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 "sysemu/sysemu.h" 32 #include "qemu/config-file.h" 33 #include "qemu/error-report.h" 34 #include "qemu/bitops.h" 35 #include "qemu/bitmap.h" 36 #include "block/block-io.h" 37 #include "block/block_int.h" 38 #include "block/qdict.h" 39 #include "scsi/constants.h" 40 #include "qemu/iov.h" 41 #include "qemu/module.h" 42 #include "qemu/option.h" 43 #include "qemu/uuid.h" 44 #include "sysemu/replay.h" 45 #include "qapi/error.h" 46 #include "qapi/qapi-commands-machine.h" 47 #include "qapi/qmp/qdict.h" 48 #include "qapi/qmp/qstring.h" 49 #include "crypto/secret.h" 50 #include "scsi/utils.h" 51 #include "trace.h" 52 53 /* Conflict between scsi/utils.h and libiscsi! :( */ 54 #define SCSI_XFER_NONE ISCSI_XFER_NONE 55 #include <iscsi/iscsi.h> 56 #define inline __attribute__((gnu_inline)) /* required for libiscsi v1.9.0 */ 57 #include <iscsi/scsi-lowlevel.h> 58 #undef inline 59 #undef SCSI_XFER_NONE 60 QEMU_BUILD_BUG_ON((int)SCSI_XFER_NONE != (int)ISCSI_XFER_NONE); 61 62 #ifdef __linux__ 63 #include <scsi/sg.h> 64 #endif 65 66 typedef struct IscsiLun { 67 struct iscsi_context *iscsi; 68 AioContext *aio_context; 69 int lun; 70 enum scsi_inquiry_peripheral_device_type type; 71 int block_size; 72 uint64_t num_blocks; 73 int events; 74 QEMUTimer *nop_timer; 75 QEMUTimer *event_timer; 76 QemuMutex mutex; 77 struct scsi_inquiry_logical_block_provisioning lbp; 78 struct scsi_inquiry_block_limits bl; 79 struct scsi_inquiry_device_designator *dd; 80 unsigned char *zeroblock; 81 /* The allocmap tracks which clusters (pages) on the iSCSI target are 82 * allocated and which are not. In case a target returns zeros for 83 * unallocated pages (iscsilun->lprz) we can directly return zeros instead 84 * of reading zeros over the wire if a read request falls within an 85 * unallocated block. As there are 3 possible states we need 2 bitmaps to 86 * track. allocmap_valid keeps track if QEMU's information about a page is 87 * valid. allocmap tracks if a page is allocated or not. In case QEMU has no 88 * valid information about a page the corresponding allocmap entry should be 89 * switched to unallocated as well to force a new lookup of the allocation 90 * status as lookups are generally skipped if a page is suspect to be 91 * allocated. If a iSCSI target is opened with cache.direct = on the 92 * allocmap_valid does not exist turning all cached information invalid so 93 * that a fresh lookup is made for any page even if allocmap entry returns 94 * it's unallocated. */ 95 unsigned long *allocmap; 96 unsigned long *allocmap_valid; 97 long allocmap_size; 98 int cluster_size; 99 bool use_16_for_rw; 100 bool write_protected; 101 bool lbpme; 102 bool lbprz; 103 bool dpofua; 104 bool has_write_same; 105 bool request_timed_out; 106 } IscsiLun; 107 108 typedef struct IscsiTask { 109 int status; 110 int complete; 111 int retries; 112 int do_retry; 113 struct scsi_task *task; 114 Coroutine *co; 115 IscsiLun *iscsilun; 116 QEMUTimer retry_timer; 117 int err_code; 118 char *err_str; 119 } IscsiTask; 120 121 typedef struct IscsiAIOCB { 122 BlockAIOCB common; 123 QEMUBH *bh; 124 IscsiLun *iscsilun; 125 struct scsi_task *task; 126 int status; 127 int64_t sector_num; 128 int nb_sectors; 129 int ret; 130 #ifdef __linux__ 131 sg_io_hdr_t *ioh; 132 #endif 133 bool cancelled; 134 } IscsiAIOCB; 135 136 /* libiscsi uses time_t so its enough to process events every second */ 137 #define EVENT_INTERVAL 1000 138 #define NOP_INTERVAL 5000 139 #define MAX_NOP_FAILURES 3 140 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times) 141 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768}; 142 143 /* this threshold is a trade-off knob to choose between 144 * the potential additional overhead of an extra GET_LBA_STATUS request 145 * vs. unnecessarily reading a lot of zero sectors over the wire. 146 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES 147 * sectors we check the allocation status of the area covered by the 148 * request first if the allocationmap indicates that the area might be 149 * unallocated. */ 150 #define ISCSI_CHECKALLOC_THRES 64 151 152 #ifdef __linux__ 153 154 static void iscsi_bh_cb(void * p)155 iscsi_bh_cb(void *p) 156 { 157 IscsiAIOCB *acb = p; 158 159 qemu_bh_delete(acb->bh); 160 161 acb->common.cb(acb->common.opaque, acb->status); 162 163 if (acb->task != NULL) { 164 scsi_free_scsi_task(acb->task); 165 acb->task = NULL; 166 } 167 168 qemu_aio_unref(acb); 169 } 170 171 static void iscsi_schedule_bh(IscsiAIOCB * acb)172 iscsi_schedule_bh(IscsiAIOCB *acb) 173 { 174 if (acb->bh) { 175 return; 176 } 177 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb); 178 qemu_bh_schedule(acb->bh); 179 } 180 181 #endif 182 iscsi_co_generic_bh_cb(void * opaque)183 static void iscsi_co_generic_bh_cb(void *opaque) 184 { 185 struct IscsiTask *iTask = opaque; 186 187 iTask->complete = 1; 188 aio_co_wake(iTask->co); 189 } 190 iscsi_retry_timer_expired(void * opaque)191 static void iscsi_retry_timer_expired(void *opaque) 192 { 193 struct IscsiTask *iTask = opaque; 194 iTask->complete = 1; 195 if (iTask->co) { 196 aio_co_wake(iTask->co); 197 } 198 } 199 exp_random(double mean)200 static inline unsigned exp_random(double mean) 201 { 202 return -mean * log((double)rand() / RAND_MAX); 203 } 204 205 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in 206 * libiscsi 1.10.0, together with other constants we need. Use it as 207 * a hint that we have to define them ourselves if needed, to keep the 208 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for 209 * the test because SCSI_STATUS_* is an enum. 210 * 211 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes 212 * an enum, check against the LIBISCSI_API_VERSION macro, which was 213 * introduced in 1.11.0. If it is present, there is no need to define 214 * anything. 215 */ 216 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \ 217 !defined(LIBISCSI_API_VERSION) 218 #define SCSI_STATUS_TASK_SET_FULL 0x28 219 #define SCSI_STATUS_TIMEOUT 0x0f000002 220 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600 221 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00 222 #endif 223 224 #ifndef LIBISCSI_API_VERSION 225 #define LIBISCSI_API_VERSION 20130701 226 #endif 227 iscsi_translate_sense(struct scsi_sense * sense)228 static int iscsi_translate_sense(struct scsi_sense *sense) 229 { 230 return scsi_sense_to_errno(sense->key, 231 (sense->ascq & 0xFF00) >> 8, 232 sense->ascq & 0xFF); 233 } 234 235 /* Called (via iscsi_service) with QemuMutex held. */ 236 static void iscsi_co_generic_cb(struct iscsi_context * iscsi,int status,void * command_data,void * opaque)237 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status, 238 void *command_data, void *opaque) 239 { 240 struct IscsiTask *iTask = opaque; 241 struct scsi_task *task = command_data; 242 243 iTask->status = status; 244 iTask->do_retry = 0; 245 iTask->err_code = 0; 246 iTask->task = task; 247 248 if (status != SCSI_STATUS_GOOD) { 249 iTask->err_code = -EIO; 250 if (iTask->retries++ < ISCSI_CMD_RETRIES) { 251 if (status == SCSI_STATUS_BUSY || 252 status == SCSI_STATUS_TIMEOUT || 253 status == SCSI_STATUS_TASK_SET_FULL) { 254 unsigned retry_time = 255 exp_random(iscsi_retry_times[iTask->retries - 1]); 256 if (status == SCSI_STATUS_TIMEOUT) { 257 /* make sure the request is rescheduled AFTER the 258 * reconnect is initiated */ 259 retry_time = EVENT_INTERVAL * 2; 260 iTask->iscsilun->request_timed_out = true; 261 } 262 error_report("iSCSI Busy/TaskSetFull/TimeOut" 263 " (retry #%u in %u ms): %s", 264 iTask->retries, retry_time, 265 iscsi_get_error(iscsi)); 266 aio_timer_init(iTask->iscsilun->aio_context, 267 &iTask->retry_timer, QEMU_CLOCK_REALTIME, 268 SCALE_MS, iscsi_retry_timer_expired, iTask); 269 timer_mod(&iTask->retry_timer, 270 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time); 271 iTask->do_retry = 1; 272 return; 273 } else if (status == SCSI_STATUS_CHECK_CONDITION) { 274 int error = iscsi_translate_sense(&task->sense); 275 if (error == EAGAIN) { 276 error_report("iSCSI CheckCondition: %s", 277 iscsi_get_error(iscsi)); 278 iTask->do_retry = 1; 279 } else { 280 iTask->err_code = -error; 281 iTask->err_str = g_strdup(iscsi_get_error(iscsi)); 282 } 283 } 284 } 285 } 286 287 if (iTask->co) { 288 replay_bh_schedule_oneshot_event(iTask->iscsilun->aio_context, 289 iscsi_co_generic_bh_cb, iTask); 290 } else { 291 iTask->complete = 1; 292 } 293 } 294 295 static void coroutine_fn iscsi_co_init_iscsitask(IscsiLun * iscsilun,struct IscsiTask * iTask)296 iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask) 297 { 298 *iTask = (struct IscsiTask) { 299 .co = qemu_coroutine_self(), 300 .iscsilun = iscsilun, 301 }; 302 } 303 304 #ifdef __linux__ 305 306 /* Called (via iscsi_service) with QemuMutex held. */ 307 static void iscsi_abort_task_cb(struct iscsi_context * iscsi,int status,void * command_data,void * private_data)308 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data, 309 void *private_data) 310 { 311 IscsiAIOCB *acb = private_data; 312 313 /* If the command callback hasn't been called yet, drop the task */ 314 if (!acb->bh) { 315 /* Call iscsi_aio_ioctl_cb() with SCSI_STATUS_CANCELLED */ 316 iscsi_scsi_cancel_task(iscsi, acb->task); 317 } 318 319 qemu_aio_unref(acb); /* acquired in iscsi_aio_cancel() */ 320 } 321 322 static void iscsi_aio_cancel(BlockAIOCB * blockacb)323 iscsi_aio_cancel(BlockAIOCB *blockacb) 324 { 325 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb; 326 IscsiLun *iscsilun = acb->iscsilun; 327 328 WITH_QEMU_LOCK_GUARD(&iscsilun->mutex) { 329 330 /* If it was cancelled or completed already, our work is done here */ 331 if (acb->cancelled || acb->status != -EINPROGRESS) { 332 return; 333 } 334 335 acb->cancelled = true; 336 337 qemu_aio_ref(acb); /* released in iscsi_abort_task_cb() */ 338 339 /* send a task mgmt call to the target to cancel the task on the target */ 340 if (iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task, 341 iscsi_abort_task_cb, acb) < 0) { 342 qemu_aio_unref(acb); /* since iscsi_abort_task_cb() won't be called */ 343 } 344 } 345 } 346 347 static const AIOCBInfo iscsi_aiocb_info = { 348 .aiocb_size = sizeof(IscsiAIOCB), 349 .cancel_async = iscsi_aio_cancel, 350 }; 351 352 #endif 353 354 static void iscsi_process_read(void *arg); 355 static void iscsi_process_write(void *arg); 356 357 /* Called with QemuMutex held. */ 358 static void iscsi_set_events(IscsiLun * iscsilun)359 iscsi_set_events(IscsiLun *iscsilun) 360 { 361 struct iscsi_context *iscsi = iscsilun->iscsi; 362 int ev = iscsi_which_events(iscsi); 363 364 if (ev != iscsilun->events) { 365 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi), 366 (ev & POLLIN) ? iscsi_process_read : NULL, 367 (ev & POLLOUT) ? iscsi_process_write : NULL, 368 NULL, NULL, 369 iscsilun); 370 iscsilun->events = ev; 371 } 372 } 373 iscsi_timed_check_events(void * opaque)374 static void iscsi_timed_check_events(void *opaque) 375 { 376 IscsiLun *iscsilun = opaque; 377 378 WITH_QEMU_LOCK_GUARD(&iscsilun->mutex) { 379 /* check for timed out requests */ 380 iscsi_service(iscsilun->iscsi, 0); 381 382 if (iscsilun->request_timed_out) { 383 iscsilun->request_timed_out = false; 384 iscsi_reconnect(iscsilun->iscsi); 385 } 386 387 /* 388 * newer versions of libiscsi may return zero events. Ensure we are 389 * able to return to service once this situation changes. 390 */ 391 iscsi_set_events(iscsilun); 392 } 393 394 timer_mod(iscsilun->event_timer, 395 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL); 396 } 397 398 static void iscsi_process_read(void * arg)399 iscsi_process_read(void *arg) 400 { 401 IscsiLun *iscsilun = arg; 402 struct iscsi_context *iscsi = iscsilun->iscsi; 403 404 qemu_mutex_lock(&iscsilun->mutex); 405 iscsi_service(iscsi, POLLIN); 406 iscsi_set_events(iscsilun); 407 qemu_mutex_unlock(&iscsilun->mutex); 408 } 409 410 static void iscsi_process_write(void * arg)411 iscsi_process_write(void *arg) 412 { 413 IscsiLun *iscsilun = arg; 414 struct iscsi_context *iscsi = iscsilun->iscsi; 415 416 qemu_mutex_lock(&iscsilun->mutex); 417 iscsi_service(iscsi, POLLOUT); 418 iscsi_set_events(iscsilun); 419 qemu_mutex_unlock(&iscsilun->mutex); 420 } 421 sector_lun2qemu(int64_t sector,IscsiLun * iscsilun)422 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun) 423 { 424 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE; 425 } 426 sector_qemu2lun(int64_t sector,IscsiLun * iscsilun)427 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun) 428 { 429 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size; 430 } 431 is_byte_request_lun_aligned(int64_t offset,int64_t bytes,IscsiLun * iscsilun)432 static bool is_byte_request_lun_aligned(int64_t offset, int64_t bytes, 433 IscsiLun *iscsilun) 434 { 435 if (offset % iscsilun->block_size || bytes % iscsilun->block_size) { 436 error_report("iSCSI misaligned request: " 437 "iscsilun->block_size %u, offset %" PRIi64 438 ", bytes %" PRIi64, 439 iscsilun->block_size, offset, bytes); 440 return false; 441 } 442 return true; 443 } 444 is_sector_request_lun_aligned(int64_t sector_num,int nb_sectors,IscsiLun * iscsilun)445 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors, 446 IscsiLun *iscsilun) 447 { 448 assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS); 449 return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS, 450 nb_sectors << BDRV_SECTOR_BITS, 451 iscsilun); 452 } 453 iscsi_allocmap_free(IscsiLun * iscsilun)454 static void iscsi_allocmap_free(IscsiLun *iscsilun) 455 { 456 g_free(iscsilun->allocmap); 457 g_free(iscsilun->allocmap_valid); 458 iscsilun->allocmap = NULL; 459 iscsilun->allocmap_valid = NULL; 460 } 461 462 iscsi_allocmap_init(IscsiLun * iscsilun,int open_flags)463 static int iscsi_allocmap_init(IscsiLun *iscsilun, int open_flags) 464 { 465 iscsi_allocmap_free(iscsilun); 466 467 assert(iscsilun->cluster_size); 468 iscsilun->allocmap_size = 469 DIV_ROUND_UP(iscsilun->num_blocks * iscsilun->block_size, 470 iscsilun->cluster_size); 471 472 iscsilun->allocmap = bitmap_try_new(iscsilun->allocmap_size); 473 if (!iscsilun->allocmap) { 474 return -ENOMEM; 475 } 476 477 if (open_flags & BDRV_O_NOCACHE) { 478 /* when cache.direct = on all allocmap entries are 479 * treated as invalid to force a relookup of the block 480 * status on every read request */ 481 return 0; 482 } 483 484 iscsilun->allocmap_valid = bitmap_try_new(iscsilun->allocmap_size); 485 if (!iscsilun->allocmap_valid) { 486 /* if we are under memory pressure free the allocmap as well */ 487 iscsi_allocmap_free(iscsilun); 488 return -ENOMEM; 489 } 490 491 return 0; 492 } 493 494 static void iscsi_allocmap_update(IscsiLun * iscsilun,int64_t offset,int64_t bytes,bool allocated,bool valid)495 iscsi_allocmap_update(IscsiLun *iscsilun, int64_t offset, 496 int64_t bytes, bool allocated, bool valid) 497 { 498 int64_t cl_num_expanded, nb_cls_expanded, cl_num_shrunk, nb_cls_shrunk; 499 500 if (iscsilun->allocmap == NULL) { 501 return; 502 } 503 /* expand to entirely contain all affected clusters */ 504 assert(iscsilun->cluster_size); 505 cl_num_expanded = offset / iscsilun->cluster_size; 506 nb_cls_expanded = DIV_ROUND_UP(offset + bytes, 507 iscsilun->cluster_size) - cl_num_expanded; 508 /* shrink to touch only completely contained clusters */ 509 cl_num_shrunk = DIV_ROUND_UP(offset, iscsilun->cluster_size); 510 nb_cls_shrunk = (offset + bytes) / iscsilun->cluster_size - cl_num_shrunk; 511 if (allocated) { 512 bitmap_set(iscsilun->allocmap, cl_num_expanded, nb_cls_expanded); 513 } else { 514 if (nb_cls_shrunk > 0) { 515 bitmap_clear(iscsilun->allocmap, cl_num_shrunk, nb_cls_shrunk); 516 } 517 } 518 519 if (iscsilun->allocmap_valid == NULL) { 520 return; 521 } 522 if (valid) { 523 if (nb_cls_shrunk > 0) { 524 bitmap_set(iscsilun->allocmap_valid, cl_num_shrunk, nb_cls_shrunk); 525 } 526 } else { 527 bitmap_clear(iscsilun->allocmap_valid, cl_num_expanded, 528 nb_cls_expanded); 529 } 530 } 531 532 static void iscsi_allocmap_set_allocated(IscsiLun * iscsilun,int64_t offset,int64_t bytes)533 iscsi_allocmap_set_allocated(IscsiLun *iscsilun, int64_t offset, 534 int64_t bytes) 535 { 536 iscsi_allocmap_update(iscsilun, offset, bytes, true, true); 537 } 538 539 static void iscsi_allocmap_set_unallocated(IscsiLun * iscsilun,int64_t offset,int64_t bytes)540 iscsi_allocmap_set_unallocated(IscsiLun *iscsilun, int64_t offset, 541 int64_t bytes) 542 { 543 /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update 544 * is ignored, so this will in effect be an iscsi_allocmap_set_invalid. 545 */ 546 iscsi_allocmap_update(iscsilun, offset, bytes, false, true); 547 } 548 iscsi_allocmap_set_invalid(IscsiLun * iscsilun,int64_t offset,int64_t bytes)549 static void iscsi_allocmap_set_invalid(IscsiLun *iscsilun, int64_t offset, 550 int64_t bytes) 551 { 552 iscsi_allocmap_update(iscsilun, offset, bytes, false, false); 553 } 554 iscsi_allocmap_invalidate(IscsiLun * iscsilun)555 static void iscsi_allocmap_invalidate(IscsiLun *iscsilun) 556 { 557 if (iscsilun->allocmap) { 558 bitmap_zero(iscsilun->allocmap, iscsilun->allocmap_size); 559 } 560 if (iscsilun->allocmap_valid) { 561 bitmap_zero(iscsilun->allocmap_valid, iscsilun->allocmap_size); 562 } 563 } 564 565 static inline bool iscsi_allocmap_is_allocated(IscsiLun * iscsilun,int64_t offset,int64_t bytes)566 iscsi_allocmap_is_allocated(IscsiLun *iscsilun, int64_t offset, 567 int64_t bytes) 568 { 569 unsigned long size; 570 if (iscsilun->allocmap == NULL) { 571 return true; 572 } 573 assert(iscsilun->cluster_size); 574 size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size); 575 return !(find_next_bit(iscsilun->allocmap, size, 576 offset / iscsilun->cluster_size) == size); 577 } 578 iscsi_allocmap_is_valid(IscsiLun * iscsilun,int64_t offset,int64_t bytes)579 static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun, 580 int64_t offset, int64_t bytes) 581 { 582 unsigned long size; 583 if (iscsilun->allocmap_valid == NULL) { 584 return false; 585 } 586 assert(iscsilun->cluster_size); 587 size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size); 588 return (find_next_zero_bit(iscsilun->allocmap_valid, size, 589 offset / iscsilun->cluster_size) == size); 590 } 591 iscsi_co_wait_for_task(IscsiTask * iTask,IscsiLun * iscsilun)592 static void coroutine_fn iscsi_co_wait_for_task(IscsiTask *iTask, 593 IscsiLun *iscsilun) 594 { 595 while (!iTask->complete) { 596 iscsi_set_events(iscsilun); 597 qemu_mutex_unlock(&iscsilun->mutex); 598 qemu_coroutine_yield(); 599 qemu_mutex_lock(&iscsilun->mutex); 600 } 601 } 602 603 static int coroutine_fn iscsi_co_writev(BlockDriverState * bs,int64_t sector_num,int nb_sectors,QEMUIOVector * iov,int flags)604 iscsi_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors, 605 QEMUIOVector *iov, int flags) 606 { 607 IscsiLun *iscsilun = bs->opaque; 608 struct IscsiTask iTask; 609 uint64_t lba; 610 uint32_t num_sectors; 611 bool fua = flags & BDRV_REQ_FUA; 612 int r = 0; 613 614 if (fua) { 615 assert(iscsilun->dpofua); 616 } 617 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 618 return -EINVAL; 619 } 620 621 if (bs->bl.max_transfer) { 622 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer); 623 } 624 625 lba = sector_qemu2lun(sector_num, iscsilun); 626 num_sectors = sector_qemu2lun(nb_sectors, iscsilun); 627 iscsi_co_init_iscsitask(iscsilun, &iTask); 628 qemu_mutex_lock(&iscsilun->mutex); 629 retry: 630 if (iscsilun->use_16_for_rw) { 631 #if LIBISCSI_API_VERSION >= (20160603) 632 iTask.task = iscsi_write16_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 633 NULL, num_sectors * iscsilun->block_size, 634 iscsilun->block_size, 0, 0, fua, 0, 0, 635 iscsi_co_generic_cb, &iTask, 636 (struct scsi_iovec *)iov->iov, iov->niov); 637 } else { 638 iTask.task = iscsi_write10_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 639 NULL, num_sectors * iscsilun->block_size, 640 iscsilun->block_size, 0, 0, fua, 0, 0, 641 iscsi_co_generic_cb, &iTask, 642 (struct scsi_iovec *)iov->iov, iov->niov); 643 } 644 #else 645 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba, 646 NULL, num_sectors * iscsilun->block_size, 647 iscsilun->block_size, 0, 0, fua, 0, 0, 648 iscsi_co_generic_cb, &iTask); 649 } else { 650 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba, 651 NULL, num_sectors * iscsilun->block_size, 652 iscsilun->block_size, 0, 0, fua, 0, 0, 653 iscsi_co_generic_cb, &iTask); 654 } 655 #endif 656 if (iTask.task == NULL) { 657 qemu_mutex_unlock(&iscsilun->mutex); 658 return -ENOMEM; 659 } 660 #if LIBISCSI_API_VERSION < (20160603) 661 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov, 662 iov->niov); 663 #endif 664 iscsi_co_wait_for_task(&iTask, iscsilun); 665 666 if (iTask.task != NULL) { 667 scsi_free_scsi_task(iTask.task); 668 iTask.task = NULL; 669 } 670 671 if (iTask.do_retry) { 672 iTask.complete = 0; 673 goto retry; 674 } 675 676 if (iTask.status != SCSI_STATUS_GOOD) { 677 iscsi_allocmap_set_invalid(iscsilun, sector_num * BDRV_SECTOR_SIZE, 678 nb_sectors * BDRV_SECTOR_SIZE); 679 error_report("iSCSI WRITE10/16 failed at lba %" PRIu64 ": %s", lba, 680 iTask.err_str); 681 r = iTask.err_code; 682 goto out_unlock; 683 } 684 685 iscsi_allocmap_set_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE, 686 nb_sectors * BDRV_SECTOR_SIZE); 687 688 out_unlock: 689 qemu_mutex_unlock(&iscsilun->mutex); 690 g_free(iTask.err_str); 691 return r; 692 } 693 694 695 iscsi_co_block_status(BlockDriverState * bs,bool want_zero,int64_t offset,int64_t bytes,int64_t * pnum,int64_t * map,BlockDriverState ** file)696 static int coroutine_fn iscsi_co_block_status(BlockDriverState *bs, 697 bool want_zero, int64_t offset, 698 int64_t bytes, int64_t *pnum, 699 int64_t *map, 700 BlockDriverState **file) 701 { 702 IscsiLun *iscsilun = bs->opaque; 703 struct scsi_get_lba_status *lbas = NULL; 704 struct scsi_lba_status_descriptor *lbasd = NULL; 705 struct IscsiTask iTask; 706 uint64_t lba, max_bytes; 707 int ret; 708 709 iscsi_co_init_iscsitask(iscsilun, &iTask); 710 711 assert(QEMU_IS_ALIGNED(offset | bytes, iscsilun->block_size)); 712 713 /* default to all sectors allocated */ 714 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 715 if (map) { 716 *map = offset; 717 } 718 *pnum = bytes; 719 720 /* LUN does not support logical block provisioning */ 721 if (!iscsilun->lbpme) { 722 goto out; 723 } 724 725 lba = offset / iscsilun->block_size; 726 max_bytes = (iscsilun->num_blocks - lba) * iscsilun->block_size; 727 728 qemu_mutex_lock(&iscsilun->mutex); 729 retry: 730 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun, 731 lba, 8 + 16, iscsi_co_generic_cb, 732 &iTask) == NULL) { 733 ret = -ENOMEM; 734 goto out_unlock; 735 } 736 iscsi_co_wait_for_task(&iTask, iscsilun); 737 738 if (iTask.do_retry) { 739 if (iTask.task != NULL) { 740 scsi_free_scsi_task(iTask.task); 741 iTask.task = NULL; 742 } 743 iTask.complete = 0; 744 goto retry; 745 } 746 747 if (iTask.status != SCSI_STATUS_GOOD) { 748 /* in case the get_lba_status_callout fails (i.e. 749 * because the device is busy or the cmd is not 750 * supported) we pretend all blocks are allocated 751 * for backwards compatibility */ 752 error_report("iSCSI GET_LBA_STATUS failed at lba %" PRIu64 ": %s", 753 lba, iTask.err_str); 754 goto out_unlock; 755 } 756 757 lbas = scsi_datain_unmarshall(iTask.task); 758 if (lbas == NULL || lbas->num_descriptors == 0) { 759 ret = -EIO; 760 goto out_unlock; 761 } 762 763 lbasd = &lbas->descriptors[0]; 764 765 if (lba != lbasd->lba) { 766 ret = -EIO; 767 goto out_unlock; 768 } 769 770 *pnum = MIN((int64_t) lbasd->num_blocks * iscsilun->block_size, max_bytes); 771 772 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED || 773 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) { 774 ret &= ~BDRV_BLOCK_DATA; 775 if (iscsilun->lbprz) { 776 ret |= BDRV_BLOCK_ZERO; 777 } 778 } 779 780 if (ret & BDRV_BLOCK_ZERO) { 781 iscsi_allocmap_set_unallocated(iscsilun, offset, *pnum); 782 } else { 783 iscsi_allocmap_set_allocated(iscsilun, offset, *pnum); 784 } 785 786 out_unlock: 787 qemu_mutex_unlock(&iscsilun->mutex); 788 g_free(iTask.err_str); 789 out: 790 if (iTask.task != NULL) { 791 scsi_free_scsi_task(iTask.task); 792 } 793 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID && file) { 794 *file = bs; 795 } 796 return ret; 797 } 798 iscsi_co_readv(BlockDriverState * bs,int64_t sector_num,int nb_sectors,QEMUIOVector * iov)799 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs, 800 int64_t sector_num, int nb_sectors, 801 QEMUIOVector *iov) 802 { 803 IscsiLun *iscsilun = bs->opaque; 804 struct IscsiTask iTask; 805 uint64_t lba; 806 uint32_t num_sectors; 807 int r = 0; 808 809 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) { 810 return -EINVAL; 811 } 812 813 if (bs->bl.max_transfer) { 814 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer); 815 } 816 817 /* if cache.direct is off and we have a valid entry in our allocation map 818 * we can skip checking the block status and directly return zeroes if 819 * the request falls within an unallocated area */ 820 if (iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE, 821 nb_sectors * BDRV_SECTOR_SIZE) && 822 !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE, 823 nb_sectors * BDRV_SECTOR_SIZE)) { 824 qemu_iovec_memset(iov, 0, 0x00, iov->size); 825 return 0; 826 } 827 828 if (nb_sectors >= ISCSI_CHECKALLOC_THRES && 829 !iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE, 830 nb_sectors * BDRV_SECTOR_SIZE) && 831 !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE, 832 nb_sectors * BDRV_SECTOR_SIZE)) { 833 int64_t pnum; 834 /* check the block status from the beginning of the cluster 835 * containing the start sector */ 836 int64_t head; 837 int ret; 838 839 assert(iscsilun->cluster_size); 840 head = (sector_num * BDRV_SECTOR_SIZE) % iscsilun->cluster_size; 841 ret = iscsi_co_block_status(bs, true, 842 sector_num * BDRV_SECTOR_SIZE - head, 843 BDRV_REQUEST_MAX_BYTES, &pnum, NULL, NULL); 844 if (ret < 0) { 845 return ret; 846 } 847 /* if the whole request falls into an unallocated area we can avoid 848 * reading and directly return zeroes instead */ 849 if (ret & BDRV_BLOCK_ZERO && 850 pnum >= nb_sectors * BDRV_SECTOR_SIZE + head) { 851 qemu_iovec_memset(iov, 0, 0x00, iov->size); 852 return 0; 853 } 854 } 855 856 lba = sector_qemu2lun(sector_num, iscsilun); 857 num_sectors = sector_qemu2lun(nb_sectors, iscsilun); 858 859 iscsi_co_init_iscsitask(iscsilun, &iTask); 860 qemu_mutex_lock(&iscsilun->mutex); 861 retry: 862 if (iscsilun->use_16_for_rw) { 863 #if LIBISCSI_API_VERSION >= (20160603) 864 iTask.task = iscsi_read16_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 865 num_sectors * iscsilun->block_size, 866 iscsilun->block_size, 0, 0, 0, 0, 0, 867 iscsi_co_generic_cb, &iTask, 868 (struct scsi_iovec *)iov->iov, iov->niov); 869 } else { 870 iTask.task = iscsi_read10_iov_task(iscsilun->iscsi, iscsilun->lun, lba, 871 num_sectors * iscsilun->block_size, 872 iscsilun->block_size, 873 0, 0, 0, 0, 0, 874 iscsi_co_generic_cb, &iTask, 875 (struct scsi_iovec *)iov->iov, iov->niov); 876 } 877 #else 878 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba, 879 num_sectors * iscsilun->block_size, 880 iscsilun->block_size, 0, 0, 0, 0, 0, 881 iscsi_co_generic_cb, &iTask); 882 } else { 883 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba, 884 num_sectors * iscsilun->block_size, 885 iscsilun->block_size, 886 0, 0, 0, 0, 0, 887 iscsi_co_generic_cb, &iTask); 888 } 889 #endif 890 if (iTask.task == NULL) { 891 qemu_mutex_unlock(&iscsilun->mutex); 892 return -ENOMEM; 893 } 894 #if LIBISCSI_API_VERSION < (20160603) 895 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov); 896 #endif 897 898 iscsi_co_wait_for_task(&iTask, iscsilun); 899 if (iTask.task != NULL) { 900 scsi_free_scsi_task(iTask.task); 901 iTask.task = NULL; 902 } 903 904 if (iTask.do_retry) { 905 iTask.complete = 0; 906 goto retry; 907 } 908 909 if (iTask.status != SCSI_STATUS_GOOD) { 910 error_report("iSCSI READ10/16 failed at lba %" PRIu64 ": %s", 911 lba, iTask.err_str); 912 r = iTask.err_code; 913 } 914 915 qemu_mutex_unlock(&iscsilun->mutex); 916 g_free(iTask.err_str); 917 return r; 918 } 919 iscsi_co_flush(BlockDriverState * bs)920 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs) 921 { 922 IscsiLun *iscsilun = bs->opaque; 923 struct IscsiTask iTask; 924 int r = 0; 925 926 iscsi_co_init_iscsitask(iscsilun, &iTask); 927 qemu_mutex_lock(&iscsilun->mutex); 928 retry: 929 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0, 930 0, iscsi_co_generic_cb, &iTask) == NULL) { 931 qemu_mutex_unlock(&iscsilun->mutex); 932 return -ENOMEM; 933 } 934 935 iscsi_co_wait_for_task(&iTask, iscsilun); 936 937 if (iTask.task != NULL) { 938 scsi_free_scsi_task(iTask.task); 939 iTask.task = NULL; 940 } 941 942 if (iTask.do_retry) { 943 iTask.complete = 0; 944 goto retry; 945 } 946 947 if (iTask.status != SCSI_STATUS_GOOD) { 948 error_report("iSCSI SYNCHRONIZECACHE10 failed: %s", iTask.err_str); 949 r = iTask.err_code; 950 } 951 952 qemu_mutex_unlock(&iscsilun->mutex); 953 g_free(iTask.err_str); 954 return r; 955 } 956 957 #ifdef __linux__ 958 /* Called (via iscsi_service) with QemuMutex held. */ 959 static void iscsi_aio_ioctl_cb(struct iscsi_context * iscsi,int status,void * command_data,void * opaque)960 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status, 961 void *command_data, void *opaque) 962 { 963 IscsiAIOCB *acb = opaque; 964 965 if (status == SCSI_STATUS_CANCELLED) { 966 if (!acb->bh) { 967 acb->status = -ECANCELED; 968 iscsi_schedule_bh(acb); 969 } 970 return; 971 } 972 973 acb->status = 0; 974 if (status < 0) { 975 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s", 976 iscsi_get_error(iscsi)); 977 acb->status = -iscsi_translate_sense(&acb->task->sense); 978 } 979 980 acb->ioh->driver_status = 0; 981 acb->ioh->host_status = 0; 982 acb->ioh->resid = 0; 983 acb->ioh->status = status; 984 985 #define SG_ERR_DRIVER_SENSE 0x08 986 987 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) { 988 int ss; 989 990 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE; 991 992 acb->ioh->sb_len_wr = acb->task->datain.size - 2; 993 ss = MIN(acb->ioh->mx_sb_len, acb->ioh->sb_len_wr); 994 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss); 995 } 996 997 iscsi_schedule_bh(acb); 998 } 999 iscsi_ioctl_bh_completion(void * opaque)1000 static void iscsi_ioctl_bh_completion(void *opaque) 1001 { 1002 IscsiAIOCB *acb = opaque; 1003 1004 qemu_bh_delete(acb->bh); 1005 acb->common.cb(acb->common.opaque, acb->ret); 1006 qemu_aio_unref(acb); 1007 } 1008 iscsi_ioctl_handle_emulated(IscsiAIOCB * acb,int req,void * buf)1009 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf) 1010 { 1011 BlockDriverState *bs = acb->common.bs; 1012 IscsiLun *iscsilun = bs->opaque; 1013 int ret = 0; 1014 1015 switch (req) { 1016 case SG_GET_VERSION_NUM: 1017 *(int *)buf = 30000; 1018 break; 1019 case SG_GET_SCSI_ID: 1020 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type; 1021 break; 1022 default: 1023 ret = -EINVAL; 1024 } 1025 assert(!acb->bh); 1026 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), 1027 iscsi_ioctl_bh_completion, acb); 1028 acb->ret = ret; 1029 qemu_bh_schedule(acb->bh); 1030 } 1031 iscsi_aio_ioctl(BlockDriverState * bs,unsigned long int req,void * buf,BlockCompletionFunc * cb,void * opaque)1032 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs, 1033 unsigned long int req, void *buf, 1034 BlockCompletionFunc *cb, void *opaque) 1035 { 1036 IscsiLun *iscsilun = bs->opaque; 1037 struct iscsi_context *iscsi = iscsilun->iscsi; 1038 struct iscsi_data data; 1039 IscsiAIOCB *acb; 1040 1041 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque); 1042 1043 acb->iscsilun = iscsilun; 1044 acb->bh = NULL; 1045 acb->status = -EINPROGRESS; 1046 acb->ioh = buf; 1047 acb->cancelled = false; 1048 1049 if (req != SG_IO) { 1050 iscsi_ioctl_handle_emulated(acb, req, buf); 1051 return &acb->common; 1052 } 1053 1054 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) { 1055 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)", 1056 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE); 1057 qemu_aio_unref(acb); 1058 return NULL; 1059 } 1060 1061 /* Must use malloc(): this is freed via scsi_free_scsi_task() */ 1062 acb->task = malloc(sizeof(struct scsi_task)); 1063 if (acb->task == NULL) { 1064 error_report("iSCSI: Failed to allocate task for scsi command. %s", 1065 iscsi_get_error(iscsi)); 1066 qemu_aio_unref(acb); 1067 return NULL; 1068 } 1069 memset(acb->task, 0, sizeof(struct scsi_task)); 1070 1071 switch (acb->ioh->dxfer_direction) { 1072 case SG_DXFER_TO_DEV: 1073 acb->task->xfer_dir = SCSI_XFER_WRITE; 1074 break; 1075 case SG_DXFER_FROM_DEV: 1076 acb->task->xfer_dir = SCSI_XFER_READ; 1077 break; 1078 default: 1079 acb->task->xfer_dir = SCSI_XFER_NONE; 1080 break; 1081 } 1082 1083 acb->task->cdb_size = acb->ioh->cmd_len; 1084 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len); 1085 acb->task->expxferlen = acb->ioh->dxfer_len; 1086 1087 data.size = 0; 1088 qemu_mutex_lock(&iscsilun->mutex); 1089 if (acb->task->xfer_dir == SCSI_XFER_WRITE) { 1090 if (acb->ioh->iovec_count == 0) { 1091 data.data = acb->ioh->dxferp; 1092 data.size = acb->ioh->dxfer_len; 1093 } else { 1094 scsi_task_set_iov_out(acb->task, 1095 (struct scsi_iovec *) acb->ioh->dxferp, 1096 acb->ioh->iovec_count); 1097 } 1098 } 1099 1100 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task, 1101 iscsi_aio_ioctl_cb, 1102 (data.size > 0) ? &data : NULL, 1103 acb) != 0) { 1104 qemu_mutex_unlock(&iscsilun->mutex); 1105 scsi_free_scsi_task(acb->task); 1106 qemu_aio_unref(acb); 1107 return NULL; 1108 } 1109 1110 /* tell libiscsi to read straight into the buffer we got from ioctl */ 1111 if (acb->task->xfer_dir == SCSI_XFER_READ) { 1112 if (acb->ioh->iovec_count == 0) { 1113 scsi_task_add_data_in_buffer(acb->task, 1114 acb->ioh->dxfer_len, 1115 acb->ioh->dxferp); 1116 } else { 1117 scsi_task_set_iov_in(acb->task, 1118 (struct scsi_iovec *) acb->ioh->dxferp, 1119 acb->ioh->iovec_count); 1120 } 1121 } 1122 1123 iscsi_set_events(iscsilun); 1124 qemu_mutex_unlock(&iscsilun->mutex); 1125 1126 return &acb->common; 1127 } 1128 1129 #endif 1130 1131 static int64_t coroutine_fn iscsi_co_getlength(BlockDriverState * bs)1132 iscsi_co_getlength(BlockDriverState *bs) 1133 { 1134 IscsiLun *iscsilun = bs->opaque; 1135 int64_t len; 1136 1137 len = iscsilun->num_blocks; 1138 len *= iscsilun->block_size; 1139 1140 return len; 1141 } 1142 1143 static int iscsi_co_pdiscard(BlockDriverState * bs,int64_t offset,int64_t bytes)1144 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, 1145 int64_t bytes) 1146 { 1147 IscsiLun *iscsilun = bs->opaque; 1148 struct IscsiTask iTask; 1149 struct unmap_list list; 1150 int r = 0; 1151 1152 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) { 1153 return -ENOTSUP; 1154 } 1155 1156 if (!iscsilun->lbp.lbpu) { 1157 /* UNMAP is not supported by the target */ 1158 return 0; 1159 } 1160 1161 /* 1162 * We don't want to overflow list.num which is uint32_t. 1163 * We rely on our max_pdiscard. 1164 */ 1165 assert(bytes / iscsilun->block_size <= UINT32_MAX); 1166 1167 list.lba = offset / iscsilun->block_size; 1168 list.num = bytes / iscsilun->block_size; 1169 1170 iscsi_co_init_iscsitask(iscsilun, &iTask); 1171 qemu_mutex_lock(&iscsilun->mutex); 1172 retry: 1173 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1, 1174 iscsi_co_generic_cb, &iTask) == NULL) { 1175 r = -ENOMEM; 1176 goto out_unlock; 1177 } 1178 1179 iscsi_co_wait_for_task(&iTask, iscsilun); 1180 1181 if (iTask.task != NULL) { 1182 scsi_free_scsi_task(iTask.task); 1183 iTask.task = NULL; 1184 } 1185 1186 if (iTask.do_retry) { 1187 iTask.complete = 0; 1188 goto retry; 1189 } 1190 1191 iscsi_allocmap_set_invalid(iscsilun, offset, bytes); 1192 1193 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) { 1194 /* the target might fail with a check condition if it 1195 is not happy with the alignment of the UNMAP request 1196 we silently fail in this case */ 1197 goto out_unlock; 1198 } 1199 1200 if (iTask.status != SCSI_STATUS_GOOD) { 1201 error_report("iSCSI UNMAP failed at lba %" PRIu64 ": %s", 1202 list.lba, iTask.err_str); 1203 r = iTask.err_code; 1204 goto out_unlock; 1205 } 1206 1207 out_unlock: 1208 qemu_mutex_unlock(&iscsilun->mutex); 1209 g_free(iTask.err_str); 1210 return r; 1211 } 1212 1213 static int iscsi_co_pwrite_zeroes(BlockDriverState * bs,int64_t offset,int64_t bytes,BdrvRequestFlags flags)1214 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, 1215 int64_t bytes, BdrvRequestFlags flags) 1216 { 1217 IscsiLun *iscsilun = bs->opaque; 1218 struct IscsiTask iTask; 1219 uint64_t lba; 1220 uint64_t nb_blocks; 1221 bool use_16_for_ws = iscsilun->use_16_for_rw; 1222 int r = 0; 1223 1224 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) { 1225 return -ENOTSUP; 1226 } 1227 1228 if (flags & BDRV_REQ_MAY_UNMAP) { 1229 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) { 1230 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */ 1231 use_16_for_ws = true; 1232 } 1233 if (use_16_for_ws && !iscsilun->lbp.lbpws) { 1234 /* WRITESAME16 with UNMAP is not supported by the target, 1235 * fall back and try WRITESAME10/16 without UNMAP */ 1236 flags &= ~BDRV_REQ_MAY_UNMAP; 1237 use_16_for_ws = iscsilun->use_16_for_rw; 1238 } 1239 } 1240 1241 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) { 1242 /* WRITESAME without UNMAP is not supported by the target */ 1243 return -ENOTSUP; 1244 } 1245 1246 lba = offset / iscsilun->block_size; 1247 nb_blocks = bytes / iscsilun->block_size; 1248 1249 if (iscsilun->zeroblock == NULL) { 1250 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size); 1251 if (iscsilun->zeroblock == NULL) { 1252 return -ENOMEM; 1253 } 1254 } 1255 1256 qemu_mutex_lock(&iscsilun->mutex); 1257 iscsi_co_init_iscsitask(iscsilun, &iTask); 1258 retry: 1259 if (use_16_for_ws) { 1260 /* 1261 * iscsi_writesame16_task num_blocks argument is uint32_t. We rely here 1262 * on our max_pwrite_zeroes limit. 1263 */ 1264 assert(nb_blocks <= UINT32_MAX); 1265 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba, 1266 iscsilun->zeroblock, iscsilun->block_size, 1267 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 1268 0, 0, iscsi_co_generic_cb, &iTask); 1269 } else { 1270 /* 1271 * iscsi_writesame10_task num_blocks argument is uint16_t. We rely here 1272 * on our max_pwrite_zeroes limit. 1273 */ 1274 assert(nb_blocks <= UINT16_MAX); 1275 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba, 1276 iscsilun->zeroblock, iscsilun->block_size, 1277 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP), 1278 0, 0, iscsi_co_generic_cb, &iTask); 1279 } 1280 if (iTask.task == NULL) { 1281 qemu_mutex_unlock(&iscsilun->mutex); 1282 return -ENOMEM; 1283 } 1284 1285 iscsi_co_wait_for_task(&iTask, iscsilun); 1286 1287 if (iTask.status == SCSI_STATUS_CHECK_CONDITION && 1288 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST && 1289 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE || 1290 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) { 1291 /* WRITE SAME is not supported by the target */ 1292 iscsilun->has_write_same = false; 1293 scsi_free_scsi_task(iTask.task); 1294 r = -ENOTSUP; 1295 goto out_unlock; 1296 } 1297 1298 if (iTask.task != NULL) { 1299 scsi_free_scsi_task(iTask.task); 1300 iTask.task = NULL; 1301 } 1302 1303 if (iTask.do_retry) { 1304 iTask.complete = 0; 1305 goto retry; 1306 } 1307 1308 if (iTask.status != SCSI_STATUS_GOOD) { 1309 iscsi_allocmap_set_invalid(iscsilun, offset, bytes); 1310 error_report("iSCSI WRITESAME10/16 failed at lba %" PRIu64 ": %s", 1311 lba, iTask.err_str); 1312 r = iTask.err_code; 1313 goto out_unlock; 1314 } 1315 1316 if (flags & BDRV_REQ_MAY_UNMAP) { 1317 iscsi_allocmap_set_invalid(iscsilun, offset, bytes); 1318 } else { 1319 iscsi_allocmap_set_allocated(iscsilun, offset, bytes); 1320 } 1321 1322 out_unlock: 1323 qemu_mutex_unlock(&iscsilun->mutex); 1324 g_free(iTask.err_str); 1325 return r; 1326 } 1327 apply_chap(struct iscsi_context * iscsi,QemuOpts * opts,Error ** errp)1328 static void apply_chap(struct iscsi_context *iscsi, QemuOpts *opts, 1329 Error **errp) 1330 { 1331 const char *user = NULL; 1332 const char *password = NULL; 1333 const char *secretid; 1334 char *secret = NULL; 1335 1336 user = qemu_opt_get(opts, "user"); 1337 if (!user) { 1338 return; 1339 } 1340 1341 secretid = qemu_opt_get(opts, "password-secret"); 1342 password = qemu_opt_get(opts, "password"); 1343 if (secretid && password) { 1344 error_setg(errp, "'password' and 'password-secret' properties are " 1345 "mutually exclusive"); 1346 return; 1347 } 1348 if (secretid) { 1349 secret = qcrypto_secret_lookup_as_utf8(secretid, errp); 1350 if (!secret) { 1351 return; 1352 } 1353 password = secret; 1354 } else if (!password) { 1355 error_setg(errp, "CHAP username specified but no password was given"); 1356 return; 1357 } else { 1358 warn_report("iSCSI block driver 'password' option is deprecated, " 1359 "use 'password-secret' instead"); 1360 } 1361 1362 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) { 1363 error_setg(errp, "Failed to set initiator username and password"); 1364 } 1365 1366 g_free(secret); 1367 } 1368 apply_header_digest(struct iscsi_context * iscsi,QemuOpts * opts,Error ** errp)1369 static void apply_header_digest(struct iscsi_context *iscsi, QemuOpts *opts, 1370 Error **errp) 1371 { 1372 const char *digest = NULL; 1373 1374 digest = qemu_opt_get(opts, "header-digest"); 1375 if (!digest) { 1376 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1377 } else if (!strcmp(digest, "crc32c")) { 1378 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C); 1379 } else if (!strcmp(digest, "none")) { 1380 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE); 1381 } else if (!strcmp(digest, "crc32c-none")) { 1382 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE); 1383 } else if (!strcmp(digest, "none-crc32c")) { 1384 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C); 1385 } else { 1386 error_setg(errp, "Invalid header-digest setting : %s", digest); 1387 } 1388 } 1389 get_initiator_name(QemuOpts * opts)1390 static char *get_initiator_name(QemuOpts *opts) 1391 { 1392 const char *name; 1393 char *iscsi_name; 1394 UuidInfo *uuid_info; 1395 1396 name = qemu_opt_get(opts, "initiator-name"); 1397 if (name) { 1398 return g_strdup(name); 1399 } 1400 1401 uuid_info = qmp_query_uuid(NULL); 1402 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) { 1403 name = qemu_get_vm_name(); 1404 } else { 1405 name = uuid_info->UUID; 1406 } 1407 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s", 1408 name ? ":" : "", name ? name : ""); 1409 qapi_free_UuidInfo(uuid_info); 1410 return iscsi_name; 1411 } 1412 iscsi_nop_timed_event(void * opaque)1413 static void iscsi_nop_timed_event(void *opaque) 1414 { 1415 IscsiLun *iscsilun = opaque; 1416 1417 QEMU_LOCK_GUARD(&iscsilun->mutex); 1418 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) { 1419 error_report("iSCSI: NOP timeout. Reconnecting..."); 1420 iscsilun->request_timed_out = true; 1421 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) { 1422 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages."); 1423 return; 1424 } 1425 1426 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1427 iscsi_set_events(iscsilun); 1428 } 1429 iscsi_readcapacity_sync(IscsiLun * iscsilun,Error ** errp)1430 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp) 1431 { 1432 struct scsi_task *task = NULL; 1433 struct scsi_readcapacity10 *rc10 = NULL; 1434 struct scsi_readcapacity16 *rc16 = NULL; 1435 int retries = ISCSI_CMD_RETRIES; 1436 1437 do { 1438 if (task != NULL) { 1439 scsi_free_scsi_task(task); 1440 task = NULL; 1441 } 1442 1443 switch (iscsilun->type) { 1444 case TYPE_DISK: 1445 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun); 1446 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1447 rc16 = scsi_datain_unmarshall(task); 1448 if (rc16 == NULL) { 1449 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data."); 1450 } else { 1451 iscsilun->block_size = rc16->block_length; 1452 iscsilun->num_blocks = rc16->returned_lba + 1; 1453 iscsilun->lbpme = !!rc16->lbpme; 1454 iscsilun->lbprz = !!rc16->lbprz; 1455 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff); 1456 } 1457 break; 1458 } 1459 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION 1460 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) { 1461 break; 1462 } 1463 /* Fall through and try READ CAPACITY(10) instead. */ 1464 case TYPE_ROM: 1465 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0); 1466 if (task != NULL && task->status == SCSI_STATUS_GOOD) { 1467 rc10 = scsi_datain_unmarshall(task); 1468 if (rc10 == NULL) { 1469 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data."); 1470 } else { 1471 iscsilun->block_size = rc10->block_size; 1472 if (rc10->lba == 0) { 1473 /* blank disk loaded */ 1474 iscsilun->num_blocks = 0; 1475 } else { 1476 iscsilun->num_blocks = rc10->lba + 1; 1477 } 1478 } 1479 } 1480 break; 1481 default: 1482 return; 1483 } 1484 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION 1485 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION 1486 && retries-- > 0); 1487 1488 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1489 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command"); 1490 } else if (!iscsilun->block_size || 1491 iscsilun->block_size % BDRV_SECTOR_SIZE) { 1492 error_setg(errp, "iSCSI: the target returned an invalid " 1493 "block size of %d.", iscsilun->block_size); 1494 } 1495 if (task) { 1496 scsi_free_scsi_task(task); 1497 } 1498 } 1499 iscsi_do_inquiry(struct iscsi_context * iscsi,int lun,int evpd,int pc,void ** inq,Error ** errp)1500 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun, 1501 int evpd, int pc, void **inq, Error **errp) 1502 { 1503 int full_size; 1504 struct scsi_task *task = NULL; 1505 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64); 1506 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1507 goto fail; 1508 } 1509 full_size = scsi_datain_getfullsize(task); 1510 if (full_size > task->datain.size) { 1511 scsi_free_scsi_task(task); 1512 1513 /* we need more data for the full list */ 1514 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size); 1515 if (task == NULL || task->status != SCSI_STATUS_GOOD) { 1516 goto fail; 1517 } 1518 } 1519 1520 *inq = scsi_datain_unmarshall(task); 1521 if (*inq == NULL) { 1522 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob"); 1523 goto fail_with_err; 1524 } 1525 1526 return task; 1527 1528 fail: 1529 error_setg(errp, "iSCSI: Inquiry command failed : %s", 1530 iscsi_get_error(iscsi)); 1531 fail_with_err: 1532 if (task != NULL) { 1533 scsi_free_scsi_task(task); 1534 } 1535 return NULL; 1536 } 1537 iscsi_detach_aio_context(BlockDriverState * bs)1538 static void iscsi_detach_aio_context(BlockDriverState *bs) 1539 { 1540 IscsiLun *iscsilun = bs->opaque; 1541 1542 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi), 1543 NULL, NULL, NULL, NULL, NULL); 1544 iscsilun->events = 0; 1545 1546 if (iscsilun->nop_timer) { 1547 timer_free(iscsilun->nop_timer); 1548 iscsilun->nop_timer = NULL; 1549 } 1550 if (iscsilun->event_timer) { 1551 timer_free(iscsilun->event_timer); 1552 iscsilun->event_timer = NULL; 1553 } 1554 } 1555 iscsi_attach_aio_context(BlockDriverState * bs,AioContext * new_context)1556 static void iscsi_attach_aio_context(BlockDriverState *bs, 1557 AioContext *new_context) 1558 { 1559 IscsiLun *iscsilun = bs->opaque; 1560 1561 iscsilun->aio_context = new_context; 1562 iscsi_set_events(iscsilun); 1563 1564 /* Set up a timer for sending out iSCSI NOPs */ 1565 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context, 1566 QEMU_CLOCK_REALTIME, SCALE_MS, 1567 iscsi_nop_timed_event, iscsilun); 1568 timer_mod(iscsilun->nop_timer, 1569 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL); 1570 1571 /* Set up a timer for periodic calls to iscsi_set_events and to 1572 * scan for command timeout */ 1573 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context, 1574 QEMU_CLOCK_REALTIME, SCALE_MS, 1575 iscsi_timed_check_events, iscsilun); 1576 timer_mod(iscsilun->event_timer, 1577 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL); 1578 } 1579 iscsi_modesense_sync(IscsiLun * iscsilun)1580 static void iscsi_modesense_sync(IscsiLun *iscsilun) 1581 { 1582 struct scsi_task *task; 1583 struct scsi_mode_sense *ms = NULL; 1584 iscsilun->write_protected = false; 1585 iscsilun->dpofua = false; 1586 1587 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun, 1588 1, SCSI_MODESENSE_PC_CURRENT, 1589 0x3F, 0, 255); 1590 if (task == NULL) { 1591 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s", 1592 iscsi_get_error(iscsilun->iscsi)); 1593 goto out; 1594 } 1595 1596 if (task->status != SCSI_STATUS_GOOD) { 1597 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable"); 1598 goto out; 1599 } 1600 ms = scsi_datain_unmarshall(task); 1601 if (!ms) { 1602 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s", 1603 iscsi_get_error(iscsilun->iscsi)); 1604 goto out; 1605 } 1606 iscsilun->write_protected = ms->device_specific_parameter & 0x80; 1607 iscsilun->dpofua = ms->device_specific_parameter & 0x10; 1608 1609 out: 1610 if (task) { 1611 scsi_free_scsi_task(task); 1612 } 1613 } 1614 iscsi_parse_iscsi_option(const char * target,QDict * options)1615 static void iscsi_parse_iscsi_option(const char *target, QDict *options) 1616 { 1617 QemuOptsList *list; 1618 QemuOpts *opts; 1619 const char *user, *password, *password_secret, *initiator_name, 1620 *header_digest, *timeout; 1621 1622 list = qemu_find_opts("iscsi"); 1623 if (!list) { 1624 return; 1625 } 1626 1627 opts = qemu_opts_find(list, target); 1628 if (opts == NULL) { 1629 opts = QTAILQ_FIRST(&list->head); 1630 if (!opts) { 1631 return; 1632 } 1633 } 1634 1635 user = qemu_opt_get(opts, "user"); 1636 if (user) { 1637 qdict_set_default_str(options, "user", user); 1638 } 1639 1640 password = qemu_opt_get(opts, "password"); 1641 if (password) { 1642 qdict_set_default_str(options, "password", password); 1643 } 1644 1645 password_secret = qemu_opt_get(opts, "password-secret"); 1646 if (password_secret) { 1647 qdict_set_default_str(options, "password-secret", password_secret); 1648 } 1649 1650 initiator_name = qemu_opt_get(opts, "initiator-name"); 1651 if (initiator_name) { 1652 qdict_set_default_str(options, "initiator-name", initiator_name); 1653 } 1654 1655 header_digest = qemu_opt_get(opts, "header-digest"); 1656 if (header_digest) { 1657 /* -iscsi takes upper case values, but QAPI only supports lower case 1658 * enum constant names, so we have to convert here. */ 1659 char *qapi_value = g_ascii_strdown(header_digest, -1); 1660 qdict_set_default_str(options, "header-digest", qapi_value); 1661 g_free(qapi_value); 1662 } 1663 1664 timeout = qemu_opt_get(opts, "timeout"); 1665 if (timeout) { 1666 qdict_set_default_str(options, "timeout", timeout); 1667 } 1668 } 1669 1670 /* 1671 * We support iscsi url's on the form 1672 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun> 1673 */ iscsi_parse_filename(const char * filename,QDict * options,Error ** errp)1674 static void iscsi_parse_filename(const char *filename, QDict *options, 1675 Error **errp) 1676 { 1677 struct iscsi_url *iscsi_url; 1678 const char *transport_name; 1679 char *lun_str; 1680 1681 iscsi_url = iscsi_parse_full_url(NULL, filename); 1682 if (iscsi_url == NULL) { 1683 error_setg(errp, "Failed to parse URL : %s", filename); 1684 return; 1685 } 1686 1687 #if LIBISCSI_API_VERSION >= (20160603) 1688 switch (iscsi_url->transport) { 1689 case TCP_TRANSPORT: 1690 transport_name = "tcp"; 1691 break; 1692 case ISER_TRANSPORT: 1693 transport_name = "iser"; 1694 break; 1695 default: 1696 error_setg(errp, "Unknown transport type (%d)", 1697 iscsi_url->transport); 1698 return; 1699 } 1700 #else 1701 transport_name = "tcp"; 1702 #endif 1703 1704 qdict_set_default_str(options, "transport", transport_name); 1705 qdict_set_default_str(options, "portal", iscsi_url->portal); 1706 qdict_set_default_str(options, "target", iscsi_url->target); 1707 1708 lun_str = g_strdup_printf("%d", iscsi_url->lun); 1709 qdict_set_default_str(options, "lun", lun_str); 1710 g_free(lun_str); 1711 1712 /* User/password from -iscsi take precedence over those from the URL */ 1713 iscsi_parse_iscsi_option(iscsi_url->target, options); 1714 1715 if (iscsi_url->user[0] != '\0') { 1716 qdict_set_default_str(options, "user", iscsi_url->user); 1717 qdict_set_default_str(options, "password", iscsi_url->passwd); 1718 } 1719 1720 iscsi_destroy_url(iscsi_url); 1721 } 1722 1723 static QemuOptsList runtime_opts = { 1724 .name = "iscsi", 1725 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), 1726 .desc = { 1727 { 1728 .name = "transport", 1729 .type = QEMU_OPT_STRING, 1730 }, 1731 { 1732 .name = "portal", 1733 .type = QEMU_OPT_STRING, 1734 }, 1735 { 1736 .name = "target", 1737 .type = QEMU_OPT_STRING, 1738 }, 1739 { 1740 .name = "user", 1741 .type = QEMU_OPT_STRING, 1742 }, 1743 { 1744 .name = "password", 1745 .type = QEMU_OPT_STRING, 1746 }, 1747 { 1748 .name = "password-secret", 1749 .type = QEMU_OPT_STRING, 1750 }, 1751 { 1752 .name = "lun", 1753 .type = QEMU_OPT_NUMBER, 1754 }, 1755 { 1756 .name = "initiator-name", 1757 .type = QEMU_OPT_STRING, 1758 }, 1759 { 1760 .name = "header-digest", 1761 .type = QEMU_OPT_STRING, 1762 }, 1763 { 1764 .name = "timeout", 1765 .type = QEMU_OPT_NUMBER, 1766 }, 1767 { /* end of list */ } 1768 }, 1769 }; 1770 iscsi_save_designator(IscsiLun * lun,struct scsi_inquiry_device_identification * inq_di)1771 static void iscsi_save_designator(IscsiLun *lun, 1772 struct scsi_inquiry_device_identification *inq_di) 1773 { 1774 struct scsi_inquiry_device_designator *desig, *copy = NULL; 1775 1776 for (desig = inq_di->designators; desig; desig = desig->next) { 1777 if (desig->association || 1778 desig->designator_type > SCSI_DESIGNATOR_TYPE_NAA) { 1779 continue; 1780 } 1781 /* NAA works better than T10 vendor ID based designator. */ 1782 if (!copy || copy->designator_type < desig->designator_type) { 1783 copy = desig; 1784 } 1785 } 1786 if (copy) { 1787 lun->dd = g_new(struct scsi_inquiry_device_designator, 1); 1788 *lun->dd = *copy; 1789 lun->dd->next = NULL; 1790 lun->dd->designator = g_malloc(copy->designator_length); 1791 memcpy(lun->dd->designator, copy->designator, copy->designator_length); 1792 } 1793 } 1794 iscsi_open(BlockDriverState * bs,QDict * options,int flags,Error ** errp)1795 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags, 1796 Error **errp) 1797 { 1798 IscsiLun *iscsilun = bs->opaque; 1799 struct iscsi_context *iscsi = NULL; 1800 struct scsi_task *task = NULL; 1801 struct scsi_inquiry_standard *inq = NULL; 1802 struct scsi_inquiry_supported_pages *inq_vpd; 1803 char *initiator_name = NULL; 1804 QemuOpts *opts; 1805 Error *local_err = NULL; 1806 const char *transport_name, *portal, *target; 1807 #if LIBISCSI_API_VERSION >= (20160603) 1808 enum iscsi_transport_type transport; 1809 #endif 1810 int i, ret = 0, timeout = 0, lun; 1811 1812 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); 1813 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 1814 ret = -EINVAL; 1815 goto out; 1816 } 1817 1818 transport_name = qemu_opt_get(opts, "transport"); 1819 portal = qemu_opt_get(opts, "portal"); 1820 target = qemu_opt_get(opts, "target"); 1821 lun = qemu_opt_get_number(opts, "lun", 0); 1822 1823 if (!transport_name || !portal || !target) { 1824 error_setg(errp, "Need all of transport, portal and target options"); 1825 ret = -EINVAL; 1826 goto out; 1827 } 1828 1829 if (!strcmp(transport_name, "tcp")) { 1830 #if LIBISCSI_API_VERSION >= (20160603) 1831 transport = TCP_TRANSPORT; 1832 } else if (!strcmp(transport_name, "iser")) { 1833 transport = ISER_TRANSPORT; 1834 #else 1835 /* TCP is what older libiscsi versions always use */ 1836 #endif 1837 } else { 1838 error_setg(errp, "Unknown transport: %s", transport_name); 1839 ret = -EINVAL; 1840 goto out; 1841 } 1842 1843 memset(iscsilun, 0, sizeof(IscsiLun)); 1844 1845 initiator_name = get_initiator_name(opts); 1846 1847 iscsi = iscsi_create_context(initiator_name); 1848 if (iscsi == NULL) { 1849 error_setg(errp, "iSCSI: Failed to create iSCSI context."); 1850 ret = -ENOMEM; 1851 goto out; 1852 } 1853 #if LIBISCSI_API_VERSION >= (20160603) 1854 if (iscsi_init_transport(iscsi, transport)) { 1855 error_setg(errp, ("Error initializing transport.")); 1856 ret = -EINVAL; 1857 goto out; 1858 } 1859 #endif 1860 if (iscsi_set_targetname(iscsi, target)) { 1861 error_setg(errp, "iSCSI: Failed to set target name."); 1862 ret = -EINVAL; 1863 goto out; 1864 } 1865 1866 /* check if we got CHAP username/password via the options */ 1867 apply_chap(iscsi, opts, &local_err); 1868 if (local_err != NULL) { 1869 error_propagate(errp, local_err); 1870 ret = -EINVAL; 1871 goto out; 1872 } 1873 1874 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) { 1875 error_setg(errp, "iSCSI: Failed to set session type to normal."); 1876 ret = -EINVAL; 1877 goto out; 1878 } 1879 1880 /* check if we got HEADER_DIGEST via the options */ 1881 apply_header_digest(iscsi, opts, &local_err); 1882 if (local_err != NULL) { 1883 error_propagate(errp, local_err); 1884 ret = -EINVAL; 1885 goto out; 1886 } 1887 1888 /* timeout handling is broken in libiscsi before 1.15.0 */ 1889 timeout = qemu_opt_get_number(opts, "timeout", 0); 1890 #if LIBISCSI_API_VERSION >= 20150621 1891 iscsi_set_timeout(iscsi, timeout); 1892 #else 1893 if (timeout) { 1894 warn_report("iSCSI: ignoring timeout value for libiscsi <1.15.0"); 1895 } 1896 #endif 1897 1898 if (iscsi_full_connect_sync(iscsi, portal, lun) != 0) { 1899 error_setg(errp, "iSCSI: Failed to connect to LUN : %s", 1900 iscsi_get_error(iscsi)); 1901 ret = -EINVAL; 1902 goto out; 1903 } 1904 1905 iscsilun->iscsi = iscsi; 1906 iscsilun->aio_context = bdrv_get_aio_context(bs); 1907 iscsilun->lun = lun; 1908 iscsilun->has_write_same = true; 1909 1910 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0, 1911 (void **) &inq, errp); 1912 if (task == NULL) { 1913 ret = -EINVAL; 1914 goto out; 1915 } 1916 iscsilun->type = inq->periperal_device_type; 1917 scsi_free_scsi_task(task); 1918 task = NULL; 1919 1920 iscsi_modesense_sync(iscsilun); 1921 if (iscsilun->dpofua) { 1922 bs->supported_write_flags = BDRV_REQ_FUA; 1923 } 1924 1925 /* Check the write protect flag of the LUN if we want to write */ 1926 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) && 1927 iscsilun->write_protected) { 1928 bdrv_graph_rdlock_main_loop(); 1929 ret = bdrv_apply_auto_read_only(bs, "LUN is write protected", errp); 1930 bdrv_graph_rdunlock_main_loop(); 1931 if (ret < 0) { 1932 goto out; 1933 } 1934 flags &= ~BDRV_O_RDWR; 1935 } 1936 1937 iscsi_readcapacity_sync(iscsilun, &local_err); 1938 if (local_err != NULL) { 1939 error_propagate(errp, local_err); 1940 ret = -EINVAL; 1941 goto out; 1942 } 1943 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun); 1944 1945 /* We don't have any emulation for devices other than disks and CD-ROMs, so 1946 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu 1947 * will try to read from the device to guess the image format. 1948 */ 1949 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) { 1950 bs->sg = true; 1951 } 1952 1953 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1954 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES, 1955 (void **) &inq_vpd, errp); 1956 if (task == NULL) { 1957 ret = -EINVAL; 1958 goto out; 1959 } 1960 for (i = 0; i < inq_vpd->num_pages; i++) { 1961 struct scsi_task *inq_task; 1962 struct scsi_inquiry_logical_block_provisioning *inq_lbp; 1963 struct scsi_inquiry_block_limits *inq_bl; 1964 struct scsi_inquiry_device_identification *inq_di; 1965 switch (inq_vpd->pages[i]) { 1966 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING: 1967 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1968 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING, 1969 (void **) &inq_lbp, errp); 1970 if (inq_task == NULL) { 1971 ret = -EINVAL; 1972 goto out; 1973 } 1974 memcpy(&iscsilun->lbp, inq_lbp, 1975 sizeof(struct scsi_inquiry_logical_block_provisioning)); 1976 scsi_free_scsi_task(inq_task); 1977 break; 1978 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS: 1979 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1980 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS, 1981 (void **) &inq_bl, errp); 1982 if (inq_task == NULL) { 1983 ret = -EINVAL; 1984 goto out; 1985 } 1986 memcpy(&iscsilun->bl, inq_bl, 1987 sizeof(struct scsi_inquiry_block_limits)); 1988 scsi_free_scsi_task(inq_task); 1989 break; 1990 case SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION: 1991 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1, 1992 SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION, 1993 (void **) &inq_di, errp); 1994 if (inq_task == NULL) { 1995 ret = -EINVAL; 1996 goto out; 1997 } 1998 iscsi_save_designator(iscsilun, inq_di); 1999 scsi_free_scsi_task(inq_task); 2000 break; 2001 default: 2002 break; 2003 } 2004 } 2005 scsi_free_scsi_task(task); 2006 task = NULL; 2007 2008 qemu_mutex_init(&iscsilun->mutex); 2009 iscsi_attach_aio_context(bs, iscsilun->aio_context); 2010 2011 /* Guess the internal cluster (page) size of the iscsi target by the means 2012 * of opt_unmap_gran. Transfer the unmap granularity only if it has a 2013 * reasonable size */ 2014 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 && 2015 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) { 2016 iscsilun->cluster_size = iscsilun->bl.opt_unmap_gran * 2017 iscsilun->block_size; 2018 if (iscsilun->lbprz) { 2019 ret = iscsi_allocmap_init(iscsilun, flags); 2020 } 2021 } 2022 2023 if (iscsilun->lbprz && iscsilun->lbp.lbpws) { 2024 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP; 2025 } 2026 2027 out: 2028 qemu_opts_del(opts); 2029 g_free(initiator_name); 2030 if (task != NULL) { 2031 scsi_free_scsi_task(task); 2032 } 2033 2034 if (ret) { 2035 if (iscsi != NULL) { 2036 if (iscsi_is_logged_in(iscsi)) { 2037 iscsi_logout_sync(iscsi); 2038 } 2039 iscsi_destroy_context(iscsi); 2040 } 2041 memset(iscsilun, 0, sizeof(IscsiLun)); 2042 } 2043 2044 return ret; 2045 } 2046 iscsi_close(BlockDriverState * bs)2047 static void iscsi_close(BlockDriverState *bs) 2048 { 2049 IscsiLun *iscsilun = bs->opaque; 2050 struct iscsi_context *iscsi = iscsilun->iscsi; 2051 2052 iscsi_detach_aio_context(bs); 2053 if (iscsi_is_logged_in(iscsi)) { 2054 iscsi_logout_sync(iscsi); 2055 } 2056 iscsi_destroy_context(iscsi); 2057 if (iscsilun->dd) { 2058 g_free(iscsilun->dd->designator); 2059 g_free(iscsilun->dd); 2060 } 2061 g_free(iscsilun->zeroblock); 2062 iscsi_allocmap_free(iscsilun); 2063 qemu_mutex_destroy(&iscsilun->mutex); 2064 memset(iscsilun, 0, sizeof(IscsiLun)); 2065 } 2066 iscsi_refresh_limits(BlockDriverState * bs,Error ** errp)2067 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp) 2068 { 2069 /* We don't actually refresh here, but just return data queried in 2070 * iscsi_open(): iscsi targets don't change their limits. */ 2071 2072 IscsiLun *iscsilun = bs->opaque; 2073 uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff; 2074 unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size); 2075 2076 assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bdrv_is_sg(bs)); 2077 2078 bs->bl.request_alignment = block_size; 2079 2080 if (iscsilun->bl.max_xfer_len) { 2081 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len); 2082 } 2083 2084 if (max_xfer_len * block_size < INT_MAX) { 2085 bs->bl.max_transfer = max_xfer_len * iscsilun->block_size; 2086 } 2087 2088 if (iscsilun->lbp.lbpu) { 2089 bs->bl.max_pdiscard = 2090 MIN_NON_ZERO(iscsilun->bl.max_unmap * iscsilun->block_size, 2091 (uint64_t)UINT32_MAX * iscsilun->block_size); 2092 bs->bl.pdiscard_alignment = 2093 iscsilun->bl.opt_unmap_gran * iscsilun->block_size; 2094 } else { 2095 bs->bl.pdiscard_alignment = iscsilun->block_size; 2096 } 2097 2098 bs->bl.max_pwrite_zeroes = 2099 MIN_NON_ZERO(iscsilun->bl.max_ws_len * iscsilun->block_size, 2100 max_xfer_len * iscsilun->block_size); 2101 2102 if (iscsilun->lbp.lbpws) { 2103 bs->bl.pwrite_zeroes_alignment = 2104 iscsilun->bl.opt_unmap_gran * iscsilun->block_size; 2105 } else { 2106 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size; 2107 } 2108 if (iscsilun->bl.opt_xfer_len && 2109 iscsilun->bl.opt_xfer_len < INT_MAX / block_size) { 2110 bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len * 2111 iscsilun->block_size); 2112 } 2113 } 2114 2115 /* Note that this will not re-establish a connection with an iSCSI target - it 2116 * is effectively a NOP. */ iscsi_reopen_prepare(BDRVReopenState * state,BlockReopenQueue * queue,Error ** errp)2117 static int iscsi_reopen_prepare(BDRVReopenState *state, 2118 BlockReopenQueue *queue, Error **errp) 2119 { 2120 IscsiLun *iscsilun = state->bs->opaque; 2121 2122 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) { 2123 error_setg(errp, "Cannot open a write protected LUN as read-write"); 2124 return -EACCES; 2125 } 2126 return 0; 2127 } 2128 iscsi_reopen_commit(BDRVReopenState * reopen_state)2129 static void iscsi_reopen_commit(BDRVReopenState *reopen_state) 2130 { 2131 IscsiLun *iscsilun = reopen_state->bs->opaque; 2132 2133 /* the cache.direct status might have changed */ 2134 if (iscsilun->allocmap != NULL) { 2135 iscsi_allocmap_init(iscsilun, reopen_state->flags); 2136 } 2137 } 2138 iscsi_co_truncate(BlockDriverState * bs,int64_t offset,bool exact,PreallocMode prealloc,BdrvRequestFlags flags,Error ** errp)2139 static int coroutine_fn iscsi_co_truncate(BlockDriverState *bs, int64_t offset, 2140 bool exact, PreallocMode prealloc, 2141 BdrvRequestFlags flags, Error **errp) 2142 { 2143 IscsiLun *iscsilun = bs->opaque; 2144 int64_t cur_length; 2145 Error *local_err = NULL; 2146 2147 if (prealloc != PREALLOC_MODE_OFF) { 2148 error_setg(errp, "Unsupported preallocation mode '%s'", 2149 PreallocMode_str(prealloc)); 2150 return -ENOTSUP; 2151 } 2152 2153 if (iscsilun->type != TYPE_DISK) { 2154 error_setg(errp, "Cannot resize non-disk iSCSI devices"); 2155 return -ENOTSUP; 2156 } 2157 2158 iscsi_readcapacity_sync(iscsilun, &local_err); 2159 if (local_err != NULL) { 2160 error_propagate(errp, local_err); 2161 return -EIO; 2162 } 2163 2164 cur_length = iscsi_co_getlength(bs); 2165 if (offset != cur_length && exact) { 2166 error_setg(errp, "Cannot resize iSCSI devices"); 2167 return -ENOTSUP; 2168 } else if (offset > cur_length) { 2169 error_setg(errp, "Cannot grow iSCSI devices"); 2170 return -EINVAL; 2171 } 2172 2173 if (iscsilun->allocmap != NULL) { 2174 iscsi_allocmap_init(iscsilun, bs->open_flags); 2175 } 2176 2177 return 0; 2178 } 2179 2180 static int coroutine_fn iscsi_co_get_info(BlockDriverState * bs,BlockDriverInfo * bdi)2181 iscsi_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2182 { 2183 IscsiLun *iscsilun = bs->opaque; 2184 bdi->cluster_size = iscsilun->cluster_size; 2185 return 0; 2186 } 2187 iscsi_co_invalidate_cache(BlockDriverState * bs,Error ** errp)2188 static void coroutine_fn iscsi_co_invalidate_cache(BlockDriverState *bs, 2189 Error **errp) 2190 { 2191 IscsiLun *iscsilun = bs->opaque; 2192 iscsi_allocmap_invalidate(iscsilun); 2193 } 2194 2195 static int coroutine_fn GRAPH_RDLOCK iscsi_co_copy_range_from(BlockDriverState * bs,BdrvChild * src,int64_t src_offset,BdrvChild * dst,int64_t dst_offset,int64_t bytes,BdrvRequestFlags read_flags,BdrvRequestFlags write_flags)2196 iscsi_co_copy_range_from(BlockDriverState *bs, 2197 BdrvChild *src, int64_t src_offset, 2198 BdrvChild *dst, int64_t dst_offset, 2199 int64_t bytes, BdrvRequestFlags read_flags, 2200 BdrvRequestFlags write_flags) 2201 { 2202 return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, 2203 read_flags, write_flags); 2204 } 2205 iscsi_xcopy_task(int param_len)2206 static struct scsi_task *iscsi_xcopy_task(int param_len) 2207 { 2208 struct scsi_task *task; 2209 2210 task = g_new0(struct scsi_task, 1); 2211 2212 task->cdb[0] = EXTENDED_COPY; 2213 task->cdb[10] = (param_len >> 24) & 0xFF; 2214 task->cdb[11] = (param_len >> 16) & 0xFF; 2215 task->cdb[12] = (param_len >> 8) & 0xFF; 2216 task->cdb[13] = param_len & 0xFF; 2217 task->cdb_size = 16; 2218 task->xfer_dir = SCSI_XFER_WRITE; 2219 task->expxferlen = param_len; 2220 2221 return task; 2222 } 2223 iscsi_populate_target_desc(unsigned char * desc,IscsiLun * lun)2224 static void iscsi_populate_target_desc(unsigned char *desc, IscsiLun *lun) 2225 { 2226 struct scsi_inquiry_device_designator *dd = lun->dd; 2227 2228 memset(desc, 0, 32); 2229 desc[0] = 0xE4; /* IDENT_DESCR_TGT_DESCR */ 2230 desc[4] = dd->code_set; 2231 desc[5] = (dd->designator_type & 0xF) 2232 | ((dd->association & 3) << 4); 2233 desc[7] = dd->designator_length; 2234 memcpy(desc + 8, dd->designator, MIN(dd->designator_length, 20)); 2235 2236 desc[28] = 0; 2237 desc[29] = (lun->block_size >> 16) & 0xFF; 2238 desc[30] = (lun->block_size >> 8) & 0xFF; 2239 desc[31] = lun->block_size & 0xFF; 2240 } 2241 iscsi_xcopy_desc_hdr(uint8_t * hdr,int dc,int cat,int src_index,int dst_index)2242 static void iscsi_xcopy_desc_hdr(uint8_t *hdr, int dc, int cat, int src_index, 2243 int dst_index) 2244 { 2245 hdr[0] = 0x02; /* BLK_TO_BLK_SEG_DESCR */ 2246 hdr[1] = ((dc << 1) | cat) & 0xFF; 2247 hdr[2] = (XCOPY_BLK2BLK_SEG_DESC_SIZE >> 8) & 0xFF; 2248 /* don't account for the first 4 bytes in descriptor header*/ 2249 hdr[3] = (XCOPY_BLK2BLK_SEG_DESC_SIZE - 4 /* SEG_DESC_SRC_INDEX_OFFSET */) & 0xFF; 2250 hdr[4] = (src_index >> 8) & 0xFF; 2251 hdr[5] = src_index & 0xFF; 2252 hdr[6] = (dst_index >> 8) & 0xFF; 2253 hdr[7] = dst_index & 0xFF; 2254 } 2255 iscsi_xcopy_populate_desc(uint8_t * desc,int dc,int cat,int src_index,int dst_index,int num_blks,uint64_t src_lba,uint64_t dst_lba)2256 static void iscsi_xcopy_populate_desc(uint8_t *desc, int dc, int cat, 2257 int src_index, int dst_index, int num_blks, 2258 uint64_t src_lba, uint64_t dst_lba) 2259 { 2260 iscsi_xcopy_desc_hdr(desc, dc, cat, src_index, dst_index); 2261 2262 /* The caller should verify the request size */ 2263 assert(num_blks < 65536); 2264 desc[10] = (num_blks >> 8) & 0xFF; 2265 desc[11] = num_blks & 0xFF; 2266 desc[12] = (src_lba >> 56) & 0xFF; 2267 desc[13] = (src_lba >> 48) & 0xFF; 2268 desc[14] = (src_lba >> 40) & 0xFF; 2269 desc[15] = (src_lba >> 32) & 0xFF; 2270 desc[16] = (src_lba >> 24) & 0xFF; 2271 desc[17] = (src_lba >> 16) & 0xFF; 2272 desc[18] = (src_lba >> 8) & 0xFF; 2273 desc[19] = src_lba & 0xFF; 2274 desc[20] = (dst_lba >> 56) & 0xFF; 2275 desc[21] = (dst_lba >> 48) & 0xFF; 2276 desc[22] = (dst_lba >> 40) & 0xFF; 2277 desc[23] = (dst_lba >> 32) & 0xFF; 2278 desc[24] = (dst_lba >> 24) & 0xFF; 2279 desc[25] = (dst_lba >> 16) & 0xFF; 2280 desc[26] = (dst_lba >> 8) & 0xFF; 2281 desc[27] = dst_lba & 0xFF; 2282 } 2283 iscsi_xcopy_populate_header(unsigned char * buf,int list_id,int str,int list_id_usage,int prio,int tgt_desc_len,int seg_desc_len,int inline_data_len)2284 static void iscsi_xcopy_populate_header(unsigned char *buf, int list_id, int str, 2285 int list_id_usage, int prio, 2286 int tgt_desc_len, 2287 int seg_desc_len, int inline_data_len) 2288 { 2289 buf[0] = list_id; 2290 buf[1] = ((str & 1) << 5) | ((list_id_usage & 3) << 3) | (prio & 7); 2291 buf[2] = (tgt_desc_len >> 8) & 0xFF; 2292 buf[3] = tgt_desc_len & 0xFF; 2293 buf[8] = (seg_desc_len >> 24) & 0xFF; 2294 buf[9] = (seg_desc_len >> 16) & 0xFF; 2295 buf[10] = (seg_desc_len >> 8) & 0xFF; 2296 buf[11] = seg_desc_len & 0xFF; 2297 buf[12] = (inline_data_len >> 24) & 0xFF; 2298 buf[13] = (inline_data_len >> 16) & 0xFF; 2299 buf[14] = (inline_data_len >> 8) & 0xFF; 2300 buf[15] = inline_data_len & 0xFF; 2301 } 2302 iscsi_xcopy_data(struct iscsi_data * data,IscsiLun * src,int64_t src_lba,IscsiLun * dst,int64_t dst_lba,uint16_t num_blocks)2303 static void iscsi_xcopy_data(struct iscsi_data *data, 2304 IscsiLun *src, int64_t src_lba, 2305 IscsiLun *dst, int64_t dst_lba, 2306 uint16_t num_blocks) 2307 { 2308 uint8_t *buf; 2309 const int src_offset = XCOPY_DESC_OFFSET; 2310 const int dst_offset = XCOPY_DESC_OFFSET + IDENT_DESCR_TGT_DESCR_SIZE; 2311 const int seg_offset = dst_offset + IDENT_DESCR_TGT_DESCR_SIZE; 2312 2313 data->size = XCOPY_DESC_OFFSET + 2314 IDENT_DESCR_TGT_DESCR_SIZE * 2 + 2315 XCOPY_BLK2BLK_SEG_DESC_SIZE; 2316 data->data = g_malloc0(data->size); 2317 buf = data->data; 2318 2319 /* Initialise the parameter list header */ 2320 iscsi_xcopy_populate_header(buf, 1, 0, 2 /* LIST_ID_USAGE_DISCARD */, 2321 0, 2 * IDENT_DESCR_TGT_DESCR_SIZE, 2322 XCOPY_BLK2BLK_SEG_DESC_SIZE, 2323 0); 2324 2325 /* Initialise CSCD list with one src + one dst descriptor */ 2326 iscsi_populate_target_desc(&buf[src_offset], src); 2327 iscsi_populate_target_desc(&buf[dst_offset], dst); 2328 2329 /* Initialise one segment descriptor */ 2330 iscsi_xcopy_populate_desc(&buf[seg_offset], 0, 0, 0, 1, num_blocks, 2331 src_lba, dst_lba); 2332 } 2333 2334 static int coroutine_fn GRAPH_RDLOCK iscsi_co_copy_range_to(BlockDriverState * bs,BdrvChild * src,int64_t src_offset,BdrvChild * dst,int64_t dst_offset,int64_t bytes,BdrvRequestFlags read_flags,BdrvRequestFlags write_flags)2335 iscsi_co_copy_range_to(BlockDriverState *bs, 2336 BdrvChild *src, int64_t src_offset, 2337 BdrvChild *dst, int64_t dst_offset, 2338 int64_t bytes, BdrvRequestFlags read_flags, 2339 BdrvRequestFlags write_flags) 2340 { 2341 IscsiLun *dst_lun = dst->bs->opaque; 2342 IscsiLun *src_lun; 2343 struct IscsiTask iscsi_task; 2344 struct iscsi_data data; 2345 int r = 0; 2346 int block_size; 2347 2348 if (src->bs->drv->bdrv_co_copy_range_to != iscsi_co_copy_range_to) { 2349 return -ENOTSUP; 2350 } 2351 src_lun = src->bs->opaque; 2352 2353 if (!src_lun->dd || !dst_lun->dd) { 2354 return -ENOTSUP; 2355 } 2356 if (!is_byte_request_lun_aligned(dst_offset, bytes, dst_lun)) { 2357 return -ENOTSUP; 2358 } 2359 if (!is_byte_request_lun_aligned(src_offset, bytes, src_lun)) { 2360 return -ENOTSUP; 2361 } 2362 if (dst_lun->block_size != src_lun->block_size || 2363 !dst_lun->block_size) { 2364 return -ENOTSUP; 2365 } 2366 2367 block_size = dst_lun->block_size; 2368 if (bytes / block_size > 65535) { 2369 return -ENOTSUP; 2370 } 2371 2372 iscsi_xcopy_data(&data, 2373 src_lun, src_offset / block_size, 2374 dst_lun, dst_offset / block_size, 2375 bytes / block_size); 2376 2377 iscsi_co_init_iscsitask(dst_lun, &iscsi_task); 2378 2379 qemu_mutex_lock(&dst_lun->mutex); 2380 iscsi_task.task = iscsi_xcopy_task(data.size); 2381 retry: 2382 if (iscsi_scsi_command_async(dst_lun->iscsi, dst_lun->lun, 2383 iscsi_task.task, iscsi_co_generic_cb, 2384 &data, 2385 &iscsi_task) != 0) { 2386 r = -EIO; 2387 goto out_unlock; 2388 } 2389 2390 iscsi_co_wait_for_task(&iscsi_task, dst_lun); 2391 2392 if (iscsi_task.do_retry) { 2393 iscsi_task.complete = 0; 2394 goto retry; 2395 } 2396 2397 if (iscsi_task.status != SCSI_STATUS_GOOD) { 2398 r = iscsi_task.err_code; 2399 goto out_unlock; 2400 } 2401 2402 out_unlock: 2403 2404 trace_iscsi_xcopy(src_lun, src_offset, dst_lun, dst_offset, bytes, r); 2405 g_free(iscsi_task.task); 2406 qemu_mutex_unlock(&dst_lun->mutex); 2407 g_free(iscsi_task.err_str); 2408 return r; 2409 } 2410 2411 2412 static const char *const iscsi_strong_runtime_opts[] = { 2413 "transport", 2414 "portal", 2415 "target", 2416 "user", 2417 "password", 2418 "password-secret", 2419 "lun", 2420 "initiator-name", 2421 "header-digest", 2422 2423 NULL 2424 }; 2425 2426 static BlockDriver bdrv_iscsi = { 2427 .format_name = "iscsi", 2428 .protocol_name = "iscsi", 2429 2430 .instance_size = sizeof(IscsiLun), 2431 .bdrv_parse_filename = iscsi_parse_filename, 2432 .bdrv_open = iscsi_open, 2433 .bdrv_close = iscsi_close, 2434 .bdrv_co_create_opts = bdrv_co_create_opts_simple, 2435 .create_opts = &bdrv_create_opts_simple, 2436 .bdrv_reopen_prepare = iscsi_reopen_prepare, 2437 .bdrv_reopen_commit = iscsi_reopen_commit, 2438 .bdrv_co_invalidate_cache = iscsi_co_invalidate_cache, 2439 2440 .bdrv_co_getlength = iscsi_co_getlength, 2441 .bdrv_co_get_info = iscsi_co_get_info, 2442 .bdrv_co_truncate = iscsi_co_truncate, 2443 .bdrv_refresh_limits = iscsi_refresh_limits, 2444 2445 .bdrv_co_block_status = iscsi_co_block_status, 2446 .bdrv_co_pdiscard = iscsi_co_pdiscard, 2447 .bdrv_co_copy_range_from = iscsi_co_copy_range_from, 2448 .bdrv_co_copy_range_to = iscsi_co_copy_range_to, 2449 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes, 2450 .bdrv_co_readv = iscsi_co_readv, 2451 .bdrv_co_writev = iscsi_co_writev, 2452 .bdrv_co_flush_to_disk = iscsi_co_flush, 2453 2454 #ifdef __linux__ 2455 .bdrv_aio_ioctl = iscsi_aio_ioctl, 2456 #endif 2457 2458 .bdrv_detach_aio_context = iscsi_detach_aio_context, 2459 .bdrv_attach_aio_context = iscsi_attach_aio_context, 2460 2461 .strong_runtime_opts = iscsi_strong_runtime_opts, 2462 }; 2463 2464 #if LIBISCSI_API_VERSION >= (20160603) 2465 static BlockDriver bdrv_iser = { 2466 .format_name = "iser", 2467 .protocol_name = "iser", 2468 2469 .instance_size = sizeof(IscsiLun), 2470 .bdrv_parse_filename = iscsi_parse_filename, 2471 .bdrv_open = iscsi_open, 2472 .bdrv_close = iscsi_close, 2473 .bdrv_co_create_opts = bdrv_co_create_opts_simple, 2474 .create_opts = &bdrv_create_opts_simple, 2475 .bdrv_reopen_prepare = iscsi_reopen_prepare, 2476 .bdrv_reopen_commit = iscsi_reopen_commit, 2477 .bdrv_co_invalidate_cache = iscsi_co_invalidate_cache, 2478 2479 .bdrv_co_getlength = iscsi_co_getlength, 2480 .bdrv_co_get_info = iscsi_co_get_info, 2481 .bdrv_co_truncate = iscsi_co_truncate, 2482 .bdrv_refresh_limits = iscsi_refresh_limits, 2483 2484 .bdrv_co_block_status = iscsi_co_block_status, 2485 .bdrv_co_pdiscard = iscsi_co_pdiscard, 2486 .bdrv_co_copy_range_from = iscsi_co_copy_range_from, 2487 .bdrv_co_copy_range_to = iscsi_co_copy_range_to, 2488 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes, 2489 .bdrv_co_readv = iscsi_co_readv, 2490 .bdrv_co_writev = iscsi_co_writev, 2491 .bdrv_co_flush_to_disk = iscsi_co_flush, 2492 2493 #ifdef __linux__ 2494 .bdrv_aio_ioctl = iscsi_aio_ioctl, 2495 #endif 2496 2497 .bdrv_detach_aio_context = iscsi_detach_aio_context, 2498 .bdrv_attach_aio_context = iscsi_attach_aio_context, 2499 2500 .strong_runtime_opts = iscsi_strong_runtime_opts, 2501 }; 2502 #endif 2503 iscsi_block_init(void)2504 static void iscsi_block_init(void) 2505 { 2506 bdrv_register(&bdrv_iscsi); 2507 #if LIBISCSI_API_VERSION >= (20160603) 2508 bdrv_register(&bdrv_iser); 2509 #endif 2510 } 2511 2512 block_init(iscsi_block_init); 2513