block/iscsi.c

   1 /*
   2  * QEMU Block driver for iSCSI images
   3  *
   4  * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
   5  * Copyright (c) 2012-2014 Peter Lieven <pl@kamp.de>
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  23  * THE SOFTWARE.
  24  */
  25
  26 #include "config-host.h"
  27
  28 #include <poll.h>
  29 #include <arpa/inet.h>
  30 #include "qemu-common.h"
  31 #include "qemu/config-file.h"
  32 #include "qemu/error-report.h"
  33 #include "qemu/bitops.h"
  34 #include "qemu/bitmap.h"
  35 #include "block/block_int.h"
  36 #include "trace.h"
  37 #include "block/scsi.h"
  38 #include "qemu/iov.h"
  39 #include "sysemu/sysemu.h"
  40 #include "qmp-commands.h"
  41
  42 #include <iscsi/iscsi.h>
  43 #include <iscsi/scsi-lowlevel.h>
  44
  45 #ifdef __linux__
  46 #include <scsi/sg.h>
  47 #include <block/scsi.h>
  48 #endif
  49
  50 typedef struct IscsiLun {
  51     struct iscsi_context *iscsi;
  52     AioContext *aio_context;
  53     int lun;
  54     enum scsi_inquiry_peripheral_device_type type;
  55     int block_size;
  56     uint64_t num_blocks;
  57     int events;
  58     QEMUTimer *nop_timer;
  59     uint8_t lbpme;
  60     uint8_t lbprz;
  61     uint8_t has_write_same;
  62     struct scsi_inquiry_logical_block_provisioning lbp;
  63     struct scsi_inquiry_block_limits bl;
  64     unsigned char *zeroblock;
  65     unsigned long *allocationmap;
  66     int cluster_sectors;
  67 } IscsiLun;
  68
  69 typedef struct IscsiTask {
  70     int status;
  71     int complete;
  72     int retries;
  73     int do_retry;
  74     struct scsi_task *task;
  75     Coroutine *co;
  76     QEMUBH *bh;
  77     IscsiLun *iscsilun;
  78 } IscsiTask;
  79
  80 typedef struct IscsiAIOCB {
  81     BlockDriverAIOCB common;
  82     QEMUIOVector *qiov;
  83     QEMUBH *bh;
  84     IscsiLun *iscsilun;
  85     struct scsi_task *task;
  86     uint8_t *buf;
  87     int status;
  88     int canceled;
  89     int retries;
  90     int64_t sector_num;
  91     int nb_sectors;
  92 #ifdef __linux__
  93     sg_io_hdr_t *ioh;
  94 #endif
  95 } IscsiAIOCB;
  96
  97 #define NOP_INTERVAL 5000
  98 #define MAX_NOP_FAILURES 3
  99 #define ISCSI_CMD_RETRIES 5
 100
 101 /* this threshhold is a trade-off knob to choose between
 102  * the potential additional overhead of an extra GET_LBA_STATUS request
 103  * vs. unnecessarily reading a lot of zero sectors over the wire.
 104  * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
 105  * sectors we check the allocation status of the area covered by the
 106  * request first if the allocationmap indicates that the area might be
 107  * unallocated. */
 108 #define ISCSI_CHECKALLOC_THRES 64
 109
 110 static void
 111 iscsi_bh_cb(void *p)
 112 {
 113     IscsiAIOCB *acb = p;
 114
 115     qemu_bh_delete(acb->bh);
 116
 117     g_free(acb->buf);
 118     acb->buf = NULL;
 119
 120     if (acb->canceled == 0) {
 121         acb->common.cb(acb->common.opaque, acb->status);
 122     }
 123
 124     if (acb->task != NULL) {
 125         scsi_free_scsi_task(acb->task);
 126         acb->task = NULL;
 127     }
 128
 129     qemu_aio_release(acb);
 130 }
 131
 132 static void
 133 iscsi_schedule_bh(IscsiAIOCB *acb)
 134 {
 135     if (acb->bh) {
 136         return;
 137     }
 138     acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb);
 139     qemu_bh_schedule(acb->bh);
 140 }
 141
 142 static void iscsi_co_generic_bh_cb(void *opaque)
 143 {
 144     struct IscsiTask *iTask = opaque;
 145     qemu_bh_delete(iTask->bh);
 146     qemu_coroutine_enter(iTask->co, NULL);
 147 }
 148
 149 static void
 150 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
 151                         void *command_data, void *opaque)
 152 {
 153     struct IscsiTask *iTask = opaque;
 154     struct scsi_task *task = command_data;
 155
 156     iTask->complete = 1;
 157     iTask->status = status;
 158     iTask->do_retry = 0;
 159     iTask->task = task;
 160
 161     if (iTask->retries-- > 0 && status == SCSI_STATUS_CHECK_CONDITION
 162         && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
 163         error_report("iSCSI CheckCondition: %s", iscsi_get_error(iscsi));
 164         iTask->do_retry = 1;
 165         goto out;
 166     }
 167
 168     if (status != SCSI_STATUS_GOOD) {
 169         error_report("iSCSI Failure: %s", iscsi_get_error(iscsi));
 170     }
 171
 172 out:
 173     if (iTask->co) {
 174         iTask->bh = aio_bh_new(iTask->iscsilun->aio_context,
 175                                iscsi_co_generic_bh_cb, iTask);
 176         qemu_bh_schedule(iTask->bh);
 177     }
 178 }
 179
 180 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
 181 {
 182     *iTask = (struct IscsiTask) {
 183         .co             = qemu_coroutine_self(),
 184         .retries        = ISCSI_CMD_RETRIES,
 185         .iscsilun       = iscsilun,
 186     };
 187 }
 188
 189 static void
 190 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
 191                     void *private_data)
 192 {
 193     IscsiAIOCB *acb = private_data;
 194
 195     acb->status = -ECANCELED;
 196     iscsi_schedule_bh(acb);
 197 }
 198
 199 static void
 200 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
 201 {
 202     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
 203     IscsiLun *iscsilun = acb->iscsilun;
 204
 205     if (acb->status != -EINPROGRESS) {
 206         return;
 207     }
 208
 209     acb->canceled = 1;
 210
 211     /* send a task mgmt call to the target to cancel the task on the target */
 212     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
 213                                      iscsi_abort_task_cb, acb);
 214
 215     while (acb->status == -EINPROGRESS) {
 216         aio_poll(iscsilun->aio_context, true);
 217     }
 218 }
 219
 220 static const AIOCBInfo iscsi_aiocb_info = {
 221     .aiocb_size         = sizeof(IscsiAIOCB),
 222     .cancel             = iscsi_aio_cancel,
 223 };
 224
 225
 226 static void iscsi_process_read(void *arg);
 227 static void iscsi_process_write(void *arg);
 228
 229 static void
 230 iscsi_set_events(IscsiLun *iscsilun)
 231 {
 232     struct iscsi_context *iscsi = iscsilun->iscsi;
 233     int ev;
 234
 235     /* We always register a read handler.  */
 236     ev = POLLIN;
 237     ev |= iscsi_which_events(iscsi);
 238     if (ev != iscsilun->events) {
 239         aio_set_fd_handler(iscsilun->aio_context,
 240                            iscsi_get_fd(iscsi),
 241                            iscsi_process_read,
 242                            (ev & POLLOUT) ? iscsi_process_write : NULL,
 243                            iscsilun);
 244
 245     }
 246
 247     iscsilun->events = ev;
 248 }
 249
 250 static void
 251 iscsi_process_read(void *arg)
 252 {
 253     IscsiLun *iscsilun = arg;
 254     struct iscsi_context *iscsi = iscsilun->iscsi;
 255
 256     iscsi_service(iscsi, POLLIN);
 257     iscsi_set_events(iscsilun);
 258 }
 259
 260 static void
 261 iscsi_process_write(void *arg)
 262 {
 263     IscsiLun *iscsilun = arg;
 264     struct iscsi_context *iscsi = iscsilun->iscsi;
 265
 266     iscsi_service(iscsi, POLLOUT);
 267     iscsi_set_events(iscsilun);
 268 }
 269
 270 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
 271 {
 272     return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
 273 }
 274
 275 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
 276 {
 277     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
 278 }
 279
 280 static bool is_request_lun_aligned(int64_t sector_num, int nb_sectors,
 281                                       IscsiLun *iscsilun)
 282 {
 283     if ((sector_num * BDRV_SECTOR_SIZE) % iscsilun->block_size ||
 284         (nb_sectors * BDRV_SECTOR_SIZE) % iscsilun->block_size) {
 285             error_report("iSCSI misaligned request: "
 286                          "iscsilun->block_size %u, sector_num %" PRIi64
 287                          ", nb_sectors %d",
 288                          iscsilun->block_size, sector_num, nb_sectors);
 289             return 0;
 290     }
 291     return 1;
 292 }
 293
 294 static void iscsi_allocationmap_set(IscsiLun *iscsilun, int64_t sector_num,
 295                                     int nb_sectors)
 296 {
 297     if (iscsilun->allocationmap == NULL) {
 298         return;
 299     }
 300     bitmap_set(iscsilun->allocationmap,
 301                sector_num / iscsilun->cluster_sectors,
 302                DIV_ROUND_UP(nb_sectors, iscsilun->cluster_sectors));
 303 }
 304
 305 static void iscsi_allocationmap_clear(IscsiLun *iscsilun, int64_t sector_num,
 306                                       int nb_sectors)
 307 {
 308     int64_t cluster_num, nb_clusters;
 309     if (iscsilun->allocationmap == NULL) {
 310         return;
 311     }
 312     cluster_num = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors);
 313     nb_clusters = (sector_num + nb_sectors) / iscsilun->cluster_sectors
 314                   - cluster_num;
 315     if (nb_clusters > 0) {
 316         bitmap_clear(iscsilun->allocationmap, cluster_num, nb_clusters);
 317     }
 318 }
 319
 320 static int coroutine_fn iscsi_co_writev(BlockDriverState *bs,
 321                                         int64_t sector_num, int nb_sectors,
 322                                         QEMUIOVector *iov)
 323 {
 324     IscsiLun *iscsilun = bs->opaque;
 325     struct IscsiTask iTask;
 326     uint64_t lba;
 327     uint32_t num_sectors;
 328     uint8_t *data = NULL;
 329     uint8_t *buf = NULL;
 330
 331     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 332         return -EINVAL;
 333     }
 334
 335     lba = sector_qemu2lun(sector_num, iscsilun);
 336     num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
 337 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 338     /* if the iovec only contains one buffer we can pass it directly */
 339     if (iov->niov == 1) {
 340         data = iov->iov[0].iov_base;
 341     } else {
 342         size_t size = MIN(nb_sectors * BDRV_SECTOR_SIZE, iov->size);
 343         buf = g_malloc(size);
 344         qemu_iovec_to_buf(iov, 0, buf, size);
 345         data = buf;
 346     }
 347 #endif
 348     iscsi_co_init_iscsitask(iscsilun, &iTask);
 349 retry:
 350     iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
 351                                     data, num_sectors * iscsilun->block_size,
 352                                     iscsilun->block_size, 0, 0, 0, 0, 0,
 353                                     iscsi_co_generic_cb, &iTask);
 354     if (iTask.task == NULL) {
 355         g_free(buf);
 356         return -ENOMEM;
 357     }
 358 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 359     scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
 360                           iov->niov);
 361 #endif
 362     while (!iTask.complete) {
 363         iscsi_set_events(iscsilun);
 364         qemu_coroutine_yield();
 365     }
 366
 367     if (iTask.task != NULL) {
 368         scsi_free_scsi_task(iTask.task);
 369         iTask.task = NULL;
 370     }
 371
 372     if (iTask.do_retry) {
 373         iTask.complete = 0;
 374         goto retry;
 375     }
 376
 377     g_free(buf);
 378
 379     if (iTask.status != SCSI_STATUS_GOOD) {
 380         return -EIO;
 381     }
 382
 383     iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors);
 384
 385     return 0;
 386 }
 387
 388
 389 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 390 static bool iscsi_allocationmap_is_allocated(IscsiLun *iscsilun,
 391                                              int64_t sector_num, int nb_sectors)
 392 {
 393     unsigned long size;
 394     if (iscsilun->allocationmap == NULL) {
 395         return true;
 396     }
 397     size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
 398     return !(find_next_bit(iscsilun->allocationmap, size,
 399                            sector_num / iscsilun->cluster_sectors) == size);
 400 }
 401
 402 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
 403                                                   int64_t sector_num,
 404                                                   int nb_sectors, int *pnum)
 405 {
 406     IscsiLun *iscsilun = bs->opaque;
 407     struct scsi_get_lba_status *lbas = NULL;
 408     struct scsi_lba_status_descriptor *lbasd = NULL;
 409     struct IscsiTask iTask;
 410     int64_t ret;
 411
 412     iscsi_co_init_iscsitask(iscsilun, &iTask);
 413
 414     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 415         ret = -EINVAL;
 416         goto out;
 417     }
 418
 419     /* default to all sectors allocated */
 420     ret = BDRV_BLOCK_DATA;
 421     ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
 422     *pnum = nb_sectors;
 423
 424     /* LUN does not support logical block provisioning */
 425     if (iscsilun->lbpme == 0) {
 426         goto out;
 427     }
 428
 429 retry:
 430     if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
 431                                   sector_qemu2lun(sector_num, iscsilun),
 432                                   8 + 16, iscsi_co_generic_cb,
 433                                   &iTask) == NULL) {
 434         ret = -ENOMEM;
 435         goto out;
 436     }
 437
 438     while (!iTask.complete) {
 439         iscsi_set_events(iscsilun);
 440         qemu_coroutine_yield();
 441     }
 442
 443     if (iTask.do_retry) {
 444         if (iTask.task != NULL) {
 445             scsi_free_scsi_task(iTask.task);
 446             iTask.task = NULL;
 447         }
 448         iTask.complete = 0;
 449         goto retry;
 450     }
 451
 452     if (iTask.status != SCSI_STATUS_GOOD) {
 453         /* in case the get_lba_status_callout fails (i.e.
 454          * because the device is busy or the cmd is not
 455          * supported) we pretend all blocks are allocated
 456          * for backwards compatibility */
 457         goto out;
 458     }
 459
 460     lbas = scsi_datain_unmarshall(iTask.task);
 461     if (lbas == NULL) {
 462         ret = -EIO;
 463         goto out;
 464     }
 465
 466     lbasd = &lbas->descriptors[0];
 467
 468     if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
 469         ret = -EIO;
 470         goto out;
 471     }
 472
 473     *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
 474
 475     if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
 476         lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
 477         ret &= ~BDRV_BLOCK_DATA;
 478         if (iscsilun->lbprz) {
 479             ret |= BDRV_BLOCK_ZERO;
 480         }
 481     }
 482
 483     if (ret & BDRV_BLOCK_ZERO) {
 484         iscsi_allocationmap_clear(iscsilun, sector_num, *pnum);
 485     } else {
 486         iscsi_allocationmap_set(iscsilun, sector_num, *pnum);
 487     }
 488
 489     if (*pnum > nb_sectors) {
 490         *pnum = nb_sectors;
 491     }
 492 out:
 493     if (iTask.task != NULL) {
 494         scsi_free_scsi_task(iTask.task);
 495     }
 496     return ret;
 497 }
 498
 499 #endif /* LIBISCSI_FEATURE_IOVECTOR */
 500
 501
 502 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
 503                                        int64_t sector_num, int nb_sectors,
 504                                        QEMUIOVector *iov)
 505 {
 506     IscsiLun *iscsilun = bs->opaque;
 507     struct IscsiTask iTask;
 508     uint64_t lba;
 509     uint32_t num_sectors;
 510 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
 511     int i;
 512 #endif
 513
 514     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 515         return -EINVAL;
 516     }
 517
 518 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 519     if (iscsilun->lbprz && nb_sectors >= ISCSI_CHECKALLOC_THRES &&
 520         !iscsi_allocationmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
 521         int64_t ret;
 522         int pnum;
 523         ret = iscsi_co_get_block_status(bs, sector_num, INT_MAX, &pnum);
 524         if (ret < 0) {
 525             return ret;
 526         }
 527         if (ret & BDRV_BLOCK_ZERO && pnum >= nb_sectors) {
 528             qemu_iovec_memset(iov, 0, 0x00, iov->size);
 529             return 0;
 530         }
 531     }
 532 #endif
 533
 534     lba = sector_qemu2lun(sector_num, iscsilun);
 535     num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
 536
 537     iscsi_co_init_iscsitask(iscsilun, &iTask);
 538 retry:
 539     switch (iscsilun->type) {
 540     case TYPE_DISK:
 541         iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
 542                                        num_sectors * iscsilun->block_size,
 543                                        iscsilun->block_size, 0, 0, 0, 0, 0,
 544                                        iscsi_co_generic_cb, &iTask);
 545         break;
 546     default:
 547         iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
 548                                        num_sectors * iscsilun->block_size,
 549                                        iscsilun->block_size,
 550 #if !defined(CONFIG_LIBISCSI_1_4) /* API change from 1.4.0 to 1.5.0 */
 551                                        0, 0, 0, 0, 0,
 552 #endif
 553                                        iscsi_co_generic_cb, &iTask);
 554         break;
 555     }
 556     if (iTask.task == NULL) {
 557         return -ENOMEM;
 558     }
 559 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 560     scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
 561 #else
 562     for (i = 0; i < iov->niov; i++) {
 563         scsi_task_add_data_in_buffer(iTask.task,
 564                                      iov->iov[i].iov_len,
 565                                      iov->iov[i].iov_base);
 566     }
 567 #endif
 568
 569     while (!iTask.complete) {
 570         iscsi_set_events(iscsilun);
 571         qemu_coroutine_yield();
 572     }
 573
 574     if (iTask.task != NULL) {
 575         scsi_free_scsi_task(iTask.task);
 576         iTask.task = NULL;
 577     }
 578
 579     if (iTask.do_retry) {
 580         iTask.complete = 0;
 581         goto retry;
 582     }
 583
 584     if (iTask.status != SCSI_STATUS_GOOD) {
 585         return -EIO;
 586     }
 587
 588     return 0;
 589 }
 590
 591 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
 592 {
 593     IscsiLun *iscsilun = bs->opaque;
 594     struct IscsiTask iTask;
 595
 596     if (bs->sg) {
 597         return 0;
 598     }
 599
 600     iscsi_co_init_iscsitask(iscsilun, &iTask);
 601
 602 retry:
 603     if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
 604                                       0, iscsi_co_generic_cb, &iTask) == NULL) {
 605         return -ENOMEM;
 606     }
 607
 608     while (!iTask.complete) {
 609         iscsi_set_events(iscsilun);
 610         qemu_coroutine_yield();
 611     }
 612
 613     if (iTask.task != NULL) {
 614         scsi_free_scsi_task(iTask.task);
 615         iTask.task = NULL;
 616     }
 617
 618     if (iTask.do_retry) {
 619         iTask.complete = 0;
 620         goto retry;
 621     }
 622
 623     if (iTask.status != SCSI_STATUS_GOOD) {
 624         return -EIO;
 625     }
 626
 627     return 0;
 628 }
 629
 630 #ifdef __linux__
 631 static void
 632 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
 633                      void *command_data, void *opaque)
 634 {
 635     IscsiAIOCB *acb = opaque;
 636
 637     g_free(acb->buf);
 638     acb->buf = NULL;
 639
 640     if (acb->canceled != 0) {
 641         return;
 642     }
 643
 644     acb->status = 0;
 645     if (status < 0) {
 646         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
 647                      iscsi_get_error(iscsi));
 648         acb->status = -EIO;
 649     }
 650
 651     acb->ioh->driver_status = 0;
 652     acb->ioh->host_status   = 0;
 653     acb->ioh->resid         = 0;
 654
 655 #define SG_ERR_DRIVER_SENSE    0x08
 656
 657     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
 658         int ss;
 659
 660         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
 661
 662         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
 663         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
 664              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
 665         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
 666     }
 667
 668     iscsi_schedule_bh(acb);
 669 }
 670
 671 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
 672         unsigned long int req, void *buf,
 673         BlockDriverCompletionFunc *cb, void *opaque)
 674 {
 675     IscsiLun *iscsilun = bs->opaque;
 676     struct iscsi_context *iscsi = iscsilun->iscsi;
 677     struct iscsi_data data;
 678     IscsiAIOCB *acb;
 679
 680     assert(req == SG_IO);
 681
 682     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
 683
 684     acb->iscsilun = iscsilun;
 685     acb->canceled    = 0;
 686     acb->bh          = NULL;
 687     acb->status      = -EINPROGRESS;
 688     acb->buf         = NULL;
 689     acb->ioh         = buf;
 690
 691     acb->task = malloc(sizeof(struct scsi_task));
 692     if (acb->task == NULL) {
 693         error_report("iSCSI: Failed to allocate task for scsi command. %s",
 694                      iscsi_get_error(iscsi));
 695         qemu_aio_release(acb);
 696         return NULL;
 697     }
 698     memset(acb->task, 0, sizeof(struct scsi_task));
 699
 700     switch (acb->ioh->dxfer_direction) {
 701     case SG_DXFER_TO_DEV:
 702         acb->task->xfer_dir = SCSI_XFER_WRITE;
 703         break;
 704     case SG_DXFER_FROM_DEV:
 705         acb->task->xfer_dir = SCSI_XFER_READ;
 706         break;
 707     default:
 708         acb->task->xfer_dir = SCSI_XFER_NONE;
 709         break;
 710     }
 711
 712     acb->task->cdb_size = acb->ioh->cmd_len;
 713     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
 714     acb->task->expxferlen = acb->ioh->dxfer_len;
 715
 716     data.size = 0;
 717     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
 718         if (acb->ioh->iovec_count == 0) {
 719             data.data = acb->ioh->dxferp;
 720             data.size = acb->ioh->dxfer_len;
 721         } else {
 722 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 723             scsi_task_set_iov_out(acb->task,
 724                                  (struct scsi_iovec *) acb->ioh->dxferp,
 725                                  acb->ioh->iovec_count);
 726 #else
 727             struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
 728
 729             acb->buf = g_malloc(acb->ioh->dxfer_len);
 730             data.data = acb->buf;
 731             data.size = iov_to_buf(iov, acb->ioh->iovec_count, 0,
 732                                    acb->buf, acb->ioh->dxfer_len);
 733 #endif
 734         }
 735     }
 736
 737     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
 738                                  iscsi_aio_ioctl_cb,
 739                                  (data.size > 0) ? &data : NULL,
 740                                  acb) != 0) {
 741         scsi_free_scsi_task(acb->task);
 742         qemu_aio_release(acb);
 743         return NULL;
 744     }
 745
 746     /* tell libiscsi to read straight into the buffer we got from ioctl */
 747     if (acb->task->xfer_dir == SCSI_XFER_READ) {
 748         if (acb->ioh->iovec_count == 0) {
 749             scsi_task_add_data_in_buffer(acb->task,
 750                                          acb->ioh->dxfer_len,
 751                                          acb->ioh->dxferp);
 752         } else {
 753 #if defined(LIBISCSI_FEATURE_IOVECTOR)
 754             scsi_task_set_iov_in(acb->task,
 755                                  (struct scsi_iovec *) acb->ioh->dxferp,
 756                                  acb->ioh->iovec_count);
 757 #else
 758             int i;
 759             for (i = 0; i < acb->ioh->iovec_count; i++) {
 760                 struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
 761
 762                 scsi_task_add_data_in_buffer(acb->task,
 763                     iov[i].iov_len,
 764                     iov[i].iov_base);
 765             }
 766 #endif
 767         }
 768     }
 769
 770     iscsi_set_events(iscsilun);
 771
 772     return &acb->common;
 773 }
 774
 775
 776 static void ioctl_cb(void *opaque, int status)
 777 {
 778     int *p_status = opaque;
 779     *p_status = status;
 780 }
 781
 782 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 783 {
 784     IscsiLun *iscsilun = bs->opaque;
 785     int status;
 786
 787     switch (req) {
 788     case SG_GET_VERSION_NUM:
 789         *(int *)buf = 30000;
 790         break;
 791     case SG_GET_SCSI_ID:
 792         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
 793         break;
 794     case SG_IO:
 795         status = -EINPROGRESS;
 796         iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status);
 797
 798         while (status == -EINPROGRESS) {
 799             aio_poll(iscsilun->aio_context, true);
 800         }
 801
 802         return 0;
 803     default:
 804         return -1;
 805     }
 806     return 0;
 807 }
 808 #endif
 809
 810 static int64_t
 811 iscsi_getlength(BlockDriverState *bs)
 812 {
 813     IscsiLun *iscsilun = bs->opaque;
 814     int64_t len;
 815
 816     len  = iscsilun->num_blocks;
 817     len *= iscsilun->block_size;
 818
 819     return len;
 820 }
 821
 822 static int
 823 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num,
 824                                    int nb_sectors)
 825 {
 826     IscsiLun *iscsilun = bs->opaque;
 827     struct IscsiTask iTask;
 828     struct unmap_list list;
 829
 830     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 831         return -EINVAL;
 832     }
 833
 834     if (!iscsilun->lbp.lbpu) {
 835         /* UNMAP is not supported by the target */
 836         return 0;
 837     }
 838
 839     list.lba = sector_qemu2lun(sector_num, iscsilun);
 840     list.num = sector_qemu2lun(nb_sectors, iscsilun);
 841
 842     iscsi_co_init_iscsitask(iscsilun, &iTask);
 843 retry:
 844     if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
 845                      iscsi_co_generic_cb, &iTask) == NULL) {
 846         return -ENOMEM;
 847     }
 848
 849     while (!iTask.complete) {
 850         iscsi_set_events(iscsilun);
 851         qemu_coroutine_yield();
 852     }
 853
 854     if (iTask.task != NULL) {
 855         scsi_free_scsi_task(iTask.task);
 856         iTask.task = NULL;
 857     }
 858
 859     if (iTask.do_retry) {
 860         iTask.complete = 0;
 861         goto retry;
 862     }
 863
 864     if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
 865         /* the target might fail with a check condition if it
 866            is not happy with the alignment of the UNMAP request
 867            we silently fail in this case */
 868         return 0;
 869     }
 870
 871     if (iTask.status != SCSI_STATUS_GOOD) {
 872         return -EIO;
 873     }
 874
 875     iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors);
 876
 877     return 0;
 878 }
 879
 880 #if defined(SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED)
 881
 882 static int
 883 coroutine_fn iscsi_co_write_zeroes(BlockDriverState *bs, int64_t sector_num,
 884                                    int nb_sectors, BdrvRequestFlags flags)
 885 {
 886     IscsiLun *iscsilun = bs->opaque;
 887     struct IscsiTask iTask;
 888     uint64_t lba;
 889     uint32_t nb_blocks;
 890
 891     if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
 892         return -EINVAL;
 893     }
 894
 895     if ((flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->lbp.lbpws) {
 896         /* WRITE SAME with UNMAP is not supported by the target,
 897          * fall back and try WRITE SAME without UNMAP */
 898         flags &= ~BDRV_REQ_MAY_UNMAP;
 899     }
 900
 901     if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
 902         /* WRITE SAME without UNMAP is not supported by the target */
 903         return -ENOTSUP;
 904     }
 905
 906     lba = sector_qemu2lun(sector_num, iscsilun);
 907     nb_blocks = sector_qemu2lun(nb_sectors, iscsilun);
 908
 909     if (iscsilun->zeroblock == NULL) {
 910         iscsilun->zeroblock = g_malloc0(iscsilun->block_size);
 911     }
 912
 913     iscsi_co_init_iscsitask(iscsilun, &iTask);
 914 retry:
 915     if (iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
 916                                iscsilun->zeroblock, iscsilun->block_size,
 917                                nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
 918                                0, 0, iscsi_co_generic_cb, &iTask) == NULL) {
 919         return -ENOMEM;
 920     }
 921
 922     while (!iTask.complete) {
 923         iscsi_set_events(iscsilun);
 924         qemu_coroutine_yield();
 925     }
 926
 927     if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
 928         iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
 929         (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
 930          iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
 931         /* WRITE SAME is not supported by the target */
 932         iscsilun->has_write_same = false;
 933         scsi_free_scsi_task(iTask.task);
 934         return -ENOTSUP;
 935     }
 936
 937     if (iTask.task != NULL) {
 938         scsi_free_scsi_task(iTask.task);
 939         iTask.task = NULL;
 940     }
 941
 942     if (iTask.do_retry) {
 943         iTask.complete = 0;
 944         goto retry;
 945     }
 946
 947     if (iTask.status != SCSI_STATUS_GOOD) {
 948         return -EIO;
 949     }
 950
 951     if (flags & BDRV_REQ_MAY_UNMAP) {
 952         iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors);
 953     } else {
 954         iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors);
 955     }
 956
 957     return 0;
 958 }
 959
 960 #endif /* SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED */
 961
 962 static void parse_chap(struct iscsi_context *iscsi, const char *target,
 963                        Error **errp)
 964 {
 965     QemuOptsList *list;
 966     QemuOpts *opts;
 967     const char *user = NULL;
 968     const char *password = NULL;
 969
 970     list = qemu_find_opts("iscsi");
 971     if (!list) {
 972         return;
 973     }
 974
 975     opts = qemu_opts_find(list, target);
 976     if (opts == NULL) {
 977         opts = QTAILQ_FIRST(&list->head);
 978         if (!opts) {
 979             return;
 980         }
 981     }
 982
 983     user = qemu_opt_get(opts, "user");
 984     if (!user) {
 985         return;
 986     }
 987
 988     password = qemu_opt_get(opts, "password");
 989     if (!password) {
 990         error_setg(errp, "CHAP username specified but no password was given");
 991         return;
 992     }
 993
 994     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
 995         error_setg(errp, "Failed to set initiator username and password");
 996     }
 997 }
 998
 999 static void parse_header_digest(struct iscsi_context *iscsi, const char *target,
1000                                 Error **errp)
1001 {
1002     QemuOptsList *list;
1003     QemuOpts *opts;
1004     const char *digest = NULL;
1005
1006     list = qemu_find_opts("iscsi");
1007     if (!list) {
1008         return;
1009     }
1010
1011     opts = qemu_opts_find(list, target);
1012     if (opts == NULL) {
1013         opts = QTAILQ_FIRST(&list->head);
1014         if (!opts) {
1015             return;
1016         }
1017     }
1018
1019     digest = qemu_opt_get(opts, "header-digest");
1020     if (!digest) {
1021         return;
1022     }
1023
1024     if (!strcmp(digest, "CRC32C")) {
1025         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1026     } else if (!strcmp(digest, "NONE")) {
1027         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1028     } else if (!strcmp(digest, "CRC32C-NONE")) {
1029         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1030     } else if (!strcmp(digest, "NONE-CRC32C")) {
1031         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1032     } else {
1033         error_setg(errp, "Invalid header-digest setting : %s", digest);
1034     }
1035 }
1036
1037 static char *parse_initiator_name(const char *target)
1038 {
1039     QemuOptsList *list;
1040     QemuOpts *opts;
1041     const char *name;
1042     char *iscsi_name;
1043     UuidInfo *uuid_info;
1044
1045     list = qemu_find_opts("iscsi");
1046     if (list) {
1047         opts = qemu_opts_find(list, target);
1048         if (!opts) {
1049             opts = QTAILQ_FIRST(&list->head);
1050         }
1051         if (opts) {
1052             name = qemu_opt_get(opts, "initiator-name");
1053             if (name) {
1054                 return g_strdup(name);
1055             }
1056         }
1057     }
1058
1059     uuid_info = qmp_query_uuid(NULL);
1060     if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1061         name = qemu_get_vm_name();
1062     } else {
1063         name = uuid_info->UUID;
1064     }
1065     iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1066                                  name ? ":" : "", name ? name : "");
1067     qapi_free_UuidInfo(uuid_info);
1068     return iscsi_name;
1069 }
1070
1071 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
1072 static void iscsi_nop_timed_event(void *opaque)
1073 {
1074     IscsiLun *iscsilun = opaque;
1075
1076     if (iscsi_get_nops_in_flight(iscsilun->iscsi) > MAX_NOP_FAILURES) {
1077         error_report("iSCSI: NOP timeout. Reconnecting...");
1078         iscsi_reconnect(iscsilun->iscsi);
1079     }
1080
1081     if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1082         error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1083         return;
1084     }
1085
1086     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1087     iscsi_set_events(iscsilun);
1088 }
1089 #endif
1090
1091 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1092 {
1093     struct scsi_task *task = NULL;
1094     struct scsi_readcapacity10 *rc10 = NULL;
1095     struct scsi_readcapacity16 *rc16 = NULL;
1096     int retries = ISCSI_CMD_RETRIES;
1097
1098     do {
1099         if (task != NULL) {
1100             scsi_free_scsi_task(task);
1101             task = NULL;
1102         }
1103
1104         switch (iscsilun->type) {
1105         case TYPE_DISK:
1106             task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1107             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1108                 rc16 = scsi_datain_unmarshall(task);
1109                 if (rc16 == NULL) {
1110                     error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1111                 } else {
1112                     iscsilun->block_size = rc16->block_length;
1113                     iscsilun->num_blocks = rc16->returned_lba + 1;
1114                     iscsilun->lbpme = rc16->lbpme;
1115                     iscsilun->lbprz = rc16->lbprz;
1116                 }
1117             }
1118             break;
1119         case TYPE_ROM:
1120             task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1121             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1122                 rc10 = scsi_datain_unmarshall(task);
1123                 if (rc10 == NULL) {
1124                     error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1125                 } else {
1126                     iscsilun->block_size = rc10->block_size;
1127                     if (rc10->lba == 0) {
1128                         /* blank disk loaded */
1129                         iscsilun->num_blocks = 0;
1130                     } else {
1131                         iscsilun->num_blocks = rc10->lba + 1;
1132                     }
1133                 }
1134             }
1135             break;
1136         default:
1137             return;
1138         }
1139     } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1140              && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1141              && retries-- > 0);
1142
1143     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1144         error_setg(errp, "iSCSI: failed to send readcapacity10 command.");
1145     }
1146     if (task) {
1147         scsi_free_scsi_task(task);
1148     }
1149 }
1150
1151 /* TODO Convert to fine grained options */
1152 static QemuOptsList runtime_opts = {
1153     .name = "iscsi",
1154     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1155     .desc = {
1156         {
1157             .name = "filename",
1158             .type = QEMU_OPT_STRING,
1159             .help = "URL to the iscsi image",
1160         },
1161         { /* end of list */ }
1162     },
1163 };
1164
1165 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1166                                           int evpd, int pc, void **inq, Error **errp)
1167 {
1168     int full_size;
1169     struct scsi_task *task = NULL;
1170     task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1171     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1172         goto fail;
1173     }
1174     full_size = scsi_datain_getfullsize(task);
1175     if (full_size > task->datain.size) {
1176         scsi_free_scsi_task(task);
1177
1178         /* we need more data for the full list */
1179         task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1180         if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1181             goto fail;
1182         }
1183     }
1184
1185     *inq = scsi_datain_unmarshall(task);
1186     if (*inq == NULL) {
1187         error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1188         goto fail_with_err;
1189     }
1190
1191     return task;
1192
1193 fail:
1194     error_setg(errp, "iSCSI: Inquiry command failed : %s",
1195                iscsi_get_error(iscsi));
1196 fail_with_err:
1197     if (task != NULL) {
1198         scsi_free_scsi_task(task);
1199     }
1200     return NULL;
1201 }
1202
1203 static void iscsi_detach_aio_context(BlockDriverState *bs)
1204 {
1205     IscsiLun *iscsilun = bs->opaque;
1206
1207     aio_set_fd_handler(iscsilun->aio_context,
1208                        iscsi_get_fd(iscsilun->iscsi),
1209                        NULL, NULL, NULL);
1210     iscsilun->events = 0;
1211
1212     if (iscsilun->nop_timer) {
1213         timer_del(iscsilun->nop_timer);
1214         timer_free(iscsilun->nop_timer);
1215         iscsilun->nop_timer = NULL;
1216     }
1217 }
1218
1219 static void iscsi_attach_aio_context(BlockDriverState *bs,
1220                                      AioContext *new_context)
1221 {
1222     IscsiLun *iscsilun = bs->opaque;
1223
1224     iscsilun->aio_context = new_context;
1225     iscsi_set_events(iscsilun);
1226
1227 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
1228     /* Set up a timer for sending out iSCSI NOPs */
1229     iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1230                                         QEMU_CLOCK_REALTIME, SCALE_MS,
1231                                         iscsi_nop_timed_event, iscsilun);
1232     timer_mod(iscsilun->nop_timer,
1233               qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1234 #endif
1235 }
1236
1237 /*
1238  * We support iscsi url's on the form
1239  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1240  *
1241  * Note: flags are currently not used by iscsi_open.  If this function
1242  * is changed such that flags are used, please examine iscsi_reopen_prepare()
1243  * to see if needs to be changed as well.
1244  */
1245 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1246                       Error **errp)
1247 {
1248     IscsiLun *iscsilun = bs->opaque;
1249     struct iscsi_context *iscsi = NULL;
1250     struct iscsi_url *iscsi_url = NULL;
1251     struct scsi_task *task = NULL;
1252     struct scsi_inquiry_standard *inq = NULL;
1253     struct scsi_inquiry_supported_pages *inq_vpd;
1254     char *initiator_name = NULL;
1255     QemuOpts *opts;
1256     Error *local_err = NULL;
1257     const char *filename;
1258     int i, ret;
1259
1260     if ((BDRV_SECTOR_SIZE % 512) != 0) {
1261         error_setg(errp, "iSCSI: Invalid BDRV_SECTOR_SIZE. "
1262                    "BDRV_SECTOR_SIZE(%lld) is not a multiple "
1263                    "of 512", BDRV_SECTOR_SIZE);
1264         return -EINVAL;
1265     }
1266
1267     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1268     qemu_opts_absorb_qdict(opts, options, &local_err);
1269     if (local_err) {
1270         error_propagate(errp, local_err);
1271         ret = -EINVAL;
1272         goto out;
1273     }
1274
1275     filename = qemu_opt_get(opts, "filename");
1276
1277     iscsi_url = iscsi_parse_full_url(iscsi, filename);
1278     if (iscsi_url == NULL) {
1279         error_setg(errp, "Failed to parse URL : %s", filename);
1280         ret = -EINVAL;
1281         goto out;
1282     }
1283
1284     memset(iscsilun, 0, sizeof(IscsiLun));
1285
1286     initiator_name = parse_initiator_name(iscsi_url->target);
1287
1288     iscsi = iscsi_create_context(initiator_name);
1289     if (iscsi == NULL) {
1290         error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1291         ret = -ENOMEM;
1292         goto out;
1293     }
1294
1295     if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1296         error_setg(errp, "iSCSI: Failed to set target name.");
1297         ret = -EINVAL;
1298         goto out;
1299     }
1300
1301     if (iscsi_url->user != NULL) {
1302         ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1303                                               iscsi_url->passwd);
1304         if (ret != 0) {
1305             error_setg(errp, "Failed to set initiator username and password");
1306             ret = -EINVAL;
1307             goto out;
1308         }
1309     }
1310
1311     /* check if we got CHAP username/password via the options */
1312     parse_chap(iscsi, iscsi_url->target, &local_err);
1313     if (local_err != NULL) {
1314         error_propagate(errp, local_err);
1315         ret = -EINVAL;
1316         goto out;
1317     }
1318
1319     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1320         error_setg(errp, "iSCSI: Failed to set session type to normal.");
1321         ret = -EINVAL;
1322         goto out;
1323     }
1324
1325     iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1326
1327     /* check if we got HEADER_DIGEST via the options */
1328     parse_header_digest(iscsi, iscsi_url->target, &local_err);
1329     if (local_err != NULL) {
1330         error_propagate(errp, local_err);
1331         ret = -EINVAL;
1332         goto out;
1333     }
1334
1335     if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1336         error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1337             iscsi_get_error(iscsi));
1338         ret = -EINVAL;
1339         goto out;
1340     }
1341
1342     iscsilun->iscsi = iscsi;
1343     iscsilun->aio_context = bdrv_get_aio_context(bs);
1344     iscsilun->lun   = iscsi_url->lun;
1345     iscsilun->has_write_same = true;
1346
1347     task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1348                             (void **) &inq, errp);
1349     if (task == NULL) {
1350         ret = -EINVAL;
1351         goto out;
1352     }
1353     iscsilun->type = inq->periperal_device_type;
1354     scsi_free_scsi_task(task);
1355     task = NULL;
1356
1357     iscsi_readcapacity_sync(iscsilun, &local_err);
1358     if (local_err != NULL) {
1359         error_propagate(errp, local_err);
1360         ret = -EINVAL;
1361         goto out;
1362     }
1363     bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1364     bs->request_alignment = iscsilun->block_size;
1365
1366     /* We don't have any emulation for devices other than disks and CD-ROMs, so
1367      * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1368      * will try to read from the device to guess the image format.
1369      */
1370     if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1371         bs->sg = 1;
1372     }
1373
1374     task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1375                             SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1376                             (void **) &inq_vpd, errp);
1377     if (task == NULL) {
1378         ret = -EINVAL;
1379         goto out;
1380     }
1381     for (i = 0; i < inq_vpd->num_pages; i++) {
1382         struct scsi_task *inq_task;
1383         struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1384         struct scsi_inquiry_block_limits *inq_bl;
1385         switch (inq_vpd->pages[i]) {
1386         case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1387             inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1388                                         SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1389                                         (void **) &inq_lbp, errp);
1390             if (inq_task == NULL) {
1391                 ret = -EINVAL;
1392                 goto out;
1393             }
1394             memcpy(&iscsilun->lbp, inq_lbp,
1395                    sizeof(struct scsi_inquiry_logical_block_provisioning));
1396             scsi_free_scsi_task(inq_task);
1397             break;
1398         case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1399             inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1400                                     SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1401                                     (void **) &inq_bl, errp);
1402             if (inq_task == NULL) {
1403                 ret = -EINVAL;
1404                 goto out;
1405             }
1406             memcpy(&iscsilun->bl, inq_bl,
1407                    sizeof(struct scsi_inquiry_block_limits));
1408             scsi_free_scsi_task(inq_task);
1409             break;
1410         default:
1411             break;
1412         }
1413     }
1414     scsi_free_scsi_task(task);
1415     task = NULL;
1416
1417     iscsi_attach_aio_context(bs, iscsilun->aio_context);
1418
1419     /* Guess the internal cluster (page) size of the iscsi target by the means
1420      * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1421      * reasonable size */
1422     if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
1423         iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1424         iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran *
1425                                      iscsilun->block_size) >> BDRV_SECTOR_BITS;
1426 #if defined(LIBISCSI_FEATURE_IOVECTOR)
1427         if (iscsilun->lbprz && !(bs->open_flags & BDRV_O_NOCACHE)) {
1428             iscsilun->allocationmap =
1429                 bitmap_new(DIV_ROUND_UP(bs->total_sectors,
1430                                         iscsilun->cluster_sectors));
1431         }
1432 #endif
1433     }
1434
1435 out:
1436     qemu_opts_del(opts);
1437     if (initiator_name != NULL) {
1438         g_free(initiator_name);
1439     }
1440     if (iscsi_url != NULL) {
1441         iscsi_destroy_url(iscsi_url);
1442     }
1443     if (task != NULL) {
1444         scsi_free_scsi_task(task);
1445     }
1446
1447     if (ret) {
1448         if (iscsi != NULL) {
1449             iscsi_destroy_context(iscsi);
1450         }
1451         memset(iscsilun, 0, sizeof(IscsiLun));
1452     }
1453     return ret;
1454 }
1455
1456 static void iscsi_close(BlockDriverState *bs)
1457 {
1458     IscsiLun *iscsilun = bs->opaque;
1459     struct iscsi_context *iscsi = iscsilun->iscsi;
1460
1461     iscsi_detach_aio_context(bs);
1462     iscsi_destroy_context(iscsi);
1463     g_free(iscsilun->zeroblock);
1464     g_free(iscsilun->allocationmap);
1465     memset(iscsilun, 0, sizeof(IscsiLun));
1466 }
1467
1468 static int iscsi_refresh_limits(BlockDriverState *bs)
1469 {
1470     IscsiLun *iscsilun = bs->opaque;
1471
1472     /* We don't actually refresh here, but just return data queried in
1473      * iscsi_open(): iscsi targets don't change their limits. */
1474     if (iscsilun->lbp.lbpu) {
1475         if (iscsilun->bl.max_unmap < 0xffffffff) {
1476             bs->bl.max_discard = sector_lun2qemu(iscsilun->bl.max_unmap,
1477                                                  iscsilun);
1478         }
1479         bs->bl.discard_alignment = sector_lun2qemu(iscsilun->bl.opt_unmap_gran,
1480                                                    iscsilun);
1481     }
1482
1483     if (iscsilun->bl.max_ws_len < 0xffffffff) {
1484         bs->bl.max_write_zeroes = sector_lun2qemu(iscsilun->bl.max_ws_len,
1485                                                   iscsilun);
1486     }
1487     if (iscsilun->lbp.lbpws) {
1488         bs->bl.write_zeroes_alignment = sector_lun2qemu(iscsilun->bl.opt_unmap_gran,
1489                                                         iscsilun);
1490     }
1491     bs->bl.opt_transfer_length = sector_lun2qemu(iscsilun->bl.opt_xfer_len,
1492                                                  iscsilun);
1493     return 0;
1494 }
1495
1496 /* Since iscsi_open() ignores bdrv_flags, there is nothing to do here in
1497  * prepare.  Note that this will not re-establish a connection with an iSCSI
1498  * target - it is effectively a NOP.  */
1499 static int iscsi_reopen_prepare(BDRVReopenState *state,
1500                                 BlockReopenQueue *queue, Error **errp)
1501 {
1502     /* NOP */
1503     return 0;
1504 }
1505
1506 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1507 {
1508     IscsiLun *iscsilun = bs->opaque;
1509     Error *local_err = NULL;
1510
1511     if (iscsilun->type != TYPE_DISK) {
1512         return -ENOTSUP;
1513     }
1514
1515     iscsi_readcapacity_sync(iscsilun, &local_err);
1516     if (local_err != NULL) {
1517         error_free(local_err);
1518         return -EIO;
1519     }
1520
1521     if (offset > iscsi_getlength(bs)) {
1522         return -EINVAL;
1523     }
1524
1525     if (iscsilun->allocationmap != NULL) {
1526         g_free(iscsilun->allocationmap);
1527         iscsilun->allocationmap =
1528             bitmap_new(DIV_ROUND_UP(bs->total_sectors,
1529                                     iscsilun->cluster_sectors));
1530     }
1531
1532     return 0;
1533 }
1534
1535 static int iscsi_create(const char *filename, QEMUOptionParameter *options,
1536                         Error **errp)
1537 {
1538     int ret = 0;
1539     int64_t total_size = 0;
1540     BlockDriverState *bs;
1541     IscsiLun *iscsilun = NULL;
1542     QDict *bs_options;
1543
1544     bs = bdrv_new("", &error_abort);
1545
1546     /* Read out options */
1547     while (options && options->name) {
1548         if (!strcmp(options->name, "size")) {
1549             total_size = options->value.n / BDRV_SECTOR_SIZE;
1550         }
1551         options++;
1552     }
1553
1554     bs->opaque = g_malloc0(sizeof(struct IscsiLun));
1555     iscsilun = bs->opaque;
1556
1557     bs_options = qdict_new();
1558     qdict_put(bs_options, "filename", qstring_from_str(filename));
1559     ret = iscsi_open(bs, bs_options, 0, NULL);
1560     QDECREF(bs_options);
1561
1562     if (ret != 0) {
1563         goto out;
1564     }
1565     iscsi_detach_aio_context(bs);
1566     if (iscsilun->type != TYPE_DISK) {
1567         ret = -ENODEV;
1568         goto out;
1569     }
1570     if (bs->total_sectors < total_size) {
1571         ret = -ENOSPC;
1572         goto out;
1573     }
1574
1575     ret = 0;
1576 out:
1577     if (iscsilun->iscsi != NULL) {
1578         iscsi_destroy_context(iscsilun->iscsi);
1579     }
1580     g_free(bs->opaque);
1581     bs->opaque = NULL;
1582     bdrv_unref(bs);
1583     return ret;
1584 }
1585
1586 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1587 {
1588     IscsiLun *iscsilun = bs->opaque;
1589     bdi->unallocated_blocks_are_zero = !!iscsilun->lbprz;
1590     bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
1591     bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE;
1592     return 0;
1593 }
1594
1595 static QEMUOptionParameter iscsi_create_options[] = {
1596     {
1597         .name = BLOCK_OPT_SIZE,
1598         .type = OPT_SIZE,
1599         .help = "Virtual disk size"
1600     },
1601     { NULL }
1602 };
1603
1604 static BlockDriver bdrv_iscsi = {
1605     .format_name     = "iscsi",
1606     .protocol_name   = "iscsi",
1607
1608     .instance_size   = sizeof(IscsiLun),
1609     .bdrv_needs_filename = true,
1610     .bdrv_file_open  = iscsi_open,
1611     .bdrv_close      = iscsi_close,
1612     .bdrv_create     = iscsi_create,
1613     .create_options  = iscsi_create_options,
1614     .bdrv_reopen_prepare  = iscsi_reopen_prepare,
1615
1616     .bdrv_getlength  = iscsi_getlength,
1617     .bdrv_get_info   = iscsi_get_info,
1618     .bdrv_truncate   = iscsi_truncate,
1619     .bdrv_refresh_limits = iscsi_refresh_limits,
1620
1621 #if defined(LIBISCSI_FEATURE_IOVECTOR)
1622     .bdrv_co_get_block_status = iscsi_co_get_block_status,
1623 #endif
1624     .bdrv_co_discard      = iscsi_co_discard,
1625 #if defined(SCSI_SENSE_ASCQ_CAPACITY_DATA_HAS_CHANGED)
1626     .bdrv_co_write_zeroes = iscsi_co_write_zeroes,
1627 #endif
1628     .bdrv_co_readv         = iscsi_co_readv,
1629     .bdrv_co_writev        = iscsi_co_writev,
1630     .bdrv_co_flush_to_disk = iscsi_co_flush,
1631
1632 #ifdef __linux__
1633     .bdrv_ioctl       = iscsi_ioctl,
1634     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
1635 #endif
1636
1637     .bdrv_detach_aio_context = iscsi_detach_aio_context,
1638     .bdrv_attach_aio_context = iscsi_attach_aio_context,
1639 };
1640
1641 static QemuOptsList qemu_iscsi_opts = {
1642     .name = "iscsi",
1643     .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1644     .desc = {
1645         {
1646             .name = "user",
1647             .type = QEMU_OPT_STRING,
1648             .help = "username for CHAP authentication to target",
1649         },{
1650             .name = "password",
1651             .type = QEMU_OPT_STRING,
1652             .help = "password for CHAP authentication to target",
1653         },{
1654             .name = "header-digest",
1655             .type = QEMU_OPT_STRING,
1656             .help = "HeaderDigest setting. "
1657                     "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1658         },{
1659             .name = "initiator-name",
1660             .type = QEMU_OPT_STRING,
1661             .help = "Initiator iqn name to use when connecting",
1662         },
1663         { /* end of list */ }
1664     },
1665 };
1666
1667 static void iscsi_block_init(void)
1668 {
1669     bdrv_register(&bdrv_iscsi);
1670     qemu_add_opts(&qemu_iscsi_opts);
1671 }
1672
1673 block_init(iscsi_block_init);