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_readv_writev_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     qemu_aio_release(acb);
  87 }
  88
  89 static int
  90 iscsi_schedule_bh(QEMUBHFunc *cb, IscsiAIOCB *acb)
  91 {
  92     acb->bh = qemu_bh_new(cb, acb);
  93     if (!acb->bh) {
  94         error_report("oom: could not create iscsi bh");
  95         return -EIO;
  96     }
  97
  98     qemu_bh_schedule(acb->bh);
  99     return 0;
 100 }
 101
 102
 103 static void
 104 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
 105                     void *private_data)
 106 {
 107 }
 108
 109 static void
 110 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
 111 {
 112     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
 113     IscsiLun *iscsilun = acb->iscsilun;
 114
 115     acb->common.cb(acb->common.opaque, -ECANCELED);
 116     acb->canceled = 1;
 117
 118     /* send a task mgmt call to the target to cancel the task on the target */
 119     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
 120                                      iscsi_abort_task_cb, NULL);
 121
 122     /* then also cancel the task locally in libiscsi */
 123     iscsi_scsi_task_cancel(iscsilun->iscsi, acb->task);
 124 }
 125
 126 static AIOPool iscsi_aio_pool = {
 127     .aiocb_size         = sizeof(IscsiAIOCB),
 128     .cancel             = iscsi_aio_cancel,
 129 };
 130
 131
 132 static void iscsi_process_read(void *arg);
 133 static void iscsi_process_write(void *arg);
 134
 135 static int iscsi_process_flush(void *arg)
 136 {
 137     IscsiLun *iscsilun = arg;
 138
 139     return iscsi_queue_length(iscsilun->iscsi) > 0;
 140 }
 141
 142 static void
 143 iscsi_set_events(IscsiLun *iscsilun)
 144 {
 145     struct iscsi_context *iscsi = iscsilun->iscsi;
 146     int ev;
 147
 148     /* We always register a read handler.  */
 149     ev = POLLIN;
 150     ev |= iscsi_which_events(iscsi);
 151     if (ev != iscsilun->events) {
 152         qemu_aio_set_fd_handler(iscsi_get_fd(iscsi),
 153                       iscsi_process_read,
 154                       (ev & POLLOUT) ? iscsi_process_write : NULL,
 155                       iscsi_process_flush,
 156                       iscsilun);
 157
 158     }
 159
 160     /* If we just added an event, the callback might be delayed
 161      * unless we call qemu_notify_event().
 162      */
 163     if (ev & ~iscsilun->events) {
 164         qemu_notify_event();
 165     }
 166     iscsilun->events = ev;
 167 }
 168
 169 static void
 170 iscsi_process_read(void *arg)
 171 {
 172     IscsiLun *iscsilun = arg;
 173     struct iscsi_context *iscsi = iscsilun->iscsi;
 174
 175     iscsi_service(iscsi, POLLIN);
 176     iscsi_set_events(iscsilun);
 177 }
 178
 179 static void
 180 iscsi_process_write(void *arg)
 181 {
 182     IscsiLun *iscsilun = arg;
 183     struct iscsi_context *iscsi = iscsilun->iscsi;
 184
 185     iscsi_service(iscsi, POLLOUT);
 186     iscsi_set_events(iscsilun);
 187 }
 188
 189
 190 static void
 191 iscsi_aio_write16_cb(struct iscsi_context *iscsi, int status,
 192                      void *command_data, void *opaque)
 193 {
 194     IscsiAIOCB *acb = opaque;
 195
 196     trace_iscsi_aio_write16_cb(iscsi, status, acb, acb->canceled);
 197
 198     g_free(acb->buf);
 199
 200     if (acb->canceled != 0) {
 201         qemu_aio_release(acb);
 202         scsi_free_scsi_task(acb->task);
 203         acb->task = NULL;
 204         return;
 205     }
 206
 207     acb->status = 0;
 208     if (status < 0) {
 209         error_report("Failed to write16 data to iSCSI lun. %s",
 210                      iscsi_get_error(iscsi));
 211         acb->status = -EIO;
 212     }
 213
 214     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 215     scsi_free_scsi_task(acb->task);
 216     acb->task = NULL;
 217 }
 218
 219 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
 220 {
 221     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
 222 }
 223
 224 static BlockDriverAIOCB *
 225 iscsi_aio_writev(BlockDriverState *bs, int64_t sector_num,
 226                  QEMUIOVector *qiov, int nb_sectors,
 227                  BlockDriverCompletionFunc *cb,
 228                  void *opaque)
 229 {
 230     IscsiLun *iscsilun = bs->opaque;
 231     struct iscsi_context *iscsi = iscsilun->iscsi;
 232     IscsiAIOCB *acb;
 233     size_t size;
 234     uint32_t num_sectors;
 235     uint64_t lba;
 236     struct iscsi_data data;
 237
 238     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 239     trace_iscsi_aio_writev(iscsi, sector_num, nb_sectors, opaque, acb);
 240
 241     acb->iscsilun = iscsilun;
 242     acb->qiov     = qiov;
 243
 244     acb->canceled   = 0;
 245
 246     /* XXX we should pass the iovec to write16 to avoid the extra copy */
 247     /* this will allow us to get rid of 'buf' completely */
 248     size = nb_sectors * BDRV_SECTOR_SIZE;
 249     acb->buf = g_malloc(size);
 250     qemu_iovec_to_buf(acb->qiov, 0, acb->buf, size);
 251
 252     acb->task = malloc(sizeof(struct scsi_task));
 253     if (acb->task == NULL) {
 254         error_report("iSCSI: Failed to allocate task for scsi WRITE16 "
 255                      "command. %s", iscsi_get_error(iscsi));
 256         qemu_aio_release(acb);
 257         return NULL;
 258     }
 259     memset(acb->task, 0, sizeof(struct scsi_task));
 260
 261     acb->task->xfer_dir = SCSI_XFER_WRITE;
 262     acb->task->cdb_size = 16;
 263     acb->task->cdb[0] = 0x8a;
 264     if (!(bs->open_flags & BDRV_O_CACHE_WB)) {
 265         /* set FUA on writes when cache mode is write through */
 266         acb->task->cdb[1] |= 0x04;
 267     }
 268     lba = sector_qemu2lun(sector_num, iscsilun);
 269     *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 270     *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 271     num_sectors = size / iscsilun->block_size;
 272     *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 273     acb->task->expxferlen = size;
 274
 275     data.data = acb->buf;
 276     data.size = size;
 277
 278     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 279                                  iscsi_aio_write16_cb,
 280                                  &data,
 281                                  acb) != 0) {
 282         scsi_free_scsi_task(acb->task);
 283         g_free(acb->buf);
 284         qemu_aio_release(acb);
 285         return NULL;
 286     }
 287
 288     iscsi_set_events(iscsilun);
 289
 290     return &acb->common;
 291 }
 292
 293 static void
 294 iscsi_aio_read16_cb(struct iscsi_context *iscsi, int status,
 295                     void *command_data, void *opaque)
 296 {
 297     IscsiAIOCB *acb = opaque;
 298
 299     trace_iscsi_aio_read16_cb(iscsi, status, acb, acb->canceled);
 300
 301     if (acb->canceled != 0) {
 302         qemu_aio_release(acb);
 303         scsi_free_scsi_task(acb->task);
 304         acb->task = NULL;
 305         return;
 306     }
 307
 308     acb->status = 0;
 309     if (status != 0) {
 310         error_report("Failed to read16 data from iSCSI lun. %s",
 311                      iscsi_get_error(iscsi));
 312         acb->status = -EIO;
 313     }
 314
 315     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 316     scsi_free_scsi_task(acb->task);
 317     acb->task = NULL;
 318 }
 319
 320 static BlockDriverAIOCB *
 321 iscsi_aio_readv(BlockDriverState *bs, int64_t sector_num,
 322                 QEMUIOVector *qiov, int nb_sectors,
 323                 BlockDriverCompletionFunc *cb,
 324                 void *opaque)
 325 {
 326     IscsiLun *iscsilun = bs->opaque;
 327     struct iscsi_context *iscsi = iscsilun->iscsi;
 328     IscsiAIOCB *acb;
 329     size_t qemu_read_size;
 330     int i;
 331     uint64_t lba;
 332     uint32_t num_sectors;
 333
 334     qemu_read_size = BDRV_SECTOR_SIZE * (size_t)nb_sectors;
 335
 336     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 337     trace_iscsi_aio_readv(iscsi, sector_num, nb_sectors, opaque, acb);
 338
 339     acb->iscsilun = iscsilun;
 340     acb->qiov     = qiov;
 341
 342     acb->canceled    = 0;
 343     acb->read_size   = qemu_read_size;
 344     acb->buf         = NULL;
 345
 346     /* If LUN blocksize is bigger than BDRV_BLOCK_SIZE a read from QEMU
 347      * may be misaligned to the LUN, so we may need to read some extra
 348      * data.
 349      */
 350     acb->read_offset = 0;
 351     if (iscsilun->block_size > BDRV_SECTOR_SIZE) {
 352         uint64_t bdrv_offset = BDRV_SECTOR_SIZE * sector_num;
 353
 354         acb->read_offset  = bdrv_offset % iscsilun->block_size;
 355     }
 356
 357     num_sectors  = (qemu_read_size + iscsilun->block_size
 358                     + acb->read_offset - 1)
 359                     / iscsilun->block_size;
 360
 361     acb->task = malloc(sizeof(struct scsi_task));
 362     if (acb->task == NULL) {
 363         error_report("iSCSI: Failed to allocate task for scsi READ16 "
 364                      "command. %s", iscsi_get_error(iscsi));
 365         qemu_aio_release(acb);
 366         return NULL;
 367     }
 368     memset(acb->task, 0, sizeof(struct scsi_task));
 369
 370     acb->task->xfer_dir = SCSI_XFER_READ;
 371     lba = sector_qemu2lun(sector_num, iscsilun);
 372     acb->task->expxferlen = qemu_read_size;
 373
 374     switch (iscsilun->type) {
 375     case TYPE_DISK:
 376         acb->task->cdb_size = 16;
 377         acb->task->cdb[0]  = 0x88;
 378         *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 379         *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 380         *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 381         break;
 382     default:
 383         acb->task->cdb_size = 10;
 384         acb->task->cdb[0]  = 0x28;
 385         *(uint32_t *)&acb->task->cdb[2] = htonl(lba);
 386         *(uint16_t *)&acb->task->cdb[7] = htons(num_sectors);
 387         break;
 388     }
 389
 390     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 391                                  iscsi_aio_read16_cb,
 392                                  NULL,
 393                                  acb) != 0) {
 394         scsi_free_scsi_task(acb->task);
 395         qemu_aio_release(acb);
 396         return NULL;
 397     }
 398
 399     for (i = 0; i < acb->qiov->niov; i++) {
 400         scsi_task_add_data_in_buffer(acb->task,
 401                 acb->qiov->iov[i].iov_len,
 402                 acb->qiov->iov[i].iov_base);
 403     }
 404
 405     iscsi_set_events(iscsilun);
 406
 407     return &acb->common;
 408 }
 409
 410
 411 static void
 412 iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
 413                      void *command_data, void *opaque)
 414 {
 415     IscsiAIOCB *acb = opaque;
 416
 417     if (acb->canceled != 0) {
 418         qemu_aio_release(acb);
 419         scsi_free_scsi_task(acb->task);
 420         acb->task = NULL;
 421         return;
 422     }
 423
 424     acb->status = 0;
 425     if (status < 0) {
 426         error_report("Failed to sync10 data on iSCSI lun. %s",
 427                      iscsi_get_error(iscsi));
 428         acb->status = -EIO;
 429     }
 430
 431     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 432     scsi_free_scsi_task(acb->task);
 433     acb->task = NULL;
 434 }
 435
 436 static BlockDriverAIOCB *
 437 iscsi_aio_flush(BlockDriverState *bs,
 438                 BlockDriverCompletionFunc *cb, void *opaque)
 439 {
 440     IscsiLun *iscsilun = bs->opaque;
 441     struct iscsi_context *iscsi = iscsilun->iscsi;
 442     IscsiAIOCB *acb;
 443
 444     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 445
 446     acb->iscsilun = iscsilun;
 447     acb->canceled   = 0;
 448
 449     acb->task = iscsi_synchronizecache10_task(iscsi, iscsilun->lun,
 450                                          0, 0, 0, 0,
 451                                          iscsi_synccache10_cb,
 452                                          acb);
 453     if (acb->task == NULL) {
 454         error_report("iSCSI: Failed to send synchronizecache10 command. %s",
 455                      iscsi_get_error(iscsi));
 456         qemu_aio_release(acb);
 457         return NULL;
 458     }
 459
 460     iscsi_set_events(iscsilun);
 461
 462     return &acb->common;
 463 }
 464
 465 static void
 466 iscsi_unmap_cb(struct iscsi_context *iscsi, int status,
 467                      void *command_data, void *opaque)
 468 {
 469     IscsiAIOCB *acb = opaque;
 470
 471     if (acb->canceled != 0) {
 472         qemu_aio_release(acb);
 473         scsi_free_scsi_task(acb->task);
 474         acb->task = NULL;
 475         return;
 476     }
 477
 478     acb->status = 0;
 479     if (status < 0) {
 480         error_report("Failed to unmap data on iSCSI lun. %s",
 481                      iscsi_get_error(iscsi));
 482         acb->status = -EIO;
 483     }
 484
 485     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 486     scsi_free_scsi_task(acb->task);
 487     acb->task = NULL;
 488 }
 489
 490 static BlockDriverAIOCB *
 491 iscsi_aio_discard(BlockDriverState *bs,
 492                   int64_t sector_num, int nb_sectors,
 493                   BlockDriverCompletionFunc *cb, void *opaque)
 494 {
 495     IscsiLun *iscsilun = bs->opaque;
 496     struct iscsi_context *iscsi = iscsilun->iscsi;
 497     IscsiAIOCB *acb;
 498     struct unmap_list list[1];
 499
 500     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 501
 502     acb->iscsilun = iscsilun;
 503     acb->canceled   = 0;
 504
 505     list[0].lba = sector_qemu2lun(sector_num, iscsilun);
 506     list[0].num = nb_sectors * BDRV_SECTOR_SIZE / iscsilun->block_size;
 507
 508     acb->task = iscsi_unmap_task(iscsi, iscsilun->lun,
 509                                  0, 0, &list[0], 1,
 510                                  iscsi_unmap_cb,
 511                                  acb);
 512     if (acb->task == NULL) {
 513         error_report("iSCSI: Failed to send unmap command. %s",
 514                      iscsi_get_error(iscsi));
 515         qemu_aio_release(acb);
 516         return NULL;
 517     }
 518
 519     iscsi_set_events(iscsilun);
 520
 521     return &acb->common;
 522 }
 523
 524 #ifdef __linux__
 525 static void
 526 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 527                      void *command_data, void *opaque)
 528 {
 529     IscsiAIOCB *acb = opaque;
 530
 531     if (acb->canceled != 0) {
 532         qemu_aio_release(acb);
 533         scsi_free_scsi_task(acb->task);
 534         acb->task = NULL;
 535         return;
 536     }
 537
 538     acb->status = 0;
 539     if (status < 0) {
 540         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 541                      iscsi_get_error(iscsi));
 542         acb->status = -EIO;
 543     }
 544
 545     acb->ioh->driver_status = 0;
 546     acb->ioh->host_status   = 0;
 547     acb->ioh->resid         = 0;
 548
 549 #define SG_ERR_DRIVER_SENSE    0x08
 550
 551     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 552         int ss;
 553
 554         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 555
 556         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 557         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 558              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 559         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 560     }
 561
 562     iscsi_schedule_bh(iscsi_readv_writev_bh_cb, acb);
 563     scsi_free_scsi_task(acb->task);
 564     acb->task = NULL;
 565 }
 566
 567 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 568         unsigned long int req, void *buf,
 569         BlockDriverCompletionFunc *cb, void *opaque)
 570 {
 571     IscsiLun *iscsilun = bs->opaque;
 572     struct iscsi_context *iscsi = iscsilun->iscsi;
 573     struct iscsi_data data;
 574     IscsiAIOCB *acb;
 575
 576     assert(req == SG_IO);
 577
 578     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 579
 580     acb->iscsilun = iscsilun;
 581     acb->canceled    = 0;
 582     acb->buf         = NULL;
 583     acb->ioh         = buf;
 584
 585     acb->task = malloc(sizeof(struct scsi_task));
 586     if (acb->task == NULL) {
 587         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 588                      iscsi_get_error(iscsi));
 589         qemu_aio_release(acb);
 590         return NULL;
 591     }
 592     memset(acb->task, 0, sizeof(struct scsi_task));
 593
 594     switch (acb->ioh->dxfer_direction) {
 595     case SG_DXFER_TO_DEV:
 596         acb->task->xfer_dir = SCSI_XFER_WRITE;
 597         break;
 598     case SG_DXFER_FROM_DEV:
 599         acb->task->xfer_dir = SCSI_XFER_READ;
 600         break;
 601     default:
 602         acb->task->xfer_dir = SCSI_XFER_NONE;
 603         break;
 604     }
 605
 606     acb->task->cdb_size = acb->ioh->cmd_len;
 607     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 608     acb->task->expxferlen = acb->ioh->dxfer_len;
 609
 610     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 611         data.data = acb->ioh->dxferp;
 612         data.size = acb->ioh->dxfer_len;
 613     }
 614     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 615                                  iscsi_aio_ioctl_cb,
 616                                  (acb->task->xfer_dir == SCSI_XFER_WRITE) ?
 617                                      &data : NULL,
 618                                  acb) != 0) {
 619         scsi_free_scsi_task(acb->task);
 620         qemu_aio_release(acb);
 621         return NULL;
 622     }
 623
 624     /* tell libiscsi to read straight into the buffer we got from ioctl */
 625     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 626         scsi_task_add_data_in_buffer(acb->task,
 627                                      acb->ioh->dxfer_len,
 628                                      acb->ioh->dxferp);
 629     }
 630
 631     iscsi_set_events(iscsilun);
 632
 633     return &acb->common;
 634 }
 635
 636 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 637 {
 638     IscsiLun *iscsilun = bs->opaque;
 639
 640     switch (req) {
 641     case SG_GET_VERSION_NUM:
 642         *(int *)buf = 30000;
 643         break;
 644     case SG_GET_SCSI_ID:
 645         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 646         break;
 647     default:
 648         return -1;
 649     }
 650     return 0;
 651 }
 652 #endif
 653
 654 static int64_t
 655 iscsi_getlength(BlockDriverState *bs)
 656 {
 657     IscsiLun *iscsilun = bs->opaque;
 658     int64_t len;
 659
 660     len  = iscsilun->num_blocks;
 661     len *= iscsilun->block_size;
 662
 663     return len;
 664 }
 665
 666 static void
 667 iscsi_readcapacity16_cb(struct iscsi_context *iscsi, int status,
 668                         void *command_data, void *opaque)
 669 {
 670     struct IscsiTask *itask = opaque;
 671     struct scsi_readcapacity16 *rc16;
 672     struct scsi_task *task = command_data;
 673
 674     if (status != 0) {
 675         error_report("iSCSI: Failed to read capacity of iSCSI lun. %s",
 676                      iscsi_get_error(iscsi));
 677         itask->status   = 1;
 678         itask->complete = 1;
 679         scsi_free_scsi_task(task);
 680         return;
 681     }
 682
 683     rc16 = scsi_datain_unmarshall(task);
 684     if (rc16 == NULL) {
 685         error_report("iSCSI: Failed to unmarshall readcapacity16 data.");
 686         itask->status   = 1;
 687         itask->complete = 1;
 688         scsi_free_scsi_task(task);
 689         return;
 690     }
 691
 692     itask->iscsilun->block_size = rc16->block_length;
 693     itask->iscsilun->num_blocks = rc16->returned_lba + 1;
 694     itask->bs->total_sectors    = itask->iscsilun->num_blocks *
 695                                itask->iscsilun->block_size / BDRV_SECTOR_SIZE ;
 696
 697     itask->status   = 0;
 698     itask->complete = 1;
 699     scsi_free_scsi_task(task);
 700 }
 701
 702 static void
 703 iscsi_readcapacity10_cb(struct iscsi_context *iscsi, int status,
 704                         void *command_data, void *opaque)
 705 {
 706     struct IscsiTask *itask = opaque;
 707     struct scsi_readcapacity10 *rc10;
 708     struct scsi_task *task = command_data;
 709
 710     if (status != 0) {
 711         error_report("iSCSI: Failed to read capacity of iSCSI lun. %s",
 712                      iscsi_get_error(iscsi));
 713         itask->status   = 1;
 714         itask->complete = 1;
 715         scsi_free_scsi_task(task);
 716         return;
 717     }
 718
 719     rc10 = scsi_datain_unmarshall(task);
 720     if (rc10 == NULL) {
 721         error_report("iSCSI: Failed to unmarshall readcapacity10 data.");
 722         itask->status   = 1;
 723         itask->complete = 1;
 724         scsi_free_scsi_task(task);
 725         return;
 726     }
 727
 728     itask->iscsilun->block_size = rc10->block_size;
 729     itask->iscsilun->num_blocks = rc10->lba + 1;
 730     itask->bs->total_sectors    = itask->iscsilun->num_blocks *
 731                                itask->iscsilun->block_size / BDRV_SECTOR_SIZE ;
 732
 733     itask->status   = 0;
 734     itask->complete = 1;
 735     scsi_free_scsi_task(task);
 736 }
 737
 738 static void
 739 iscsi_inquiry_cb(struct iscsi_context *iscsi, int status, void *command_data,
 740                  void *opaque)
 741 {
 742     struct IscsiTask *itask = opaque;
 743     struct scsi_task *task = command_data;
 744     struct scsi_inquiry_standard *inq;
 745
 746     if (status != 0) {
 747         itask->status   = 1;
 748         itask->complete = 1;
 749         scsi_free_scsi_task(task);
 750         return;
 751     }
 752
 753     inq = scsi_datain_unmarshall(task);
 754     if (inq == NULL) {
 755         error_report("iSCSI: Failed to unmarshall inquiry data.");
 756         itask->status   = 1;
 757         itask->complete = 1;
 758         scsi_free_scsi_task(task);
 759         return;
 760     }
 761
 762     itask->iscsilun->type = inq->periperal_device_type;
 763
 764     scsi_free_scsi_task(task);
 765
 766     switch (itask->iscsilun->type) {
 767     case TYPE_DISK:
 768         task = iscsi_readcapacity16_task(iscsi, itask->iscsilun->lun,
 769                                    iscsi_readcapacity16_cb, opaque);
 770         if (task == NULL) {
 771             error_report("iSCSI: failed to send readcapacity16 command.");
 772             itask->status   = 1;
 773             itask->complete = 1;
 774             return;
 775         }
 776         break;
 777     case TYPE_ROM:
 778         task = iscsi_readcapacity10_task(iscsi, itask->iscsilun->lun,
 779                                    0, 0,
 780                                    iscsi_readcapacity10_cb, opaque);
 781         if (task == NULL) {
 782             error_report("iSCSI: failed to send readcapacity16 command.");
 783             itask->status   = 1;
 784             itask->complete = 1;
 785             return;
 786         }
 787         break;
 788     default:
 789         itask->status   = 0;
 790         itask->complete = 1;
 791     }
 792 }
 793
 794 static void
 795 iscsi_connect_cb(struct iscsi_context *iscsi, int status, void *command_data,
 796                  void *opaque)
 797 {
 798     struct IscsiTask *itask = opaque;
 799     struct scsi_task *task;
 800
 801     if (status != 0) {
 802         itask->status   = 1;
 803         itask->complete = 1;
 804         return;
 805     }
 806
 807     task = iscsi_inquiry_task(iscsi, itask->iscsilun->lun,
 808                               0, 0, 36,
 809                               iscsi_inquiry_cb, opaque);
 810     if (task == NULL) {
 811         error_report("iSCSI: failed to send inquiry command.");
 812         itask->status   = 1;
 813         itask->complete = 1;
 814         return;
 815     }
 816 }
 817
 818 static int parse_chap(struct iscsi_context *iscsi, const char *target)
 819 {
 820     QemuOptsList *list;
 821     QemuOpts *opts;
 822     const char *user = NULL;
 823     const char *password = NULL;
 824
 825     list = qemu_find_opts("iscsi");
 826     if (!list) {
 827         return 0;
 828     }
 829
 830     opts = qemu_opts_find(list, target);
 831     if (opts == NULL) {
 832         opts = QTAILQ_FIRST(&list->head);
 833         if (!opts) {
 834             return 0;
 835         }
 836     }
 837
 838     user = qemu_opt_get(opts, "user");
 839     if (!user) {
 840         return 0;
 841     }
 842
 843     password = qemu_opt_get(opts, "password");
 844     if (!password) {
 845         error_report("CHAP username specified but no password was given");
 846         return -1;
 847     }
 848
 849     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
 850         error_report("Failed to set initiator username and password");
 851         return -1;
 852     }
 853
 854     return 0;
 855 }
 856
 857 static void parse_header_digest(struct iscsi_context *iscsi, const char *target)
 858 {
 859     QemuOptsList *list;
 860     QemuOpts *opts;
 861     const char *digest = NULL;
 862
 863     list = qemu_find_opts("iscsi");
 864     if (!list) {
 865         return;
 866     }
 867
 868     opts = qemu_opts_find(list, target);
 869     if (opts == NULL) {
 870         opts = QTAILQ_FIRST(&list->head);
 871         if (!opts) {
 872             return;
 873         }
 874     }
 875
 876     digest = qemu_opt_get(opts, "header-digest");
 877     if (!digest) {
 878         return;
 879     }
 880
 881     if (!strcmp(digest, "CRC32C")) {
 882         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
 883     } else if (!strcmp(digest, "NONE")) {
 884         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
 885     } else if (!strcmp(digest, "CRC32C-NONE")) {
 886         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
 887     } else if (!strcmp(digest, "NONE-CRC32C")) {
 888         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 889     } else {
 890         error_report("Invalid header-digest setting : %s", digest);
 891     }
 892 }
 893
 894 static char *parse_initiator_name(const char *target)
 895 {
 896     QemuOptsList *list;
 897     QemuOpts *opts;
 898     const char *name = NULL;
 899     const char *iscsi_name = qemu_get_vm_name();
 900
 901     list = qemu_find_opts("iscsi");
 902     if (list) {
 903         opts = qemu_opts_find(list, target);
 904         if (!opts) {
 905             opts = QTAILQ_FIRST(&list->head);
 906         }
 907         if (opts) {
 908             name = qemu_opt_get(opts, "initiator-name");
 909         }
 910     }
 911
 912     if (name) {
 913         return g_strdup(name);
 914     } else {
 915         return g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
 916                                iscsi_name ? ":" : "",
 917                                iscsi_name ? iscsi_name : "");
 918     }
 919 }
 920
 921 /*
 922  * We support iscsi url's on the form
 923  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
 924  */
 925 static int iscsi_open(BlockDriverState *bs, const char *filename, int flags)
 926 {
 927     IscsiLun *iscsilun = bs->opaque;
 928     struct iscsi_context *iscsi = NULL;
 929     struct iscsi_url *iscsi_url = NULL;
 930     struct IscsiTask task;
 931     char *initiator_name = NULL;
 932     int ret;
 933
 934     if ((BDRV_SECTOR_SIZE % 512) != 0) {
 935         error_report("iSCSI: Invalid BDRV_SECTOR_SIZE. "
 936                      "BDRV_SECTOR_SIZE(%lld) is not a multiple "
 937                      "of 512", BDRV_SECTOR_SIZE);
 938         return -EINVAL;
 939     }
 940
 941     iscsi_url = iscsi_parse_full_url(iscsi, filename);
 942     if (iscsi_url == NULL) {
 943         error_report("Failed to parse URL : %s %s", filename,
 944                      iscsi_get_error(iscsi));
 945         ret = -EINVAL;
 946         goto out;
 947     }
 948
 949     memset(iscsilun, 0, sizeof(IscsiLun));
 950
 951     initiator_name = parse_initiator_name(iscsi_url->target);
 952
 953     iscsi = iscsi_create_context(initiator_name);
 954     if (iscsi == NULL) {
 955         error_report("iSCSI: Failed to create iSCSI context.");
 956         ret = -ENOMEM;
 957         goto out;
 958     }
 959
 960     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
 961         error_report("iSCSI: Failed to set target name.");
 962         ret = -EINVAL;
 963         goto out;
 964     }
 965
 966     if (iscsi_url->user != NULL) {
 967         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
 968                                               iscsi_url->passwd);
 969         if (ret != 0) {
 970             error_report("Failed to set initiator username and password");
 971             ret = -EINVAL;
 972             goto out;
 973         }
 974     }
 975
 976     /* check if we got CHAP username/password via the options */
 977     if (parse_chap(iscsi, iscsi_url->target) != 0) {
 978         error_report("iSCSI: Failed to set CHAP user/password");
 979         ret = -EINVAL;
 980         goto out;
 981     }
 982
 983     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
 984         error_report("iSCSI: Failed to set session type to normal.");
 985         ret = -EINVAL;
 986         goto out;
 987     }
 988
 989     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 990
 991     /* check if we got HEADER_DIGEST via the options */
 992     parse_header_digest(iscsi, iscsi_url->target);
 993
 994     task.iscsilun = iscsilun;
 995     task.status = 0;
 996     task.complete = 0;
 997     task.bs = bs;
 998
 999     iscsilun->iscsi = iscsi;
1000     iscsilun->lun   = iscsi_url->lun;
1001
1002     if (iscsi_full_connect_async(iscsi, iscsi_url->portal, iscsi_url->lun,
1003                                  iscsi_connect_cb, &task)
1004         != 0) {
1005         error_report("iSCSI: Failed to start async connect.");
1006         ret = -EINVAL;
1007         goto out;
1008     }
1009
1010     while (!task.complete) {
1011         iscsi_set_events(iscsilun);
1012         qemu_aio_wait();
1013     }
1014     if (task.status != 0) {
1015         error_report("iSCSI: Failed to connect to LUN : %s",
1016                      iscsi_get_error(iscsi));
1017         ret = -EINVAL;
1018         goto out;
1019     }
1020
1021     /* Medium changer or tape. We dont have any emulation for this so this must
1022      * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
1023      * to read from the device to guess the image format.
1024      */
1025     if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
1026         iscsilun->type == TYPE_TAPE) {
1027         bs->sg = 1;
1028     }
1029
1030     ret = 0;
1031
1032 out:
1033     if (initiator_name != NULL) {
1034         g_free(initiator_name);
1035     }
1036     if (iscsi_url != NULL) {
1037         iscsi_destroy_url(iscsi_url);
1038     }
1039
1040     if (ret) {
1041         if (iscsi != NULL) {
1042             iscsi_destroy_context(iscsi);
1043         }
1044         memset(iscsilun, 0, sizeof(IscsiLun));
1045     }
1046     return ret;
1047 }
1048
1049 static void iscsi_close(BlockDriverState *bs)
1050 {
1051     IscsiLun *iscsilun = bs->opaque;
1052     struct iscsi_context *iscsi = iscsilun->iscsi;
1053
1054     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL, NULL);
1055     iscsi_destroy_context(iscsi);
1056     memset(iscsilun, 0, sizeof(IscsiLun));
1057 }
1058
1059 static BlockDriver bdrv_iscsi = {
1060     .format_name     = "iscsi",
1061     .protocol_name   = "iscsi",
1062
1063     .instance_size   = sizeof(IscsiLun),
1064     .bdrv_file_open  = iscsi_open,
1065     .bdrv_close      = iscsi_close,
1066
1067     .bdrv_getlength  = iscsi_getlength,
1068
1069     .bdrv_aio_readv  = iscsi_aio_readv,
1070     .bdrv_aio_writev = iscsi_aio_writev,
1071     .bdrv_aio_flush  = iscsi_aio_flush,
1072
1073     .bdrv_aio_discard = iscsi_aio_discard,
1074
1075 #ifdef __linux__
1076     .bdrv_ioctl       = iscsi_ioctl,
1077     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1078 #endif
1079 };
1080
1081 static void iscsi_block_init(void)
1082 {
1083     bdrv_register(&bdrv_iscsi);
1084 }
1085
1086 block_init(iscsi_block_init);