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_bh_cb(void *p)
  77 {
  78     IscsiAIOCB *acb = p;
  79
  80     qemu_bh_delete(acb->bh);
  81
  82     if (acb->canceled == 0) {
  83         acb->common.cb(acb->common.opaque, acb->status);
  84     }
  85
  86     if (acb->task != NULL) {
  87         scsi_free_scsi_task(acb->task);
  88         acb->task = NULL;
  89     }
  90
  91     qemu_aio_release(acb);
  92 }
  93
  94 static void
  95 iscsi_schedule_bh(IscsiAIOCB *acb)
  96 {
  97     if (acb->bh) {
  98         return;
  99     }
 100     acb->bh = qemu_bh_new(iscsi_bh_cb, acb);
 101     qemu_bh_schedule(acb->bh);
 102 }
 103
 104
 105 static void
 106 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
 107                     void *private_data)
 108 {
 109     IscsiAIOCB *acb = private_data;
 110
 111     acb->status = -ECANCELED;
 112     iscsi_schedule_bh(acb);
 113 }
 114
 115 static void
 116 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
 117 {
 118     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
 119     IscsiLun *iscsilun = acb->iscsilun;
 120
 121     if (acb->status != -EINPROGRESS) {
 122         return;
 123     }
 124
 125     acb->canceled = 1;
 126
 127     /* send a task mgmt call to the target to cancel the task on the target */
 128     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
 129                                      iscsi_abort_task_cb, acb);
 130
 131     while (acb->status == -EINPROGRESS) {
 132         qemu_aio_wait();
 133     }
 134 }
 135
 136 static AIOPool iscsi_aio_pool = {
 137     .aiocb_size         = sizeof(IscsiAIOCB),
 138     .cancel             = iscsi_aio_cancel,
 139 };
 140
 141
 142 static void iscsi_process_read(void *arg);
 143 static void iscsi_process_write(void *arg);
 144
 145 static int iscsi_process_flush(void *arg)
 146 {
 147     IscsiLun *iscsilun = arg;
 148
 149     return iscsi_queue_length(iscsilun->iscsi) > 0;
 150 }
 151
 152 static void
 153 iscsi_set_events(IscsiLun *iscsilun)
 154 {
 155     struct iscsi_context *iscsi = iscsilun->iscsi;
 156     int ev;
 157
 158     /* We always register a read handler.  */
 159     ev = POLLIN;
 160     ev |= iscsi_which_events(iscsi);
 161     if (ev != iscsilun->events) {
 162         qemu_aio_set_fd_handler(iscsi_get_fd(iscsi),
 163                       iscsi_process_read,
 164                       (ev & POLLOUT) ? iscsi_process_write : NULL,
 165                       iscsi_process_flush,
 166                       iscsilun);
 167
 168     }
 169
 170     iscsilun->events = ev;
 171 }
 172
 173 static void
 174 iscsi_process_read(void *arg)
 175 {
 176     IscsiLun *iscsilun = arg;
 177     struct iscsi_context *iscsi = iscsilun->iscsi;
 178
 179     iscsi_service(iscsi, POLLIN);
 180     iscsi_set_events(iscsilun);
 181 }
 182
 183 static void
 184 iscsi_process_write(void *arg)
 185 {
 186     IscsiLun *iscsilun = arg;
 187     struct iscsi_context *iscsi = iscsilun->iscsi;
 188
 189     iscsi_service(iscsi, POLLOUT);
 190     iscsi_set_events(iscsilun);
 191 }
 192
 193
 194 static void
 195 iscsi_aio_write16_cb(struct iscsi_context *iscsi, int status,
 196                      void *command_data, void *opaque)
 197 {
 198     IscsiAIOCB *acb = opaque;
 199
 200     trace_iscsi_aio_write16_cb(iscsi, status, acb, acb->canceled);
 201
 202     g_free(acb->buf);
 203
 204     if (acb->canceled != 0) {
 205         return;
 206     }
 207
 208     acb->status = 0;
 209     if (status < 0) {
 210         error_report("Failed to write16 data to iSCSI lun. %s",
 211                      iscsi_get_error(iscsi));
 212         acb->status = -EIO;
 213     }
 214
 215     iscsi_schedule_bh(acb);
 216 }
 217
 218 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
 219 {
 220     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
 221 }
 222
 223 static BlockDriverAIOCB *
 224 iscsi_aio_writev(BlockDriverState *bs, int64_t sector_num,
 225                  QEMUIOVector *qiov, int nb_sectors,
 226                  BlockDriverCompletionFunc *cb,
 227                  void *opaque)
 228 {
 229     IscsiLun *iscsilun = bs->opaque;
 230     struct iscsi_context *iscsi = iscsilun->iscsi;
 231     IscsiAIOCB *acb;
 232     size_t size;
 233     uint32_t num_sectors;
 234     uint64_t lba;
 235     struct iscsi_data data;
 236
 237     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 238     trace_iscsi_aio_writev(iscsi, sector_num, nb_sectors, opaque, acb);
 239
 240     acb->iscsilun = iscsilun;
 241     acb->qiov     = qiov;
 242
 243     acb->canceled   = 0;
 244     acb->bh         = NULL;
 245     acb->status     = -EINPROGRESS;
 246
 247     /* XXX we should pass the iovec to write16 to avoid the extra copy */
 248     /* this will allow us to get rid of 'buf' completely */
 249     size = nb_sectors * BDRV_SECTOR_SIZE;
 250     acb->buf = g_malloc(size);
 251     qemu_iovec_to_buf(acb->qiov, 0, acb->buf, size);
 252
 253     acb->task = malloc(sizeof(struct scsi_task));
 254     if (acb->task == NULL) {
 255         error_report("iSCSI: Failed to allocate task for scsi WRITE16 "
 256                      "command. %s", iscsi_get_error(iscsi));
 257         qemu_aio_release(acb);
 258         return NULL;
 259     }
 260     memset(acb->task, 0, sizeof(struct scsi_task));
 261
 262     acb->task->xfer_dir = SCSI_XFER_WRITE;
 263     acb->task->cdb_size = 16;
 264     acb->task->cdb[0] = 0x8a;
 265     lba = sector_qemu2lun(sector_num, iscsilun);
 266     *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 267     *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 268     num_sectors = size / iscsilun->block_size;
 269     *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 270     acb->task->expxferlen = size;
 271
 272     data.data = acb->buf;
 273     data.size = size;
 274
 275     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 276                                  iscsi_aio_write16_cb,
 277                                  &data,
 278                                  acb) != 0) {
 279         scsi_free_scsi_task(acb->task);
 280         g_free(acb->buf);
 281         qemu_aio_release(acb);
 282         return NULL;
 283     }
 284
 285     iscsi_set_events(iscsilun);
 286
 287     return &acb->common;
 288 }
 289
 290 static void
 291 iscsi_aio_read16_cb(struct iscsi_context *iscsi, int status,
 292                     void *command_data, void *opaque)
 293 {
 294     IscsiAIOCB *acb = opaque;
 295
 296     trace_iscsi_aio_read16_cb(iscsi, status, acb, acb->canceled);
 297
 298     if (acb->canceled != 0) {
 299         return;
 300     }
 301
 302     acb->status = 0;
 303     if (status != 0) {
 304         error_report("Failed to read16 data from iSCSI lun. %s",
 305                      iscsi_get_error(iscsi));
 306         acb->status = -EIO;
 307     }
 308
 309     iscsi_schedule_bh(acb);
 310 }
 311
 312 static BlockDriverAIOCB *
 313 iscsi_aio_readv(BlockDriverState *bs, int64_t sector_num,
 314                 QEMUIOVector *qiov, int nb_sectors,
 315                 BlockDriverCompletionFunc *cb,
 316                 void *opaque)
 317 {
 318     IscsiLun *iscsilun = bs->opaque;
 319     struct iscsi_context *iscsi = iscsilun->iscsi;
 320     IscsiAIOCB *acb;
 321     size_t qemu_read_size;
 322     int i;
 323     uint64_t lba;
 324     uint32_t num_sectors;
 325
 326     qemu_read_size = BDRV_SECTOR_SIZE * (size_t)nb_sectors;
 327
 328     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 329     trace_iscsi_aio_readv(iscsi, sector_num, nb_sectors, opaque, acb);
 330
 331     acb->iscsilun = iscsilun;
 332     acb->qiov     = qiov;
 333
 334     acb->canceled    = 0;
 335     acb->bh          = NULL;
 336     acb->status      = -EINPROGRESS;
 337     acb->read_size   = qemu_read_size;
 338     acb->buf         = NULL;
 339
 340     /* If LUN blocksize is bigger than BDRV_BLOCK_SIZE a read from QEMU
 341      * may be misaligned to the LUN, so we may need to read some extra
 342      * data.
 343      */
 344     acb->read_offset = 0;
 345     if (iscsilun->block_size > BDRV_SECTOR_SIZE) {
 346         uint64_t bdrv_offset = BDRV_SECTOR_SIZE * sector_num;
 347
 348         acb->read_offset  = bdrv_offset % iscsilun->block_size;
 349     }
 350
 351     num_sectors  = (qemu_read_size + iscsilun->block_size
 352                     + acb->read_offset - 1)
 353                     / iscsilun->block_size;
 354
 355     acb->task = malloc(sizeof(struct scsi_task));
 356     if (acb->task == NULL) {
 357         error_report("iSCSI: Failed to allocate task for scsi READ16 "
 358                      "command. %s", iscsi_get_error(iscsi));
 359         qemu_aio_release(acb);
 360         return NULL;
 361     }
 362     memset(acb->task, 0, sizeof(struct scsi_task));
 363
 364     acb->task->xfer_dir = SCSI_XFER_READ;
 365     lba = sector_qemu2lun(sector_num, iscsilun);
 366     acb->task->expxferlen = qemu_read_size;
 367
 368     switch (iscsilun->type) {
 369     case TYPE_DISK:
 370         acb->task->cdb_size = 16;
 371         acb->task->cdb[0]  = 0x88;
 372         *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 373         *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 374         *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 375         break;
 376     default:
 377         acb->task->cdb_size = 10;
 378         acb->task->cdb[0]  = 0x28;
 379         *(uint32_t *)&acb->task->cdb[2] = htonl(lba);
 380         *(uint16_t *)&acb->task->cdb[7] = htons(num_sectors);
 381         break;
 382     }
 383
 384     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 385                                  iscsi_aio_read16_cb,
 386                                  NULL,
 387                                  acb) != 0) {
 388         scsi_free_scsi_task(acb->task);
 389         qemu_aio_release(acb);
 390         return NULL;
 391     }
 392
 393     for (i = 0; i < acb->qiov->niov; i++) {
 394         scsi_task_add_data_in_buffer(acb->task,
 395                 acb->qiov->iov[i].iov_len,
 396                 acb->qiov->iov[i].iov_base);
 397     }
 398
 399     iscsi_set_events(iscsilun);
 400
 401     return &acb->common;
 402 }
 403
 404
 405 static void
 406 iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
 407                      void *command_data, void *opaque)
 408 {
 409     IscsiAIOCB *acb = opaque;
 410
 411     if (acb->canceled != 0) {
 412         return;
 413     }
 414
 415     acb->status = 0;
 416     if (status < 0) {
 417         error_report("Failed to sync10 data on iSCSI lun. %s",
 418                      iscsi_get_error(iscsi));
 419         acb->status = -EIO;
 420     }
 421
 422     iscsi_schedule_bh(acb);
 423 }
 424
 425 static BlockDriverAIOCB *
 426 iscsi_aio_flush(BlockDriverState *bs,
 427                 BlockDriverCompletionFunc *cb, void *opaque)
 428 {
 429     IscsiLun *iscsilun = bs->opaque;
 430     struct iscsi_context *iscsi = iscsilun->iscsi;
 431     IscsiAIOCB *acb;
 432
 433     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 434
 435     acb->iscsilun = iscsilun;
 436     acb->canceled   = 0;
 437     acb->bh         = NULL;
 438     acb->status     = -EINPROGRESS;
 439
 440     acb->task = iscsi_synchronizecache10_task(iscsi, iscsilun->lun,
 441                                          0, 0, 0, 0,
 442                                          iscsi_synccache10_cb,
 443                                          acb);
 444     if (acb->task == NULL) {
 445         error_report("iSCSI: Failed to send synchronizecache10 command. %s",
 446                      iscsi_get_error(iscsi));
 447         qemu_aio_release(acb);
 448         return NULL;
 449     }
 450
 451     iscsi_set_events(iscsilun);
 452
 453     return &acb->common;
 454 }
 455
 456 static void
 457 iscsi_unmap_cb(struct iscsi_context *iscsi, int status,
 458                      void *command_data, void *opaque)
 459 {
 460     IscsiAIOCB *acb = opaque;
 461
 462     if (acb->canceled != 0) {
 463         return;
 464     }
 465
 466     acb->status = 0;
 467     if (status < 0) {
 468         error_report("Failed to unmap data on iSCSI lun. %s",
 469                      iscsi_get_error(iscsi));
 470         acb->status = -EIO;
 471     }
 472
 473     iscsi_schedule_bh(acb);
 474 }
 475
 476 static BlockDriverAIOCB *
 477 iscsi_aio_discard(BlockDriverState *bs,
 478                   int64_t sector_num, int nb_sectors,
 479                   BlockDriverCompletionFunc *cb, void *opaque)
 480 {
 481     IscsiLun *iscsilun = bs->opaque;
 482     struct iscsi_context *iscsi = iscsilun->iscsi;
 483     IscsiAIOCB *acb;
 484     struct unmap_list list[1];
 485
 486     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 487
 488     acb->iscsilun = iscsilun;
 489     acb->canceled   = 0;
 490     acb->bh         = NULL;
 491     acb->status     = -EINPROGRESS;
 492
 493     list[0].lba = sector_qemu2lun(sector_num, iscsilun);
 494     list[0].num = nb_sectors * BDRV_SECTOR_SIZE / iscsilun->block_size;
 495
 496     acb->task = iscsi_unmap_task(iscsi, iscsilun->lun,
 497                                  0, 0, &list[0], 1,
 498                                  iscsi_unmap_cb,
 499                                  acb);
 500     if (acb->task == NULL) {
 501         error_report("iSCSI: Failed to send unmap command. %s",
 502                      iscsi_get_error(iscsi));
 503         qemu_aio_release(acb);
 504         return NULL;
 505     }
 506
 507     iscsi_set_events(iscsilun);
 508
 509     return &acb->common;
 510 }
 511
 512 #ifdef __linux__
 513 static void
 514 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 515                      void *command_data, void *opaque)
 516 {
 517     IscsiAIOCB *acb = opaque;
 518
 519     if (acb->canceled != 0) {
 520         return;
 521     }
 522
 523     acb->status = 0;
 524     if (status < 0) {
 525         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 526                      iscsi_get_error(iscsi));
 527         acb->status = -EIO;
 528     }
 529
 530     acb->ioh->driver_status = 0;
 531     acb->ioh->host_status   = 0;
 532     acb->ioh->resid         = 0;
 533
 534 #define SG_ERR_DRIVER_SENSE    0x08
 535
 536     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 537         int ss;
 538
 539         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 540
 541         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 542         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 543              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 544         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 545     }
 546
 547     iscsi_schedule_bh(acb);
 548 }
 549
 550 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 551         unsigned long int req, void *buf,
 552         BlockDriverCompletionFunc *cb, void *opaque)
 553 {
 554     IscsiLun *iscsilun = bs->opaque;
 555     struct iscsi_context *iscsi = iscsilun->iscsi;
 556     struct iscsi_data data;
 557     IscsiAIOCB *acb;
 558
 559     assert(req == SG_IO);
 560
 561     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 562
 563     acb->iscsilun = iscsilun;
 564     acb->canceled    = 0;
 565     acb->bh          = NULL;
 566     acb->status      = -EINPROGRESS;
 567     acb->buf         = NULL;
 568     acb->ioh         = buf;
 569
 570     acb->task = malloc(sizeof(struct scsi_task));
 571     if (acb->task == NULL) {
 572         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 573                      iscsi_get_error(iscsi));
 574         qemu_aio_release(acb);
 575         return NULL;
 576     }
 577     memset(acb->task, 0, sizeof(struct scsi_task));
 578
 579     switch (acb->ioh->dxfer_direction) {
 580     case SG_DXFER_TO_DEV:
 581         acb->task->xfer_dir = SCSI_XFER_WRITE;
 582         break;
 583     case SG_DXFER_FROM_DEV:
 584         acb->task->xfer_dir = SCSI_XFER_READ;
 585         break;
 586     default:
 587         acb->task->xfer_dir = SCSI_XFER_NONE;
 588         break;
 589     }
 590
 591     acb->task->cdb_size = acb->ioh->cmd_len;
 592     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 593     acb->task->expxferlen = acb->ioh->dxfer_len;
 594
 595     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 596         data.data = acb->ioh->dxferp;
 597         data.size = acb->ioh->dxfer_len;
 598     }
 599     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 600                                  iscsi_aio_ioctl_cb,
 601                                  (acb->task->xfer_dir == SCSI_XFER_WRITE) ?
 602                                      &data : NULL,
 603                                  acb) != 0) {
 604         scsi_free_scsi_task(acb->task);
 605         qemu_aio_release(acb);
 606         return NULL;
 607     }
 608
 609     /* tell libiscsi to read straight into the buffer we got from ioctl */
 610     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 611         scsi_task_add_data_in_buffer(acb->task,
 612                                      acb->ioh->dxfer_len,
 613                                      acb->ioh->dxferp);
 614     }
 615
 616     iscsi_set_events(iscsilun);
 617
 618     return &acb->common;
 619 }
 620
 621
 622 static void ioctl_cb(void *opaque, int status)
 623 {
 624     int *p_status = opaque;
 625     *p_status = status;
 626 }
 627
 628 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 629 {
 630     IscsiLun *iscsilun = bs->opaque;
 631     int status;
 632
 633     switch (req) {
 634     case SG_GET_VERSION_NUM:
 635         *(int *)buf = 30000;
 636         break;
 637     case SG_GET_SCSI_ID:
 638         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 639         break;
 640     case SG_IO:
 641         status = -EINPROGRESS;
 642         iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status);
 643
 644         while (status == -EINPROGRESS) {
 645             qemu_aio_wait();
 646         }
 647
 648         return 0;
 649     default:
 650         return -1;
 651     }
 652     return 0;
 653 }
 654 #endif
 655
 656 static int64_t
 657 iscsi_getlength(BlockDriverState *bs)
 658 {
 659     IscsiLun *iscsilun = bs->opaque;
 660     int64_t len;
 661
 662     len  = iscsilun->num_blocks;
 663     len *= iscsilun->block_size;
 664
 665     return len;
 666 }
 667
 668 static void
 669 iscsi_readcapacity16_cb(struct iscsi_context *iscsi, int status,
 670                         void *command_data, void *opaque)
 671 {
 672     struct IscsiTask *itask = opaque;
 673     struct scsi_readcapacity16 *rc16;
 674     struct scsi_task *task = command_data;
 675
 676     if (status != 0) {
 677         error_report("iSCSI: Failed to read capacity of iSCSI lun. %s",
 678                      iscsi_get_error(iscsi));
 679         itask->status   = 1;
 680         itask->complete = 1;
 681         scsi_free_scsi_task(task);
 682         return;
 683     }
 684
 685     rc16 = scsi_datain_unmarshall(task);
 686     if (rc16 == NULL) {
 687         error_report("iSCSI: Failed to unmarshall readcapacity16 data.");
 688         itask->status   = 1;
 689         itask->complete = 1;
 690         scsi_free_scsi_task(task);
 691         return;
 692     }
 693
 694     itask->iscsilun->block_size = rc16->block_length;
 695     itask->iscsilun->num_blocks = rc16->returned_lba + 1;
 696     itask->bs->total_sectors    = itask->iscsilun->num_blocks *
 697                                itask->iscsilun->block_size / BDRV_SECTOR_SIZE ;
 698
 699     itask->status   = 0;
 700     itask->complete = 1;
 701     scsi_free_scsi_task(task);
 702 }
 703
 704 static void
 705 iscsi_readcapacity10_cb(struct iscsi_context *iscsi, int status,
 706                         void *command_data, void *opaque)
 707 {
 708     struct IscsiTask *itask = opaque;
 709     struct scsi_readcapacity10 *rc10;
 710     struct scsi_task *task = command_data;
 711
 712     if (status != 0) {
 713         error_report("iSCSI: Failed to read capacity of iSCSI lun. %s",
 714                      iscsi_get_error(iscsi));
 715         itask->status   = 1;
 716         itask->complete = 1;
 717         scsi_free_scsi_task(task);
 718         return;
 719     }
 720
 721     rc10 = scsi_datain_unmarshall(task);
 722     if (rc10 == NULL) {
 723         error_report("iSCSI: Failed to unmarshall readcapacity10 data.");
 724         itask->status   = 1;
 725         itask->complete = 1;
 726         scsi_free_scsi_task(task);
 727         return;
 728     }
 729
 730     itask->iscsilun->block_size = rc10->block_size;
 731     if (rc10->lba == 0) {
 732         /* blank disk loaded */
 733         itask->iscsilun->num_blocks = 0;
 734     } else {
 735         itask->iscsilun->num_blocks = rc10->lba + 1;
 736     }
 737     itask->bs->total_sectors    = itask->iscsilun->num_blocks *
 738                                itask->iscsilun->block_size / BDRV_SECTOR_SIZE ;
 739
 740     itask->status   = 0;
 741     itask->complete = 1;
 742     scsi_free_scsi_task(task);
 743 }
 744
 745 static void
 746 iscsi_inquiry_cb(struct iscsi_context *iscsi, int status, void *command_data,
 747                  void *opaque)
 748 {
 749     struct IscsiTask *itask = opaque;
 750     struct scsi_task *task = command_data;
 751     struct scsi_inquiry_standard *inq;
 752
 753     if (status != 0) {
 754         itask->status   = 1;
 755         itask->complete = 1;
 756         scsi_free_scsi_task(task);
 757         return;
 758     }
 759
 760     inq = scsi_datain_unmarshall(task);
 761     if (inq == NULL) {
 762         error_report("iSCSI: Failed to unmarshall inquiry data.");
 763         itask->status   = 1;
 764         itask->complete = 1;
 765         scsi_free_scsi_task(task);
 766         return;
 767     }
 768
 769     itask->iscsilun->type = inq->periperal_device_type;
 770
 771     scsi_free_scsi_task(task);
 772
 773     switch (itask->iscsilun->type) {
 774     case TYPE_DISK:
 775         task = iscsi_readcapacity16_task(iscsi, itask->iscsilun->lun,
 776                                    iscsi_readcapacity16_cb, opaque);
 777         if (task == NULL) {
 778             error_report("iSCSI: failed to send readcapacity16 command.");
 779             itask->status   = 1;
 780             itask->complete = 1;
 781             return;
 782         }
 783         break;
 784     case TYPE_ROM:
 785         task = iscsi_readcapacity10_task(iscsi, itask->iscsilun->lun,
 786                                    0, 0,
 787                                    iscsi_readcapacity10_cb, opaque);
 788         if (task == NULL) {
 789             error_report("iSCSI: failed to send readcapacity16 command.");
 790             itask->status   = 1;
 791             itask->complete = 1;
 792             return;
 793         }
 794         break;
 795     default:
 796         itask->status   = 0;
 797         itask->complete = 1;
 798     }
 799 }
 800
 801 static void
 802 iscsi_connect_cb(struct iscsi_context *iscsi, int status, void *command_data,
 803                  void *opaque)
 804 {
 805     struct IscsiTask *itask = opaque;
 806     struct scsi_task *task;
 807
 808     if (status != 0) {
 809         itask->status   = 1;
 810         itask->complete = 1;
 811         return;
 812     }
 813
 814     task = iscsi_inquiry_task(iscsi, itask->iscsilun->lun,
 815                               0, 0, 36,
 816                               iscsi_inquiry_cb, opaque);
 817     if (task == NULL) {
 818         error_report("iSCSI: failed to send inquiry command.");
 819         itask->status   = 1;
 820         itask->complete = 1;
 821         return;
 822     }
 823 }
 824
 825 static int parse_chap(struct iscsi_context *iscsi, const char *target)
 826 {
 827     QemuOptsList *list;
 828     QemuOpts *opts;
 829     const char *user = NULL;
 830     const char *password = NULL;
 831
 832     list = qemu_find_opts("iscsi");
 833     if (!list) {
 834         return 0;
 835     }
 836
 837     opts = qemu_opts_find(list, target);
 838     if (opts == NULL) {
 839         opts = QTAILQ_FIRST(&list->head);
 840         if (!opts) {
 841             return 0;
 842         }
 843     }
 844
 845     user = qemu_opt_get(opts, "user");
 846     if (!user) {
 847         return 0;
 848     }
 849
 850     password = qemu_opt_get(opts, "password");
 851     if (!password) {
 852         error_report("CHAP username specified but no password was given");
 853         return -1;
 854     }
 855
 856     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
 857         error_report("Failed to set initiator username and password");
 858         return -1;
 859     }
 860
 861     return 0;
 862 }
 863
 864 static void parse_header_digest(struct iscsi_context *iscsi, const char *target)
 865 {
 866     QemuOptsList *list;
 867     QemuOpts *opts;
 868     const char *digest = NULL;
 869
 870     list = qemu_find_opts("iscsi");
 871     if (!list) {
 872         return;
 873     }
 874
 875     opts = qemu_opts_find(list, target);
 876     if (opts == NULL) {
 877         opts = QTAILQ_FIRST(&list->head);
 878         if (!opts) {
 879             return;
 880         }
 881     }
 882
 883     digest = qemu_opt_get(opts, "header-digest");
 884     if (!digest) {
 885         return;
 886     }
 887
 888     if (!strcmp(digest, "CRC32C")) {
 889         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
 890     } else if (!strcmp(digest, "NONE")) {
 891         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
 892     } else if (!strcmp(digest, "CRC32C-NONE")) {
 893         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
 894     } else if (!strcmp(digest, "NONE-CRC32C")) {
 895         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 896     } else {
 897         error_report("Invalid header-digest setting : %s", digest);
 898     }
 899 }
 900
 901 static char *parse_initiator_name(const char *target)
 902 {
 903     QemuOptsList *list;
 904     QemuOpts *opts;
 905     const char *name = NULL;
 906     const char *iscsi_name = qemu_get_vm_name();
 907
 908     list = qemu_find_opts("iscsi");
 909     if (list) {
 910         opts = qemu_opts_find(list, target);
 911         if (!opts) {
 912             opts = QTAILQ_FIRST(&list->head);
 913         }
 914         if (opts) {
 915             name = qemu_opt_get(opts, "initiator-name");
 916         }
 917     }
 918
 919     if (name) {
 920         return g_strdup(name);
 921     } else {
 922         return g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
 923                                iscsi_name ? ":" : "",
 924                                iscsi_name ? iscsi_name : "");
 925     }
 926 }
 927
 928 /*
 929  * We support iscsi url's on the form
 930  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
 931  */
 932 static int iscsi_open(BlockDriverState *bs, const char *filename, int flags)
 933 {
 934     IscsiLun *iscsilun = bs->opaque;
 935     struct iscsi_context *iscsi = NULL;
 936     struct iscsi_url *iscsi_url = NULL;
 937     struct IscsiTask task;
 938     char *initiator_name = NULL;
 939     int ret;
 940
 941     if ((BDRV_SECTOR_SIZE % 512) != 0) {
 942         error_report("iSCSI: Invalid BDRV_SECTOR_SIZE. "
 943                      "BDRV_SECTOR_SIZE(%lld) is not a multiple "
 944                      "of 512", BDRV_SECTOR_SIZE);
 945         return -EINVAL;
 946     }
 947
 948     iscsi_url = iscsi_parse_full_url(iscsi, filename);
 949     if (iscsi_url == NULL) {
 950         error_report("Failed to parse URL : %s %s", filename,
 951                      iscsi_get_error(iscsi));
 952         ret = -EINVAL;
 953         goto out;
 954     }
 955
 956     memset(iscsilun, 0, sizeof(IscsiLun));
 957
 958     initiator_name = parse_initiator_name(iscsi_url->target);
 959
 960     iscsi = iscsi_create_context(initiator_name);
 961     if (iscsi == NULL) {
 962         error_report("iSCSI: Failed to create iSCSI context.");
 963         ret = -ENOMEM;
 964         goto out;
 965     }
 966
 967     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
 968         error_report("iSCSI: Failed to set target name.");
 969         ret = -EINVAL;
 970         goto out;
 971     }
 972
 973     if (iscsi_url->user != NULL) {
 974         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
 975                                               iscsi_url->passwd);
 976         if (ret != 0) {
 977             error_report("Failed to set initiator username and password");
 978             ret = -EINVAL;
 979             goto out;
 980         }
 981     }
 982
 983     /* check if we got CHAP username/password via the options */
 984     if (parse_chap(iscsi, iscsi_url->target) != 0) {
 985         error_report("iSCSI: Failed to set CHAP user/password");
 986         ret = -EINVAL;
 987         goto out;
 988     }
 989
 990     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
 991         error_report("iSCSI: Failed to set session type to normal.");
 992         ret = -EINVAL;
 993         goto out;
 994     }
 995
 996     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 997
 998     /* check if we got HEADER_DIGEST via the options */
 999     parse_header_digest(iscsi, iscsi_url->target);
1000
1001     task.iscsilun = iscsilun;
1002     task.status = 0;
1003     task.complete = 0;
1004     task.bs = bs;
1005
1006     iscsilun->iscsi = iscsi;
1007     iscsilun->lun   = iscsi_url->lun;
1008
1009     if (iscsi_full_connect_async(iscsi, iscsi_url->portal, iscsi_url->lun,
1010                                  iscsi_connect_cb, &task)
1011         != 0) {
1012         error_report("iSCSI: Failed to start async connect.");
1013         ret = -EINVAL;
1014         goto out;
1015     }
1016
1017     while (!task.complete) {
1018         iscsi_set_events(iscsilun);
1019         qemu_aio_wait();
1020     }
1021     if (task.status != 0) {
1022         error_report("iSCSI: Failed to connect to LUN : %s",
1023                      iscsi_get_error(iscsi));
1024         ret = -EINVAL;
1025         goto out;
1026     }
1027
1028     /* Medium changer or tape. We dont have any emulation for this so this must
1029      * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
1030      * to read from the device to guess the image format.
1031      */
1032     if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
1033         iscsilun->type == TYPE_TAPE) {
1034         bs->sg = 1;
1035     }
1036
1037     ret = 0;
1038
1039 out:
1040     if (initiator_name != NULL) {
1041         g_free(initiator_name);
1042     }
1043     if (iscsi_url != NULL) {
1044         iscsi_destroy_url(iscsi_url);
1045     }
1046
1047     if (ret) {
1048         if (iscsi != NULL) {
1049             iscsi_destroy_context(iscsi);
1050         }
1051         memset(iscsilun, 0, sizeof(IscsiLun));
1052     }
1053     return ret;
1054 }
1055
1056 static void iscsi_close(BlockDriverState *bs)
1057 {
1058     IscsiLun *iscsilun = bs->opaque;
1059     struct iscsi_context *iscsi = iscsilun->iscsi;
1060
1061     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL, NULL);
1062     iscsi_destroy_context(iscsi);
1063     memset(iscsilun, 0, sizeof(IscsiLun));
1064 }
1065
1066 static BlockDriver bdrv_iscsi = {
1067     .format_name     = "iscsi",
1068     .protocol_name   = "iscsi",
1069
1070     .instance_size   = sizeof(IscsiLun),
1071     .bdrv_file_open  = iscsi_open,
1072     .bdrv_close      = iscsi_close,
1073
1074     .bdrv_getlength  = iscsi_getlength,
1075
1076     .bdrv_aio_readv  = iscsi_aio_readv,
1077     .bdrv_aio_writev = iscsi_aio_writev,
1078     .bdrv_aio_flush  = iscsi_aio_flush,
1079
1080     .bdrv_aio_discard = iscsi_aio_discard,
1081
1082 #ifdef __linux__
1083     .bdrv_ioctl       = iscsi_ioctl,
1084     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1085 #endif
1086 };
1087
1088 static void iscsi_block_init(void)
1089 {
1090     bdrv_register(&bdrv_iscsi);
1091 }
1092
1093 block_init(iscsi_block_init);