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, INT_MAX, &pnum, &file); 665 if (ret < 0) { 666 return ret; 667 } 668 if (ret & BDRV_BLOCK_ZERO && pnum >= nb_sectors) { 669 qemu_iovec_memset(iov, 0, 0x00, iov->size); 670 return 0; 671 } 672 } 673 674 lba = sector_qemu2lun(sector_num, iscsilun); 675 num_sectors = sector_qemu2lun(nb_sectors, iscsilun); 676 677 iscsi_co_init_iscsitask(iscsilun, &iTask); 678 retry: 679 if (iscsilun->use_16_for_rw) { 680 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba, 681 num_sectors * iscsilun->block_size, 682 iscsilun->block_size, 0, 0, 0, 0, 0, 683 iscsi_co_generic_cb, &iTask); 684 } else { 685 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba, 686 num_sectors * iscsilun->block_size, 687 iscsilun->block_size, 688 0, 0, 0, 0, 0, 689 iscsi_co_generic_cb, &iTask); 690 } 691 if (iTask.task == NULL) { 692 return -ENOMEM; 693 } 694 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov); 695 696 while (!iTask.complete) { 697 iscsi_set_events(iscsilun); 698 qemu_coroutine_yield(); 699 } 700 701 if (iTask.task != NULL) { 702 scsi_free_scsi_task(iTask.task); 703 iTask.task = NULL; 704 } 705 706 if (iTask.do_retry) { 707 iTask.complete = 0; 708 goto retry; 709 } 710 711 if (iTask.status != SCSI_STATUS_GOOD) { 712 return iTask.err_code; 713 } 714 715 return 0; 716 } 717 718 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs) 719 { 720 IscsiLun *iscsilun = bs->opaque; 721 struct IscsiTask iTask; 722 723 iscsi_co_init_iscsitask(iscsilun, &iTask); 724 retry: 725 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0, 726 0, iscsi_co_generic_cb, &iTask) == NULL) { 727 return -ENOMEM; 728 } 729 730 while (!iTask.complete) { 731 iscsi_set_events(iscsilun); 732 qemu_coroutine_yield(); 733 } 734 735 if (iTask.task != NULL) { 736 scsi_free_scsi_task(iTask.task); 737 iTask.task = NULL; 738 } 739 740 if (iTask.do_retry) { 741 iTask.complete = 0; 742 goto retry; 743 } 744 745 if (iTask.status != SCSI_STATUS_GOOD) { 746 return iTask.err_code; 747 } 748 749 return 0; 750 } 751 752 #ifdef __linux__ 753 static void 754 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status, 755 void *command_data, void *opaque) 756 { 757 IscsiAIOCB *acb = opaque; 758 759 g_free(acb->buf); 760 acb->buf = NULL; 761 762 acb->status = 0; 763 if (status < 0) { 764 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s", 765 iscsi_get_error(iscsi)); 766 acb->status = iscsi_translate_sense(&acb->task->sense); 767 } 768 769 acb->ioh->driver_status = 0; 770 acb->ioh->host_status = 0; 771 acb->ioh->resid = 0; 772 acb->ioh->status = status; 773 774 #define SG_ERR_DRIVER_SENSE 0x08 775 776 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) { 777 int ss; 778 779 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE; 780 781 acb->ioh->sb_len_wr = acb->task->datain.size - 2; 782 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ? 783 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr; 784 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss); 785 } 786 787 iscsi_schedule_bh(acb); 788 } 789 790 static void iscsi_ioctl_bh_completion(void *opaque) 791 { 792 IscsiAIOCB *acb = opaque; 793 794 qemu_bh_delete(acb->bh); 795 acb->common.cb(acb->common.opaque, acb->ret); 796 qemu_aio_unref(acb); 797 } 798 799 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf) 800 { 801 BlockDriverState *bs = acb->common.bs; 802 IscsiLun *iscsilun = bs->opaque; 803 int ret = 0; 804 805 switch (req) { 806 case SG_GET_VERSION_NUM: 807 *(int *)buf = 30000; 808 break; 809 case SG_GET_SCSI_ID: 810 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type; 811 break; 812 default: 813 ret = -EINVAL; 814 } 815 assert(!acb->bh); 816 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), 817 iscsi_ioctl_bh_completion, acb); 818 acb->ret = ret; 819 qemu_bh_schedule(acb->bh); 820 } 821 822 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs, 823 unsigned long int req, void *buf, 824 BlockCompletionFunc *cb, void *opaque) 825 { 826 IscsiLun *iscsilun = bs->opaque; 827 struct iscsi_context *iscsi = iscsilun->iscsi; 828 struct iscsi_data data; 829 IscsiAIOCB *acb; 830 831 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque); 832 833 acb->iscsilun = iscsilun; 834 acb->bh = NULL; 835 acb->status = -EINPROGRESS; 836 acb->buf = NULL; 837 acb->ioh = buf; 838 839 if (req != SG_IO) { 840 iscsi_ioctl_handle_emulated(acb, req, buf); 841 return &acb->common; 842 } 843 844 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) { 845 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)", 846 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE); 847 qemu_aio_unref(acb); 848 return NULL; 849 } 850 851 acb->task = malloc(sizeof(struct scsi_task)); 852 if (acb->task == NULL) { 853 error_report("iSCSI: Failed to allocate task for scsi command. %s", 854 iscsi_get_error(iscsi)); 855 qemu_aio_unref(acb); 856 return NULL; 857 } 858 memset(acb->task, 0, sizeof(struct scsi_task)); 859 860 switch (acb->ioh->dxfer_direction) { 861 case SG_DXFER_TO_DEV: 862 acb->task->xfer_dir = SCSI_XFER_WRITE; 863 break; 864 case SG_DXFER_FROM_DEV: 865 acb->task->xfer_dir = SCSI_XFER_READ; 866 break; 867 default: 868 acb->task->xfer_dir = SCSI_XFER_NONE; 869 break; 870 } 871 872 acb->task->cdb_size = acb->ioh->cmd_len; 873 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len); 874 acb->task->expxferlen = acb->ioh->dxfer_len; 875 876 data.size = 0; 877 if (acb->task->xfer_dir == SCSI_XFER_WRITE) { 878 if (acb->ioh->iovec_count == 0) { 879 data.data = acb->ioh->dxferp; 880 data.size = acb->ioh->dxfer_len; 881 } else { 882 scsi_task_set_iov_out(acb->task, 883 (struct scsi_iovec *) acb->ioh->dxferp, 884 acb->ioh->iovec_count); 885 } 886 } 887 888 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task, 889 iscsi_aio_ioctl_cb, 890 (data.size > 0) ? &data : NULL, 891 acb) != 0) { 892 scsi_free_scsi_task(acb->task); 893 qemu_aio_unref(acb); 894 return NULL; 895 } 896 897 /* tell libiscsi to read straight into the buffer we got from ioctl */ 898 if (acb->task->xfer_dir == SCSI_XFER_READ) { 899 if (acb->ioh->iovec_count == 0) { 900 scsi_task_add_data_in_buffer(acb->task, 901 acb->ioh->dxfer_len, 902 acb->ioh->dxferp); 903 } else { 904 scsi_task_set_iov_in(acb->task, 905 (struct scsi_iovec *) acb->ioh->dxferp, 906 acb->ioh->iovec_count); 907 } 908 } 909 910 iscsi_set_events(iscsilun); 911 912 return &acb->common; 913 } 914 915 #endif 916 917 static int64_t 918 iscsi_getlength(BlockDriverState *bs) 919 { 920 IscsiLun *iscsilun = bs->opaque; 921 int64_t len; 922 923 len = iscsilun->num_blocks; 924 len *= iscsilun->block_size; 925 926 return len; 927 } 928 929 static int 930 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num, 931 int nb_sectors) 932 { 933 IscsiLun *iscsilun = bs->opaque; 934 struct IscsiTask iTask; 935 struct unmap_list list; 936 937 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 938 return -EINVAL; 939 } 940 941 if (!iscsilun->lbp.lbpu) { 942 /* UNMAP is not supported by the target */ 943 return 0; 944 } 945 946 list.lba = sector_qemu2lun(sector_num, iscsilun); 947 list.num = sector_qemu2lun(nb_sectors, iscsilun); 948 949 iscsi_co_init_iscsitask(iscsilun, &iTask); 950 retry: 951 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1, 952 iscsi_co_generic_cb, &iTask) == NULL) { 953 return -ENOMEM; 954 } 955 956 while (!iTask.complete) { 957 iscsi_set_events(iscsilun); 958 qemu_coroutine_yield(); 959 } 960 961 if (iTask.task != NULL) { 962 scsi_free_scsi_task(iTask.task); 963 iTask.task = NULL; 964 } 965 966 if (iTask.do_retry) { 967 iTask.complete = 0; 968 goto retry; 969 } 970 971 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) { 972 /* the target might fail with a check condition if it 973 is not happy with the alignment of the UNMAP request 974 we silently fail in this case */ 975 return 0; 976 } 977 978 if (iTask.status != SCSI_STATUS_GOOD) { 979 return iTask.err_code; 980 } 981 982 iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors); 983 984 return 0; 985 } 986 987 static int 988 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, 989 int count, BdrvRequestFlags flags) 990 { 991 IscsiLun *iscsilun = bs->opaque; 992 struct IscsiTask iTask; 993 uint64_t lba; 994 uint32_t nb_blocks; 995 bool use_16_for_ws = iscsilun->use_16_for_rw; 996 997 if (!is_byte_request_lun_aligned(offset, count, iscsilun)) { 998 return -ENOTSUP; 999 } 1000 1001 if (flags & BDRV_REQ_MAY_UNMAP) { 1002 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) { 1003 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */ 1004 use_16_for_ws = true; 1005 } 1006 if (use_16_for_ws && !iscsilun->lbp.lbpws) { 1007 /* WRITESAME16 with UNMAP is not supported by the target, 1008 * fall back and try WRITESAME10/16 without UNMAP */ 1009 flags &= ~BDRV_REQ_MAY_UNMAP; 1010 use_16_for_ws = iscsilun->use_16_for_rw; 1011 } 1012 } 1013 1014 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) { 1015 /* WRITESAME without UNMAP is not supported by the target */ 1016 return -ENOTSUP; 1017 } 1018 1019 lba = offset / iscsilun->block_size; 1020 nb_blocks = count / iscsilun->block_size; 1021 1022 if (iscsilun->zeroblock == NULL) { 1023 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size); 1024 if (iscsilun->zeroblock == NULL) { 1025 return -ENOMEM; 1026 } 1027 } 1028 1029 iscsi_co_init_iscsitask(iscsilun, &iTask); 1030 retry: 1031 if (use_16_for_ws) { 1032 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba, 1033 iscsilun->zeroblock, iscsilun->block_size, 1034 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 1035 0, 0, iscsi_co_generic_cb, &iTask); 1036 } else { 1037 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba, 1038 iscsilun->zeroblock, iscsilun->block_size, 1039 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 1040 0, 0, iscsi_co_generic_cb, &iTask); 1041 } 1042 if (iTask.task == NULL) { 1043 return -ENOMEM; 1044 } 1045 1046 while (!iTask.complete) { 1047 iscsi_set_events(iscsilun); 1048 qemu_coroutine_yield(); 1049 } 1050 1051 if (iTask.status == SCSI_STATUS_CHECK_CONDITION && 1052 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST && 1053 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE || 1054 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) { 1055 /* WRITE SAME is not supported by the target */ 1056 iscsilun->has_write_same = false; 1057 scsi_free_scsi_task(iTask.task); 1058 return -ENOTSUP; 1059 } 1060 1061 if (iTask.task != NULL) { 1062 scsi_free_scsi_task(iTask.task); 1063 iTask.task = NULL; 1064 } 1065 1066 if (iTask.do_retry) { 1067 iTask.complete = 0; 1068 goto retry; 1069 } 1070 1071 if (iTask.status != SCSI_STATUS_GOOD) { 1072 return iTask.err_code; 1073 } 1074 1075 if (flags & BDRV_REQ_MAY_UNMAP) { 1076 iscsi_allocationmap_clear(iscsilun, offset >> BDRV_SECTOR_BITS, 1077 count >> BDRV_SECTOR_BITS); 1078 } else { 1079 iscsi_allocationmap_set(iscsilun, offset >> BDRV_SECTOR_BITS, 1080 count >> BDRV_SECTOR_BITS); 1081 } 1082 1083 return 0; 1084 } 1085 1086 static void parse_chap(struct iscsi_context *iscsi, const char *target, 1087 Error **errp) 1088 { 1089 QemuOptsList *list; 1090 QemuOpts *opts; 1091 const char *user = NULL; 1092 const char *password = NULL; 1093 const char *secretid; 1094 char *secret = NULL; 1095 1096 list = qemu_find_opts("iscsi"); 1097 if (!list) { 1098 return; 1099 } 1100 1101 opts = qemu_opts_find(list, target); 1102 if (opts == NULL) { 1103 opts = QTAILQ_FIRST(&list->head); 1104 if (!opts) { 1105 return; 1106 } 1107 } 1108 1109 user = qemu_opt_get(opts, "user"); 1110 if (!user) { 1111 return; 1112 } 1113 1114 secretid = qemu_opt_get(opts, "password-secret"); 1115 password = qemu_opt_get(opts, "password"); 1116 if (secretid && password) { 1117 error_setg(errp, "'password' and 'password-secret' properties are " 1118 "mutually exclusive"); 1119 return; 1120 } 1121 if (secretid) { 1122 secret = qcrypto_secret_lookup_as_utf8(secretid, errp); 1123 if (!secret) { 1124 return; 1125 } 1126 password = secret; 1127 } else if (!password) { 1128 error_setg(errp, "CHAP username specified but no password was given"); 1129 return; 1130 } 1131 1132 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) { 1133 error_setg(errp, "Failed to set initiator username and password"); 1134 } 1135 1136 g_free(secret); 1137 } 1138 1139 static void parse_header_digest(struct iscsi_context *iscsi, const char *target, 1140 Error **errp) 1141 { 1142 QemuOptsList *list; 1143 QemuOpts *opts; 1144 const char *digest = NULL; 1145 1146 list = qemu_find_opts("iscsi"); 1147 if (!list) { 1148 return; 1149 } 1150 1151 opts = qemu_opts_find(list, target); 1152 if (opts == NULL) { 1153 opts = QTAILQ_FIRST(&list->head); 1154 if (!opts) { 1155 return; 1156 } 1157 } 1158 1159 digest = qemu_opt_get(opts, "header-digest"); 1160 if (!digest) { 1161 return; 1162 } 1163 1164 if (!strcmp(digest, "CRC32C")) { 1165 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C); 1166 } else if (!strcmp(digest, "NONE")) { 1167 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE); 1168 } else if (!strcmp(digest, "CRC32C-NONE")) { 1169 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE); 1170 } else if (!strcmp(digest, "NONE-CRC32C")) { 1171 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1172 } else { 1173 error_setg(errp, "Invalid header-digest setting : %s", digest); 1174 } 1175 } 1176 1177 static char *parse_initiator_name(const char *target) 1178 { 1179 QemuOptsList *list; 1180 QemuOpts *opts; 1181 const char *name; 1182 char *iscsi_name; 1183 UuidInfo *uuid_info; 1184 1185 list = qemu_find_opts("iscsi"); 1186 if (list) { 1187 opts = qemu_opts_find(list, target); 1188 if (!opts) { 1189 opts = QTAILQ_FIRST(&list->head); 1190 } 1191 if (opts) { 1192 name = qemu_opt_get(opts, "initiator-name"); 1193 if (name) { 1194 return g_strdup(name); 1195 } 1196 } 1197 } 1198 1199 uuid_info = qmp_query_uuid(NULL); 1200 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) { 1201 name = qemu_get_vm_name(); 1202 } else { 1203 name = uuid_info->UUID; 1204 } 1205 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s", 1206 name ? ":" : "", name ? name : ""); 1207 qapi_free_UuidInfo(uuid_info); 1208 return iscsi_name; 1209 } 1210 1211 static int parse_timeout(const char *target) 1212 { 1213 QemuOptsList *list; 1214 QemuOpts *opts; 1215 const char *timeout; 1216 1217 list = qemu_find_opts("iscsi"); 1218 if (list) { 1219 opts = qemu_opts_find(list, target); 1220 if (!opts) { 1221 opts = QTAILQ_FIRST(&list->head); 1222 } 1223 if (opts) { 1224 timeout = qemu_opt_get(opts, "timeout"); 1225 if (timeout) { 1226 return atoi(timeout); 1227 } 1228 } 1229 } 1230 1231 return 0; 1232 } 1233 1234 static void iscsi_nop_timed_event(void *opaque) 1235 { 1236 IscsiLun *iscsilun = opaque; 1237 1238 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) { 1239 error_report("iSCSI: NOP timeout. Reconnecting..."); 1240 iscsilun->request_timed_out = true; 1241 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) { 1242 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages."); 1243 return; 1244 } 1245 1246 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1247 iscsi_set_events(iscsilun); 1248 } 1249 1250 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp) 1251 { 1252 struct scsi_task *task = NULL; 1253 struct scsi_readcapacity10 *rc10 = NULL; 1254 struct scsi_readcapacity16 *rc16 = NULL; 1255 int retries = ISCSI_CMD_RETRIES; 1256 1257 do { 1258 if (task != NULL) { 1259 scsi_free_scsi_task(task); 1260 task = NULL; 1261 } 1262 1263 switch (iscsilun->type) { 1264 case TYPE_DISK: 1265 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun); 1266 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1267 rc16 = scsi_datain_unmarshall(task); 1268 if (rc16 == NULL) { 1269 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data."); 1270 } else { 1271 iscsilun->block_size = rc16->block_length; 1272 iscsilun->num_blocks = rc16->returned_lba + 1; 1273 iscsilun->lbpme = !!rc16->lbpme; 1274 iscsilun->lbprz = !!rc16->lbprz; 1275 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff); 1276 } 1277 break; 1278 } 1279 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION 1280 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) { 1281 break; 1282 } 1283 /* Fall through and try READ CAPACITY(10) instead. */ 1284 case TYPE_ROM: 1285 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0); 1286 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1287 rc10 = scsi_datain_unmarshall(task); 1288 if (rc10 == NULL) { 1289 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data."); 1290 } else { 1291 iscsilun->block_size = rc10->block_size; 1292 if (rc10->lba == 0) { 1293 /* blank disk loaded */ 1294 iscsilun->num_blocks = 0; 1295 } else { 1296 iscsilun->num_blocks = rc10->lba + 1; 1297 } 1298 } 1299 } 1300 break; 1301 default: 1302 return; 1303 } 1304 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION 1305 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION 1306 && retries-- > 0); 1307 1308 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1309 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command"); 1310 } else if (!iscsilun->block_size || 1311 iscsilun->block_size % BDRV_SECTOR_SIZE) { 1312 error_setg(errp, "iSCSI: the target returned an invalid " 1313 "block size of %d.", iscsilun->block_size); 1314 } 1315 if (task) { 1316 scsi_free_scsi_task(task); 1317 } 1318 } 1319 1320 /* TODO Convert to fine grained options */ 1321 static QemuOptsList runtime_opts = { 1322 .name = "iscsi", 1323 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), 1324 .desc = { 1325 { 1326 .name = "filename", 1327 .type = QEMU_OPT_STRING, 1328 .help = "URL to the iscsi image", 1329 }, 1330 { /* end of list */ } 1331 }, 1332 }; 1333 1334 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun, 1335 int evpd, int pc, void **inq, Error **errp) 1336 { 1337 int full_size; 1338 struct scsi_task *task = NULL; 1339 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64); 1340 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1341 goto fail; 1342 } 1343 full_size = scsi_datain_getfullsize(task); 1344 if (full_size > task->datain.size) { 1345 scsi_free_scsi_task(task); 1346 1347 /* we need more data for the full list */ 1348 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size); 1349 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1350 goto fail; 1351 } 1352 } 1353 1354 *inq = scsi_datain_unmarshall(task); 1355 if (*inq == NULL) { 1356 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob"); 1357 goto fail_with_err; 1358 } 1359 1360 return task; 1361 1362 fail: 1363 error_setg(errp, "iSCSI: Inquiry command failed : %s", 1364 iscsi_get_error(iscsi)); 1365 fail_with_err: 1366 if (task != NULL) { 1367 scsi_free_scsi_task(task); 1368 } 1369 return NULL; 1370 } 1371 1372 static void iscsi_detach_aio_context(BlockDriverState *bs) 1373 { 1374 IscsiLun *iscsilun = bs->opaque; 1375 1376 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi), 1377 false, NULL, NULL, NULL); 1378 iscsilun->events = 0; 1379 1380 if (iscsilun->nop_timer) { 1381 timer_del(iscsilun->nop_timer); 1382 timer_free(iscsilun->nop_timer); 1383 iscsilun->nop_timer = NULL; 1384 } 1385 if (iscsilun->event_timer) { 1386 timer_del(iscsilun->event_timer); 1387 timer_free(iscsilun->event_timer); 1388 iscsilun->event_timer = NULL; 1389 } 1390 } 1391 1392 static void iscsi_attach_aio_context(BlockDriverState *bs, 1393 AioContext *new_context) 1394 { 1395 IscsiLun *iscsilun = bs->opaque; 1396 1397 iscsilun->aio_context = new_context; 1398 iscsi_set_events(iscsilun); 1399 1400 /* Set up a timer for sending out iSCSI NOPs */ 1401 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context, 1402 QEMU_CLOCK_REALTIME, SCALE_MS, 1403 iscsi_nop_timed_event, iscsilun); 1404 timer_mod(iscsilun->nop_timer, 1405 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1406 1407 /* Set up a timer for periodic calls to iscsi_set_events and to 1408 * scan for command timeout */ 1409 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context, 1410 QEMU_CLOCK_REALTIME, SCALE_MS, 1411 iscsi_timed_check_events, iscsilun); 1412 timer_mod(iscsilun->event_timer, 1413 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL); 1414 } 1415 1416 static void iscsi_modesense_sync(IscsiLun *iscsilun) 1417 { 1418 struct scsi_task *task; 1419 struct scsi_mode_sense *ms = NULL; 1420 iscsilun->write_protected = false; 1421 iscsilun->dpofua = false; 1422 1423 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun, 1424 1, SCSI_MODESENSE_PC_CURRENT, 1425 0x3F, 0, 255); 1426 if (task == NULL) { 1427 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s", 1428 iscsi_get_error(iscsilun->iscsi)); 1429 goto out; 1430 } 1431 1432 if (task->status != SCSI_STATUS_GOOD) { 1433 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable"); 1434 goto out; 1435 } 1436 ms = scsi_datain_unmarshall(task); 1437 if (!ms) { 1438 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s", 1439 iscsi_get_error(iscsilun->iscsi)); 1440 goto out; 1441 } 1442 iscsilun->write_protected = ms->device_specific_parameter & 0x80; 1443 iscsilun->dpofua = ms->device_specific_parameter & 0x10; 1444 1445 out: 1446 if (task) { 1447 scsi_free_scsi_task(task); 1448 } 1449 } 1450 1451 /* 1452 * We support iscsi url's on the form 1453 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun> 1454 */ 1455 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags, 1456 Error **errp) 1457 { 1458 IscsiLun *iscsilun = bs->opaque; 1459 struct iscsi_context *iscsi = NULL; 1460 struct iscsi_url *iscsi_url = NULL; 1461 struct scsi_task *task = NULL; 1462 struct scsi_inquiry_standard *inq = NULL; 1463 struct scsi_inquiry_supported_pages *inq_vpd; 1464 char *initiator_name = NULL; 1465 QemuOpts *opts; 1466 Error *local_err = NULL; 1467 const char *filename; 1468 int i, ret = 0, timeout = 0; 1469 1470 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); 1471 qemu_opts_absorb_qdict(opts, options, &local_err); 1472 if (local_err) { 1473 error_propagate(errp, local_err); 1474 ret = -EINVAL; 1475 goto out; 1476 } 1477 1478 filename = qemu_opt_get(opts, "filename"); 1479 1480 iscsi_url = iscsi_parse_full_url(iscsi, filename); 1481 if (iscsi_url == NULL) { 1482 error_setg(errp, "Failed to parse URL : %s", filename); 1483 ret = -EINVAL; 1484 goto out; 1485 } 1486 1487 memset(iscsilun, 0, sizeof(IscsiLun)); 1488 1489 initiator_name = parse_initiator_name(iscsi_url->target); 1490 1491 iscsi = iscsi_create_context(initiator_name); 1492 if (iscsi == NULL) { 1493 error_setg(errp, "iSCSI: Failed to create iSCSI context."); 1494 ret = -ENOMEM; 1495 goto out; 1496 } 1497 1498 if (iscsi_set_targetname(iscsi, iscsi_url->target)) { 1499 error_setg(errp, "iSCSI: Failed to set target name."); 1500 ret = -EINVAL; 1501 goto out; 1502 } 1503 1504 if (iscsi_url->user[0] != '\0') { 1505 ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user, 1506 iscsi_url->passwd); 1507 if (ret != 0) { 1508 error_setg(errp, "Failed to set initiator username and password"); 1509 ret = -EINVAL; 1510 goto out; 1511 } 1512 } 1513 1514 /* check if we got CHAP username/password via the options */ 1515 parse_chap(iscsi, iscsi_url->target, &local_err); 1516 if (local_err != NULL) { 1517 error_propagate(errp, local_err); 1518 ret = -EINVAL; 1519 goto out; 1520 } 1521 1522 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) { 1523 error_setg(errp, "iSCSI: Failed to set session type to normal."); 1524 ret = -EINVAL; 1525 goto out; 1526 } 1527 1528 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1529 1530 /* check if we got HEADER_DIGEST via the options */ 1531 parse_header_digest(iscsi, iscsi_url->target, &local_err); 1532 if (local_err != NULL) { 1533 error_propagate(errp, local_err); 1534 ret = -EINVAL; 1535 goto out; 1536 } 1537 1538 /* timeout handling is broken in libiscsi before 1.15.0 */ 1539 timeout = parse_timeout(iscsi_url->target); 1540 #if defined(LIBISCSI_API_VERSION) && LIBISCSI_API_VERSION >= 20150621 1541 iscsi_set_timeout(iscsi, timeout); 1542 #else 1543 if (timeout) { 1544 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0"); 1545 } 1546 #endif 1547 1548 if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) { 1549 error_setg(errp, "iSCSI: Failed to connect to LUN : %s", 1550 iscsi_get_error(iscsi)); 1551 ret = -EINVAL; 1552 goto out; 1553 } 1554 1555 iscsilun->iscsi = iscsi; 1556 iscsilun->aio_context = bdrv_get_aio_context(bs); 1557 iscsilun->lun = iscsi_url->lun; 1558 iscsilun->has_write_same = true; 1559 1560 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0, 1561 (void **) &inq, errp); 1562 if (task == NULL) { 1563 ret = -EINVAL; 1564 goto out; 1565 } 1566 iscsilun->type = inq->periperal_device_type; 1567 scsi_free_scsi_task(task); 1568 task = NULL; 1569 1570 iscsi_modesense_sync(iscsilun); 1571 if (iscsilun->dpofua) { 1572 bs->supported_write_flags = BDRV_REQ_FUA; 1573 } 1574 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP; 1575 1576 /* Check the write protect flag of the LUN if we want to write */ 1577 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) && 1578 iscsilun->write_protected) { 1579 error_setg(errp, "Cannot open a write protected LUN as read-write"); 1580 ret = -EACCES; 1581 goto out; 1582 } 1583 1584 iscsi_readcapacity_sync(iscsilun, &local_err); 1585 if (local_err != NULL) { 1586 error_propagate(errp, local_err); 1587 ret = -EINVAL; 1588 goto out; 1589 } 1590 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun); 1591 bs->request_alignment = iscsilun->block_size; 1592 1593 /* We don't have any emulation for devices other than disks and CD-ROMs, so 1594 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu 1595 * will try to read from the device to guess the image format. 1596 */ 1597 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) { 1598 bs->sg = 1; 1599 } 1600 1601 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1602 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES, 1603 (void **) &inq_vpd, errp); 1604 if (task == NULL) { 1605 ret = -EINVAL; 1606 goto out; 1607 } 1608 for (i = 0; i < inq_vpd->num_pages; i++) { 1609 struct scsi_task *inq_task; 1610 struct scsi_inquiry_logical_block_provisioning *inq_lbp; 1611 struct scsi_inquiry_block_limits *inq_bl; 1612 switch (inq_vpd->pages[i]) { 1613 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING: 1614 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1615 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING, 1616 (void **) &inq_lbp, errp); 1617 if (inq_task == NULL) { 1618 ret = -EINVAL; 1619 goto out; 1620 } 1621 memcpy(&iscsilun->lbp, inq_lbp, 1622 sizeof(struct scsi_inquiry_logical_block_provisioning)); 1623 scsi_free_scsi_task(inq_task); 1624 break; 1625 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS: 1626 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1627 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS, 1628 (void **) &inq_bl, errp); 1629 if (inq_task == NULL) { 1630 ret = -EINVAL; 1631 goto out; 1632 } 1633 memcpy(&iscsilun->bl, inq_bl, 1634 sizeof(struct scsi_inquiry_block_limits)); 1635 scsi_free_scsi_task(inq_task); 1636 break; 1637 default: 1638 break; 1639 } 1640 } 1641 scsi_free_scsi_task(task); 1642 task = NULL; 1643 1644 iscsi_attach_aio_context(bs, iscsilun->aio_context); 1645 1646 /* Guess the internal cluster (page) size of the iscsi target by the means 1647 * of opt_unmap_gran. Transfer the unmap granularity only if it has a 1648 * reasonable size */ 1649 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 && 1650 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) { 1651 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran * 1652 iscsilun->block_size) >> BDRV_SECTOR_BITS; 1653 if (iscsilun->lbprz) { 1654 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun); 1655 if (iscsilun->allocationmap == NULL) { 1656 ret = -ENOMEM; 1657 } 1658 } 1659 } 1660 1661 out: 1662 qemu_opts_del(opts); 1663 g_free(initiator_name); 1664 if (iscsi_url != NULL) { 1665 iscsi_destroy_url(iscsi_url); 1666 } 1667 if (task != NULL) { 1668 scsi_free_scsi_task(task); 1669 } 1670 1671 if (ret) { 1672 if (iscsi != NULL) { 1673 if (iscsi_is_logged_in(iscsi)) { 1674 iscsi_logout_sync(iscsi); 1675 } 1676 iscsi_destroy_context(iscsi); 1677 } 1678 memset(iscsilun, 0, sizeof(IscsiLun)); 1679 } 1680 return ret; 1681 } 1682 1683 static void iscsi_close(BlockDriverState *bs) 1684 { 1685 IscsiLun *iscsilun = bs->opaque; 1686 struct iscsi_context *iscsi = iscsilun->iscsi; 1687 1688 iscsi_detach_aio_context(bs); 1689 if (iscsi_is_logged_in(iscsi)) { 1690 iscsi_logout_sync(iscsi); 1691 } 1692 iscsi_destroy_context(iscsi); 1693 g_free(iscsilun->zeroblock); 1694 g_free(iscsilun->allocationmap); 1695 memset(iscsilun, 0, sizeof(IscsiLun)); 1696 } 1697 1698 static int sector_limits_lun2qemu(int64_t sector, IscsiLun *iscsilun) 1699 { 1700 return MIN(sector_lun2qemu(sector, iscsilun), INT_MAX / 2 + 1); 1701 } 1702 1703 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp) 1704 { 1705 /* We don't actually refresh here, but just return data queried in 1706 * iscsi_open(): iscsi targets don't change their limits. */ 1707 1708 IscsiLun *iscsilun = bs->opaque; 1709 uint32_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff; 1710 1711 if (iscsilun->bl.max_xfer_len) { 1712 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len); 1713 } 1714 1715 bs->bl.max_transfer_length = sector_limits_lun2qemu(max_xfer_len, iscsilun); 1716 1717 if (iscsilun->lbp.lbpu) { 1718 if (iscsilun->bl.max_unmap < 0xffffffff) { 1719 bs->bl.max_discard = 1720 sector_limits_lun2qemu(iscsilun->bl.max_unmap, iscsilun); 1721 } 1722 bs->bl.discard_alignment = 1723 sector_limits_lun2qemu(iscsilun->bl.opt_unmap_gran, iscsilun); 1724 } else { 1725 bs->bl.discard_alignment = iscsilun->block_size >> BDRV_SECTOR_BITS; 1726 } 1727 1728 if (iscsilun->bl.max_ws_len < 0xffffffff / iscsilun->block_size) { 1729 bs->bl.max_pwrite_zeroes = 1730 iscsilun->bl.max_ws_len * iscsilun->block_size; 1731 } 1732 if (iscsilun->lbp.lbpws) { 1733 bs->bl.pwrite_zeroes_alignment = 1734 iscsilun->bl.opt_unmap_gran * iscsilun->block_size; 1735 } else { 1736 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size; 1737 } 1738 bs->bl.opt_transfer_length = 1739 sector_limits_lun2qemu(iscsilun->bl.opt_xfer_len, iscsilun); 1740 } 1741 1742 /* Note that this will not re-establish a connection with an iSCSI target - it 1743 * is effectively a NOP. */ 1744 static int iscsi_reopen_prepare(BDRVReopenState *state, 1745 BlockReopenQueue *queue, Error **errp) 1746 { 1747 IscsiLun *iscsilun = state->bs->opaque; 1748 1749 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) { 1750 error_setg(errp, "Cannot open a write protected LUN as read-write"); 1751 return -EACCES; 1752 } 1753 return 0; 1754 } 1755 1756 static int iscsi_truncate(BlockDriverState *bs, int64_t offset) 1757 { 1758 IscsiLun *iscsilun = bs->opaque; 1759 Error *local_err = NULL; 1760 1761 if (iscsilun->type != TYPE_DISK) { 1762 return -ENOTSUP; 1763 } 1764 1765 iscsi_readcapacity_sync(iscsilun, &local_err); 1766 if (local_err != NULL) { 1767 error_free(local_err); 1768 return -EIO; 1769 } 1770 1771 if (offset > iscsi_getlength(bs)) { 1772 return -EINVAL; 1773 } 1774 1775 if (iscsilun->allocationmap != NULL) { 1776 g_free(iscsilun->allocationmap); 1777 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun); 1778 } 1779 1780 return 0; 1781 } 1782 1783 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp) 1784 { 1785 int ret = 0; 1786 int64_t total_size = 0; 1787 BlockDriverState *bs; 1788 IscsiLun *iscsilun = NULL; 1789 QDict *bs_options; 1790 1791 bs = bdrv_new(); 1792 1793 /* Read out options */ 1794 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 1795 BDRV_SECTOR_SIZE); 1796 bs->opaque = g_new0(struct IscsiLun, 1); 1797 iscsilun = bs->opaque; 1798 1799 bs_options = qdict_new(); 1800 qdict_put(bs_options, "filename", qstring_from_str(filename)); 1801 ret = iscsi_open(bs, bs_options, 0, NULL); 1802 QDECREF(bs_options); 1803 1804 if (ret != 0) { 1805 goto out; 1806 } 1807 iscsi_detach_aio_context(bs); 1808 if (iscsilun->type != TYPE_DISK) { 1809 ret = -ENODEV; 1810 goto out; 1811 } 1812 if (bs->total_sectors < total_size) { 1813 ret = -ENOSPC; 1814 goto out; 1815 } 1816 1817 ret = 0; 1818 out: 1819 if (iscsilun->iscsi != NULL) { 1820 iscsi_destroy_context(iscsilun->iscsi); 1821 } 1822 g_free(bs->opaque); 1823 bs->opaque = NULL; 1824 bdrv_unref(bs); 1825 return ret; 1826 } 1827 1828 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 1829 { 1830 IscsiLun *iscsilun = bs->opaque; 1831 bdi->unallocated_blocks_are_zero = iscsilun->lbprz; 1832 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws; 1833 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE; 1834 return 0; 1835 } 1836 1837 static QemuOptsList iscsi_create_opts = { 1838 .name = "iscsi-create-opts", 1839 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head), 1840 .desc = { 1841 { 1842 .name = BLOCK_OPT_SIZE, 1843 .type = QEMU_OPT_SIZE, 1844 .help = "Virtual disk size" 1845 }, 1846 { /* end of list */ } 1847 } 1848 }; 1849 1850 static BlockDriver bdrv_iscsi = { 1851 .format_name = "iscsi", 1852 .protocol_name = "iscsi", 1853 1854 .instance_size = sizeof(IscsiLun), 1855 .bdrv_needs_filename = true, 1856 .bdrv_file_open = iscsi_open, 1857 .bdrv_close = iscsi_close, 1858 .bdrv_create = iscsi_create, 1859 .create_opts = &iscsi_create_opts, 1860 .bdrv_reopen_prepare = iscsi_reopen_prepare, 1861 1862 .bdrv_getlength = iscsi_getlength, 1863 .bdrv_get_info = iscsi_get_info, 1864 .bdrv_truncate = iscsi_truncate, 1865 .bdrv_refresh_limits = iscsi_refresh_limits, 1866 1867 .bdrv_co_get_block_status = iscsi_co_get_block_status, 1868 .bdrv_co_discard = iscsi_co_discard, 1869 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes, 1870 .bdrv_co_readv = iscsi_co_readv, 1871 .bdrv_co_writev_flags = iscsi_co_writev_flags, 1872 .bdrv_co_flush_to_disk = iscsi_co_flush, 1873 1874 #ifdef __linux__ 1875 .bdrv_aio_ioctl = iscsi_aio_ioctl, 1876 #endif 1877 1878 .bdrv_detach_aio_context = iscsi_detach_aio_context, 1879 .bdrv_attach_aio_context = iscsi_attach_aio_context, 1880 }; 1881 1882 static QemuOptsList qemu_iscsi_opts = { 1883 .name = "iscsi", 1884 .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head), 1885 .desc = { 1886 { 1887 .name = "user", 1888 .type = QEMU_OPT_STRING, 1889 .help = "username for CHAP authentication to target", 1890 },{ 1891 .name = "password", 1892 .type = QEMU_OPT_STRING, 1893 .help = "password for CHAP authentication to target", 1894 },{ 1895 .name = "password-secret", 1896 .type = QEMU_OPT_STRING, 1897 .help = "ID of the secret providing password for CHAP " 1898 "authentication to target", 1899 },{ 1900 .name = "header-digest", 1901 .type = QEMU_OPT_STRING, 1902 .help = "HeaderDigest setting. " 1903 "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}", 1904 },{ 1905 .name = "initiator-name", 1906 .type = QEMU_OPT_STRING, 1907 .help = "Initiator iqn name to use when connecting", 1908 },{ 1909 .name = "timeout", 1910 .type = QEMU_OPT_NUMBER, 1911 .help = "Request timeout in seconds (default 0 = no timeout)", 1912 }, 1913 { /* end of list */ } 1914 }, 1915 }; 1916 1917 static void iscsi_block_init(void) 1918 { 1919 bdrv_register(&bdrv_iscsi); 1920 qemu_add_opts(&qemu_iscsi_opts); 1921 } 1922 1923 block_init(iscsi_block_init); 1924