1 /* 2 * QEMU Block driver for iSCSI images 3 * 4 * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com> 5 * Copyright (c) 2012-2017 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/qdict.h" 38 #include "scsi/constants.h" 39 #include "qemu/iov.h" 40 #include "qemu/module.h" 41 #include "qemu/option.h" 42 #include "qemu/uuid.h" 43 #include "sysemu/replay.h" 44 #include "qapi/error.h" 45 #include "qapi/qapi-commands-machine.h" 46 #include "qapi/qmp/qdict.h" 47 #include "qapi/qmp/qstring.h" 48 #include "crypto/secret.h" 49 #include "scsi/utils.h" 50 #include "trace.h" 51 52 /* Conflict between scsi/utils.h and libiscsi! :( */ 53 #define SCSI_XFER_NONE ISCSI_XFER_NONE 54 #include <iscsi/iscsi.h> 55 #define inline __attribute__((gnu_inline)) /* required for libiscsi v1.9.0 */ 56 #include <iscsi/scsi-lowlevel.h> 57 #undef inline 58 #undef SCSI_XFER_NONE 59 QEMU_BUILD_BUG_ON((int)SCSI_XFER_NONE != (int)ISCSI_XFER_NONE); 60 61 #ifdef __linux__ 62 #include <scsi/sg.h> 63 #endif 64 65 typedef struct IscsiLun { 66 struct iscsi_context *iscsi; 67 AioContext *aio_context; 68 int lun; 69 enum scsi_inquiry_peripheral_device_type type; 70 int block_size; 71 uint64_t num_blocks; 72 int events; 73 QEMUTimer *nop_timer; 74 QEMUTimer *event_timer; 75 QemuMutex mutex; 76 struct scsi_inquiry_logical_block_provisioning lbp; 77 struct scsi_inquiry_block_limits bl; 78 struct scsi_inquiry_device_designator *dd; 79 unsigned char *zeroblock; 80 /* The allocmap tracks which clusters (pages) on the iSCSI target are 81 * allocated and which are not. In case a target returns zeros for 82 * unallocated pages (iscsilun->lprz) we can directly return zeros instead 83 * of reading zeros over the wire if a read request falls within an 84 * unallocated block. As there are 3 possible states we need 2 bitmaps to 85 * track. allocmap_valid keeps track if QEMU's information about a page is 86 * valid. allocmap tracks if a page is allocated or not. In case QEMU has no 87 * valid information about a page the corresponding allocmap entry should be 88 * switched to unallocated as well to force a new lookup of the allocation 89 * status as lookups are generally skipped if a page is suspect to be 90 * allocated. If a iSCSI target is opened with cache.direct = on the 91 * allocmap_valid does not exist turning all cached information invalid so 92 * that a fresh lookup is made for any page even if allocmap entry returns 93 * it's unallocated. */ 94 unsigned long *allocmap; 95 unsigned long *allocmap_valid; 96 long allocmap_size; 97 int cluster_size; 98 bool use_16_for_rw; 99 bool write_protected; 100 bool lbpme; 101 bool lbprz; 102 bool dpofua; 103 bool has_write_same; 104 bool request_timed_out; 105 } IscsiLun; 106 107 typedef struct IscsiTask { 108 int status; 109 int complete; 110 int retries; 111 int do_retry; 112 struct scsi_task *task; 113 Coroutine *co; 114 IscsiLun *iscsilun; 115 QEMUTimer retry_timer; 116 int err_code; 117 char *err_str; 118 } IscsiTask; 119 120 typedef struct IscsiAIOCB { 121 BlockAIOCB common; 122 QEMUBH *bh; 123 IscsiLun *iscsilun; 124 struct scsi_task *task; 125 int status; 126 int64_t sector_num; 127 int nb_sectors; 128 int ret; 129 #ifdef __linux__ 130 sg_io_hdr_t *ioh; 131 #endif 132 bool cancelled; 133 } IscsiAIOCB; 134 135 /* libiscsi uses time_t so its enough to process events every second */ 136 #define EVENT_INTERVAL 1000 137 #define NOP_INTERVAL 5000 138 #define MAX_NOP_FAILURES 3 139 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times) 140 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768}; 141 142 /* this threshold is a trade-off knob to choose between 143 * the potential additional overhead of an extra GET_LBA_STATUS request 144 * vs. unnecessarily reading a lot of zero sectors over the wire. 145 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES 146 * sectors we check the allocation status of the area covered by the 147 * request first if the allocationmap indicates that the area might be 148 * unallocated. */ 149 #define ISCSI_CHECKALLOC_THRES 64 150 151 #ifdef __linux__ 152 153 static void 154 iscsi_bh_cb(void *p) 155 { 156 IscsiAIOCB *acb = p; 157 158 qemu_bh_delete(acb->bh); 159 160 acb->common.cb(acb->common.opaque, acb->status); 161 162 if (acb->task != NULL) { 163 scsi_free_scsi_task(acb->task); 164 acb->task = NULL; 165 } 166 167 qemu_aio_unref(acb); 168 } 169 170 static void 171 iscsi_schedule_bh(IscsiAIOCB *acb) 172 { 173 if (acb->bh) { 174 return; 175 } 176 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb); 177 qemu_bh_schedule(acb->bh); 178 } 179 180 #endif 181 182 static void iscsi_co_generic_bh_cb(void *opaque) 183 { 184 struct IscsiTask *iTask = opaque; 185 186 iTask->complete = 1; 187 aio_co_wake(iTask->co); 188 } 189 190 static void iscsi_retry_timer_expired(void *opaque) 191 { 192 struct IscsiTask *iTask = opaque; 193 iTask->complete = 1; 194 if (iTask->co) { 195 aio_co_wake(iTask->co); 196 } 197 } 198 199 static inline unsigned exp_random(double mean) 200 { 201 return -mean * log((double)rand() / RAND_MAX); 202 } 203 204 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in 205 * libiscsi 1.10.0, together with other constants we need. Use it as 206 * a hint that we have to define them ourselves if needed, to keep the 207 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for 208 * the test because SCSI_STATUS_* is an enum. 209 * 210 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes 211 * an enum, check against the LIBISCSI_API_VERSION macro, which was 212 * introduced in 1.11.0. If it is present, there is no need to define 213 * anything. 214 */ 215 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \ 216 !defined(LIBISCSI_API_VERSION) 217 #define SCSI_STATUS_TASK_SET_FULL 0x28 218 #define SCSI_STATUS_TIMEOUT 0x0f000002 219 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600 220 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00 221 #endif 222 223 #ifndef LIBISCSI_API_VERSION 224 #define LIBISCSI_API_VERSION 20130701 225 #endif 226 227 static int iscsi_translate_sense(struct scsi_sense *sense) 228 { 229 return scsi_sense_to_errno(sense->key, 230 (sense->ascq & 0xFF00) >> 8, 231 sense->ascq & 0xFF); 232 } 233 234 /* Called (via iscsi_service) with QemuMutex held. */ 235 static void 236 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status, 237 void *command_data, void *opaque) 238 { 239 struct IscsiTask *iTask = opaque; 240 struct scsi_task *task = command_data; 241 242 iTask->status = status; 243 iTask->do_retry = 0; 244 iTask->err_code = 0; 245 iTask->task = task; 246 247 if (status != SCSI_STATUS_GOOD) { 248 iTask->err_code = -EIO; 249 if (iTask->retries++ < ISCSI_CMD_RETRIES) { 250 if (status == SCSI_STATUS_BUSY || 251 status == SCSI_STATUS_TIMEOUT || 252 status == SCSI_STATUS_TASK_SET_FULL) { 253 unsigned retry_time = 254 exp_random(iscsi_retry_times[iTask->retries - 1]); 255 if (status == SCSI_STATUS_TIMEOUT) { 256 /* make sure the request is rescheduled AFTER the 257 * reconnect is initiated */ 258 retry_time = EVENT_INTERVAL * 2; 259 iTask->iscsilun->request_timed_out = true; 260 } 261 error_report("iSCSI Busy/TaskSetFull/TimeOut" 262 " (retry #%u in %u ms): %s", 263 iTask->retries, retry_time, 264 iscsi_get_error(iscsi)); 265 aio_timer_init(iTask->iscsilun->aio_context, 266 &iTask->retry_timer, QEMU_CLOCK_REALTIME, 267 SCALE_MS, iscsi_retry_timer_expired, iTask); 268 timer_mod(&iTask->retry_timer, 269 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time); 270 iTask->do_retry = 1; 271 } else if (status == SCSI_STATUS_CHECK_CONDITION) { 272 int error = iscsi_translate_sense(&task->sense); 273 if (error == EAGAIN) { 274 error_report("iSCSI CheckCondition: %s", 275 iscsi_get_error(iscsi)); 276 iTask->do_retry = 1; 277 } else { 278 iTask->err_code = -error; 279 iTask->err_str = g_strdup(iscsi_get_error(iscsi)); 280 } 281 } 282 } 283 } 284 285 if (iTask->co) { 286 replay_bh_schedule_oneshot_event(iTask->iscsilun->aio_context, 287 iscsi_co_generic_bh_cb, iTask); 288 } else { 289 iTask->complete = 1; 290 } 291 } 292 293 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask) 294 { 295 *iTask = (struct IscsiTask) { 296 .co = qemu_coroutine_self(), 297 .iscsilun = iscsilun, 298 }; 299 } 300 301 #ifdef __linux__ 302 303 /* Called (via iscsi_service) with QemuMutex held. */ 304 static void 305 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data, 306 void *private_data) 307 { 308 IscsiAIOCB *acb = private_data; 309 310 /* If the command callback hasn't been called yet, drop the task */ 311 if (!acb->bh) { 312 /* Call iscsi_aio_ioctl_cb() with SCSI_STATUS_CANCELLED */ 313 iscsi_scsi_cancel_task(iscsi, acb->task); 314 } 315 316 qemu_aio_unref(acb); /* acquired in iscsi_aio_cancel() */ 317 } 318 319 static void 320 iscsi_aio_cancel(BlockAIOCB *blockacb) 321 { 322 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb; 323 IscsiLun *iscsilun = acb->iscsilun; 324 325 WITH_QEMU_LOCK_GUARD(&iscsilun->mutex) { 326 327 /* If it was cancelled or completed already, our work is done here */ 328 if (acb->cancelled || acb->status != -EINPROGRESS) { 329 return; 330 } 331 332 acb->cancelled = true; 333 334 qemu_aio_ref(acb); /* released in iscsi_abort_task_cb() */ 335 336 /* send a task mgmt call to the target to cancel the task on the target */ 337 if (iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task, 338 iscsi_abort_task_cb, acb) < 0) { 339 qemu_aio_unref(acb); /* since iscsi_abort_task_cb() won't be called */ 340 } 341 } 342 } 343 344 static const AIOCBInfo iscsi_aiocb_info = { 345 .aiocb_size = sizeof(IscsiAIOCB), 346 .cancel_async = iscsi_aio_cancel, 347 }; 348 349 #endif 350 351 static void iscsi_process_read(void *arg); 352 static void iscsi_process_write(void *arg); 353 354 /* Called with QemuMutex held. */ 355 static void 356 iscsi_set_events(IscsiLun *iscsilun) 357 { 358 struct iscsi_context *iscsi = iscsilun->iscsi; 359 int ev = iscsi_which_events(iscsi); 360 361 if (ev != iscsilun->events) { 362 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi), 363 false, 364 (ev & POLLIN) ? iscsi_process_read : NULL, 365 (ev & POLLOUT) ? iscsi_process_write : NULL, 366 NULL, 367 iscsilun); 368 iscsilun->events = ev; 369 } 370 } 371 372 static void iscsi_timed_check_events(void *opaque) 373 { 374 IscsiLun *iscsilun = opaque; 375 376 WITH_QEMU_LOCK_GUARD(&iscsilun->mutex) { 377 /* check for timed out requests */ 378 iscsi_service(iscsilun->iscsi, 0); 379 380 if (iscsilun->request_timed_out) { 381 iscsilun->request_timed_out = false; 382 iscsi_reconnect(iscsilun->iscsi); 383 } 384 385 /* 386 * newer versions of libiscsi may return zero events. Ensure we are 387 * able to return to service once this situation changes. 388 */ 389 iscsi_set_events(iscsilun); 390 } 391 392 timer_mod(iscsilun->event_timer, 393 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL); 394 } 395 396 static void 397 iscsi_process_read(void *arg) 398 { 399 IscsiLun *iscsilun = arg; 400 struct iscsi_context *iscsi = iscsilun->iscsi; 401 402 qemu_mutex_lock(&iscsilun->mutex); 403 iscsi_service(iscsi, POLLIN); 404 iscsi_set_events(iscsilun); 405 qemu_mutex_unlock(&iscsilun->mutex); 406 } 407 408 static void 409 iscsi_process_write(void *arg) 410 { 411 IscsiLun *iscsilun = arg; 412 struct iscsi_context *iscsi = iscsilun->iscsi; 413 414 qemu_mutex_lock(&iscsilun->mutex); 415 iscsi_service(iscsi, POLLOUT); 416 iscsi_set_events(iscsilun); 417 qemu_mutex_unlock(&iscsilun->mutex); 418 } 419 420 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun) 421 { 422 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE; 423 } 424 425 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun) 426 { 427 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size; 428 } 429 430 static bool is_byte_request_lun_aligned(int64_t offset, int64_t bytes, 431 IscsiLun *iscsilun) 432 { 433 if (offset % iscsilun->block_size || bytes % iscsilun->block_size) { 434 error_report("iSCSI misaligned request: " 435 "iscsilun->block_size %u, offset %" PRIi64 436 ", bytes %" PRIi64, 437 iscsilun->block_size, offset, bytes); 438 return false; 439 } 440 return true; 441 } 442 443 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors, 444 IscsiLun *iscsilun) 445 { 446 assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS); 447 return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS, 448 nb_sectors << BDRV_SECTOR_BITS, 449 iscsilun); 450 } 451 452 static void iscsi_allocmap_free(IscsiLun *iscsilun) 453 { 454 g_free(iscsilun->allocmap); 455 g_free(iscsilun->allocmap_valid); 456 iscsilun->allocmap = NULL; 457 iscsilun->allocmap_valid = NULL; 458 } 459 460 461 static int iscsi_allocmap_init(IscsiLun *iscsilun, int open_flags) 462 { 463 iscsi_allocmap_free(iscsilun); 464 465 assert(iscsilun->cluster_size); 466 iscsilun->allocmap_size = 467 DIV_ROUND_UP(iscsilun->num_blocks * iscsilun->block_size, 468 iscsilun->cluster_size); 469 470 iscsilun->allocmap = bitmap_try_new(iscsilun->allocmap_size); 471 if (!iscsilun->allocmap) { 472 return -ENOMEM; 473 } 474 475 if (open_flags & BDRV_O_NOCACHE) { 476 /* when cache.direct = on all allocmap entries are 477 * treated as invalid to force a relookup of the block 478 * status on every read request */ 479 return 0; 480 } 481 482 iscsilun->allocmap_valid = bitmap_try_new(iscsilun->allocmap_size); 483 if (!iscsilun->allocmap_valid) { 484 /* if we are under memory pressure free the allocmap as well */ 485 iscsi_allocmap_free(iscsilun); 486 return -ENOMEM; 487 } 488 489 return 0; 490 } 491 492 static void 493 iscsi_allocmap_update(IscsiLun *iscsilun, int64_t offset, 494 int64_t bytes, bool allocated, bool valid) 495 { 496 int64_t cl_num_expanded, nb_cls_expanded, cl_num_shrunk, nb_cls_shrunk; 497 498 if (iscsilun->allocmap == NULL) { 499 return; 500 } 501 /* expand to entirely contain all affected clusters */ 502 assert(iscsilun->cluster_size); 503 cl_num_expanded = offset / iscsilun->cluster_size; 504 nb_cls_expanded = DIV_ROUND_UP(offset + bytes, 505 iscsilun->cluster_size) - cl_num_expanded; 506 /* shrink to touch only completely contained clusters */ 507 cl_num_shrunk = DIV_ROUND_UP(offset, iscsilun->cluster_size); 508 nb_cls_shrunk = (offset + bytes) / iscsilun->cluster_size - cl_num_shrunk; 509 if (allocated) { 510 bitmap_set(iscsilun->allocmap, cl_num_expanded, nb_cls_expanded); 511 } else { 512 if (nb_cls_shrunk > 0) { 513 bitmap_clear(iscsilun->allocmap, cl_num_shrunk, nb_cls_shrunk); 514 } 515 } 516 517 if (iscsilun->allocmap_valid == NULL) { 518 return; 519 } 520 if (valid) { 521 if (nb_cls_shrunk > 0) { 522 bitmap_set(iscsilun->allocmap_valid, cl_num_shrunk, nb_cls_shrunk); 523 } 524 } else { 525 bitmap_clear(iscsilun->allocmap_valid, cl_num_expanded, 526 nb_cls_expanded); 527 } 528 } 529 530 static void 531 iscsi_allocmap_set_allocated(IscsiLun *iscsilun, int64_t offset, 532 int64_t bytes) 533 { 534 iscsi_allocmap_update(iscsilun, offset, bytes, true, true); 535 } 536 537 static void 538 iscsi_allocmap_set_unallocated(IscsiLun *iscsilun, int64_t offset, 539 int64_t bytes) 540 { 541 /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update 542 * is ignored, so this will in effect be an iscsi_allocmap_set_invalid. 543 */ 544 iscsi_allocmap_update(iscsilun, offset, bytes, false, true); 545 } 546 547 static void iscsi_allocmap_set_invalid(IscsiLun *iscsilun, int64_t offset, 548 int64_t bytes) 549 { 550 iscsi_allocmap_update(iscsilun, offset, bytes, false, false); 551 } 552 553 static void iscsi_allocmap_invalidate(IscsiLun *iscsilun) 554 { 555 if (iscsilun->allocmap) { 556 bitmap_zero(iscsilun->allocmap, iscsilun->allocmap_size); 557 } 558 if (iscsilun->allocmap_valid) { 559 bitmap_zero(iscsilun->allocmap_valid, iscsilun->allocmap_size); 560 } 561 } 562 563 static inline bool 564 iscsi_allocmap_is_allocated(IscsiLun *iscsilun, int64_t offset, 565 int64_t bytes) 566 { 567 unsigned long size; 568 if (iscsilun->allocmap == NULL) { 569 return true; 570 } 571 assert(iscsilun->cluster_size); 572 size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size); 573 return !(find_next_bit(iscsilun->allocmap, size, 574 offset / iscsilun->cluster_size) == size); 575 } 576 577 static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun, 578 int64_t offset, int64_t bytes) 579 { 580 unsigned long size; 581 if (iscsilun->allocmap_valid == NULL) { 582 return false; 583 } 584 assert(iscsilun->cluster_size); 585 size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size); 586 return (find_next_zero_bit(iscsilun->allocmap_valid, size, 587 offset / iscsilun->cluster_size) == size); 588 } 589 590 static void coroutine_fn iscsi_co_wait_for_task(IscsiTask *iTask, 591 IscsiLun *iscsilun) 592 { 593 while (!iTask->complete) { 594 iscsi_set_events(iscsilun); 595 qemu_mutex_unlock(&iscsilun->mutex); 596 qemu_coroutine_yield(); 597 qemu_mutex_lock(&iscsilun->mutex); 598 } 599 } 600 601 static int coroutine_fn 602 iscsi_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors, 603 QEMUIOVector *iov, int flags) 604 { 605 IscsiLun *iscsilun = bs->opaque; 606 struct IscsiTask iTask; 607 uint64_t lba; 608 uint32_t num_sectors; 609 bool fua = flags & BDRV_REQ_FUA; 610 int r = 0; 611 612 if (fua) { 613 assert(iscsilun->dpofua); 614 } 615 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 616 return -EINVAL; 617 } 618 619 if (bs->bl.max_transfer) { 620 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer); 621 } 622 623 lba = sector_qemu2lun(sector_num, iscsilun); 624 num_sectors = sector_qemu2lun(nb_sectors, iscsilun); 625 iscsi_co_init_iscsitask(iscsilun, &iTask); 626 qemu_mutex_lock(&iscsilun->mutex); 627 retry: 628 if (iscsilun->use_16_for_rw) { 629 #if LIBISCSI_API_VERSION >= (20160603) 630 iTask.task = iscsi_write16_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 631 NULL, num_sectors * iscsilun->block_size, 632 iscsilun->block_size, 0, 0, fua, 0, 0, 633 iscsi_co_generic_cb, &iTask, 634 (struct scsi_iovec *)iov->iov, iov->niov); 635 } else { 636 iTask.task = iscsi_write10_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 637 NULL, num_sectors * iscsilun->block_size, 638 iscsilun->block_size, 0, 0, fua, 0, 0, 639 iscsi_co_generic_cb, &iTask, 640 (struct scsi_iovec *)iov->iov, iov->niov); 641 } 642 #else 643 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba, 644 NULL, num_sectors * iscsilun->block_size, 645 iscsilun->block_size, 0, 0, fua, 0, 0, 646 iscsi_co_generic_cb, &iTask); 647 } else { 648 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba, 649 NULL, num_sectors * iscsilun->block_size, 650 iscsilun->block_size, 0, 0, fua, 0, 0, 651 iscsi_co_generic_cb, &iTask); 652 } 653 #endif 654 if (iTask.task == NULL) { 655 qemu_mutex_unlock(&iscsilun->mutex); 656 return -ENOMEM; 657 } 658 #if LIBISCSI_API_VERSION < (20160603) 659 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov, 660 iov->niov); 661 #endif 662 iscsi_co_wait_for_task(&iTask, iscsilun); 663 664 if (iTask.task != NULL) { 665 scsi_free_scsi_task(iTask.task); 666 iTask.task = NULL; 667 } 668 669 if (iTask.do_retry) { 670 iTask.complete = 0; 671 goto retry; 672 } 673 674 if (iTask.status != SCSI_STATUS_GOOD) { 675 iscsi_allocmap_set_invalid(iscsilun, sector_num * BDRV_SECTOR_SIZE, 676 nb_sectors * BDRV_SECTOR_SIZE); 677 error_report("iSCSI WRITE10/16 failed at lba %" PRIu64 ": %s", lba, 678 iTask.err_str); 679 r = iTask.err_code; 680 goto out_unlock; 681 } 682 683 iscsi_allocmap_set_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE, 684 nb_sectors * BDRV_SECTOR_SIZE); 685 686 out_unlock: 687 qemu_mutex_unlock(&iscsilun->mutex); 688 g_free(iTask.err_str); 689 return r; 690 } 691 692 693 694 static int coroutine_fn iscsi_co_block_status(BlockDriverState *bs, 695 bool want_zero, int64_t offset, 696 int64_t bytes, int64_t *pnum, 697 int64_t *map, 698 BlockDriverState **file) 699 { 700 IscsiLun *iscsilun = bs->opaque; 701 struct scsi_get_lba_status *lbas = NULL; 702 struct scsi_lba_status_descriptor *lbasd = NULL; 703 struct IscsiTask iTask; 704 uint64_t lba, max_bytes; 705 int ret; 706 707 iscsi_co_init_iscsitask(iscsilun, &iTask); 708 709 assert(QEMU_IS_ALIGNED(offset | bytes, iscsilun->block_size)); 710 711 /* default to all sectors allocated */ 712 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 713 if (map) { 714 *map = offset; 715 } 716 *pnum = bytes; 717 718 /* LUN does not support logical block provisioning */ 719 if (!iscsilun->lbpme) { 720 goto out; 721 } 722 723 lba = offset / iscsilun->block_size; 724 max_bytes = (iscsilun->num_blocks - lba) * iscsilun->block_size; 725 726 qemu_mutex_lock(&iscsilun->mutex); 727 retry: 728 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun, 729 lba, 8 + 16, iscsi_co_generic_cb, 730 &iTask) == NULL) { 731 ret = -ENOMEM; 732 goto out_unlock; 733 } 734 iscsi_co_wait_for_task(&iTask, iscsilun); 735 736 if (iTask.do_retry) { 737 if (iTask.task != NULL) { 738 scsi_free_scsi_task(iTask.task); 739 iTask.task = NULL; 740 } 741 iTask.complete = 0; 742 goto retry; 743 } 744 745 if (iTask.status != SCSI_STATUS_GOOD) { 746 /* in case the get_lba_status_callout fails (i.e. 747 * because the device is busy or the cmd is not 748 * supported) we pretend all blocks are allocated 749 * for backwards compatibility */ 750 error_report("iSCSI GET_LBA_STATUS failed at lba %" PRIu64 ": %s", 751 lba, iTask.err_str); 752 goto out_unlock; 753 } 754 755 lbas = scsi_datain_unmarshall(iTask.task); 756 if (lbas == NULL || lbas->num_descriptors == 0) { 757 ret = -EIO; 758 goto out_unlock; 759 } 760 761 lbasd = &lbas->descriptors[0]; 762 763 if (lba != lbasd->lba) { 764 ret = -EIO; 765 goto out_unlock; 766 } 767 768 *pnum = MIN((int64_t) lbasd->num_blocks * iscsilun->block_size, max_bytes); 769 770 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED || 771 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) { 772 ret &= ~BDRV_BLOCK_DATA; 773 if (iscsilun->lbprz) { 774 ret |= BDRV_BLOCK_ZERO; 775 } 776 } 777 778 if (ret & BDRV_BLOCK_ZERO) { 779 iscsi_allocmap_set_unallocated(iscsilun, offset, *pnum); 780 } else { 781 iscsi_allocmap_set_allocated(iscsilun, offset, *pnum); 782 } 783 784 out_unlock: 785 qemu_mutex_unlock(&iscsilun->mutex); 786 g_free(iTask.err_str); 787 out: 788 if (iTask.task != NULL) { 789 scsi_free_scsi_task(iTask.task); 790 } 791 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID && file) { 792 *file = bs; 793 } 794 return ret; 795 } 796 797 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs, 798 int64_t sector_num, int nb_sectors, 799 QEMUIOVector *iov) 800 { 801 IscsiLun *iscsilun = bs->opaque; 802 struct IscsiTask iTask; 803 uint64_t lba; 804 uint32_t num_sectors; 805 int r = 0; 806 807 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 808 return -EINVAL; 809 } 810 811 if (bs->bl.max_transfer) { 812 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer); 813 } 814 815 /* if cache.direct is off and we have a valid entry in our allocation map 816 * we can skip checking the block status and directly return zeroes if 817 * the request falls within an unallocated area */ 818 if (iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE, 819 nb_sectors * BDRV_SECTOR_SIZE) && 820 !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE, 821 nb_sectors * BDRV_SECTOR_SIZE)) { 822 qemu_iovec_memset(iov, 0, 0x00, iov->size); 823 return 0; 824 } 825 826 if (nb_sectors >= ISCSI_CHECKALLOC_THRES && 827 !iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE, 828 nb_sectors * BDRV_SECTOR_SIZE) && 829 !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE, 830 nb_sectors * BDRV_SECTOR_SIZE)) { 831 int64_t pnum; 832 /* check the block status from the beginning of the cluster 833 * containing the start sector */ 834 int64_t head; 835 int ret; 836 837 assert(iscsilun->cluster_size); 838 head = (sector_num * BDRV_SECTOR_SIZE) % iscsilun->cluster_size; 839 ret = iscsi_co_block_status(bs, true, 840 sector_num * BDRV_SECTOR_SIZE - head, 841 BDRV_REQUEST_MAX_BYTES, &pnum, NULL, NULL); 842 if (ret < 0) { 843 return ret; 844 } 845 /* if the whole request falls into an unallocated area we can avoid 846 * reading and directly return zeroes instead */ 847 if (ret & BDRV_BLOCK_ZERO && 848 pnum >= nb_sectors * BDRV_SECTOR_SIZE + head) { 849 qemu_iovec_memset(iov, 0, 0x00, iov->size); 850 return 0; 851 } 852 } 853 854 lba = sector_qemu2lun(sector_num, iscsilun); 855 num_sectors = sector_qemu2lun(nb_sectors, iscsilun); 856 857 iscsi_co_init_iscsitask(iscsilun, &iTask); 858 qemu_mutex_lock(&iscsilun->mutex); 859 retry: 860 if (iscsilun->use_16_for_rw) { 861 #if LIBISCSI_API_VERSION >= (20160603) 862 iTask.task = iscsi_read16_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 863 num_sectors * iscsilun->block_size, 864 iscsilun->block_size, 0, 0, 0, 0, 0, 865 iscsi_co_generic_cb, &iTask, 866 (struct scsi_iovec *)iov->iov, iov->niov); 867 } else { 868 iTask.task = iscsi_read10_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 869 num_sectors * iscsilun->block_size, 870 iscsilun->block_size, 871 0, 0, 0, 0, 0, 872 iscsi_co_generic_cb, &iTask, 873 (struct scsi_iovec *)iov->iov, iov->niov); 874 } 875 #else 876 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba, 877 num_sectors * iscsilun->block_size, 878 iscsilun->block_size, 0, 0, 0, 0, 0, 879 iscsi_co_generic_cb, &iTask); 880 } else { 881 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba, 882 num_sectors * iscsilun->block_size, 883 iscsilun->block_size, 884 0, 0, 0, 0, 0, 885 iscsi_co_generic_cb, &iTask); 886 } 887 #endif 888 if (iTask.task == NULL) { 889 qemu_mutex_unlock(&iscsilun->mutex); 890 return -ENOMEM; 891 } 892 #if LIBISCSI_API_VERSION < (20160603) 893 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov); 894 #endif 895 896 iscsi_co_wait_for_task(&iTask, iscsilun); 897 if (iTask.task != NULL) { 898 scsi_free_scsi_task(iTask.task); 899 iTask.task = NULL; 900 } 901 902 if (iTask.do_retry) { 903 iTask.complete = 0; 904 goto retry; 905 } 906 907 if (iTask.status != SCSI_STATUS_GOOD) { 908 error_report("iSCSI READ10/16 failed at lba %" PRIu64 ": %s", 909 lba, iTask.err_str); 910 r = iTask.err_code; 911 } 912 913 qemu_mutex_unlock(&iscsilun->mutex); 914 g_free(iTask.err_str); 915 return r; 916 } 917 918 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs) 919 { 920 IscsiLun *iscsilun = bs->opaque; 921 struct IscsiTask iTask; 922 int r = 0; 923 924 iscsi_co_init_iscsitask(iscsilun, &iTask); 925 qemu_mutex_lock(&iscsilun->mutex); 926 retry: 927 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0, 928 0, iscsi_co_generic_cb, &iTask) == NULL) { 929 qemu_mutex_unlock(&iscsilun->mutex); 930 return -ENOMEM; 931 } 932 933 iscsi_co_wait_for_task(&iTask, iscsilun); 934 935 if (iTask.task != NULL) { 936 scsi_free_scsi_task(iTask.task); 937 iTask.task = NULL; 938 } 939 940 if (iTask.do_retry) { 941 iTask.complete = 0; 942 goto retry; 943 } 944 945 if (iTask.status != SCSI_STATUS_GOOD) { 946 error_report("iSCSI SYNCHRONIZECACHE10 failed: %s", iTask.err_str); 947 r = iTask.err_code; 948 } 949 950 qemu_mutex_unlock(&iscsilun->mutex); 951 g_free(iTask.err_str); 952 return r; 953 } 954 955 #ifdef __linux__ 956 /* Called (via iscsi_service) with QemuMutex held. */ 957 static void 958 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status, 959 void *command_data, void *opaque) 960 { 961 IscsiAIOCB *acb = opaque; 962 963 if (status == SCSI_STATUS_CANCELLED) { 964 if (!acb->bh) { 965 acb->status = -ECANCELED; 966 iscsi_schedule_bh(acb); 967 } 968 return; 969 } 970 971 acb->status = 0; 972 if (status < 0) { 973 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s", 974 iscsi_get_error(iscsi)); 975 acb->status = -iscsi_translate_sense(&acb->task->sense); 976 } 977 978 acb->ioh->driver_status = 0; 979 acb->ioh->host_status = 0; 980 acb->ioh->resid = 0; 981 acb->ioh->status = status; 982 983 #define SG_ERR_DRIVER_SENSE 0x08 984 985 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) { 986 int ss; 987 988 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE; 989 990 acb->ioh->sb_len_wr = acb->task->datain.size - 2; 991 ss = MIN(acb->ioh->mx_sb_len, acb->ioh->sb_len_wr); 992 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss); 993 } 994 995 iscsi_schedule_bh(acb); 996 } 997 998 static void iscsi_ioctl_bh_completion(void *opaque) 999 { 1000 IscsiAIOCB *acb = opaque; 1001 1002 qemu_bh_delete(acb->bh); 1003 acb->common.cb(acb->common.opaque, acb->ret); 1004 qemu_aio_unref(acb); 1005 } 1006 1007 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf) 1008 { 1009 BlockDriverState *bs = acb->common.bs; 1010 IscsiLun *iscsilun = bs->opaque; 1011 int ret = 0; 1012 1013 switch (req) { 1014 case SG_GET_VERSION_NUM: 1015 *(int *)buf = 30000; 1016 break; 1017 case SG_GET_SCSI_ID: 1018 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type; 1019 break; 1020 default: 1021 ret = -EINVAL; 1022 } 1023 assert(!acb->bh); 1024 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), 1025 iscsi_ioctl_bh_completion, acb); 1026 acb->ret = ret; 1027 qemu_bh_schedule(acb->bh); 1028 } 1029 1030 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs, 1031 unsigned long int req, void *buf, 1032 BlockCompletionFunc *cb, void *opaque) 1033 { 1034 IscsiLun *iscsilun = bs->opaque; 1035 struct iscsi_context *iscsi = iscsilun->iscsi; 1036 struct iscsi_data data; 1037 IscsiAIOCB *acb; 1038 1039 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque); 1040 1041 acb->iscsilun = iscsilun; 1042 acb->bh = NULL; 1043 acb->status = -EINPROGRESS; 1044 acb->ioh = buf; 1045 acb->cancelled = false; 1046 1047 if (req != SG_IO) { 1048 iscsi_ioctl_handle_emulated(acb, req, buf); 1049 return &acb->common; 1050 } 1051 1052 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) { 1053 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)", 1054 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE); 1055 qemu_aio_unref(acb); 1056 return NULL; 1057 } 1058 1059 acb->task = malloc(sizeof(struct scsi_task)); 1060 if (acb->task == NULL) { 1061 error_report("iSCSI: Failed to allocate task for scsi command. %s", 1062 iscsi_get_error(iscsi)); 1063 qemu_aio_unref(acb); 1064 return NULL; 1065 } 1066 memset(acb->task, 0, sizeof(struct scsi_task)); 1067 1068 switch (acb->ioh->dxfer_direction) { 1069 case SG_DXFER_TO_DEV: 1070 acb->task->xfer_dir = SCSI_XFER_WRITE; 1071 break; 1072 case SG_DXFER_FROM_DEV: 1073 acb->task->xfer_dir = SCSI_XFER_READ; 1074 break; 1075 default: 1076 acb->task->xfer_dir = SCSI_XFER_NONE; 1077 break; 1078 } 1079 1080 acb->task->cdb_size = acb->ioh->cmd_len; 1081 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len); 1082 acb->task->expxferlen = acb->ioh->dxfer_len; 1083 1084 data.size = 0; 1085 qemu_mutex_lock(&iscsilun->mutex); 1086 if (acb->task->xfer_dir == SCSI_XFER_WRITE) { 1087 if (acb->ioh->iovec_count == 0) { 1088 data.data = acb->ioh->dxferp; 1089 data.size = acb->ioh->dxfer_len; 1090 } else { 1091 scsi_task_set_iov_out(acb->task, 1092 (struct scsi_iovec *) acb->ioh->dxferp, 1093 acb->ioh->iovec_count); 1094 } 1095 } 1096 1097 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task, 1098 iscsi_aio_ioctl_cb, 1099 (data.size > 0) ? &data : NULL, 1100 acb) != 0) { 1101 qemu_mutex_unlock(&iscsilun->mutex); 1102 scsi_free_scsi_task(acb->task); 1103 qemu_aio_unref(acb); 1104 return NULL; 1105 } 1106 1107 /* tell libiscsi to read straight into the buffer we got from ioctl */ 1108 if (acb->task->xfer_dir == SCSI_XFER_READ) { 1109 if (acb->ioh->iovec_count == 0) { 1110 scsi_task_add_data_in_buffer(acb->task, 1111 acb->ioh->dxfer_len, 1112 acb->ioh->dxferp); 1113 } else { 1114 scsi_task_set_iov_in(acb->task, 1115 (struct scsi_iovec *) acb->ioh->dxferp, 1116 acb->ioh->iovec_count); 1117 } 1118 } 1119 1120 iscsi_set_events(iscsilun); 1121 qemu_mutex_unlock(&iscsilun->mutex); 1122 1123 return &acb->common; 1124 } 1125 1126 #endif 1127 1128 static int64_t 1129 iscsi_getlength(BlockDriverState *bs) 1130 { 1131 IscsiLun *iscsilun = bs->opaque; 1132 int64_t len; 1133 1134 len = iscsilun->num_blocks; 1135 len *= iscsilun->block_size; 1136 1137 return len; 1138 } 1139 1140 static int 1141 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, 1142 int64_t bytes) 1143 { 1144 IscsiLun *iscsilun = bs->opaque; 1145 struct IscsiTask iTask; 1146 struct unmap_list list; 1147 int r = 0; 1148 1149 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) { 1150 return -ENOTSUP; 1151 } 1152 1153 if (!iscsilun->lbp.lbpu) { 1154 /* UNMAP is not supported by the target */ 1155 return 0; 1156 } 1157 1158 /* 1159 * We don't want to overflow list.num which is uint32_t. 1160 * We rely on our max_pdiscard. 1161 */ 1162 assert(bytes / iscsilun->block_size <= UINT32_MAX); 1163 1164 list.lba = offset / iscsilun->block_size; 1165 list.num = bytes / iscsilun->block_size; 1166 1167 iscsi_co_init_iscsitask(iscsilun, &iTask); 1168 qemu_mutex_lock(&iscsilun->mutex); 1169 retry: 1170 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1, 1171 iscsi_co_generic_cb, &iTask) == NULL) { 1172 r = -ENOMEM; 1173 goto out_unlock; 1174 } 1175 1176 iscsi_co_wait_for_task(&iTask, iscsilun); 1177 1178 if (iTask.task != NULL) { 1179 scsi_free_scsi_task(iTask.task); 1180 iTask.task = NULL; 1181 } 1182 1183 if (iTask.do_retry) { 1184 iTask.complete = 0; 1185 goto retry; 1186 } 1187 1188 iscsi_allocmap_set_invalid(iscsilun, offset, bytes); 1189 1190 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) { 1191 /* the target might fail with a check condition if it 1192 is not happy with the alignment of the UNMAP request 1193 we silently fail in this case */ 1194 goto out_unlock; 1195 } 1196 1197 if (iTask.status != SCSI_STATUS_GOOD) { 1198 error_report("iSCSI UNMAP failed at lba %" PRIu64 ": %s", 1199 list.lba, iTask.err_str); 1200 r = iTask.err_code; 1201 goto out_unlock; 1202 } 1203 1204 out_unlock: 1205 qemu_mutex_unlock(&iscsilun->mutex); 1206 g_free(iTask.err_str); 1207 return r; 1208 } 1209 1210 static int 1211 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, 1212 int64_t bytes, BdrvRequestFlags flags) 1213 { 1214 IscsiLun *iscsilun = bs->opaque; 1215 struct IscsiTask iTask; 1216 uint64_t lba; 1217 uint64_t nb_blocks; 1218 bool use_16_for_ws = iscsilun->use_16_for_rw; 1219 int r = 0; 1220 1221 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) { 1222 return -ENOTSUP; 1223 } 1224 1225 if (flags & BDRV_REQ_MAY_UNMAP) { 1226 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) { 1227 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */ 1228 use_16_for_ws = true; 1229 } 1230 if (use_16_for_ws && !iscsilun->lbp.lbpws) { 1231 /* WRITESAME16 with UNMAP is not supported by the target, 1232 * fall back and try WRITESAME10/16 without UNMAP */ 1233 flags &= ~BDRV_REQ_MAY_UNMAP; 1234 use_16_for_ws = iscsilun->use_16_for_rw; 1235 } 1236 } 1237 1238 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) { 1239 /* WRITESAME without UNMAP is not supported by the target */ 1240 return -ENOTSUP; 1241 } 1242 1243 lba = offset / iscsilun->block_size; 1244 nb_blocks = bytes / iscsilun->block_size; 1245 1246 if (iscsilun->zeroblock == NULL) { 1247 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size); 1248 if (iscsilun->zeroblock == NULL) { 1249 return -ENOMEM; 1250 } 1251 } 1252 1253 qemu_mutex_lock(&iscsilun->mutex); 1254 iscsi_co_init_iscsitask(iscsilun, &iTask); 1255 retry: 1256 if (use_16_for_ws) { 1257 /* 1258 * iscsi_writesame16_task num_blocks argument is uint32_t. We rely here 1259 * on our max_pwrite_zeroes limit. 1260 */ 1261 assert(nb_blocks <= UINT32_MAX); 1262 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba, 1263 iscsilun->zeroblock, iscsilun->block_size, 1264 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 1265 0, 0, iscsi_co_generic_cb, &iTask); 1266 } else { 1267 /* 1268 * iscsi_writesame10_task num_blocks argument is uint16_t. We rely here 1269 * on our max_pwrite_zeroes limit. 1270 */ 1271 assert(nb_blocks <= UINT16_MAX); 1272 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba, 1273 iscsilun->zeroblock, iscsilun->block_size, 1274 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 1275 0, 0, iscsi_co_generic_cb, &iTask); 1276 } 1277 if (iTask.task == NULL) { 1278 qemu_mutex_unlock(&iscsilun->mutex); 1279 return -ENOMEM; 1280 } 1281 1282 iscsi_co_wait_for_task(&iTask, iscsilun); 1283 1284 if (iTask.status == SCSI_STATUS_CHECK_CONDITION && 1285 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST && 1286 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE || 1287 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) { 1288 /* WRITE SAME is not supported by the target */ 1289 iscsilun->has_write_same = false; 1290 scsi_free_scsi_task(iTask.task); 1291 r = -ENOTSUP; 1292 goto out_unlock; 1293 } 1294 1295 if (iTask.task != NULL) { 1296 scsi_free_scsi_task(iTask.task); 1297 iTask.task = NULL; 1298 } 1299 1300 if (iTask.do_retry) { 1301 iTask.complete = 0; 1302 goto retry; 1303 } 1304 1305 if (iTask.status != SCSI_STATUS_GOOD) { 1306 iscsi_allocmap_set_invalid(iscsilun, offset, bytes); 1307 error_report("iSCSI WRITESAME10/16 failed at lba %" PRIu64 ": %s", 1308 lba, iTask.err_str); 1309 r = iTask.err_code; 1310 goto out_unlock; 1311 } 1312 1313 if (flags & BDRV_REQ_MAY_UNMAP) { 1314 iscsi_allocmap_set_invalid(iscsilun, offset, bytes); 1315 } else { 1316 iscsi_allocmap_set_allocated(iscsilun, offset, bytes); 1317 } 1318 1319 out_unlock: 1320 qemu_mutex_unlock(&iscsilun->mutex); 1321 g_free(iTask.err_str); 1322 return r; 1323 } 1324 1325 static void apply_chap(struct iscsi_context *iscsi, QemuOpts *opts, 1326 Error **errp) 1327 { 1328 const char *user = NULL; 1329 const char *password = NULL; 1330 const char *secretid; 1331 char *secret = NULL; 1332 1333 user = qemu_opt_get(opts, "user"); 1334 if (!user) { 1335 return; 1336 } 1337 1338 secretid = qemu_opt_get(opts, "password-secret"); 1339 password = qemu_opt_get(opts, "password"); 1340 if (secretid && password) { 1341 error_setg(errp, "'password' and 'password-secret' properties are " 1342 "mutually exclusive"); 1343 return; 1344 } 1345 if (secretid) { 1346 secret = qcrypto_secret_lookup_as_utf8(secretid, errp); 1347 if (!secret) { 1348 return; 1349 } 1350 password = secret; 1351 } else if (!password) { 1352 error_setg(errp, "CHAP username specified but no password was given"); 1353 return; 1354 } 1355 1356 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) { 1357 error_setg(errp, "Failed to set initiator username and password"); 1358 } 1359 1360 g_free(secret); 1361 } 1362 1363 static void apply_header_digest(struct iscsi_context *iscsi, QemuOpts *opts, 1364 Error **errp) 1365 { 1366 const char *digest = NULL; 1367 1368 digest = qemu_opt_get(opts, "header-digest"); 1369 if (!digest) { 1370 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1371 } else if (!strcmp(digest, "crc32c")) { 1372 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C); 1373 } else if (!strcmp(digest, "none")) { 1374 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE); 1375 } else if (!strcmp(digest, "crc32c-none")) { 1376 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE); 1377 } else if (!strcmp(digest, "none-crc32c")) { 1378 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1379 } else { 1380 error_setg(errp, "Invalid header-digest setting : %s", digest); 1381 } 1382 } 1383 1384 static char *get_initiator_name(QemuOpts *opts) 1385 { 1386 const char *name; 1387 char *iscsi_name; 1388 UuidInfo *uuid_info; 1389 1390 name = qemu_opt_get(opts, "initiator-name"); 1391 if (name) { 1392 return g_strdup(name); 1393 } 1394 1395 uuid_info = qmp_query_uuid(NULL); 1396 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) { 1397 name = qemu_get_vm_name(); 1398 } else { 1399 name = uuid_info->UUID; 1400 } 1401 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s", 1402 name ? ":" : "", name ? name : ""); 1403 qapi_free_UuidInfo(uuid_info); 1404 return iscsi_name; 1405 } 1406 1407 static void iscsi_nop_timed_event(void *opaque) 1408 { 1409 IscsiLun *iscsilun = opaque; 1410 1411 QEMU_LOCK_GUARD(&iscsilun->mutex); 1412 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) { 1413 error_report("iSCSI: NOP timeout. Reconnecting..."); 1414 iscsilun->request_timed_out = true; 1415 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) { 1416 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages."); 1417 return; 1418 } 1419 1420 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1421 iscsi_set_events(iscsilun); 1422 } 1423 1424 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp) 1425 { 1426 struct scsi_task *task = NULL; 1427 struct scsi_readcapacity10 *rc10 = NULL; 1428 struct scsi_readcapacity16 *rc16 = NULL; 1429 int retries = ISCSI_CMD_RETRIES; 1430 1431 do { 1432 if (task != NULL) { 1433 scsi_free_scsi_task(task); 1434 task = NULL; 1435 } 1436 1437 switch (iscsilun->type) { 1438 case TYPE_DISK: 1439 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun); 1440 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1441 rc16 = scsi_datain_unmarshall(task); 1442 if (rc16 == NULL) { 1443 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data."); 1444 } else { 1445 iscsilun->block_size = rc16->block_length; 1446 iscsilun->num_blocks = rc16->returned_lba + 1; 1447 iscsilun->lbpme = !!rc16->lbpme; 1448 iscsilun->lbprz = !!rc16->lbprz; 1449 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff); 1450 } 1451 break; 1452 } 1453 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION 1454 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) { 1455 break; 1456 } 1457 /* Fall through and try READ CAPACITY(10) instead. */ 1458 case TYPE_ROM: 1459 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0); 1460 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1461 rc10 = scsi_datain_unmarshall(task); 1462 if (rc10 == NULL) { 1463 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data."); 1464 } else { 1465 iscsilun->block_size = rc10->block_size; 1466 if (rc10->lba == 0) { 1467 /* blank disk loaded */ 1468 iscsilun->num_blocks = 0; 1469 } else { 1470 iscsilun->num_blocks = rc10->lba + 1; 1471 } 1472 } 1473 } 1474 break; 1475 default: 1476 return; 1477 } 1478 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION 1479 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION 1480 && retries-- > 0); 1481 1482 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1483 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command"); 1484 } else if (!iscsilun->block_size || 1485 iscsilun->block_size % BDRV_SECTOR_SIZE) { 1486 error_setg(errp, "iSCSI: the target returned an invalid " 1487 "block size of %d.", iscsilun->block_size); 1488 } 1489 if (task) { 1490 scsi_free_scsi_task(task); 1491 } 1492 } 1493 1494 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun, 1495 int evpd, int pc, void **inq, Error **errp) 1496 { 1497 int full_size; 1498 struct scsi_task *task = NULL; 1499 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64); 1500 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1501 goto fail; 1502 } 1503 full_size = scsi_datain_getfullsize(task); 1504 if (full_size > task->datain.size) { 1505 scsi_free_scsi_task(task); 1506 1507 /* we need more data for the full list */ 1508 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size); 1509 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1510 goto fail; 1511 } 1512 } 1513 1514 *inq = scsi_datain_unmarshall(task); 1515 if (*inq == NULL) { 1516 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob"); 1517 goto fail_with_err; 1518 } 1519 1520 return task; 1521 1522 fail: 1523 error_setg(errp, "iSCSI: Inquiry command failed : %s", 1524 iscsi_get_error(iscsi)); 1525 fail_with_err: 1526 if (task != NULL) { 1527 scsi_free_scsi_task(task); 1528 } 1529 return NULL; 1530 } 1531 1532 static void iscsi_detach_aio_context(BlockDriverState *bs) 1533 { 1534 IscsiLun *iscsilun = bs->opaque; 1535 1536 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi), 1537 false, NULL, NULL, NULL, NULL); 1538 iscsilun->events = 0; 1539 1540 if (iscsilun->nop_timer) { 1541 timer_free(iscsilun->nop_timer); 1542 iscsilun->nop_timer = NULL; 1543 } 1544 if (iscsilun->event_timer) { 1545 timer_free(iscsilun->event_timer); 1546 iscsilun->event_timer = NULL; 1547 } 1548 } 1549 1550 static void iscsi_attach_aio_context(BlockDriverState *bs, 1551 AioContext *new_context) 1552 { 1553 IscsiLun *iscsilun = bs->opaque; 1554 1555 iscsilun->aio_context = new_context; 1556 iscsi_set_events(iscsilun); 1557 1558 /* Set up a timer for sending out iSCSI NOPs */ 1559 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context, 1560 QEMU_CLOCK_REALTIME, SCALE_MS, 1561 iscsi_nop_timed_event, iscsilun); 1562 timer_mod(iscsilun->nop_timer, 1563 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1564 1565 /* Set up a timer for periodic calls to iscsi_set_events and to 1566 * scan for command timeout */ 1567 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context, 1568 QEMU_CLOCK_REALTIME, SCALE_MS, 1569 iscsi_timed_check_events, iscsilun); 1570 timer_mod(iscsilun->event_timer, 1571 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL); 1572 } 1573 1574 static void iscsi_modesense_sync(IscsiLun *iscsilun) 1575 { 1576 struct scsi_task *task; 1577 struct scsi_mode_sense *ms = NULL; 1578 iscsilun->write_protected = false; 1579 iscsilun->dpofua = false; 1580 1581 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun, 1582 1, SCSI_MODESENSE_PC_CURRENT, 1583 0x3F, 0, 255); 1584 if (task == NULL) { 1585 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s", 1586 iscsi_get_error(iscsilun->iscsi)); 1587 goto out; 1588 } 1589 1590 if (task->status != SCSI_STATUS_GOOD) { 1591 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable"); 1592 goto out; 1593 } 1594 ms = scsi_datain_unmarshall(task); 1595 if (!ms) { 1596 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s", 1597 iscsi_get_error(iscsilun->iscsi)); 1598 goto out; 1599 } 1600 iscsilun->write_protected = ms->device_specific_parameter & 0x80; 1601 iscsilun->dpofua = ms->device_specific_parameter & 0x10; 1602 1603 out: 1604 if (task) { 1605 scsi_free_scsi_task(task); 1606 } 1607 } 1608 1609 static void iscsi_parse_iscsi_option(const char *target, QDict *options) 1610 { 1611 QemuOptsList *list; 1612 QemuOpts *opts; 1613 const char *user, *password, *password_secret, *initiator_name, 1614 *header_digest, *timeout; 1615 1616 list = qemu_find_opts("iscsi"); 1617 if (!list) { 1618 return; 1619 } 1620 1621 opts = qemu_opts_find(list, target); 1622 if (opts == NULL) { 1623 opts = QTAILQ_FIRST(&list->head); 1624 if (!opts) { 1625 return; 1626 } 1627 } 1628 1629 user = qemu_opt_get(opts, "user"); 1630 if (user) { 1631 qdict_set_default_str(options, "user", user); 1632 } 1633 1634 password = qemu_opt_get(opts, "password"); 1635 if (password) { 1636 qdict_set_default_str(options, "password", password); 1637 } 1638 1639 password_secret = qemu_opt_get(opts, "password-secret"); 1640 if (password_secret) { 1641 qdict_set_default_str(options, "password-secret", password_secret); 1642 } 1643 1644 initiator_name = qemu_opt_get(opts, "initiator-name"); 1645 if (initiator_name) { 1646 qdict_set_default_str(options, "initiator-name", initiator_name); 1647 } 1648 1649 header_digest = qemu_opt_get(opts, "header-digest"); 1650 if (header_digest) { 1651 /* -iscsi takes upper case values, but QAPI only supports lower case 1652 * enum constant names, so we have to convert here. */ 1653 char *qapi_value = g_ascii_strdown(header_digest, -1); 1654 qdict_set_default_str(options, "header-digest", qapi_value); 1655 g_free(qapi_value); 1656 } 1657 1658 timeout = qemu_opt_get(opts, "timeout"); 1659 if (timeout) { 1660 qdict_set_default_str(options, "timeout", timeout); 1661 } 1662 } 1663 1664 /* 1665 * We support iscsi url's on the form 1666 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun> 1667 */ 1668 static void iscsi_parse_filename(const char *filename, QDict *options, 1669 Error **errp) 1670 { 1671 struct iscsi_url *iscsi_url; 1672 const char *transport_name; 1673 char *lun_str; 1674 1675 iscsi_url = iscsi_parse_full_url(NULL, filename); 1676 if (iscsi_url == NULL) { 1677 error_setg(errp, "Failed to parse URL : %s", filename); 1678 return; 1679 } 1680 1681 #if LIBISCSI_API_VERSION >= (20160603) 1682 switch (iscsi_url->transport) { 1683 case TCP_TRANSPORT: 1684 transport_name = "tcp"; 1685 break; 1686 case ISER_TRANSPORT: 1687 transport_name = "iser"; 1688 break; 1689 default: 1690 error_setg(errp, "Unknown transport type (%d)", 1691 iscsi_url->transport); 1692 return; 1693 } 1694 #else 1695 transport_name = "tcp"; 1696 #endif 1697 1698 qdict_set_default_str(options, "transport", transport_name); 1699 qdict_set_default_str(options, "portal", iscsi_url->portal); 1700 qdict_set_default_str(options, "target", iscsi_url->target); 1701 1702 lun_str = g_strdup_printf("%d", iscsi_url->lun); 1703 qdict_set_default_str(options, "lun", lun_str); 1704 g_free(lun_str); 1705 1706 /* User/password from -iscsi take precedence over those from the URL */ 1707 iscsi_parse_iscsi_option(iscsi_url->target, options); 1708 1709 if (iscsi_url->user[0] != '\0') { 1710 qdict_set_default_str(options, "user", iscsi_url->user); 1711 qdict_set_default_str(options, "password", iscsi_url->passwd); 1712 } 1713 1714 iscsi_destroy_url(iscsi_url); 1715 } 1716 1717 static QemuOptsList runtime_opts = { 1718 .name = "iscsi", 1719 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), 1720 .desc = { 1721 { 1722 .name = "transport", 1723 .type = QEMU_OPT_STRING, 1724 }, 1725 { 1726 .name = "portal", 1727 .type = QEMU_OPT_STRING, 1728 }, 1729 { 1730 .name = "target", 1731 .type = QEMU_OPT_STRING, 1732 }, 1733 { 1734 .name = "user", 1735 .type = QEMU_OPT_STRING, 1736 }, 1737 { 1738 .name = "password", 1739 .type = QEMU_OPT_STRING, 1740 }, 1741 { 1742 .name = "password-secret", 1743 .type = QEMU_OPT_STRING, 1744 }, 1745 { 1746 .name = "lun", 1747 .type = QEMU_OPT_NUMBER, 1748 }, 1749 { 1750 .name = "initiator-name", 1751 .type = QEMU_OPT_STRING, 1752 }, 1753 { 1754 .name = "header-digest", 1755 .type = QEMU_OPT_STRING, 1756 }, 1757 { 1758 .name = "timeout", 1759 .type = QEMU_OPT_NUMBER, 1760 }, 1761 { /* end of list */ } 1762 }, 1763 }; 1764 1765 static void iscsi_save_designator(IscsiLun *lun, 1766 struct scsi_inquiry_device_identification *inq_di) 1767 { 1768 struct scsi_inquiry_device_designator *desig, *copy = NULL; 1769 1770 for (desig = inq_di->designators; desig; desig = desig->next) { 1771 if (desig->association || 1772 desig->designator_type > SCSI_DESIGNATOR_TYPE_NAA) { 1773 continue; 1774 } 1775 /* NAA works better than T10 vendor ID based designator. */ 1776 if (!copy || copy->designator_type < desig->designator_type) { 1777 copy = desig; 1778 } 1779 } 1780 if (copy) { 1781 lun->dd = g_new(struct scsi_inquiry_device_designator, 1); 1782 *lun->dd = *copy; 1783 lun->dd->next = NULL; 1784 lun->dd->designator = g_malloc(copy->designator_length); 1785 memcpy(lun->dd->designator, copy->designator, copy->designator_length); 1786 } 1787 } 1788 1789 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags, 1790 Error **errp) 1791 { 1792 IscsiLun *iscsilun = bs->opaque; 1793 struct iscsi_context *iscsi = NULL; 1794 struct scsi_task *task = NULL; 1795 struct scsi_inquiry_standard *inq = NULL; 1796 struct scsi_inquiry_supported_pages *inq_vpd; 1797 char *initiator_name = NULL; 1798 QemuOpts *opts; 1799 Error *local_err = NULL; 1800 const char *transport_name, *portal, *target; 1801 #if LIBISCSI_API_VERSION >= (20160603) 1802 enum iscsi_transport_type transport; 1803 #endif 1804 int i, ret = 0, timeout = 0, lun; 1805 1806 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); 1807 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 1808 ret = -EINVAL; 1809 goto out; 1810 } 1811 1812 transport_name = qemu_opt_get(opts, "transport"); 1813 portal = qemu_opt_get(opts, "portal"); 1814 target = qemu_opt_get(opts, "target"); 1815 lun = qemu_opt_get_number(opts, "lun", 0); 1816 1817 if (!transport_name || !portal || !target) { 1818 error_setg(errp, "Need all of transport, portal and target options"); 1819 ret = -EINVAL; 1820 goto out; 1821 } 1822 1823 if (!strcmp(transport_name, "tcp")) { 1824 #if LIBISCSI_API_VERSION >= (20160603) 1825 transport = TCP_TRANSPORT; 1826 } else if (!strcmp(transport_name, "iser")) { 1827 transport = ISER_TRANSPORT; 1828 #else 1829 /* TCP is what older libiscsi versions always use */ 1830 #endif 1831 } else { 1832 error_setg(errp, "Unknown transport: %s", transport_name); 1833 ret = -EINVAL; 1834 goto out; 1835 } 1836 1837 memset(iscsilun, 0, sizeof(IscsiLun)); 1838 1839 initiator_name = get_initiator_name(opts); 1840 1841 iscsi = iscsi_create_context(initiator_name); 1842 if (iscsi == NULL) { 1843 error_setg(errp, "iSCSI: Failed to create iSCSI context."); 1844 ret = -ENOMEM; 1845 goto out; 1846 } 1847 #if LIBISCSI_API_VERSION >= (20160603) 1848 if (iscsi_init_transport(iscsi, transport)) { 1849 error_setg(errp, ("Error initializing transport.")); 1850 ret = -EINVAL; 1851 goto out; 1852 } 1853 #endif 1854 if (iscsi_set_targetname(iscsi, target)) { 1855 error_setg(errp, "iSCSI: Failed to set target name."); 1856 ret = -EINVAL; 1857 goto out; 1858 } 1859 1860 /* check if we got CHAP username/password via the options */ 1861 apply_chap(iscsi, opts, &local_err); 1862 if (local_err != NULL) { 1863 error_propagate(errp, local_err); 1864 ret = -EINVAL; 1865 goto out; 1866 } 1867 1868 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) { 1869 error_setg(errp, "iSCSI: Failed to set session type to normal."); 1870 ret = -EINVAL; 1871 goto out; 1872 } 1873 1874 /* check if we got HEADER_DIGEST via the options */ 1875 apply_header_digest(iscsi, opts, &local_err); 1876 if (local_err != NULL) { 1877 error_propagate(errp, local_err); 1878 ret = -EINVAL; 1879 goto out; 1880 } 1881 1882 /* timeout handling is broken in libiscsi before 1.15.0 */ 1883 timeout = qemu_opt_get_number(opts, "timeout", 0); 1884 #if LIBISCSI_API_VERSION >= 20150621 1885 iscsi_set_timeout(iscsi, timeout); 1886 #else 1887 if (timeout) { 1888 warn_report("iSCSI: ignoring timeout value for libiscsi <1.15.0"); 1889 } 1890 #endif 1891 1892 if (iscsi_full_connect_sync(iscsi, portal, lun) != 0) { 1893 error_setg(errp, "iSCSI: Failed to connect to LUN : %s", 1894 iscsi_get_error(iscsi)); 1895 ret = -EINVAL; 1896 goto out; 1897 } 1898 1899 iscsilun->iscsi = iscsi; 1900 iscsilun->aio_context = bdrv_get_aio_context(bs); 1901 iscsilun->lun = lun; 1902 iscsilun->has_write_same = true; 1903 1904 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0, 1905 (void **) &inq, errp); 1906 if (task == NULL) { 1907 ret = -EINVAL; 1908 goto out; 1909 } 1910 iscsilun->type = inq->periperal_device_type; 1911 scsi_free_scsi_task(task); 1912 task = NULL; 1913 1914 iscsi_modesense_sync(iscsilun); 1915 if (iscsilun->dpofua) { 1916 bs->supported_write_flags = BDRV_REQ_FUA; 1917 } 1918 1919 /* Check the write protect flag of the LUN if we want to write */ 1920 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) && 1921 iscsilun->write_protected) { 1922 ret = bdrv_apply_auto_read_only(bs, "LUN is write protected", errp); 1923 if (ret < 0) { 1924 goto out; 1925 } 1926 flags &= ~BDRV_O_RDWR; 1927 } 1928 1929 iscsi_readcapacity_sync(iscsilun, &local_err); 1930 if (local_err != NULL) { 1931 error_propagate(errp, local_err); 1932 ret = -EINVAL; 1933 goto out; 1934 } 1935 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun); 1936 1937 /* We don't have any emulation for devices other than disks and CD-ROMs, so 1938 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu 1939 * will try to read from the device to guess the image format. 1940 */ 1941 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) { 1942 bs->sg = true; 1943 } 1944 1945 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1946 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES, 1947 (void **) &inq_vpd, errp); 1948 if (task == NULL) { 1949 ret = -EINVAL; 1950 goto out; 1951 } 1952 for (i = 0; i < inq_vpd->num_pages; i++) { 1953 struct scsi_task *inq_task; 1954 struct scsi_inquiry_logical_block_provisioning *inq_lbp; 1955 struct scsi_inquiry_block_limits *inq_bl; 1956 struct scsi_inquiry_device_identification *inq_di; 1957 switch (inq_vpd->pages[i]) { 1958 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING: 1959 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1960 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING, 1961 (void **) &inq_lbp, errp); 1962 if (inq_task == NULL) { 1963 ret = -EINVAL; 1964 goto out; 1965 } 1966 memcpy(&iscsilun->lbp, inq_lbp, 1967 sizeof(struct scsi_inquiry_logical_block_provisioning)); 1968 scsi_free_scsi_task(inq_task); 1969 break; 1970 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS: 1971 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1972 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS, 1973 (void **) &inq_bl, errp); 1974 if (inq_task == NULL) { 1975 ret = -EINVAL; 1976 goto out; 1977 } 1978 memcpy(&iscsilun->bl, inq_bl, 1979 sizeof(struct scsi_inquiry_block_limits)); 1980 scsi_free_scsi_task(inq_task); 1981 break; 1982 case SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION: 1983 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1984 SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION, 1985 (void **) &inq_di, errp); 1986 if (inq_task == NULL) { 1987 ret = -EINVAL; 1988 goto out; 1989 } 1990 iscsi_save_designator(iscsilun, inq_di); 1991 scsi_free_scsi_task(inq_task); 1992 break; 1993 default: 1994 break; 1995 } 1996 } 1997 scsi_free_scsi_task(task); 1998 task = NULL; 1999 2000 qemu_mutex_init(&iscsilun->mutex); 2001 iscsi_attach_aio_context(bs, iscsilun->aio_context); 2002 2003 /* Guess the internal cluster (page) size of the iscsi target by the means 2004 * of opt_unmap_gran. Transfer the unmap granularity only if it has a 2005 * reasonable size */ 2006 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 && 2007 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) { 2008 iscsilun->cluster_size = iscsilun->bl.opt_unmap_gran * 2009 iscsilun->block_size; 2010 if (iscsilun->lbprz) { 2011 ret = iscsi_allocmap_init(iscsilun, flags); 2012 } 2013 } 2014 2015 if (iscsilun->lbprz && iscsilun->lbp.lbpws) { 2016 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP; 2017 } 2018 2019 out: 2020 qemu_opts_del(opts); 2021 g_free(initiator_name); 2022 if (task != NULL) { 2023 scsi_free_scsi_task(task); 2024 } 2025 2026 if (ret) { 2027 if (iscsi != NULL) { 2028 if (iscsi_is_logged_in(iscsi)) { 2029 iscsi_logout_sync(iscsi); 2030 } 2031 iscsi_destroy_context(iscsi); 2032 } 2033 memset(iscsilun, 0, sizeof(IscsiLun)); 2034 } 2035 2036 return ret; 2037 } 2038 2039 static void iscsi_close(BlockDriverState *bs) 2040 { 2041 IscsiLun *iscsilun = bs->opaque; 2042 struct iscsi_context *iscsi = iscsilun->iscsi; 2043 2044 iscsi_detach_aio_context(bs); 2045 if (iscsi_is_logged_in(iscsi)) { 2046 iscsi_logout_sync(iscsi); 2047 } 2048 iscsi_destroy_context(iscsi); 2049 if (iscsilun->dd) { 2050 g_free(iscsilun->dd->designator); 2051 g_free(iscsilun->dd); 2052 } 2053 g_free(iscsilun->zeroblock); 2054 iscsi_allocmap_free(iscsilun); 2055 qemu_mutex_destroy(&iscsilun->mutex); 2056 memset(iscsilun, 0, sizeof(IscsiLun)); 2057 } 2058 2059 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp) 2060 { 2061 /* We don't actually refresh here, but just return data queried in 2062 * iscsi_open(): iscsi targets don't change their limits. */ 2063 2064 IscsiLun *iscsilun = bs->opaque; 2065 uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff; 2066 unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size); 2067 2068 assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bs->sg); 2069 2070 bs->bl.request_alignment = block_size; 2071 2072 if (iscsilun->bl.max_xfer_len) { 2073 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len); 2074 } 2075 2076 if (max_xfer_len * block_size < INT_MAX) { 2077 bs->bl.max_transfer = max_xfer_len * iscsilun->block_size; 2078 } 2079 2080 if (iscsilun->lbp.lbpu) { 2081 bs->bl.max_pdiscard = 2082 MIN_NON_ZERO(iscsilun->bl.max_unmap * iscsilun->block_size, 2083 (uint64_t)UINT32_MAX * iscsilun->block_size); 2084 bs->bl.pdiscard_alignment = 2085 iscsilun->bl.opt_unmap_gran * iscsilun->block_size; 2086 } else { 2087 bs->bl.pdiscard_alignment = iscsilun->block_size; 2088 } 2089 2090 bs->bl.max_pwrite_zeroes = 2091 MIN_NON_ZERO(iscsilun->bl.max_ws_len * iscsilun->block_size, 2092 max_xfer_len * iscsilun->block_size); 2093 2094 if (iscsilun->lbp.lbpws) { 2095 bs->bl.pwrite_zeroes_alignment = 2096 iscsilun->bl.opt_unmap_gran * iscsilun->block_size; 2097 } else { 2098 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size; 2099 } 2100 if (iscsilun->bl.opt_xfer_len && 2101 iscsilun->bl.opt_xfer_len < INT_MAX / block_size) { 2102 bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len * 2103 iscsilun->block_size); 2104 } 2105 } 2106 2107 /* Note that this will not re-establish a connection with an iSCSI target - it 2108 * is effectively a NOP. */ 2109 static int iscsi_reopen_prepare(BDRVReopenState *state, 2110 BlockReopenQueue *queue, Error **errp) 2111 { 2112 IscsiLun *iscsilun = state->bs->opaque; 2113 2114 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) { 2115 error_setg(errp, "Cannot open a write protected LUN as read-write"); 2116 return -EACCES; 2117 } 2118 return 0; 2119 } 2120 2121 static void iscsi_reopen_commit(BDRVReopenState *reopen_state) 2122 { 2123 IscsiLun *iscsilun = reopen_state->bs->opaque; 2124 2125 /* the cache.direct status might have changed */ 2126 if (iscsilun->allocmap != NULL) { 2127 iscsi_allocmap_init(iscsilun, reopen_state->flags); 2128 } 2129 } 2130 2131 static int coroutine_fn iscsi_co_truncate(BlockDriverState *bs, int64_t offset, 2132 bool exact, PreallocMode prealloc, 2133 BdrvRequestFlags flags, Error **errp) 2134 { 2135 IscsiLun *iscsilun = bs->opaque; 2136 int64_t cur_length; 2137 Error *local_err = NULL; 2138 2139 if (prealloc != PREALLOC_MODE_OFF) { 2140 error_setg(errp, "Unsupported preallocation mode '%s'", 2141 PreallocMode_str(prealloc)); 2142 return -ENOTSUP; 2143 } 2144 2145 if (iscsilun->type != TYPE_DISK) { 2146 error_setg(errp, "Cannot resize non-disk iSCSI devices"); 2147 return -ENOTSUP; 2148 } 2149 2150 iscsi_readcapacity_sync(iscsilun, &local_err); 2151 if (local_err != NULL) { 2152 error_propagate(errp, local_err); 2153 return -EIO; 2154 } 2155 2156 cur_length = iscsi_getlength(bs); 2157 if (offset != cur_length && exact) { 2158 error_setg(errp, "Cannot resize iSCSI devices"); 2159 return -ENOTSUP; 2160 } else if (offset > cur_length) { 2161 error_setg(errp, "Cannot grow iSCSI devices"); 2162 return -EINVAL; 2163 } 2164 2165 if (iscsilun->allocmap != NULL) { 2166 iscsi_allocmap_init(iscsilun, bs->open_flags); 2167 } 2168 2169 return 0; 2170 } 2171 2172 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2173 { 2174 IscsiLun *iscsilun = bs->opaque; 2175 bdi->cluster_size = iscsilun->cluster_size; 2176 return 0; 2177 } 2178 2179 static void coroutine_fn iscsi_co_invalidate_cache(BlockDriverState *bs, 2180 Error **errp) 2181 { 2182 IscsiLun *iscsilun = bs->opaque; 2183 iscsi_allocmap_invalidate(iscsilun); 2184 } 2185 2186 static int coroutine_fn iscsi_co_copy_range_from(BlockDriverState *bs, 2187 BdrvChild *src, 2188 int64_t src_offset, 2189 BdrvChild *dst, 2190 int64_t dst_offset, 2191 int64_t bytes, 2192 BdrvRequestFlags read_flags, 2193 BdrvRequestFlags write_flags) 2194 { 2195 return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, 2196 read_flags, write_flags); 2197 } 2198 2199 static struct scsi_task *iscsi_xcopy_task(int param_len) 2200 { 2201 struct scsi_task *task; 2202 2203 task = g_new0(struct scsi_task, 1); 2204 2205 task->cdb[0] = EXTENDED_COPY; 2206 task->cdb[10] = (param_len >> 24) & 0xFF; 2207 task->cdb[11] = (param_len >> 16) & 0xFF; 2208 task->cdb[12] = (param_len >> 8) & 0xFF; 2209 task->cdb[13] = param_len & 0xFF; 2210 task->cdb_size = 16; 2211 task->xfer_dir = SCSI_XFER_WRITE; 2212 task->expxferlen = param_len; 2213 2214 return task; 2215 } 2216 2217 static void iscsi_populate_target_desc(unsigned char *desc, IscsiLun *lun) 2218 { 2219 struct scsi_inquiry_device_designator *dd = lun->dd; 2220 2221 memset(desc, 0, 32); 2222 desc[0] = 0xE4; /* IDENT_DESCR_TGT_DESCR */ 2223 desc[4] = dd->code_set; 2224 desc[5] = (dd->designator_type & 0xF) 2225 | ((dd->association & 3) << 4); 2226 desc[7] = dd->designator_length; 2227 memcpy(desc + 8, dd->designator, MIN(dd->designator_length, 20)); 2228 2229 desc[28] = 0; 2230 desc[29] = (lun->block_size >> 16) & 0xFF; 2231 desc[30] = (lun->block_size >> 8) & 0xFF; 2232 desc[31] = lun->block_size & 0xFF; 2233 } 2234 2235 static void iscsi_xcopy_desc_hdr(uint8_t *hdr, int dc, int cat, int src_index, 2236 int dst_index) 2237 { 2238 hdr[0] = 0x02; /* BLK_TO_BLK_SEG_DESCR */ 2239 hdr[1] = ((dc << 1) | cat) & 0xFF; 2240 hdr[2] = (XCOPY_BLK2BLK_SEG_DESC_SIZE >> 8) & 0xFF; 2241 /* don't account for the first 4 bytes in descriptor header*/ 2242 hdr[3] = (XCOPY_BLK2BLK_SEG_DESC_SIZE - 4 /* SEG_DESC_SRC_INDEX_OFFSET */) & 0xFF; 2243 hdr[4] = (src_index >> 8) & 0xFF; 2244 hdr[5] = src_index & 0xFF; 2245 hdr[6] = (dst_index >> 8) & 0xFF; 2246 hdr[7] = dst_index & 0xFF; 2247 } 2248 2249 static void iscsi_xcopy_populate_desc(uint8_t *desc, int dc, int cat, 2250 int src_index, int dst_index, int num_blks, 2251 uint64_t src_lba, uint64_t dst_lba) 2252 { 2253 iscsi_xcopy_desc_hdr(desc, dc, cat, src_index, dst_index); 2254 2255 /* The caller should verify the request size */ 2256 assert(num_blks < 65536); 2257 desc[10] = (num_blks >> 8) & 0xFF; 2258 desc[11] = num_blks & 0xFF; 2259 desc[12] = (src_lba >> 56) & 0xFF; 2260 desc[13] = (src_lba >> 48) & 0xFF; 2261 desc[14] = (src_lba >> 40) & 0xFF; 2262 desc[15] = (src_lba >> 32) & 0xFF; 2263 desc[16] = (src_lba >> 24) & 0xFF; 2264 desc[17] = (src_lba >> 16) & 0xFF; 2265 desc[18] = (src_lba >> 8) & 0xFF; 2266 desc[19] = src_lba & 0xFF; 2267 desc[20] = (dst_lba >> 56) & 0xFF; 2268 desc[21] = (dst_lba >> 48) & 0xFF; 2269 desc[22] = (dst_lba >> 40) & 0xFF; 2270 desc[23] = (dst_lba >> 32) & 0xFF; 2271 desc[24] = (dst_lba >> 24) & 0xFF; 2272 desc[25] = (dst_lba >> 16) & 0xFF; 2273 desc[26] = (dst_lba >> 8) & 0xFF; 2274 desc[27] = dst_lba & 0xFF; 2275 } 2276 2277 static void iscsi_xcopy_populate_header(unsigned char *buf, int list_id, int str, 2278 int list_id_usage, int prio, 2279 int tgt_desc_len, 2280 int seg_desc_len, int inline_data_len) 2281 { 2282 buf[0] = list_id; 2283 buf[1] = ((str & 1) << 5) | ((list_id_usage & 3) << 3) | (prio & 7); 2284 buf[2] = (tgt_desc_len >> 8) & 0xFF; 2285 buf[3] = tgt_desc_len & 0xFF; 2286 buf[8] = (seg_desc_len >> 24) & 0xFF; 2287 buf[9] = (seg_desc_len >> 16) & 0xFF; 2288 buf[10] = (seg_desc_len >> 8) & 0xFF; 2289 buf[11] = seg_desc_len & 0xFF; 2290 buf[12] = (inline_data_len >> 24) & 0xFF; 2291 buf[13] = (inline_data_len >> 16) & 0xFF; 2292 buf[14] = (inline_data_len >> 8) & 0xFF; 2293 buf[15] = inline_data_len & 0xFF; 2294 } 2295 2296 static void iscsi_xcopy_data(struct iscsi_data *data, 2297 IscsiLun *src, int64_t src_lba, 2298 IscsiLun *dst, int64_t dst_lba, 2299 uint16_t num_blocks) 2300 { 2301 uint8_t *buf; 2302 const int src_offset = XCOPY_DESC_OFFSET; 2303 const int dst_offset = XCOPY_DESC_OFFSET + IDENT_DESCR_TGT_DESCR_SIZE; 2304 const int seg_offset = dst_offset + IDENT_DESCR_TGT_DESCR_SIZE; 2305 2306 data->size = XCOPY_DESC_OFFSET + 2307 IDENT_DESCR_TGT_DESCR_SIZE * 2 + 2308 XCOPY_BLK2BLK_SEG_DESC_SIZE; 2309 data->data = g_malloc0(data->size); 2310 buf = data->data; 2311 2312 /* Initialise the parameter list header */ 2313 iscsi_xcopy_populate_header(buf, 1, 0, 2 /* LIST_ID_USAGE_DISCARD */, 2314 0, 2 * IDENT_DESCR_TGT_DESCR_SIZE, 2315 XCOPY_BLK2BLK_SEG_DESC_SIZE, 2316 0); 2317 2318 /* Initialise CSCD list with one src + one dst descriptor */ 2319 iscsi_populate_target_desc(&buf[src_offset], src); 2320 iscsi_populate_target_desc(&buf[dst_offset], dst); 2321 2322 /* Initialise one segment descriptor */ 2323 iscsi_xcopy_populate_desc(&buf[seg_offset], 0, 0, 0, 1, num_blocks, 2324 src_lba, dst_lba); 2325 } 2326 2327 static int coroutine_fn iscsi_co_copy_range_to(BlockDriverState *bs, 2328 BdrvChild *src, 2329 int64_t src_offset, 2330 BdrvChild *dst, 2331 int64_t dst_offset, 2332 int64_t bytes, 2333 BdrvRequestFlags read_flags, 2334 BdrvRequestFlags write_flags) 2335 { 2336 IscsiLun *dst_lun = dst->bs->opaque; 2337 IscsiLun *src_lun; 2338 struct IscsiTask iscsi_task; 2339 struct iscsi_data data; 2340 int r = 0; 2341 int block_size; 2342 2343 if (src->bs->drv->bdrv_co_copy_range_to != iscsi_co_copy_range_to) { 2344 return -ENOTSUP; 2345 } 2346 src_lun = src->bs->opaque; 2347 2348 if (!src_lun->dd || !dst_lun->dd) { 2349 return -ENOTSUP; 2350 } 2351 if (!is_byte_request_lun_aligned(dst_offset, bytes, dst_lun)) { 2352 return -ENOTSUP; 2353 } 2354 if (!is_byte_request_lun_aligned(src_offset, bytes, src_lun)) { 2355 return -ENOTSUP; 2356 } 2357 if (dst_lun->block_size != src_lun->block_size || 2358 !dst_lun->block_size) { 2359 return -ENOTSUP; 2360 } 2361 2362 block_size = dst_lun->block_size; 2363 if (bytes / block_size > 65535) { 2364 return -ENOTSUP; 2365 } 2366 2367 iscsi_xcopy_data(&data, 2368 src_lun, src_offset / block_size, 2369 dst_lun, dst_offset / block_size, 2370 bytes / block_size); 2371 2372 iscsi_co_init_iscsitask(dst_lun, &iscsi_task); 2373 2374 qemu_mutex_lock(&dst_lun->mutex); 2375 iscsi_task.task = iscsi_xcopy_task(data.size); 2376 retry: 2377 if (iscsi_scsi_command_async(dst_lun->iscsi, dst_lun->lun, 2378 iscsi_task.task, iscsi_co_generic_cb, 2379 &data, 2380 &iscsi_task) != 0) { 2381 r = -EIO; 2382 goto out_unlock; 2383 } 2384 2385 iscsi_co_wait_for_task(&iscsi_task, dst_lun); 2386 2387 if (iscsi_task.do_retry) { 2388 iscsi_task.complete = 0; 2389 goto retry; 2390 } 2391 2392 if (iscsi_task.status != SCSI_STATUS_GOOD) { 2393 r = iscsi_task.err_code; 2394 goto out_unlock; 2395 } 2396 2397 out_unlock: 2398 2399 trace_iscsi_xcopy(src_lun, src_offset, dst_lun, dst_offset, bytes, r); 2400 g_free(iscsi_task.task); 2401 qemu_mutex_unlock(&dst_lun->mutex); 2402 g_free(iscsi_task.err_str); 2403 return r; 2404 } 2405 2406 2407 static const char *const iscsi_strong_runtime_opts[] = { 2408 "transport", 2409 "portal", 2410 "target", 2411 "user", 2412 "password", 2413 "password-secret", 2414 "lun", 2415 "initiator-name", 2416 "header-digest", 2417 2418 NULL 2419 }; 2420 2421 static BlockDriver bdrv_iscsi = { 2422 .format_name = "iscsi", 2423 .protocol_name = "iscsi", 2424 2425 .instance_size = sizeof(IscsiLun), 2426 .bdrv_parse_filename = iscsi_parse_filename, 2427 .bdrv_file_open = iscsi_open, 2428 .bdrv_close = iscsi_close, 2429 .bdrv_co_create_opts = bdrv_co_create_opts_simple, 2430 .create_opts = &bdrv_create_opts_simple, 2431 .bdrv_reopen_prepare = iscsi_reopen_prepare, 2432 .bdrv_reopen_commit = iscsi_reopen_commit, 2433 .bdrv_co_invalidate_cache = iscsi_co_invalidate_cache, 2434 2435 .bdrv_getlength = iscsi_getlength, 2436 .bdrv_get_info = iscsi_get_info, 2437 .bdrv_co_truncate = iscsi_co_truncate, 2438 .bdrv_refresh_limits = iscsi_refresh_limits, 2439 2440 .bdrv_co_block_status = iscsi_co_block_status, 2441 .bdrv_co_pdiscard = iscsi_co_pdiscard, 2442 .bdrv_co_copy_range_from = iscsi_co_copy_range_from, 2443 .bdrv_co_copy_range_to = iscsi_co_copy_range_to, 2444 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes, 2445 .bdrv_co_readv = iscsi_co_readv, 2446 .bdrv_co_writev = iscsi_co_writev, 2447 .bdrv_co_flush_to_disk = iscsi_co_flush, 2448 2449 #ifdef __linux__ 2450 .bdrv_aio_ioctl = iscsi_aio_ioctl, 2451 #endif 2452 2453 .bdrv_detach_aio_context = iscsi_detach_aio_context, 2454 .bdrv_attach_aio_context = iscsi_attach_aio_context, 2455 2456 .strong_runtime_opts = iscsi_strong_runtime_opts, 2457 }; 2458 2459 #if LIBISCSI_API_VERSION >= (20160603) 2460 static BlockDriver bdrv_iser = { 2461 .format_name = "iser", 2462 .protocol_name = "iser", 2463 2464 .instance_size = sizeof(IscsiLun), 2465 .bdrv_parse_filename = iscsi_parse_filename, 2466 .bdrv_file_open = iscsi_open, 2467 .bdrv_close = iscsi_close, 2468 .bdrv_co_create_opts = bdrv_co_create_opts_simple, 2469 .create_opts = &bdrv_create_opts_simple, 2470 .bdrv_reopen_prepare = iscsi_reopen_prepare, 2471 .bdrv_reopen_commit = iscsi_reopen_commit, 2472 .bdrv_co_invalidate_cache = iscsi_co_invalidate_cache, 2473 2474 .bdrv_getlength = iscsi_getlength, 2475 .bdrv_get_info = iscsi_get_info, 2476 .bdrv_co_truncate = iscsi_co_truncate, 2477 .bdrv_refresh_limits = iscsi_refresh_limits, 2478 2479 .bdrv_co_block_status = iscsi_co_block_status, 2480 .bdrv_co_pdiscard = iscsi_co_pdiscard, 2481 .bdrv_co_copy_range_from = iscsi_co_copy_range_from, 2482 .bdrv_co_copy_range_to = iscsi_co_copy_range_to, 2483 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes, 2484 .bdrv_co_readv = iscsi_co_readv, 2485 .bdrv_co_writev = iscsi_co_writev, 2486 .bdrv_co_flush_to_disk = iscsi_co_flush, 2487 2488 #ifdef __linux__ 2489 .bdrv_aio_ioctl = iscsi_aio_ioctl, 2490 #endif 2491 2492 .bdrv_detach_aio_context = iscsi_detach_aio_context, 2493 .bdrv_attach_aio_context = iscsi_attach_aio_context, 2494 2495 .strong_runtime_opts = iscsi_strong_runtime_opts, 2496 }; 2497 #endif 2498 2499 static void iscsi_block_init(void) 2500 { 2501 bdrv_register(&bdrv_iscsi); 2502 #if LIBISCSI_API_VERSION >= (20160603) 2503 bdrv_register(&bdrv_iser); 2504 #endif 2505 } 2506 2507 block_init(iscsi_block_init); 2508