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 "qemu/osdep.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 #include "crypto/secret.h" 43 44 #include <iscsi/iscsi.h> 45 #include <iscsi/scsi-lowlevel.h> 46 47 #ifdef __linux__ 48 #include <scsi/sg.h> 49 #include <block/scsi.h> 50 #endif 51 52 typedef struct IscsiLun { 53 struct iscsi_context *iscsi; 54 AioContext *aio_context; 55 int lun; 56 enum scsi_inquiry_peripheral_device_type type; 57 int block_size; 58 uint64_t num_blocks; 59 int events; 60 QEMUTimer *nop_timer; 61 QEMUTimer *event_timer; 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 bool use_16_for_rw; 68 bool write_protected; 69 bool lbpme; 70 bool lbprz; 71 bool dpofua; 72 bool has_write_same; 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 int err_code; 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 int ret; 100 #ifdef __linux__ 101 sg_io_hdr_t *ioh; 102 #endif 103 } IscsiAIOCB; 104 105 /* libiscsi uses time_t so its enough to process events every second */ 106 #define EVENT_INTERVAL 1000 107 #define NOP_INTERVAL 5000 108 #define MAX_NOP_FAILURES 3 109 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times) 110 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768}; 111 112 /* this threshold is a trade-off knob to choose between 113 * the potential additional overhead of an extra GET_LBA_STATUS request 114 * vs. unnecessarily reading a lot of zero sectors over the wire. 115 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES 116 * sectors we check the allocation status of the area covered by the 117 * request first if the allocationmap indicates that the area might be 118 * unallocated. */ 119 #define ISCSI_CHECKALLOC_THRES 64 120 121 static void 122 iscsi_bh_cb(void *p) 123 { 124 IscsiAIOCB *acb = p; 125 126 qemu_bh_delete(acb->bh); 127 128 g_free(acb->buf); 129 acb->buf = NULL; 130 131 acb->common.cb(acb->common.opaque, acb->status); 132 133 if (acb->task != NULL) { 134 scsi_free_scsi_task(acb->task); 135 acb->task = NULL; 136 } 137 138 qemu_aio_unref(acb); 139 } 140 141 static void 142 iscsi_schedule_bh(IscsiAIOCB *acb) 143 { 144 if (acb->bh) { 145 return; 146 } 147 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb); 148 qemu_bh_schedule(acb->bh); 149 } 150 151 static void iscsi_co_generic_bh_cb(void *opaque) 152 { 153 struct IscsiTask *iTask = opaque; 154 iTask->complete = 1; 155 qemu_bh_delete(iTask->bh); 156 qemu_coroutine_enter(iTask->co, NULL); 157 } 158 159 static void iscsi_retry_timer_expired(void *opaque) 160 { 161 struct IscsiTask *iTask = opaque; 162 iTask->complete = 1; 163 if (iTask->co) { 164 qemu_coroutine_enter(iTask->co, NULL); 165 } 166 } 167 168 static inline unsigned exp_random(double mean) 169 { 170 return -mean * log((double)rand() / RAND_MAX); 171 } 172 173 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in 174 * libiscsi 1.10.0, together with other constants we need. Use it as 175 * a hint that we have to define them ourselves if needed, to keep the 176 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for 177 * the test because SCSI_STATUS_* is an enum. 178 * 179 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes 180 * an enum, check against the LIBISCSI_API_VERSION macro, which was 181 * introduced in 1.11.0. If it is present, there is no need to define 182 * anything. 183 */ 184 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \ 185 !defined(LIBISCSI_API_VERSION) 186 #define SCSI_STATUS_TASK_SET_FULL 0x28 187 #define SCSI_STATUS_TIMEOUT 0x0f000002 188 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600 189 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00 190 #endif 191 192 static int iscsi_translate_sense(struct scsi_sense *sense) 193 { 194 int ret; 195 196 switch (sense->key) { 197 case SCSI_SENSE_NOT_READY: 198 return -EBUSY; 199 case SCSI_SENSE_DATA_PROTECTION: 200 return -EACCES; 201 case SCSI_SENSE_COMMAND_ABORTED: 202 return -ECANCELED; 203 case SCSI_SENSE_ILLEGAL_REQUEST: 204 /* Parse ASCQ */ 205 break; 206 default: 207 return -EIO; 208 } 209 switch (sense->ascq) { 210 case SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR: 211 case SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE: 212 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB: 213 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST: 214 ret = -EINVAL; 215 break; 216 case SCSI_SENSE_ASCQ_LBA_OUT_OF_RANGE: 217 ret = -ENOSPC; 218 break; 219 case SCSI_SENSE_ASCQ_LOGICAL_UNIT_NOT_SUPPORTED: 220 ret = -ENOTSUP; 221 break; 222 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT: 223 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_CLOSED: 224 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_OPEN: 225 ret = -ENOMEDIUM; 226 break; 227 case SCSI_SENSE_ASCQ_WRITE_PROTECTED: 228 ret = -EACCES; 229 break; 230 default: 231 ret = -EIO; 232 break; 233 } 234 return ret; 235 } 236 237 static void 238 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status, 239 void *command_data, void *opaque) 240 { 241 struct IscsiTask *iTask = opaque; 242 struct scsi_task *task = command_data; 243 244 iTask->status = status; 245 iTask->do_retry = 0; 246 iTask->task = task; 247 248 if (status != SCSI_STATUS_GOOD) { 249 if (iTask->retries++ < ISCSI_CMD_RETRIES) { 250 if (status == SCSI_STATUS_CHECK_CONDITION 251 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) { 252 error_report("iSCSI CheckCondition: %s", 253 iscsi_get_error(iscsi)); 254 iTask->do_retry = 1; 255 goto out; 256 } 257 if (status == SCSI_STATUS_BUSY || 258 status == SCSI_STATUS_TIMEOUT || 259 status == SCSI_STATUS_TASK_SET_FULL) { 260 unsigned retry_time = 261 exp_random(iscsi_retry_times[iTask->retries - 1]); 262 if (status == SCSI_STATUS_TIMEOUT) { 263 /* make sure the request is rescheduled AFTER the 264 * reconnect is initiated */ 265 retry_time = EVENT_INTERVAL * 2; 266 iTask->iscsilun->request_timed_out = true; 267 } 268 error_report("iSCSI Busy/TaskSetFull/TimeOut" 269 " (retry #%u in %u ms): %s", 270 iTask->retries, retry_time, 271 iscsi_get_error(iscsi)); 272 aio_timer_init(iTask->iscsilun->aio_context, 273 &iTask->retry_timer, QEMU_CLOCK_REALTIME, 274 SCALE_MS, iscsi_retry_timer_expired, iTask); 275 timer_mod(&iTask->retry_timer, 276 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time); 277 iTask->do_retry = 1; 278 return; 279 } 280 } 281 iTask->err_code = iscsi_translate_sense(&task->sense); 282 error_report("iSCSI Failure: %s", iscsi_get_error(iscsi)); 283 } 284 285 out: 286 if (iTask->co) { 287 iTask->bh = aio_bh_new(iTask->iscsilun->aio_context, 288 iscsi_co_generic_bh_cb, iTask); 289 qemu_bh_schedule(iTask->bh); 290 } else { 291 iTask->complete = 1; 292 } 293 } 294 295 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask) 296 { 297 *iTask = (struct IscsiTask) { 298 .co = qemu_coroutine_self(), 299 .iscsilun = iscsilun, 300 }; 301 } 302 303 static void 304 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data, 305 void *private_data) 306 { 307 IscsiAIOCB *acb = private_data; 308 309 acb->status = -ECANCELED; 310 iscsi_schedule_bh(acb); 311 } 312 313 static void 314 iscsi_aio_cancel(BlockAIOCB *blockacb) 315 { 316 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb; 317 IscsiLun *iscsilun = acb->iscsilun; 318 319 if (acb->status != -EINPROGRESS) { 320 return; 321 } 322 323 /* send a task mgmt call to the target to cancel the task on the target */ 324 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task, 325 iscsi_abort_task_cb, acb); 326 327 } 328 329 static const AIOCBInfo iscsi_aiocb_info = { 330 .aiocb_size = sizeof(IscsiAIOCB), 331 .cancel_async = iscsi_aio_cancel, 332 }; 333 334 335 static void iscsi_process_read(void *arg); 336 static void iscsi_process_write(void *arg); 337 338 static void 339 iscsi_set_events(IscsiLun *iscsilun) 340 { 341 struct iscsi_context *iscsi = iscsilun->iscsi; 342 int ev = iscsi_which_events(iscsi); 343 344 if (ev != iscsilun->events) { 345 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi), 346 false, 347 (ev & POLLIN) ? iscsi_process_read : NULL, 348 (ev & POLLOUT) ? iscsi_process_write : NULL, 349 iscsilun); 350 iscsilun->events = ev; 351 } 352 } 353 354 static void iscsi_timed_check_events(void *opaque) 355 { 356 IscsiLun *iscsilun = opaque; 357 358 /* check for timed out requests */ 359 iscsi_service(iscsilun->iscsi, 0); 360 361 if (iscsilun->request_timed_out) { 362 iscsilun->request_timed_out = false; 363 iscsi_reconnect(iscsilun->iscsi); 364 } 365 366 /* newer versions of libiscsi may return zero events. Ensure we are able 367 * to return to service once this situation changes. */ 368 iscsi_set_events(iscsilun); 369 370 timer_mod(iscsilun->event_timer, 371 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL); 372 } 373 374 static void 375 iscsi_process_read(void *arg) 376 { 377 IscsiLun *iscsilun = arg; 378 struct iscsi_context *iscsi = iscsilun->iscsi; 379 380 iscsi_service(iscsi, POLLIN); 381 iscsi_set_events(iscsilun); 382 } 383 384 static void 385 iscsi_process_write(void *arg) 386 { 387 IscsiLun *iscsilun = arg; 388 struct iscsi_context *iscsi = iscsilun->iscsi; 389 390 iscsi_service(iscsi, POLLOUT); 391 iscsi_set_events(iscsilun); 392 } 393 394 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun) 395 { 396 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE; 397 } 398 399 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun) 400 { 401 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size; 402 } 403 404 static bool is_byte_request_lun_aligned(int64_t offset, int count, 405 IscsiLun *iscsilun) 406 { 407 if (offset % iscsilun->block_size || count % iscsilun->block_size) { 408 error_report("iSCSI misaligned request: " 409 "iscsilun->block_size %u, offset %" PRIi64 410 ", count %d", 411 iscsilun->block_size, offset, count); 412 return false; 413 } 414 return true; 415 } 416 417 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors, 418 IscsiLun *iscsilun) 419 { 420 assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS); 421 return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS, 422 nb_sectors << BDRV_SECTOR_BITS, 423 iscsilun); 424 } 425 426 static unsigned long *iscsi_allocationmap_init(IscsiLun *iscsilun) 427 { 428 return bitmap_try_new(DIV_ROUND_UP(sector_lun2qemu(iscsilun->num_blocks, 429 iscsilun), 430 iscsilun->cluster_sectors)); 431 } 432 433 static void iscsi_allocationmap_set(IscsiLun *iscsilun, int64_t sector_num, 434 int nb_sectors) 435 { 436 if (iscsilun->allocationmap == NULL) { 437 return; 438 } 439 bitmap_set(iscsilun->allocationmap, 440 sector_num / iscsilun->cluster_sectors, 441 DIV_ROUND_UP(nb_sectors, iscsilun->cluster_sectors)); 442 } 443 444 static void iscsi_allocationmap_clear(IscsiLun *iscsilun, int64_t sector_num, 445 int nb_sectors) 446 { 447 int64_t cluster_num, nb_clusters; 448 if (iscsilun->allocationmap == NULL) { 449 return; 450 } 451 cluster_num = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors); 452 nb_clusters = (sector_num + nb_sectors) / iscsilun->cluster_sectors 453 - cluster_num; 454 if (nb_clusters > 0) { 455 bitmap_clear(iscsilun->allocationmap, cluster_num, nb_clusters); 456 } 457 } 458 459 static int coroutine_fn 460 iscsi_co_writev_flags(BlockDriverState *bs, int64_t sector_num, int nb_sectors, 461 QEMUIOVector *iov, int flags) 462 { 463 IscsiLun *iscsilun = bs->opaque; 464 struct IscsiTask iTask; 465 uint64_t lba; 466 uint32_t num_sectors; 467 bool fua = flags & BDRV_REQ_FUA; 468 469 if (fua) { 470 assert(iscsilun->dpofua); 471 } 472 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 473 return -EINVAL; 474 } 475 476 if (bs->bl.max_transfer_length && nb_sectors > bs->bl.max_transfer_length) { 477 error_report("iSCSI Error: Write of %d sectors exceeds max_xfer_len " 478 "of %d sectors", nb_sectors, bs->bl.max_transfer_length); 479 return -EINVAL; 480 } 481 482 lba = sector_qemu2lun(sector_num, iscsilun); 483 num_sectors = sector_qemu2lun(nb_sectors, iscsilun); 484 iscsi_co_init_iscsitask(iscsilun, &iTask); 485 retry: 486 if (iscsilun->use_16_for_rw) { 487 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba, 488 NULL, num_sectors * iscsilun->block_size, 489 iscsilun->block_size, 0, 0, fua, 0, 0, 490 iscsi_co_generic_cb, &iTask); 491 } else { 492 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba, 493 NULL, num_sectors * iscsilun->block_size, 494 iscsilun->block_size, 0, 0, fua, 0, 0, 495 iscsi_co_generic_cb, &iTask); 496 } 497 if (iTask.task == NULL) { 498 return -ENOMEM; 499 } 500 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov, 501 iov->niov); 502 while (!iTask.complete) { 503 iscsi_set_events(iscsilun); 504 qemu_coroutine_yield(); 505 } 506 507 if (iTask.task != NULL) { 508 scsi_free_scsi_task(iTask.task); 509 iTask.task = NULL; 510 } 511 512 if (iTask.do_retry) { 513 iTask.complete = 0; 514 goto retry; 515 } 516 517 if (iTask.status != SCSI_STATUS_GOOD) { 518 return iTask.err_code; 519 } 520 521 iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors); 522 523 return 0; 524 } 525 526 527 static bool iscsi_allocationmap_is_allocated(IscsiLun *iscsilun, 528 int64_t sector_num, int nb_sectors) 529 { 530 unsigned long size; 531 if (iscsilun->allocationmap == NULL) { 532 return true; 533 } 534 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors); 535 return !(find_next_bit(iscsilun->allocationmap, size, 536 sector_num / iscsilun->cluster_sectors) == size); 537 } 538 539 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs, 540 int64_t sector_num, 541 int nb_sectors, int *pnum, 542 BlockDriverState **file) 543 { 544 IscsiLun *iscsilun = bs->opaque; 545 struct scsi_get_lba_status *lbas = NULL; 546 struct scsi_lba_status_descriptor *lbasd = NULL; 547 struct IscsiTask iTask; 548 int64_t ret; 549 550 iscsi_co_init_iscsitask(iscsilun, &iTask); 551 552 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 553 ret = -EINVAL; 554 goto out; 555 } 556 557 /* default to all sectors allocated */ 558 ret = BDRV_BLOCK_DATA; 559 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID; 560 *pnum = nb_sectors; 561 562 /* LUN does not support logical block provisioning */ 563 if (!iscsilun->lbpme) { 564 goto out; 565 } 566 567 retry: 568 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun, 569 sector_qemu2lun(sector_num, iscsilun), 570 8 + 16, iscsi_co_generic_cb, 571 &iTask) == NULL) { 572 ret = -ENOMEM; 573 goto out; 574 } 575 576 while (!iTask.complete) { 577 iscsi_set_events(iscsilun); 578 qemu_coroutine_yield(); 579 } 580 581 if (iTask.do_retry) { 582 if (iTask.task != NULL) { 583 scsi_free_scsi_task(iTask.task); 584 iTask.task = NULL; 585 } 586 iTask.complete = 0; 587 goto retry; 588 } 589 590 if (iTask.status != SCSI_STATUS_GOOD) { 591 /* in case the get_lba_status_callout fails (i.e. 592 * because the device is busy or the cmd is not 593 * supported) we pretend all blocks are allocated 594 * for backwards compatibility */ 595 goto out; 596 } 597 598 lbas = scsi_datain_unmarshall(iTask.task); 599 if (lbas == NULL) { 600 ret = -EIO; 601 goto out; 602 } 603 604 lbasd = &lbas->descriptors[0]; 605 606 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) { 607 ret = -EIO; 608 goto out; 609 } 610 611 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun); 612 613 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED || 614 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) { 615 ret &= ~BDRV_BLOCK_DATA; 616 if (iscsilun->lbprz) { 617 ret |= BDRV_BLOCK_ZERO; 618 } 619 } 620 621 if (ret & BDRV_BLOCK_ZERO) { 622 iscsi_allocationmap_clear(iscsilun, sector_num, *pnum); 623 } else { 624 iscsi_allocationmap_set(iscsilun, sector_num, *pnum); 625 } 626 627 if (*pnum > nb_sectors) { 628 *pnum = nb_sectors; 629 } 630 out: 631 if (iTask.task != NULL) { 632 scsi_free_scsi_task(iTask.task); 633 } 634 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) { 635 *file = bs; 636 } 637 return ret; 638 } 639 640 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs, 641 int64_t sector_num, int nb_sectors, 642 QEMUIOVector *iov) 643 { 644 IscsiLun *iscsilun = bs->opaque; 645 struct IscsiTask iTask; 646 uint64_t lba; 647 uint32_t num_sectors; 648 649 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 650 return -EINVAL; 651 } 652 653 if (bs->bl.max_transfer_length && nb_sectors > bs->bl.max_transfer_length) { 654 error_report("iSCSI Error: Read of %d sectors exceeds max_xfer_len " 655 "of %d sectors", nb_sectors, bs->bl.max_transfer_length); 656 return -EINVAL; 657 } 658 659 if (iscsilun->lbprz && nb_sectors >= ISCSI_CHECKALLOC_THRES && 660 !iscsi_allocationmap_is_allocated(iscsilun, sector_num, nb_sectors)) { 661 int64_t ret; 662 int pnum; 663 BlockDriverState *file; 664 ret = iscsi_co_get_block_status(bs, sector_num, 665 BDRV_REQUEST_MAX_SECTORS, &pnum, &file); 666 if (ret < 0) { 667 return ret; 668 } 669 if (ret & BDRV_BLOCK_ZERO && pnum >= nb_sectors) { 670 qemu_iovec_memset(iov, 0, 0x00, iov->size); 671 return 0; 672 } 673 } 674 675 lba = sector_qemu2lun(sector_num, iscsilun); 676 num_sectors = sector_qemu2lun(nb_sectors, iscsilun); 677 678 iscsi_co_init_iscsitask(iscsilun, &iTask); 679 retry: 680 if (iscsilun->use_16_for_rw) { 681 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba, 682 num_sectors * iscsilun->block_size, 683 iscsilun->block_size, 0, 0, 0, 0, 0, 684 iscsi_co_generic_cb, &iTask); 685 } else { 686 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba, 687 num_sectors * iscsilun->block_size, 688 iscsilun->block_size, 689 0, 0, 0, 0, 0, 690 iscsi_co_generic_cb, &iTask); 691 } 692 if (iTask.task == NULL) { 693 return -ENOMEM; 694 } 695 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov); 696 697 while (!iTask.complete) { 698 iscsi_set_events(iscsilun); 699 qemu_coroutine_yield(); 700 } 701 702 if (iTask.task != NULL) { 703 scsi_free_scsi_task(iTask.task); 704 iTask.task = NULL; 705 } 706 707 if (iTask.do_retry) { 708 iTask.complete = 0; 709 goto retry; 710 } 711 712 if (iTask.status != SCSI_STATUS_GOOD) { 713 return iTask.err_code; 714 } 715 716 return 0; 717 } 718 719 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs) 720 { 721 IscsiLun *iscsilun = bs->opaque; 722 struct IscsiTask iTask; 723 724 iscsi_co_init_iscsitask(iscsilun, &iTask); 725 retry: 726 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0, 727 0, iscsi_co_generic_cb, &iTask) == NULL) { 728 return -ENOMEM; 729 } 730 731 while (!iTask.complete) { 732 iscsi_set_events(iscsilun); 733 qemu_coroutine_yield(); 734 } 735 736 if (iTask.task != NULL) { 737 scsi_free_scsi_task(iTask.task); 738 iTask.task = NULL; 739 } 740 741 if (iTask.do_retry) { 742 iTask.complete = 0; 743 goto retry; 744 } 745 746 if (iTask.status != SCSI_STATUS_GOOD) { 747 return iTask.err_code; 748 } 749 750 return 0; 751 } 752 753 #ifdef __linux__ 754 static void 755 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status, 756 void *command_data, void *opaque) 757 { 758 IscsiAIOCB *acb = opaque; 759 760 g_free(acb->buf); 761 acb->buf = NULL; 762 763 acb->status = 0; 764 if (status < 0) { 765 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s", 766 iscsi_get_error(iscsi)); 767 acb->status = iscsi_translate_sense(&acb->task->sense); 768 } 769 770 acb->ioh->driver_status = 0; 771 acb->ioh->host_status = 0; 772 acb->ioh->resid = 0; 773 acb->ioh->status = status; 774 775 #define SG_ERR_DRIVER_SENSE 0x08 776 777 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) { 778 int ss; 779 780 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE; 781 782 acb->ioh->sb_len_wr = acb->task->datain.size - 2; 783 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ? 784 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr; 785 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss); 786 } 787 788 iscsi_schedule_bh(acb); 789 } 790 791 static void iscsi_ioctl_bh_completion(void *opaque) 792 { 793 IscsiAIOCB *acb = opaque; 794 795 qemu_bh_delete(acb->bh); 796 acb->common.cb(acb->common.opaque, acb->ret); 797 qemu_aio_unref(acb); 798 } 799 800 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf) 801 { 802 BlockDriverState *bs = acb->common.bs; 803 IscsiLun *iscsilun = bs->opaque; 804 int ret = 0; 805 806 switch (req) { 807 case SG_GET_VERSION_NUM: 808 *(int *)buf = 30000; 809 break; 810 case SG_GET_SCSI_ID: 811 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type; 812 break; 813 default: 814 ret = -EINVAL; 815 } 816 assert(!acb->bh); 817 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), 818 iscsi_ioctl_bh_completion, acb); 819 acb->ret = ret; 820 qemu_bh_schedule(acb->bh); 821 } 822 823 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs, 824 unsigned long int req, void *buf, 825 BlockCompletionFunc *cb, void *opaque) 826 { 827 IscsiLun *iscsilun = bs->opaque; 828 struct iscsi_context *iscsi = iscsilun->iscsi; 829 struct iscsi_data data; 830 IscsiAIOCB *acb; 831 832 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque); 833 834 acb->iscsilun = iscsilun; 835 acb->bh = NULL; 836 acb->status = -EINPROGRESS; 837 acb->buf = NULL; 838 acb->ioh = buf; 839 840 if (req != SG_IO) { 841 iscsi_ioctl_handle_emulated(acb, req, buf); 842 return &acb->common; 843 } 844 845 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) { 846 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)", 847 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE); 848 qemu_aio_unref(acb); 849 return NULL; 850 } 851 852 acb->task = malloc(sizeof(struct scsi_task)); 853 if (acb->task == NULL) { 854 error_report("iSCSI: Failed to allocate task for scsi command. %s", 855 iscsi_get_error(iscsi)); 856 qemu_aio_unref(acb); 857 return NULL; 858 } 859 memset(acb->task, 0, sizeof(struct scsi_task)); 860 861 switch (acb->ioh->dxfer_direction) { 862 case SG_DXFER_TO_DEV: 863 acb->task->xfer_dir = SCSI_XFER_WRITE; 864 break; 865 case SG_DXFER_FROM_DEV: 866 acb->task->xfer_dir = SCSI_XFER_READ; 867 break; 868 default: 869 acb->task->xfer_dir = SCSI_XFER_NONE; 870 break; 871 } 872 873 acb->task->cdb_size = acb->ioh->cmd_len; 874 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len); 875 acb->task->expxferlen = acb->ioh->dxfer_len; 876 877 data.size = 0; 878 if (acb->task->xfer_dir == SCSI_XFER_WRITE) { 879 if (acb->ioh->iovec_count == 0) { 880 data.data = acb->ioh->dxferp; 881 data.size = acb->ioh->dxfer_len; 882 } else { 883 scsi_task_set_iov_out(acb->task, 884 (struct scsi_iovec *) acb->ioh->dxferp, 885 acb->ioh->iovec_count); 886 } 887 } 888 889 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task, 890 iscsi_aio_ioctl_cb, 891 (data.size > 0) ? &data : NULL, 892 acb) != 0) { 893 scsi_free_scsi_task(acb->task); 894 qemu_aio_unref(acb); 895 return NULL; 896 } 897 898 /* tell libiscsi to read straight into the buffer we got from ioctl */ 899 if (acb->task->xfer_dir == SCSI_XFER_READ) { 900 if (acb->ioh->iovec_count == 0) { 901 scsi_task_add_data_in_buffer(acb->task, 902 acb->ioh->dxfer_len, 903 acb->ioh->dxferp); 904 } else { 905 scsi_task_set_iov_in(acb->task, 906 (struct scsi_iovec *) acb->ioh->dxferp, 907 acb->ioh->iovec_count); 908 } 909 } 910 911 iscsi_set_events(iscsilun); 912 913 return &acb->common; 914 } 915 916 #endif 917 918 static int64_t 919 iscsi_getlength(BlockDriverState *bs) 920 { 921 IscsiLun *iscsilun = bs->opaque; 922 int64_t len; 923 924 len = iscsilun->num_blocks; 925 len *= iscsilun->block_size; 926 927 return len; 928 } 929 930 static int 931 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num, 932 int nb_sectors) 933 { 934 IscsiLun *iscsilun = bs->opaque; 935 struct IscsiTask iTask; 936 struct unmap_list list; 937 938 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 939 return -EINVAL; 940 } 941 942 if (!iscsilun->lbp.lbpu) { 943 /* UNMAP is not supported by the target */ 944 return 0; 945 } 946 947 list.lba = sector_qemu2lun(sector_num, iscsilun); 948 list.num = sector_qemu2lun(nb_sectors, iscsilun); 949 950 iscsi_co_init_iscsitask(iscsilun, &iTask); 951 retry: 952 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1, 953 iscsi_co_generic_cb, &iTask) == NULL) { 954 return -ENOMEM; 955 } 956 957 while (!iTask.complete) { 958 iscsi_set_events(iscsilun); 959 qemu_coroutine_yield(); 960 } 961 962 if (iTask.task != NULL) { 963 scsi_free_scsi_task(iTask.task); 964 iTask.task = NULL; 965 } 966 967 if (iTask.do_retry) { 968 iTask.complete = 0; 969 goto retry; 970 } 971 972 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) { 973 /* the target might fail with a check condition if it 974 is not happy with the alignment of the UNMAP request 975 we silently fail in this case */ 976 return 0; 977 } 978 979 if (iTask.status != SCSI_STATUS_GOOD) { 980 return iTask.err_code; 981 } 982 983 iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors); 984 985 return 0; 986 } 987 988 static int 989 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, 990 int count, BdrvRequestFlags flags) 991 { 992 IscsiLun *iscsilun = bs->opaque; 993 struct IscsiTask iTask; 994 uint64_t lba; 995 uint32_t nb_blocks; 996 bool use_16_for_ws = iscsilun->use_16_for_rw; 997 998 if (!is_byte_request_lun_aligned(offset, count, iscsilun)) { 999 return -ENOTSUP; 1000 } 1001 1002 if (flags & BDRV_REQ_MAY_UNMAP) { 1003 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) { 1004 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */ 1005 use_16_for_ws = true; 1006 } 1007 if (use_16_for_ws && !iscsilun->lbp.lbpws) { 1008 /* WRITESAME16 with UNMAP is not supported by the target, 1009 * fall back and try WRITESAME10/16 without UNMAP */ 1010 flags &= ~BDRV_REQ_MAY_UNMAP; 1011 use_16_for_ws = iscsilun->use_16_for_rw; 1012 } 1013 } 1014 1015 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) { 1016 /* WRITESAME without UNMAP is not supported by the target */ 1017 return -ENOTSUP; 1018 } 1019 1020 lba = offset / iscsilun->block_size; 1021 nb_blocks = count / iscsilun->block_size; 1022 1023 if (iscsilun->zeroblock == NULL) { 1024 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size); 1025 if (iscsilun->zeroblock == NULL) { 1026 return -ENOMEM; 1027 } 1028 } 1029 1030 iscsi_co_init_iscsitask(iscsilun, &iTask); 1031 retry: 1032 if (use_16_for_ws) { 1033 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba, 1034 iscsilun->zeroblock, iscsilun->block_size, 1035 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 1036 0, 0, iscsi_co_generic_cb, &iTask); 1037 } else { 1038 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba, 1039 iscsilun->zeroblock, iscsilun->block_size, 1040 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 1041 0, 0, iscsi_co_generic_cb, &iTask); 1042 } 1043 if (iTask.task == NULL) { 1044 return -ENOMEM; 1045 } 1046 1047 while (!iTask.complete) { 1048 iscsi_set_events(iscsilun); 1049 qemu_coroutine_yield(); 1050 } 1051 1052 if (iTask.status == SCSI_STATUS_CHECK_CONDITION && 1053 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST && 1054 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE || 1055 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) { 1056 /* WRITE SAME is not supported by the target */ 1057 iscsilun->has_write_same = false; 1058 scsi_free_scsi_task(iTask.task); 1059 return -ENOTSUP; 1060 } 1061 1062 if (iTask.task != NULL) { 1063 scsi_free_scsi_task(iTask.task); 1064 iTask.task = NULL; 1065 } 1066 1067 if (iTask.do_retry) { 1068 iTask.complete = 0; 1069 goto retry; 1070 } 1071 1072 if (iTask.status != SCSI_STATUS_GOOD) { 1073 return iTask.err_code; 1074 } 1075 1076 if (flags & BDRV_REQ_MAY_UNMAP) { 1077 iscsi_allocationmap_clear(iscsilun, offset >> BDRV_SECTOR_BITS, 1078 count >> BDRV_SECTOR_BITS); 1079 } else { 1080 iscsi_allocationmap_set(iscsilun, offset >> BDRV_SECTOR_BITS, 1081 count >> BDRV_SECTOR_BITS); 1082 } 1083 1084 return 0; 1085 } 1086 1087 static void parse_chap(struct iscsi_context *iscsi, const char *target, 1088 Error **errp) 1089 { 1090 QemuOptsList *list; 1091 QemuOpts *opts; 1092 const char *user = NULL; 1093 const char *password = NULL; 1094 const char *secretid; 1095 char *secret = NULL; 1096 1097 list = qemu_find_opts("iscsi"); 1098 if (!list) { 1099 return; 1100 } 1101 1102 opts = qemu_opts_find(list, target); 1103 if (opts == NULL) { 1104 opts = QTAILQ_FIRST(&list->head); 1105 if (!opts) { 1106 return; 1107 } 1108 } 1109 1110 user = qemu_opt_get(opts, "user"); 1111 if (!user) { 1112 return; 1113 } 1114 1115 secretid = qemu_opt_get(opts, "password-secret"); 1116 password = qemu_opt_get(opts, "password"); 1117 if (secretid && password) { 1118 error_setg(errp, "'password' and 'password-secret' properties are " 1119 "mutually exclusive"); 1120 return; 1121 } 1122 if (secretid) { 1123 secret = qcrypto_secret_lookup_as_utf8(secretid, errp); 1124 if (!secret) { 1125 return; 1126 } 1127 password = secret; 1128 } else if (!password) { 1129 error_setg(errp, "CHAP username specified but no password was given"); 1130 return; 1131 } 1132 1133 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) { 1134 error_setg(errp, "Failed to set initiator username and password"); 1135 } 1136 1137 g_free(secret); 1138 } 1139 1140 static void parse_header_digest(struct iscsi_context *iscsi, const char *target, 1141 Error **errp) 1142 { 1143 QemuOptsList *list; 1144 QemuOpts *opts; 1145 const char *digest = NULL; 1146 1147 list = qemu_find_opts("iscsi"); 1148 if (!list) { 1149 return; 1150 } 1151 1152 opts = qemu_opts_find(list, target); 1153 if (opts == NULL) { 1154 opts = QTAILQ_FIRST(&list->head); 1155 if (!opts) { 1156 return; 1157 } 1158 } 1159 1160 digest = qemu_opt_get(opts, "header-digest"); 1161 if (!digest) { 1162 return; 1163 } 1164 1165 if (!strcmp(digest, "CRC32C")) { 1166 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C); 1167 } else if (!strcmp(digest, "NONE")) { 1168 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE); 1169 } else if (!strcmp(digest, "CRC32C-NONE")) { 1170 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE); 1171 } else if (!strcmp(digest, "NONE-CRC32C")) { 1172 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1173 } else { 1174 error_setg(errp, "Invalid header-digest setting : %s", digest); 1175 } 1176 } 1177 1178 static char *parse_initiator_name(const char *target) 1179 { 1180 QemuOptsList *list; 1181 QemuOpts *opts; 1182 const char *name; 1183 char *iscsi_name; 1184 UuidInfo *uuid_info; 1185 1186 list = qemu_find_opts("iscsi"); 1187 if (list) { 1188 opts = qemu_opts_find(list, target); 1189 if (!opts) { 1190 opts = QTAILQ_FIRST(&list->head); 1191 } 1192 if (opts) { 1193 name = qemu_opt_get(opts, "initiator-name"); 1194 if (name) { 1195 return g_strdup(name); 1196 } 1197 } 1198 } 1199 1200 uuid_info = qmp_query_uuid(NULL); 1201 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) { 1202 name = qemu_get_vm_name(); 1203 } else { 1204 name = uuid_info->UUID; 1205 } 1206 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s", 1207 name ? ":" : "", name ? name : ""); 1208 qapi_free_UuidInfo(uuid_info); 1209 return iscsi_name; 1210 } 1211 1212 static int parse_timeout(const char *target) 1213 { 1214 QemuOptsList *list; 1215 QemuOpts *opts; 1216 const char *timeout; 1217 1218 list = qemu_find_opts("iscsi"); 1219 if (list) { 1220 opts = qemu_opts_find(list, target); 1221 if (!opts) { 1222 opts = QTAILQ_FIRST(&list->head); 1223 } 1224 if (opts) { 1225 timeout = qemu_opt_get(opts, "timeout"); 1226 if (timeout) { 1227 return atoi(timeout); 1228 } 1229 } 1230 } 1231 1232 return 0; 1233 } 1234 1235 static void iscsi_nop_timed_event(void *opaque) 1236 { 1237 IscsiLun *iscsilun = opaque; 1238 1239 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) { 1240 error_report("iSCSI: NOP timeout. Reconnecting..."); 1241 iscsilun->request_timed_out = true; 1242 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) { 1243 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages."); 1244 return; 1245 } 1246 1247 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1248 iscsi_set_events(iscsilun); 1249 } 1250 1251 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp) 1252 { 1253 struct scsi_task *task = NULL; 1254 struct scsi_readcapacity10 *rc10 = NULL; 1255 struct scsi_readcapacity16 *rc16 = NULL; 1256 int retries = ISCSI_CMD_RETRIES; 1257 1258 do { 1259 if (task != NULL) { 1260 scsi_free_scsi_task(task); 1261 task = NULL; 1262 } 1263 1264 switch (iscsilun->type) { 1265 case TYPE_DISK: 1266 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun); 1267 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1268 rc16 = scsi_datain_unmarshall(task); 1269 if (rc16 == NULL) { 1270 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data."); 1271 } else { 1272 iscsilun->block_size = rc16->block_length; 1273 iscsilun->num_blocks = rc16->returned_lba + 1; 1274 iscsilun->lbpme = !!rc16->lbpme; 1275 iscsilun->lbprz = !!rc16->lbprz; 1276 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff); 1277 } 1278 break; 1279 } 1280 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION 1281 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) { 1282 break; 1283 } 1284 /* Fall through and try READ CAPACITY(10) instead. */ 1285 case TYPE_ROM: 1286 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0); 1287 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1288 rc10 = scsi_datain_unmarshall(task); 1289 if (rc10 == NULL) { 1290 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data."); 1291 } else { 1292 iscsilun->block_size = rc10->block_size; 1293 if (rc10->lba == 0) { 1294 /* blank disk loaded */ 1295 iscsilun->num_blocks = 0; 1296 } else { 1297 iscsilun->num_blocks = rc10->lba + 1; 1298 } 1299 } 1300 } 1301 break; 1302 default: 1303 return; 1304 } 1305 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION 1306 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION 1307 && retries-- > 0); 1308 1309 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1310 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command"); 1311 } else if (!iscsilun->block_size || 1312 iscsilun->block_size % BDRV_SECTOR_SIZE) { 1313 error_setg(errp, "iSCSI: the target returned an invalid " 1314 "block size of %d.", iscsilun->block_size); 1315 } 1316 if (task) { 1317 scsi_free_scsi_task(task); 1318 } 1319 } 1320 1321 /* TODO Convert to fine grained options */ 1322 static QemuOptsList runtime_opts = { 1323 .name = "iscsi", 1324 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), 1325 .desc = { 1326 { 1327 .name = "filename", 1328 .type = QEMU_OPT_STRING, 1329 .help = "URL to the iscsi image", 1330 }, 1331 { /* end of list */ } 1332 }, 1333 }; 1334 1335 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun, 1336 int evpd, int pc, void **inq, Error **errp) 1337 { 1338 int full_size; 1339 struct scsi_task *task = NULL; 1340 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64); 1341 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1342 goto fail; 1343 } 1344 full_size = scsi_datain_getfullsize(task); 1345 if (full_size > task->datain.size) { 1346 scsi_free_scsi_task(task); 1347 1348 /* we need more data for the full list */ 1349 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size); 1350 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1351 goto fail; 1352 } 1353 } 1354 1355 *inq = scsi_datain_unmarshall(task); 1356 if (*inq == NULL) { 1357 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob"); 1358 goto fail_with_err; 1359 } 1360 1361 return task; 1362 1363 fail: 1364 error_setg(errp, "iSCSI: Inquiry command failed : %s", 1365 iscsi_get_error(iscsi)); 1366 fail_with_err: 1367 if (task != NULL) { 1368 scsi_free_scsi_task(task); 1369 } 1370 return NULL; 1371 } 1372 1373 static void iscsi_detach_aio_context(BlockDriverState *bs) 1374 { 1375 IscsiLun *iscsilun = bs->opaque; 1376 1377 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi), 1378 false, NULL, NULL, NULL); 1379 iscsilun->events = 0; 1380 1381 if (iscsilun->nop_timer) { 1382 timer_del(iscsilun->nop_timer); 1383 timer_free(iscsilun->nop_timer); 1384 iscsilun->nop_timer = NULL; 1385 } 1386 if (iscsilun->event_timer) { 1387 timer_del(iscsilun->event_timer); 1388 timer_free(iscsilun->event_timer); 1389 iscsilun->event_timer = NULL; 1390 } 1391 } 1392 1393 static void iscsi_attach_aio_context(BlockDriverState *bs, 1394 AioContext *new_context) 1395 { 1396 IscsiLun *iscsilun = bs->opaque; 1397 1398 iscsilun->aio_context = new_context; 1399 iscsi_set_events(iscsilun); 1400 1401 /* Set up a timer for sending out iSCSI NOPs */ 1402 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context, 1403 QEMU_CLOCK_REALTIME, SCALE_MS, 1404 iscsi_nop_timed_event, iscsilun); 1405 timer_mod(iscsilun->nop_timer, 1406 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1407 1408 /* Set up a timer for periodic calls to iscsi_set_events and to 1409 * scan for command timeout */ 1410 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context, 1411 QEMU_CLOCK_REALTIME, SCALE_MS, 1412 iscsi_timed_check_events, iscsilun); 1413 timer_mod(iscsilun->event_timer, 1414 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL); 1415 } 1416 1417 static void iscsi_modesense_sync(IscsiLun *iscsilun) 1418 { 1419 struct scsi_task *task; 1420 struct scsi_mode_sense *ms = NULL; 1421 iscsilun->write_protected = false; 1422 iscsilun->dpofua = false; 1423 1424 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun, 1425 1, SCSI_MODESENSE_PC_CURRENT, 1426 0x3F, 0, 255); 1427 if (task == NULL) { 1428 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s", 1429 iscsi_get_error(iscsilun->iscsi)); 1430 goto out; 1431 } 1432 1433 if (task->status != SCSI_STATUS_GOOD) { 1434 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable"); 1435 goto out; 1436 } 1437 ms = scsi_datain_unmarshall(task); 1438 if (!ms) { 1439 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s", 1440 iscsi_get_error(iscsilun->iscsi)); 1441 goto out; 1442 } 1443 iscsilun->write_protected = ms->device_specific_parameter & 0x80; 1444 iscsilun->dpofua = ms->device_specific_parameter & 0x10; 1445 1446 out: 1447 if (task) { 1448 scsi_free_scsi_task(task); 1449 } 1450 } 1451 1452 /* 1453 * We support iscsi url's on the form 1454 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun> 1455 */ 1456 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags, 1457 Error **errp) 1458 { 1459 IscsiLun *iscsilun = bs->opaque; 1460 struct iscsi_context *iscsi = NULL; 1461 struct iscsi_url *iscsi_url = NULL; 1462 struct scsi_task *task = NULL; 1463 struct scsi_inquiry_standard *inq = NULL; 1464 struct scsi_inquiry_supported_pages *inq_vpd; 1465 char *initiator_name = NULL; 1466 QemuOpts *opts; 1467 Error *local_err = NULL; 1468 const char *filename; 1469 int i, ret = 0, timeout = 0; 1470 1471 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); 1472 qemu_opts_absorb_qdict(opts, options, &local_err); 1473 if (local_err) { 1474 error_propagate(errp, local_err); 1475 ret = -EINVAL; 1476 goto out; 1477 } 1478 1479 filename = qemu_opt_get(opts, "filename"); 1480 1481 iscsi_url = iscsi_parse_full_url(iscsi, filename); 1482 if (iscsi_url == NULL) { 1483 error_setg(errp, "Failed to parse URL : %s", filename); 1484 ret = -EINVAL; 1485 goto out; 1486 } 1487 1488 memset(iscsilun, 0, sizeof(IscsiLun)); 1489 1490 initiator_name = parse_initiator_name(iscsi_url->target); 1491 1492 iscsi = iscsi_create_context(initiator_name); 1493 if (iscsi == NULL) { 1494 error_setg(errp, "iSCSI: Failed to create iSCSI context."); 1495 ret = -ENOMEM; 1496 goto out; 1497 } 1498 1499 if (iscsi_set_targetname(iscsi, iscsi_url->target)) { 1500 error_setg(errp, "iSCSI: Failed to set target name."); 1501 ret = -EINVAL; 1502 goto out; 1503 } 1504 1505 if (iscsi_url->user[0] != '\0') { 1506 ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user, 1507 iscsi_url->passwd); 1508 if (ret != 0) { 1509 error_setg(errp, "Failed to set initiator username and password"); 1510 ret = -EINVAL; 1511 goto out; 1512 } 1513 } 1514 1515 /* check if we got CHAP username/password via the options */ 1516 parse_chap(iscsi, iscsi_url->target, &local_err); 1517 if (local_err != NULL) { 1518 error_propagate(errp, local_err); 1519 ret = -EINVAL; 1520 goto out; 1521 } 1522 1523 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) { 1524 error_setg(errp, "iSCSI: Failed to set session type to normal."); 1525 ret = -EINVAL; 1526 goto out; 1527 } 1528 1529 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1530 1531 /* check if we got HEADER_DIGEST via the options */ 1532 parse_header_digest(iscsi, iscsi_url->target, &local_err); 1533 if (local_err != NULL) { 1534 error_propagate(errp, local_err); 1535 ret = -EINVAL; 1536 goto out; 1537 } 1538 1539 /* timeout handling is broken in libiscsi before 1.15.0 */ 1540 timeout = parse_timeout(iscsi_url->target); 1541 #if defined(LIBISCSI_API_VERSION) && LIBISCSI_API_VERSION >= 20150621 1542 iscsi_set_timeout(iscsi, timeout); 1543 #else 1544 if (timeout) { 1545 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0"); 1546 } 1547 #endif 1548 1549 if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) { 1550 error_setg(errp, "iSCSI: Failed to connect to LUN : %s", 1551 iscsi_get_error(iscsi)); 1552 ret = -EINVAL; 1553 goto out; 1554 } 1555 1556 iscsilun->iscsi = iscsi; 1557 iscsilun->aio_context = bdrv_get_aio_context(bs); 1558 iscsilun->lun = iscsi_url->lun; 1559 iscsilun->has_write_same = true; 1560 1561 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0, 1562 (void **) &inq, errp); 1563 if (task == NULL) { 1564 ret = -EINVAL; 1565 goto out; 1566 } 1567 iscsilun->type = inq->periperal_device_type; 1568 scsi_free_scsi_task(task); 1569 task = NULL; 1570 1571 iscsi_modesense_sync(iscsilun); 1572 if (iscsilun->dpofua) { 1573 bs->supported_write_flags = BDRV_REQ_FUA; 1574 } 1575 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP; 1576 1577 /* Check the write protect flag of the LUN if we want to write */ 1578 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) && 1579 iscsilun->write_protected) { 1580 error_setg(errp, "Cannot open a write protected LUN as read-write"); 1581 ret = -EACCES; 1582 goto out; 1583 } 1584 1585 iscsi_readcapacity_sync(iscsilun, &local_err); 1586 if (local_err != NULL) { 1587 error_propagate(errp, local_err); 1588 ret = -EINVAL; 1589 goto out; 1590 } 1591 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun); 1592 bs->request_alignment = iscsilun->block_size; 1593 1594 /* We don't have any emulation for devices other than disks and CD-ROMs, so 1595 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu 1596 * will try to read from the device to guess the image format. 1597 */ 1598 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) { 1599 bs->sg = 1; 1600 } 1601 1602 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1603 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES, 1604 (void **) &inq_vpd, errp); 1605 if (task == NULL) { 1606 ret = -EINVAL; 1607 goto out; 1608 } 1609 for (i = 0; i < inq_vpd->num_pages; i++) { 1610 struct scsi_task *inq_task; 1611 struct scsi_inquiry_logical_block_provisioning *inq_lbp; 1612 struct scsi_inquiry_block_limits *inq_bl; 1613 switch (inq_vpd->pages[i]) { 1614 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING: 1615 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1616 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING, 1617 (void **) &inq_lbp, errp); 1618 if (inq_task == NULL) { 1619 ret = -EINVAL; 1620 goto out; 1621 } 1622 memcpy(&iscsilun->lbp, inq_lbp, 1623 sizeof(struct scsi_inquiry_logical_block_provisioning)); 1624 scsi_free_scsi_task(inq_task); 1625 break; 1626 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS: 1627 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1628 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS, 1629 (void **) &inq_bl, errp); 1630 if (inq_task == NULL) { 1631 ret = -EINVAL; 1632 goto out; 1633 } 1634 memcpy(&iscsilun->bl, inq_bl, 1635 sizeof(struct scsi_inquiry_block_limits)); 1636 scsi_free_scsi_task(inq_task); 1637 break; 1638 default: 1639 break; 1640 } 1641 } 1642 scsi_free_scsi_task(task); 1643 task = NULL; 1644 1645 iscsi_attach_aio_context(bs, iscsilun->aio_context); 1646 1647 /* Guess the internal cluster (page) size of the iscsi target by the means 1648 * of opt_unmap_gran. Transfer the unmap granularity only if it has a 1649 * reasonable size */ 1650 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 && 1651 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) { 1652 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran * 1653 iscsilun->block_size) >> BDRV_SECTOR_BITS; 1654 if (iscsilun->lbprz) { 1655 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun); 1656 if (iscsilun->allocationmap == NULL) { 1657 ret = -ENOMEM; 1658 } 1659 } 1660 } 1661 1662 out: 1663 qemu_opts_del(opts); 1664 g_free(initiator_name); 1665 if (iscsi_url != NULL) { 1666 iscsi_destroy_url(iscsi_url); 1667 } 1668 if (task != NULL) { 1669 scsi_free_scsi_task(task); 1670 } 1671 1672 if (ret) { 1673 if (iscsi != NULL) { 1674 if (iscsi_is_logged_in(iscsi)) { 1675 iscsi_logout_sync(iscsi); 1676 } 1677 iscsi_destroy_context(iscsi); 1678 } 1679 memset(iscsilun, 0, sizeof(IscsiLun)); 1680 } 1681 return ret; 1682 } 1683 1684 static void iscsi_close(BlockDriverState *bs) 1685 { 1686 IscsiLun *iscsilun = bs->opaque; 1687 struct iscsi_context *iscsi = iscsilun->iscsi; 1688 1689 iscsi_detach_aio_context(bs); 1690 if (iscsi_is_logged_in(iscsi)) { 1691 iscsi_logout_sync(iscsi); 1692 } 1693 iscsi_destroy_context(iscsi); 1694 g_free(iscsilun->zeroblock); 1695 g_free(iscsilun->allocationmap); 1696 memset(iscsilun, 0, sizeof(IscsiLun)); 1697 } 1698 1699 static int sector_limits_lun2qemu(int64_t sector, IscsiLun *iscsilun) 1700 { 1701 return MIN(sector_lun2qemu(sector, iscsilun), INT_MAX / 2 + 1); 1702 } 1703 1704 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp) 1705 { 1706 /* We don't actually refresh here, but just return data queried in 1707 * iscsi_open(): iscsi targets don't change their limits. */ 1708 1709 IscsiLun *iscsilun = bs->opaque; 1710 uint32_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff; 1711 1712 if (iscsilun->bl.max_xfer_len) { 1713 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len); 1714 } 1715 1716 bs->bl.max_transfer_length = sector_limits_lun2qemu(max_xfer_len, iscsilun); 1717 1718 if (iscsilun->lbp.lbpu) { 1719 if (iscsilun->bl.max_unmap < 0xffffffff) { 1720 bs->bl.max_discard = 1721 sector_limits_lun2qemu(iscsilun->bl.max_unmap, iscsilun); 1722 } 1723 bs->bl.discard_alignment = 1724 sector_limits_lun2qemu(iscsilun->bl.opt_unmap_gran, iscsilun); 1725 } else { 1726 bs->bl.discard_alignment = iscsilun->block_size >> BDRV_SECTOR_BITS; 1727 } 1728 1729 if (iscsilun->bl.max_ws_len < 0xffffffff / iscsilun->block_size) { 1730 bs->bl.max_pwrite_zeroes = 1731 iscsilun->bl.max_ws_len * iscsilun->block_size; 1732 } 1733 if (iscsilun->lbp.lbpws) { 1734 bs->bl.pwrite_zeroes_alignment = 1735 iscsilun->bl.opt_unmap_gran * iscsilun->block_size; 1736 } else { 1737 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size; 1738 } 1739 bs->bl.opt_transfer_length = 1740 sector_limits_lun2qemu(iscsilun->bl.opt_xfer_len, iscsilun); 1741 } 1742 1743 /* Note that this will not re-establish a connection with an iSCSI target - it 1744 * is effectively a NOP. */ 1745 static int iscsi_reopen_prepare(BDRVReopenState *state, 1746 BlockReopenQueue *queue, Error **errp) 1747 { 1748 IscsiLun *iscsilun = state->bs->opaque; 1749 1750 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) { 1751 error_setg(errp, "Cannot open a write protected LUN as read-write"); 1752 return -EACCES; 1753 } 1754 return 0; 1755 } 1756 1757 static int iscsi_truncate(BlockDriverState *bs, int64_t offset) 1758 { 1759 IscsiLun *iscsilun = bs->opaque; 1760 Error *local_err = NULL; 1761 1762 if (iscsilun->type != TYPE_DISK) { 1763 return -ENOTSUP; 1764 } 1765 1766 iscsi_readcapacity_sync(iscsilun, &local_err); 1767 if (local_err != NULL) { 1768 error_free(local_err); 1769 return -EIO; 1770 } 1771 1772 if (offset > iscsi_getlength(bs)) { 1773 return -EINVAL; 1774 } 1775 1776 if (iscsilun->allocationmap != NULL) { 1777 g_free(iscsilun->allocationmap); 1778 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun); 1779 } 1780 1781 return 0; 1782 } 1783 1784 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp) 1785 { 1786 int ret = 0; 1787 int64_t total_size = 0; 1788 BlockDriverState *bs; 1789 IscsiLun *iscsilun = NULL; 1790 QDict *bs_options; 1791 1792 bs = bdrv_new(); 1793 1794 /* Read out options */ 1795 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 1796 BDRV_SECTOR_SIZE); 1797 bs->opaque = g_new0(struct IscsiLun, 1); 1798 iscsilun = bs->opaque; 1799 1800 bs_options = qdict_new(); 1801 qdict_put(bs_options, "filename", qstring_from_str(filename)); 1802 ret = iscsi_open(bs, bs_options, 0, NULL); 1803 QDECREF(bs_options); 1804 1805 if (ret != 0) { 1806 goto out; 1807 } 1808 iscsi_detach_aio_context(bs); 1809 if (iscsilun->type != TYPE_DISK) { 1810 ret = -ENODEV; 1811 goto out; 1812 } 1813 if (bs->total_sectors < total_size) { 1814 ret = -ENOSPC; 1815 goto out; 1816 } 1817 1818 ret = 0; 1819 out: 1820 if (iscsilun->iscsi != NULL) { 1821 iscsi_destroy_context(iscsilun->iscsi); 1822 } 1823 g_free(bs->opaque); 1824 bs->opaque = NULL; 1825 bdrv_unref(bs); 1826 return ret; 1827 } 1828 1829 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 1830 { 1831 IscsiLun *iscsilun = bs->opaque; 1832 bdi->unallocated_blocks_are_zero = iscsilun->lbprz; 1833 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws; 1834 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE; 1835 return 0; 1836 } 1837 1838 static QemuOptsList iscsi_create_opts = { 1839 .name = "iscsi-create-opts", 1840 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head), 1841 .desc = { 1842 { 1843 .name = BLOCK_OPT_SIZE, 1844 .type = QEMU_OPT_SIZE, 1845 .help = "Virtual disk size" 1846 }, 1847 { /* end of list */ } 1848 } 1849 }; 1850 1851 static BlockDriver bdrv_iscsi = { 1852 .format_name = "iscsi", 1853 .protocol_name = "iscsi", 1854 1855 .instance_size = sizeof(IscsiLun), 1856 .bdrv_needs_filename = true, 1857 .bdrv_file_open = iscsi_open, 1858 .bdrv_close = iscsi_close, 1859 .bdrv_create = iscsi_create, 1860 .create_opts = &iscsi_create_opts, 1861 .bdrv_reopen_prepare = iscsi_reopen_prepare, 1862 1863 .bdrv_getlength = iscsi_getlength, 1864 .bdrv_get_info = iscsi_get_info, 1865 .bdrv_truncate = iscsi_truncate, 1866 .bdrv_refresh_limits = iscsi_refresh_limits, 1867 1868 .bdrv_co_get_block_status = iscsi_co_get_block_status, 1869 .bdrv_co_discard = iscsi_co_discard, 1870 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes, 1871 .bdrv_co_readv = iscsi_co_readv, 1872 .bdrv_co_writev_flags = iscsi_co_writev_flags, 1873 .bdrv_co_flush_to_disk = iscsi_co_flush, 1874 1875 #ifdef __linux__ 1876 .bdrv_aio_ioctl = iscsi_aio_ioctl, 1877 #endif 1878 1879 .bdrv_detach_aio_context = iscsi_detach_aio_context, 1880 .bdrv_attach_aio_context = iscsi_attach_aio_context, 1881 }; 1882 1883 static QemuOptsList qemu_iscsi_opts = { 1884 .name = "iscsi", 1885 .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head), 1886 .desc = { 1887 { 1888 .name = "user", 1889 .type = QEMU_OPT_STRING, 1890 .help = "username for CHAP authentication to target", 1891 },{ 1892 .name = "password", 1893 .type = QEMU_OPT_STRING, 1894 .help = "password for CHAP authentication to target", 1895 },{ 1896 .name = "password-secret", 1897 .type = QEMU_OPT_STRING, 1898 .help = "ID of the secret providing password for CHAP " 1899 "authentication to target", 1900 },{ 1901 .name = "header-digest", 1902 .type = QEMU_OPT_STRING, 1903 .help = "HeaderDigest setting. " 1904 "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}", 1905 },{ 1906 .name = "initiator-name", 1907 .type = QEMU_OPT_STRING, 1908 .help = "Initiator iqn name to use when connecting", 1909 },{ 1910 .name = "timeout", 1911 .type = QEMU_OPT_NUMBER, 1912 .help = "Request timeout in seconds (default 0 = no timeout)", 1913 }, 1914 { /* end of list */ } 1915 }, 1916 }; 1917 1918 static void iscsi_block_init(void) 1919 { 1920 bdrv_register(&bdrv_iscsi); 1921 qemu_add_opts(&qemu_iscsi_opts); 1922 } 1923 1924 block_init(iscsi_block_init); 1925