block/iscsi.c

   1 /*
   2  * QEMU Block driver for iSCSI images
   3  *
   4  * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25 #include "config-host.h"
  26
  27 #include <poll.h>
  28 #include <arpa/inet.h>
  29 #include "qemu-common.h"
  30 #include "qemu/config-file.h"
  31 #include "qemu/error-report.h"
  32 #include "block/block_int.h"
  33 #include "trace.h"
  34 #include "hw/scsi-defs.h"
  35
  36 #include <iscsi/iscsi.h>
  37 #include <iscsi/scsi-lowlevel.h>
  38
  39 #ifdef __linux__
  40 #include <scsi/sg.h>
  41 #include <hw/scsi-defs.h>
  42 #endif
  43
  44 typedef struct IscsiLun {
  45     struct iscsi_context *iscsi;
  46     int lun;
  47     enum scsi_inquiry_peripheral_device_type type;
  48     int block_size;
  49     uint64_t num_blocks;
  50     int events;
  51     QEMUTimer *nop_timer;
  52 } IscsiLun;
  53
  54 typedef struct IscsiAIOCB {
  55     BlockDriverAIOCB common;
  56     QEMUIOVector *qiov;
  57     QEMUBH *bh;
  58     IscsiLun *iscsilun;
  59     struct scsi_task *task;
  60     uint8_t *buf;
  61     int status;
  62     int canceled;
  63     size_t read_size;
  64     size_t read_offset;
  65 #ifdef __linux__
  66     sg_io_hdr_t *ioh;
  67 #endif
  68 } IscsiAIOCB;
  69
  70 #define NOP_INTERVAL 5000
  71 #define MAX_NOP_FAILURES 3
  72
  73 static void
  74 iscsi_bh_cb(void *p)
  75 {
  76     IscsiAIOCB *acb = p;
  77
  78     qemu_bh_delete(acb->bh);
  79
  80     g_free(acb->buf);
  81     acb->buf = NULL;
  82
  83     if (acb->canceled == 0) {
  84         acb->common.cb(acb->common.opaque, acb->status);
  85     }
  86
  87     if (acb->task != NULL) {
  88         scsi_free_scsi_task(acb->task);
  89         acb->task = NULL;
  90     }
  91
  92     qemu_aio_release(acb);
  93 }
  94
  95 static void
  96 iscsi_schedule_bh(IscsiAIOCB *acb)
  97 {
  98     if (acb->bh) {
  99         return;
 100     }
 101     acb->bh = qemu_bh_new(iscsi_bh_cb, acb);
 102     qemu_bh_schedule(acb->bh);
 103 }
 104
 105
 106 static void
 107 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
 108                     void *private_data)
 109 {
 110     IscsiAIOCB *acb = private_data;
 111
 112     acb->status = -ECANCELED;
 113     iscsi_schedule_bh(acb);
 114 }
 115
 116 static void
 117 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
 118 {
 119     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
 120     IscsiLun *iscsilun = acb->iscsilun;
 121
 122     if (acb->status != -EINPROGRESS) {
 123         return;
 124     }
 125
 126     acb->canceled = 1;
 127
 128     /* send a task mgmt call to the target to cancel the task on the target */
 129     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
 130                                      iscsi_abort_task_cb, acb);
 131
 132     while (acb->status == -EINPROGRESS) {
 133         qemu_aio_wait();
 134     }
 135 }
 136
 137 static const AIOCBInfo iscsi_aiocb_info = {
 138     .aiocb_size         = sizeof(IscsiAIOCB),
 139     .cancel             = iscsi_aio_cancel,
 140 };
 141
 142
 143 static void iscsi_process_read(void *arg);
 144 static void iscsi_process_write(void *arg);
 145
 146 static int iscsi_process_flush(void *arg)
 147 {
 148     IscsiLun *iscsilun = arg;
 149
 150     return iscsi_queue_length(iscsilun->iscsi) > 0;
 151 }
 152
 153 static void
 154 iscsi_set_events(IscsiLun *iscsilun)
 155 {
 156     struct iscsi_context *iscsi = iscsilun->iscsi;
 157     int ev;
 158
 159     /* We always register a read handler.  */
 160     ev = POLLIN;
 161     ev |= iscsi_which_events(iscsi);
 162     if (ev != iscsilun->events) {
 163         qemu_aio_set_fd_handler(iscsi_get_fd(iscsi),
 164                       iscsi_process_read,
 165                       (ev & POLLOUT) ? iscsi_process_write : NULL,
 166                       iscsi_process_flush,
 167                       iscsilun);
 168
 169     }
 170
 171     iscsilun->events = ev;
 172 }
 173
 174 static void
 175 iscsi_process_read(void *arg)
 176 {
 177     IscsiLun *iscsilun = arg;
 178     struct iscsi_context *iscsi = iscsilun->iscsi;
 179
 180     iscsi_service(iscsi, POLLIN);
 181     iscsi_set_events(iscsilun);
 182 }
 183
 184 static void
 185 iscsi_process_write(void *arg)
 186 {
 187     IscsiLun *iscsilun = arg;
 188     struct iscsi_context *iscsi = iscsilun->iscsi;
 189
 190     iscsi_service(iscsi, POLLOUT);
 191     iscsi_set_events(iscsilun);
 192 }
 193
 194
 195 static void
 196 iscsi_aio_write16_cb(struct iscsi_context *iscsi, int status,
 197                      void *command_data, void *opaque)
 198 {
 199     IscsiAIOCB *acb = opaque;
 200
 201     trace_iscsi_aio_write16_cb(iscsi, status, acb, acb->canceled);
 202
 203     g_free(acb->buf);
 204     acb->buf = NULL;
 205
 206     if (acb->canceled != 0) {
 207         return;
 208     }
 209
 210     acb->status = 0;
 211     if (status < 0) {
 212         error_report("Failed to write16 data to iSCSI lun. %s",
 213                      iscsi_get_error(iscsi));
 214         acb->status = -EIO;
 215     }
 216
 217     iscsi_schedule_bh(acb);
 218 }
 219
 220 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
 221 {
 222     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
 223 }
 224
 225 static BlockDriverAIOCB *
 226 iscsi_aio_writev(BlockDriverState *bs, int64_t sector_num,
 227                  QEMUIOVector *qiov, int nb_sectors,
 228                  BlockDriverCompletionFunc *cb,
 229                  void *opaque)
 230 {
 231     IscsiLun *iscsilun = bs->opaque;
 232     struct iscsi_context *iscsi = iscsilun->iscsi;
 233     IscsiAIOCB *acb;
 234     size_t size;
 235     uint32_t num_sectors;
 236     uint64_t lba;
 237 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 238     struct iscsi_data data;
 239 #endif
 240     int ret;
 241
 242     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 243     trace_iscsi_aio_writev(iscsi, sector_num, nb_sectors, opaque, acb);
 244
 245     acb->iscsilun = iscsilun;
 246     acb->qiov     = qiov;
 247
 248     acb->canceled   = 0;
 249     acb->bh         = NULL;
 250     acb->status     = -EINPROGRESS;
 251     acb->buf        = NULL;
 252
 253     /* this will allow us to get rid of 'buf' completely */
 254     size = nb_sectors * BDRV_SECTOR_SIZE;
 255
 256 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 257     data.size = MIN(size, acb->qiov->size);
 258
 259     /* if the iovec only contains one buffer we can pass it directly */
 260     if (acb->qiov->niov == 1) {
 261         data.data = acb->qiov->iov[0].iov_base;
 262     } else {
 263         acb->buf = g_malloc(data.size);
 264         qemu_iovec_to_buf(acb->qiov, 0, acb->buf, data.size);
 265         data.data = acb->buf;
 266     }
 267 #endif
 268
 269     acb->task = malloc(sizeof(struct scsi_task));
 270     if (acb->task == NULL) {
 271         error_report("iSCSI: Failed to allocate task for scsi WRITE16 "
 272                      "command. %s", iscsi_get_error(iscsi));
 273         qemu_aio_release(acb);
 274         return NULL;
 275     }
 276     memset(acb->task, 0, sizeof(struct scsi_task));
 277
 278     acb->task->xfer_dir = SCSI_XFER_WRITE;
 279     acb->task->cdb_size = 16;
 280     acb->task->cdb[0] = 0x8a;
 281     lba = sector_qemu2lun(sector_num, iscsilun);
 282     *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 283     *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 284     num_sectors = size / iscsilun->block_size;
 285     *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 286     acb->task->expxferlen = size;
 287
 288 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 289     ret = iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 290                                    iscsi_aio_write16_cb,
 291                                    NULL,
 292                                    acb);
 293 #else
 294     ret = iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 295                                    iscsi_aio_write16_cb,
 296                                    &data,
 297                                    acb);
 298 #endif
 299     if (ret != 0) {
 300         scsi_free_scsi_task(acb->task);
 301         g_free(acb->buf);
 302         qemu_aio_release(acb);
 303         return NULL;
 304     }
 305
 306 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 307     scsi_task_set_iov_out(acb->task, (struct scsi_iovec*) acb->qiov->iov, acb->qiov->niov);
 308 #endif
 309
 310     iscsi_set_events(iscsilun);
 311
 312     return &acb->common;
 313 }
 314
 315 static void
 316 iscsi_aio_read16_cb(struct iscsi_context *iscsi, int status,
 317                     void *command_data, void *opaque)
 318 {
 319     IscsiAIOCB *acb = opaque;
 320
 321     trace_iscsi_aio_read16_cb(iscsi, status, acb, acb->canceled);
 322
 323     if (acb->canceled != 0) {
 324         return;
 325     }
 326
 327     acb->status = 0;
 328     if (status != 0) {
 329         error_report("Failed to read16 data from iSCSI lun. %s",
 330                      iscsi_get_error(iscsi));
 331         acb->status = -EIO;
 332     }
 333
 334     iscsi_schedule_bh(acb);
 335 }
 336
 337 static BlockDriverAIOCB *
 338 iscsi_aio_readv(BlockDriverState *bs, int64_t sector_num,
 339                 QEMUIOVector *qiov, int nb_sectors,
 340                 BlockDriverCompletionFunc *cb,
 341                 void *opaque)
 342 {
 343     IscsiLun *iscsilun = bs->opaque;
 344     struct iscsi_context *iscsi = iscsilun->iscsi;
 345     IscsiAIOCB *acb;
 346     size_t qemu_read_size;
 347 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 348     int i;
 349 #endif
 350     int ret;
 351     uint64_t lba;
 352     uint32_t num_sectors;
 353
 354     qemu_read_size = BDRV_SECTOR_SIZE * (size_t)nb_sectors;
 355
 356     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 357     trace_iscsi_aio_readv(iscsi, sector_num, nb_sectors, opaque, acb);
 358
 359     acb->iscsilun = iscsilun;
 360     acb->qiov     = qiov;
 361
 362     acb->canceled    = 0;
 363     acb->bh          = NULL;
 364     acb->status      = -EINPROGRESS;
 365     acb->read_size   = qemu_read_size;
 366     acb->buf         = NULL;
 367
 368     /* If LUN blocksize is bigger than BDRV_BLOCK_SIZE a read from QEMU
 369      * may be misaligned to the LUN, so we may need to read some extra
 370      * data.
 371      */
 372     acb->read_offset = 0;
 373     if (iscsilun->block_size > BDRV_SECTOR_SIZE) {
 374         uint64_t bdrv_offset = BDRV_SECTOR_SIZE * sector_num;
 375
 376         acb->read_offset  = bdrv_offset % iscsilun->block_size;
 377     }
 378
 379     num_sectors  = (qemu_read_size + iscsilun->block_size
 380                     + acb->read_offset - 1)
 381                     / iscsilun->block_size;
 382
 383     acb->task = malloc(sizeof(struct scsi_task));
 384     if (acb->task == NULL) {
 385         error_report("iSCSI: Failed to allocate task for scsi READ16 "
 386                      "command. %s", iscsi_get_error(iscsi));
 387         qemu_aio_release(acb);
 388         return NULL;
 389     }
 390     memset(acb->task, 0, sizeof(struct scsi_task));
 391
 392     acb->task->xfer_dir = SCSI_XFER_READ;
 393     lba = sector_qemu2lun(sector_num, iscsilun);
 394     acb->task->expxferlen = qemu_read_size;
 395
 396     switch (iscsilun->type) {
 397     case TYPE_DISK:
 398         acb->task->cdb_size = 16;
 399         acb->task->cdb[0]  = 0x88;
 400         *(uint32_t *)&acb->task->cdb[2]  = htonl(lba >> 32);
 401         *(uint32_t *)&acb->task->cdb[6]  = htonl(lba & 0xffffffff);
 402         *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
 403         break;
 404     default:
 405         acb->task->cdb_size = 10;
 406         acb->task->cdb[0]  = 0x28;
 407         *(uint32_t *)&acb->task->cdb[2] = htonl(lba);
 408         *(uint16_t *)&acb->task->cdb[7] = htons(num_sectors);
 409         break;
 410     }
 411
 412     ret = iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 413                                    iscsi_aio_read16_cb,
 414                                    NULL,
 415                                    acb);
 416     if (ret != 0) {
 417         scsi_free_scsi_task(acb->task);
 418         qemu_aio_release(acb);
 419         return NULL;
 420     }
 421
 422 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 423     scsi_task_set_iov_in(acb->task, (struct scsi_iovec*) acb->qiov->iov, acb->qiov->niov);
 424 #else
 425     for (i = 0; i < acb->qiov->niov; i++) {
 426         scsi_task_add_data_in_buffer(acb->task,
 427                 acb->qiov->iov[i].iov_len,
 428                 acb->qiov->iov[i].iov_base);
 429     }
 430 #endif
 431
 432     iscsi_set_events(iscsilun);
 433
 434     return &acb->common;
 435 }
 436
 437
 438 static void
 439 iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
 440                      void *command_data, void *opaque)
 441 {
 442     IscsiAIOCB *acb = opaque;
 443
 444     if (acb->canceled != 0) {
 445         return;
 446     }
 447
 448     acb->status = 0;
 449     if (status < 0) {
 450         error_report("Failed to sync10 data on iSCSI lun. %s",
 451                      iscsi_get_error(iscsi));
 452         acb->status = -EIO;
 453     }
 454
 455     iscsi_schedule_bh(acb);
 456 }
 457
 458 static BlockDriverAIOCB *
 459 iscsi_aio_flush(BlockDriverState *bs,
 460                 BlockDriverCompletionFunc *cb, void *opaque)
 461 {
 462     IscsiLun *iscsilun = bs->opaque;
 463     struct iscsi_context *iscsi = iscsilun->iscsi;
 464     IscsiAIOCB *acb;
 465
 466     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 467
 468     acb->iscsilun = iscsilun;
 469     acb->canceled   = 0;
 470     acb->bh         = NULL;
 471     acb->status     = -EINPROGRESS;
 472     acb->buf        = NULL;
 473
 474     acb->task = iscsi_synchronizecache10_task(iscsi, iscsilun->lun,
 475                                          0, 0, 0, 0,
 476                                          iscsi_synccache10_cb,
 477                                          acb);
 478     if (acb->task == NULL) {
 479         error_report("iSCSI: Failed to send synchronizecache10 command. %s",
 480                      iscsi_get_error(iscsi));
 481         qemu_aio_release(acb);
 482         return NULL;
 483     }
 484
 485     iscsi_set_events(iscsilun);
 486
 487     return &acb->common;
 488 }
 489
 490 static void
 491 iscsi_unmap_cb(struct iscsi_context *iscsi, int status,
 492                      void *command_data, void *opaque)
 493 {
 494     IscsiAIOCB *acb = opaque;
 495
 496     if (acb->canceled != 0) {
 497         return;
 498     }
 499
 500     acb->status = 0;
 501     if (status < 0) {
 502         error_report("Failed to unmap data on iSCSI lun. %s",
 503                      iscsi_get_error(iscsi));
 504         acb->status = -EIO;
 505     }
 506
 507     iscsi_schedule_bh(acb);
 508 }
 509
 510 static BlockDriverAIOCB *
 511 iscsi_aio_discard(BlockDriverState *bs,
 512                   int64_t sector_num, int nb_sectors,
 513                   BlockDriverCompletionFunc *cb, void *opaque)
 514 {
 515     IscsiLun *iscsilun = bs->opaque;
 516     struct iscsi_context *iscsi = iscsilun->iscsi;
 517     IscsiAIOCB *acb;
 518     struct unmap_list list[1];
 519
 520     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 521
 522     acb->iscsilun = iscsilun;
 523     acb->canceled   = 0;
 524     acb->bh         = NULL;
 525     acb->status     = -EINPROGRESS;
 526     acb->buf        = NULL;
 527
 528     list[0].lba = sector_qemu2lun(sector_num, iscsilun);
 529     list[0].num = nb_sectors * BDRV_SECTOR_SIZE / iscsilun->block_size;
 530
 531     acb->task = iscsi_unmap_task(iscsi, iscsilun->lun,
 532                                  0, 0, &list[0], 1,
 533                                  iscsi_unmap_cb,
 534                                  acb);
 535     if (acb->task == NULL) {
 536         error_report("iSCSI: Failed to send unmap command. %s",
 537                      iscsi_get_error(iscsi));
 538         qemu_aio_release(acb);
 539         return NULL;
 540     }
 541
 542     iscsi_set_events(iscsilun);
 543
 544     return &acb->common;
 545 }
 546
 547 #ifdef __linux__
 548 static void
 549 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 550                      void *command_data, void *opaque)
 551 {
 552     IscsiAIOCB *acb = opaque;
 553
 554     if (acb->canceled != 0) {
 555         return;
 556     }
 557
 558     acb->status = 0;
 559     if (status < 0) {
 560         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 561                      iscsi_get_error(iscsi));
 562         acb->status = -EIO;
 563     }
 564
 565     acb->ioh->driver_status = 0;
 566     acb->ioh->host_status   = 0;
 567     acb->ioh->resid         = 0;
 568
 569 #define SG_ERR_DRIVER_SENSE    0x08
 570
 571     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 572         int ss;
 573
 574         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 575
 576         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 577         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 578              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 579         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 580     }
 581
 582     iscsi_schedule_bh(acb);
 583 }
 584
 585 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 586         unsigned long int req, void *buf,
 587         BlockDriverCompletionFunc *cb, void *opaque)
 588 {
 589     IscsiLun *iscsilun = bs->opaque;
 590     struct iscsi_context *iscsi = iscsilun->iscsi;
 591     struct iscsi_data data;
 592     IscsiAIOCB *acb;
 593
 594     assert(req == SG_IO);
 595
 596     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 597
 598     acb->iscsilun = iscsilun;
 599     acb->canceled    = 0;
 600     acb->bh          = NULL;
 601     acb->status      = -EINPROGRESS;
 602     acb->buf         = NULL;
 603     acb->ioh         = buf;
 604
 605     acb->task = malloc(sizeof(struct scsi_task));
 606     if (acb->task == NULL) {
 607         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 608                      iscsi_get_error(iscsi));
 609         qemu_aio_release(acb);
 610         return NULL;
 611     }
 612     memset(acb->task, 0, sizeof(struct scsi_task));
 613
 614     switch (acb->ioh->dxfer_direction) {
 615     case SG_DXFER_TO_DEV:
 616         acb->task->xfer_dir = SCSI_XFER_WRITE;
 617         break;
 618     case SG_DXFER_FROM_DEV:
 619         acb->task->xfer_dir = SCSI_XFER_READ;
 620         break;
 621     default:
 622         acb->task->xfer_dir = SCSI_XFER_NONE;
 623         break;
 624     }
 625
 626     acb->task->cdb_size = acb->ioh->cmd_len;
 627     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 628     acb->task->expxferlen = acb->ioh->dxfer_len;
 629
 630     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 631         data.data = acb->ioh->dxferp;
 632         data.size = acb->ioh->dxfer_len;
 633     }
 634     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 635                                  iscsi_aio_ioctl_cb,
 636                                  (acb->task->xfer_dir == SCSI_XFER_WRITE) ?
 637                                      &data : NULL,
 638                                  acb) != 0) {
 639         scsi_free_scsi_task(acb->task);
 640         qemu_aio_release(acb);
 641         return NULL;
 642     }
 643
 644     /* tell libiscsi to read straight into the buffer we got from ioctl */
 645     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 646         scsi_task_add_data_in_buffer(acb->task,
 647                                      acb->ioh->dxfer_len,
 648                                      acb->ioh->dxferp);
 649     }
 650
 651     iscsi_set_events(iscsilun);
 652
 653     return &acb->common;
 654 }
 655
 656
 657 static void ioctl_cb(void *opaque, int status)
 658 {
 659     int *p_status = opaque;
 660     *p_status = status;
 661 }
 662
 663 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 664 {
 665     IscsiLun *iscsilun = bs->opaque;
 666     int status;
 667
 668     switch (req) {
 669     case SG_GET_VERSION_NUM:
 670         *(int *)buf = 30000;
 671         break;
 672     case SG_GET_SCSI_ID:
 673         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 674         break;
 675     case SG_IO:
 676         status = -EINPROGRESS;
 677         iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status);
 678
 679         while (status == -EINPROGRESS) {
 680             qemu_aio_wait();
 681         }
 682
 683         return 0;
 684     default:
 685         return -1;
 686     }
 687     return 0;
 688 }
 689 #endif
 690
 691 static int64_t
 692 iscsi_getlength(BlockDriverState *bs)
 693 {
 694     IscsiLun *iscsilun = bs->opaque;
 695     int64_t len;
 696
 697     len  = iscsilun->num_blocks;
 698     len *= iscsilun->block_size;
 699
 700     return len;
 701 }
 702
 703 static int parse_chap(struct iscsi_context *iscsi, const char *target)
 704 {
 705     QemuOptsList *list;
 706     QemuOpts *opts;
 707     const char *user = NULL;
 708     const char *password = NULL;
 709
 710     list = qemu_find_opts("iscsi");
 711     if (!list) {
 712         return 0;
 713     }
 714
 715     opts = qemu_opts_find(list, target);
 716     if (opts == NULL) {
 717         opts = QTAILQ_FIRST(&list->head);
 718         if (!opts) {
 719             return 0;
 720         }
 721     }
 722
 723     user = qemu_opt_get(opts, "user");
 724     if (!user) {
 725         return 0;
 726     }
 727
 728     password = qemu_opt_get(opts, "password");
 729     if (!password) {
 730         error_report("CHAP username specified but no password was given");
 731         return -1;
 732     }
 733
 734     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
 735         error_report("Failed to set initiator username and password");
 736         return -1;
 737     }
 738
 739     return 0;
 740 }
 741
 742 static void parse_header_digest(struct iscsi_context *iscsi, const char *target)
 743 {
 744     QemuOptsList *list;
 745     QemuOpts *opts;
 746     const char *digest = NULL;
 747
 748     list = qemu_find_opts("iscsi");
 749     if (!list) {
 750         return;
 751     }
 752
 753     opts = qemu_opts_find(list, target);
 754     if (opts == NULL) {
 755         opts = QTAILQ_FIRST(&list->head);
 756         if (!opts) {
 757             return;
 758         }
 759     }
 760
 761     digest = qemu_opt_get(opts, "header-digest");
 762     if (!digest) {
 763         return;
 764     }
 765
 766     if (!strcmp(digest, "CRC32C")) {
 767         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
 768     } else if (!strcmp(digest, "NONE")) {
 769         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
 770     } else if (!strcmp(digest, "CRC32C-NONE")) {
 771         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
 772     } else if (!strcmp(digest, "NONE-CRC32C")) {
 773         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 774     } else {
 775         error_report("Invalid header-digest setting : %s", digest);
 776     }
 777 }
 778
 779 static char *parse_initiator_name(const char *target)
 780 {
 781     QemuOptsList *list;
 782     QemuOpts *opts;
 783     const char *name = NULL;
 784     const char *iscsi_name = qemu_get_vm_name();
 785
 786     list = qemu_find_opts("iscsi");
 787     if (list) {
 788         opts = qemu_opts_find(list, target);
 789         if (!opts) {
 790             opts = QTAILQ_FIRST(&list->head);
 791         }
 792         if (opts) {
 793             name = qemu_opt_get(opts, "initiator-name");
 794         }
 795     }
 796
 797     if (name) {
 798         return g_strdup(name);
 799     } else {
 800         return g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
 801                                iscsi_name ? ":" : "",
 802                                iscsi_name ? iscsi_name : "");
 803     }
 804 }
 805
 806 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
 807 static void iscsi_nop_timed_event(void *opaque)
 808 {
 809     IscsiLun *iscsilun = opaque;
 810
 811     if (iscsi_get_nops_in_flight(iscsilun->iscsi) > MAX_NOP_FAILURES) {
 812         error_report("iSCSI: NOP timeout. Reconnecting...");
 813         iscsi_reconnect(iscsilun->iscsi);
 814     }
 815
 816     if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
 817         error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
 818         return;
 819     }
 820
 821     qemu_mod_timer(iscsilun->nop_timer, qemu_get_clock_ms(rt_clock) + NOP_INTERVAL);
 822     iscsi_set_events(iscsilun);
 823 }
 824 #endif
 825
 826 /*
 827  * We support iscsi url's on the form
 828  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
 829  */
 830 static int iscsi_open(BlockDriverState *bs, const char *filename, int flags)
 831 {
 832     IscsiLun *iscsilun = bs->opaque;
 833     struct iscsi_context *iscsi = NULL;
 834     struct iscsi_url *iscsi_url = NULL;
 835     struct scsi_task *task = NULL;
 836     struct scsi_inquiry_standard *inq = NULL;
 837     struct scsi_readcapacity10 *rc10 = NULL;
 838     struct scsi_readcapacity16 *rc16 = NULL;
 839     char *initiator_name = NULL;
 840     int ret;
 841
 842     if ((BDRV_SECTOR_SIZE % 512) != 0) {
 843         error_report("iSCSI: Invalid BDRV_SECTOR_SIZE. "
 844                      "BDRV_SECTOR_SIZE(%lld) is not a multiple "
 845                      "of 512", BDRV_SECTOR_SIZE);
 846         return -EINVAL;
 847     }
 848
 849     iscsi_url = iscsi_parse_full_url(iscsi, filename);
 850     if (iscsi_url == NULL) {
 851         error_report("Failed to parse URL : %s", filename);
 852         ret = -EINVAL;
 853         goto out;
 854     }
 855
 856     memset(iscsilun, 0, sizeof(IscsiLun));
 857
 858     initiator_name = parse_initiator_name(iscsi_url->target);
 859
 860     iscsi = iscsi_create_context(initiator_name);
 861     if (iscsi == NULL) {
 862         error_report("iSCSI: Failed to create iSCSI context.");
 863         ret = -ENOMEM;
 864         goto out;
 865     }
 866
 867     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
 868         error_report("iSCSI: Failed to set target name.");
 869         ret = -EINVAL;
 870         goto out;
 871     }
 872
 873     if (iscsi_url->user != NULL) {
 874         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
 875                                               iscsi_url->passwd);
 876         if (ret != 0) {
 877             error_report("Failed to set initiator username and password");
 878             ret = -EINVAL;
 879             goto out;
 880         }
 881     }
 882
 883     /* check if we got CHAP username/password via the options */
 884     if (parse_chap(iscsi, iscsi_url->target) != 0) {
 885         error_report("iSCSI: Failed to set CHAP user/password");
 886         ret = -EINVAL;
 887         goto out;
 888     }
 889
 890     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
 891         error_report("iSCSI: Failed to set session type to normal.");
 892         ret = -EINVAL;
 893         goto out;
 894     }
 895
 896     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
 897
 898     /* check if we got HEADER_DIGEST via the options */
 899     parse_header_digest(iscsi, iscsi_url->target);
 900
 901     if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
 902         error_report("iSCSI: Failed to connect to LUN : %s",
 903             iscsi_get_error(iscsi));
 904         ret = -EINVAL;
 905         goto out;
 906     }
 907
 908     iscsilun->iscsi = iscsi;
 909     iscsilun->lun   = iscsi_url->lun;
 910
 911     task = iscsi_inquiry_sync(iscsi, iscsilun->lun, 0, 0, 36);
 912
 913     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
 914         error_report("iSCSI: failed to send inquiry command.");
 915         ret = -EINVAL;
 916         goto out;
 917     }
 918
 919     inq = scsi_datain_unmarshall(task);
 920     if (inq == NULL) {
 921         error_report("iSCSI: Failed to unmarshall inquiry data.");
 922         ret = -EINVAL;
 923         goto out;
 924     }
 925
 926     iscsilun->type = inq->periperal_device_type;
 927
 928     scsi_free_scsi_task(task);
 929
 930     switch (iscsilun->type) {
 931     case TYPE_DISK:
 932         task = iscsi_readcapacity16_sync(iscsi, iscsilun->lun);
 933         if (task == NULL || task->status != SCSI_STATUS_GOOD) {
 934             error_report("iSCSI: failed to send readcapacity16 command.");
 935             ret = -EINVAL;
 936             goto out;
 937         }
 938         rc16 = scsi_datain_unmarshall(task);
 939         if (rc16 == NULL) {
 940             error_report("iSCSI: Failed to unmarshall readcapacity16 data.");
 941             ret = -EINVAL;
 942             goto out;
 943         }
 944         iscsilun->block_size = rc16->block_length;
 945         iscsilun->num_blocks = rc16->returned_lba + 1;
 946         break;
 947     case TYPE_ROM:
 948         task = iscsi_readcapacity10_sync(iscsi, iscsilun->lun, 0, 0);
 949         if (task == NULL || task->status != SCSI_STATUS_GOOD) {
 950             error_report("iSCSI: failed to send readcapacity10 command.");
 951             ret = -EINVAL;
 952             goto out;
 953         }
 954         rc10 = scsi_datain_unmarshall(task);
 955         if (rc10 == NULL) {
 956             error_report("iSCSI: Failed to unmarshall readcapacity10 data.");
 957             ret = -EINVAL;
 958             goto out;
 959         }
 960         iscsilun->block_size = rc10->block_size;
 961         if (rc10->lba == 0) {
 962             /* blank disk loaded */
 963             iscsilun->num_blocks = 0;
 964         } else {
 965             iscsilun->num_blocks = rc10->lba + 1;
 966         }
 967         break;
 968     default:
 969         break;
 970     }
 971
 972     bs->total_sectors    = iscsilun->num_blocks *
 973                            iscsilun->block_size / BDRV_SECTOR_SIZE ;
 974
 975     /* Medium changer or tape. We dont have any emulation for this so this must
 976      * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
 977      * to read from the device to guess the image format.
 978      */
 979     if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
 980         iscsilun->type == TYPE_TAPE) {
 981         bs->sg = 1;
 982     }
 983
 984     ret = 0;
 985
 986 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
 987     /* Set up a timer for sending out iSCSI NOPs */
 988     iscsilun->nop_timer = qemu_new_timer_ms(rt_clock, iscsi_nop_timed_event, iscsilun);
 989     qemu_mod_timer(iscsilun->nop_timer, qemu_get_clock_ms(rt_clock) + NOP_INTERVAL);
 990 #endif
 991
 992 out:
 993     if (initiator_name != NULL) {
 994         g_free(initiator_name);
 995     }
 996     if (iscsi_url != NULL) {
 997         iscsi_destroy_url(iscsi_url);
 998     }
 999     if (task != NULL) {
1000         scsi_free_scsi_task(task);
1001     }
1002
1003     if (ret) {
1004         if (iscsi != NULL) {
1005             iscsi_destroy_context(iscsi);
1006         }
1007         memset(iscsilun, 0, sizeof(IscsiLun));
1008     }
1009     return ret;
1010 }
1011
1012 static void iscsi_close(BlockDriverState *bs)
1013 {
1014     IscsiLun *iscsilun = bs->opaque;
1015     struct iscsi_context *iscsi = iscsilun->iscsi;
1016
1017     if (iscsilun->nop_timer) {
1018         qemu_del_timer(iscsilun->nop_timer);
1019         qemu_free_timer(iscsilun->nop_timer);
1020     }
1021     qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL, NULL);
1022     iscsi_destroy_context(iscsi);
1023     memset(iscsilun, 0, sizeof(IscsiLun));
1024 }
1025
1026 static int iscsi_has_zero_init(BlockDriverState *bs)
1027 {
1028     return 0;
1029 }
1030
1031 static int iscsi_create(const char *filename, QEMUOptionParameter *options)
1032 {
1033     int ret = 0;
1034     int64_t total_size = 0;
1035     BlockDriverState bs;
1036     IscsiLun *iscsilun = NULL;
1037
1038     memset(&bs, 0, sizeof(BlockDriverState));
1039
1040     /* Read out options */
1041     while (options && options->name) {
1042         if (!strcmp(options->name, "size")) {
1043             total_size = options->value.n / BDRV_SECTOR_SIZE;
1044         }
1045         options++;
1046     }
1047
1048     bs.opaque = g_malloc0(sizeof(struct IscsiLun));
1049     iscsilun = bs.opaque;
1050
1051     ret = iscsi_open(&bs, filename, 0);
1052     if (ret != 0) {
1053         goto out;
1054     }
1055     if (iscsilun->nop_timer) {
1056         qemu_del_timer(iscsilun->nop_timer);
1057         qemu_free_timer(iscsilun->nop_timer);
1058     }
1059     if (iscsilun->type != TYPE_DISK) {
1060         ret = -ENODEV;
1061         goto out;
1062     }
1063     if (bs.total_sectors < total_size) {
1064         ret = -ENOSPC;
1065     }
1066
1067     ret = 0;
1068 out:
1069     if (iscsilun->iscsi != NULL) {
1070         iscsi_destroy_context(iscsilun->iscsi);
1071     }
1072     g_free(bs.opaque);
1073     return ret;
1074 }
1075
1076 static QEMUOptionParameter iscsi_create_options[] = {
1077     {
1078         .name = BLOCK_OPT_SIZE,
1079         .type = OPT_SIZE,
1080         .help = "Virtual disk size"
1081     },
1082     { NULL }
1083 };
1084
1085 static BlockDriver bdrv_iscsi = {
1086     .format_name     = "iscsi",
1087     .protocol_name   = "iscsi",
1088
1089     .instance_size   = sizeof(IscsiLun),
1090     .bdrv_file_open  = iscsi_open,
1091     .bdrv_close      = iscsi_close,
1092     .bdrv_create     = iscsi_create,
1093     .create_options  = iscsi_create_options,
1094
1095     .bdrv_getlength  = iscsi_getlength,
1096
1097     .bdrv_aio_readv  = iscsi_aio_readv,
1098     .bdrv_aio_writev = iscsi_aio_writev,
1099     .bdrv_aio_flush  = iscsi_aio_flush,
1100
1101     .bdrv_aio_discard = iscsi_aio_discard,
1102     .bdrv_has_zero_init = iscsi_has_zero_init,
1103
1104 #ifdef __linux__
1105     .bdrv_ioctl       = iscsi_ioctl,
1106     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1107 #endif
1108 };
1109
1110 static QemuOptsList qemu_iscsi_opts = {
1111     .name = "iscsi",
1112     .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1113     .desc = {
1114         {
1115             .name = "user",
1116             .type = QEMU_OPT_STRING,
1117             .help = "username for CHAP authentication to target",
1118         },{
1119             .name = "password",
1120             .type = QEMU_OPT_STRING,
1121             .help = "password for CHAP authentication to target",
1122         },{
1123             .name = "header-digest",
1124             .type = QEMU_OPT_STRING,
1125             .help = "HeaderDigest setting. "
1126                     "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1127         },{
1128             .name = "initiator-name",
1129             .type = QEMU_OPT_STRING,
1130             .help = "Initiator iqn name to use when connecting",
1131         },
1132         { /* end of list */ }
1133     },
1134 };
1135
1136 static void iscsi_block_init(void)
1137 {
1138     bdrv_register(&bdrv_iscsi);
1139     qemu_add_opts(&qemu_iscsi_opts);
1140 }
1141
1142 block_init(iscsi_block_init);