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 "qapi/error.h" 41 #include "qapi/qapi-commands-misc.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_size; 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 assert(iscsilun->cluster_size); 434 iscsilun->allocmap_size = 435 DIV_ROUND_UP(iscsilun->num_blocks * iscsilun->block_size, 436 iscsilun->cluster_size); 437 438 iscsilun->allocmap = bitmap_try_new(iscsilun->allocmap_size); 439 if (!iscsilun->allocmap) { 440 return -ENOMEM; 441 } 442 443 if (open_flags & BDRV_O_NOCACHE) { 444 /* when cache.direct = on all allocmap entries are 445 * treated as invalid to force a relookup of the block 446 * status on every read request */ 447 return 0; 448 } 449 450 iscsilun->allocmap_valid = bitmap_try_new(iscsilun->allocmap_size); 451 if (!iscsilun->allocmap_valid) { 452 /* if we are under memory pressure free the allocmap as well */ 453 iscsi_allocmap_free(iscsilun); 454 return -ENOMEM; 455 } 456 457 return 0; 458 } 459 460 static void 461 iscsi_allocmap_update(IscsiLun *iscsilun, int64_t offset, 462 int64_t bytes, bool allocated, bool valid) 463 { 464 int64_t cl_num_expanded, nb_cls_expanded, cl_num_shrunk, nb_cls_shrunk; 465 466 if (iscsilun->allocmap == NULL) { 467 return; 468 } 469 /* expand to entirely contain all affected clusters */ 470 assert(iscsilun->cluster_size); 471 cl_num_expanded = offset / iscsilun->cluster_size; 472 nb_cls_expanded = DIV_ROUND_UP(offset + bytes, 473 iscsilun->cluster_size) - cl_num_expanded; 474 /* shrink to touch only completely contained clusters */ 475 cl_num_shrunk = DIV_ROUND_UP(offset, iscsilun->cluster_size); 476 nb_cls_shrunk = (offset + bytes) / iscsilun->cluster_size - cl_num_shrunk; 477 if (allocated) { 478 bitmap_set(iscsilun->allocmap, cl_num_expanded, nb_cls_expanded); 479 } else { 480 if (nb_cls_shrunk > 0) { 481 bitmap_clear(iscsilun->allocmap, cl_num_shrunk, nb_cls_shrunk); 482 } 483 } 484 485 if (iscsilun->allocmap_valid == NULL) { 486 return; 487 } 488 if (valid) { 489 if (nb_cls_shrunk > 0) { 490 bitmap_set(iscsilun->allocmap_valid, cl_num_shrunk, nb_cls_shrunk); 491 } 492 } else { 493 bitmap_clear(iscsilun->allocmap_valid, cl_num_expanded, 494 nb_cls_expanded); 495 } 496 } 497 498 static void 499 iscsi_allocmap_set_allocated(IscsiLun *iscsilun, int64_t offset, 500 int64_t bytes) 501 { 502 iscsi_allocmap_update(iscsilun, offset, bytes, true, true); 503 } 504 505 static void 506 iscsi_allocmap_set_unallocated(IscsiLun *iscsilun, int64_t offset, 507 int64_t bytes) 508 { 509 /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update 510 * is ignored, so this will in effect be an iscsi_allocmap_set_invalid. 511 */ 512 iscsi_allocmap_update(iscsilun, offset, bytes, false, true); 513 } 514 515 static void iscsi_allocmap_set_invalid(IscsiLun *iscsilun, int64_t offset, 516 int64_t bytes) 517 { 518 iscsi_allocmap_update(iscsilun, offset, bytes, false, false); 519 } 520 521 static void iscsi_allocmap_invalidate(IscsiLun *iscsilun) 522 { 523 if (iscsilun->allocmap) { 524 bitmap_zero(iscsilun->allocmap, iscsilun->allocmap_size); 525 } 526 if (iscsilun->allocmap_valid) { 527 bitmap_zero(iscsilun->allocmap_valid, iscsilun->allocmap_size); 528 } 529 } 530 531 static inline bool 532 iscsi_allocmap_is_allocated(IscsiLun *iscsilun, int64_t offset, 533 int64_t bytes) 534 { 535 unsigned long size; 536 if (iscsilun->allocmap == NULL) { 537 return true; 538 } 539 assert(iscsilun->cluster_size); 540 size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size); 541 return !(find_next_bit(iscsilun->allocmap, size, 542 offset / iscsilun->cluster_size) == size); 543 } 544 545 static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun, 546 int64_t offset, int64_t bytes) 547 { 548 unsigned long size; 549 if (iscsilun->allocmap_valid == NULL) { 550 return false; 551 } 552 assert(iscsilun->cluster_size); 553 size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size); 554 return (find_next_zero_bit(iscsilun->allocmap_valid, size, 555 offset / iscsilun->cluster_size) == size); 556 } 557 558 static int coroutine_fn 559 iscsi_co_writev_flags(BlockDriverState *bs, int64_t sector_num, int nb_sectors, 560 QEMUIOVector *iov, int flags) 561 { 562 IscsiLun *iscsilun = bs->opaque; 563 struct IscsiTask iTask; 564 uint64_t lba; 565 uint32_t num_sectors; 566 bool fua = flags & BDRV_REQ_FUA; 567 int r = 0; 568 569 if (fua) { 570 assert(iscsilun->dpofua); 571 } 572 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 573 return -EINVAL; 574 } 575 576 if (bs->bl.max_transfer) { 577 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer); 578 } 579 580 lba = sector_qemu2lun(sector_num, iscsilun); 581 num_sectors = sector_qemu2lun(nb_sectors, iscsilun); 582 iscsi_co_init_iscsitask(iscsilun, &iTask); 583 qemu_mutex_lock(&iscsilun->mutex); 584 retry: 585 if (iscsilun->use_16_for_rw) { 586 #if LIBISCSI_API_VERSION >= (20160603) 587 iTask.task = iscsi_write16_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 588 NULL, num_sectors * iscsilun->block_size, 589 iscsilun->block_size, 0, 0, fua, 0, 0, 590 iscsi_co_generic_cb, &iTask, 591 (struct scsi_iovec *)iov->iov, iov->niov); 592 } else { 593 iTask.task = iscsi_write10_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 594 NULL, num_sectors * iscsilun->block_size, 595 iscsilun->block_size, 0, 0, fua, 0, 0, 596 iscsi_co_generic_cb, &iTask, 597 (struct scsi_iovec *)iov->iov, iov->niov); 598 } 599 #else 600 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba, 601 NULL, num_sectors * iscsilun->block_size, 602 iscsilun->block_size, 0, 0, fua, 0, 0, 603 iscsi_co_generic_cb, &iTask); 604 } else { 605 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba, 606 NULL, num_sectors * iscsilun->block_size, 607 iscsilun->block_size, 0, 0, fua, 0, 0, 608 iscsi_co_generic_cb, &iTask); 609 } 610 #endif 611 if (iTask.task == NULL) { 612 qemu_mutex_unlock(&iscsilun->mutex); 613 return -ENOMEM; 614 } 615 #if LIBISCSI_API_VERSION < (20160603) 616 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov, 617 iov->niov); 618 #endif 619 while (!iTask.complete) { 620 iscsi_set_events(iscsilun); 621 qemu_mutex_unlock(&iscsilun->mutex); 622 qemu_coroutine_yield(); 623 qemu_mutex_lock(&iscsilun->mutex); 624 } 625 626 if (iTask.task != NULL) { 627 scsi_free_scsi_task(iTask.task); 628 iTask.task = NULL; 629 } 630 631 if (iTask.do_retry) { 632 iTask.complete = 0; 633 goto retry; 634 } 635 636 if (iTask.status != SCSI_STATUS_GOOD) { 637 iscsi_allocmap_set_invalid(iscsilun, sector_num * BDRV_SECTOR_SIZE, 638 nb_sectors * BDRV_SECTOR_SIZE); 639 error_report("iSCSI WRITE10/16 failed at lba %" PRIu64 ": %s", lba, 640 iTask.err_str); 641 r = iTask.err_code; 642 goto out_unlock; 643 } 644 645 iscsi_allocmap_set_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE, 646 nb_sectors * BDRV_SECTOR_SIZE); 647 648 out_unlock: 649 qemu_mutex_unlock(&iscsilun->mutex); 650 g_free(iTask.err_str); 651 return r; 652 } 653 654 655 656 static int coroutine_fn iscsi_co_block_status(BlockDriverState *bs, 657 bool want_zero, int64_t offset, 658 int64_t bytes, int64_t *pnum, 659 int64_t *map, 660 BlockDriverState **file) 661 { 662 IscsiLun *iscsilun = bs->opaque; 663 struct scsi_get_lba_status *lbas = NULL; 664 struct scsi_lba_status_descriptor *lbasd = NULL; 665 struct IscsiTask iTask; 666 uint64_t lba; 667 int ret; 668 669 iscsi_co_init_iscsitask(iscsilun, &iTask); 670 671 assert(QEMU_IS_ALIGNED(offset | bytes, iscsilun->block_size)); 672 673 /* default to all sectors allocated */ 674 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 675 if (map) { 676 *map = offset; 677 } 678 *pnum = bytes; 679 680 /* LUN does not support logical block provisioning */ 681 if (!iscsilun->lbpme) { 682 goto out; 683 } 684 685 lba = offset / iscsilun->block_size; 686 687 qemu_mutex_lock(&iscsilun->mutex); 688 retry: 689 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun, 690 lba, 8 + 16, iscsi_co_generic_cb, 691 &iTask) == NULL) { 692 ret = -ENOMEM; 693 goto out_unlock; 694 } 695 696 while (!iTask.complete) { 697 iscsi_set_events(iscsilun); 698 qemu_mutex_unlock(&iscsilun->mutex); 699 qemu_coroutine_yield(); 700 qemu_mutex_lock(&iscsilun->mutex); 701 } 702 703 if (iTask.do_retry) { 704 if (iTask.task != NULL) { 705 scsi_free_scsi_task(iTask.task); 706 iTask.task = NULL; 707 } 708 iTask.complete = 0; 709 goto retry; 710 } 711 712 if (iTask.status != SCSI_STATUS_GOOD) { 713 /* in case the get_lba_status_callout fails (i.e. 714 * because the device is busy or the cmd is not 715 * supported) we pretend all blocks are allocated 716 * for backwards compatibility */ 717 error_report("iSCSI GET_LBA_STATUS failed at lba %" PRIu64 ": %s", 718 lba, iTask.err_str); 719 goto out_unlock; 720 } 721 722 lbas = scsi_datain_unmarshall(iTask.task); 723 if (lbas == NULL) { 724 ret = -EIO; 725 goto out_unlock; 726 } 727 728 lbasd = &lbas->descriptors[0]; 729 730 if (lba != lbasd->lba) { 731 ret = -EIO; 732 goto out_unlock; 733 } 734 735 *pnum = lbasd->num_blocks * iscsilun->block_size; 736 737 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED || 738 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) { 739 ret &= ~BDRV_BLOCK_DATA; 740 if (iscsilun->lbprz) { 741 ret |= BDRV_BLOCK_ZERO; 742 } 743 } 744 745 if (ret & BDRV_BLOCK_ZERO) { 746 iscsi_allocmap_set_unallocated(iscsilun, offset, *pnum); 747 } else { 748 iscsi_allocmap_set_allocated(iscsilun, offset, *pnum); 749 } 750 751 if (*pnum > bytes) { 752 *pnum = bytes; 753 } 754 out_unlock: 755 qemu_mutex_unlock(&iscsilun->mutex); 756 g_free(iTask.err_str); 757 out: 758 if (iTask.task != NULL) { 759 scsi_free_scsi_task(iTask.task); 760 } 761 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID && file) { 762 *file = bs; 763 } 764 return ret; 765 } 766 767 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs, 768 int64_t sector_num, int nb_sectors, 769 QEMUIOVector *iov) 770 { 771 IscsiLun *iscsilun = bs->opaque; 772 struct IscsiTask iTask; 773 uint64_t lba; 774 uint32_t num_sectors; 775 int r = 0; 776 777 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 778 return -EINVAL; 779 } 780 781 if (bs->bl.max_transfer) { 782 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer); 783 } 784 785 /* if cache.direct is off and we have a valid entry in our allocation map 786 * we can skip checking the block status and directly return zeroes if 787 * the request falls within an unallocated area */ 788 if (iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE, 789 nb_sectors * BDRV_SECTOR_SIZE) && 790 !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE, 791 nb_sectors * BDRV_SECTOR_SIZE)) { 792 qemu_iovec_memset(iov, 0, 0x00, iov->size); 793 return 0; 794 } 795 796 if (nb_sectors >= ISCSI_CHECKALLOC_THRES && 797 !iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE, 798 nb_sectors * BDRV_SECTOR_SIZE) && 799 !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE, 800 nb_sectors * BDRV_SECTOR_SIZE)) { 801 int64_t pnum; 802 /* check the block status from the beginning of the cluster 803 * containing the start sector */ 804 int64_t head; 805 int ret; 806 807 assert(iscsilun->cluster_size); 808 head = (sector_num * BDRV_SECTOR_SIZE) % iscsilun->cluster_size; 809 ret = iscsi_co_block_status(bs, true, 810 sector_num * BDRV_SECTOR_SIZE - head, 811 BDRV_REQUEST_MAX_BYTES, &pnum, NULL, NULL); 812 if (ret < 0) { 813 return ret; 814 } 815 /* if the whole request falls into an unallocated area we can avoid 816 * reading and directly return zeroes instead */ 817 if (ret & BDRV_BLOCK_ZERO && 818 pnum >= nb_sectors * BDRV_SECTOR_SIZE + head) { 819 qemu_iovec_memset(iov, 0, 0x00, iov->size); 820 return 0; 821 } 822 } 823 824 lba = sector_qemu2lun(sector_num, iscsilun); 825 num_sectors = sector_qemu2lun(nb_sectors, iscsilun); 826 827 iscsi_co_init_iscsitask(iscsilun, &iTask); 828 qemu_mutex_lock(&iscsilun->mutex); 829 retry: 830 if (iscsilun->use_16_for_rw) { 831 #if LIBISCSI_API_VERSION >= (20160603) 832 iTask.task = iscsi_read16_iov_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 (struct scsi_iovec *)iov->iov, iov->niov); 837 } else { 838 iTask.task = iscsi_read10_iov_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 (struct scsi_iovec *)iov->iov, iov->niov); 844 } 845 #else 846 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba, 847 num_sectors * iscsilun->block_size, 848 iscsilun->block_size, 0, 0, 0, 0, 0, 849 iscsi_co_generic_cb, &iTask); 850 } else { 851 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba, 852 num_sectors * iscsilun->block_size, 853 iscsilun->block_size, 854 0, 0, 0, 0, 0, 855 iscsi_co_generic_cb, &iTask); 856 } 857 #endif 858 if (iTask.task == NULL) { 859 qemu_mutex_unlock(&iscsilun->mutex); 860 return -ENOMEM; 861 } 862 #if LIBISCSI_API_VERSION < (20160603) 863 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov); 864 #endif 865 while (!iTask.complete) { 866 iscsi_set_events(iscsilun); 867 qemu_mutex_unlock(&iscsilun->mutex); 868 qemu_coroutine_yield(); 869 qemu_mutex_lock(&iscsilun->mutex); 870 } 871 872 if (iTask.task != NULL) { 873 scsi_free_scsi_task(iTask.task); 874 iTask.task = NULL; 875 } 876 877 if (iTask.do_retry) { 878 iTask.complete = 0; 879 goto retry; 880 } 881 882 if (iTask.status != SCSI_STATUS_GOOD) { 883 error_report("iSCSI READ10/16 failed at lba %" PRIu64 ": %s", 884 lba, iTask.err_str); 885 r = iTask.err_code; 886 } 887 888 qemu_mutex_unlock(&iscsilun->mutex); 889 g_free(iTask.err_str); 890 return r; 891 } 892 893 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs) 894 { 895 IscsiLun *iscsilun = bs->opaque; 896 struct IscsiTask iTask; 897 int r = 0; 898 899 iscsi_co_init_iscsitask(iscsilun, &iTask); 900 qemu_mutex_lock(&iscsilun->mutex); 901 retry: 902 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0, 903 0, iscsi_co_generic_cb, &iTask) == NULL) { 904 qemu_mutex_unlock(&iscsilun->mutex); 905 return -ENOMEM; 906 } 907 908 while (!iTask.complete) { 909 iscsi_set_events(iscsilun); 910 qemu_mutex_unlock(&iscsilun->mutex); 911 qemu_coroutine_yield(); 912 qemu_mutex_lock(&iscsilun->mutex); 913 } 914 915 if (iTask.task != NULL) { 916 scsi_free_scsi_task(iTask.task); 917 iTask.task = NULL; 918 } 919 920 if (iTask.do_retry) { 921 iTask.complete = 0; 922 goto retry; 923 } 924 925 if (iTask.status != SCSI_STATUS_GOOD) { 926 error_report("iSCSI SYNCHRONIZECACHE10 failed: %s", iTask.err_str); 927 r = iTask.err_code; 928 } 929 930 qemu_mutex_unlock(&iscsilun->mutex); 931 g_free(iTask.err_str); 932 return r; 933 } 934 935 #ifdef __linux__ 936 /* Called (via iscsi_service) with QemuMutex held. */ 937 static void 938 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status, 939 void *command_data, void *opaque) 940 { 941 IscsiAIOCB *acb = opaque; 942 943 g_free(acb->buf); 944 acb->buf = NULL; 945 946 acb->status = 0; 947 if (status < 0) { 948 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s", 949 iscsi_get_error(iscsi)); 950 acb->status = iscsi_translate_sense(&acb->task->sense); 951 } 952 953 acb->ioh->driver_status = 0; 954 acb->ioh->host_status = 0; 955 acb->ioh->resid = 0; 956 acb->ioh->status = status; 957 958 #define SG_ERR_DRIVER_SENSE 0x08 959 960 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) { 961 int ss; 962 963 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE; 964 965 acb->ioh->sb_len_wr = acb->task->datain.size - 2; 966 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ? 967 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr; 968 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss); 969 } 970 971 iscsi_schedule_bh(acb); 972 } 973 974 static void iscsi_ioctl_bh_completion(void *opaque) 975 { 976 IscsiAIOCB *acb = opaque; 977 978 qemu_bh_delete(acb->bh); 979 acb->common.cb(acb->common.opaque, acb->ret); 980 qemu_aio_unref(acb); 981 } 982 983 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf) 984 { 985 BlockDriverState *bs = acb->common.bs; 986 IscsiLun *iscsilun = bs->opaque; 987 int ret = 0; 988 989 switch (req) { 990 case SG_GET_VERSION_NUM: 991 *(int *)buf = 30000; 992 break; 993 case SG_GET_SCSI_ID: 994 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type; 995 break; 996 default: 997 ret = -EINVAL; 998 } 999 assert(!acb->bh); 1000 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), 1001 iscsi_ioctl_bh_completion, acb); 1002 acb->ret = ret; 1003 qemu_bh_schedule(acb->bh); 1004 } 1005 1006 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs, 1007 unsigned long int req, void *buf, 1008 BlockCompletionFunc *cb, void *opaque) 1009 { 1010 IscsiLun *iscsilun = bs->opaque; 1011 struct iscsi_context *iscsi = iscsilun->iscsi; 1012 struct iscsi_data data; 1013 IscsiAIOCB *acb; 1014 1015 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque); 1016 1017 acb->iscsilun = iscsilun; 1018 acb->bh = NULL; 1019 acb->status = -EINPROGRESS; 1020 acb->buf = NULL; 1021 acb->ioh = buf; 1022 1023 if (req != SG_IO) { 1024 iscsi_ioctl_handle_emulated(acb, req, buf); 1025 return &acb->common; 1026 } 1027 1028 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) { 1029 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)", 1030 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE); 1031 qemu_aio_unref(acb); 1032 return NULL; 1033 } 1034 1035 acb->task = malloc(sizeof(struct scsi_task)); 1036 if (acb->task == NULL) { 1037 error_report("iSCSI: Failed to allocate task for scsi command. %s", 1038 iscsi_get_error(iscsi)); 1039 qemu_aio_unref(acb); 1040 return NULL; 1041 } 1042 memset(acb->task, 0, sizeof(struct scsi_task)); 1043 1044 switch (acb->ioh->dxfer_direction) { 1045 case SG_DXFER_TO_DEV: 1046 acb->task->xfer_dir = SCSI_XFER_WRITE; 1047 break; 1048 case SG_DXFER_FROM_DEV: 1049 acb->task->xfer_dir = SCSI_XFER_READ; 1050 break; 1051 default: 1052 acb->task->xfer_dir = SCSI_XFER_NONE; 1053 break; 1054 } 1055 1056 acb->task->cdb_size = acb->ioh->cmd_len; 1057 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len); 1058 acb->task->expxferlen = acb->ioh->dxfer_len; 1059 1060 data.size = 0; 1061 qemu_mutex_lock(&iscsilun->mutex); 1062 if (acb->task->xfer_dir == SCSI_XFER_WRITE) { 1063 if (acb->ioh->iovec_count == 0) { 1064 data.data = acb->ioh->dxferp; 1065 data.size = acb->ioh->dxfer_len; 1066 } else { 1067 scsi_task_set_iov_out(acb->task, 1068 (struct scsi_iovec *) acb->ioh->dxferp, 1069 acb->ioh->iovec_count); 1070 } 1071 } 1072 1073 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task, 1074 iscsi_aio_ioctl_cb, 1075 (data.size > 0) ? &data : NULL, 1076 acb) != 0) { 1077 qemu_mutex_unlock(&iscsilun->mutex); 1078 scsi_free_scsi_task(acb->task); 1079 qemu_aio_unref(acb); 1080 return NULL; 1081 } 1082 1083 /* tell libiscsi to read straight into the buffer we got from ioctl */ 1084 if (acb->task->xfer_dir == SCSI_XFER_READ) { 1085 if (acb->ioh->iovec_count == 0) { 1086 scsi_task_add_data_in_buffer(acb->task, 1087 acb->ioh->dxfer_len, 1088 acb->ioh->dxferp); 1089 } else { 1090 scsi_task_set_iov_in(acb->task, 1091 (struct scsi_iovec *) acb->ioh->dxferp, 1092 acb->ioh->iovec_count); 1093 } 1094 } 1095 1096 iscsi_set_events(iscsilun); 1097 qemu_mutex_unlock(&iscsilun->mutex); 1098 1099 return &acb->common; 1100 } 1101 1102 #endif 1103 1104 static int64_t 1105 iscsi_getlength(BlockDriverState *bs) 1106 { 1107 IscsiLun *iscsilun = bs->opaque; 1108 int64_t len; 1109 1110 len = iscsilun->num_blocks; 1111 len *= iscsilun->block_size; 1112 1113 return len; 1114 } 1115 1116 static int 1117 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 1118 { 1119 IscsiLun *iscsilun = bs->opaque; 1120 struct IscsiTask iTask; 1121 struct unmap_list list; 1122 int r = 0; 1123 1124 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) { 1125 return -ENOTSUP; 1126 } 1127 1128 if (!iscsilun->lbp.lbpu) { 1129 /* UNMAP is not supported by the target */ 1130 return 0; 1131 } 1132 1133 list.lba = offset / iscsilun->block_size; 1134 list.num = bytes / iscsilun->block_size; 1135 1136 iscsi_co_init_iscsitask(iscsilun, &iTask); 1137 qemu_mutex_lock(&iscsilun->mutex); 1138 retry: 1139 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1, 1140 iscsi_co_generic_cb, &iTask) == NULL) { 1141 r = -ENOMEM; 1142 goto out_unlock; 1143 } 1144 1145 while (!iTask.complete) { 1146 iscsi_set_events(iscsilun); 1147 qemu_mutex_unlock(&iscsilun->mutex); 1148 qemu_coroutine_yield(); 1149 qemu_mutex_lock(&iscsilun->mutex); 1150 } 1151 1152 if (iTask.task != NULL) { 1153 scsi_free_scsi_task(iTask.task); 1154 iTask.task = NULL; 1155 } 1156 1157 if (iTask.do_retry) { 1158 iTask.complete = 0; 1159 goto retry; 1160 } 1161 1162 iscsi_allocmap_set_invalid(iscsilun, offset, bytes); 1163 1164 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) { 1165 /* the target might fail with a check condition if it 1166 is not happy with the alignment of the UNMAP request 1167 we silently fail in this case */ 1168 goto out_unlock; 1169 } 1170 1171 if (iTask.status != SCSI_STATUS_GOOD) { 1172 error_report("iSCSI UNMAP failed at lba %" PRIu64 ": %s", 1173 list.lba, iTask.err_str); 1174 r = iTask.err_code; 1175 goto out_unlock; 1176 } 1177 1178 out_unlock: 1179 qemu_mutex_unlock(&iscsilun->mutex); 1180 g_free(iTask.err_str); 1181 return r; 1182 } 1183 1184 static int 1185 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, 1186 int bytes, BdrvRequestFlags flags) 1187 { 1188 IscsiLun *iscsilun = bs->opaque; 1189 struct IscsiTask iTask; 1190 uint64_t lba; 1191 uint32_t nb_blocks; 1192 bool use_16_for_ws = iscsilun->use_16_for_rw; 1193 int r = 0; 1194 1195 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) { 1196 return -ENOTSUP; 1197 } 1198 1199 if (flags & BDRV_REQ_MAY_UNMAP) { 1200 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) { 1201 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */ 1202 use_16_for_ws = true; 1203 } 1204 if (use_16_for_ws && !iscsilun->lbp.lbpws) { 1205 /* WRITESAME16 with UNMAP is not supported by the target, 1206 * fall back and try WRITESAME10/16 without UNMAP */ 1207 flags &= ~BDRV_REQ_MAY_UNMAP; 1208 use_16_for_ws = iscsilun->use_16_for_rw; 1209 } 1210 } 1211 1212 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) { 1213 /* WRITESAME without UNMAP is not supported by the target */ 1214 return -ENOTSUP; 1215 } 1216 1217 lba = offset / iscsilun->block_size; 1218 nb_blocks = bytes / iscsilun->block_size; 1219 1220 if (iscsilun->zeroblock == NULL) { 1221 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size); 1222 if (iscsilun->zeroblock == NULL) { 1223 return -ENOMEM; 1224 } 1225 } 1226 1227 qemu_mutex_lock(&iscsilun->mutex); 1228 iscsi_co_init_iscsitask(iscsilun, &iTask); 1229 retry: 1230 if (use_16_for_ws) { 1231 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba, 1232 iscsilun->zeroblock, iscsilun->block_size, 1233 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 1234 0, 0, iscsi_co_generic_cb, &iTask); 1235 } else { 1236 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba, 1237 iscsilun->zeroblock, iscsilun->block_size, 1238 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 1239 0, 0, iscsi_co_generic_cb, &iTask); 1240 } 1241 if (iTask.task == NULL) { 1242 qemu_mutex_unlock(&iscsilun->mutex); 1243 return -ENOMEM; 1244 } 1245 1246 while (!iTask.complete) { 1247 iscsi_set_events(iscsilun); 1248 qemu_mutex_unlock(&iscsilun->mutex); 1249 qemu_coroutine_yield(); 1250 qemu_mutex_lock(&iscsilun->mutex); 1251 } 1252 1253 if (iTask.status == SCSI_STATUS_CHECK_CONDITION && 1254 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST && 1255 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE || 1256 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) { 1257 /* WRITE SAME is not supported by the target */ 1258 iscsilun->has_write_same = false; 1259 scsi_free_scsi_task(iTask.task); 1260 r = -ENOTSUP; 1261 goto out_unlock; 1262 } 1263 1264 if (iTask.task != NULL) { 1265 scsi_free_scsi_task(iTask.task); 1266 iTask.task = NULL; 1267 } 1268 1269 if (iTask.do_retry) { 1270 iTask.complete = 0; 1271 goto retry; 1272 } 1273 1274 if (iTask.status != SCSI_STATUS_GOOD) { 1275 iscsi_allocmap_set_invalid(iscsilun, offset, bytes); 1276 error_report("iSCSI WRITESAME10/16 failed at lba %" PRIu64 ": %s", 1277 lba, iTask.err_str); 1278 r = iTask.err_code; 1279 goto out_unlock; 1280 } 1281 1282 if (flags & BDRV_REQ_MAY_UNMAP) { 1283 iscsi_allocmap_set_invalid(iscsilun, offset, bytes); 1284 } else { 1285 iscsi_allocmap_set_allocated(iscsilun, offset, bytes); 1286 } 1287 1288 out_unlock: 1289 qemu_mutex_unlock(&iscsilun->mutex); 1290 g_free(iTask.err_str); 1291 return r; 1292 } 1293 1294 static void apply_chap(struct iscsi_context *iscsi, QemuOpts *opts, 1295 Error **errp) 1296 { 1297 const char *user = NULL; 1298 const char *password = NULL; 1299 const char *secretid; 1300 char *secret = NULL; 1301 1302 user = qemu_opt_get(opts, "user"); 1303 if (!user) { 1304 return; 1305 } 1306 1307 secretid = qemu_opt_get(opts, "password-secret"); 1308 password = qemu_opt_get(opts, "password"); 1309 if (secretid && password) { 1310 error_setg(errp, "'password' and 'password-secret' properties are " 1311 "mutually exclusive"); 1312 return; 1313 } 1314 if (secretid) { 1315 secret = qcrypto_secret_lookup_as_utf8(secretid, errp); 1316 if (!secret) { 1317 return; 1318 } 1319 password = secret; 1320 } else if (!password) { 1321 error_setg(errp, "CHAP username specified but no password was given"); 1322 return; 1323 } 1324 1325 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) { 1326 error_setg(errp, "Failed to set initiator username and password"); 1327 } 1328 1329 g_free(secret); 1330 } 1331 1332 static void apply_header_digest(struct iscsi_context *iscsi, QemuOpts *opts, 1333 Error **errp) 1334 { 1335 const char *digest = NULL; 1336 1337 digest = qemu_opt_get(opts, "header-digest"); 1338 if (!digest) { 1339 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1340 } else if (!strcmp(digest, "crc32c")) { 1341 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C); 1342 } else if (!strcmp(digest, "none")) { 1343 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE); 1344 } else if (!strcmp(digest, "crc32c-none")) { 1345 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE); 1346 } else if (!strcmp(digest, "none-crc32c")) { 1347 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1348 } else { 1349 error_setg(errp, "Invalid header-digest setting : %s", digest); 1350 } 1351 } 1352 1353 static char *get_initiator_name(QemuOpts *opts) 1354 { 1355 const char *name; 1356 char *iscsi_name; 1357 UuidInfo *uuid_info; 1358 1359 name = qemu_opt_get(opts, "initiator-name"); 1360 if (name) { 1361 return g_strdup(name); 1362 } 1363 1364 uuid_info = qmp_query_uuid(NULL); 1365 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) { 1366 name = qemu_get_vm_name(); 1367 } else { 1368 name = uuid_info->UUID; 1369 } 1370 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s", 1371 name ? ":" : "", name ? name : ""); 1372 qapi_free_UuidInfo(uuid_info); 1373 return iscsi_name; 1374 } 1375 1376 static void iscsi_nop_timed_event(void *opaque) 1377 { 1378 IscsiLun *iscsilun = opaque; 1379 1380 qemu_mutex_lock(&iscsilun->mutex); 1381 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) { 1382 error_report("iSCSI: NOP timeout. Reconnecting..."); 1383 iscsilun->request_timed_out = true; 1384 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) { 1385 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages."); 1386 goto out; 1387 } 1388 1389 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1390 iscsi_set_events(iscsilun); 1391 1392 out: 1393 qemu_mutex_unlock(&iscsilun->mutex); 1394 } 1395 1396 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp) 1397 { 1398 struct scsi_task *task = NULL; 1399 struct scsi_readcapacity10 *rc10 = NULL; 1400 struct scsi_readcapacity16 *rc16 = NULL; 1401 int retries = ISCSI_CMD_RETRIES; 1402 1403 do { 1404 if (task != NULL) { 1405 scsi_free_scsi_task(task); 1406 task = NULL; 1407 } 1408 1409 switch (iscsilun->type) { 1410 case TYPE_DISK: 1411 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun); 1412 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1413 rc16 = scsi_datain_unmarshall(task); 1414 if (rc16 == NULL) { 1415 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data."); 1416 } else { 1417 iscsilun->block_size = rc16->block_length; 1418 iscsilun->num_blocks = rc16->returned_lba + 1; 1419 iscsilun->lbpme = !!rc16->lbpme; 1420 iscsilun->lbprz = !!rc16->lbprz; 1421 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff); 1422 } 1423 break; 1424 } 1425 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION 1426 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) { 1427 break; 1428 } 1429 /* Fall through and try READ CAPACITY(10) instead. */ 1430 case TYPE_ROM: 1431 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0); 1432 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1433 rc10 = scsi_datain_unmarshall(task); 1434 if (rc10 == NULL) { 1435 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data."); 1436 } else { 1437 iscsilun->block_size = rc10->block_size; 1438 if (rc10->lba == 0) { 1439 /* blank disk loaded */ 1440 iscsilun->num_blocks = 0; 1441 } else { 1442 iscsilun->num_blocks = rc10->lba + 1; 1443 } 1444 } 1445 } 1446 break; 1447 default: 1448 return; 1449 } 1450 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION 1451 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION 1452 && retries-- > 0); 1453 1454 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1455 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command"); 1456 } else if (!iscsilun->block_size || 1457 iscsilun->block_size % BDRV_SECTOR_SIZE) { 1458 error_setg(errp, "iSCSI: the target returned an invalid " 1459 "block size of %d.", iscsilun->block_size); 1460 } 1461 if (task) { 1462 scsi_free_scsi_task(task); 1463 } 1464 } 1465 1466 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun, 1467 int evpd, int pc, void **inq, Error **errp) 1468 { 1469 int full_size; 1470 struct scsi_task *task = NULL; 1471 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64); 1472 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1473 goto fail; 1474 } 1475 full_size = scsi_datain_getfullsize(task); 1476 if (full_size > task->datain.size) { 1477 scsi_free_scsi_task(task); 1478 1479 /* we need more data for the full list */ 1480 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size); 1481 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1482 goto fail; 1483 } 1484 } 1485 1486 *inq = scsi_datain_unmarshall(task); 1487 if (*inq == NULL) { 1488 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob"); 1489 goto fail_with_err; 1490 } 1491 1492 return task; 1493 1494 fail: 1495 error_setg(errp, "iSCSI: Inquiry command failed : %s", 1496 iscsi_get_error(iscsi)); 1497 fail_with_err: 1498 if (task != NULL) { 1499 scsi_free_scsi_task(task); 1500 } 1501 return NULL; 1502 } 1503 1504 static void iscsi_detach_aio_context(BlockDriverState *bs) 1505 { 1506 IscsiLun *iscsilun = bs->opaque; 1507 1508 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi), 1509 false, NULL, NULL, NULL, NULL); 1510 iscsilun->events = 0; 1511 1512 if (iscsilun->nop_timer) { 1513 timer_del(iscsilun->nop_timer); 1514 timer_free(iscsilun->nop_timer); 1515 iscsilun->nop_timer = NULL; 1516 } 1517 if (iscsilun->event_timer) { 1518 timer_del(iscsilun->event_timer); 1519 timer_free(iscsilun->event_timer); 1520 iscsilun->event_timer = NULL; 1521 } 1522 } 1523 1524 static void iscsi_attach_aio_context(BlockDriverState *bs, 1525 AioContext *new_context) 1526 { 1527 IscsiLun *iscsilun = bs->opaque; 1528 1529 iscsilun->aio_context = new_context; 1530 iscsi_set_events(iscsilun); 1531 1532 /* Set up a timer for sending out iSCSI NOPs */ 1533 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context, 1534 QEMU_CLOCK_REALTIME, SCALE_MS, 1535 iscsi_nop_timed_event, iscsilun); 1536 timer_mod(iscsilun->nop_timer, 1537 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1538 1539 /* Set up a timer for periodic calls to iscsi_set_events and to 1540 * scan for command timeout */ 1541 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context, 1542 QEMU_CLOCK_REALTIME, SCALE_MS, 1543 iscsi_timed_check_events, iscsilun); 1544 timer_mod(iscsilun->event_timer, 1545 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL); 1546 } 1547 1548 static void iscsi_modesense_sync(IscsiLun *iscsilun) 1549 { 1550 struct scsi_task *task; 1551 struct scsi_mode_sense *ms = NULL; 1552 iscsilun->write_protected = false; 1553 iscsilun->dpofua = false; 1554 1555 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun, 1556 1, SCSI_MODESENSE_PC_CURRENT, 1557 0x3F, 0, 255); 1558 if (task == NULL) { 1559 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s", 1560 iscsi_get_error(iscsilun->iscsi)); 1561 goto out; 1562 } 1563 1564 if (task->status != SCSI_STATUS_GOOD) { 1565 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable"); 1566 goto out; 1567 } 1568 ms = scsi_datain_unmarshall(task); 1569 if (!ms) { 1570 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s", 1571 iscsi_get_error(iscsilun->iscsi)); 1572 goto out; 1573 } 1574 iscsilun->write_protected = ms->device_specific_parameter & 0x80; 1575 iscsilun->dpofua = ms->device_specific_parameter & 0x10; 1576 1577 out: 1578 if (task) { 1579 scsi_free_scsi_task(task); 1580 } 1581 } 1582 1583 static void iscsi_parse_iscsi_option(const char *target, QDict *options) 1584 { 1585 QemuOptsList *list; 1586 QemuOpts *opts; 1587 const char *user, *password, *password_secret, *initiator_name, 1588 *header_digest, *timeout; 1589 1590 list = qemu_find_opts("iscsi"); 1591 if (!list) { 1592 return; 1593 } 1594 1595 opts = qemu_opts_find(list, target); 1596 if (opts == NULL) { 1597 opts = QTAILQ_FIRST(&list->head); 1598 if (!opts) { 1599 return; 1600 } 1601 } 1602 1603 user = qemu_opt_get(opts, "user"); 1604 if (user) { 1605 qdict_set_default_str(options, "user", user); 1606 } 1607 1608 password = qemu_opt_get(opts, "password"); 1609 if (password) { 1610 qdict_set_default_str(options, "password", password); 1611 } 1612 1613 password_secret = qemu_opt_get(opts, "password-secret"); 1614 if (password_secret) { 1615 qdict_set_default_str(options, "password-secret", password_secret); 1616 } 1617 1618 initiator_name = qemu_opt_get(opts, "initiator-name"); 1619 if (initiator_name) { 1620 qdict_set_default_str(options, "initiator-name", initiator_name); 1621 } 1622 1623 header_digest = qemu_opt_get(opts, "header-digest"); 1624 if (header_digest) { 1625 /* -iscsi takes upper case values, but QAPI only supports lower case 1626 * enum constant names, so we have to convert here. */ 1627 char *qapi_value = g_ascii_strdown(header_digest, -1); 1628 qdict_set_default_str(options, "header-digest", qapi_value); 1629 g_free(qapi_value); 1630 } 1631 1632 timeout = qemu_opt_get(opts, "timeout"); 1633 if (timeout) { 1634 qdict_set_default_str(options, "timeout", timeout); 1635 } 1636 } 1637 1638 /* 1639 * We support iscsi url's on the form 1640 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun> 1641 */ 1642 static void iscsi_parse_filename(const char *filename, QDict *options, 1643 Error **errp) 1644 { 1645 struct iscsi_url *iscsi_url; 1646 const char *transport_name; 1647 char *lun_str; 1648 1649 iscsi_url = iscsi_parse_full_url(NULL, filename); 1650 if (iscsi_url == NULL) { 1651 error_setg(errp, "Failed to parse URL : %s", filename); 1652 return; 1653 } 1654 1655 #if LIBISCSI_API_VERSION >= (20160603) 1656 switch (iscsi_url->transport) { 1657 case TCP_TRANSPORT: 1658 transport_name = "tcp"; 1659 break; 1660 case ISER_TRANSPORT: 1661 transport_name = "iser"; 1662 break; 1663 default: 1664 error_setg(errp, "Unknown transport type (%d)", 1665 iscsi_url->transport); 1666 return; 1667 } 1668 #else 1669 transport_name = "tcp"; 1670 #endif 1671 1672 qdict_set_default_str(options, "transport", transport_name); 1673 qdict_set_default_str(options, "portal", iscsi_url->portal); 1674 qdict_set_default_str(options, "target", iscsi_url->target); 1675 1676 lun_str = g_strdup_printf("%d", iscsi_url->lun); 1677 qdict_set_default_str(options, "lun", lun_str); 1678 g_free(lun_str); 1679 1680 /* User/password from -iscsi take precedence over those from the URL */ 1681 iscsi_parse_iscsi_option(iscsi_url->target, options); 1682 1683 if (iscsi_url->user[0] != '\0') { 1684 qdict_set_default_str(options, "user", iscsi_url->user); 1685 qdict_set_default_str(options, "password", iscsi_url->passwd); 1686 } 1687 1688 iscsi_destroy_url(iscsi_url); 1689 } 1690 1691 static QemuOptsList runtime_opts = { 1692 .name = "iscsi", 1693 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), 1694 .desc = { 1695 { 1696 .name = "transport", 1697 .type = QEMU_OPT_STRING, 1698 }, 1699 { 1700 .name = "portal", 1701 .type = QEMU_OPT_STRING, 1702 }, 1703 { 1704 .name = "target", 1705 .type = QEMU_OPT_STRING, 1706 }, 1707 { 1708 .name = "user", 1709 .type = QEMU_OPT_STRING, 1710 }, 1711 { 1712 .name = "password", 1713 .type = QEMU_OPT_STRING, 1714 }, 1715 { 1716 .name = "password-secret", 1717 .type = QEMU_OPT_STRING, 1718 }, 1719 { 1720 .name = "lun", 1721 .type = QEMU_OPT_NUMBER, 1722 }, 1723 { 1724 .name = "initiator-name", 1725 .type = QEMU_OPT_STRING, 1726 }, 1727 { 1728 .name = "header-digest", 1729 .type = QEMU_OPT_STRING, 1730 }, 1731 { 1732 .name = "timeout", 1733 .type = QEMU_OPT_NUMBER, 1734 }, 1735 { 1736 .name = "filename", 1737 .type = QEMU_OPT_STRING, 1738 }, 1739 { /* end of list */ } 1740 }, 1741 }; 1742 1743 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags, 1744 Error **errp) 1745 { 1746 IscsiLun *iscsilun = bs->opaque; 1747 struct iscsi_context *iscsi = NULL; 1748 struct scsi_task *task = NULL; 1749 struct scsi_inquiry_standard *inq = NULL; 1750 struct scsi_inquiry_supported_pages *inq_vpd; 1751 char *initiator_name = NULL; 1752 QemuOpts *opts; 1753 Error *local_err = NULL; 1754 const char *transport_name, *portal, *target, *filename; 1755 #if LIBISCSI_API_VERSION >= (20160603) 1756 enum iscsi_transport_type transport; 1757 #endif 1758 int i, ret = 0, timeout = 0, lun; 1759 1760 /* If we are given a filename, parse the filename, with precedence given to 1761 * filename encoded options */ 1762 filename = qdict_get_try_str(options, "filename"); 1763 if (filename) { 1764 warn_report("'filename' option specified. " 1765 "This is an unsupported option, and may be deprecated " 1766 "in the future"); 1767 iscsi_parse_filename(filename, options, &local_err); 1768 if (local_err) { 1769 ret = -EINVAL; 1770 error_propagate(errp, local_err); 1771 goto exit; 1772 } 1773 } 1774 1775 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); 1776 qemu_opts_absorb_qdict(opts, options, &local_err); 1777 if (local_err) { 1778 error_propagate(errp, local_err); 1779 ret = -EINVAL; 1780 goto out; 1781 } 1782 1783 transport_name = qemu_opt_get(opts, "transport"); 1784 portal = qemu_opt_get(opts, "portal"); 1785 target = qemu_opt_get(opts, "target"); 1786 lun = qemu_opt_get_number(opts, "lun", 0); 1787 1788 if (!transport_name || !portal || !target) { 1789 error_setg(errp, "Need all of transport, portal and target options"); 1790 ret = -EINVAL; 1791 goto out; 1792 } 1793 1794 if (!strcmp(transport_name, "tcp")) { 1795 #if LIBISCSI_API_VERSION >= (20160603) 1796 transport = TCP_TRANSPORT; 1797 } else if (!strcmp(transport_name, "iser")) { 1798 transport = ISER_TRANSPORT; 1799 #else 1800 /* TCP is what older libiscsi versions always use */ 1801 #endif 1802 } else { 1803 error_setg(errp, "Unknown transport: %s", transport_name); 1804 ret = -EINVAL; 1805 goto out; 1806 } 1807 1808 memset(iscsilun, 0, sizeof(IscsiLun)); 1809 1810 initiator_name = get_initiator_name(opts); 1811 1812 iscsi = iscsi_create_context(initiator_name); 1813 if (iscsi == NULL) { 1814 error_setg(errp, "iSCSI: Failed to create iSCSI context."); 1815 ret = -ENOMEM; 1816 goto out; 1817 } 1818 #if LIBISCSI_API_VERSION >= (20160603) 1819 if (iscsi_init_transport(iscsi, transport)) { 1820 error_setg(errp, ("Error initializing transport.")); 1821 ret = -EINVAL; 1822 goto out; 1823 } 1824 #endif 1825 if (iscsi_set_targetname(iscsi, target)) { 1826 error_setg(errp, "iSCSI: Failed to set target name."); 1827 ret = -EINVAL; 1828 goto out; 1829 } 1830 1831 /* check if we got CHAP username/password via the options */ 1832 apply_chap(iscsi, opts, &local_err); 1833 if (local_err != NULL) { 1834 error_propagate(errp, local_err); 1835 ret = -EINVAL; 1836 goto out; 1837 } 1838 1839 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) { 1840 error_setg(errp, "iSCSI: Failed to set session type to normal."); 1841 ret = -EINVAL; 1842 goto out; 1843 } 1844 1845 /* check if we got HEADER_DIGEST via the options */ 1846 apply_header_digest(iscsi, opts, &local_err); 1847 if (local_err != NULL) { 1848 error_propagate(errp, local_err); 1849 ret = -EINVAL; 1850 goto out; 1851 } 1852 1853 /* timeout handling is broken in libiscsi before 1.15.0 */ 1854 timeout = qemu_opt_get_number(opts, "timeout", 0); 1855 #if LIBISCSI_API_VERSION >= 20150621 1856 iscsi_set_timeout(iscsi, timeout); 1857 #else 1858 if (timeout) { 1859 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0"); 1860 } 1861 #endif 1862 1863 if (iscsi_full_connect_sync(iscsi, portal, lun) != 0) { 1864 error_setg(errp, "iSCSI: Failed to connect to LUN : %s", 1865 iscsi_get_error(iscsi)); 1866 ret = -EINVAL; 1867 goto out; 1868 } 1869 1870 iscsilun->iscsi = iscsi; 1871 iscsilun->aio_context = bdrv_get_aio_context(bs); 1872 iscsilun->lun = lun; 1873 iscsilun->has_write_same = true; 1874 1875 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0, 1876 (void **) &inq, errp); 1877 if (task == NULL) { 1878 ret = -EINVAL; 1879 goto out; 1880 } 1881 iscsilun->type = inq->periperal_device_type; 1882 scsi_free_scsi_task(task); 1883 task = NULL; 1884 1885 iscsi_modesense_sync(iscsilun); 1886 if (iscsilun->dpofua) { 1887 bs->supported_write_flags = BDRV_REQ_FUA; 1888 } 1889 1890 /* Check the write protect flag of the LUN if we want to write */ 1891 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) && 1892 iscsilun->write_protected) { 1893 error_setg(errp, "Cannot open a write protected LUN as read-write"); 1894 ret = -EACCES; 1895 goto out; 1896 } 1897 1898 iscsi_readcapacity_sync(iscsilun, &local_err); 1899 if (local_err != NULL) { 1900 error_propagate(errp, local_err); 1901 ret = -EINVAL; 1902 goto out; 1903 } 1904 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun); 1905 1906 /* We don't have any emulation for devices other than disks and CD-ROMs, so 1907 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu 1908 * will try to read from the device to guess the image format. 1909 */ 1910 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) { 1911 bs->sg = true; 1912 } 1913 1914 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1915 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES, 1916 (void **) &inq_vpd, errp); 1917 if (task == NULL) { 1918 ret = -EINVAL; 1919 goto out; 1920 } 1921 for (i = 0; i < inq_vpd->num_pages; i++) { 1922 struct scsi_task *inq_task; 1923 struct scsi_inquiry_logical_block_provisioning *inq_lbp; 1924 struct scsi_inquiry_block_limits *inq_bl; 1925 switch (inq_vpd->pages[i]) { 1926 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING: 1927 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1928 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING, 1929 (void **) &inq_lbp, errp); 1930 if (inq_task == NULL) { 1931 ret = -EINVAL; 1932 goto out; 1933 } 1934 memcpy(&iscsilun->lbp, inq_lbp, 1935 sizeof(struct scsi_inquiry_logical_block_provisioning)); 1936 scsi_free_scsi_task(inq_task); 1937 break; 1938 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS: 1939 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1940 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS, 1941 (void **) &inq_bl, errp); 1942 if (inq_task == NULL) { 1943 ret = -EINVAL; 1944 goto out; 1945 } 1946 memcpy(&iscsilun->bl, inq_bl, 1947 sizeof(struct scsi_inquiry_block_limits)); 1948 scsi_free_scsi_task(inq_task); 1949 break; 1950 default: 1951 break; 1952 } 1953 } 1954 scsi_free_scsi_task(task); 1955 task = NULL; 1956 1957 qemu_mutex_init(&iscsilun->mutex); 1958 iscsi_attach_aio_context(bs, iscsilun->aio_context); 1959 1960 /* Guess the internal cluster (page) size of the iscsi target by the means 1961 * of opt_unmap_gran. Transfer the unmap granularity only if it has a 1962 * reasonable size */ 1963 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 && 1964 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) { 1965 iscsilun->cluster_size = iscsilun->bl.opt_unmap_gran * 1966 iscsilun->block_size; 1967 if (iscsilun->lbprz) { 1968 ret = iscsi_allocmap_init(iscsilun, bs->open_flags); 1969 } 1970 } 1971 1972 if (iscsilun->lbprz && iscsilun->lbp.lbpws) { 1973 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP; 1974 } 1975 1976 out: 1977 qemu_opts_del(opts); 1978 g_free(initiator_name); 1979 if (task != NULL) { 1980 scsi_free_scsi_task(task); 1981 } 1982 1983 if (ret) { 1984 if (iscsi != NULL) { 1985 if (iscsi_is_logged_in(iscsi)) { 1986 iscsi_logout_sync(iscsi); 1987 } 1988 iscsi_destroy_context(iscsi); 1989 } 1990 memset(iscsilun, 0, sizeof(IscsiLun)); 1991 } 1992 exit: 1993 return ret; 1994 } 1995 1996 static void iscsi_close(BlockDriverState *bs) 1997 { 1998 IscsiLun *iscsilun = bs->opaque; 1999 struct iscsi_context *iscsi = iscsilun->iscsi; 2000 2001 iscsi_detach_aio_context(bs); 2002 if (iscsi_is_logged_in(iscsi)) { 2003 iscsi_logout_sync(iscsi); 2004 } 2005 iscsi_destroy_context(iscsi); 2006 g_free(iscsilun->zeroblock); 2007 iscsi_allocmap_free(iscsilun); 2008 qemu_mutex_destroy(&iscsilun->mutex); 2009 memset(iscsilun, 0, sizeof(IscsiLun)); 2010 } 2011 2012 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp) 2013 { 2014 /* We don't actually refresh here, but just return data queried in 2015 * iscsi_open(): iscsi targets don't change their limits. */ 2016 2017 IscsiLun *iscsilun = bs->opaque; 2018 uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff; 2019 unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size); 2020 2021 assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bs->sg); 2022 2023 bs->bl.request_alignment = block_size; 2024 2025 if (iscsilun->bl.max_xfer_len) { 2026 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len); 2027 } 2028 2029 if (max_xfer_len * block_size < INT_MAX) { 2030 bs->bl.max_transfer = max_xfer_len * iscsilun->block_size; 2031 } 2032 2033 if (iscsilun->lbp.lbpu) { 2034 if (iscsilun->bl.max_unmap < 0xffffffff / block_size) { 2035 bs->bl.max_pdiscard = 2036 iscsilun->bl.max_unmap * iscsilun->block_size; 2037 } 2038 bs->bl.pdiscard_alignment = 2039 iscsilun->bl.opt_unmap_gran * iscsilun->block_size; 2040 } else { 2041 bs->bl.pdiscard_alignment = iscsilun->block_size; 2042 } 2043 2044 if (iscsilun->bl.max_ws_len < 0xffffffff / block_size) { 2045 bs->bl.max_pwrite_zeroes = 2046 iscsilun->bl.max_ws_len * iscsilun->block_size; 2047 } 2048 if (iscsilun->lbp.lbpws) { 2049 bs->bl.pwrite_zeroes_alignment = 2050 iscsilun->bl.opt_unmap_gran * iscsilun->block_size; 2051 } else { 2052 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size; 2053 } 2054 if (iscsilun->bl.opt_xfer_len && 2055 iscsilun->bl.opt_xfer_len < INT_MAX / block_size) { 2056 bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len * 2057 iscsilun->block_size); 2058 } 2059 } 2060 2061 /* Note that this will not re-establish a connection with an iSCSI target - it 2062 * is effectively a NOP. */ 2063 static int iscsi_reopen_prepare(BDRVReopenState *state, 2064 BlockReopenQueue *queue, Error **errp) 2065 { 2066 IscsiLun *iscsilun = state->bs->opaque; 2067 2068 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) { 2069 error_setg(errp, "Cannot open a write protected LUN as read-write"); 2070 return -EACCES; 2071 } 2072 return 0; 2073 } 2074 2075 static void iscsi_reopen_commit(BDRVReopenState *reopen_state) 2076 { 2077 IscsiLun *iscsilun = reopen_state->bs->opaque; 2078 2079 /* the cache.direct status might have changed */ 2080 if (iscsilun->allocmap != NULL) { 2081 iscsi_allocmap_init(iscsilun, reopen_state->flags); 2082 } 2083 } 2084 2085 static int iscsi_truncate(BlockDriverState *bs, int64_t offset, 2086 PreallocMode prealloc, Error **errp) 2087 { 2088 IscsiLun *iscsilun = bs->opaque; 2089 Error *local_err = NULL; 2090 2091 if (prealloc != PREALLOC_MODE_OFF) { 2092 error_setg(errp, "Unsupported preallocation mode '%s'", 2093 PreallocMode_str(prealloc)); 2094 return -ENOTSUP; 2095 } 2096 2097 if (iscsilun->type != TYPE_DISK) { 2098 error_setg(errp, "Cannot resize non-disk iSCSI devices"); 2099 return -ENOTSUP; 2100 } 2101 2102 iscsi_readcapacity_sync(iscsilun, &local_err); 2103 if (local_err != NULL) { 2104 error_propagate(errp, local_err); 2105 return -EIO; 2106 } 2107 2108 if (offset > iscsi_getlength(bs)) { 2109 error_setg(errp, "Cannot grow iSCSI devices"); 2110 return -EINVAL; 2111 } 2112 2113 if (iscsilun->allocmap != NULL) { 2114 iscsi_allocmap_init(iscsilun, bs->open_flags); 2115 } 2116 2117 return 0; 2118 } 2119 2120 static int coroutine_fn iscsi_co_create_opts(const char *filename, QemuOpts *opts, 2121 Error **errp) 2122 { 2123 int ret = 0; 2124 int64_t total_size = 0; 2125 BlockDriverState *bs; 2126 IscsiLun *iscsilun = NULL; 2127 QDict *bs_options; 2128 Error *local_err = NULL; 2129 2130 bs = bdrv_new(); 2131 2132 /* Read out options */ 2133 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2134 BDRV_SECTOR_SIZE); 2135 bs->opaque = g_new0(struct IscsiLun, 1); 2136 iscsilun = bs->opaque; 2137 2138 bs_options = qdict_new(); 2139 iscsi_parse_filename(filename, bs_options, &local_err); 2140 if (local_err) { 2141 error_propagate(errp, local_err); 2142 ret = -EINVAL; 2143 } else { 2144 ret = iscsi_open(bs, bs_options, 0, NULL); 2145 } 2146 QDECREF(bs_options); 2147 2148 if (ret != 0) { 2149 goto out; 2150 } 2151 iscsi_detach_aio_context(bs); 2152 if (iscsilun->type != TYPE_DISK) { 2153 ret = -ENODEV; 2154 goto out; 2155 } 2156 if (bs->total_sectors < total_size) { 2157 ret = -ENOSPC; 2158 goto out; 2159 } 2160 2161 ret = 0; 2162 out: 2163 if (iscsilun->iscsi != NULL) { 2164 iscsi_destroy_context(iscsilun->iscsi); 2165 } 2166 g_free(bs->opaque); 2167 bs->opaque = NULL; 2168 bdrv_unref(bs); 2169 return ret; 2170 } 2171 2172 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2173 { 2174 IscsiLun *iscsilun = bs->opaque; 2175 bdi->unallocated_blocks_are_zero = iscsilun->lbprz; 2176 bdi->cluster_size = iscsilun->cluster_size; 2177 return 0; 2178 } 2179 2180 static void coroutine_fn iscsi_co_invalidate_cache(BlockDriverState *bs, 2181 Error **errp) 2182 { 2183 IscsiLun *iscsilun = bs->opaque; 2184 iscsi_allocmap_invalidate(iscsilun); 2185 } 2186 2187 static QemuOptsList iscsi_create_opts = { 2188 .name = "iscsi-create-opts", 2189 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head), 2190 .desc = { 2191 { 2192 .name = BLOCK_OPT_SIZE, 2193 .type = QEMU_OPT_SIZE, 2194 .help = "Virtual disk size" 2195 }, 2196 { /* end of list */ } 2197 } 2198 }; 2199 2200 static BlockDriver bdrv_iscsi = { 2201 .format_name = "iscsi", 2202 .protocol_name = "iscsi", 2203 2204 .instance_size = sizeof(IscsiLun), 2205 .bdrv_parse_filename = iscsi_parse_filename, 2206 .bdrv_file_open = iscsi_open, 2207 .bdrv_close = iscsi_close, 2208 .bdrv_co_create_opts = iscsi_co_create_opts, 2209 .create_opts = &iscsi_create_opts, 2210 .bdrv_reopen_prepare = iscsi_reopen_prepare, 2211 .bdrv_reopen_commit = iscsi_reopen_commit, 2212 .bdrv_co_invalidate_cache = iscsi_co_invalidate_cache, 2213 2214 .bdrv_getlength = iscsi_getlength, 2215 .bdrv_get_info = iscsi_get_info, 2216 .bdrv_truncate = iscsi_truncate, 2217 .bdrv_refresh_limits = iscsi_refresh_limits, 2218 2219 .bdrv_co_block_status = iscsi_co_block_status, 2220 .bdrv_co_pdiscard = iscsi_co_pdiscard, 2221 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes, 2222 .bdrv_co_readv = iscsi_co_readv, 2223 .bdrv_co_writev_flags = iscsi_co_writev_flags, 2224 .bdrv_co_flush_to_disk = iscsi_co_flush, 2225 2226 #ifdef __linux__ 2227 .bdrv_aio_ioctl = iscsi_aio_ioctl, 2228 #endif 2229 2230 .bdrv_detach_aio_context = iscsi_detach_aio_context, 2231 .bdrv_attach_aio_context = iscsi_attach_aio_context, 2232 }; 2233 2234 #if LIBISCSI_API_VERSION >= (20160603) 2235 static BlockDriver bdrv_iser = { 2236 .format_name = "iser", 2237 .protocol_name = "iser", 2238 2239 .instance_size = sizeof(IscsiLun), 2240 .bdrv_parse_filename = iscsi_parse_filename, 2241 .bdrv_file_open = iscsi_open, 2242 .bdrv_close = iscsi_close, 2243 .bdrv_co_create_opts = iscsi_co_create_opts, 2244 .create_opts = &iscsi_create_opts, 2245 .bdrv_reopen_prepare = iscsi_reopen_prepare, 2246 .bdrv_reopen_commit = iscsi_reopen_commit, 2247 .bdrv_co_invalidate_cache = iscsi_co_invalidate_cache, 2248 2249 .bdrv_getlength = iscsi_getlength, 2250 .bdrv_get_info = iscsi_get_info, 2251 .bdrv_truncate = iscsi_truncate, 2252 .bdrv_refresh_limits = iscsi_refresh_limits, 2253 2254 .bdrv_co_block_status = iscsi_co_block_status, 2255 .bdrv_co_pdiscard = iscsi_co_pdiscard, 2256 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes, 2257 .bdrv_co_readv = iscsi_co_readv, 2258 .bdrv_co_writev_flags = iscsi_co_writev_flags, 2259 .bdrv_co_flush_to_disk = iscsi_co_flush, 2260 2261 #ifdef __linux__ 2262 .bdrv_aio_ioctl = iscsi_aio_ioctl, 2263 #endif 2264 2265 .bdrv_detach_aio_context = iscsi_detach_aio_context, 2266 .bdrv_attach_aio_context = iscsi_attach_aio_context, 2267 }; 2268 #endif 2269 2270 static void iscsi_block_init(void) 2271 { 2272 bdrv_register(&bdrv_iscsi); 2273 #if LIBISCSI_API_VERSION >= (20160603) 2274 bdrv_register(&bdrv_iser); 2275 #endif 2276 } 2277 2278 block_init(iscsi_block_init); 2279