280     if (iscsi_url->user != NULL) {
1281         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1282                                               iscsi_url->passwd);
1283         if (ret != 0) {
1284             error_report("Failed to set initiator username and password");
1285             ret = -EINVAL;
1286             goto out;
1287         }
1288     }
1289
1290     /* check if we got CHAP username/password via the options */
1291     if (parse_chap(iscsi, iscsi_url->target) != 0) {
1292         error_report("iSCSI: Failed to set CHAP user/password");
1293         ret = -EINVAL;
1294         goto out;
1295     }
1296
1297     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1298         error_report("iSCSI: Failed to set session type to normal.");
1299         ret = -EINVAL;
1300         goto out;
1301     }
1302
1303     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1304
1305     /* check if we got HEADER_DIGEST via the options */
1306     parse_header_digest(iscsi, iscsi_url->target);
1307
1308     if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1309         error_report("iSCSI: Failed to connect to LUN : %s",
1310             iscsi_get_error(iscsi));
1311         ret = -EINVAL;
1312         goto out;
1313     }
1314
1315     iscsilun->iscsi = iscsi;
1316     iscsilun->lun   = iscsi_url->lun;
1317
1318     task = iscsi_inquiry_sync(iscsi, iscsilun->lun, 0, 0, 36);
1319
1320     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1321         error_report("iSCSI: failed to send inquiry command.");
1322         ret = -EINVAL;
1323         goto out;
1324     }
1325
1326     inq = scsi_datain_unmarshall(task);
1327     if (inq == NULL) {
1328         error_report("iSCSI: Failed to unmarshall inquiry data.");
1329         ret = -EINVAL;
1330         goto out;
1331     }
1332
1333     iscsilun->type = inq->periperal_device_type;
1334
1335     if ((ret = iscsi_readcapacity_sync(iscsilun)) != 0) {
1336         goto out;
1337     }
1338     bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1339
1340     /* Medium changer or tape. We dont have any emulation for this so this must
1341      * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
1342      * to read from the device to guess the image format.
1343      */
1344     if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
1345         iscsilun->type == TYPE_TAPE) {
1346         bs->sg = 1;
1347     }
1348
1349     if (iscsilun->lbpme) {
1350         struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1351         task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1352                                 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING);
1353         if (task == NULL) {
1354             ret = -EINVAL;
1355             goto out;
1356         }
1357         inq_lbp = scsi_datain_unmarshall(task);
1358         if (inq_lbp == NULL) {
1359             error_report("iSCSI: failed to unmarshall inquiry datain blob");
1360             ret = -EINVAL;
1361             goto out;
1362         }
1363         memcpy(&iscsilun->lbp, inq_lbp,
1364                sizeof(struct scsi_inquiry_logical_block_provisioning));
1365         scsi_free_scsi_task(task);
1366         task = NULL;
1367     }
1368
1369     if (iscsilun->lbp.lbpu || iscsilun->lbp.lbpws) {
1370         struct scsi_inquiry_block_limits *inq_bl;
1371         task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1372                                 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS);
1373         if (task == NULL) {
1374             ret = -EINVAL;
1375             goto out;
1376         }
1377         inq_bl = scsi_datain_unmarshall(task);
1378         if (inq_bl == NULL) {
1379             error_report("iSCSI: failed to unmarshall inquiry datain blob");
1380             ret = -EINVAL;
1381             goto out;
1382         }
1383         memcpy(&iscsilun->bl, inq_bl,
1384                sizeof(struct scsi_inquiry_block_limits));
1385         scsi_free_scsi_task(task);
1386         task = NULL;
1387     }
1388
1389 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
1390     /* Set up a timer for sending out iSCSI NOPs */
1391     iscsilun->nop_timer = timer_new_ms(QEMU_CLOCK_REALTIME, iscsi_nop_timed_event, iscsilun);
1392     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1393 #endif
1394
1395 out:
1396     qemu_opts_del(opts);
1397     if (initiator_name != NULL) {
1398         g_free(initiator_name);
1399     }
1400     if (iscsi_url != NULL) {
1401         iscsi_destroy_url(iscsi_url);
1402     }
1403     if (task != NULL) {
1404         scsi_free_scsi_task(task);
1405     }
1406
1407     if (ret) {
1408         if (iscsi != NULL) {
1409             iscsi_destroy_context(iscsi);
1410         }
1411         memset(iscsilun, 0, sizeof(IscsiLun));
1412     }
1413     return ret;
1414 }
1415
1416 static void iscsi_close(BlockDriverState *bs)
1417 {
1418     IscsiLun *iscsilun = bs->opaque;
1419     struct iscsi_context *iscsi = iscsilun->iscsi;
1420
1421     if (iscsilun->nop_timer) {
1422         timer_del(iscsilun->nop_timer);
1423         timer_free(iscsilun->nop_timer);
1424     }
1425     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL);
1426     iscsi_destroy_context(iscsi);
1427     memset(iscsilun, 0, sizeof(IscsiLun));
1428 }
1429
1430 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1431 {
1432     IscsiLun *iscsilun = bs->opaque;
1433     int ret = 0;
1434
1435     if (iscsilun->type != TYPE_DISK) {
1436         return -ENOTSUP;
1437     }
1438
1439     if ((ret = iscsi_readcapacity_sync(iscsilun)) != 0) {
1440         return ret;
1441     }
1442
1443     if (offset > iscsi_getlength(bs)) {
1444         return -EINVAL;
1445     }
1446
1447     return 0;
1448 }
1449
1450 static int iscsi_has_zero_init(BlockDriverState *bs)
1451 {
1452     return 0;
1453 }
1454
1455 static int iscsi_create(const char *filename, QEMUOptionParameter *options,
1456                         Error **errp)
1457 {
1458     int ret = 0;
1459     int64_t total_size = 0;
1460     BlockDriverState *bs;
1461     IscsiLun *iscsilun = NULL;
1462     QDict *bs_options;
1463
1464     bs = bdrv_new("");
1465
1466     /* Read out options */
1467     while (options && options->name) {
1468         if (!strcmp(options->name, "size")) {
1469             total_size = options->value.n / BDRV_SECTOR_SIZE;
1470         }
1471         options++;
1472     }
1473
1474     bs->opaque = g_malloc0(sizeof(struct IscsiLun));
1475     iscsilun = bs->opaque;
1476
1477     bs_options = qdict_new();
1478     qdict_put(bs_options, "filename", qstring_from_str(filename));
1479     ret = iscsi_open(bs, bs_options, 0, NULL);
1480     QDECREF(bs_options);
1481
1482     if (ret != 0) {
1483         goto out;
1484     }
1485     if (iscsilun->nop_timer) {
1486         timer_del(iscsilun->nop_timer);
1487         timer_free(iscsilun->nop_timer);
1488     }
1489     if (iscsilun->type != TYPE_DISK) {
1490         ret = -ENODEV;
1491         goto out;
1492     }
1493     if (bs->total_sectors < total_size) {
1494         ret = -ENOSPC;
1495         goto out;
1496     }
1497
1498     ret = 0;
1499 out:
1500     if (iscsilun->iscsi != NULL) {
1501         iscsi_destroy_context(iscsilun->iscsi);
1502     }
1503     g_free(bs->opaque);
1504     bs->opaque = NULL;
1505     bdrv_unref(bs);
1506     return ret;
1507 }
1508
1509 static QEMUOptionParameter iscsi_create_options[] = {
1510     {
1511         .name = BLOCK_OPT_SIZE,
1512         .type = OPT_SIZE,
1513         .help = "Virtual disk size"
1514     },
1515     { NULL }
1516 };
1517
1518 static BlockDriver bdrv_iscsi = {
1519     .format_name     = "iscsi",
1520     .protocol_name   = "iscsi",
1521
1522     .instance_size   = sizeof(IscsiLun),
1523     .bdrv_needs_filename = true,
1524     .bdrv_file_open  = iscsi_open,
1525     .bdrv_close      = iscsi_close,
1526     .bdrv_create     = iscsi_create,
1527     .create_options  = iscsi_create_options,
1528
1529     .bdrv_getlength  = iscsi_getlength,
1530     .bdrv_truncate   = iscsi_truncate,
1531
1532 #if defined(LIBISCSI_FEATURE_IOVECTOR)
1533     .bdrv_co_get_block_status = iscsi_co_get_block_status,
1534 #endif
1535     .bdrv_co_discard      = iscsi_co_discard,
1536
1537     .bdrv_aio_readv  = iscsi_aio_readv,
1538     .bdrv_aio_writev = iscsi_aio_writev,
1539     .bdrv_aio_flush  = iscsi_aio_flush,
1540
1541     .bdrv_has_zero_init = iscsi_has_zero_init,
1542
1543 #ifdef __linux__
1544     .bdrv_ioctl       = iscsi_ioctl,
1545     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1546 #endif
1547 };
1548
1549 static QemuOptsList qemu_iscsi_opts = {
1550     .name = "iscsi",
1551     .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1552     .desc = {
1553         {
1554             .name = "user",
1555             .type = QEMU_OPT_STRING,
1556             .help = "username for CHAP authentication to target",
1557         },{
1558             .name = "password",
1559             .type = QEMU_OPT_STRING,
1560             .help = "password for CHAP authentication to target",
1561         },{
1562             .name = "header-digest",
1563             .type = QEMU_OPT_STRING,
1564             .help = "HeaderDigest setting. "
1565                     "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1566         },{
1567             .name = "initiator-name",
1568             .type = QEMU_OPT_STRING,
1569             .help = "Initiator iqn name to use when connecting",
1570         },
1571         { /* end of list */ }
1572     },
1573 };
1574
1575 static void iscsi_block_init(void)
1576 {
1577     bdrv_register(&bdrv_iscsi);
1578     qemu_add_opts(&qemu_iscsi_opts);
1579 }
1580
1581 block_init(iscsi_block_init);