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
  36 #include <iscsi/iscsi.h>
  37 #include <iscsi/scsi-lowlevel.h>
  38
  39 #ifdef __linux__
  40 #include <scsi/sg.h>
  41 #include <block/scsi.h>
  42 #endif
  43
  44 typedef struct IscsiLun {
  45     struct iscsi_context *iscsi;
  46     int lun;
  47     enum scsi_inquiry_peripheral_device_type type;
  48     int block_size;
  49     uint64_t num_blocks;
  50     int events;
  51     QEMUTimer *nop_timer;
  52 } IscsiLun;
  53
  54 typedef struct IscsiAIOCB {
  55     BlockDriverAIOCB common;
  56     QEMUIOVector *qiov;
  57     QEMUBH *bh;
  58     IscsiLun *iscsilun;
  59     struct scsi_task *task;
  60     uint8_t *buf;
  61     int status;
  62     int canceled;
  63     int retries;
  64     size_t read_size;
  65     size_t read_offset;
  66     int64_t sector_num;
  67     int nb_sectors;
  68 #ifdef __linux__
  69     sg_io_hdr_t *ioh;
  70 #endif
  71 } IscsiAIOCB;
  72
  73 #define NOP_INTERVAL 5000
  74 #define MAX_NOP_FAILURES 3
  75 #define ISCSI_CMD_RETRIES 5
  76
  77 static void
  78 iscsi_bh_cb(void *p)
  79 {
  80     IscsiAIOCB *acb = p;
  81
  82     qemu_bh_delete(acb->bh);
  83
  84     g_free(acb->buf);
  85     acb->buf = NULL;
  86
  87     if (acb->canceled == 0) {
  88         acb->common.cb(acb->common.opaque, acb->status);
  89     }
  90
  91     if (acb->task != NULL) {
  92         scsi_free_scsi_task(acb->task);
  93         acb->task = NULL;
  94     }
  95
  96     qemu_aio_release(acb);
  97 }
  98
  99 static void
 100 iscsi_schedule_bh(IscsiAIOCB *acb)
 101 {
 102     if (acb->bh) {
 103         return;
 104     }
 105     acb->bh = qemu_bh_new(iscsi_bh_cb, acb);
 106     qemu_bh_schedule(acb->bh);
 107 }
 108
 109
 110 static void
 111 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
 112                     void *private_data)
 113 {
 114     IscsiAIOCB *acb = private_data;
 115
 116     acb->status = -ECANCELED;
 117     iscsi_schedule_bh(acb);
 118 }
 119
 120 static void
 121 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
 122 {
 123     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
 124     IscsiLun *iscsilun = acb->iscsilun;
 125
 126     if (acb->status != -EINPROGRESS) {
 127         return;
 128     }
 129
 130     acb->canceled = 1;
 131
 132     /* send a task mgmt call to the target to cancel the task on the target */
 133     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
 134                                      iscsi_abort_task_cb, acb);
 135
 136     while (acb->status == -EINPROGRESS) {
 137         qemu_aio_wait();
 138     }
 139 }
 140
 141 static const AIOCBInfo iscsi_aiocb_info = {
 142     .aiocb_size         = sizeof(IscsiAIOCB),
 143     .cancel             = iscsi_aio_cancel,
 144 };
 145
 146
 147 static void iscsi_process_read(void *arg);
 148 static void iscsi_process_write(void *arg);
 149
 150 static int iscsi_process_flush(void *arg)
 151 {
 152     IscsiLun *iscsilun = arg;
 153
 154     return iscsi_queue_length(iscsilun->iscsi) > 0;
 155 }
 156
 157 static void
 158 iscsi_set_events(IscsiLun *iscsilun)
 159 {
 160     struct iscsi_context *iscsi = iscsilun->iscsi;
 161     int ev;
 162
 163     /* We always register a read handler.  */
 164     ev = POLLIN;
 165     ev |= iscsi_which_events(iscsi);
 166     if (ev != iscsilun->events) {
 167         qemu_aio_set_fd_handler(iscsi_get_fd(iscsi),
 168                       iscsi_process_read,
 169                       (ev & POLLOUT) ? iscsi_process_write : NULL,
 170                       iscsi_process_flush,
 171                       iscsilun);
 172
 173     }
 174
 175     iscsilun->events = ev;
 176 }
 177
 178 static void
 179 iscsi_process_read(void *arg)
 180 {
 181     IscsiLun *iscsilun = arg;
 182     struct iscsi_context *iscsi = iscsilun->iscsi;
 183
 184     iscsi_service(iscsi, POLLIN);
 185     iscsi_set_events(iscsilun);
 186 }
 187
 188 static void
 189 iscsi_process_write(void *arg)
 190 {
 191     IscsiLun *iscsilun = arg;
 192     struct iscsi_context *iscsi = iscsilun->iscsi;
 193
 194     iscsi_service(iscsi, POLLOUT);
 195     iscsi_set_events(iscsilun);
 196 }
 197
 198 static int
 199 iscsi_aio_writev_acb(IscsiAIOCB *acb);
 200
 201 static void
 202 iscsi_aio_write16_cb(struct iscsi_context *iscsi, int status,
 203                      void *command_data, void *opaque)
 204 {
 205     IscsiAIOCB *acb = opaque;
 206
 207     trace_iscsi_aio_write16_cb(iscsi, status, acb, acb->canceled);
 208
 209     g_free(acb->buf);
 210     acb->buf = NULL;
 211
 212     if (acb->canceled != 0) {
 213         return;
 214     }
 215
 216     acb->status = 0;
 217     if (status != 0) {
 218         if (status == SCSI_STATUS_CHECK_CONDITION
 219             && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
 220             && acb->retries-- > 0) {
 221             scsi_free_scsi_task(acb->task);
 222             acb->task = NULL;
 223             if (iscsi_aio_writev_acb(acb) == 0) {
 224                 iscsi_set_events(acb->iscsilun);
 225                 return;
 226             }
 227         }
 228         error_report("Failed to write16 data to iSCSI lun. %s",
 229                      iscsi_get_error(iscsi));
 230         acb->status = -EIO;
 231     }
 232
 233     iscsi_schedule_bh(acb);
 234 }
 235
 236 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
 237 {
 238     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
 239 }
 240
 241 static int
 242 iscsi_aio_writev_acb(IscsiAIOCB *acb)
 243 {
 244     struct iscsi_context *iscsi = acb->iscsilun->iscsi;
 245     size_t size;
 246     uint32_t num_sectors;
 247     uint64_t lba;
 248 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 249     struct iscsi_data data;
 250 #endif
 251     int ret;
 252
 253     acb->canceled   = 0;
 254     acb->bh         = NULL;
 255     acb->status     = -EINPROGRESS;
 256     acb->buf        = NULL;
 257
 258     /* this will allow us to get rid of 'buf' completely */
 259     size = acb->nb_sectors * BDRV_SECTOR_SIZE;
 260
 261 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 262     data.size = MIN(size, acb->qiov->size);
 263
 264     /* if the iovec only contains one buffer we can pass it directly */
 265     if (acb->qiov->niov == 1) {
 266         data.data = acb->qiov->iov[0].iov_base;
 267     } else {
 268         acb->buf = g_malloc(data.size);
 269         qemu_iovec_to_buf(acb->qiov, 0, acb->buf, data.size);
 270         data.data = acb->buf;
 271     }
 272 #endif
 273
 274     acb->task = malloc(sizeof(struct scsi_task));
 275     if (acb->task == NULL) {
 276         error_report("iSCSI: Failed to allocate task for scsi WRITE16 "
 277                      "command. %s", iscsi_get_error(iscsi));
 278         return -1;
 279     }
 280     memset(acb->task, 0, sizeof(struct scsi_task));
 281
 282     acb->task->xfer_dir = SCSI_XFER_WRITE;
 283     acb->task->cdb_size = 16;
 284     acb->task->cdb[0] = 0x8a;
 285     lba = sector_qemu2lun(acb->sector_num, acb->iscsilun);
 286     *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 287     *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 288     num_sectors = size / acb->iscsilun->block_size;
 289     *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 290     acb->task->expxferlen = size;
 291
 292 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 293     ret = iscsi_scsi_command_async(iscsi, acb->iscsilun->lun, acb->task,
 294                                    iscsi_aio_write16_cb,
 295                                    NULL,
 296                                    acb);
 297 #else
 298     ret = iscsi_scsi_command_async(iscsi, acb->iscsilun->lun, acb->task,
 299                                    iscsi_aio_write16_cb,
 300                                    &data,
 301                                    acb);
 302 #endif
 303     if (ret != 0) {
 304         scsi_free_scsi_task(acb->task);
 305         g_free(acb->buf);
 306         return -1;
 307     }
 308
 309 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 310     scsi_task_set_iov_out(acb->task, (struct scsi_iovec*) acb->qiov->iov, acb->qiov->niov);
 311 #endif
 312
 313     return 0;
 314 }
 315
 316 static BlockDriverAIOCB *
 317 iscsi_aio_writev(BlockDriverState *bs, int64_t sector_num,
 318                  QEMUIOVector *qiov, int nb_sectors,
 319                  BlockDriverCompletionFunc *cb,
 320                  void *opaque)
 321 {
 322     IscsiLun *iscsilun = bs->opaque;
 323     IscsiAIOCB *acb;
 324
 325     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 326     trace_iscsi_aio_writev(iscsilun->iscsi, sector_num, nb_sectors, opaque, acb);
 327
 328     acb->iscsilun    = iscsilun;
 329     acb->qiov        = qiov;
 330     acb->nb_sectors  = nb_sectors;
 331     acb->sector_num  = sector_num;
 332     acb->retries     = ISCSI_CMD_RETRIES;
 333
 334     if (iscsi_aio_writev_acb(acb) != 0) {
 335         qemu_aio_release(acb);
 336         return NULL;
 337     }
 338
 339     iscsi_set_events(iscsilun);
 340     return &acb->common;
 341 }
 342
 343 static int
 344 iscsi_aio_readv_acb(IscsiAIOCB *acb);
 345
 346 static void
 347 iscsi_aio_read16_cb(struct iscsi_context *iscsi, int status,
 348                     void *command_data, void *opaque)
 349 {
 350     IscsiAIOCB *acb = opaque;
 351
 352     trace_iscsi_aio_read16_cb(iscsi, status, acb, acb->canceled);
 353
 354     if (acb->canceled != 0) {
 355         return;
 356     }
 357
 358     acb->status = 0;
 359     if (status != 0) {
 360         if (status == SCSI_STATUS_CHECK_CONDITION
 361             && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
 362             && acb->retries-- > 0) {
 363             scsi_free_scsi_task(acb->task);
 364             acb->task = NULL;
 365             if (iscsi_aio_readv_acb(acb) == 0) {
 366                 iscsi_set_events(acb->iscsilun);
 367                 return;
 368             }
 369         }
 370         error_report("Failed to read16 data from iSCSI lun. %s",
 371                      iscsi_get_error(iscsi));
 372         acb->status = -EIO;
 373     }
 374
 375     iscsi_schedule_bh(acb);
 376 }
 377
 378 static int
 379 iscsi_aio_readv_acb(IscsiAIOCB *acb)
 380 {
 381     struct iscsi_context *iscsi = acb->iscsilun->iscsi;
 382     uint64_t lba;
 383     uint32_t num_sectors;
 384     int ret;
 385 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 386     int i;
 387 #endif
 388
 389     acb->canceled    = 0;
 390     acb->bh          = NULL;
 391     acb->status      = -EINPROGRESS;
 392     acb->buf         = NULL;
 393
 394     /* If LUN blocksize is bigger than BDRV_BLOCK_SIZE a read from QEMU
 395      * may be misaligned to the LUN, so we may need to read some extra
 396      * data.
 397      */
 398     acb->read_offset = 0;
 399     if (acb->iscsilun->block_size > BDRV_SECTOR_SIZE) {
 400         uint64_t bdrv_offset = BDRV_SECTOR_SIZE * acb->sector_num;
 401
 402         acb->read_offset  = bdrv_offset % acb->iscsilun->block_size;
 403     }
 404
 405     num_sectors  = (acb->read_size + acb->iscsilun->block_size
 406                     + acb->read_offset - 1)
 407                     / acb->iscsilun->block_size;
 408
 409     acb->task = malloc(sizeof(struct scsi_task));
 410     if (acb->task == NULL) {
 411         error_report("iSCSI: Failed to allocate task for scsi READ16 "
 412                      "command. %s", iscsi_get_error(iscsi));
 413         return -1;
 414     }
 415     memset(acb->task, 0, sizeof(struct scsi_task));
 416
 417     acb->task->xfer_dir = SCSI_XFER_READ;
 418     lba = sector_qemu2lun(acb->sector_num, acb->iscsilun);
 419     acb->task->expxferlen = acb->read_size;
 420
 421     switch (acb->iscsilun->type) {
 422     case TYPE_DISK:
 423         acb->task->cdb_size = 16;
 424         acb->task->cdb[0]  = 0x88;
 425         *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 426         *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 427         *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 428         break;
 429     default:
 430         acb->task->cdb_size = 10;
 431         acb->task->cdb[0]  = 0x28;
 432         *(uint32_t *)&acb->task->cdb[2] = htonl(lba);
 433         *(uint16_t *)&acb->task->cdb[7] = htons(num_sectors);
 434         break;
 435     }
 436
 437     ret = iscsi_scsi_command_async(iscsi, acb->iscsilun->lun, acb->task,
 438                                    iscsi_aio_read16_cb,
 439                                    NULL,
 440                                    acb);
 441     if (ret != 0) {
 442         scsi_free_scsi_task(acb->task);
 443         return -1;
 444     }
 445
 446 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 447     scsi_task_set_iov_in(acb->task, (struct scsi_iovec*) acb->qiov->iov, acb->qiov->niov);
 448 #else
 449     for (i = 0; i < acb->qiov->niov; i++) {
 450         scsi_task_add_data_in_buffer(acb->task,
 451                 acb->qiov->iov[i].iov_len,
 452                 acb->qiov->iov[i].iov_base);
 453     }
 454 #endif
 455     return 0;
 456 }
 457
 458 static BlockDriverAIOCB *
 459 iscsi_aio_readv(BlockDriverState *bs, int64_t sector_num,
 460                 QEMUIOVector *qiov, int nb_sectors,
 461                 BlockDriverCompletionFunc *cb,
 462                 void *opaque)
 463 {
 464     IscsiLun *iscsilun = bs->opaque;
 465     IscsiAIOCB *acb;
 466
 467     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 468     trace_iscsi_aio_readv(iscsilun->iscsi, sector_num, nb_sectors, opaque, acb);
 469
 470     acb->nb_sectors  = nb_sectors;
 471     acb->sector_num  = sector_num;
 472     acb->iscsilun    = iscsilun;
 473     acb->qiov        = qiov;
 474     acb->read_size   = BDRV_SECTOR_SIZE * (size_t)acb->nb_sectors;
 475     acb->retries     = ISCSI_CMD_RETRIES;
 476
 477     if (iscsi_aio_readv_acb(acb) != 0) {
 478         qemu_aio_release(acb);
 479         return NULL;
 480     }
 481
 482     iscsi_set_events(iscsilun);
 483     return &acb->common;
 484 }
 485
 486 static int
 487 iscsi_aio_flush_acb(IscsiAIOCB *acb);
 488
 489 static void
 490 iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
 491                      void *command_data, void *opaque)
 492 {
 493     IscsiAIOCB *acb = opaque;
 494
 495     if (acb->canceled != 0) {
 496         return;
 497     }
 498
 499     acb->status = 0;
 500     if (status != 0) {
 501         if (status == SCSI_STATUS_CHECK_CONDITION
 502             && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
 503             && acb->retries-- > 0) {
 504             scsi_free_scsi_task(acb->task);
 505             acb->task = NULL;
 506             if (iscsi_aio_flush_acb(acb) == 0) {
 507                 iscsi_set_events(acb->iscsilun);
 508                 return;
 509             }
 510         }
 511         error_report("Failed to sync10 data on iSCSI lun. %s",
 512                      iscsi_get_error(iscsi));
 513         acb->status = -EIO;
 514     }
 515
 516     iscsi_schedule_bh(acb);
 517 }
 518
 519 static int
 520 iscsi_aio_flush_acb(IscsiAIOCB *acb)
 521 {
 522     struct iscsi_context *iscsi = acb->iscsilun->iscsi;
 523
 524     acb->canceled   = 0;
 525     acb->bh         = NULL;
 526     acb->status     = -EINPROGRESS;
 527     acb->buf        = NULL;
 528
 529     acb->task = iscsi_synchronizecache10_task(iscsi, acb->iscsilun->lun,
 530                                          0, 0, 0, 0,
 531                                          iscsi_synccache10_cb,
 532                                          acb);
 533     if (acb->task == NULL) {
 534         error_report("iSCSI: Failed to send synchronizecache10 command. %s",
 535                      iscsi_get_error(iscsi));
 536         return -1;
 537     }
 538
 539     return 0;
 540 }
 541
 542 static BlockDriverAIOCB *
 543 iscsi_aio_flush(BlockDriverState *bs,
 544                 BlockDriverCompletionFunc *cb, void *opaque)
 545 {
 546     IscsiLun *iscsilun = bs->opaque;
 547
 548     IscsiAIOCB *acb;
 549
 550     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 551
 552     acb->iscsilun    = iscsilun;
 553     acb->retries     = ISCSI_CMD_RETRIES;
 554
 555     if (iscsi_aio_flush_acb(acb) != 0) {
 556         qemu_aio_release(acb);
 557         return NULL;
 558     }
 559
 560     iscsi_set_events(iscsilun);
 561
 562     return &acb->common;
 563 }
 564
 565 static int iscsi_aio_discard_acb(IscsiAIOCB *acb);
 566
 567 static void
 568 iscsi_unmap_cb(struct iscsi_context *iscsi, int status,
 569                      void *command_data, void *opaque)
 570 {
 571     IscsiAIOCB *acb = opaque;
 572
 573     if (acb->canceled != 0) {
 574         return;
 575     }
 576
 577     acb->status = 0;
 578     if (status != 0) {
 579         if (status == SCSI_STATUS_CHECK_CONDITION
 580             && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
 581             && acb->retries-- > 0) {
 582             scsi_free_scsi_task(acb->task);
 583             acb->task = NULL;
 584             if (iscsi_aio_discard_acb(acb) == 0) {
 585                 iscsi_set_events(acb->iscsilun);
 586                 return;
 587             }
 588         }
 589         error_report("Failed to unmap data on iSCSI lun. %s",
 590                      iscsi_get_error(iscsi));
 591         acb->status = -EIO;
 592     }
 593
 594     iscsi_schedule_bh(acb);
 595 }
 596
 597 static int iscsi_aio_discard_acb(IscsiAIOCB *acb) {
 598     struct iscsi_context *iscsi = acb->iscsilun->iscsi;
 599     struct unmap_list list[1];
 600
 601     acb->canceled   = 0;
 602     acb->bh         = NULL;
 603     acb->status     = -EINPROGRESS;
 604     acb->buf        = NULL;
 605
 606     list[0].lba = sector_qemu2lun(acb->sector_num, acb->iscsilun);
 607     list[0].num = acb->nb_sectors * BDRV_SECTOR_SIZE / acb->iscsilun->block_size;
 608
 609     acb->task = iscsi_unmap_task(iscsi, acb->iscsilun->lun,
 610                                  0, 0, &list[0], 1,
 611                                  iscsi_unmap_cb,
 612                                  acb);
 613     if (acb->task == NULL) {
 614         error_report("iSCSI: Failed to send unmap command. %s",
 615                      iscsi_get_error(iscsi));
 616         return -1;
 617     }
 618
 619     return 0;
 620 }
 621
 622 static BlockDriverAIOCB *
 623 iscsi_aio_discard(BlockDriverState *bs,
 624                   int64_t sector_num, int nb_sectors,
 625                   BlockDriverCompletionFunc *cb, void *opaque)
 626 {
 627     IscsiLun *iscsilun = bs->opaque;
 628     IscsiAIOCB *acb;
 629
 630     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 631
 632     acb->iscsilun    = iscsilun;
 633     acb->nb_sectors  = nb_sectors;
 634     acb->sector_num  = sector_num;
 635     acb->retries     = ISCSI_CMD_RETRIES;
 636
 637     if (iscsi_aio_discard_acb(acb) != 0) {
 638         qemu_aio_release(acb);
 639         return NULL;
 640     }
 641
 642     iscsi_set_events(iscsilun);
 643
 644     return &acb->common;
 645 }
 646
 647 #ifdef __linux__
 648 static void
 649 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 650                      void *command_data, void *opaque)
 651 {
 652     IscsiAIOCB *acb = opaque;
 653
 654     if (acb->canceled != 0) {
 655         return;
 656     }
 657
 658     acb->status = 0;
 659     if (status < 0) {
 660         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 661                      iscsi_get_error(iscsi));
 662         acb->status = -EIO;
 663     }
 664
 665     acb->ioh->driver_status = 0;
 666     acb->ioh->host_status   = 0;
 667     acb->ioh->resid         = 0;
 668
 669 #define SG_ERR_DRIVER_SENSE    0x08
 670
 671     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 672         int ss;
 673
 674         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 675
 676         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 677         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 678              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 679         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 680     }
 681
 682     iscsi_schedule_bh(acb);
 683 }
 684
 685 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 686         unsigned long int req, void *buf,
 687         BlockDriverCompletionFunc *cb, void *opaque)
 688 {
 689     IscsiLun *iscsilun = bs->opaque;
 690     struct iscsi_context *iscsi = iscsilun->iscsi;
 691     struct iscsi_data data;
 692     IscsiAIOCB *acb;
 693
 694     assert(req == SG_IO);
 695
 696     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 697
 698     acb->iscsilun = iscsilun;
 699     acb->canceled    = 0;
 700     acb->bh          = NULL;
 701     acb->status      = -EINPROGRESS;
 702     acb->buf         = NULL;
 703     acb->ioh         = buf;
 704
 705     acb->task = malloc(sizeof(struct scsi_task));
 706     if (acb->task == NULL) {
 707         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 708                      iscsi_get_error(iscsi));
 709         qemu_aio_release(acb);
 710         return NULL;
 711     }
 712     memset(acb->task, 0, sizeof(struct scsi_task));
 713
 714     switch (acb->ioh->dxfer_direction) {
 715     case SG_DXFER_TO_DEV:
 716         acb->task->xfer_dir = SCSI_XFER_WRITE;
 717         break;
 718     case SG_DXFER_FROM_DEV:
 719         acb->task->xfer_dir = SCSI_XFER_READ;
 720         break;
 721     default:
 722         acb->task->xfer_dir = SCSI_XFER_NONE;
 723         break;
 724     }
 725
 726     acb->task->cdb_size = acb->ioh->cmd_len;
 727     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 728     acb->task->expxferlen = acb->ioh->dxfer_len;
 729
 730     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 731         data.data = acb->ioh->dxferp;
 732         data.size = acb->ioh->dxfer_len;
 733     }
 734     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 735                                  iscsi_aio_ioctl_cb,
 736                                  (acb->task->xfer_dir == SCSI_XFER_WRITE) ?
 737                                      &data : NULL,
 738                                  acb) != 0) {
 739         scsi_free_scsi_task(acb->task);
 740         qemu_aio_release(acb);
 741         return NULL;
 742     }
 743
 744     /* tell libiscsi to read straight into the buffer we got from ioctl */
 745     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 746         scsi_task_add_data_in_buffer(acb->task,
 747                                      acb->ioh->dxfer_len,
 748                                      acb->ioh->dxferp);
 749     }
 750
 751     iscsi_set_events(iscsilun);
 752
 753     return &acb->common;
 754 }
 755
 756
 757 static void ioctl_cb(void *opaque, int status)
 758 {
 759     int *p_status = opaque;
 760     *p_status = status;
 761 }
 762
 763 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 764 {
 765     IscsiLun *iscsilun = bs->opaque;
 766     int status;
 767
 768     switch (req) {
 769     case SG_GET_VERSION_NUM:
 770         *(int *)buf = 30000;
 771         break;
 772     case SG_GET_SCSI_ID:
 773         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 774         break;
 775     case SG_IO:
 776         status = -EINPROGRESS;
 777         iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status);
 778
 779         while (status == -EINPROGRESS) {
 780             qemu_aio_wait();
 781         }
 782
 783         return 0;
 784     default:
 785         return -1;
 786     }
 787     return 0;
 788 }
 789 #endif
 790
 791 static int64_t
 792 iscsi_getlength(BlockDriverState *bs)
 793 {
 794     IscsiLun *iscsilun = bs->opaque;
 795     int64_t len;
 796
 797     len  = iscsilun->num_blocks;
 798     len *= iscsilun->block_size;
 799
 800     return len;
 801 }
 802
 803 static int parse_chap(struct iscsi_context *iscsi, const char *target)
 804 {
 805     QemuOptsList *list;
 806     QemuOpts *opts;
 807     const char *user = NULL;
 808     const char *password = NULL;
 809
 810     list = qemu_find_opts("iscsi");
 811     if (!list) {
 812         return 0;
 813     }
 814
 815     opts = qemu_opts_find(list, target);
 816     if (opts == NULL) {
 817         opts = QTAILQ_FIRST(&list->head);
 818         if (!opts) {
 819             return 0;
 820         }
 821     }
 822
 823     user = qemu_opt_get(opts, "user");
 824     if (!user) {
 825         return 0;
 826     }
 827
 828     password = qemu_opt_get(opts, "password");
 829     if (!password) {
 830         error_report("CHAP username specified but no password was given");
 831         return -1;
 832     }
 833
 834     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
 835         error_report("Failed to set initiator username and password");
 836         return -1;
 837     }
 838
 839     return 0;
 840 }
 841
 842 static void parse_header_digest(struct iscsi_context *iscsi, const char *target)
 843 {
 844     QemuOptsList *list;
 845     QemuOpts *opts;
 846     const char *digest = NULL;
 847
 848     list = qemu_find_opts("iscsi");
 849     if (!list) {
 850         return;
 851     }
 852
 853     opts = qemu_opts_find(list, target);
 854     if (opts == NULL) {
 855         opts = QTAILQ_FIRST(&list->head);
 856         if (!opts) {
 857             return;
 858         }
 859     }
 860
 861     digest = qemu_opt_get(opts, "header-digest");
 862     if (!digest) {
 863         return;
 864     }
 865
 866     if (!strcmp(digest, "CRC32C")) {
 867         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
 868     } else if (!strcmp(digest, "NONE")) {
 869         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
 870     } else if (!strcmp(digest, "CRC32C-NONE")) {
 871         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
 872     } else if (!strcmp(digest, "NONE-CRC32C")) {
 873         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 874     } else {
 875         error_report("Invalid header-digest setting : %s", digest);
 876     }
 877 }
 878
 879 static char *parse_initiator_name(const char *target)
 880 {
 881     QemuOptsList *list;
 882     QemuOpts *opts;
 883     const char *name = NULL;
 884     const char *iscsi_name = qemu_get_vm_name();
 885
 886     list = qemu_find_opts("iscsi");
 887     if (list) {
 888         opts = qemu_opts_find(list, target);
 889         if (!opts) {
 890             opts = QTAILQ_FIRST(&list->head);
 891         }
 892         if (opts) {
 893             name = qemu_opt_get(opts, "initiator-name");
 894         }
 895     }
 896
 897     if (name) {
 898         return g_strdup(name);
 899     } else {
 900         return g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
 901                                iscsi_name ? ":" : "",
 902                                iscsi_name ? iscsi_name : "");
 903     }
 904 }
 905
 906 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
 907 static void iscsi_nop_timed_event(void *opaque)
 908 {
 909     IscsiLun *iscsilun = opaque;
 910
 911     if (iscsi_get_nops_in_flight(iscsilun->iscsi) > MAX_NOP_FAILURES) {
 912         error_report("iSCSI: NOP timeout. Reconnecting...");
 913         iscsi_reconnect(iscsilun->iscsi);
 914     }
 915
 916     if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
 917         error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
 918         return;
 919     }
 920
 921     qemu_mod_timer(iscsilun->nop_timer, qemu_get_clock_ms(rt_clock) + NOP_INTERVAL);
 922     iscsi_set_events(iscsilun);
 923 }
 924 #endif
 925
 926 static int iscsi_readcapacity_sync(IscsiLun *iscsilun)
 927 {
 928     struct scsi_task *task = NULL;
 929     struct scsi_readcapacity10 *rc10 = NULL;
 930     struct scsi_readcapacity16 *rc16 = NULL;
 931     int ret = 0;
 932     int retries = ISCSI_CMD_RETRIES;
 933
 934     do {
 935         if (task != NULL) {
 936             scsi_free_scsi_task(task);
 937             task = NULL;
 938         }
 939
 940         switch (iscsilun->type) {
 941         case TYPE_DISK:
 942             task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
 943             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
 944                 rc16 = scsi_datain_unmarshall(task);
 945                 if (rc16 == NULL) {
 946                     error_report("iSCSI: Failed to unmarshall readcapacity16 data.");
 947                     ret = -EINVAL;
 948                 } else {
 949                     iscsilun->block_size = rc16->block_length;
 950                     iscsilun->num_blocks = rc16->returned_lba + 1;
 951                 }
 952             }
 953             break;
 954         case TYPE_ROM:
 955             task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
 956             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
 957                 rc10 = scsi_datain_unmarshall(task);
 958                 if (rc10 == NULL) {
 959                     error_report("iSCSI: Failed to unmarshall readcapacity10 data.");
 960                     ret = -EINVAL;
 961                 } else {
 962                     iscsilun->block_size = rc10->block_size;
 963                     if (rc10->lba == 0) {
 964                         /* blank disk loaded */
 965                         iscsilun->num_blocks = 0;
 966                     } else {
 967                         iscsilun->num_blocks = rc10->lba + 1;
 968                     }
 969                 }
 970             }
 971             break;
 972         default:
 973             return 0;
 974         }
 975     } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
 976              && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
 977              && retries-- > 0);
 978
 979     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
 980         error_report("iSCSI: failed to send readcapacity10 command.");
 981         ret = -EINVAL;
 982     }
 983     if (task) {
 984         scsi_free_scsi_task(task);
 985     }
 986     return ret;
 987 }
 988
 989 /* TODO Convert to fine grained options */
 990 static QemuOptsList runtime_opts = {
 991     .name = "iscsi",
 992     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 993     .desc = {
 994         {
 995             .name = "filename",
 996             .type = QEMU_OPT_STRING,
 997             .help = "URL to the iscsi image",
 998         },
 999         { /* end of list */ }
1000     },
1001 };
1002
1003 /*
1004  * We support iscsi url's on the form
1005  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1006  */
1007 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags)
1008 {
1009     IscsiLun *iscsilun = bs->opaque;
1010     struct iscsi_context *iscsi = NULL;
1011     struct iscsi_url *iscsi_url = NULL;
1012     struct scsi_task *task = NULL;
1013     struct scsi_inquiry_standard *inq = NULL;
1014     char *initiator_name = NULL;
1015     QemuOpts *opts;
1016     Error *local_err = NULL;
1017     const char *filename;
1018     int ret;
1019
1020     if ((BDRV_SECTOR_SIZE % 512) != 0) {
1021         error_report("iSCSI: Invalid BDRV_SECTOR_SIZE. "
1022                      "BDRV_SECTOR_SIZE(%lld) is not a multiple "
1023                      "of 512", BDRV_SECTOR_SIZE);
1024         return -EINVAL;
1025     }
1026
1027     opts = qemu_opts_create_nofail(&runtime_opts);
1028     qemu_opts_absorb_qdict(opts, options, &local_err);
1029     if (error_is_set(&local_err)) {
1030         qerror_report_err(local_err);
1031         error_free(local_err);
1032         ret = -EINVAL;
1033         goto out;
1034     }
1035
1036     filename = qemu_opt_get(opts, "filename");
1037
1038
1039     iscsi_url = iscsi_parse_full_url(iscsi, filename);
1040     if (iscsi_url == NULL) {
1041         error_report("Failed to parse URL : %s", filename);
1042         ret = -EINVAL;
1043         goto out;
1044     }
1045
1046     memset(iscsilun, 0, sizeof(IscsiLun));
1047
1048     initiator_name = parse_initiator_name(iscsi_url->target);
1049
1050     iscsi = iscsi_create_context(initiator_name);
1051     if (iscsi == NULL) {
1052         error_report("iSCSI: Failed to create iSCSI context.");
1053         ret = -ENOMEM;
1054         goto out;
1055     }
1056
1057     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1058         error_report("iSCSI: Failed to set target name.");
1059         ret = -EINVAL;
1060         goto out;
1061     }
1062
1063     if (iscsi_url->user != NULL) {
1064         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1065                                               iscsi_url->passwd);
1066         if (ret != 0) {
1067             error_report("Failed to set initiator username and password");
1068             ret = -EINVAL;
1069             goto out;
1070         }
1071     }
1072
1073     /* check if we got CHAP username/password via the options */
1074     if (parse_chap(iscsi, iscsi_url->target) != 0) {
1075         error_report("iSCSI: Failed to set CHAP user/password");
1076         ret = -EINVAL;
1077         goto out;
1078     }
1079
1080     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1081         error_report("iSCSI: Failed to set session type to normal.");
1082         ret = -EINVAL;
1083         goto out;
1084     }
1085
1086     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1087
1088     /* check if we got HEADER_DIGEST via the options */
1089     parse_header_digest(iscsi, iscsi_url->target);
1090
1091     if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1092         error_report("iSCSI: Failed to connect to LUN : %s",
1093             iscsi_get_error(iscsi));
1094         ret = -EINVAL;
1095         goto out;
1096     }
1097
1098     iscsilun->iscsi = iscsi;
1099     iscsilun->lun   = iscsi_url->lun;
1100
1101     task = iscsi_inquiry_sync(iscsi, iscsilun->lun, 0, 0, 36);
1102
1103     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1104         error_report("iSCSI: failed to send inquiry command.");
1105         ret = -EINVAL;
1106         goto out;
1107     }
1108
1109     inq = scsi_datain_unmarshall(task);
1110     if (inq == NULL) {
1111         error_report("iSCSI: Failed to unmarshall inquiry data.");
1112         ret = -EINVAL;
1113         goto out;
1114     }
1115
1116     iscsilun->type = inq->periperal_device_type;
1117
1118     if ((ret = iscsi_readcapacity_sync(iscsilun)) != 0) {
1119         goto out;
1120     }
1121     bs->total_sectors    = iscsilun->num_blocks *
1122                            iscsilun->block_size / BDRV_SECTOR_SIZE ;
1123
1124     /* Medium changer or tape. We dont have any emulation for this so this must
1125      * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
1126      * to read from the device to guess the image format.
1127      */
1128     if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
1129         iscsilun->type == TYPE_TAPE) {
1130         bs->sg = 1;
1131     }
1132
1133 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
1134     /* Set up a timer for sending out iSCSI NOPs */
1135     iscsilun->nop_timer = qemu_new_timer_ms(rt_clock, iscsi_nop_timed_event, iscsilun);
1136     qemu_mod_timer(iscsilun->nop_timer, qemu_get_clock_ms(rt_clock) + NOP_INTERVAL);
1137 #endif
1138
1139 out:
1140     qemu_opts_del(opts);
1141     if (initiator_name != NULL) {
1142         g_free(initiator_name);
1143     }
1144     if (iscsi_url != NULL) {
1145         iscsi_destroy_url(iscsi_url);
1146     }
1147     if (task != NULL) {
1148         scsi_free_scsi_task(task);
1149     }
1150
1151     if (ret) {
1152         if (iscsi != NULL) {
1153             iscsi_destroy_context(iscsi);
1154         }
1155         memset(iscsilun, 0, sizeof(IscsiLun));
1156     }
1157     return ret;
1158 }
1159
1160 static void iscsi_close(BlockDriverState *bs)
1161 {
1162     IscsiLun *iscsilun = bs->opaque;
1163     struct iscsi_context *iscsi = iscsilun->iscsi;
1164
1165     if (iscsilun->nop_timer) {
1166         qemu_del_timer(iscsilun->nop_timer);
1167         qemu_free_timer(iscsilun->nop_timer);
1168     }
1169     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL, NULL);
1170     iscsi_destroy_context(iscsi);
1171     memset(iscsilun, 0, sizeof(IscsiLun));
1172 }
1173
1174 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1175 {
1176     IscsiLun *iscsilun = bs->opaque;
1177     int ret = 0;
1178
1179     if (iscsilun->type != TYPE_DISK) {
1180         return -ENOTSUP;
1181     }
1182
1183     if ((ret = iscsi_readcapacity_sync(iscsilun)) != 0) {
1184         return ret;
1185     }
1186
1187     if (offset > iscsi_getlength(bs)) {
1188         return -EINVAL;
1189     }
1190
1191     return 0;
1192 }
1193
1194 static int iscsi_has_zero_init(BlockDriverState *bs)
1195 {
1196     return 0;
1197 }
1198
1199 static int iscsi_create(const char *filename, QEMUOptionParameter *options)
1200 {
1201     int ret = 0;
1202     int64_t total_size = 0;
1203     BlockDriverState bs;
1204     IscsiLun *iscsilun = NULL;
1205     QDict *bs_options;
1206
1207     memset(&bs, 0, sizeof(BlockDriverState));
1208
1209     /* Read out options */
1210     while (options && options->name) {
1211         if (!strcmp(options->name, "size")) {
1212             total_size = options->value.n / BDRV_SECTOR_SIZE;
1213         }
1214         options++;
1215     }
1216
1217     bs.opaque = g_malloc0(sizeof(struct IscsiLun));
1218     iscsilun = bs.opaque;
1219
1220     bs_options = qdict_new();
1221     qdict_put(bs_options, "filename", qstring_from_str(filename));
1222     ret = iscsi_open(&bs, bs_options, 0);
1223     QDECREF(bs_options);
1224
1225     if (ret != 0) {
1226         goto out;
1227     }
1228     if (iscsilun->nop_timer) {
1229         qemu_del_timer(iscsilun->nop_timer);
1230         qemu_free_timer(iscsilun->nop_timer);
1231     }
1232     if (iscsilun->type != TYPE_DISK) {
1233         ret = -ENODEV;
1234         goto out;
1235     }
1236     if (bs.total_sectors < total_size) {
1237         ret = -ENOSPC;
1238     }
1239
1240     ret = 0;
1241 out:
1242     if (iscsilun->iscsi != NULL) {
1243         iscsi_destroy_context(iscsilun->iscsi);
1244     }
1245     g_free(bs.opaque);
1246     return ret;
1247 }
1248
1249 static QEMUOptionParameter iscsi_create_options[] = {
1250     {
1251         .name = BLOCK_OPT_SIZE,
1252         .type = OPT_SIZE,
1253         .help = "Virtual disk size"
1254     },
1255     { NULL }
1256 };
1257
1258 static BlockDriver bdrv_iscsi = {
1259     .format_name     = "iscsi",
1260     .protocol_name   = "iscsi",
1261
1262     .instance_size   = sizeof(IscsiLun),
1263     .bdrv_file_open  = iscsi_open,
1264     .bdrv_close      = iscsi_close,
1265     .bdrv_create     = iscsi_create,
1266     .create_options  = iscsi_create_options,
1267
1268     .bdrv_getlength  = iscsi_getlength,
1269     .bdrv_truncate   = iscsi_truncate,
1270
1271     .bdrv_aio_readv  = iscsi_aio_readv,
1272     .bdrv_aio_writev = iscsi_aio_writev,
1273     .bdrv_aio_flush  = iscsi_aio_flush,
1274
1275     .bdrv_aio_discard = iscsi_aio_discard,
1276     .bdrv_has_zero_init = iscsi_has_zero_init,
1277
1278 #ifdef __linux__
1279     .bdrv_ioctl       = iscsi_ioctl,
1280     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1281 #endif
1282 };
1283
1284 static QemuOptsList qemu_iscsi_opts = {
1285     .name = "iscsi",
1286     .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1287     .desc = {
1288         {
1289             .name = "user",
1290             .type = QEMU_OPT_STRING,
1291             .help = "username for CHAP authentication to target",
1292         },{
1293             .name = "password",
1294             .type = QEMU_OPT_STRING,
1295             .help = "password for CHAP authentication to target",
1296         },{
1297             .name = "header-digest",
1298             .type = QEMU_OPT_STRING,
1299             .help = "HeaderDigest setting. "
1300                     "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1301         },{
1302             .name = "initiator-name",
1303             .type = QEMU_OPT_STRING,
1304             .help = "Initiator iqn name to use when connecting",
1305         },
1306         { /* end of list */ }
1307     },
1308 };
1309
1310 static void iscsi_block_init(void)
1311 {
1312     bdrv_register(&bdrv_iscsi);
1313     qemu_add_opts(&qemu_iscsi_opts);
1314 }
1315
1316 block_init(iscsi_block_init);