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