1 /* 2 * QEMU Block driver for iSCSI images 3 * 4 * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com> 5 * Copyright (c) 2012-2015 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 <math.h> 30 #include <arpa/inet.h> 31 #include "qemu-common.h" 32 #include "qemu/config-file.h" 33 #include "qemu/error-report.h" 34 #include "qemu/bitops.h" 35 #include "qemu/bitmap.h" 36 #include "block/block_int.h" 37 #include "block/scsi.h" 38 #include "qemu/iov.h" 39 #include "sysemu/sysemu.h" 40 #include "qmp-commands.h" 41 #include "qapi/qmp/qstring.h" 42 43 #include <iscsi/iscsi.h> 44 #include <iscsi/scsi-lowlevel.h> 45 46 #ifdef __linux__ 47 #include <scsi/sg.h> 48 #include <block/scsi.h> 49 #endif 50 51 typedef struct IscsiLun { 52 struct iscsi_context *iscsi; 53 AioContext *aio_context; 54 int lun; 55 enum scsi_inquiry_peripheral_device_type type; 56 int block_size; 57 uint64_t num_blocks; 58 int events; 59 QEMUTimer *nop_timer; 60 QEMUTimer *event_timer; 61 struct scsi_inquiry_logical_block_provisioning lbp; 62 struct scsi_inquiry_block_limits bl; 63 unsigned char *zeroblock; 64 unsigned long *allocationmap; 65 int cluster_sectors; 66 bool use_16_for_rw; 67 bool write_protected; 68 bool lbpme; 69 bool lbprz; 70 bool dpofua; 71 bool has_write_same; 72 bool force_next_flush; 73 bool request_timed_out; 74 } IscsiLun; 75 76 typedef struct IscsiTask { 77 int status; 78 int complete; 79 int retries; 80 int do_retry; 81 struct scsi_task *task; 82 Coroutine *co; 83 QEMUBH *bh; 84 IscsiLun *iscsilun; 85 QEMUTimer retry_timer; 86 bool force_next_flush; 87 } IscsiTask; 88 89 typedef struct IscsiAIOCB { 90 BlockAIOCB common; 91 QEMUIOVector *qiov; 92 QEMUBH *bh; 93 IscsiLun *iscsilun; 94 struct scsi_task *task; 95 uint8_t *buf; 96 int status; 97 int64_t sector_num; 98 int nb_sectors; 99 #ifdef __linux__ 100 sg_io_hdr_t *ioh; 101 #endif 102 } IscsiAIOCB; 103 104 /* libiscsi uses time_t so its enough to process events every second */ 105 #define EVENT_INTERVAL 1000 106 #define NOP_INTERVAL 5000 107 #define MAX_NOP_FAILURES 3 108 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times) 109 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768}; 110 111 /* this threshold is a trade-off knob to choose between 112 * the potential additional overhead of an extra GET_LBA_STATUS request 113 * vs. unnecessarily reading a lot of zero sectors over the wire. 114 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES 115 * sectors we check the allocation status of the area covered by the 116 * request first if the allocationmap indicates that the area might be 117 * unallocated. */ 118 #define ISCSI_CHECKALLOC_THRES 64 119 120 static void 121 iscsi_bh_cb(void *p) 122 { 123 IscsiAIOCB *acb = p; 124 125 qemu_bh_delete(acb->bh); 126 127 g_free(acb->buf); 128 acb->buf = NULL; 129 130 acb->common.cb(acb->common.opaque, acb->status); 131 132 if (acb->task != NULL) { 133 scsi_free_scsi_task(acb->task); 134 acb->task = NULL; 135 } 136 137 qemu_aio_unref(acb); 138 } 139 140 static void 141 iscsi_schedule_bh(IscsiAIOCB *acb) 142 { 143 if (acb->bh) { 144 return; 145 } 146 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb); 147 qemu_bh_schedule(acb->bh); 148 } 149 150 static void iscsi_co_generic_bh_cb(void *opaque) 151 { 152 struct IscsiTask *iTask = opaque; 153 iTask->complete = 1; 154 qemu_bh_delete(iTask->bh); 155 qemu_coroutine_enter(iTask->co, NULL); 156 } 157 158 static void iscsi_retry_timer_expired(void *opaque) 159 { 160 struct IscsiTask *iTask = opaque; 161 iTask->complete = 1; 162 if (iTask->co) { 163 qemu_coroutine_enter(iTask->co, NULL); 164 } 165 } 166 167 static inline unsigned exp_random(double mean) 168 { 169 return -mean * log((double)rand() / RAND_MAX); 170 } 171 172 /* SCSI_STATUS_TASK_SET_FULL and SCSI_STATUS_TIMEOUT were introduced 173 * in libiscsi 1.10.0 as part of an enum. The LIBISCSI_API_VERSION 174 * macro was introduced in 1.11.0. So use the API_VERSION macro as 175 * a hint that the macros are defined and define them ourselves 176 * otherwise to keep the required libiscsi version at 1.9.0 */ 177 #if !defined(LIBISCSI_API_VERSION) 178 #define QEMU_SCSI_STATUS_TASK_SET_FULL 0x28 179 #define QEMU_SCSI_STATUS_TIMEOUT 0x0f000002 180 #else 181 #define QEMU_SCSI_STATUS_TASK_SET_FULL SCSI_STATUS_TASK_SET_FULL 182 #define QEMU_SCSI_STATUS_TIMEOUT SCSI_STATUS_TIMEOUT 183 #endif 184 185 static void 186 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status, 187 void *command_data, void *opaque) 188 { 189 struct IscsiTask *iTask = opaque; 190 struct scsi_task *task = command_data; 191 192 iTask->status = status; 193 iTask->do_retry = 0; 194 iTask->task = task; 195 196 if (status != SCSI_STATUS_GOOD) { 197 if (iTask->retries++ < ISCSI_CMD_RETRIES) { 198 if (status == SCSI_STATUS_CHECK_CONDITION 199 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) { 200 error_report("iSCSI CheckCondition: %s", 201 iscsi_get_error(iscsi)); 202 iTask->do_retry = 1; 203 goto out; 204 } 205 if (status == SCSI_STATUS_BUSY || 206 status == QEMU_SCSI_STATUS_TIMEOUT || 207 status == QEMU_SCSI_STATUS_TASK_SET_FULL) { 208 unsigned retry_time = 209 exp_random(iscsi_retry_times[iTask->retries - 1]); 210 if (status == QEMU_SCSI_STATUS_TIMEOUT) { 211 /* make sure the request is rescheduled AFTER the 212 * reconnect is initiated */ 213 retry_time = EVENT_INTERVAL * 2; 214 iTask->iscsilun->request_timed_out = true; 215 } 216 error_report("iSCSI Busy/TaskSetFull/TimeOut" 217 " (retry #%u in %u ms): %s", 218 iTask->retries, retry_time, 219 iscsi_get_error(iscsi)); 220 aio_timer_init(iTask->iscsilun->aio_context, 221 &iTask->retry_timer, QEMU_CLOCK_REALTIME, 222 SCALE_MS, iscsi_retry_timer_expired, iTask); 223 timer_mod(&iTask->retry_timer, 224 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time); 225 iTask->do_retry = 1; 226 return; 227 } 228 } 229 error_report("iSCSI Failure: %s", iscsi_get_error(iscsi)); 230 } else { 231 iTask->iscsilun->force_next_flush |= iTask->force_next_flush; 232 } 233 234 out: 235 if (iTask->co) { 236 iTask->bh = aio_bh_new(iTask->iscsilun->aio_context, 237 iscsi_co_generic_bh_cb, iTask); 238 qemu_bh_schedule(iTask->bh); 239 } else { 240 iTask->complete = 1; 241 } 242 } 243 244 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask) 245 { 246 *iTask = (struct IscsiTask) { 247 .co = qemu_coroutine_self(), 248 .iscsilun = iscsilun, 249 }; 250 } 251 252 static void 253 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data, 254 void *private_data) 255 { 256 IscsiAIOCB *acb = private_data; 257 258 acb->status = -ECANCELED; 259 iscsi_schedule_bh(acb); 260 } 261 262 static void 263 iscsi_aio_cancel(BlockAIOCB *blockacb) 264 { 265 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb; 266 IscsiLun *iscsilun = acb->iscsilun; 267 268 if (acb->status != -EINPROGRESS) { 269 return; 270 } 271 272 /* send a task mgmt call to the target to cancel the task on the target */ 273 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task, 274 iscsi_abort_task_cb, acb); 275 276 } 277 278 static const AIOCBInfo iscsi_aiocb_info = { 279 .aiocb_size = sizeof(IscsiAIOCB), 280 .cancel_async = iscsi_aio_cancel, 281 }; 282 283 284 static void iscsi_process_read(void *arg); 285 static void iscsi_process_write(void *arg); 286 287 static void 288 iscsi_set_events(IscsiLun *iscsilun) 289 { 290 struct iscsi_context *iscsi = iscsilun->iscsi; 291 int ev = iscsi_which_events(iscsi); 292 293 if (ev != iscsilun->events) { 294 aio_set_fd_handler(iscsilun->aio_context, 295 iscsi_get_fd(iscsi), 296 (ev & POLLIN) ? iscsi_process_read : NULL, 297 (ev & POLLOUT) ? iscsi_process_write : NULL, 298 iscsilun); 299 iscsilun->events = ev; 300 } 301 } 302 303 static void iscsi_timed_check_events(void *opaque) 304 { 305 IscsiLun *iscsilun = opaque; 306 307 /* check for timed out requests */ 308 iscsi_service(iscsilun->iscsi, 0); 309 310 if (iscsilun->request_timed_out) { 311 iscsilun->request_timed_out = false; 312 iscsi_reconnect(iscsilun->iscsi); 313 } 314 315 /* newer versions of libiscsi may return zero events. Ensure we are able 316 * to return to service once this situation changes. */ 317 iscsi_set_events(iscsilun); 318 319 timer_mod(iscsilun->event_timer, 320 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL); 321 } 322 323 static void 324 iscsi_process_read(void *arg) 325 { 326 IscsiLun *iscsilun = arg; 327 struct iscsi_context *iscsi = iscsilun->iscsi; 328 329 iscsi_service(iscsi, POLLIN); 330 iscsi_set_events(iscsilun); 331 } 332 333 static void 334 iscsi_process_write(void *arg) 335 { 336 IscsiLun *iscsilun = arg; 337 struct iscsi_context *iscsi = iscsilun->iscsi; 338 339 iscsi_service(iscsi, POLLOUT); 340 iscsi_set_events(iscsilun); 341 } 342 343 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun) 344 { 345 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE; 346 } 347 348 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun) 349 { 350 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size; 351 } 352 353 static bool is_request_lun_aligned(int64_t sector_num, int nb_sectors, 354 IscsiLun *iscsilun) 355 { 356 if ((sector_num * BDRV_SECTOR_SIZE) % iscsilun->block_size || 357 (nb_sectors * BDRV_SECTOR_SIZE) % iscsilun->block_size) { 358 error_report("iSCSI misaligned request: " 359 "iscsilun->block_size %u, sector_num %" PRIi64 360 ", nb_sectors %d", 361 iscsilun->block_size, sector_num, nb_sectors); 362 return 0; 363 } 364 return 1; 365 } 366 367 static unsigned long *iscsi_allocationmap_init(IscsiLun *iscsilun) 368 { 369 return bitmap_try_new(DIV_ROUND_UP(sector_lun2qemu(iscsilun->num_blocks, 370 iscsilun), 371 iscsilun->cluster_sectors)); 372 } 373 374 static void iscsi_allocationmap_set(IscsiLun *iscsilun, int64_t sector_num, 375 int nb_sectors) 376 { 377 if (iscsilun->allocationmap == NULL) { 378 return; 379 } 380 bitmap_set(iscsilun->allocationmap, 381 sector_num / iscsilun->cluster_sectors, 382 DIV_ROUND_UP(nb_sectors, iscsilun->cluster_sectors)); 383 } 384 385 static void iscsi_allocationmap_clear(IscsiLun *iscsilun, int64_t sector_num, 386 int nb_sectors) 387 { 388 int64_t cluster_num, nb_clusters; 389 if (iscsilun->allocationmap == NULL) { 390 return; 391 } 392 cluster_num = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors); 393 nb_clusters = (sector_num + nb_sectors) / iscsilun->cluster_sectors 394 - cluster_num; 395 if (nb_clusters > 0) { 396 bitmap_clear(iscsilun->allocationmap, cluster_num, nb_clusters); 397 } 398 } 399 400 static int coroutine_fn iscsi_co_writev(BlockDriverState *bs, 401 int64_t sector_num, int nb_sectors, 402 QEMUIOVector *iov) 403 { 404 IscsiLun *iscsilun = bs->opaque; 405 struct IscsiTask iTask; 406 uint64_t lba; 407 uint32_t num_sectors; 408 int fua; 409 410 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 411 return -EINVAL; 412 } 413 414 if (bs->bl.max_transfer_length && nb_sectors > bs->bl.max_transfer_length) { 415 error_report("iSCSI Error: Write of %d sectors exceeds max_xfer_len " 416 "of %d sectors", nb_sectors, bs->bl.max_transfer_length); 417 return -EINVAL; 418 } 419 420 lba = sector_qemu2lun(sector_num, iscsilun); 421 num_sectors = sector_qemu2lun(nb_sectors, iscsilun); 422 iscsi_co_init_iscsitask(iscsilun, &iTask); 423 retry: 424 fua = iscsilun->dpofua && !bs->enable_write_cache; 425 iTask.force_next_flush = !fua; 426 if (iscsilun->use_16_for_rw) { 427 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba, 428 NULL, num_sectors * iscsilun->block_size, 429 iscsilun->block_size, 0, 0, fua, 0, 0, 430 iscsi_co_generic_cb, &iTask); 431 } else { 432 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba, 433 NULL, num_sectors * iscsilun->block_size, 434 iscsilun->block_size, 0, 0, fua, 0, 0, 435 iscsi_co_generic_cb, &iTask); 436 } 437 if (iTask.task == NULL) { 438 return -ENOMEM; 439 } 440 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov, 441 iov->niov); 442 while (!iTask.complete) { 443 iscsi_set_events(iscsilun); 444 qemu_coroutine_yield(); 445 } 446 447 if (iTask.task != NULL) { 448 scsi_free_scsi_task(iTask.task); 449 iTask.task = NULL; 450 } 451 452 if (iTask.do_retry) { 453 iTask.complete = 0; 454 goto retry; 455 } 456 457 if (iTask.status != SCSI_STATUS_GOOD) { 458 return -EIO; 459 } 460 461 iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors); 462 463 return 0; 464 } 465 466 467 static bool iscsi_allocationmap_is_allocated(IscsiLun *iscsilun, 468 int64_t sector_num, int nb_sectors) 469 { 470 unsigned long size; 471 if (iscsilun->allocationmap == NULL) { 472 return true; 473 } 474 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors); 475 return !(find_next_bit(iscsilun->allocationmap, size, 476 sector_num / iscsilun->cluster_sectors) == size); 477 } 478 479 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs, 480 int64_t sector_num, 481 int nb_sectors, int *pnum) 482 { 483 IscsiLun *iscsilun = bs->opaque; 484 struct scsi_get_lba_status *lbas = NULL; 485 struct scsi_lba_status_descriptor *lbasd = NULL; 486 struct IscsiTask iTask; 487 int64_t ret; 488 489 iscsi_co_init_iscsitask(iscsilun, &iTask); 490 491 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 492 ret = -EINVAL; 493 goto out; 494 } 495 496 /* default to all sectors allocated */ 497 ret = BDRV_BLOCK_DATA; 498 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID; 499 *pnum = nb_sectors; 500 501 /* LUN does not support logical block provisioning */ 502 if (!iscsilun->lbpme) { 503 goto out; 504 } 505 506 retry: 507 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun, 508 sector_qemu2lun(sector_num, iscsilun), 509 8 + 16, iscsi_co_generic_cb, 510 &iTask) == NULL) { 511 ret = -ENOMEM; 512 goto out; 513 } 514 515 while (!iTask.complete) { 516 iscsi_set_events(iscsilun); 517 qemu_coroutine_yield(); 518 } 519 520 if (iTask.do_retry) { 521 if (iTask.task != NULL) { 522 scsi_free_scsi_task(iTask.task); 523 iTask.task = NULL; 524 } 525 iTask.complete = 0; 526 goto retry; 527 } 528 529 if (iTask.status != SCSI_STATUS_GOOD) { 530 /* in case the get_lba_status_callout fails (i.e. 531 * because the device is busy or the cmd is not 532 * supported) we pretend all blocks are allocated 533 * for backwards compatibility */ 534 goto out; 535 } 536 537 lbas = scsi_datain_unmarshall(iTask.task); 538 if (lbas == NULL) { 539 ret = -EIO; 540 goto out; 541 } 542 543 lbasd = &lbas->descriptors[0]; 544 545 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) { 546 ret = -EIO; 547 goto out; 548 } 549 550 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun); 551 552 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED || 553 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) { 554 ret &= ~BDRV_BLOCK_DATA; 555 if (iscsilun->lbprz) { 556 ret |= BDRV_BLOCK_ZERO; 557 } 558 } 559 560 if (ret & BDRV_BLOCK_ZERO) { 561 iscsi_allocationmap_clear(iscsilun, sector_num, *pnum); 562 } else { 563 iscsi_allocationmap_set(iscsilun, sector_num, *pnum); 564 } 565 566 if (*pnum > nb_sectors) { 567 *pnum = nb_sectors; 568 } 569 out: 570 if (iTask.task != NULL) { 571 scsi_free_scsi_task(iTask.task); 572 } 573 return ret; 574 } 575 576 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs, 577 int64_t sector_num, int nb_sectors, 578 QEMUIOVector *iov) 579 { 580 IscsiLun *iscsilun = bs->opaque; 581 struct IscsiTask iTask; 582 uint64_t lba; 583 uint32_t num_sectors; 584 585 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 586 return -EINVAL; 587 } 588 589 if (bs->bl.max_transfer_length && nb_sectors > bs->bl.max_transfer_length) { 590 error_report("iSCSI Error: Read of %d sectors exceeds max_xfer_len " 591 "of %d sectors", nb_sectors, bs->bl.max_transfer_length); 592 return -EINVAL; 593 } 594 595 if (iscsilun->lbprz && nb_sectors >= ISCSI_CHECKALLOC_THRES && 596 !iscsi_allocationmap_is_allocated(iscsilun, sector_num, nb_sectors)) { 597 int64_t ret; 598 int pnum; 599 ret = iscsi_co_get_block_status(bs, sector_num, INT_MAX, &pnum); 600 if (ret < 0) { 601 return ret; 602 } 603 if (ret & BDRV_BLOCK_ZERO && pnum >= nb_sectors) { 604 qemu_iovec_memset(iov, 0, 0x00, iov->size); 605 return 0; 606 } 607 } 608 609 lba = sector_qemu2lun(sector_num, iscsilun); 610 num_sectors = sector_qemu2lun(nb_sectors, iscsilun); 611 612 iscsi_co_init_iscsitask(iscsilun, &iTask); 613 retry: 614 if (iscsilun->use_16_for_rw) { 615 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba, 616 num_sectors * iscsilun->block_size, 617 iscsilun->block_size, 0, 0, 0, 0, 0, 618 iscsi_co_generic_cb, &iTask); 619 } else { 620 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba, 621 num_sectors * iscsilun->block_size, 622 iscsilun->block_size, 623 0, 0, 0, 0, 0, 624 iscsi_co_generic_cb, &iTask); 625 } 626 if (iTask.task == NULL) { 627 return -ENOMEM; 628 } 629 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov); 630 631 while (!iTask.complete) { 632 iscsi_set_events(iscsilun); 633 qemu_coroutine_yield(); 634 } 635 636 if (iTask.task != NULL) { 637 scsi_free_scsi_task(iTask.task); 638 iTask.task = NULL; 639 } 640 641 if (iTask.do_retry) { 642 iTask.complete = 0; 643 goto retry; 644 } 645 646 if (iTask.status != SCSI_STATUS_GOOD) { 647 return -EIO; 648 } 649 650 return 0; 651 } 652 653 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs) 654 { 655 IscsiLun *iscsilun = bs->opaque; 656 struct IscsiTask iTask; 657 658 if (!iscsilun->force_next_flush) { 659 return 0; 660 } 661 iscsilun->force_next_flush = false; 662 663 iscsi_co_init_iscsitask(iscsilun, &iTask); 664 retry: 665 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0, 666 0, iscsi_co_generic_cb, &iTask) == NULL) { 667 return -ENOMEM; 668 } 669 670 while (!iTask.complete) { 671 iscsi_set_events(iscsilun); 672 qemu_coroutine_yield(); 673 } 674 675 if (iTask.task != NULL) { 676 scsi_free_scsi_task(iTask.task); 677 iTask.task = NULL; 678 } 679 680 if (iTask.do_retry) { 681 iTask.complete = 0; 682 goto retry; 683 } 684 685 if (iTask.status != SCSI_STATUS_GOOD) { 686 return -EIO; 687 } 688 689 return 0; 690 } 691 692 #ifdef __linux__ 693 static void 694 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status, 695 void *command_data, void *opaque) 696 { 697 IscsiAIOCB *acb = opaque; 698 699 g_free(acb->buf); 700 acb->buf = NULL; 701 702 acb->status = 0; 703 if (status < 0) { 704 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s", 705 iscsi_get_error(iscsi)); 706 acb->status = -EIO; 707 } 708 709 acb->ioh->driver_status = 0; 710 acb->ioh->host_status = 0; 711 acb->ioh->resid = 0; 712 713 #define SG_ERR_DRIVER_SENSE 0x08 714 715 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) { 716 int ss; 717 718 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE; 719 720 acb->ioh->sb_len_wr = acb->task->datain.size - 2; 721 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ? 722 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr; 723 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss); 724 } 725 726 iscsi_schedule_bh(acb); 727 } 728 729 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs, 730 unsigned long int req, void *buf, 731 BlockCompletionFunc *cb, void *opaque) 732 { 733 IscsiLun *iscsilun = bs->opaque; 734 struct iscsi_context *iscsi = iscsilun->iscsi; 735 struct iscsi_data data; 736 IscsiAIOCB *acb; 737 738 assert(req == SG_IO); 739 740 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque); 741 742 acb->iscsilun = iscsilun; 743 acb->bh = NULL; 744 acb->status = -EINPROGRESS; 745 acb->buf = NULL; 746 acb->ioh = buf; 747 748 acb->task = malloc(sizeof(struct scsi_task)); 749 if (acb->task == NULL) { 750 error_report("iSCSI: Failed to allocate task for scsi command. %s", 751 iscsi_get_error(iscsi)); 752 qemu_aio_unref(acb); 753 return NULL; 754 } 755 memset(acb->task, 0, sizeof(struct scsi_task)); 756 757 switch (acb->ioh->dxfer_direction) { 758 case SG_DXFER_TO_DEV: 759 acb->task->xfer_dir = SCSI_XFER_WRITE; 760 break; 761 case SG_DXFER_FROM_DEV: 762 acb->task->xfer_dir = SCSI_XFER_READ; 763 break; 764 default: 765 acb->task->xfer_dir = SCSI_XFER_NONE; 766 break; 767 } 768 769 acb->task->cdb_size = acb->ioh->cmd_len; 770 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len); 771 acb->task->expxferlen = acb->ioh->dxfer_len; 772 773 data.size = 0; 774 if (acb->task->xfer_dir == SCSI_XFER_WRITE) { 775 if (acb->ioh->iovec_count == 0) { 776 data.data = acb->ioh->dxferp; 777 data.size = acb->ioh->dxfer_len; 778 } else { 779 scsi_task_set_iov_out(acb->task, 780 (struct scsi_iovec *) acb->ioh->dxferp, 781 acb->ioh->iovec_count); 782 } 783 } 784 785 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task, 786 iscsi_aio_ioctl_cb, 787 (data.size > 0) ? &data : NULL, 788 acb) != 0) { 789 scsi_free_scsi_task(acb->task); 790 qemu_aio_unref(acb); 791 return NULL; 792 } 793 794 /* tell libiscsi to read straight into the buffer we got from ioctl */ 795 if (acb->task->xfer_dir == SCSI_XFER_READ) { 796 if (acb->ioh->iovec_count == 0) { 797 scsi_task_add_data_in_buffer(acb->task, 798 acb->ioh->dxfer_len, 799 acb->ioh->dxferp); 800 } else { 801 scsi_task_set_iov_in(acb->task, 802 (struct scsi_iovec *) acb->ioh->dxferp, 803 acb->ioh->iovec_count); 804 } 805 } 806 807 iscsi_set_events(iscsilun); 808 809 return &acb->common; 810 } 811 812 static void ioctl_cb(void *opaque, int status) 813 { 814 int *p_status = opaque; 815 *p_status = status; 816 } 817 818 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) 819 { 820 IscsiLun *iscsilun = bs->opaque; 821 int status; 822 823 switch (req) { 824 case SG_GET_VERSION_NUM: 825 *(int *)buf = 30000; 826 break; 827 case SG_GET_SCSI_ID: 828 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type; 829 break; 830 case SG_IO: 831 status = -EINPROGRESS; 832 iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status); 833 834 while (status == -EINPROGRESS) { 835 aio_poll(iscsilun->aio_context, true); 836 } 837 838 return 0; 839 default: 840 return -1; 841 } 842 return 0; 843 } 844 #endif 845 846 static int64_t 847 iscsi_getlength(BlockDriverState *bs) 848 { 849 IscsiLun *iscsilun = bs->opaque; 850 int64_t len; 851 852 len = iscsilun->num_blocks; 853 len *= iscsilun->block_size; 854 855 return len; 856 } 857 858 static int 859 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num, 860 int nb_sectors) 861 { 862 IscsiLun *iscsilun = bs->opaque; 863 struct IscsiTask iTask; 864 struct unmap_list list; 865 866 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 867 return -EINVAL; 868 } 869 870 if (!iscsilun->lbp.lbpu) { 871 /* UNMAP is not supported by the target */ 872 return 0; 873 } 874 875 list.lba = sector_qemu2lun(sector_num, iscsilun); 876 list.num = sector_qemu2lun(nb_sectors, iscsilun); 877 878 iscsi_co_init_iscsitask(iscsilun, &iTask); 879 retry: 880 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1, 881 iscsi_co_generic_cb, &iTask) == NULL) { 882 return -ENOMEM; 883 } 884 885 while (!iTask.complete) { 886 iscsi_set_events(iscsilun); 887 qemu_coroutine_yield(); 888 } 889 890 if (iTask.task != NULL) { 891 scsi_free_scsi_task(iTask.task); 892 iTask.task = NULL; 893 } 894 895 if (iTask.do_retry) { 896 iTask.complete = 0; 897 goto retry; 898 } 899 900 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) { 901 /* the target might fail with a check condition if it 902 is not happy with the alignment of the UNMAP request 903 we silently fail in this case */ 904 return 0; 905 } 906 907 if (iTask.status != SCSI_STATUS_GOOD) { 908 return -EIO; 909 } 910 911 iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors); 912 913 return 0; 914 } 915 916 static int 917 coroutine_fn iscsi_co_write_zeroes(BlockDriverState *bs, int64_t sector_num, 918 int nb_sectors, BdrvRequestFlags flags) 919 { 920 IscsiLun *iscsilun = bs->opaque; 921 struct IscsiTask iTask; 922 uint64_t lba; 923 uint32_t nb_blocks; 924 bool use_16_for_ws = iscsilun->use_16_for_rw; 925 926 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 927 return -EINVAL; 928 } 929 930 if (flags & BDRV_REQ_MAY_UNMAP) { 931 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) { 932 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */ 933 use_16_for_ws = true; 934 } 935 if (use_16_for_ws && !iscsilun->lbp.lbpws) { 936 /* WRITESAME16 with UNMAP is not supported by the target, 937 * fall back and try WRITESAME10/16 without UNMAP */ 938 flags &= ~BDRV_REQ_MAY_UNMAP; 939 use_16_for_ws = iscsilun->use_16_for_rw; 940 } 941 } 942 943 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) { 944 /* WRITESAME without UNMAP is not supported by the target */ 945 return -ENOTSUP; 946 } 947 948 lba = sector_qemu2lun(sector_num, iscsilun); 949 nb_blocks = sector_qemu2lun(nb_sectors, iscsilun); 950 951 if (iscsilun->zeroblock == NULL) { 952 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size); 953 if (iscsilun->zeroblock == NULL) { 954 return -ENOMEM; 955 } 956 } 957 958 iscsi_co_init_iscsitask(iscsilun, &iTask); 959 iTask.force_next_flush = true; 960 retry: 961 if (use_16_for_ws) { 962 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba, 963 iscsilun->zeroblock, iscsilun->block_size, 964 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 965 0, 0, iscsi_co_generic_cb, &iTask); 966 } else { 967 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba, 968 iscsilun->zeroblock, iscsilun->block_size, 969 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 970 0, 0, iscsi_co_generic_cb, &iTask); 971 } 972 if (iTask.task == NULL) { 973 return -ENOMEM; 974 } 975 976 while (!iTask.complete) { 977 iscsi_set_events(iscsilun); 978 qemu_coroutine_yield(); 979 } 980 981 if (iTask.status == SCSI_STATUS_CHECK_CONDITION && 982 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST && 983 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE || 984 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) { 985 /* WRITE SAME is not supported by the target */ 986 iscsilun->has_write_same = false; 987 scsi_free_scsi_task(iTask.task); 988 return -ENOTSUP; 989 } 990 991 if (iTask.task != NULL) { 992 scsi_free_scsi_task(iTask.task); 993 iTask.task = NULL; 994 } 995 996 if (iTask.do_retry) { 997 iTask.complete = 0; 998 goto retry; 999 } 1000 1001 if (iTask.status != SCSI_STATUS_GOOD) { 1002 return -EIO; 1003 } 1004 1005 if (flags & BDRV_REQ_MAY_UNMAP) { 1006 iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors); 1007 } else { 1008 iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors); 1009 } 1010 1011 return 0; 1012 } 1013 1014 static void parse_chap(struct iscsi_context *iscsi, const char *target, 1015 Error **errp) 1016 { 1017 QemuOptsList *list; 1018 QemuOpts *opts; 1019 const char *user = NULL; 1020 const char *password = NULL; 1021 1022 list = qemu_find_opts("iscsi"); 1023 if (!list) { 1024 return; 1025 } 1026 1027 opts = qemu_opts_find(list, target); 1028 if (opts == NULL) { 1029 opts = QTAILQ_FIRST(&list->head); 1030 if (!opts) { 1031 return; 1032 } 1033 } 1034 1035 user = qemu_opt_get(opts, "user"); 1036 if (!user) { 1037 return; 1038 } 1039 1040 password = qemu_opt_get(opts, "password"); 1041 if (!password) { 1042 error_setg(errp, "CHAP username specified but no password was given"); 1043 return; 1044 } 1045 1046 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) { 1047 error_setg(errp, "Failed to set initiator username and password"); 1048 } 1049 } 1050 1051 static void parse_header_digest(struct iscsi_context *iscsi, const char *target, 1052 Error **errp) 1053 { 1054 QemuOptsList *list; 1055 QemuOpts *opts; 1056 const char *digest = NULL; 1057 1058 list = qemu_find_opts("iscsi"); 1059 if (!list) { 1060 return; 1061 } 1062 1063 opts = qemu_opts_find(list, target); 1064 if (opts == NULL) { 1065 opts = QTAILQ_FIRST(&list->head); 1066 if (!opts) { 1067 return; 1068 } 1069 } 1070 1071 digest = qemu_opt_get(opts, "header-digest"); 1072 if (!digest) { 1073 return; 1074 } 1075 1076 if (!strcmp(digest, "CRC32C")) { 1077 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C); 1078 } else if (!strcmp(digest, "NONE")) { 1079 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE); 1080 } else if (!strcmp(digest, "CRC32C-NONE")) { 1081 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE); 1082 } else if (!strcmp(digest, "NONE-CRC32C")) { 1083 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1084 } else { 1085 error_setg(errp, "Invalid header-digest setting : %s", digest); 1086 } 1087 } 1088 1089 static char *parse_initiator_name(const char *target) 1090 { 1091 QemuOptsList *list; 1092 QemuOpts *opts; 1093 const char *name; 1094 char *iscsi_name; 1095 UuidInfo *uuid_info; 1096 1097 list = qemu_find_opts("iscsi"); 1098 if (list) { 1099 opts = qemu_opts_find(list, target); 1100 if (!opts) { 1101 opts = QTAILQ_FIRST(&list->head); 1102 } 1103 if (opts) { 1104 name = qemu_opt_get(opts, "initiator-name"); 1105 if (name) { 1106 return g_strdup(name); 1107 } 1108 } 1109 } 1110 1111 uuid_info = qmp_query_uuid(NULL); 1112 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) { 1113 name = qemu_get_vm_name(); 1114 } else { 1115 name = uuid_info->UUID; 1116 } 1117 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s", 1118 name ? ":" : "", name ? name : ""); 1119 qapi_free_UuidInfo(uuid_info); 1120 return iscsi_name; 1121 } 1122 1123 static int parse_timeout(const char *target) 1124 { 1125 QemuOptsList *list; 1126 QemuOpts *opts; 1127 const char *timeout; 1128 1129 list = qemu_find_opts("iscsi"); 1130 if (list) { 1131 opts = qemu_opts_find(list, target); 1132 if (!opts) { 1133 opts = QTAILQ_FIRST(&list->head); 1134 } 1135 if (opts) { 1136 timeout = qemu_opt_get(opts, "timeout"); 1137 if (timeout) { 1138 return atoi(timeout); 1139 } 1140 } 1141 } 1142 1143 return 0; 1144 } 1145 1146 static void iscsi_nop_timed_event(void *opaque) 1147 { 1148 IscsiLun *iscsilun = opaque; 1149 1150 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) { 1151 error_report("iSCSI: NOP timeout. Reconnecting..."); 1152 iscsilun->request_timed_out = true; 1153 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) { 1154 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages."); 1155 return; 1156 } 1157 1158 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1159 iscsi_set_events(iscsilun); 1160 } 1161 1162 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp) 1163 { 1164 struct scsi_task *task = NULL; 1165 struct scsi_readcapacity10 *rc10 = NULL; 1166 struct scsi_readcapacity16 *rc16 = NULL; 1167 int retries = ISCSI_CMD_RETRIES; 1168 1169 do { 1170 if (task != NULL) { 1171 scsi_free_scsi_task(task); 1172 task = NULL; 1173 } 1174 1175 switch (iscsilun->type) { 1176 case TYPE_DISK: 1177 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun); 1178 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1179 rc16 = scsi_datain_unmarshall(task); 1180 if (rc16 == NULL) { 1181 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data."); 1182 } else { 1183 iscsilun->block_size = rc16->block_length; 1184 iscsilun->num_blocks = rc16->returned_lba + 1; 1185 iscsilun->lbpme = !!rc16->lbpme; 1186 iscsilun->lbprz = !!rc16->lbprz; 1187 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff); 1188 } 1189 } 1190 break; 1191 case TYPE_ROM: 1192 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0); 1193 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1194 rc10 = scsi_datain_unmarshall(task); 1195 if (rc10 == NULL) { 1196 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data."); 1197 } else { 1198 iscsilun->block_size = rc10->block_size; 1199 if (rc10->lba == 0) { 1200 /* blank disk loaded */ 1201 iscsilun->num_blocks = 0; 1202 } else { 1203 iscsilun->num_blocks = rc10->lba + 1; 1204 } 1205 } 1206 } 1207 break; 1208 default: 1209 return; 1210 } 1211 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION 1212 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION 1213 && retries-- > 0); 1214 1215 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1216 error_setg(errp, "iSCSI: failed to send readcapacity10 command."); 1217 } 1218 if (task) { 1219 scsi_free_scsi_task(task); 1220 } 1221 } 1222 1223 /* TODO Convert to fine grained options */ 1224 static QemuOptsList runtime_opts = { 1225 .name = "iscsi", 1226 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), 1227 .desc = { 1228 { 1229 .name = "filename", 1230 .type = QEMU_OPT_STRING, 1231 .help = "URL to the iscsi image", 1232 }, 1233 { /* end of list */ } 1234 }, 1235 }; 1236 1237 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun, 1238 int evpd, int pc, void **inq, Error **errp) 1239 { 1240 int full_size; 1241 struct scsi_task *task = NULL; 1242 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64); 1243 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1244 goto fail; 1245 } 1246 full_size = scsi_datain_getfullsize(task); 1247 if (full_size > task->datain.size) { 1248 scsi_free_scsi_task(task); 1249 1250 /* we need more data for the full list */ 1251 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size); 1252 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1253 goto fail; 1254 } 1255 } 1256 1257 *inq = scsi_datain_unmarshall(task); 1258 if (*inq == NULL) { 1259 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob"); 1260 goto fail_with_err; 1261 } 1262 1263 return task; 1264 1265 fail: 1266 error_setg(errp, "iSCSI: Inquiry command failed : %s", 1267 iscsi_get_error(iscsi)); 1268 fail_with_err: 1269 if (task != NULL) { 1270 scsi_free_scsi_task(task); 1271 } 1272 return NULL; 1273 } 1274 1275 static void iscsi_detach_aio_context(BlockDriverState *bs) 1276 { 1277 IscsiLun *iscsilun = bs->opaque; 1278 1279 aio_set_fd_handler(iscsilun->aio_context, 1280 iscsi_get_fd(iscsilun->iscsi), 1281 NULL, NULL, NULL); 1282 iscsilun->events = 0; 1283 1284 if (iscsilun->nop_timer) { 1285 timer_del(iscsilun->nop_timer); 1286 timer_free(iscsilun->nop_timer); 1287 iscsilun->nop_timer = NULL; 1288 } 1289 if (iscsilun->event_timer) { 1290 timer_del(iscsilun->event_timer); 1291 timer_free(iscsilun->event_timer); 1292 iscsilun->event_timer = NULL; 1293 } 1294 } 1295 1296 static void iscsi_attach_aio_context(BlockDriverState *bs, 1297 AioContext *new_context) 1298 { 1299 IscsiLun *iscsilun = bs->opaque; 1300 1301 iscsilun->aio_context = new_context; 1302 iscsi_set_events(iscsilun); 1303 1304 /* Set up a timer for sending out iSCSI NOPs */ 1305 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context, 1306 QEMU_CLOCK_REALTIME, SCALE_MS, 1307 iscsi_nop_timed_event, iscsilun); 1308 timer_mod(iscsilun->nop_timer, 1309 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1310 1311 /* Set up a timer for periodic calls to iscsi_set_events and to 1312 * scan for command timeout */ 1313 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context, 1314 QEMU_CLOCK_REALTIME, SCALE_MS, 1315 iscsi_timed_check_events, iscsilun); 1316 timer_mod(iscsilun->event_timer, 1317 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL); 1318 } 1319 1320 static void iscsi_modesense_sync(IscsiLun *iscsilun) 1321 { 1322 struct scsi_task *task; 1323 struct scsi_mode_sense *ms = NULL; 1324 iscsilun->write_protected = false; 1325 iscsilun->dpofua = false; 1326 1327 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun, 1328 1, SCSI_MODESENSE_PC_CURRENT, 1329 0x3F, 0, 255); 1330 if (task == NULL) { 1331 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s", 1332 iscsi_get_error(iscsilun->iscsi)); 1333 goto out; 1334 } 1335 1336 if (task->status != SCSI_STATUS_GOOD) { 1337 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable"); 1338 goto out; 1339 } 1340 ms = scsi_datain_unmarshall(task); 1341 if (!ms) { 1342 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s", 1343 iscsi_get_error(iscsilun->iscsi)); 1344 goto out; 1345 } 1346 iscsilun->write_protected = ms->device_specific_parameter & 0x80; 1347 iscsilun->dpofua = ms->device_specific_parameter & 0x10; 1348 1349 out: 1350 if (task) { 1351 scsi_free_scsi_task(task); 1352 } 1353 } 1354 1355 /* 1356 * We support iscsi url's on the form 1357 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun> 1358 */ 1359 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags, 1360 Error **errp) 1361 { 1362 IscsiLun *iscsilun = bs->opaque; 1363 struct iscsi_context *iscsi = NULL; 1364 struct iscsi_url *iscsi_url = NULL; 1365 struct scsi_task *task = NULL; 1366 struct scsi_inquiry_standard *inq = NULL; 1367 struct scsi_inquiry_supported_pages *inq_vpd; 1368 char *initiator_name = NULL; 1369 QemuOpts *opts; 1370 Error *local_err = NULL; 1371 const char *filename; 1372 int i, ret = 0, timeout = 0; 1373 1374 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); 1375 qemu_opts_absorb_qdict(opts, options, &local_err); 1376 if (local_err) { 1377 error_propagate(errp, local_err); 1378 ret = -EINVAL; 1379 goto out; 1380 } 1381 1382 filename = qemu_opt_get(opts, "filename"); 1383 1384 iscsi_url = iscsi_parse_full_url(iscsi, filename); 1385 if (iscsi_url == NULL) { 1386 error_setg(errp, "Failed to parse URL : %s", filename); 1387 ret = -EINVAL; 1388 goto out; 1389 } 1390 1391 memset(iscsilun, 0, sizeof(IscsiLun)); 1392 1393 initiator_name = parse_initiator_name(iscsi_url->target); 1394 1395 iscsi = iscsi_create_context(initiator_name); 1396 if (iscsi == NULL) { 1397 error_setg(errp, "iSCSI: Failed to create iSCSI context."); 1398 ret = -ENOMEM; 1399 goto out; 1400 } 1401 1402 if (iscsi_set_targetname(iscsi, iscsi_url->target)) { 1403 error_setg(errp, "iSCSI: Failed to set target name."); 1404 ret = -EINVAL; 1405 goto out; 1406 } 1407 1408 if (iscsi_url->user[0] != '\0') { 1409 ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user, 1410 iscsi_url->passwd); 1411 if (ret != 0) { 1412 error_setg(errp, "Failed to set initiator username and password"); 1413 ret = -EINVAL; 1414 goto out; 1415 } 1416 } 1417 1418 /* check if we got CHAP username/password via the options */ 1419 parse_chap(iscsi, iscsi_url->target, &local_err); 1420 if (local_err != NULL) { 1421 error_propagate(errp, local_err); 1422 ret = -EINVAL; 1423 goto out; 1424 } 1425 1426 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) { 1427 error_setg(errp, "iSCSI: Failed to set session type to normal."); 1428 ret = -EINVAL; 1429 goto out; 1430 } 1431 1432 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1433 1434 /* check if we got HEADER_DIGEST via the options */ 1435 parse_header_digest(iscsi, iscsi_url->target, &local_err); 1436 if (local_err != NULL) { 1437 error_propagate(errp, local_err); 1438 ret = -EINVAL; 1439 goto out; 1440 } 1441 1442 /* timeout handling is broken in libiscsi before 1.15.0 */ 1443 timeout = parse_timeout(iscsi_url->target); 1444 #if defined(LIBISCSI_API_VERSION) && LIBISCSI_API_VERSION >= 20150621 1445 iscsi_set_timeout(iscsi, timeout); 1446 #else 1447 if (timeout) { 1448 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0"); 1449 } 1450 #endif 1451 1452 if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) { 1453 error_setg(errp, "iSCSI: Failed to connect to LUN : %s", 1454 iscsi_get_error(iscsi)); 1455 ret = -EINVAL; 1456 goto out; 1457 } 1458 1459 iscsilun->iscsi = iscsi; 1460 iscsilun->aio_context = bdrv_get_aio_context(bs); 1461 iscsilun->lun = iscsi_url->lun; 1462 iscsilun->has_write_same = true; 1463 1464 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0, 1465 (void **) &inq, errp); 1466 if (task == NULL) { 1467 ret = -EINVAL; 1468 goto out; 1469 } 1470 iscsilun->type = inq->periperal_device_type; 1471 scsi_free_scsi_task(task); 1472 task = NULL; 1473 1474 iscsi_modesense_sync(iscsilun); 1475 1476 /* Check the write protect flag of the LUN if we want to write */ 1477 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) && 1478 iscsilun->write_protected) { 1479 error_setg(errp, "Cannot open a write protected LUN as read-write"); 1480 ret = -EACCES; 1481 goto out; 1482 } 1483 1484 iscsi_readcapacity_sync(iscsilun, &local_err); 1485 if (local_err != NULL) { 1486 error_propagate(errp, local_err); 1487 ret = -EINVAL; 1488 goto out; 1489 } 1490 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun); 1491 bs->request_alignment = iscsilun->block_size; 1492 1493 /* We don't have any emulation for devices other than disks and CD-ROMs, so 1494 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu 1495 * will try to read from the device to guess the image format. 1496 */ 1497 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) { 1498 bs->sg = 1; 1499 } 1500 1501 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1502 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES, 1503 (void **) &inq_vpd, errp); 1504 if (task == NULL) { 1505 ret = -EINVAL; 1506 goto out; 1507 } 1508 for (i = 0; i < inq_vpd->num_pages; i++) { 1509 struct scsi_task *inq_task; 1510 struct scsi_inquiry_logical_block_provisioning *inq_lbp; 1511 struct scsi_inquiry_block_limits *inq_bl; 1512 switch (inq_vpd->pages[i]) { 1513 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING: 1514 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1515 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING, 1516 (void **) &inq_lbp, errp); 1517 if (inq_task == NULL) { 1518 ret = -EINVAL; 1519 goto out; 1520 } 1521 memcpy(&iscsilun->lbp, inq_lbp, 1522 sizeof(struct scsi_inquiry_logical_block_provisioning)); 1523 scsi_free_scsi_task(inq_task); 1524 break; 1525 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS: 1526 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1527 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS, 1528 (void **) &inq_bl, errp); 1529 if (inq_task == NULL) { 1530 ret = -EINVAL; 1531 goto out; 1532 } 1533 memcpy(&iscsilun->bl, inq_bl, 1534 sizeof(struct scsi_inquiry_block_limits)); 1535 scsi_free_scsi_task(inq_task); 1536 break; 1537 default: 1538 break; 1539 } 1540 } 1541 scsi_free_scsi_task(task); 1542 task = NULL; 1543 1544 iscsi_attach_aio_context(bs, iscsilun->aio_context); 1545 1546 /* Guess the internal cluster (page) size of the iscsi target by the means 1547 * of opt_unmap_gran. Transfer the unmap granularity only if it has a 1548 * reasonable size */ 1549 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 && 1550 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) { 1551 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran * 1552 iscsilun->block_size) >> BDRV_SECTOR_BITS; 1553 if (iscsilun->lbprz) { 1554 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun); 1555 if (iscsilun->allocationmap == NULL) { 1556 ret = -ENOMEM; 1557 } 1558 } 1559 } 1560 1561 out: 1562 qemu_opts_del(opts); 1563 g_free(initiator_name); 1564 if (iscsi_url != NULL) { 1565 iscsi_destroy_url(iscsi_url); 1566 } 1567 if (task != NULL) { 1568 scsi_free_scsi_task(task); 1569 } 1570 1571 if (ret) { 1572 if (iscsi != NULL) { 1573 if (iscsi_is_logged_in(iscsi)) { 1574 iscsi_logout_sync(iscsi); 1575 } 1576 iscsi_destroy_context(iscsi); 1577 } 1578 memset(iscsilun, 0, sizeof(IscsiLun)); 1579 } 1580 return ret; 1581 } 1582 1583 static void iscsi_close(BlockDriverState *bs) 1584 { 1585 IscsiLun *iscsilun = bs->opaque; 1586 struct iscsi_context *iscsi = iscsilun->iscsi; 1587 1588 iscsi_detach_aio_context(bs); 1589 if (iscsi_is_logged_in(iscsi)) { 1590 iscsi_logout_sync(iscsi); 1591 } 1592 iscsi_destroy_context(iscsi); 1593 g_free(iscsilun->zeroblock); 1594 g_free(iscsilun->allocationmap); 1595 memset(iscsilun, 0, sizeof(IscsiLun)); 1596 } 1597 1598 static int sector_limits_lun2qemu(int64_t sector, IscsiLun *iscsilun) 1599 { 1600 return MIN(sector_lun2qemu(sector, iscsilun), INT_MAX / 2 + 1); 1601 } 1602 1603 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp) 1604 { 1605 /* We don't actually refresh here, but just return data queried in 1606 * iscsi_open(): iscsi targets don't change their limits. */ 1607 1608 IscsiLun *iscsilun = bs->opaque; 1609 uint32_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff; 1610 1611 if (iscsilun->bl.max_xfer_len) { 1612 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len); 1613 } 1614 1615 bs->bl.max_transfer_length = sector_limits_lun2qemu(max_xfer_len, iscsilun); 1616 1617 if (iscsilun->lbp.lbpu) { 1618 if (iscsilun->bl.max_unmap < 0xffffffff) { 1619 bs->bl.max_discard = 1620 sector_limits_lun2qemu(iscsilun->bl.max_unmap, iscsilun); 1621 } 1622 bs->bl.discard_alignment = 1623 sector_limits_lun2qemu(iscsilun->bl.opt_unmap_gran, iscsilun); 1624 } 1625 1626 if (iscsilun->bl.max_ws_len < 0xffffffff) { 1627 bs->bl.max_write_zeroes = 1628 sector_limits_lun2qemu(iscsilun->bl.max_ws_len, iscsilun); 1629 } 1630 if (iscsilun->lbp.lbpws) { 1631 bs->bl.write_zeroes_alignment = 1632 sector_limits_lun2qemu(iscsilun->bl.opt_unmap_gran, iscsilun); 1633 } 1634 bs->bl.opt_transfer_length = 1635 sector_limits_lun2qemu(iscsilun->bl.opt_xfer_len, iscsilun); 1636 } 1637 1638 /* Note that this will not re-establish a connection with an iSCSI target - it 1639 * is effectively a NOP. */ 1640 static int iscsi_reopen_prepare(BDRVReopenState *state, 1641 BlockReopenQueue *queue, Error **errp) 1642 { 1643 IscsiLun *iscsilun = state->bs->opaque; 1644 1645 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) { 1646 error_setg(errp, "Cannot open a write protected LUN as read-write"); 1647 return -EACCES; 1648 } 1649 return 0; 1650 } 1651 1652 static int iscsi_truncate(BlockDriverState *bs, int64_t offset) 1653 { 1654 IscsiLun *iscsilun = bs->opaque; 1655 Error *local_err = NULL; 1656 1657 if (iscsilun->type != TYPE_DISK) { 1658 return -ENOTSUP; 1659 } 1660 1661 iscsi_readcapacity_sync(iscsilun, &local_err); 1662 if (local_err != NULL) { 1663 error_free(local_err); 1664 return -EIO; 1665 } 1666 1667 if (offset > iscsi_getlength(bs)) { 1668 return -EINVAL; 1669 } 1670 1671 if (iscsilun->allocationmap != NULL) { 1672 g_free(iscsilun->allocationmap); 1673 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun); 1674 } 1675 1676 return 0; 1677 } 1678 1679 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp) 1680 { 1681 int ret = 0; 1682 int64_t total_size = 0; 1683 BlockDriverState *bs; 1684 IscsiLun *iscsilun = NULL; 1685 QDict *bs_options; 1686 1687 bs = bdrv_new(); 1688 1689 /* Read out options */ 1690 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 1691 BDRV_SECTOR_SIZE); 1692 bs->opaque = g_new0(struct IscsiLun, 1); 1693 iscsilun = bs->opaque; 1694 1695 bs_options = qdict_new(); 1696 qdict_put(bs_options, "filename", qstring_from_str(filename)); 1697 ret = iscsi_open(bs, bs_options, 0, NULL); 1698 QDECREF(bs_options); 1699 1700 if (ret != 0) { 1701 goto out; 1702 } 1703 iscsi_detach_aio_context(bs); 1704 if (iscsilun->type != TYPE_DISK) { 1705 ret = -ENODEV; 1706 goto out; 1707 } 1708 if (bs->total_sectors < total_size) { 1709 ret = -ENOSPC; 1710 goto out; 1711 } 1712 1713 ret = 0; 1714 out: 1715 if (iscsilun->iscsi != NULL) { 1716 iscsi_destroy_context(iscsilun->iscsi); 1717 } 1718 g_free(bs->opaque); 1719 bs->opaque = NULL; 1720 bdrv_unref(bs); 1721 return ret; 1722 } 1723 1724 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 1725 { 1726 IscsiLun *iscsilun = bs->opaque; 1727 bdi->unallocated_blocks_are_zero = iscsilun->lbprz; 1728 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws; 1729 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE; 1730 return 0; 1731 } 1732 1733 static QemuOptsList iscsi_create_opts = { 1734 .name = "iscsi-create-opts", 1735 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head), 1736 .desc = { 1737 { 1738 .name = BLOCK_OPT_SIZE, 1739 .type = QEMU_OPT_SIZE, 1740 .help = "Virtual disk size" 1741 }, 1742 { /* end of list */ } 1743 } 1744 }; 1745 1746 static BlockDriver bdrv_iscsi = { 1747 .format_name = "iscsi", 1748 .protocol_name = "iscsi", 1749 1750 .instance_size = sizeof(IscsiLun), 1751 .bdrv_needs_filename = true, 1752 .bdrv_file_open = iscsi_open, 1753 .bdrv_close = iscsi_close, 1754 .bdrv_create = iscsi_create, 1755 .create_opts = &iscsi_create_opts, 1756 .bdrv_reopen_prepare = iscsi_reopen_prepare, 1757 1758 .bdrv_getlength = iscsi_getlength, 1759 .bdrv_get_info = iscsi_get_info, 1760 .bdrv_truncate = iscsi_truncate, 1761 .bdrv_refresh_limits = iscsi_refresh_limits, 1762 1763 .bdrv_co_get_block_status = iscsi_co_get_block_status, 1764 .bdrv_co_discard = iscsi_co_discard, 1765 .bdrv_co_write_zeroes = iscsi_co_write_zeroes, 1766 .bdrv_co_readv = iscsi_co_readv, 1767 .bdrv_co_writev = iscsi_co_writev, 1768 .bdrv_co_flush_to_disk = iscsi_co_flush, 1769 1770 #ifdef __linux__ 1771 .bdrv_ioctl = iscsi_ioctl, 1772 .bdrv_aio_ioctl = iscsi_aio_ioctl, 1773 #endif 1774 1775 .bdrv_detach_aio_context = iscsi_detach_aio_context, 1776 .bdrv_attach_aio_context = iscsi_attach_aio_context, 1777 }; 1778 1779 static QemuOptsList qemu_iscsi_opts = { 1780 .name = "iscsi", 1781 .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head), 1782 .desc = { 1783 { 1784 .name = "user", 1785 .type = QEMU_OPT_STRING, 1786 .help = "username for CHAP authentication to target", 1787 },{ 1788 .name = "password", 1789 .type = QEMU_OPT_STRING, 1790 .help = "password for CHAP authentication to target", 1791 },{ 1792 .name = "header-digest", 1793 .type = QEMU_OPT_STRING, 1794 .help = "HeaderDigest setting. " 1795 "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}", 1796 },{ 1797 .name = "initiator-name", 1798 .type = QEMU_OPT_STRING, 1799 .help = "Initiator iqn name to use when connecting", 1800 },{ 1801 .name = "timeout", 1802 .type = QEMU_OPT_NUMBER, 1803 .help = "Request timeout in seconds (default 0 = no timeout)", 1804 }, 1805 { /* end of list */ } 1806 }, 1807 }; 1808 1809 static void iscsi_block_init(void) 1810 { 1811 bdrv_register(&bdrv_iscsi); 1812 qemu_add_opts(&qemu_iscsi_opts); 1813 } 1814 1815 block_init(iscsi_block_init); 1816