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/config-file.h" 32 #include "qemu/error-report.h" 33 #include "qemu/bitops.h" 34 #include "qemu/bitmap.h" 35 #include "block/block_int.h" 36 #include "scsi/constants.h" 37 #include "qemu/iov.h" 38 #include "qemu/option.h" 39 #include "qemu/uuid.h" 40 #include "qmp-commands.h" 41 #include "qapi/error.h" 42 #include "qapi/qmp/qdict.h" 43 #include "qapi/qmp/qstring.h" 44 #include "crypto/secret.h" 45 #include "scsi/utils.h" 46 47 /* Conflict between scsi/utils.h and libiscsi! :( */ 48 #define SCSI_XFER_NONE ISCSI_XFER_NONE 49 #include <iscsi/iscsi.h> 50 #include <iscsi/scsi-lowlevel.h> 51 #undef SCSI_XFER_NONE 52 QEMU_BUILD_BUG_ON((int)SCSI_XFER_NONE != (int)ISCSI_XFER_NONE); 53 54 #ifdef __linux__ 55 #include <scsi/sg.h> 56 #endif 57 58 typedef struct IscsiLun { 59 struct iscsi_context *iscsi; 60 AioContext *aio_context; 61 int lun; 62 enum scsi_inquiry_peripheral_device_type type; 63 int block_size; 64 uint64_t num_blocks; 65 int events; 66 QEMUTimer *nop_timer; 67 QEMUTimer *event_timer; 68 QemuMutex mutex; 69 struct scsi_inquiry_logical_block_provisioning lbp; 70 struct scsi_inquiry_block_limits bl; 71 unsigned char *zeroblock; 72 /* The allocmap tracks which clusters (pages) on the iSCSI target are 73 * allocated and which are not. In case a target returns zeros for 74 * unallocated pages (iscsilun->lprz) we can directly return zeros instead 75 * of reading zeros over the wire if a read request falls within an 76 * unallocated block. As there are 3 possible states we need 2 bitmaps to 77 * track. allocmap_valid keeps track if QEMU's information about a page is 78 * valid. allocmap tracks if a page is allocated or not. In case QEMU has no 79 * valid information about a page the corresponding allocmap entry should be 80 * switched to unallocated as well to force a new lookup of the allocation 81 * status as lookups are generally skipped if a page is suspect to be 82 * allocated. If a iSCSI target is opened with cache.direct = on the 83 * allocmap_valid does not exist turning all cached information invalid so 84 * that a fresh lookup is made for any page even if allocmap entry returns 85 * it's unallocated. */ 86 unsigned long *allocmap; 87 unsigned long *allocmap_valid; 88 long allocmap_size; 89 int cluster_sectors; 90 bool use_16_for_rw; 91 bool write_protected; 92 bool lbpme; 93 bool lbprz; 94 bool dpofua; 95 bool has_write_same; 96 bool request_timed_out; 97 } IscsiLun; 98 99 typedef struct IscsiTask { 100 int status; 101 int complete; 102 int retries; 103 int do_retry; 104 struct scsi_task *task; 105 Coroutine *co; 106 IscsiLun *iscsilun; 107 QEMUTimer retry_timer; 108 int err_code; 109 char *err_str; 110 } IscsiTask; 111 112 typedef struct IscsiAIOCB { 113 BlockAIOCB common; 114 QEMUBH *bh; 115 IscsiLun *iscsilun; 116 struct scsi_task *task; 117 uint8_t *buf; 118 int status; 119 int64_t sector_num; 120 int nb_sectors; 121 int ret; 122 #ifdef __linux__ 123 sg_io_hdr_t *ioh; 124 #endif 125 } IscsiAIOCB; 126 127 /* libiscsi uses time_t so its enough to process events every second */ 128 #define EVENT_INTERVAL 1000 129 #define NOP_INTERVAL 5000 130 #define MAX_NOP_FAILURES 3 131 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times) 132 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768}; 133 134 /* this threshold is a trade-off knob to choose between 135 * the potential additional overhead of an extra GET_LBA_STATUS request 136 * vs. unnecessarily reading a lot of zero sectors over the wire. 137 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES 138 * sectors we check the allocation status of the area covered by the 139 * request first if the allocationmap indicates that the area might be 140 * unallocated. */ 141 #define ISCSI_CHECKALLOC_THRES 64 142 143 static void 144 iscsi_bh_cb(void *p) 145 { 146 IscsiAIOCB *acb = p; 147 148 qemu_bh_delete(acb->bh); 149 150 g_free(acb->buf); 151 acb->buf = NULL; 152 153 acb->common.cb(acb->common.opaque, acb->status); 154 155 if (acb->task != NULL) { 156 scsi_free_scsi_task(acb->task); 157 acb->task = NULL; 158 } 159 160 qemu_aio_unref(acb); 161 } 162 163 static void 164 iscsi_schedule_bh(IscsiAIOCB *acb) 165 { 166 if (acb->bh) { 167 return; 168 } 169 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb); 170 qemu_bh_schedule(acb->bh); 171 } 172 173 static void iscsi_co_generic_bh_cb(void *opaque) 174 { 175 struct IscsiTask *iTask = opaque; 176 177 iTask->complete = 1; 178 aio_co_wake(iTask->co); 179 } 180 181 static void iscsi_retry_timer_expired(void *opaque) 182 { 183 struct IscsiTask *iTask = opaque; 184 iTask->complete = 1; 185 if (iTask->co) { 186 aio_co_wake(iTask->co); 187 } 188 } 189 190 static inline unsigned exp_random(double mean) 191 { 192 return -mean * log((double)rand() / RAND_MAX); 193 } 194 195 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in 196 * libiscsi 1.10.0, together with other constants we need. Use it as 197 * a hint that we have to define them ourselves if needed, to keep the 198 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for 199 * the test because SCSI_STATUS_* is an enum. 200 * 201 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes 202 * an enum, check against the LIBISCSI_API_VERSION macro, which was 203 * introduced in 1.11.0. If it is present, there is no need to define 204 * anything. 205 */ 206 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \ 207 !defined(LIBISCSI_API_VERSION) 208 #define SCSI_STATUS_TASK_SET_FULL 0x28 209 #define SCSI_STATUS_TIMEOUT 0x0f000002 210 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600 211 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00 212 #endif 213 214 #ifndef LIBISCSI_API_VERSION 215 #define LIBISCSI_API_VERSION 20130701 216 #endif 217 218 static int iscsi_translate_sense(struct scsi_sense *sense) 219 { 220 return - scsi_sense_to_errno(sense->key, 221 (sense->ascq & 0xFF00) >> 8, 222 sense->ascq & 0xFF); 223 } 224 225 /* Called (via iscsi_service) with QemuMutex held. */ 226 static void 227 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status, 228 void *command_data, void *opaque) 229 { 230 struct IscsiTask *iTask = opaque; 231 struct scsi_task *task = command_data; 232 233 iTask->status = status; 234 iTask->do_retry = 0; 235 iTask->task = task; 236 237 if (status != SCSI_STATUS_GOOD) { 238 if (iTask->retries++ < ISCSI_CMD_RETRIES) { 239 if (status == SCSI_STATUS_CHECK_CONDITION 240 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) { 241 error_report("iSCSI CheckCondition: %s", 242 iscsi_get_error(iscsi)); 243 iTask->do_retry = 1; 244 goto out; 245 } 246 if (status == SCSI_STATUS_BUSY || 247 status == SCSI_STATUS_TIMEOUT || 248 status == SCSI_STATUS_TASK_SET_FULL) { 249 unsigned retry_time = 250 exp_random(iscsi_retry_times[iTask->retries - 1]); 251 if (status == SCSI_STATUS_TIMEOUT) { 252 /* make sure the request is rescheduled AFTER the 253 * reconnect is initiated */ 254 retry_time = EVENT_INTERVAL * 2; 255 iTask->iscsilun->request_timed_out = true; 256 } 257 error_report("iSCSI Busy/TaskSetFull/TimeOut" 258 " (retry #%u in %u ms): %s", 259 iTask->retries, retry_time, 260 iscsi_get_error(iscsi)); 261 aio_timer_init(iTask->iscsilun->aio_context, 262 &iTask->retry_timer, QEMU_CLOCK_REALTIME, 263 SCALE_MS, iscsi_retry_timer_expired, iTask); 264 timer_mod(&iTask->retry_timer, 265 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time); 266 iTask->do_retry = 1; 267 return; 268 } 269 } 270 iTask->err_code = iscsi_translate_sense(&task->sense); 271 iTask->err_str = g_strdup(iscsi_get_error(iscsi)); 272 } 273 274 out: 275 if (iTask->co) { 276 aio_bh_schedule_oneshot(iTask->iscsilun->aio_context, 277 iscsi_co_generic_bh_cb, iTask); 278 } else { 279 iTask->complete = 1; 280 } 281 } 282 283 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask) 284 { 285 *iTask = (struct IscsiTask) { 286 .co = qemu_coroutine_self(), 287 .iscsilun = iscsilun, 288 }; 289 } 290 291 static void 292 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data, 293 void *private_data) 294 { 295 IscsiAIOCB *acb = private_data; 296 297 acb->status = -ECANCELED; 298 iscsi_schedule_bh(acb); 299 } 300 301 static void 302 iscsi_aio_cancel(BlockAIOCB *blockacb) 303 { 304 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb; 305 IscsiLun *iscsilun = acb->iscsilun; 306 307 if (acb->status != -EINPROGRESS) { 308 return; 309 } 310 311 /* send a task mgmt call to the target to cancel the task on the target */ 312 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task, 313 iscsi_abort_task_cb, acb); 314 315 } 316 317 static const AIOCBInfo iscsi_aiocb_info = { 318 .aiocb_size = sizeof(IscsiAIOCB), 319 .cancel_async = iscsi_aio_cancel, 320 }; 321 322 323 static void iscsi_process_read(void *arg); 324 static void iscsi_process_write(void *arg); 325 326 /* Called with QemuMutex held. */ 327 static void 328 iscsi_set_events(IscsiLun *iscsilun) 329 { 330 struct iscsi_context *iscsi = iscsilun->iscsi; 331 int ev = iscsi_which_events(iscsi); 332 333 if (ev != iscsilun->events) { 334 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi), 335 false, 336 (ev & POLLIN) ? iscsi_process_read : NULL, 337 (ev & POLLOUT) ? iscsi_process_write : NULL, 338 NULL, 339 iscsilun); 340 iscsilun->events = ev; 341 } 342 } 343 344 static void iscsi_timed_check_events(void *opaque) 345 { 346 IscsiLun *iscsilun = opaque; 347 348 /* check for timed out requests */ 349 iscsi_service(iscsilun->iscsi, 0); 350 351 if (iscsilun->request_timed_out) { 352 iscsilun->request_timed_out = false; 353 iscsi_reconnect(iscsilun->iscsi); 354 } 355 356 /* newer versions of libiscsi may return zero events. Ensure we are able 357 * to return to service once this situation changes. */ 358 iscsi_set_events(iscsilun); 359 360 timer_mod(iscsilun->event_timer, 361 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL); 362 } 363 364 static void 365 iscsi_process_read(void *arg) 366 { 367 IscsiLun *iscsilun = arg; 368 struct iscsi_context *iscsi = iscsilun->iscsi; 369 370 qemu_mutex_lock(&iscsilun->mutex); 371 iscsi_service(iscsi, POLLIN); 372 iscsi_set_events(iscsilun); 373 qemu_mutex_unlock(&iscsilun->mutex); 374 } 375 376 static void 377 iscsi_process_write(void *arg) 378 { 379 IscsiLun *iscsilun = arg; 380 struct iscsi_context *iscsi = iscsilun->iscsi; 381 382 qemu_mutex_lock(&iscsilun->mutex); 383 iscsi_service(iscsi, POLLOUT); 384 iscsi_set_events(iscsilun); 385 qemu_mutex_unlock(&iscsilun->mutex); 386 } 387 388 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun) 389 { 390 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE; 391 } 392 393 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun) 394 { 395 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size; 396 } 397 398 static bool is_byte_request_lun_aligned(int64_t offset, int count, 399 IscsiLun *iscsilun) 400 { 401 if (offset % iscsilun->block_size || count % iscsilun->block_size) { 402 error_report("iSCSI misaligned request: " 403 "iscsilun->block_size %u, offset %" PRIi64 404 ", count %d", 405 iscsilun->block_size, offset, count); 406 return false; 407 } 408 return true; 409 } 410 411 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors, 412 IscsiLun *iscsilun) 413 { 414 assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS); 415 return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS, 416 nb_sectors << BDRV_SECTOR_BITS, 417 iscsilun); 418 } 419 420 static void iscsi_allocmap_free(IscsiLun *iscsilun) 421 { 422 g_free(iscsilun->allocmap); 423 g_free(iscsilun->allocmap_valid); 424 iscsilun->allocmap = NULL; 425 iscsilun->allocmap_valid = NULL; 426 } 427 428 429 static int iscsi_allocmap_init(IscsiLun *iscsilun, int open_flags) 430 { 431 iscsi_allocmap_free(iscsilun); 432 433 iscsilun->allocmap_size = 434 DIV_ROUND_UP(sector_lun2qemu(iscsilun->num_blocks, iscsilun), 435 iscsilun->cluster_sectors); 436 437 iscsilun->allocmap = bitmap_try_new(iscsilun->allocmap_size); 438 if (!iscsilun->allocmap) { 439 return -ENOMEM; 440 } 441 442 if (open_flags & BDRV_O_NOCACHE) { 443 /* in case that cache.direct = on all allocmap entries are 444 * treated as invalid to force a relookup of the block 445 * status on every read request */ 446 return 0; 447 } 448 449 iscsilun->allocmap_valid = bitmap_try_new(iscsilun->allocmap_size); 450 if (!iscsilun->allocmap_valid) { 451 /* if we are under memory pressure free the allocmap as well */ 452 iscsi_allocmap_free(iscsilun); 453 return -ENOMEM; 454 } 455 456 return 0; 457 } 458 459 static void 460 iscsi_allocmap_update(IscsiLun *iscsilun, int64_t sector_num, 461 int nb_sectors, bool allocated, bool valid) 462 { 463 int64_t cl_num_expanded, nb_cls_expanded, cl_num_shrunk, nb_cls_shrunk; 464 465 if (iscsilun->allocmap == NULL) { 466 return; 467 } 468 /* expand to entirely contain all affected clusters */ 469 cl_num_expanded = sector_num / iscsilun->cluster_sectors; 470 nb_cls_expanded = DIV_ROUND_UP(sector_num + nb_sectors, 471 iscsilun->cluster_sectors) - cl_num_expanded; 472 /* shrink to touch only completely contained clusters */ 473 cl_num_shrunk = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors); 474 nb_cls_shrunk = (sector_num + nb_sectors) / iscsilun->cluster_sectors 475 - cl_num_shrunk; 476 if (allocated) { 477 bitmap_set(iscsilun->allocmap, cl_num_expanded, nb_cls_expanded); 478 } else { 479 if (nb_cls_shrunk > 0) { 480 bitmap_clear(iscsilun->allocmap, cl_num_shrunk, nb_cls_shrunk); 481 } 482 } 483 484 if (iscsilun->allocmap_valid == NULL) { 485 return; 486 } 487 if (valid) { 488 if (nb_cls_shrunk > 0) { 489 bitmap_set(iscsilun->allocmap_valid, cl_num_shrunk, nb_cls_shrunk); 490 } 491 } else { 492 bitmap_clear(iscsilun->allocmap_valid, cl_num_expanded, 493 nb_cls_expanded); 494 } 495 } 496 497 static void 498 iscsi_allocmap_set_allocated(IscsiLun *iscsilun, int64_t sector_num, 499 int nb_sectors) 500 { 501 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, true, true); 502 } 503 504 static void 505 iscsi_allocmap_set_unallocated(IscsiLun *iscsilun, int64_t sector_num, 506 int nb_sectors) 507 { 508 /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update 509 * is ignored, so this will in effect be an iscsi_allocmap_set_invalid. 510 */ 511 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, false, true); 512 } 513 514 static void iscsi_allocmap_set_invalid(IscsiLun *iscsilun, int64_t sector_num, 515 int nb_sectors) 516 { 517 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, false, false); 518 } 519 520 static void iscsi_allocmap_invalidate(IscsiLun *iscsilun) 521 { 522 if (iscsilun->allocmap) { 523 bitmap_zero(iscsilun->allocmap, iscsilun->allocmap_size); 524 } 525 if (iscsilun->allocmap_valid) { 526 bitmap_zero(iscsilun->allocmap_valid, iscsilun->allocmap_size); 527 } 528 } 529 530 static inline bool 531 iscsi_allocmap_is_allocated(IscsiLun *iscsilun, int64_t sector_num, 532 int nb_sectors) 533 { 534 unsigned long size; 535 if (iscsilun->allocmap == NULL) { 536 return true; 537 } 538 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors); 539 return !(find_next_bit(iscsilun->allocmap, size, 540 sector_num / iscsilun->cluster_sectors) == size); 541 } 542 543 static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun, 544 int64_t sector_num, int nb_sectors) 545 { 546 unsigned long size; 547 if (iscsilun->allocmap_valid == NULL) { 548 return false; 549 } 550 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors); 551 return (find_next_zero_bit(iscsilun->allocmap_valid, size, 552 sector_num / iscsilun->cluster_sectors) == size); 553 } 554 555 static int coroutine_fn 556 iscsi_co_writev_flags(BlockDriverState *bs, int64_t sector_num, int nb_sectors, 557 QEMUIOVector *iov, int flags) 558 { 559 IscsiLun *iscsilun = bs->opaque; 560 struct IscsiTask iTask; 561 uint64_t lba; 562 uint32_t num_sectors; 563 bool fua = flags & BDRV_REQ_FUA; 564 int r = 0; 565 566 if (fua) { 567 assert(iscsilun->dpofua); 568 } 569 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 570 return -EINVAL; 571 } 572 573 if (bs->bl.max_transfer) { 574 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer); 575 } 576 577 lba = sector_qemu2lun(sector_num, iscsilun); 578 num_sectors = sector_qemu2lun(nb_sectors, iscsilun); 579 iscsi_co_init_iscsitask(iscsilun, &iTask); 580 qemu_mutex_lock(&iscsilun->mutex); 581 retry: 582 if (iscsilun->use_16_for_rw) { 583 #if LIBISCSI_API_VERSION >= (20160603) 584 iTask.task = iscsi_write16_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 585 NULL, num_sectors * iscsilun->block_size, 586 iscsilun->block_size, 0, 0, fua, 0, 0, 587 iscsi_co_generic_cb, &iTask, 588 (struct scsi_iovec *)iov->iov, iov->niov); 589 } else { 590 iTask.task = iscsi_write10_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 591 NULL, num_sectors * iscsilun->block_size, 592 iscsilun->block_size, 0, 0, fua, 0, 0, 593 iscsi_co_generic_cb, &iTask, 594 (struct scsi_iovec *)iov->iov, iov->niov); 595 } 596 #else 597 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba, 598 NULL, num_sectors * iscsilun->block_size, 599 iscsilun->block_size, 0, 0, fua, 0, 0, 600 iscsi_co_generic_cb, &iTask); 601 } else { 602 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba, 603 NULL, num_sectors * iscsilun->block_size, 604 iscsilun->block_size, 0, 0, fua, 0, 0, 605 iscsi_co_generic_cb, &iTask); 606 } 607 #endif 608 if (iTask.task == NULL) { 609 qemu_mutex_unlock(&iscsilun->mutex); 610 return -ENOMEM; 611 } 612 #if LIBISCSI_API_VERSION < (20160603) 613 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov, 614 iov->niov); 615 #endif 616 while (!iTask.complete) { 617 iscsi_set_events(iscsilun); 618 qemu_mutex_unlock(&iscsilun->mutex); 619 qemu_coroutine_yield(); 620 qemu_mutex_lock(&iscsilun->mutex); 621 } 622 623 if (iTask.task != NULL) { 624 scsi_free_scsi_task(iTask.task); 625 iTask.task = NULL; 626 } 627 628 if (iTask.do_retry) { 629 iTask.complete = 0; 630 goto retry; 631 } 632 633 if (iTask.status != SCSI_STATUS_GOOD) { 634 iscsi_allocmap_set_invalid(iscsilun, sector_num, nb_sectors); 635 error_report("iSCSI WRITE10/16 failed at lba %" PRIu64 ": %s", lba, 636 iTask.err_str); 637 r = iTask.err_code; 638 goto out_unlock; 639 } 640 641 iscsi_allocmap_set_allocated(iscsilun, sector_num, nb_sectors); 642 643 out_unlock: 644 qemu_mutex_unlock(&iscsilun->mutex); 645 g_free(iTask.err_str); 646 return r; 647 } 648 649 650 651 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs, 652 int64_t sector_num, 653 int nb_sectors, int *pnum, 654 BlockDriverState **file) 655 { 656 IscsiLun *iscsilun = bs->opaque; 657 struct scsi_get_lba_status *lbas = NULL; 658 struct scsi_lba_status_descriptor *lbasd = NULL; 659 struct IscsiTask iTask; 660 uint64_t lba; 661 int64_t ret; 662 663 iscsi_co_init_iscsitask(iscsilun, &iTask); 664 665 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 666 ret = -EINVAL; 667 goto out; 668 } 669 670 /* default to all sectors allocated */ 671 ret = BDRV_BLOCK_DATA; 672 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID; 673 *pnum = nb_sectors; 674 675 /* LUN does not support logical block provisioning */ 676 if (!iscsilun->lbpme) { 677 goto out; 678 } 679 680 lba = sector_qemu2lun(sector_num, iscsilun); 681 682 qemu_mutex_lock(&iscsilun->mutex); 683 retry: 684 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun, 685 lba, 8 + 16, iscsi_co_generic_cb, 686 &iTask) == NULL) { 687 ret = -ENOMEM; 688 goto out_unlock; 689 } 690 691 while (!iTask.complete) { 692 iscsi_set_events(iscsilun); 693 qemu_mutex_unlock(&iscsilun->mutex); 694 qemu_coroutine_yield(); 695 qemu_mutex_lock(&iscsilun->mutex); 696 } 697 698 if (iTask.do_retry) { 699 if (iTask.task != NULL) { 700 scsi_free_scsi_task(iTask.task); 701 iTask.task = NULL; 702 } 703 iTask.complete = 0; 704 goto retry; 705 } 706 707 if (iTask.status != SCSI_STATUS_GOOD) { 708 /* in case the get_lba_status_callout fails (i.e. 709 * because the device is busy or the cmd is not 710 * supported) we pretend all blocks are allocated 711 * for backwards compatibility */ 712 error_report("iSCSI GET_LBA_STATUS failed at lba %" PRIu64 ": %s", 713 lba, iTask.err_str); 714 goto out_unlock; 715 } 716 717 lbas = scsi_datain_unmarshall(iTask.task); 718 if (lbas == NULL) { 719 ret = -EIO; 720 goto out_unlock; 721 } 722 723 lbasd = &lbas->descriptors[0]; 724 725 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) { 726 ret = -EIO; 727 goto out_unlock; 728 } 729 730 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun); 731 732 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED || 733 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) { 734 ret &= ~BDRV_BLOCK_DATA; 735 if (iscsilun->lbprz) { 736 ret |= BDRV_BLOCK_ZERO; 737 } 738 } 739 740 if (ret & BDRV_BLOCK_ZERO) { 741 iscsi_allocmap_set_unallocated(iscsilun, sector_num, *pnum); 742 } else { 743 iscsi_allocmap_set_allocated(iscsilun, sector_num, *pnum); 744 } 745 746 if (*pnum > nb_sectors) { 747 *pnum = nb_sectors; 748 } 749 out_unlock: 750 qemu_mutex_unlock(&iscsilun->mutex); 751 g_free(iTask.err_str); 752 out: 753 if (iTask.task != NULL) { 754 scsi_free_scsi_task(iTask.task); 755 } 756 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) { 757 *file = bs; 758 } 759 return ret; 760 } 761 762 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs, 763 int64_t sector_num, int nb_sectors, 764 QEMUIOVector *iov) 765 { 766 IscsiLun *iscsilun = bs->opaque; 767 struct IscsiTask iTask; 768 uint64_t lba; 769 uint32_t num_sectors; 770 int r = 0; 771 772 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 773 return -EINVAL; 774 } 775 776 if (bs->bl.max_transfer) { 777 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer); 778 } 779 780 /* if cache.direct is off and we have a valid entry in our allocation map 781 * we can skip checking the block status and directly return zeroes if 782 * the request falls within an unallocated area */ 783 if (iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) && 784 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) { 785 qemu_iovec_memset(iov, 0, 0x00, iov->size); 786 return 0; 787 } 788 789 if (nb_sectors >= ISCSI_CHECKALLOC_THRES && 790 !iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) && 791 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) { 792 int pnum; 793 BlockDriverState *file; 794 /* check the block status from the beginning of the cluster 795 * containing the start sector */ 796 int64_t ret = iscsi_co_get_block_status(bs, 797 sector_num - sector_num % iscsilun->cluster_sectors, 798 BDRV_REQUEST_MAX_SECTORS, &pnum, &file); 799 if (ret < 0) { 800 return ret; 801 } 802 /* if the whole request falls into an unallocated area we can avoid 803 * to read and directly return zeroes instead */ 804 if (ret & BDRV_BLOCK_ZERO && 805 pnum >= nb_sectors + sector_num % iscsilun->cluster_sectors) { 806 qemu_iovec_memset(iov, 0, 0x00, iov->size); 807 return 0; 808 } 809 } 810 811 lba = sector_qemu2lun(sector_num, iscsilun); 812 num_sectors = sector_qemu2lun(nb_sectors, iscsilun); 813 814 iscsi_co_init_iscsitask(iscsilun, &iTask); 815 qemu_mutex_lock(&iscsilun->mutex); 816 retry: 817 if (iscsilun->use_16_for_rw) { 818 #if LIBISCSI_API_VERSION >= (20160603) 819 iTask.task = iscsi_read16_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 820 num_sectors * iscsilun->block_size, 821 iscsilun->block_size, 0, 0, 0, 0, 0, 822 iscsi_co_generic_cb, &iTask, 823 (struct scsi_iovec *)iov->iov, iov->niov); 824 } else { 825 iTask.task = iscsi_read10_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 826 num_sectors * iscsilun->block_size, 827 iscsilun->block_size, 828 0, 0, 0, 0, 0, 829 iscsi_co_generic_cb, &iTask, 830 (struct scsi_iovec *)iov->iov, iov->niov); 831 } 832 #else 833 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba, 834 num_sectors * iscsilun->block_size, 835 iscsilun->block_size, 0, 0, 0, 0, 0, 836 iscsi_co_generic_cb, &iTask); 837 } else { 838 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba, 839 num_sectors * iscsilun->block_size, 840 iscsilun->block_size, 841 0, 0, 0, 0, 0, 842 iscsi_co_generic_cb, &iTask); 843 } 844 #endif 845 if (iTask.task == NULL) { 846 qemu_mutex_unlock(&iscsilun->mutex); 847 return -ENOMEM; 848 } 849 #if LIBISCSI_API_VERSION < (20160603) 850 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov); 851 #endif 852 while (!iTask.complete) { 853 iscsi_set_events(iscsilun); 854 qemu_mutex_unlock(&iscsilun->mutex); 855 qemu_coroutine_yield(); 856 qemu_mutex_lock(&iscsilun->mutex); 857 } 858 859 if (iTask.task != NULL) { 860 scsi_free_scsi_task(iTask.task); 861 iTask.task = NULL; 862 } 863 864 if (iTask.do_retry) { 865 iTask.complete = 0; 866 goto retry; 867 } 868 869 if (iTask.status != SCSI_STATUS_GOOD) { 870 error_report("iSCSI READ10/16 failed at lba %" PRIu64 ": %s", 871 lba, iTask.err_str); 872 r = iTask.err_code; 873 } 874 875 qemu_mutex_unlock(&iscsilun->mutex); 876 g_free(iTask.err_str); 877 return r; 878 } 879 880 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs) 881 { 882 IscsiLun *iscsilun = bs->opaque; 883 struct IscsiTask iTask; 884 int r = 0; 885 886 iscsi_co_init_iscsitask(iscsilun, &iTask); 887 qemu_mutex_lock(&iscsilun->mutex); 888 retry: 889 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0, 890 0, iscsi_co_generic_cb, &iTask) == NULL) { 891 qemu_mutex_unlock(&iscsilun->mutex); 892 return -ENOMEM; 893 } 894 895 while (!iTask.complete) { 896 iscsi_set_events(iscsilun); 897 qemu_mutex_unlock(&iscsilun->mutex); 898 qemu_coroutine_yield(); 899 qemu_mutex_lock(&iscsilun->mutex); 900 } 901 902 if (iTask.task != NULL) { 903 scsi_free_scsi_task(iTask.task); 904 iTask.task = NULL; 905 } 906 907 if (iTask.do_retry) { 908 iTask.complete = 0; 909 goto retry; 910 } 911 912 if (iTask.status != SCSI_STATUS_GOOD) { 913 error_report("iSCSI SYNCHRONIZECACHE10 failed: %s", iTask.err_str); 914 r = iTask.err_code; 915 } 916 917 qemu_mutex_unlock(&iscsilun->mutex); 918 g_free(iTask.err_str); 919 return r; 920 } 921 922 #ifdef __linux__ 923 /* Called (via iscsi_service) with QemuMutex held. */ 924 static void 925 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status, 926 void *command_data, void *opaque) 927 { 928 IscsiAIOCB *acb = opaque; 929 930 g_free(acb->buf); 931 acb->buf = NULL; 932 933 acb->status = 0; 934 if (status < 0) { 935 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s", 936 iscsi_get_error(iscsi)); 937 acb->status = iscsi_translate_sense(&acb->task->sense); 938 } 939 940 acb->ioh->driver_status = 0; 941 acb->ioh->host_status = 0; 942 acb->ioh->resid = 0; 943 acb->ioh->status = status; 944 945 #define SG_ERR_DRIVER_SENSE 0x08 946 947 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) { 948 int ss; 949 950 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE; 951 952 acb->ioh->sb_len_wr = acb->task->datain.size - 2; 953 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ? 954 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr; 955 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss); 956 } 957 958 iscsi_schedule_bh(acb); 959 } 960 961 static void iscsi_ioctl_bh_completion(void *opaque) 962 { 963 IscsiAIOCB *acb = opaque; 964 965 qemu_bh_delete(acb->bh); 966 acb->common.cb(acb->common.opaque, acb->ret); 967 qemu_aio_unref(acb); 968 } 969 970 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf) 971 { 972 BlockDriverState *bs = acb->common.bs; 973 IscsiLun *iscsilun = bs->opaque; 974 int ret = 0; 975 976 switch (req) { 977 case SG_GET_VERSION_NUM: 978 *(int *)buf = 30000; 979 break; 980 case SG_GET_SCSI_ID: 981 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type; 982 break; 983 default: 984 ret = -EINVAL; 985 } 986 assert(!acb->bh); 987 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), 988 iscsi_ioctl_bh_completion, acb); 989 acb->ret = ret; 990 qemu_bh_schedule(acb->bh); 991 } 992 993 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs, 994 unsigned long int req, void *buf, 995 BlockCompletionFunc *cb, void *opaque) 996 { 997 IscsiLun *iscsilun = bs->opaque; 998 struct iscsi_context *iscsi = iscsilun->iscsi; 999 struct iscsi_data data; 1000 IscsiAIOCB *acb; 1001 1002 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque); 1003 1004 acb->iscsilun = iscsilun; 1005 acb->bh = NULL; 1006 acb->status = -EINPROGRESS; 1007 acb->buf = NULL; 1008 acb->ioh = buf; 1009 1010 if (req != SG_IO) { 1011 iscsi_ioctl_handle_emulated(acb, req, buf); 1012 return &acb->common; 1013 } 1014 1015 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) { 1016 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)", 1017 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE); 1018 qemu_aio_unref(acb); 1019 return NULL; 1020 } 1021 1022 acb->task = malloc(sizeof(struct scsi_task)); 1023 if (acb->task == NULL) { 1024 error_report("iSCSI: Failed to allocate task for scsi command. %s", 1025 iscsi_get_error(iscsi)); 1026 qemu_aio_unref(acb); 1027 return NULL; 1028 } 1029 memset(acb->task, 0, sizeof(struct scsi_task)); 1030 1031 switch (acb->ioh->dxfer_direction) { 1032 case SG_DXFER_TO_DEV: 1033 acb->task->xfer_dir = SCSI_XFER_WRITE; 1034 break; 1035 case SG_DXFER_FROM_DEV: 1036 acb->task->xfer_dir = SCSI_XFER_READ; 1037 break; 1038 default: 1039 acb->task->xfer_dir = SCSI_XFER_NONE; 1040 break; 1041 } 1042 1043 acb->task->cdb_size = acb->ioh->cmd_len; 1044 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len); 1045 acb->task->expxferlen = acb->ioh->dxfer_len; 1046 1047 data.size = 0; 1048 qemu_mutex_lock(&iscsilun->mutex); 1049 if (acb->task->xfer_dir == SCSI_XFER_WRITE) { 1050 if (acb->ioh->iovec_count == 0) { 1051 data.data = acb->ioh->dxferp; 1052 data.size = acb->ioh->dxfer_len; 1053 } else { 1054 scsi_task_set_iov_out(acb->task, 1055 (struct scsi_iovec *) acb->ioh->dxferp, 1056 acb->ioh->iovec_count); 1057 } 1058 } 1059 1060 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task, 1061 iscsi_aio_ioctl_cb, 1062 (data.size > 0) ? &data : NULL, 1063 acb) != 0) { 1064 qemu_mutex_unlock(&iscsilun->mutex); 1065 scsi_free_scsi_task(acb->task); 1066 qemu_aio_unref(acb); 1067 return NULL; 1068 } 1069 1070 /* tell libiscsi to read straight into the buffer we got from ioctl */ 1071 if (acb->task->xfer_dir == SCSI_XFER_READ) { 1072 if (acb->ioh->iovec_count == 0) { 1073 scsi_task_add_data_in_buffer(acb->task, 1074 acb->ioh->dxfer_len, 1075 acb->ioh->dxferp); 1076 } else { 1077 scsi_task_set_iov_in(acb->task, 1078 (struct scsi_iovec *) acb->ioh->dxferp, 1079 acb->ioh->iovec_count); 1080 } 1081 } 1082 1083 iscsi_set_events(iscsilun); 1084 qemu_mutex_unlock(&iscsilun->mutex); 1085 1086 return &acb->common; 1087 } 1088 1089 #endif 1090 1091 static int64_t 1092 iscsi_getlength(BlockDriverState *bs) 1093 { 1094 IscsiLun *iscsilun = bs->opaque; 1095 int64_t len; 1096 1097 len = iscsilun->num_blocks; 1098 len *= iscsilun->block_size; 1099 1100 return len; 1101 } 1102 1103 static int 1104 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 1105 { 1106 IscsiLun *iscsilun = bs->opaque; 1107 struct IscsiTask iTask; 1108 struct unmap_list list; 1109 int r = 0; 1110 1111 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) { 1112 return -ENOTSUP; 1113 } 1114 1115 if (!iscsilun->lbp.lbpu) { 1116 /* UNMAP is not supported by the target */ 1117 return 0; 1118 } 1119 1120 list.lba = offset / iscsilun->block_size; 1121 list.num = bytes / iscsilun->block_size; 1122 1123 iscsi_co_init_iscsitask(iscsilun, &iTask); 1124 qemu_mutex_lock(&iscsilun->mutex); 1125 retry: 1126 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1, 1127 iscsi_co_generic_cb, &iTask) == NULL) { 1128 r = -ENOMEM; 1129 goto out_unlock; 1130 } 1131 1132 while (!iTask.complete) { 1133 iscsi_set_events(iscsilun); 1134 qemu_mutex_unlock(&iscsilun->mutex); 1135 qemu_coroutine_yield(); 1136 qemu_mutex_lock(&iscsilun->mutex); 1137 } 1138 1139 if (iTask.task != NULL) { 1140 scsi_free_scsi_task(iTask.task); 1141 iTask.task = NULL; 1142 } 1143 1144 if (iTask.do_retry) { 1145 iTask.complete = 0; 1146 goto retry; 1147 } 1148 1149 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS, 1150 bytes >> BDRV_SECTOR_BITS); 1151 1152 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) { 1153 /* the target might fail with a check condition if it 1154 is not happy with the alignment of the UNMAP request 1155 we silently fail in this case */ 1156 goto out_unlock; 1157 } 1158 1159 if (iTask.status != SCSI_STATUS_GOOD) { 1160 error_report("iSCSI UNMAP failed at lba %" PRIu64 ": %s", 1161 list.lba, iTask.err_str); 1162 r = iTask.err_code; 1163 goto out_unlock; 1164 } 1165 1166 out_unlock: 1167 qemu_mutex_unlock(&iscsilun->mutex); 1168 g_free(iTask.err_str); 1169 return r; 1170 } 1171 1172 static int 1173 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, 1174 int bytes, BdrvRequestFlags flags) 1175 { 1176 IscsiLun *iscsilun = bs->opaque; 1177 struct IscsiTask iTask; 1178 uint64_t lba; 1179 uint32_t nb_blocks; 1180 bool use_16_for_ws = iscsilun->use_16_for_rw; 1181 int r = 0; 1182 1183 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) { 1184 return -ENOTSUP; 1185 } 1186 1187 if (flags & BDRV_REQ_MAY_UNMAP) { 1188 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) { 1189 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */ 1190 use_16_for_ws = true; 1191 } 1192 if (use_16_for_ws && !iscsilun->lbp.lbpws) { 1193 /* WRITESAME16 with UNMAP is not supported by the target, 1194 * fall back and try WRITESAME10/16 without UNMAP */ 1195 flags &= ~BDRV_REQ_MAY_UNMAP; 1196 use_16_for_ws = iscsilun->use_16_for_rw; 1197 } 1198 } 1199 1200 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) { 1201 /* WRITESAME without UNMAP is not supported by the target */ 1202 return -ENOTSUP; 1203 } 1204 1205 lba = offset / iscsilun->block_size; 1206 nb_blocks = bytes / iscsilun->block_size; 1207 1208 if (iscsilun->zeroblock == NULL) { 1209 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size); 1210 if (iscsilun->zeroblock == NULL) { 1211 return -ENOMEM; 1212 } 1213 } 1214 1215 qemu_mutex_lock(&iscsilun->mutex); 1216 iscsi_co_init_iscsitask(iscsilun, &iTask); 1217 retry: 1218 if (use_16_for_ws) { 1219 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba, 1220 iscsilun->zeroblock, iscsilun->block_size, 1221 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 1222 0, 0, iscsi_co_generic_cb, &iTask); 1223 } else { 1224 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba, 1225 iscsilun->zeroblock, iscsilun->block_size, 1226 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 1227 0, 0, iscsi_co_generic_cb, &iTask); 1228 } 1229 if (iTask.task == NULL) { 1230 qemu_mutex_unlock(&iscsilun->mutex); 1231 return -ENOMEM; 1232 } 1233 1234 while (!iTask.complete) { 1235 iscsi_set_events(iscsilun); 1236 qemu_mutex_unlock(&iscsilun->mutex); 1237 qemu_coroutine_yield(); 1238 qemu_mutex_lock(&iscsilun->mutex); 1239 } 1240 1241 if (iTask.status == SCSI_STATUS_CHECK_CONDITION && 1242 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST && 1243 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE || 1244 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) { 1245 /* WRITE SAME is not supported by the target */ 1246 iscsilun->has_write_same = false; 1247 scsi_free_scsi_task(iTask.task); 1248 r = -ENOTSUP; 1249 goto out_unlock; 1250 } 1251 1252 if (iTask.task != NULL) { 1253 scsi_free_scsi_task(iTask.task); 1254 iTask.task = NULL; 1255 } 1256 1257 if (iTask.do_retry) { 1258 iTask.complete = 0; 1259 goto retry; 1260 } 1261 1262 if (iTask.status != SCSI_STATUS_GOOD) { 1263 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS, 1264 bytes >> BDRV_SECTOR_BITS); 1265 error_report("iSCSI WRITESAME10/16 failed at lba %" PRIu64 ": %s", 1266 lba, iTask.err_str); 1267 r = iTask.err_code; 1268 goto out_unlock; 1269 } 1270 1271 if (flags & BDRV_REQ_MAY_UNMAP) { 1272 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS, 1273 bytes >> BDRV_SECTOR_BITS); 1274 } else { 1275 iscsi_allocmap_set_allocated(iscsilun, offset >> BDRV_SECTOR_BITS, 1276 bytes >> BDRV_SECTOR_BITS); 1277 } 1278 1279 out_unlock: 1280 qemu_mutex_unlock(&iscsilun->mutex); 1281 g_free(iTask.err_str); 1282 return r; 1283 } 1284 1285 static void apply_chap(struct iscsi_context *iscsi, QemuOpts *opts, 1286 Error **errp) 1287 { 1288 const char *user = NULL; 1289 const char *password = NULL; 1290 const char *secretid; 1291 char *secret = NULL; 1292 1293 user = qemu_opt_get(opts, "user"); 1294 if (!user) { 1295 return; 1296 } 1297 1298 secretid = qemu_opt_get(opts, "password-secret"); 1299 password = qemu_opt_get(opts, "password"); 1300 if (secretid && password) { 1301 error_setg(errp, "'password' and 'password-secret' properties are " 1302 "mutually exclusive"); 1303 return; 1304 } 1305 if (secretid) { 1306 secret = qcrypto_secret_lookup_as_utf8(secretid, errp); 1307 if (!secret) { 1308 return; 1309 } 1310 password = secret; 1311 } else if (!password) { 1312 error_setg(errp, "CHAP username specified but no password was given"); 1313 return; 1314 } 1315 1316 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) { 1317 error_setg(errp, "Failed to set initiator username and password"); 1318 } 1319 1320 g_free(secret); 1321 } 1322 1323 static void apply_header_digest(struct iscsi_context *iscsi, QemuOpts *opts, 1324 Error **errp) 1325 { 1326 const char *digest = NULL; 1327 1328 digest = qemu_opt_get(opts, "header-digest"); 1329 if (!digest) { 1330 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1331 } else if (!strcmp(digest, "crc32c")) { 1332 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C); 1333 } else if (!strcmp(digest, "none")) { 1334 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE); 1335 } else if (!strcmp(digest, "crc32c-none")) { 1336 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE); 1337 } else if (!strcmp(digest, "none-crc32c")) { 1338 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1339 } else { 1340 error_setg(errp, "Invalid header-digest setting : %s", digest); 1341 } 1342 } 1343 1344 static char *get_initiator_name(QemuOpts *opts) 1345 { 1346 const char *name; 1347 char *iscsi_name; 1348 UuidInfo *uuid_info; 1349 1350 name = qemu_opt_get(opts, "initiator-name"); 1351 if (name) { 1352 return g_strdup(name); 1353 } 1354 1355 uuid_info = qmp_query_uuid(NULL); 1356 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) { 1357 name = qemu_get_vm_name(); 1358 } else { 1359 name = uuid_info->UUID; 1360 } 1361 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s", 1362 name ? ":" : "", name ? name : ""); 1363 qapi_free_UuidInfo(uuid_info); 1364 return iscsi_name; 1365 } 1366 1367 static void iscsi_nop_timed_event(void *opaque) 1368 { 1369 IscsiLun *iscsilun = opaque; 1370 1371 qemu_mutex_lock(&iscsilun->mutex); 1372 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) { 1373 error_report("iSCSI: NOP timeout. Reconnecting..."); 1374 iscsilun->request_timed_out = true; 1375 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) { 1376 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages."); 1377 goto out; 1378 } 1379 1380 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1381 iscsi_set_events(iscsilun); 1382 1383 out: 1384 qemu_mutex_unlock(&iscsilun->mutex); 1385 } 1386 1387 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp) 1388 { 1389 struct scsi_task *task = NULL; 1390 struct scsi_readcapacity10 *rc10 = NULL; 1391 struct scsi_readcapacity16 *rc16 = NULL; 1392 int retries = ISCSI_CMD_RETRIES; 1393 1394 do { 1395 if (task != NULL) { 1396 scsi_free_scsi_task(task); 1397 task = NULL; 1398 } 1399 1400 switch (iscsilun->type) { 1401 case TYPE_DISK: 1402 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun); 1403 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1404 rc16 = scsi_datain_unmarshall(task); 1405 if (rc16 == NULL) { 1406 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data."); 1407 } else { 1408 iscsilun->block_size = rc16->block_length; 1409 iscsilun->num_blocks = rc16->returned_lba + 1; 1410 iscsilun->lbpme = !!rc16->lbpme; 1411 iscsilun->lbprz = !!rc16->lbprz; 1412 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff); 1413 } 1414 break; 1415 } 1416 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION 1417 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) { 1418 break; 1419 } 1420 /* Fall through and try READ CAPACITY(10) instead. */ 1421 case TYPE_ROM: 1422 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0); 1423 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1424 rc10 = scsi_datain_unmarshall(task); 1425 if (rc10 == NULL) { 1426 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data."); 1427 } else { 1428 iscsilun->block_size = rc10->block_size; 1429 if (rc10->lba == 0) { 1430 /* blank disk loaded */ 1431 iscsilun->num_blocks = 0; 1432 } else { 1433 iscsilun->num_blocks = rc10->lba + 1; 1434 } 1435 } 1436 } 1437 break; 1438 default: 1439 return; 1440 } 1441 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION 1442 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION 1443 && retries-- > 0); 1444 1445 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1446 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command"); 1447 } else if (!iscsilun->block_size || 1448 iscsilun->block_size % BDRV_SECTOR_SIZE) { 1449 error_setg(errp, "iSCSI: the target returned an invalid " 1450 "block size of %d.", iscsilun->block_size); 1451 } 1452 if (task) { 1453 scsi_free_scsi_task(task); 1454 } 1455 } 1456 1457 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun, 1458 int evpd, int pc, void **inq, Error **errp) 1459 { 1460 int full_size; 1461 struct scsi_task *task = NULL; 1462 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64); 1463 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1464 goto fail; 1465 } 1466 full_size = scsi_datain_getfullsize(task); 1467 if (full_size > task->datain.size) { 1468 scsi_free_scsi_task(task); 1469 1470 /* we need more data for the full list */ 1471 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size); 1472 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1473 goto fail; 1474 } 1475 } 1476 1477 *inq = scsi_datain_unmarshall(task); 1478 if (*inq == NULL) { 1479 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob"); 1480 goto fail_with_err; 1481 } 1482 1483 return task; 1484 1485 fail: 1486 error_setg(errp, "iSCSI: Inquiry command failed : %s", 1487 iscsi_get_error(iscsi)); 1488 fail_with_err: 1489 if (task != NULL) { 1490 scsi_free_scsi_task(task); 1491 } 1492 return NULL; 1493 } 1494 1495 static void iscsi_detach_aio_context(BlockDriverState *bs) 1496 { 1497 IscsiLun *iscsilun = bs->opaque; 1498 1499 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi), 1500 false, NULL, NULL, NULL, NULL); 1501 iscsilun->events = 0; 1502 1503 if (iscsilun->nop_timer) { 1504 timer_del(iscsilun->nop_timer); 1505 timer_free(iscsilun->nop_timer); 1506 iscsilun->nop_timer = NULL; 1507 } 1508 if (iscsilun->event_timer) { 1509 timer_del(iscsilun->event_timer); 1510 timer_free(iscsilun->event_timer); 1511 iscsilun->event_timer = NULL; 1512 } 1513 } 1514 1515 static void iscsi_attach_aio_context(BlockDriverState *bs, 1516 AioContext *new_context) 1517 { 1518 IscsiLun *iscsilun = bs->opaque; 1519 1520 iscsilun->aio_context = new_context; 1521 iscsi_set_events(iscsilun); 1522 1523 /* Set up a timer for sending out iSCSI NOPs */ 1524 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context, 1525 QEMU_CLOCK_REALTIME, SCALE_MS, 1526 iscsi_nop_timed_event, iscsilun); 1527 timer_mod(iscsilun->nop_timer, 1528 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1529 1530 /* Set up a timer for periodic calls to iscsi_set_events and to 1531 * scan for command timeout */ 1532 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context, 1533 QEMU_CLOCK_REALTIME, SCALE_MS, 1534 iscsi_timed_check_events, iscsilun); 1535 timer_mod(iscsilun->event_timer, 1536 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL); 1537 } 1538 1539 static void iscsi_modesense_sync(IscsiLun *iscsilun) 1540 { 1541 struct scsi_task *task; 1542 struct scsi_mode_sense *ms = NULL; 1543 iscsilun->write_protected = false; 1544 iscsilun->dpofua = false; 1545 1546 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun, 1547 1, SCSI_MODESENSE_PC_CURRENT, 1548 0x3F, 0, 255); 1549 if (task == NULL) { 1550 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s", 1551 iscsi_get_error(iscsilun->iscsi)); 1552 goto out; 1553 } 1554 1555 if (task->status != SCSI_STATUS_GOOD) { 1556 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable"); 1557 goto out; 1558 } 1559 ms = scsi_datain_unmarshall(task); 1560 if (!ms) { 1561 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s", 1562 iscsi_get_error(iscsilun->iscsi)); 1563 goto out; 1564 } 1565 iscsilun->write_protected = ms->device_specific_parameter & 0x80; 1566 iscsilun->dpofua = ms->device_specific_parameter & 0x10; 1567 1568 out: 1569 if (task) { 1570 scsi_free_scsi_task(task); 1571 } 1572 } 1573 1574 static void iscsi_parse_iscsi_option(const char *target, QDict *options) 1575 { 1576 QemuOptsList *list; 1577 QemuOpts *opts; 1578 const char *user, *password, *password_secret, *initiator_name, 1579 *header_digest, *timeout; 1580 1581 list = qemu_find_opts("iscsi"); 1582 if (!list) { 1583 return; 1584 } 1585 1586 opts = qemu_opts_find(list, target); 1587 if (opts == NULL) { 1588 opts = QTAILQ_FIRST(&list->head); 1589 if (!opts) { 1590 return; 1591 } 1592 } 1593 1594 user = qemu_opt_get(opts, "user"); 1595 if (user) { 1596 qdict_set_default_str(options, "user", user); 1597 } 1598 1599 password = qemu_opt_get(opts, "password"); 1600 if (password) { 1601 qdict_set_default_str(options, "password", password); 1602 } 1603 1604 password_secret = qemu_opt_get(opts, "password-secret"); 1605 if (password_secret) { 1606 qdict_set_default_str(options, "password-secret", password_secret); 1607 } 1608 1609 initiator_name = qemu_opt_get(opts, "initiator-name"); 1610 if (initiator_name) { 1611 qdict_set_default_str(options, "initiator-name", initiator_name); 1612 } 1613 1614 header_digest = qemu_opt_get(opts, "header-digest"); 1615 if (header_digest) { 1616 /* -iscsi takes upper case values, but QAPI only supports lower case 1617 * enum constant names, so we have to convert here. */ 1618 char *qapi_value = g_ascii_strdown(header_digest, -1); 1619 qdict_set_default_str(options, "header-digest", qapi_value); 1620 g_free(qapi_value); 1621 } 1622 1623 timeout = qemu_opt_get(opts, "timeout"); 1624 if (timeout) { 1625 qdict_set_default_str(options, "timeout", timeout); 1626 } 1627 } 1628 1629 /* 1630 * We support iscsi url's on the form 1631 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun> 1632 */ 1633 static void iscsi_parse_filename(const char *filename, QDict *options, 1634 Error **errp) 1635 { 1636 struct iscsi_url *iscsi_url; 1637 const char *transport_name; 1638 char *lun_str; 1639 1640 iscsi_url = iscsi_parse_full_url(NULL, filename); 1641 if (iscsi_url == NULL) { 1642 error_setg(errp, "Failed to parse URL : %s", filename); 1643 return; 1644 } 1645 1646 #if LIBISCSI_API_VERSION >= (20160603) 1647 switch (iscsi_url->transport) { 1648 case TCP_TRANSPORT: 1649 transport_name = "tcp"; 1650 break; 1651 case ISER_TRANSPORT: 1652 transport_name = "iser"; 1653 break; 1654 default: 1655 error_setg(errp, "Unknown transport type (%d)", 1656 iscsi_url->transport); 1657 return; 1658 } 1659 #else 1660 transport_name = "tcp"; 1661 #endif 1662 1663 qdict_set_default_str(options, "transport", transport_name); 1664 qdict_set_default_str(options, "portal", iscsi_url->portal); 1665 qdict_set_default_str(options, "target", iscsi_url->target); 1666 1667 lun_str = g_strdup_printf("%d", iscsi_url->lun); 1668 qdict_set_default_str(options, "lun", lun_str); 1669 g_free(lun_str); 1670 1671 /* User/password from -iscsi take precedence over those from the URL */ 1672 iscsi_parse_iscsi_option(iscsi_url->target, options); 1673 1674 if (iscsi_url->user[0] != '\0') { 1675 qdict_set_default_str(options, "user", iscsi_url->user); 1676 qdict_set_default_str(options, "password", iscsi_url->passwd); 1677 } 1678 1679 iscsi_destroy_url(iscsi_url); 1680 } 1681 1682 static QemuOptsList runtime_opts = { 1683 .name = "iscsi", 1684 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), 1685 .desc = { 1686 { 1687 .name = "transport", 1688 .type = QEMU_OPT_STRING, 1689 }, 1690 { 1691 .name = "portal", 1692 .type = QEMU_OPT_STRING, 1693 }, 1694 { 1695 .name = "target", 1696 .type = QEMU_OPT_STRING, 1697 }, 1698 { 1699 .name = "user", 1700 .type = QEMU_OPT_STRING, 1701 }, 1702 { 1703 .name = "password", 1704 .type = QEMU_OPT_STRING, 1705 }, 1706 { 1707 .name = "password-secret", 1708 .type = QEMU_OPT_STRING, 1709 }, 1710 { 1711 .name = "lun", 1712 .type = QEMU_OPT_NUMBER, 1713 }, 1714 { 1715 .name = "initiator-name", 1716 .type = QEMU_OPT_STRING, 1717 }, 1718 { 1719 .name = "header-digest", 1720 .type = QEMU_OPT_STRING, 1721 }, 1722 { 1723 .name = "timeout", 1724 .type = QEMU_OPT_NUMBER, 1725 }, 1726 { 1727 .name = "filename", 1728 .type = QEMU_OPT_STRING, 1729 }, 1730 { /* end of list */ } 1731 }, 1732 }; 1733 1734 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags, 1735 Error **errp) 1736 { 1737 IscsiLun *iscsilun = bs->opaque; 1738 struct iscsi_context *iscsi = NULL; 1739 struct scsi_task *task = NULL; 1740 struct scsi_inquiry_standard *inq = NULL; 1741 struct scsi_inquiry_supported_pages *inq_vpd; 1742 char *initiator_name = NULL; 1743 QemuOpts *opts; 1744 Error *local_err = NULL; 1745 const char *transport_name, *portal, *target, *filename; 1746 #if LIBISCSI_API_VERSION >= (20160603) 1747 enum iscsi_transport_type transport; 1748 #endif 1749 int i, ret = 0, timeout = 0, lun; 1750 1751 /* If we are given a filename, parse the filename, with precedence given to 1752 * filename encoded options */ 1753 filename = qdict_get_try_str(options, "filename"); 1754 if (filename) { 1755 warn_report("'filename' option specified. " 1756 "This is an unsupported option, and may be deprecated " 1757 "in the future"); 1758 iscsi_parse_filename(filename, options, &local_err); 1759 if (local_err) { 1760 ret = -EINVAL; 1761 error_propagate(errp, local_err); 1762 goto exit; 1763 } 1764 } 1765 1766 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); 1767 qemu_opts_absorb_qdict(opts, options, &local_err); 1768 if (local_err) { 1769 error_propagate(errp, local_err); 1770 ret = -EINVAL; 1771 goto out; 1772 } 1773 1774 transport_name = qemu_opt_get(opts, "transport"); 1775 portal = qemu_opt_get(opts, "portal"); 1776 target = qemu_opt_get(opts, "target"); 1777 lun = qemu_opt_get_number(opts, "lun", 0); 1778 1779 if (!transport_name || !portal || !target) { 1780 error_setg(errp, "Need all of transport, portal and target options"); 1781 ret = -EINVAL; 1782 goto out; 1783 } 1784 1785 if (!strcmp(transport_name, "tcp")) { 1786 #if LIBISCSI_API_VERSION >= (20160603) 1787 transport = TCP_TRANSPORT; 1788 } else if (!strcmp(transport_name, "iser")) { 1789 transport = ISER_TRANSPORT; 1790 #else 1791 /* TCP is what older libiscsi versions always use */ 1792 #endif 1793 } else { 1794 error_setg(errp, "Unknown transport: %s", transport_name); 1795 ret = -EINVAL; 1796 goto out; 1797 } 1798 1799 memset(iscsilun, 0, sizeof(IscsiLun)); 1800 1801 initiator_name = get_initiator_name(opts); 1802 1803 iscsi = iscsi_create_context(initiator_name); 1804 if (iscsi == NULL) { 1805 error_setg(errp, "iSCSI: Failed to create iSCSI context."); 1806 ret = -ENOMEM; 1807 goto out; 1808 } 1809 #if LIBISCSI_API_VERSION >= (20160603) 1810 if (iscsi_init_transport(iscsi, transport)) { 1811 error_setg(errp, ("Error initializing transport.")); 1812 ret = -EINVAL; 1813 goto out; 1814 } 1815 #endif 1816 if (iscsi_set_targetname(iscsi, target)) { 1817 error_setg(errp, "iSCSI: Failed to set target name."); 1818 ret = -EINVAL; 1819 goto out; 1820 } 1821 1822 /* check if we got CHAP username/password via the options */ 1823 apply_chap(iscsi, opts, &local_err); 1824 if (local_err != NULL) { 1825 error_propagate(errp, local_err); 1826 ret = -EINVAL; 1827 goto out; 1828 } 1829 1830 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) { 1831 error_setg(errp, "iSCSI: Failed to set session type to normal."); 1832 ret = -EINVAL; 1833 goto out; 1834 } 1835 1836 /* check if we got HEADER_DIGEST via the options */ 1837 apply_header_digest(iscsi, opts, &local_err); 1838 if (local_err != NULL) { 1839 error_propagate(errp, local_err); 1840 ret = -EINVAL; 1841 goto out; 1842 } 1843 1844 /* timeout handling is broken in libiscsi before 1.15.0 */ 1845 timeout = qemu_opt_get_number(opts, "timeout", 0); 1846 #if LIBISCSI_API_VERSION >= 20150621 1847 iscsi_set_timeout(iscsi, timeout); 1848 #else 1849 if (timeout) { 1850 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0"); 1851 } 1852 #endif 1853 1854 if (iscsi_full_connect_sync(iscsi, portal, lun) != 0) { 1855 error_setg(errp, "iSCSI: Failed to connect to LUN : %s", 1856 iscsi_get_error(iscsi)); 1857 ret = -EINVAL; 1858 goto out; 1859 } 1860 1861 iscsilun->iscsi = iscsi; 1862 iscsilun->aio_context = bdrv_get_aio_context(bs); 1863 iscsilun->lun = lun; 1864 iscsilun->has_write_same = true; 1865 1866 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0, 1867 (void **) &inq, errp); 1868 if (task == NULL) { 1869 ret = -EINVAL; 1870 goto out; 1871 } 1872 iscsilun->type = inq->periperal_device_type; 1873 scsi_free_scsi_task(task); 1874 task = NULL; 1875 1876 iscsi_modesense_sync(iscsilun); 1877 if (iscsilun->dpofua) { 1878 bs->supported_write_flags = BDRV_REQ_FUA; 1879 } 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 if (iscsilun->lbprz && iscsilun->lbp.lbpws) { 1964 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP; 1965 } 1966 1967 out: 1968 qemu_opts_del(opts); 1969 g_free(initiator_name); 1970 if (task != NULL) { 1971 scsi_free_scsi_task(task); 1972 } 1973 1974 if (ret) { 1975 if (iscsi != NULL) { 1976 if (iscsi_is_logged_in(iscsi)) { 1977 iscsi_logout_sync(iscsi); 1978 } 1979 iscsi_destroy_context(iscsi); 1980 } 1981 memset(iscsilun, 0, sizeof(IscsiLun)); 1982 } 1983 exit: 1984 return ret; 1985 } 1986 1987 static void iscsi_close(BlockDriverState *bs) 1988 { 1989 IscsiLun *iscsilun = bs->opaque; 1990 struct iscsi_context *iscsi = iscsilun->iscsi; 1991 1992 iscsi_detach_aio_context(bs); 1993 if (iscsi_is_logged_in(iscsi)) { 1994 iscsi_logout_sync(iscsi); 1995 } 1996 iscsi_destroy_context(iscsi); 1997 g_free(iscsilun->zeroblock); 1998 iscsi_allocmap_free(iscsilun); 1999 qemu_mutex_destroy(&iscsilun->mutex); 2000 memset(iscsilun, 0, sizeof(IscsiLun)); 2001 } 2002 2003 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp) 2004 { 2005 /* We don't actually refresh here, but just return data queried in 2006 * iscsi_open(): iscsi targets don't change their limits. */ 2007 2008 IscsiLun *iscsilun = bs->opaque; 2009 uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff; 2010 unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size); 2011 2012 assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bs->sg); 2013 2014 bs->bl.request_alignment = block_size; 2015 2016 if (iscsilun->bl.max_xfer_len) { 2017 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len); 2018 } 2019 2020 if (max_xfer_len * block_size < INT_MAX) { 2021 bs->bl.max_transfer = max_xfer_len * iscsilun->block_size; 2022 } 2023 2024 if (iscsilun->lbp.lbpu) { 2025 if (iscsilun->bl.max_unmap < 0xffffffff / block_size) { 2026 bs->bl.max_pdiscard = 2027 iscsilun->bl.max_unmap * iscsilun->block_size; 2028 } 2029 bs->bl.pdiscard_alignment = 2030 iscsilun->bl.opt_unmap_gran * iscsilun->block_size; 2031 } else { 2032 bs->bl.pdiscard_alignment = iscsilun->block_size; 2033 } 2034 2035 if (iscsilun->bl.max_ws_len < 0xffffffff / block_size) { 2036 bs->bl.max_pwrite_zeroes = 2037 iscsilun->bl.max_ws_len * iscsilun->block_size; 2038 } 2039 if (iscsilun->lbp.lbpws) { 2040 bs->bl.pwrite_zeroes_alignment = 2041 iscsilun->bl.opt_unmap_gran * iscsilun->block_size; 2042 } else { 2043 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size; 2044 } 2045 if (iscsilun->bl.opt_xfer_len && 2046 iscsilun->bl.opt_xfer_len < INT_MAX / block_size) { 2047 bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len * 2048 iscsilun->block_size); 2049 } 2050 } 2051 2052 /* Note that this will not re-establish a connection with an iSCSI target - it 2053 * is effectively a NOP. */ 2054 static int iscsi_reopen_prepare(BDRVReopenState *state, 2055 BlockReopenQueue *queue, Error **errp) 2056 { 2057 IscsiLun *iscsilun = state->bs->opaque; 2058 2059 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) { 2060 error_setg(errp, "Cannot open a write protected LUN as read-write"); 2061 return -EACCES; 2062 } 2063 return 0; 2064 } 2065 2066 static void iscsi_reopen_commit(BDRVReopenState *reopen_state) 2067 { 2068 IscsiLun *iscsilun = reopen_state->bs->opaque; 2069 2070 /* the cache.direct status might have changed */ 2071 if (iscsilun->allocmap != NULL) { 2072 iscsi_allocmap_init(iscsilun, reopen_state->flags); 2073 } 2074 } 2075 2076 static int iscsi_truncate(BlockDriverState *bs, int64_t offset, 2077 PreallocMode prealloc, Error **errp) 2078 { 2079 IscsiLun *iscsilun = bs->opaque; 2080 Error *local_err = NULL; 2081 2082 if (prealloc != PREALLOC_MODE_OFF) { 2083 error_setg(errp, "Unsupported preallocation mode '%s'", 2084 PreallocMode_str(prealloc)); 2085 return -ENOTSUP; 2086 } 2087 2088 if (iscsilun->type != TYPE_DISK) { 2089 error_setg(errp, "Cannot resize non-disk iSCSI devices"); 2090 return -ENOTSUP; 2091 } 2092 2093 iscsi_readcapacity_sync(iscsilun, &local_err); 2094 if (local_err != NULL) { 2095 error_propagate(errp, local_err); 2096 return -EIO; 2097 } 2098 2099 if (offset > iscsi_getlength(bs)) { 2100 error_setg(errp, "Cannot grow iSCSI devices"); 2101 return -EINVAL; 2102 } 2103 2104 if (iscsilun->allocmap != NULL) { 2105 iscsi_allocmap_init(iscsilun, bs->open_flags); 2106 } 2107 2108 return 0; 2109 } 2110 2111 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp) 2112 { 2113 int ret = 0; 2114 int64_t total_size = 0; 2115 BlockDriverState *bs; 2116 IscsiLun *iscsilun = NULL; 2117 QDict *bs_options; 2118 Error *local_err = NULL; 2119 2120 bs = bdrv_new(); 2121 2122 /* Read out options */ 2123 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2124 BDRV_SECTOR_SIZE); 2125 bs->opaque = g_new0(struct IscsiLun, 1); 2126 iscsilun = bs->opaque; 2127 2128 bs_options = qdict_new(); 2129 iscsi_parse_filename(filename, bs_options, &local_err); 2130 if (local_err) { 2131 error_propagate(errp, local_err); 2132 ret = -EINVAL; 2133 } else { 2134 ret = iscsi_open(bs, bs_options, 0, NULL); 2135 } 2136 QDECREF(bs_options); 2137 2138 if (ret != 0) { 2139 goto out; 2140 } 2141 iscsi_detach_aio_context(bs); 2142 if (iscsilun->type != TYPE_DISK) { 2143 ret = -ENODEV; 2144 goto out; 2145 } 2146 if (bs->total_sectors < total_size) { 2147 ret = -ENOSPC; 2148 goto out; 2149 } 2150 2151 ret = 0; 2152 out: 2153 if (iscsilun->iscsi != NULL) { 2154 iscsi_destroy_context(iscsilun->iscsi); 2155 } 2156 g_free(bs->opaque); 2157 bs->opaque = NULL; 2158 bdrv_unref(bs); 2159 return ret; 2160 } 2161 2162 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2163 { 2164 IscsiLun *iscsilun = bs->opaque; 2165 bdi->unallocated_blocks_are_zero = iscsilun->lbprz; 2166 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE; 2167 return 0; 2168 } 2169 2170 static void iscsi_invalidate_cache(BlockDriverState *bs, 2171 Error **errp) 2172 { 2173 IscsiLun *iscsilun = bs->opaque; 2174 iscsi_allocmap_invalidate(iscsilun); 2175 } 2176 2177 static QemuOptsList iscsi_create_opts = { 2178 .name = "iscsi-create-opts", 2179 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head), 2180 .desc = { 2181 { 2182 .name = BLOCK_OPT_SIZE, 2183 .type = QEMU_OPT_SIZE, 2184 .help = "Virtual disk size" 2185 }, 2186 { /* end of list */ } 2187 } 2188 }; 2189 2190 static BlockDriver bdrv_iscsi = { 2191 .format_name = "iscsi", 2192 .protocol_name = "iscsi", 2193 2194 .instance_size = sizeof(IscsiLun), 2195 .bdrv_parse_filename = iscsi_parse_filename, 2196 .bdrv_file_open = iscsi_open, 2197 .bdrv_close = iscsi_close, 2198 .bdrv_create = iscsi_create, 2199 .create_opts = &iscsi_create_opts, 2200 .bdrv_reopen_prepare = iscsi_reopen_prepare, 2201 .bdrv_reopen_commit = iscsi_reopen_commit, 2202 .bdrv_invalidate_cache = iscsi_invalidate_cache, 2203 2204 .bdrv_getlength = iscsi_getlength, 2205 .bdrv_get_info = iscsi_get_info, 2206 .bdrv_truncate = iscsi_truncate, 2207 .bdrv_refresh_limits = iscsi_refresh_limits, 2208 2209 .bdrv_co_get_block_status = iscsi_co_get_block_status, 2210 .bdrv_co_pdiscard = iscsi_co_pdiscard, 2211 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes, 2212 .bdrv_co_readv = iscsi_co_readv, 2213 .bdrv_co_writev_flags = iscsi_co_writev_flags, 2214 .bdrv_co_flush_to_disk = iscsi_co_flush, 2215 2216 #ifdef __linux__ 2217 .bdrv_aio_ioctl = iscsi_aio_ioctl, 2218 #endif 2219 2220 .bdrv_detach_aio_context = iscsi_detach_aio_context, 2221 .bdrv_attach_aio_context = iscsi_attach_aio_context, 2222 }; 2223 2224 #if LIBISCSI_API_VERSION >= (20160603) 2225 static BlockDriver bdrv_iser = { 2226 .format_name = "iser", 2227 .protocol_name = "iser", 2228 2229 .instance_size = sizeof(IscsiLun), 2230 .bdrv_parse_filename = iscsi_parse_filename, 2231 .bdrv_file_open = iscsi_open, 2232 .bdrv_close = iscsi_close, 2233 .bdrv_create = iscsi_create, 2234 .create_opts = &iscsi_create_opts, 2235 .bdrv_reopen_prepare = iscsi_reopen_prepare, 2236 .bdrv_reopen_commit = iscsi_reopen_commit, 2237 .bdrv_invalidate_cache = iscsi_invalidate_cache, 2238 2239 .bdrv_getlength = iscsi_getlength, 2240 .bdrv_get_info = iscsi_get_info, 2241 .bdrv_truncate = iscsi_truncate, 2242 .bdrv_refresh_limits = iscsi_refresh_limits, 2243 2244 .bdrv_co_get_block_status = iscsi_co_get_block_status, 2245 .bdrv_co_pdiscard = iscsi_co_pdiscard, 2246 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes, 2247 .bdrv_co_readv = iscsi_co_readv, 2248 .bdrv_co_writev_flags = iscsi_co_writev_flags, 2249 .bdrv_co_flush_to_disk = iscsi_co_flush, 2250 2251 #ifdef __linux__ 2252 .bdrv_aio_ioctl = iscsi_aio_ioctl, 2253 #endif 2254 2255 .bdrv_detach_aio_context = iscsi_detach_aio_context, 2256 .bdrv_attach_aio_context = iscsi_attach_aio_context, 2257 }; 2258 #endif 2259 2260 static void iscsi_block_init(void) 2261 { 2262 bdrv_register(&bdrv_iscsi); 2263 #if LIBISCSI_API_VERSION >= (20160603) 2264 bdrv_register(&bdrv_iser); 2265 #endif 2266 } 2267 2268 block_init(iscsi_block_init); 2269