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     if (!(bs->open_flags & BDRV_O_CACHE_WB)) {
 266         /* set FUA on writes when cache mode is write through */
 267         acb->task->cdb[1] |= 0x04;
 268     }
 269     lba = sector_qemu2lun(sector_num, iscsilun);
 270     *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 271     *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 272     num_sectors = size / iscsilun->block_size;
 273     *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 274     acb->task->expxferlen = size;
 275
 276     data.data = acb->buf;
 277     data.size = size;
 278
 279     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 280                                  iscsi_aio_write16_cb,
 281                                  &data,
 282                                  acb) != 0) {
 283         scsi_free_scsi_task(acb->task);
 284         g_free(acb->buf);
 285         qemu_aio_release(acb);
 286         return NULL;
 287     }
 288
 289     iscsi_set_events(iscsilun);
 290
 291     return &acb->common;
 292 }
 293
 294 static void
 295 iscsi_aio_read16_cb(struct iscsi_context *iscsi, int status,
 296                     void *command_data, void *opaque)
 297 {
 298     IscsiAIOCB *acb = opaque;
 299
 300     trace_iscsi_aio_read16_cb(iscsi, status, acb, acb->canceled);
 301
 302     if (acb->canceled != 0) {
 303         return;
 304     }
 305
 306     acb->status = 0;
 307     if (status != 0) {
 308         error_report("Failed to read16 data from iSCSI lun. %s",
 309                      iscsi_get_error(iscsi));
 310         acb->status = -EIO;
 311     }
 312
 313     iscsi_schedule_bh(acb);
 314 }
 315
 316 static BlockDriverAIOCB *
 317 iscsi_aio_readv(BlockDriverState *bs, int64_t sector_num,
 318                 QEMUIOVector *qiov, int nb_sectors,
 319                 BlockDriverCompletionFunc *cb,
 320                 void *opaque)
 321 {
 322     IscsiLun *iscsilun = bs->opaque;
 323     struct iscsi_context *iscsi = iscsilun->iscsi;
 324     IscsiAIOCB *acb;
 325     size_t qemu_read_size;
 326     int i;
 327     uint64_t lba;
 328     uint32_t num_sectors;
 329
 330     qemu_read_size = BDRV_SECTOR_SIZE * (size_t)nb_sectors;
 331
 332     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 333     trace_iscsi_aio_readv(iscsi, sector_num, nb_sectors, opaque, acb);
 334
 335     acb->iscsilun = iscsilun;
 336     acb->qiov     = qiov;
 337
 338     acb->canceled    = 0;
 339     acb->bh          = NULL;
 340     acb->status      = -EINPROGRESS;
 341     acb->read_size   = qemu_read_size;
 342     acb->buf         = NULL;
 343
 344     /* If LUN blocksize is bigger than BDRV_BLOCK_SIZE a read from QEMU
 345      * may be misaligned to the LUN, so we may need to read some extra
 346      * data.
 347      */
 348     acb->read_offset = 0;
 349     if (iscsilun->block_size > BDRV_SECTOR_SIZE) {
 350         uint64_t bdrv_offset = BDRV_SECTOR_SIZE * sector_num;
 351
 352         acb->read_offset  = bdrv_offset % iscsilun->block_size;
 353     }
 354
 355     num_sectors  = (qemu_read_size + iscsilun->block_size
 356                     + acb->read_offset - 1)
 357                     / iscsilun->block_size;
 358
 359     acb->task = malloc(sizeof(struct scsi_task));
 360     if (acb->task == NULL) {
 361         error_report("iSCSI: Failed to allocate task for scsi READ16 "
 362                      "command. %s", iscsi_get_error(iscsi));
 363         qemu_aio_release(acb);
 364         return NULL;
 365     }
 366     memset(acb->task, 0, sizeof(struct scsi_task));
 367
 368     acb->task->xfer_dir = SCSI_XFER_READ;
 369     lba = sector_qemu2lun(sector_num, iscsilun);
 370     acb->task->expxferlen = qemu_read_size;
 371
 372     switch (iscsilun->type) {
 373     case TYPE_DISK:
 374         acb->task->cdb_size = 16;
 375         acb->task->cdb[0]  = 0x88;
 376         *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 377         *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 378         *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 379         break;
 380     default:
 381         acb->task->cdb_size = 10;
 382         acb->task->cdb[0]  = 0x28;
 383         *(uint32_t *)&acb->task->cdb[2] = htonl(lba);
 384         *(uint16_t *)&acb->task->cdb[7] = htons(num_sectors);
 385         break;
 386     }
 387
 388     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 389                                  iscsi_aio_read16_cb,
 390                                  NULL,
 391                                  acb) != 0) {
 392         scsi_free_scsi_task(acb->task);
 393         qemu_aio_release(acb);
 394         return NULL;
 395     }
 396
 397     for (i = 0; i < acb->qiov->niov; i++) {
 398         scsi_task_add_data_in_buffer(acb->task,
 399                 acb->qiov->iov[i].iov_len,
 400                 acb->qiov->iov[i].iov_base);
 401     }
 402
 403     iscsi_set_events(iscsilun);
 404
 405     return &acb->common;
 406 }
 407
 408
 409 static void
 410 iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
 411                      void *command_data, void *opaque)
 412 {
 413     IscsiAIOCB *acb = opaque;
 414
 415     if (acb->canceled != 0) {
 416         return;
 417     }
 418
 419     acb->status = 0;
 420     if (status < 0) {
 421         error_report("Failed to sync10 data on iSCSI lun. %s",
 422                      iscsi_get_error(iscsi));
 423         acb->status = -EIO;
 424     }
 425
 426     iscsi_schedule_bh(acb);
 427 }
 428
 429 static BlockDriverAIOCB *
 430 iscsi_aio_flush(BlockDriverState *bs,
 431                 BlockDriverCompletionFunc *cb, void *opaque)
 432 {
 433     IscsiLun *iscsilun = bs->opaque;
 434     struct iscsi_context *iscsi = iscsilun->iscsi;
 435     IscsiAIOCB *acb;
 436
 437     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 438
 439     acb->iscsilun = iscsilun;
 440     acb->canceled   = 0;
 441     acb->bh         = NULL;
 442     acb->status     = -EINPROGRESS;
 443
 444     acb->task = iscsi_synchronizecache10_task(iscsi, iscsilun->lun,
 445                                          0, 0, 0, 0,
 446                                          iscsi_synccache10_cb,
 447                                          acb);
 448     if (acb->task == NULL) {
 449         error_report("iSCSI: Failed to send synchronizecache10 command. %s",
 450                      iscsi_get_error(iscsi));
 451         qemu_aio_release(acb);
 452         return NULL;
 453     }
 454
 455     iscsi_set_events(iscsilun);
 456
 457     return &acb->common;
 458 }
 459
 460 static void
 461 iscsi_unmap_cb(struct iscsi_context *iscsi, int status,
 462                      void *command_data, void *opaque)
 463 {
 464     IscsiAIOCB *acb = opaque;
 465
 466     if (acb->canceled != 0) {
 467         return;
 468     }
 469
 470     acb->status = 0;
 471     if (status < 0) {
 472         error_report("Failed to unmap data on iSCSI lun. %s",
 473                      iscsi_get_error(iscsi));
 474         acb->status = -EIO;
 475     }
 476
 477     iscsi_schedule_bh(acb);
 478 }
 479
 480 static BlockDriverAIOCB *
 481 iscsi_aio_discard(BlockDriverState *bs,
 482                   int64_t sector_num, int nb_sectors,
 483                   BlockDriverCompletionFunc *cb, void *opaque)
 484 {
 485     IscsiLun *iscsilun = bs->opaque;
 486     struct iscsi_context *iscsi = iscsilun->iscsi;
 487     IscsiAIOCB *acb;
 488     struct unmap_list list[1];
 489
 490     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 491
 492     acb->iscsilun = iscsilun;
 493     acb->canceled   = 0;
 494     acb->bh         = NULL;
 495     acb->status     = -EINPROGRESS;
 496
 497     list[0].lba = sector_qemu2lun(sector_num, iscsilun);
 498     list[0].num = nb_sectors * BDRV_SECTOR_SIZE / iscsilun->block_size;
 499
 500     acb->task = iscsi_unmap_task(iscsi, iscsilun->lun,
 501                                  0, 0, &list[0], 1,
 502                                  iscsi_unmap_cb,
 503                                  acb);
 504     if (acb->task == NULL) {
 505         error_report("iSCSI: Failed to send unmap command. %s",
 506                      iscsi_get_error(iscsi));
 507         qemu_aio_release(acb);
 508         return NULL;
 509     }
 510
 511     iscsi_set_events(iscsilun);
 512
 513     return &acb->common;
 514 }
 515
 516 #ifdef __linux__
 517 static void
 518 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 519                      void *command_data, void *opaque)
 520 {
 521     IscsiAIOCB *acb = opaque;
 522
 523     if (acb->canceled != 0) {
 524         return;
 525     }
 526
 527     acb->status = 0;
 528     if (status < 0) {
 529         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 530                      iscsi_get_error(iscsi));
 531         acb->status = -EIO;
 532     }
 533
 534     acb->ioh->driver_status = 0;
 535     acb->ioh->host_status   = 0;
 536     acb->ioh->resid         = 0;
 537
 538 #define SG_ERR_DRIVER_SENSE    0x08
 539
 540     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 541         int ss;
 542
 543         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 544
 545         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 546         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 547              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 548         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 549     }
 550
 551     iscsi_schedule_bh(acb);
 552 }
 553
 554 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 555         unsigned long int req, void *buf,
 556         BlockDriverCompletionFunc *cb, void *opaque)
 557 {
 558     IscsiLun *iscsilun = bs->opaque;
 559     struct iscsi_context *iscsi = iscsilun->iscsi;
 560     struct iscsi_data data;
 561     IscsiAIOCB *acb;
 562
 563     assert(req == SG_IO);
 564
 565     acb = qemu_aio_get(&iscsi_aio_pool, bs, cb, opaque);
 566
 567     acb->iscsilun = iscsilun;
 568     acb->canceled    = 0;
 569     acb->bh          = NULL;
 570     acb->status      = -EINPROGRESS;
 571     acb->buf         = NULL;
 572     acb->ioh         = buf;
 573
 574     acb->task = malloc(sizeof(struct scsi_task));
 575     if (acb->task == NULL) {
 576         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 577                      iscsi_get_error(iscsi));
 578         qemu_aio_release(acb);
 579         return NULL;
 580     }
 581     memset(acb->task, 0, sizeof(struct scsi_task));
 582
 583     switch (acb->ioh->dxfer_direction) {
 584     case SG_DXFER_TO_DEV:
 585         acb->task->xfer_dir = SCSI_XFER_WRITE;
 586         break;
 587     case SG_DXFER_FROM_DEV:
 588         acb->task->xfer_dir = SCSI_XFER_READ;
 589         break;
 590     default:
 591         acb->task->xfer_dir = SCSI_XFER_NONE;
 592         break;
 593     }
 594
 595     acb->task->cdb_size = acb->ioh->cmd_len;
 596     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 597     acb->task->expxferlen = acb->ioh->dxfer_len;
 598
 599     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 600         data.data = acb->ioh->dxferp;
 601         data.size = acb->ioh->dxfer_len;
 602     }
 603     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 604                                  iscsi_aio_ioctl_cb,
 605                                  (acb->task->xfer_dir == SCSI_XFER_WRITE) ?
 606                                      &data : NULL,
 607                                  acb) != 0) {
 608         scsi_free_scsi_task(acb->task);
 609         qemu_aio_release(acb);
 610         return NULL;
 611     }
 612
 613     /* tell libiscsi to read straight into the buffer we got from ioctl */
 614     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 615         scsi_task_add_data_in_buffer(acb->task,
 616                                      acb->ioh->dxfer_len,
 617                                      acb->ioh->dxferp);
 618     }
 619
 620     iscsi_set_events(iscsilun);
 621
 622     return &acb->common;
 623 }
 624
 625
 626 static void ioctl_cb(void *opaque, int status)
 627 {
 628     int *p_status = opaque;
 629     *p_status = status;
 630 }
 631
 632 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 633 {
 634     IscsiLun *iscsilun = bs->opaque;
 635     int status;
 636
 637     switch (req) {
 638     case SG_GET_VERSION_NUM:
 639         *(int *)buf = 30000;
 640         break;
 641     case SG_GET_SCSI_ID:
 642         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 643         break;
 644     case SG_IO:
 645         status = -EINPROGRESS;
 646         iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status);
 647
 648         while (status == -EINPROGRESS) {
 649             qemu_aio_wait();
 650         }
 651
 652         return 0;
 653     default:
 654         return -1;
 655     }
 656     return 0;
 657 }
 658 #endif
 659
 660 static int64_t
 661 iscsi_getlength(BlockDriverState *bs)
 662 {
 663     IscsiLun *iscsilun = bs->opaque;
 664     int64_t len;
 665
 666     len  = iscsilun->num_blocks;
 667     len *= iscsilun->block_size;
 668
 669     return len;
 670 }
 671
 672 static void
 673 iscsi_readcapacity16_cb(struct iscsi_context *iscsi, int status,
 674                         void *command_data, void *opaque)
 675 {
 676     struct IscsiTask *itask = opaque;
 677     struct scsi_readcapacity16 *rc16;
 678     struct scsi_task *task = command_data;
 679
 680     if (status != 0) {
 681         error_report("iSCSI: Failed to read capacity of iSCSI lun. %s",
 682                      iscsi_get_error(iscsi));
 683         itask->status   = 1;
 684         itask->complete = 1;
 685         scsi_free_scsi_task(task);
 686         return;
 687     }
 688
 689     rc16 = scsi_datain_unmarshall(task);
 690     if (rc16 == NULL) {
 691         error_report("iSCSI: Failed to unmarshall readcapacity16 data.");
 692         itask->status   = 1;
 693         itask->complete = 1;
 694         scsi_free_scsi_task(task);
 695         return;
 696     }
 697
 698     itask->iscsilun->block_size = rc16->block_length;
 699     itask->iscsilun->num_blocks = rc16->returned_lba + 1;
 700     itask->bs->total_sectors    = itask->iscsilun->num_blocks *
 701                                itask->iscsilun->block_size / BDRV_SECTOR_SIZE ;
 702
 703     itask->status   = 0;
 704     itask->complete = 1;
 705     scsi_free_scsi_task(task);
 706 }
 707
 708 static void
 709 iscsi_readcapacity10_cb(struct iscsi_context *iscsi, int status,
 710                         void *command_data, void *opaque)
 711 {
 712     struct IscsiTask *itask = opaque;
 713     struct scsi_readcapacity10 *rc10;
 714     struct scsi_task *task = command_data;
 715
 716     if (status != 0) {
 717         error_report("iSCSI: Failed to read capacity of iSCSI lun. %s",
 718                      iscsi_get_error(iscsi));
 719         itask->status   = 1;
 720         itask->complete = 1;
 721         scsi_free_scsi_task(task);
 722         return;
 723     }
 724
 725     rc10 = scsi_datain_unmarshall(task);
 726     if (rc10 == NULL) {
 727         error_report("iSCSI: Failed to unmarshall readcapacity10 data.");
 728         itask->status   = 1;
 729         itask->complete = 1;
 730         scsi_free_scsi_task(task);
 731         return;
 732     }
 733
 734     itask->iscsilun->block_size = rc10->block_size;
 735     if (rc10->lba == 0) {
 736         /* blank disk loaded */
 737         itask->iscsilun->num_blocks = 0;
 738     } else {
 739         itask->iscsilun->num_blocks = rc10->lba + 1;
 740     }
 741     itask->bs->total_sectors    = itask->iscsilun->num_blocks *
 742                                itask->iscsilun->block_size / BDRV_SECTOR_SIZE ;
 743
 744     itask->status   = 0;
 745     itask->complete = 1;
 746     scsi_free_scsi_task(task);
 747 }
 748
 749 static void
 750 iscsi_inquiry_cb(struct iscsi_context *iscsi, int status, void *command_data,
 751                  void *opaque)
 752 {
 753     struct IscsiTask *itask = opaque;
 754     struct scsi_task *task = command_data;
 755     struct scsi_inquiry_standard *inq;
 756
 757     if (status != 0) {
 758         itask->status   = 1;
 759         itask->complete = 1;
 760         scsi_free_scsi_task(task);
 761         return;
 762     }
 763
 764     inq = scsi_datain_unmarshall(task);
 765     if (inq == NULL) {
 766         error_report("iSCSI: Failed to unmarshall inquiry data.");
 767         itask->status   = 1;
 768         itask->complete = 1;
 769         scsi_free_scsi_task(task);
 770         return;
 771     }
 772
 773     itask->iscsilun->type = inq->periperal_device_type;
 774
 775     scsi_free_scsi_task(task);
 776
 777     switch (itask->iscsilun->type) {
 778     case TYPE_DISK:
 779         task = iscsi_readcapacity16_task(iscsi, itask->iscsilun->lun,
 780                                    iscsi_readcapacity16_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     case TYPE_ROM:
 789         task = iscsi_readcapacity10_task(iscsi, itask->iscsilun->lun,
 790                                    0, 0,
 791                                    iscsi_readcapacity10_cb, opaque);
 792         if (task == NULL) {
 793             error_report("iSCSI: failed to send readcapacity16 command.");
 794             itask->status   = 1;
 795             itask->complete = 1;
 796             return;
 797         }
 798         break;
 799     default:
 800         itask->status   = 0;
 801         itask->complete = 1;
 802     }
 803 }
 804
 805 static void
 806 iscsi_connect_cb(struct iscsi_context *iscsi, int status, void *command_data,
 807                  void *opaque)
 808 {
 809     struct IscsiTask *itask = opaque;
 810     struct scsi_task *task;
 811
 812     if (status != 0) {
 813         itask->status   = 1;
 814         itask->complete = 1;
 815         return;
 816     }
 817
 818     task = iscsi_inquiry_task(iscsi, itask->iscsilun->lun,
 819                               0, 0, 36,
 820                               iscsi_inquiry_cb, opaque);
 821     if (task == NULL) {
 822         error_report("iSCSI: failed to send inquiry command.");
 823         itask->status   = 1;
 824         itask->complete = 1;
 825         return;
 826     }
 827 }
 828
 829 static int parse_chap(struct iscsi_context *iscsi, const char *target)
 830 {
 831     QemuOptsList *list;
 832     QemuOpts *opts;
 833     const char *user = NULL;
 834     const char *password = NULL;
 835
 836     list = qemu_find_opts("iscsi");
 837     if (!list) {
 838         return 0;
 839     }
 840
 841     opts = qemu_opts_find(list, target);
 842     if (opts == NULL) {
 843         opts = QTAILQ_FIRST(&list->head);
 844         if (!opts) {
 845             return 0;
 846         }
 847     }
 848
 849     user = qemu_opt_get(opts, "user");
 850     if (!user) {
 851         return 0;
 852     }
 853
 854     password = qemu_opt_get(opts, "password");
 855     if (!password) {
 856         error_report("CHAP username specified but no password was given");
 857         return -1;
 858     }
 859
 860     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
 861         error_report("Failed to set initiator username and password");
 862         return -1;
 863     }
 864
 865     return 0;
 866 }
 867
 868 static void parse_header_digest(struct iscsi_context *iscsi, const char *target)
 869 {
 870     QemuOptsList *list;
 871     QemuOpts *opts;
 872     const char *digest = NULL;
 873
 874     list = qemu_find_opts("iscsi");
 875     if (!list) {
 876         return;
 877     }
 878
 879     opts = qemu_opts_find(list, target);
 880     if (opts == NULL) {
 881         opts = QTAILQ_FIRST(&list->head);
 882         if (!opts) {
 883             return;
 884         }
 885     }
 886
 887     digest = qemu_opt_get(opts, "header-digest");
 888     if (!digest) {
 889         return;
 890     }
 891
 892     if (!strcmp(digest, "CRC32C")) {
 893         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
 894     } else if (!strcmp(digest, "NONE")) {
 895         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
 896     } else if (!strcmp(digest, "CRC32C-NONE")) {
 897         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
 898     } else if (!strcmp(digest, "NONE-CRC32C")) {
 899         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 900     } else {
 901         error_report("Invalid header-digest setting : %s", digest);
 902     }
 903 }
 904
 905 static char *parse_initiator_name(const char *target)
 906 {
 907     QemuOptsList *list;
 908     QemuOpts *opts;
 909     const char *name = NULL;
 910     const char *iscsi_name = qemu_get_vm_name();
 911
 912     list = qemu_find_opts("iscsi");
 913     if (list) {
 914         opts = qemu_opts_find(list, target);
 915         if (!opts) {
 916             opts = QTAILQ_FIRST(&list->head);
 917         }
 918         if (opts) {
 919             name = qemu_opt_get(opts, "initiator-name");
 920         }
 921     }
 922
 923     if (name) {
 924         return g_strdup(name);
 925     } else {
 926         return g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
 927                                iscsi_name ? ":" : "",
 928                                iscsi_name ? iscsi_name : "");
 929     }
 930 }
 931
 932 /*
 933  * We support iscsi url's on the form
 934  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
 935  */
 936 static int iscsi_open(BlockDriverState *bs, const char *filename, int flags)
 937 {
 938     IscsiLun *iscsilun = bs->opaque;
 939     struct iscsi_context *iscsi = NULL;
 940     struct iscsi_url *iscsi_url = NULL;
 941     struct IscsiTask task;
 942     char *initiator_name = NULL;
 943     int ret;
 944
 945     if ((BDRV_SECTOR_SIZE % 512) != 0) {
 946         error_report("iSCSI: Invalid BDRV_SECTOR_SIZE. "
 947                      "BDRV_SECTOR_SIZE(%lld) is not a multiple "
 948                      "of 512", BDRV_SECTOR_SIZE);
 949         return -EINVAL;
 950     }
 951
 952     iscsi_url = iscsi_parse_full_url(iscsi, filename);
 953     if (iscsi_url == NULL) {
 954         error_report("Failed to parse URL : %s %s", filename,
 955                      iscsi_get_error(iscsi));
 956         ret = -EINVAL;
 957         goto out;
 958     }
 959
 960     memset(iscsilun, 0, sizeof(IscsiLun));
 961
 962     initiator_name = parse_initiator_name(iscsi_url->target);
 963
 964     iscsi = iscsi_create_context(initiator_name);
 965     if (iscsi == NULL) {
 966         error_report("iSCSI: Failed to create iSCSI context.");
 967         ret = -ENOMEM;
 968         goto out;
 969     }
 970
 971     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
 972         error_report("iSCSI: Failed to set target name.");
 973         ret = -EINVAL;
 974         goto out;
 975     }
 976
 977     if (iscsi_url->user != NULL) {
 978         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
 979                                               iscsi_url->passwd);
 980         if (ret != 0) {
 981             error_report("Failed to set initiator username and password");
 982             ret = -EINVAL;
 983             goto out;
 984         }
 985     }
 986
 987     /* check if we got CHAP username/password via the options */
 988     if (parse_chap(iscsi, iscsi_url->target) != 0) {
 989         error_report("iSCSI: Failed to set CHAP user/password");
 990         ret = -EINVAL;
 991         goto out;
 992     }
 993
 994     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
 995         error_report("iSCSI: Failed to set session type to normal.");
 996         ret = -EINVAL;
 997         goto out;
 998     }
 999
1000     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1001
1002     /* check if we got HEADER_DIGEST via the options */
1003     parse_header_digest(iscsi, iscsi_url->target);
1004
1005     task.iscsilun = iscsilun;
1006     task.status = 0;
1007     task.complete = 0;
1008     task.bs = bs;
1009
1010     iscsilun->iscsi = iscsi;
1011     iscsilun->lun   = iscsi_url->lun;
1012
1013     if (iscsi_full_connect_async(iscsi, iscsi_url->portal, iscsi_url->lun,
1014                                  iscsi_connect_cb, &task)
1015         != 0) {
1016         error_report("iSCSI: Failed to start async connect.");
1017         ret = -EINVAL;
1018         goto out;
1019     }
1020
1021     while (!task.complete) {
1022         iscsi_set_events(iscsilun);
1023         qemu_aio_wait();
1024     }
1025     if (task.status != 0) {
1026         error_report("iSCSI: Failed to connect to LUN : %s",
1027                      iscsi_get_error(iscsi));
1028         ret = -EINVAL;
1029         goto out;
1030     }
1031
1032     /* Medium changer or tape. We dont have any emulation for this so this must
1033      * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
1034      * to read from the device to guess the image format.
1035      */
1036     if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
1037         iscsilun->type == TYPE_TAPE) {
1038         bs->sg = 1;
1039     }
1040
1041     ret = 0;
1042
1043 out:
1044     if (initiator_name != NULL) {
1045         g_free(initiator_name);
1046     }
1047     if (iscsi_url != NULL) {
1048         iscsi_destroy_url(iscsi_url);
1049     }
1050
1051     if (ret) {
1052         if (iscsi != NULL) {
1053             iscsi_destroy_context(iscsi);
1054         }
1055         memset(iscsilun, 0, sizeof(IscsiLun));
1056     }
1057     return ret;
1058 }
1059
1060 static void iscsi_close(BlockDriverState *bs)
1061 {
1062     IscsiLun *iscsilun = bs->opaque;
1063     struct iscsi_context *iscsi = iscsilun->iscsi;
1064
1065     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL, NULL);
1066     iscsi_destroy_context(iscsi);
1067     memset(iscsilun, 0, sizeof(IscsiLun));
1068 }
1069
1070 static BlockDriver bdrv_iscsi = {
1071     .format_name     = "iscsi",
1072     .protocol_name   = "iscsi",
1073
1074     .instance_size   = sizeof(IscsiLun),
1075     .bdrv_file_open  = iscsi_open,
1076     .bdrv_close      = iscsi_close,
1077
1078     .bdrv_getlength  = iscsi_getlength,
1079
1080     .bdrv_aio_readv  = iscsi_aio_readv,
1081     .bdrv_aio_writev = iscsi_aio_writev,
1082     .bdrv_aio_flush  = iscsi_aio_flush,
1083
1084     .bdrv_aio_discard = iscsi_aio_discard,
1085
1086 #ifdef __linux__
1087     .bdrv_ioctl       = iscsi_ioctl,
1088     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1089 #endif
1090 };
1091
1092 static void iscsi_block_init(void)
1093 {
1094     bdrv_register(&bdrv_iscsi);
1095 }
1096
1097 block_init(iscsi_block_init);