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-error.h"
  31 #include "block_int.h"
  32 #include "trace.h"
  33 #include "hw/scsi-defs.h"
  34
  35 #include <iscsi/iscsi.h>
  36 #include <iscsi/scsi-lowlevel.h>
  37
  38 #ifdef __linux__
  39 #include <scsi/sg.h>
  40 #include <hw/scsi-defs.h>
  41 #endif
  42
  43 typedef struct IscsiLun {
  44     struct iscsi_context *iscsi;
  45     int lun;
  46     enum scsi_inquiry_peripheral_device_type type;
  47     int block_size;
  48     uint64_t num_blocks;
  49     int events;
  50 } IscsiLun;
  51
  52 typedef struct IscsiAIOCB {
  53     BlockDriverAIOCB common;
  54     QEMUIOVector *qiov;
  55     QEMUBH *bh;
  56     IscsiLun *iscsilun;
  57     struct scsi_task *task;
  58     uint8_t *buf;
  59     int status;
  60     int canceled;
  61     size_t read_size;
  62     size_t read_offset;
  63 #ifdef __linux__
  64     sg_io_hdr_t *ioh;
  65 #endif
  66 } IscsiAIOCB;
  67
  68 struct IscsiTask {
  69     IscsiLun *iscsilun;
  70     BlockDriverState *bs;
  71     int status;
  72     int complete;
  73 };
  74
  75 static void
  76 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
  77                     void *private_data)
  78 {
  79     IscsiAIOCB *acb = (IscsiAIOCB *)private_data;
  80
  81     scsi_free_scsi_task(acb->task);
  82     acb->task = NULL;
  83 }
  84
  85 static void
  86 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
  87 {
  88     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
  89     IscsiLun *iscsilun = acb->iscsilun;
  90
  91     acb->canceled = 1;
  92
  93     acb->common.cb(acb->common.opaque, -ECANCELED);
  94
  95     /* send a task mgmt call to the target to cancel the task on the target
  96      * this also cancels the task in libiscsi
  97      */
  98     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
  99                                      iscsi_abort_task_cb, &acb);
 100 }
 101
 102 static AIOPool iscsi_aio_pool = {
 103     .aiocb_size         = sizeof(IscsiAIOCB),
 104     .cancel             = iscsi_aio_cancel,
 105 };
 106
 107
 108 static void iscsi_process_read(void *arg);
 109 static void iscsi_process_write(void *arg);
 110
 111 static int iscsi_process_flush(void *arg)
 112 {
 113     IscsiLun *iscsilun = arg;
 114
 115     return iscsi_queue_length(iscsilun->iscsi) > 0;
 116 }
 117
 118 static void
 119 iscsi_set_events(IscsiLun *iscsilun)
 120 {
 121     struct iscsi_context *iscsi = iscsilun->iscsi;
 122     int ev;
 123
 124     /* We always register a read handler.  */
 125     ev = POLLIN;
 126     ev |= iscsi_which_events(iscsi);
 127     if (ev != iscsilun->events) {
 128         qemu_aio_set_fd_handler(iscsi_get_fd(iscsi),
 129                       iscsi_process_read,
 130                       (ev & POLLOUT) ? iscsi_process_write : NULL,
 131                       iscsi_process_flush,
 132                       iscsilun);
 133
 134     }
 135
 136     /* If we just added an event, the callback might be delayed
 137      * unless we call qemu_notify_event().
 138      */
 139     if (ev & ~iscsilun->events) {
 140         qemu_notify_event();
 141     }
 142     iscsilun->events = ev;
 143 }
 144
 145 static void
 146 iscsi_process_read(void *arg)
 147 {
 148     IscsiLun *iscsilun = arg;
 149     struct iscsi_context *iscsi = iscsilun->iscsi;
 150
 151     iscsi_service(iscsi, POLLIN);
 152     iscsi_set_events(iscsilun);
 153 }
 154
 155 static void
 156 iscsi_process_write(void *arg)
 157 {
 158     IscsiLun *iscsilun = arg;
 159     struct iscsi_context *iscsi = iscsilun->iscsi;
 160
 161     iscsi_service(iscsi, POLLOUT);
 162     iscsi_set_events(iscsilun);
 163 }
 164
 165
 166 static int
 167 iscsi_schedule_bh(QEMUBHFunc *cb, IscsiAIOCB *acb)
 168 {
 169     acb->bh = qemu_bh_new(cb, acb);
 170     if (!acb->bh) {
 171         error_report("oom: could not create iscsi bh");
 172         return -EIO;
 173     }
 174
 175     qemu_bh_schedule(acb->bh);
 176     return 0;
 177 }
 178
 179 static void
 180 iscsi_readv_writev_bh_cb(void *p)
 181 {
 182     IscsiAIOCB *acb = p;
 183
 184     qemu_bh_delete(acb->bh);
 185
 186     if (!acb->canceled) {
 187         acb->common.cb(acb->common.opaque, acb->status);
 188     }
 189
 190     qemu_aio_release(acb);
 191
 192     if (acb->canceled) {
 193         return;
 194     }
 195
 196     scsi_free_scsi_task(acb->task);
 197     acb->task = NULL;
 198 }
 199
 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
 211     if (acb->canceled) {
 212         qemu_aio_release(acb);
 213         return;
 214     }
 215
 216     acb->status = 0;
 217     if (status < 0) {
 218         error_report("Failed to write16 data to iSCSI lun. %s",
 219                      iscsi_get_error(iscsi));
 220         acb->status = -EIO;
 221     }
 222
 223     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 224 }
 225
 226 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
 227 {
 228     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
 229 }
 230
 231 static BlockDriverAIOCB *
 232 iscsi_aio_writev(BlockDriverState *bs, int64_t sector_num,
 233                  QEMUIOVector *qiov, int nb_sectors,
 234                  BlockDriverCompletionFunc *cb,
 235                  void *opaque)
 236 {
 237     IscsiLun *iscsilun = bs->opaque;
 238     struct iscsi_context *iscsi = iscsilun->iscsi;
 239     IscsiAIOCB *acb;
 240     size_t size;
 241     uint32_t num_sectors;
 242     uint64_t lba;
 243     struct iscsi_data data;
 244
 245     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 246     trace_iscsi_aio_writev(iscsi, sector_num, nb_sectors, opaque, acb);
 247
 248     acb->iscsilun = iscsilun;
 249     acb->qiov     = qiov;
 250
 251     acb->canceled   = 0;
 252
 253     /* XXX we should pass the iovec to write16 to avoid the extra copy */
 254     /* this will allow us to get rid of 'buf' completely */
 255     size = nb_sectors * BDRV_SECTOR_SIZE;
 256     acb->buf = g_malloc(size);
 257     qemu_iovec_to_buf(acb->qiov, 0, acb->buf, size);
 258
 259     acb->task = malloc(sizeof(struct scsi_task));
 260     if (acb->task == NULL) {
 261         error_report("iSCSI: Failed to allocate task for scsi WRITE16 "
 262                      "command. %s", iscsi_get_error(iscsi));
 263         qemu_aio_release(acb);
 264         return NULL;
 265     }
 266     memset(acb->task, 0, sizeof(struct scsi_task));
 267
 268     acb->task->xfer_dir = SCSI_XFER_WRITE;
 269     acb->task->cdb_size = 16;
 270     acb->task->cdb[0] = 0x8a;
 271     if (!(bs->open_flags & BDRV_O_CACHE_WB)) {
 272         /* set FUA on writes when cache mode is write through */
 273         acb->task->cdb[1] |= 0x04;
 274     }
 275     lba = sector_qemu2lun(sector_num, iscsilun);
 276     *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 277     *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 278     num_sectors = size / iscsilun->block_size;
 279     *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 280     acb->task->expxferlen = size;
 281
 282     data.data = acb->buf;
 283     data.size = size;
 284
 285     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 286                                  iscsi_aio_write16_cb,
 287                                  &data,
 288                                  acb) != 0) {
 289         scsi_free_scsi_task(acb->task);
 290         g_free(acb->buf);
 291         qemu_aio_release(acb);
 292         return NULL;
 293     }
 294
 295     iscsi_set_events(iscsilun);
 296
 297     return &acb->common;
 298 }
 299
 300 static void
 301 iscsi_aio_read16_cb(struct iscsi_context *iscsi, int status,
 302                     void *command_data, void *opaque)
 303 {
 304     IscsiAIOCB *acb = opaque;
 305
 306     trace_iscsi_aio_read16_cb(iscsi, status, acb, acb->canceled);
 307
 308     if (acb->canceled) {
 309         qemu_aio_release(acb);
 310         return;
 311     }
 312
 313     acb->status = 0;
 314     if (status != 0) {
 315         error_report("Failed to read16 data from iSCSI lun. %s",
 316                      iscsi_get_error(iscsi));
 317         acb->status = -EIO;
 318     }
 319
 320     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 321 }
 322
 323 static BlockDriverAIOCB *
 324 iscsi_aio_readv(BlockDriverState *bs, int64_t sector_num,
 325                 QEMUIOVector *qiov, int nb_sectors,
 326                 BlockDriverCompletionFunc *cb,
 327                 void *opaque)
 328 {
 329     IscsiLun *iscsilun = bs->opaque;
 330     struct iscsi_context *iscsi = iscsilun->iscsi;
 331     IscsiAIOCB *acb;
 332     size_t qemu_read_size;
 333     int i;
 334     uint64_t lba;
 335     uint32_t num_sectors;
 336
 337     qemu_read_size = BDRV_SECTOR_SIZE * (size_t)nb_sectors;
 338
 339     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 340     trace_iscsi_aio_readv(iscsi, sector_num, nb_sectors, opaque, acb);
 341
 342     acb->iscsilun = iscsilun;
 343     acb->qiov     = qiov;
 344
 345     acb->canceled    = 0;
 346     acb->read_size   = qemu_read_size;
 347     acb->buf         = NULL;
 348
 349     /* If LUN blocksize is bigger than BDRV_BLOCK_SIZE a read from QEMU
 350      * may be misaligned to the LUN, so we may need to read some extra
 351      * data.
 352      */
 353     acb->read_offset = 0;
 354     if (iscsilun->block_size > BDRV_SECTOR_SIZE) {
 355         uint64_t bdrv_offset = BDRV_SECTOR_SIZE * sector_num;
 356
 357         acb->read_offset  = bdrv_offset % iscsilun->block_size;
 358     }
 359
 360     num_sectors  = (qemu_read_size + iscsilun->block_size
 361                     + acb->read_offset - 1)
 362                     / iscsilun->block_size;
 363
 364     acb->task = malloc(sizeof(struct scsi_task));
 365     if (acb->task == NULL) {
 366         error_report("iSCSI: Failed to allocate task for scsi READ16 "
 367                      "command. %s", iscsi_get_error(iscsi));
 368         qemu_aio_release(acb);
 369         return NULL;
 370     }
 371     memset(acb->task, 0, sizeof(struct scsi_task));
 372
 373     acb->task->xfer_dir = SCSI_XFER_READ;
 374     lba = sector_qemu2lun(sector_num, iscsilun);
 375     acb->task->expxferlen = qemu_read_size;
 376
 377     switch (iscsilun->type) {
 378     case TYPE_DISK:
 379         acb->task->cdb_size = 16;
 380         acb->task->cdb[0]  = 0x88;
 381         *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 382         *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 383         *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 384         break;
 385     default:
 386         acb->task->cdb_size = 10;
 387         acb->task->cdb[0]  = 0x28;
 388         *(uint32_t *)&acb->task->cdb[2] = htonl(lba);
 389         *(uint16_t *)&acb->task->cdb[7] = htons(num_sectors);
 390         break;
 391     }
 392
 393     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 394                                  iscsi_aio_read16_cb,
 395                                  NULL,
 396                                  acb) != 0) {
 397         scsi_free_scsi_task(acb->task);
 398         qemu_aio_release(acb);
 399         return NULL;
 400     }
 401
 402     for (i = 0; i < acb->qiov->niov; i++) {
 403         scsi_task_add_data_in_buffer(acb->task,
 404                 acb->qiov->iov[i].iov_len,
 405                 acb->qiov->iov[i].iov_base);
 406     }
 407
 408     iscsi_set_events(iscsilun);
 409
 410     return &acb->common;
 411 }
 412
 413
 414 static void
 415 iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
 416                      void *command_data, void *opaque)
 417 {
 418     IscsiAIOCB *acb = opaque;
 419
 420     if (acb->canceled) {
 421         qemu_aio_release(acb);
 422         return;
 423     }
 424
 425     acb->status = 0;
 426     if (status < 0) {
 427         error_report("Failed to sync10 data on iSCSI lun. %s",
 428                      iscsi_get_error(iscsi));
 429         acb->status = -EIO;
 430     }
 431
 432     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 433 }
 434
 435 static BlockDriverAIOCB *
 436 iscsi_aio_flush(BlockDriverState *bs,
 437                 BlockDriverCompletionFunc *cb, void *opaque)
 438 {
 439     IscsiLun *iscsilun = bs->opaque;
 440     struct iscsi_context *iscsi = iscsilun->iscsi;
 441     IscsiAIOCB *acb;
 442
 443     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 444
 445     acb->iscsilun = iscsilun;
 446     acb->canceled   = 0;
 447
 448     acb->task = iscsi_synchronizecache10_task(iscsi, iscsilun->lun,
 449                                          0, 0, 0, 0,
 450                                          iscsi_synccache10_cb,
 451                                          acb);
 452     if (acb->task == NULL) {
 453         error_report("iSCSI: Failed to send synchronizecache10 command. %s",
 454                      iscsi_get_error(iscsi));
 455         qemu_aio_release(acb);
 456         return NULL;
 457     }
 458
 459     iscsi_set_events(iscsilun);
 460
 461     return &acb->common;
 462 }
 463
 464 static void
 465 iscsi_unmap_cb(struct iscsi_context *iscsi, int status,
 466                      void *command_data, void *opaque)
 467 {
 468     IscsiAIOCB *acb = opaque;
 469
 470     if (acb->canceled) {
 471         qemu_aio_release(acb);
 472         return;
 473     }
 474
 475     acb->status = 0;
 476     if (status < 0) {
 477         error_report("Failed to unmap data on iSCSI lun. %s",
 478                      iscsi_get_error(iscsi));
 479         acb->status = -EIO;
 480     }
 481
 482     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 483 }
 484
 485 static BlockDriverAIOCB *
 486 iscsi_aio_discard(BlockDriverState *bs,
 487                   int64_t sector_num, int nb_sectors,
 488                   BlockDriverCompletionFunc *cb, void *opaque)
 489 {
 490     IscsiLun *iscsilun = bs->opaque;
 491     struct iscsi_context *iscsi = iscsilun->iscsi;
 492     IscsiAIOCB *acb;
 493     struct unmap_list list[1];
 494
 495     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 496
 497     acb->iscsilun = iscsilun;
 498     acb->canceled   = 0;
 499
 500     list[0].lba = sector_qemu2lun(sector_num, iscsilun);
 501     list[0].num = nb_sectors * BDRV_SECTOR_SIZE / iscsilun->block_size;
 502
 503     acb->task = iscsi_unmap_task(iscsi, iscsilun->lun,
 504                                  0, 0, &list[0], 1,
 505                                  iscsi_unmap_cb,
 506                                  acb);
 507     if (acb->task == NULL) {
 508         error_report("iSCSI: Failed to send unmap command. %s",
 509                      iscsi_get_error(iscsi));
 510         qemu_aio_release(acb);
 511         return NULL;
 512     }
 513
 514     iscsi_set_events(iscsilun);
 515
 516     return &acb->common;
 517 }
 518
 519 #ifdef __linux__
 520 static void
 521 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 522                      void *command_data, void *opaque)
 523 {
 524     IscsiAIOCB *acb = opaque;
 525
 526     if (acb->canceled) {
 527         qemu_aio_release(acb);
 528         return;
 529     }
 530
 531     acb->status = 0;
 532     if (status < 0) {
 533         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 534                      iscsi_get_error(iscsi));
 535         acb->status = -EIO;
 536     }
 537
 538     acb->ioh->driver_status = 0;
 539     acb->ioh->host_status   = 0;
 540     acb->ioh->resid         = 0;
 541
 542 #define SG_ERR_DRIVER_SENSE    0x08
 543
 544     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 545         int ss;
 546
 547         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 548
 549         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 550         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 551              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 552         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 553     }
 554
 555     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 556 }
 557
 558 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 559         unsigned long int req, void *buf,
 560         BlockDriverCompletionFunc *cb, void *opaque)
 561 {
 562     IscsiLun *iscsilun = bs->opaque;
 563     struct iscsi_context *iscsi = iscsilun->iscsi;
 564     struct iscsi_data data;
 565     IscsiAIOCB *acb;
 566
 567     assert(req == SG_IO);
 568
 569     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 570
 571     acb->iscsilun = iscsilun;
 572     acb->canceled    = 0;
 573     acb->buf         = NULL;
 574     acb->ioh         = buf;
 575
 576     acb->task = malloc(sizeof(struct scsi_task));
 577     if (acb->task == NULL) {
 578         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 579                      iscsi_get_error(iscsi));
 580         qemu_aio_release(acb);
 581         return NULL;
 582     }
 583     memset(acb->task, 0, sizeof(struct scsi_task));
 584
 585     switch (acb->ioh->dxfer_direction) {
 586     case SG_DXFER_TO_DEV:
 587         acb->task->xfer_dir = SCSI_XFER_WRITE;
 588         break;
 589     case SG_DXFER_FROM_DEV:
 590         acb->task->xfer_dir = SCSI_XFER_READ;
 591         break;
 592     default:
 593         acb->task->xfer_dir = SCSI_XFER_NONE;
 594         break;
 595     }
 596
 597     acb->task->cdb_size = acb->ioh->cmd_len;
 598     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 599     acb->task->expxferlen = acb->ioh->dxfer_len;
 600
 601     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 602         data.data = acb->ioh->dxferp;
 603         data.size = acb->ioh->dxfer_len;
 604     }
 605     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 606                                  iscsi_aio_ioctl_cb,
 607                                  (acb->task->xfer_dir == SCSI_XFER_WRITE) ?
 608                                      &data : NULL,
 609                                  acb) != 0) {
 610         scsi_free_scsi_task(acb->task);
 611         qemu_aio_release(acb);
 612         return NULL;
 613     }
 614
 615     /* tell libiscsi to read straight into the buffer we got from ioctl */
 616     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 617         scsi_task_add_data_in_buffer(acb->task,
 618                                      acb->ioh->dxfer_len,
 619                                      acb->ioh->dxferp);
 620     }
 621
 622     iscsi_set_events(iscsilun);
 623
 624     return &acb->common;
 625 }
 626
 627 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 628 {
 629     IscsiLun *iscsilun = bs->opaque;
 630
 631     switch (req) {
 632     case SG_GET_VERSION_NUM:
 633         *(int *)buf = 30000;
 634         break;
 635     case SG_GET_SCSI_ID:
 636         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 637         break;
 638     default:
 639         return -1;
 640     }
 641     return 0;
 642 }
 643 #endif
 644
 645 static int64_t
 646 iscsi_getlength(BlockDriverState *bs)
 647 {
 648     IscsiLun *iscsilun = bs->opaque;
 649     int64_t len;
 650
 651     len  = iscsilun->num_blocks;
 652     len *= iscsilun->block_size;
 653
 654     return len;
 655 }
 656
 657 static void
 658 iscsi_readcapacity16_cb(struct iscsi_context *iscsi, int status,
 659                         void *command_data, void *opaque)
 660 {
 661     struct IscsiTask *itask = opaque;
 662     struct scsi_readcapacity16 *rc16;
 663     struct scsi_task *task = command_data;
 664
 665     if (status != 0) {
 666         error_report("iSCSI: Failed to read capacity of iSCSI lun. %s",
 667                      iscsi_get_error(iscsi));
 668         itask->status   = 1;
 669         itask->complete = 1;
 670         scsi_free_scsi_task(task);
 671         return;
 672     }
 673
 674     rc16 = scsi_datain_unmarshall(task);
 675     if (rc16 == NULL) {
 676         error_report("iSCSI: Failed to unmarshall readcapacity16 data.");
 677         itask->status   = 1;
 678         itask->complete = 1;
 679         scsi_free_scsi_task(task);
 680         return;
 681     }
 682
 683     itask->iscsilun->block_size = rc16->block_length;
 684     itask->iscsilun->num_blocks = rc16->returned_lba + 1;
 685     itask->bs->total_sectors    = itask->iscsilun->num_blocks *
 686                                itask->iscsilun->block_size / BDRV_SECTOR_SIZE ;
 687
 688     itask->status   = 0;
 689     itask->complete = 1;
 690     scsi_free_scsi_task(task);
 691 }
 692
 693 static void
 694 iscsi_readcapacity10_cb(struct iscsi_context *iscsi, int status,
 695                         void *command_data, void *opaque)
 696 {
 697     struct IscsiTask *itask = opaque;
 698     struct scsi_readcapacity10 *rc10;
 699     struct scsi_task *task = command_data;
 700
 701     if (status != 0) {
 702         error_report("iSCSI: Failed to read capacity of iSCSI lun. %s",
 703                      iscsi_get_error(iscsi));
 704         itask->status   = 1;
 705         itask->complete = 1;
 706         scsi_free_scsi_task(task);
 707         return;
 708     }
 709
 710     rc10 = scsi_datain_unmarshall(task);
 711     if (rc10 == NULL) {
 712         error_report("iSCSI: Failed to unmarshall readcapacity10 data.");
 713         itask->status   = 1;
 714         itask->complete = 1;
 715         scsi_free_scsi_task(task);
 716         return;
 717     }
 718
 719     itask->iscsilun->block_size = rc10->block_size;
 720     itask->iscsilun->num_blocks = rc10->lba + 1;
 721     itask->bs->total_sectors    = itask->iscsilun->num_blocks *
 722                                itask->iscsilun->block_size / BDRV_SECTOR_SIZE ;
 723
 724     itask->status   = 0;
 725     itask->complete = 1;
 726     scsi_free_scsi_task(task);
 727 }
 728
 729 static void
 730 iscsi_inquiry_cb(struct iscsi_context *iscsi, int status, void *command_data,
 731                  void *opaque)
 732 {
 733     struct IscsiTask *itask = opaque;
 734     struct scsi_task *task = command_data;
 735     struct scsi_inquiry_standard *inq;
 736
 737     if (status != 0) {
 738         itask->status   = 1;
 739         itask->complete = 1;
 740         scsi_free_scsi_task(task);
 741         return;
 742     }
 743
 744     inq = scsi_datain_unmarshall(task);
 745     if (inq == NULL) {
 746         error_report("iSCSI: Failed to unmarshall inquiry data.");
 747         itask->status   = 1;
 748         itask->complete = 1;
 749         scsi_free_scsi_task(task);
 750         return;
 751     }
 752
 753     itask->iscsilun->type = inq->periperal_device_type;
 754
 755     scsi_free_scsi_task(task);
 756
 757     switch (itask->iscsilun->type) {
 758     case TYPE_DISK:
 759         task = iscsi_readcapacity16_task(iscsi, itask->iscsilun->lun,
 760                                    iscsi_readcapacity16_cb, opaque);
 761         if (task == NULL) {
 762             error_report("iSCSI: failed to send readcapacity16 command.");
 763             itask->status   = 1;
 764             itask->complete = 1;
 765             return;
 766         }
 767         break;
 768     case TYPE_ROM:
 769         task = iscsi_readcapacity10_task(iscsi, itask->iscsilun->lun,
 770                                    0, 0,
 771                                    iscsi_readcapacity10_cb, opaque);
 772         if (task == NULL) {
 773             error_report("iSCSI: failed to send readcapacity16 command.");
 774             itask->status   = 1;
 775             itask->complete = 1;
 776             return;
 777         }
 778         break;
 779     default:
 780         itask->status   = 0;
 781         itask->complete = 1;
 782     }
 783 }
 784
 785 static void
 786 iscsi_connect_cb(struct iscsi_context *iscsi, int status, void *command_data,
 787                  void *opaque)
 788 {
 789     struct IscsiTask *itask = opaque;
 790     struct scsi_task *task;
 791
 792     if (status != 0) {
 793         itask->status   = 1;
 794         itask->complete = 1;
 795         return;
 796     }
 797
 798     task = iscsi_inquiry_task(iscsi, itask->iscsilun->lun,
 799                               0, 0, 36,
 800                               iscsi_inquiry_cb, opaque);
 801     if (task == NULL) {
 802         error_report("iSCSI: failed to send inquiry command.");
 803         itask->status   = 1;
 804         itask->complete = 1;
 805         return;
 806     }
 807 }
 808
 809 static int parse_chap(struct iscsi_context *iscsi, const char *target)
 810 {
 811     QemuOptsList *list;
 812     QemuOpts *opts;
 813     const char *user = NULL;
 814     const char *password = NULL;
 815
 816     list = qemu_find_opts("iscsi");
 817     if (!list) {
 818         return 0;
 819     }
 820
 821     opts = qemu_opts_find(list, target);
 822     if (opts == NULL) {
 823         opts = QTAILQ_FIRST(&list->head);
 824         if (!opts) {
 825             return 0;
 826         }
 827     }
 828
 829     user = qemu_opt_get(opts, "user");
 830     if (!user) {
 831         return 0;
 832     }
 833
 834     password = qemu_opt_get(opts, "password");
 835     if (!password) {
 836         error_report("CHAP username specified but no password was given");
 837         return -1;
 838     }
 839
 840     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
 841         error_report("Failed to set initiator username and password");
 842         return -1;
 843     }
 844
 845     return 0;
 846 }
 847
 848 static void parse_header_digest(struct iscsi_context *iscsi, const char *target)
 849 {
 850     QemuOptsList *list;
 851     QemuOpts *opts;
 852     const char *digest = NULL;
 853
 854     list = qemu_find_opts("iscsi");
 855     if (!list) {
 856         return;
 857     }
 858
 859     opts = qemu_opts_find(list, target);
 860     if (opts == NULL) {
 861         opts = QTAILQ_FIRST(&list->head);
 862         if (!opts) {
 863             return;
 864         }
 865     }
 866
 867     digest = qemu_opt_get(opts, "header-digest");
 868     if (!digest) {
 869         return;
 870     }
 871
 872     if (!strcmp(digest, "CRC32C")) {
 873         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
 874     } else if (!strcmp(digest, "NONE")) {
 875         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
 876     } else if (!strcmp(digest, "CRC32C-NONE")) {
 877         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
 878     } else if (!strcmp(digest, "NONE-CRC32C")) {
 879         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 880     } else {
 881         error_report("Invalid header-digest setting : %s", digest);
 882     }
 883 }
 884
 885 static char *parse_initiator_name(const char *target)
 886 {
 887     QemuOptsList *list;
 888     QemuOpts *opts;
 889     const char *name = NULL;
 890     const char *iscsi_name = qemu_get_vm_name();
 891
 892     list = qemu_find_opts("iscsi");
 893     if (list) {
 894         opts = qemu_opts_find(list, target);
 895         if (!opts) {
 896             opts = QTAILQ_FIRST(&list->head);
 897         }
 898         if (opts) {
 899             name = qemu_opt_get(opts, "initiator-name");
 900         }
 901     }
 902
 903     if (name) {
 904         return g_strdup(name);
 905     } else {
 906         return g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
 907                                iscsi_name ? ":" : "",
 908                                iscsi_name ? iscsi_name : "");
 909     }
 910 }
 911
 912 /*
 913  * We support iscsi url's on the form
 914  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
 915  */
 916 static int iscsi_open(BlockDriverState *bs, const char *filename, int flags)
 917 {
 918     IscsiLun *iscsilun = bs->opaque;
 919     struct iscsi_context *iscsi = NULL;
 920     struct iscsi_url *iscsi_url = NULL;
 921     struct IscsiTask task;
 922     char *initiator_name = NULL;
 923     int ret;
 924
 925     if ((BDRV_SECTOR_SIZE % 512) != 0) {
 926         error_report("iSCSI: Invalid BDRV_SECTOR_SIZE. "
 927                      "BDRV_SECTOR_SIZE(%lld) is not a multiple "
 928                      "of 512", BDRV_SECTOR_SIZE);
 929         return -EINVAL;
 930     }
 931
 932     iscsi_url = iscsi_parse_full_url(iscsi, filename);
 933     if (iscsi_url == NULL) {
 934         error_report("Failed to parse URL : %s %s", filename,
 935                      iscsi_get_error(iscsi));
 936         ret = -EINVAL;
 937         goto out;
 938     }
 939
 940     memset(iscsilun, 0, sizeof(IscsiLun));
 941
 942     initiator_name = parse_initiator_name(iscsi_url->target);
 943
 944     iscsi = iscsi_create_context(initiator_name);
 945     if (iscsi == NULL) {
 946         error_report("iSCSI: Failed to create iSCSI context.");
 947         ret = -ENOMEM;
 948         goto out;
 949     }
 950
 951     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
 952         error_report("iSCSI: Failed to set target name.");
 953         ret = -EINVAL;
 954         goto out;
 955     }
 956
 957     if (iscsi_url->user != NULL) {
 958         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
 959                                               iscsi_url->passwd);
 960         if (ret != 0) {
 961             error_report("Failed to set initiator username and password");
 962             ret = -EINVAL;
 963             goto out;
 964         }
 965     }
 966
 967     /* check if we got CHAP username/password via the options */
 968     if (parse_chap(iscsi, iscsi_url->target) != 0) {
 969         error_report("iSCSI: Failed to set CHAP user/password");
 970         ret = -EINVAL;
 971         goto out;
 972     }
 973
 974     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
 975         error_report("iSCSI: Failed to set session type to normal.");
 976         ret = -EINVAL;
 977         goto out;
 978     }
 979
 980     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 981
 982     /* check if we got HEADER_DIGEST via the options */
 983     parse_header_digest(iscsi, iscsi_url->target);
 984
 985     task.iscsilun = iscsilun;
 986     task.status = 0;
 987     task.complete = 0;
 988     task.bs = bs;
 989
 990     iscsilun->iscsi = iscsi;
 991     iscsilun->lun   = iscsi_url->lun;
 992
 993     if (iscsi_full_connect_async(iscsi, iscsi_url->portal, iscsi_url->lun,
 994                                  iscsi_connect_cb, &task)
 995         != 0) {
 996         error_report("iSCSI: Failed to start async connect.");
 997         ret = -EINVAL;
 998         goto out;
 999     }
1000
1001     while (!task.complete) {
1002         iscsi_set_events(iscsilun);
1003         qemu_aio_wait();
1004     }
1005     if (task.status != 0) {
1006         error_report("iSCSI: Failed to connect to LUN : %s",
1007                      iscsi_get_error(iscsi));
1008         ret = -EINVAL;
1009         goto out;
1010     }
1011
1012     /* Medium changer or tape. We dont have any emulation for this so this must
1013      * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
1014      * to read from the device to guess the image format.
1015      */
1016     if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
1017         iscsilun->type == TYPE_TAPE) {
1018         bs->sg = 1;
1019     }
1020
1021     ret = 0;
1022
1023 out:
1024     if (initiator_name != NULL) {
1025         g_free(initiator_name);
1026     }
1027     if (iscsi_url != NULL) {
1028         iscsi_destroy_url(iscsi_url);
1029     }
1030
1031     if (ret) {
1032         if (iscsi != NULL) {
1033             iscsi_destroy_context(iscsi);
1034         }
1035         memset(iscsilun, 0, sizeof(IscsiLun));
1036     }
1037     return ret;
1038 }
1039
1040 static void iscsi_close(BlockDriverState *bs)
1041 {
1042     IscsiLun *iscsilun = bs->opaque;
1043     struct iscsi_context *iscsi = iscsilun->iscsi;
1044
1045     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL, NULL);
1046     iscsi_destroy_context(iscsi);
1047     memset(iscsilun, 0, sizeof(IscsiLun));
1048 }
1049
1050 static BlockDriver bdrv_iscsi = {
1051     .format_name     = "iscsi",
1052     .protocol_name   = "iscsi",
1053
1054     .instance_size   = sizeof(IscsiLun),
1055     .bdrv_file_open  = iscsi_open,
1056     .bdrv_close      = iscsi_close,
1057
1058     .bdrv_getlength  = iscsi_getlength,
1059
1060     .bdrv_aio_readv  = iscsi_aio_readv,
1061     .bdrv_aio_writev = iscsi_aio_writev,
1062     .bdrv_aio_flush  = iscsi_aio_flush,
1063
1064     .bdrv_aio_discard = iscsi_aio_discard,
1065
1066 #ifdef __linux__
1067     .bdrv_ioctl       = iscsi_ioctl,
1068     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1069 #endif
1070 };
1071
1072 static void iscsi_block_init(void)
1073 {
1074     bdrv_register(&bdrv_iscsi);
1075 }
1076
1077 block_init(iscsi_block_init);