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