2 * QEMU Block driver for iSCSI images
4 * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
5 * Copyright (c) 2012-2016 Peter Lieven <pl@kamp.de>
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:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
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
26 #include "qemu/osdep.h"
30 #include <arpa/inet.h>
31 #include "qemu-common.h"
32 #include "qemu/config-file.h"
33 #include "qemu/error-report.h"
34 #include "qemu/bitops.h"
35 #include "qemu/bitmap.h"
36 #include "block/block_int.h"
37 #include "block/scsi.h"
39 #include "qemu/uuid.h"
40 #include "qmp-commands.h"
41 #include "qapi/qmp/qstring.h"
42 #include "crypto/secret.h"
44 #include <iscsi/iscsi.h>
45 #include <iscsi/scsi-lowlevel.h>
51 typedef struct IscsiLun
{
52 struct iscsi_context
*iscsi
;
53 AioContext
*aio_context
;
55 enum scsi_inquiry_peripheral_device_type type
;
60 QEMUTimer
*event_timer
;
61 struct scsi_inquiry_logical_block_provisioning lbp
;
62 struct scsi_inquiry_block_limits bl
;
63 unsigned char *zeroblock
;
64 /* The allocmap tracks which clusters (pages) on the iSCSI target are
65 * allocated and which are not. In case a target returns zeros for
66 * unallocated pages (iscsilun->lprz) we can directly return zeros instead
67 * of reading zeros over the wire if a read request falls within an
68 * unallocated block. As there are 3 possible states we need 2 bitmaps to
69 * track. allocmap_valid keeps track if QEMU's information about a page is
70 * valid. allocmap tracks if a page is allocated or not. In case QEMU has no
71 * valid information about a page the corresponding allocmap entry should be
72 * switched to unallocated as well to force a new lookup of the allocation
73 * status as lookups are generally skipped if a page is suspect to be
74 * allocated. If a iSCSI target is opened with cache.direct = on the
75 * allocmap_valid does not exist turning all cached information invalid so
76 * that a fresh lookup is made for any page even if allocmap entry returns
77 * it's unallocated. */
78 unsigned long *allocmap
;
79 unsigned long *allocmap_valid
;
88 bool request_timed_out
;
91 typedef struct IscsiTask
{
96 struct scsi_task
*task
;
99 QEMUTimer retry_timer
;
103 typedef struct IscsiAIOCB
{
108 struct scsi_task
*task
;
119 /* libiscsi uses time_t so its enough to process events every second */
120 #define EVENT_INTERVAL 1000
121 #define NOP_INTERVAL 5000
122 #define MAX_NOP_FAILURES 3
123 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times)
124 static const unsigned iscsi_retry_times
[] = {8, 32, 128, 512, 2048, 8192, 32768};
126 /* this threshold is a trade-off knob to choose between
127 * the potential additional overhead of an extra GET_LBA_STATUS request
128 * vs. unnecessarily reading a lot of zero sectors over the wire.
129 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
130 * sectors we check the allocation status of the area covered by the
131 * request first if the allocationmap indicates that the area might be
133 #define ISCSI_CHECKALLOC_THRES 64
140 qemu_bh_delete(acb
->bh
);
145 acb
->common
.cb(acb
->common
.opaque
, acb
->status
);
147 if (acb
->task
!= NULL
) {
148 scsi_free_scsi_task(acb
->task
);
156 iscsi_schedule_bh(IscsiAIOCB
*acb
)
161 acb
->bh
= aio_bh_new(acb
->iscsilun
->aio_context
, iscsi_bh_cb
, acb
);
162 qemu_bh_schedule(acb
->bh
);
165 static void iscsi_co_generic_bh_cb(void *opaque
)
167 struct IscsiTask
*iTask
= opaque
;
170 aio_co_wake(iTask
->co
);
173 static void iscsi_retry_timer_expired(void *opaque
)
175 struct IscsiTask
*iTask
= opaque
;
178 aio_co_wake(iTask
->co
);
182 static inline unsigned exp_random(double mean
)
184 return -mean
* log((double)rand() / RAND_MAX
);
187 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in
188 * libiscsi 1.10.0, together with other constants we need. Use it as
189 * a hint that we have to define them ourselves if needed, to keep the
190 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for
191 * the test because SCSI_STATUS_* is an enum.
193 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes
194 * an enum, check against the LIBISCSI_API_VERSION macro, which was
195 * introduced in 1.11.0. If it is present, there is no need to define
198 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \
199 !defined(LIBISCSI_API_VERSION)
200 #define SCSI_STATUS_TASK_SET_FULL 0x28
201 #define SCSI_STATUS_TIMEOUT 0x0f000002
202 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600
203 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00
206 #ifndef LIBISCSI_API_VERSION
207 #define LIBISCSI_API_VERSION 20130701
210 static int iscsi_translate_sense(struct scsi_sense
*sense
)
214 switch (sense
->key
) {
215 case SCSI_SENSE_NOT_READY
:
217 case SCSI_SENSE_DATA_PROTECTION
:
219 case SCSI_SENSE_COMMAND_ABORTED
:
221 case SCSI_SENSE_ILLEGAL_REQUEST
:
227 switch (sense
->ascq
) {
228 case SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR
:
229 case SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE
:
230 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB
:
231 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST
:
234 case SCSI_SENSE_ASCQ_LBA_OUT_OF_RANGE
:
237 case SCSI_SENSE_ASCQ_LOGICAL_UNIT_NOT_SUPPORTED
:
240 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT
:
241 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_CLOSED
:
242 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_OPEN
:
245 case SCSI_SENSE_ASCQ_WRITE_PROTECTED
:
256 iscsi_co_generic_cb(struct iscsi_context
*iscsi
, int status
,
257 void *command_data
, void *opaque
)
259 struct IscsiTask
*iTask
= opaque
;
260 struct scsi_task
*task
= command_data
;
262 iTask
->status
= status
;
266 if (status
!= SCSI_STATUS_GOOD
) {
267 if (iTask
->retries
++ < ISCSI_CMD_RETRIES
) {
268 if (status
== SCSI_STATUS_CHECK_CONDITION
269 && task
->sense
.key
== SCSI_SENSE_UNIT_ATTENTION
) {
270 error_report("iSCSI CheckCondition: %s",
271 iscsi_get_error(iscsi
));
275 if (status
== SCSI_STATUS_BUSY
||
276 status
== SCSI_STATUS_TIMEOUT
||
277 status
== SCSI_STATUS_TASK_SET_FULL
) {
278 unsigned retry_time
=
279 exp_random(iscsi_retry_times
[iTask
->retries
- 1]);
280 if (status
== SCSI_STATUS_TIMEOUT
) {
281 /* make sure the request is rescheduled AFTER the
282 * reconnect is initiated */
283 retry_time
= EVENT_INTERVAL
* 2;
284 iTask
->iscsilun
->request_timed_out
= true;
286 error_report("iSCSI Busy/TaskSetFull/TimeOut"
287 " (retry #%u in %u ms): %s",
288 iTask
->retries
, retry_time
,
289 iscsi_get_error(iscsi
));
290 aio_timer_init(iTask
->iscsilun
->aio_context
,
291 &iTask
->retry_timer
, QEMU_CLOCK_REALTIME
,
292 SCALE_MS
, iscsi_retry_timer_expired
, iTask
);
293 timer_mod(&iTask
->retry_timer
,
294 qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) + retry_time
);
299 iTask
->err_code
= iscsi_translate_sense(&task
->sense
);
300 error_report("iSCSI Failure: %s", iscsi_get_error(iscsi
));
305 aio_bh_schedule_oneshot(iTask
->iscsilun
->aio_context
,
306 iscsi_co_generic_bh_cb
, iTask
);
312 static void iscsi_co_init_iscsitask(IscsiLun
*iscsilun
, struct IscsiTask
*iTask
)
314 *iTask
= (struct IscsiTask
) {
315 .co
= qemu_coroutine_self(),
316 .iscsilun
= iscsilun
,
321 iscsi_abort_task_cb(struct iscsi_context
*iscsi
, int status
, void *command_data
,
324 IscsiAIOCB
*acb
= private_data
;
326 acb
->status
= -ECANCELED
;
327 iscsi_schedule_bh(acb
);
331 iscsi_aio_cancel(BlockAIOCB
*blockacb
)
333 IscsiAIOCB
*acb
= (IscsiAIOCB
*)blockacb
;
334 IscsiLun
*iscsilun
= acb
->iscsilun
;
336 if (acb
->status
!= -EINPROGRESS
) {
340 /* send a task mgmt call to the target to cancel the task on the target */
341 iscsi_task_mgmt_abort_task_async(iscsilun
->iscsi
, acb
->task
,
342 iscsi_abort_task_cb
, acb
);
346 static const AIOCBInfo iscsi_aiocb_info
= {
347 .aiocb_size
= sizeof(IscsiAIOCB
),
348 .cancel_async
= iscsi_aio_cancel
,
352 static void iscsi_process_read(void *arg
);
353 static void iscsi_process_write(void *arg
);
356 iscsi_set_events(IscsiLun
*iscsilun
)
358 struct iscsi_context
*iscsi
= iscsilun
->iscsi
;
359 int ev
= iscsi_which_events(iscsi
);
361 if (ev
!= iscsilun
->events
) {
362 aio_set_fd_handler(iscsilun
->aio_context
, iscsi_get_fd(iscsi
),
364 (ev
& POLLIN
) ? iscsi_process_read
: NULL
,
365 (ev
& POLLOUT
) ? iscsi_process_write
: NULL
,
368 iscsilun
->events
= ev
;
372 static void iscsi_timed_check_events(void *opaque
)
374 IscsiLun
*iscsilun
= opaque
;
376 /* check for timed out requests */
377 iscsi_service(iscsilun
->iscsi
, 0);
379 if (iscsilun
->request_timed_out
) {
380 iscsilun
->request_timed_out
= false;
381 iscsi_reconnect(iscsilun
->iscsi
);
384 /* newer versions of libiscsi may return zero events. Ensure we are able
385 * to return to service once this situation changes. */
386 iscsi_set_events(iscsilun
);
388 timer_mod(iscsilun
->event_timer
,
389 qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) + EVENT_INTERVAL
);
393 iscsi_process_read(void *arg
)
395 IscsiLun
*iscsilun
= arg
;
396 struct iscsi_context
*iscsi
= iscsilun
->iscsi
;
398 aio_context_acquire(iscsilun
->aio_context
);
399 iscsi_service(iscsi
, POLLIN
);
400 iscsi_set_events(iscsilun
);
401 aio_context_release(iscsilun
->aio_context
);
405 iscsi_process_write(void *arg
)
407 IscsiLun
*iscsilun
= arg
;
408 struct iscsi_context
*iscsi
= iscsilun
->iscsi
;
410 aio_context_acquire(iscsilun
->aio_context
);
411 iscsi_service(iscsi
, POLLOUT
);
412 iscsi_set_events(iscsilun
);
413 aio_context_release(iscsilun
->aio_context
);
416 static int64_t sector_lun2qemu(int64_t sector
, IscsiLun
*iscsilun
)
418 return sector
* iscsilun
->block_size
/ BDRV_SECTOR_SIZE
;
421 static int64_t sector_qemu2lun(int64_t sector
, IscsiLun
*iscsilun
)
423 return sector
* BDRV_SECTOR_SIZE
/ iscsilun
->block_size
;
426 static bool is_byte_request_lun_aligned(int64_t offset
, int count
,
429 if (offset
% iscsilun
->block_size
|| count
% iscsilun
->block_size
) {
430 error_report("iSCSI misaligned request: "
431 "iscsilun->block_size %u, offset %" PRIi64
433 iscsilun
->block_size
, offset
, count
);
439 static bool is_sector_request_lun_aligned(int64_t sector_num
, int nb_sectors
,
442 assert(nb_sectors
<= BDRV_REQUEST_MAX_SECTORS
);
443 return is_byte_request_lun_aligned(sector_num
<< BDRV_SECTOR_BITS
,
444 nb_sectors
<< BDRV_SECTOR_BITS
,
448 static void iscsi_allocmap_free(IscsiLun
*iscsilun
)
450 g_free(iscsilun
->allocmap
);
451 g_free(iscsilun
->allocmap_valid
);
452 iscsilun
->allocmap
= NULL
;
453 iscsilun
->allocmap_valid
= NULL
;
457 static int iscsi_allocmap_init(IscsiLun
*iscsilun
, int open_flags
)
459 iscsi_allocmap_free(iscsilun
);
461 iscsilun
->allocmap_size
=
462 DIV_ROUND_UP(sector_lun2qemu(iscsilun
->num_blocks
, iscsilun
),
463 iscsilun
->cluster_sectors
);
465 iscsilun
->allocmap
= bitmap_try_new(iscsilun
->allocmap_size
);
466 if (!iscsilun
->allocmap
) {
470 if (open_flags
& BDRV_O_NOCACHE
) {
471 /* in case that cache.direct = on all allocmap entries are
472 * treated as invalid to force a relookup of the block
473 * status on every read request */
477 iscsilun
->allocmap_valid
= bitmap_try_new(iscsilun
->allocmap_size
);
478 if (!iscsilun
->allocmap_valid
) {
479 /* if we are under memory pressure free the allocmap as well */
480 iscsi_allocmap_free(iscsilun
);
488 iscsi_allocmap_update(IscsiLun
*iscsilun
, int64_t sector_num
,
489 int nb_sectors
, bool allocated
, bool valid
)
491 int64_t cl_num_expanded
, nb_cls_expanded
, cl_num_shrunk
, nb_cls_shrunk
;
493 if (iscsilun
->allocmap
== NULL
) {
496 /* expand to entirely contain all affected clusters */
497 cl_num_expanded
= sector_num
/ iscsilun
->cluster_sectors
;
498 nb_cls_expanded
= DIV_ROUND_UP(sector_num
+ nb_sectors
,
499 iscsilun
->cluster_sectors
) - cl_num_expanded
;
500 /* shrink to touch only completely contained clusters */
501 cl_num_shrunk
= DIV_ROUND_UP(sector_num
, iscsilun
->cluster_sectors
);
502 nb_cls_shrunk
= (sector_num
+ nb_sectors
) / iscsilun
->cluster_sectors
505 bitmap_set(iscsilun
->allocmap
, cl_num_expanded
, nb_cls_expanded
);
507 if (nb_cls_shrunk
> 0) {
508 bitmap_clear(iscsilun
->allocmap
, cl_num_shrunk
, nb_cls_shrunk
);
512 if (iscsilun
->allocmap_valid
== NULL
) {
516 if (nb_cls_shrunk
> 0) {
517 bitmap_set(iscsilun
->allocmap_valid
, cl_num_shrunk
, nb_cls_shrunk
);
520 bitmap_clear(iscsilun
->allocmap_valid
, cl_num_expanded
,
526 iscsi_allocmap_set_allocated(IscsiLun
*iscsilun
, int64_t sector_num
,
529 iscsi_allocmap_update(iscsilun
, sector_num
, nb_sectors
, true, true);
533 iscsi_allocmap_set_unallocated(IscsiLun
*iscsilun
, int64_t sector_num
,
536 /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update
537 * is ignored, so this will in effect be an iscsi_allocmap_set_invalid.
539 iscsi_allocmap_update(iscsilun
, sector_num
, nb_sectors
, false, true);
542 static void iscsi_allocmap_set_invalid(IscsiLun
*iscsilun
, int64_t sector_num
,
545 iscsi_allocmap_update(iscsilun
, sector_num
, nb_sectors
, false, false);
548 static void iscsi_allocmap_invalidate(IscsiLun
*iscsilun
)
550 if (iscsilun
->allocmap
) {
551 bitmap_zero(iscsilun
->allocmap
, iscsilun
->allocmap_size
);
553 if (iscsilun
->allocmap_valid
) {
554 bitmap_zero(iscsilun
->allocmap_valid
, iscsilun
->allocmap_size
);
559 iscsi_allocmap_is_allocated(IscsiLun
*iscsilun
, int64_t sector_num
,
563 if (iscsilun
->allocmap
== NULL
) {
566 size
= DIV_ROUND_UP(sector_num
+ nb_sectors
, iscsilun
->cluster_sectors
);
567 return !(find_next_bit(iscsilun
->allocmap
, size
,
568 sector_num
/ iscsilun
->cluster_sectors
) == size
);
571 static inline bool iscsi_allocmap_is_valid(IscsiLun
*iscsilun
,
572 int64_t sector_num
, int nb_sectors
)
575 if (iscsilun
->allocmap_valid
== NULL
) {
578 size
= DIV_ROUND_UP(sector_num
+ nb_sectors
, iscsilun
->cluster_sectors
);
579 return (find_next_zero_bit(iscsilun
->allocmap_valid
, size
,
580 sector_num
/ iscsilun
->cluster_sectors
) == size
);
583 static int coroutine_fn
584 iscsi_co_writev_flags(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
585 QEMUIOVector
*iov
, int flags
)
587 IscsiLun
*iscsilun
= bs
->opaque
;
588 struct IscsiTask iTask
;
590 uint32_t num_sectors
;
591 bool fua
= flags
& BDRV_REQ_FUA
;
594 assert(iscsilun
->dpofua
);
596 if (!is_sector_request_lun_aligned(sector_num
, nb_sectors
, iscsilun
)) {
600 if (bs
->bl
.max_transfer
) {
601 assert(nb_sectors
<< BDRV_SECTOR_BITS
<= bs
->bl
.max_transfer
);
604 lba
= sector_qemu2lun(sector_num
, iscsilun
);
605 num_sectors
= sector_qemu2lun(nb_sectors
, iscsilun
);
606 iscsi_co_init_iscsitask(iscsilun
, &iTask
);
608 if (iscsilun
->use_16_for_rw
) {
609 #if LIBISCSI_API_VERSION >= (20160603)
610 iTask
.task
= iscsi_write16_iov_task(iscsilun
->iscsi
, iscsilun
->lun
, lba
,
611 NULL
, num_sectors
* iscsilun
->block_size
,
612 iscsilun
->block_size
, 0, 0, fua
, 0, 0,
613 iscsi_co_generic_cb
, &iTask
,
614 (struct scsi_iovec
*)iov
->iov
, iov
->niov
);
616 iTask
.task
= iscsi_write10_iov_task(iscsilun
->iscsi
, iscsilun
->lun
, lba
,
617 NULL
, num_sectors
* iscsilun
->block_size
,
618 iscsilun
->block_size
, 0, 0, fua
, 0, 0,
619 iscsi_co_generic_cb
, &iTask
,
620 (struct scsi_iovec
*)iov
->iov
, iov
->niov
);
623 iTask
.task
= iscsi_write16_task(iscsilun
->iscsi
, iscsilun
->lun
, lba
,
624 NULL
, num_sectors
* iscsilun
->block_size
,
625 iscsilun
->block_size
, 0, 0, fua
, 0, 0,
626 iscsi_co_generic_cb
, &iTask
);
628 iTask
.task
= iscsi_write10_task(iscsilun
->iscsi
, iscsilun
->lun
, lba
,
629 NULL
, num_sectors
* iscsilun
->block_size
,
630 iscsilun
->block_size
, 0, 0, fua
, 0, 0,
631 iscsi_co_generic_cb
, &iTask
);
634 if (iTask
.task
== NULL
) {
637 #if LIBISCSI_API_VERSION < (20160603)
638 scsi_task_set_iov_out(iTask
.task
, (struct scsi_iovec
*) iov
->iov
,
641 while (!iTask
.complete
) {
642 iscsi_set_events(iscsilun
);
643 qemu_coroutine_yield();
646 if (iTask
.task
!= NULL
) {
647 scsi_free_scsi_task(iTask
.task
);
651 if (iTask
.do_retry
) {
656 if (iTask
.status
!= SCSI_STATUS_GOOD
) {
657 iscsi_allocmap_set_invalid(iscsilun
, sector_num
, nb_sectors
);
658 return iTask
.err_code
;
661 iscsi_allocmap_set_allocated(iscsilun
, sector_num
, nb_sectors
);
668 static int64_t coroutine_fn
iscsi_co_get_block_status(BlockDriverState
*bs
,
670 int nb_sectors
, int *pnum
,
671 BlockDriverState
**file
)
673 IscsiLun
*iscsilun
= bs
->opaque
;
674 struct scsi_get_lba_status
*lbas
= NULL
;
675 struct scsi_lba_status_descriptor
*lbasd
= NULL
;
676 struct IscsiTask iTask
;
679 iscsi_co_init_iscsitask(iscsilun
, &iTask
);
681 if (!is_sector_request_lun_aligned(sector_num
, nb_sectors
, iscsilun
)) {
686 /* default to all sectors allocated */
687 ret
= BDRV_BLOCK_DATA
;
688 ret
|= (sector_num
<< BDRV_SECTOR_BITS
) | BDRV_BLOCK_OFFSET_VALID
;
691 /* LUN does not support logical block provisioning */
692 if (!iscsilun
->lbpme
) {
697 if (iscsi_get_lba_status_task(iscsilun
->iscsi
, iscsilun
->lun
,
698 sector_qemu2lun(sector_num
, iscsilun
),
699 8 + 16, iscsi_co_generic_cb
,
705 while (!iTask
.complete
) {
706 iscsi_set_events(iscsilun
);
707 qemu_coroutine_yield();
710 if (iTask
.do_retry
) {
711 if (iTask
.task
!= NULL
) {
712 scsi_free_scsi_task(iTask
.task
);
719 if (iTask
.status
!= SCSI_STATUS_GOOD
) {
720 /* in case the get_lba_status_callout fails (i.e.
721 * because the device is busy or the cmd is not
722 * supported) we pretend all blocks are allocated
723 * for backwards compatibility */
727 lbas
= scsi_datain_unmarshall(iTask
.task
);
733 lbasd
= &lbas
->descriptors
[0];
735 if (sector_qemu2lun(sector_num
, iscsilun
) != lbasd
->lba
) {
740 *pnum
= sector_lun2qemu(lbasd
->num_blocks
, iscsilun
);
742 if (lbasd
->provisioning
== SCSI_PROVISIONING_TYPE_DEALLOCATED
||
743 lbasd
->provisioning
== SCSI_PROVISIONING_TYPE_ANCHORED
) {
744 ret
&= ~BDRV_BLOCK_DATA
;
745 if (iscsilun
->lbprz
) {
746 ret
|= BDRV_BLOCK_ZERO
;
750 if (ret
& BDRV_BLOCK_ZERO
) {
751 iscsi_allocmap_set_unallocated(iscsilun
, sector_num
, *pnum
);
753 iscsi_allocmap_set_allocated(iscsilun
, sector_num
, *pnum
);
756 if (*pnum
> nb_sectors
) {
760 if (iTask
.task
!= NULL
) {
761 scsi_free_scsi_task(iTask
.task
);
763 if (ret
> 0 && ret
& BDRV_BLOCK_OFFSET_VALID
) {
769 static int coroutine_fn
iscsi_co_readv(BlockDriverState
*bs
,
770 int64_t sector_num
, int nb_sectors
,
773 IscsiLun
*iscsilun
= bs
->opaque
;
774 struct IscsiTask iTask
;
776 uint32_t num_sectors
;
778 if (!is_sector_request_lun_aligned(sector_num
, nb_sectors
, iscsilun
)) {
782 if (bs
->bl
.max_transfer
) {
783 assert(nb_sectors
<< BDRV_SECTOR_BITS
<= bs
->bl
.max_transfer
);
786 /* if cache.direct is off and we have a valid entry in our allocation map
787 * we can skip checking the block status and directly return zeroes if
788 * the request falls within an unallocated area */
789 if (iscsi_allocmap_is_valid(iscsilun
, sector_num
, nb_sectors
) &&
790 !iscsi_allocmap_is_allocated(iscsilun
, sector_num
, nb_sectors
)) {
791 qemu_iovec_memset(iov
, 0, 0x00, iov
->size
);
795 if (nb_sectors
>= ISCSI_CHECKALLOC_THRES
&&
796 !iscsi_allocmap_is_valid(iscsilun
, sector_num
, nb_sectors
) &&
797 !iscsi_allocmap_is_allocated(iscsilun
, sector_num
, nb_sectors
)) {
799 BlockDriverState
*file
;
800 /* check the block status from the beginning of the cluster
801 * containing the start sector */
802 int64_t ret
= iscsi_co_get_block_status(bs
,
803 sector_num
- sector_num
% iscsilun
->cluster_sectors
,
804 BDRV_REQUEST_MAX_SECTORS
, &pnum
, &file
);
808 /* if the whole request falls into an unallocated area we can avoid
809 * to read and directly return zeroes instead */
810 if (ret
& BDRV_BLOCK_ZERO
&&
811 pnum
>= nb_sectors
+ sector_num
% iscsilun
->cluster_sectors
) {
812 qemu_iovec_memset(iov
, 0, 0x00, iov
->size
);
817 lba
= sector_qemu2lun(sector_num
, iscsilun
);
818 num_sectors
= sector_qemu2lun(nb_sectors
, iscsilun
);
820 iscsi_co_init_iscsitask(iscsilun
, &iTask
);
822 if (iscsilun
->use_16_for_rw
) {
823 #if LIBISCSI_API_VERSION >= (20160603)
824 iTask
.task
= iscsi_read16_iov_task(iscsilun
->iscsi
, iscsilun
->lun
, lba
,
825 num_sectors
* iscsilun
->block_size
,
826 iscsilun
->block_size
, 0, 0, 0, 0, 0,
827 iscsi_co_generic_cb
, &iTask
,
828 (struct scsi_iovec
*)iov
->iov
, iov
->niov
);
830 iTask
.task
= iscsi_read10_iov_task(iscsilun
->iscsi
, iscsilun
->lun
, lba
,
831 num_sectors
* iscsilun
->block_size
,
832 iscsilun
->block_size
,
834 iscsi_co_generic_cb
, &iTask
,
835 (struct scsi_iovec
*)iov
->iov
, iov
->niov
);
838 iTask
.task
= iscsi_read16_task(iscsilun
->iscsi
, iscsilun
->lun
, lba
,
839 num_sectors
* iscsilun
->block_size
,
840 iscsilun
->block_size
, 0, 0, 0, 0, 0,
841 iscsi_co_generic_cb
, &iTask
);
843 iTask
.task
= iscsi_read10_task(iscsilun
->iscsi
, iscsilun
->lun
, lba
,
844 num_sectors
* iscsilun
->block_size
,
845 iscsilun
->block_size
,
847 iscsi_co_generic_cb
, &iTask
);
850 if (iTask
.task
== NULL
) {
853 #if LIBISCSI_API_VERSION < (20160603)
854 scsi_task_set_iov_in(iTask
.task
, (struct scsi_iovec
*) iov
->iov
, iov
->niov
);
856 while (!iTask
.complete
) {
857 iscsi_set_events(iscsilun
);
858 qemu_coroutine_yield();
861 if (iTask
.task
!= NULL
) {
862 scsi_free_scsi_task(iTask
.task
);
866 if (iTask
.do_retry
) {
871 if (iTask
.status
!= SCSI_STATUS_GOOD
) {
872 return iTask
.err_code
;
878 static int coroutine_fn
iscsi_co_flush(BlockDriverState
*bs
)
880 IscsiLun
*iscsilun
= bs
->opaque
;
881 struct IscsiTask iTask
;
883 iscsi_co_init_iscsitask(iscsilun
, &iTask
);
885 if (iscsi_synchronizecache10_task(iscsilun
->iscsi
, iscsilun
->lun
, 0, 0, 0,
886 0, iscsi_co_generic_cb
, &iTask
) == NULL
) {
890 while (!iTask
.complete
) {
891 iscsi_set_events(iscsilun
);
892 qemu_coroutine_yield();
895 if (iTask
.task
!= NULL
) {
896 scsi_free_scsi_task(iTask
.task
);
900 if (iTask
.do_retry
) {
905 if (iTask
.status
!= SCSI_STATUS_GOOD
) {
906 return iTask
.err_code
;
914 iscsi_aio_ioctl_cb(struct iscsi_context
*iscsi
, int status
,
915 void *command_data
, void *opaque
)
917 IscsiAIOCB
*acb
= opaque
;
924 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
925 iscsi_get_error(iscsi
));
926 acb
->status
= iscsi_translate_sense(&acb
->task
->sense
);
929 acb
->ioh
->driver_status
= 0;
930 acb
->ioh
->host_status
= 0;
932 acb
->ioh
->status
= status
;
934 #define SG_ERR_DRIVER_SENSE 0x08
936 if (status
== SCSI_STATUS_CHECK_CONDITION
&& acb
->task
->datain
.size
>= 2) {
939 acb
->ioh
->driver_status
|= SG_ERR_DRIVER_SENSE
;
941 acb
->ioh
->sb_len_wr
= acb
->task
->datain
.size
- 2;
942 ss
= (acb
->ioh
->mx_sb_len
>= acb
->ioh
->sb_len_wr
) ?
943 acb
->ioh
->mx_sb_len
: acb
->ioh
->sb_len_wr
;
944 memcpy(acb
->ioh
->sbp
, &acb
->task
->datain
.data
[2], ss
);
947 iscsi_schedule_bh(acb
);
950 static void iscsi_ioctl_bh_completion(void *opaque
)
952 IscsiAIOCB
*acb
= opaque
;
954 qemu_bh_delete(acb
->bh
);
955 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
959 static void iscsi_ioctl_handle_emulated(IscsiAIOCB
*acb
, int req
, void *buf
)
961 BlockDriverState
*bs
= acb
->common
.bs
;
962 IscsiLun
*iscsilun
= bs
->opaque
;
966 case SG_GET_VERSION_NUM
:
970 ((struct sg_scsi_id
*)buf
)->scsi_type
= iscsilun
->type
;
976 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
),
977 iscsi_ioctl_bh_completion
, acb
);
979 qemu_bh_schedule(acb
->bh
);
982 static BlockAIOCB
*iscsi_aio_ioctl(BlockDriverState
*bs
,
983 unsigned long int req
, void *buf
,
984 BlockCompletionFunc
*cb
, void *opaque
)
986 IscsiLun
*iscsilun
= bs
->opaque
;
987 struct iscsi_context
*iscsi
= iscsilun
->iscsi
;
988 struct iscsi_data data
;
991 acb
= qemu_aio_get(&iscsi_aiocb_info
, bs
, cb
, opaque
);
993 acb
->iscsilun
= iscsilun
;
995 acb
->status
= -EINPROGRESS
;
1000 iscsi_ioctl_handle_emulated(acb
, req
, buf
);
1001 return &acb
->common
;
1004 if (acb
->ioh
->cmd_len
> SCSI_CDB_MAX_SIZE
) {
1005 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)",
1006 acb
->ioh
->cmd_len
, SCSI_CDB_MAX_SIZE
);
1007 qemu_aio_unref(acb
);
1011 acb
->task
= malloc(sizeof(struct scsi_task
));
1012 if (acb
->task
== NULL
) {
1013 error_report("iSCSI: Failed to allocate task for scsi command. %s",
1014 iscsi_get_error(iscsi
));
1015 qemu_aio_unref(acb
);
1018 memset(acb
->task
, 0, sizeof(struct scsi_task
));
1020 switch (acb
->ioh
->dxfer_direction
) {
1021 case SG_DXFER_TO_DEV
:
1022 acb
->task
->xfer_dir
= SCSI_XFER_WRITE
;
1024 case SG_DXFER_FROM_DEV
:
1025 acb
->task
->xfer_dir
= SCSI_XFER_READ
;
1028 acb
->task
->xfer_dir
= SCSI_XFER_NONE
;
1032 acb
->task
->cdb_size
= acb
->ioh
->cmd_len
;
1033 memcpy(&acb
->task
->cdb
[0], acb
->ioh
->cmdp
, acb
->ioh
->cmd_len
);
1034 acb
->task
->expxferlen
= acb
->ioh
->dxfer_len
;
1037 if (acb
->task
->xfer_dir
== SCSI_XFER_WRITE
) {
1038 if (acb
->ioh
->iovec_count
== 0) {
1039 data
.data
= acb
->ioh
->dxferp
;
1040 data
.size
= acb
->ioh
->dxfer_len
;
1042 scsi_task_set_iov_out(acb
->task
,
1043 (struct scsi_iovec
*) acb
->ioh
->dxferp
,
1044 acb
->ioh
->iovec_count
);
1048 if (iscsi_scsi_command_async(iscsi
, iscsilun
->lun
, acb
->task
,
1050 (data
.size
> 0) ? &data
: NULL
,
1052 scsi_free_scsi_task(acb
->task
);
1053 qemu_aio_unref(acb
);
1057 /* tell libiscsi to read straight into the buffer we got from ioctl */
1058 if (acb
->task
->xfer_dir
== SCSI_XFER_READ
) {
1059 if (acb
->ioh
->iovec_count
== 0) {
1060 scsi_task_add_data_in_buffer(acb
->task
,
1061 acb
->ioh
->dxfer_len
,
1064 scsi_task_set_iov_in(acb
->task
,
1065 (struct scsi_iovec
*) acb
->ioh
->dxferp
,
1066 acb
->ioh
->iovec_count
);
1070 iscsi_set_events(iscsilun
);
1072 return &acb
->common
;
1078 iscsi_getlength(BlockDriverState
*bs
)
1080 IscsiLun
*iscsilun
= bs
->opaque
;
1083 len
= iscsilun
->num_blocks
;
1084 len
*= iscsilun
->block_size
;
1090 coroutine_fn
iscsi_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int count
)
1092 IscsiLun
*iscsilun
= bs
->opaque
;
1093 struct IscsiTask iTask
;
1094 struct unmap_list list
;
1096 if (!is_byte_request_lun_aligned(offset
, count
, iscsilun
)) {
1100 if (!iscsilun
->lbp
.lbpu
) {
1101 /* UNMAP is not supported by the target */
1105 list
.lba
= offset
/ iscsilun
->block_size
;
1106 list
.num
= count
/ iscsilun
->block_size
;
1108 iscsi_co_init_iscsitask(iscsilun
, &iTask
);
1110 if (iscsi_unmap_task(iscsilun
->iscsi
, iscsilun
->lun
, 0, 0, &list
, 1,
1111 iscsi_co_generic_cb
, &iTask
) == NULL
) {
1115 while (!iTask
.complete
) {
1116 iscsi_set_events(iscsilun
);
1117 qemu_coroutine_yield();
1120 if (iTask
.task
!= NULL
) {
1121 scsi_free_scsi_task(iTask
.task
);
1125 if (iTask
.do_retry
) {
1130 if (iTask
.status
== SCSI_STATUS_CHECK_CONDITION
) {
1131 /* the target might fail with a check condition if it
1132 is not happy with the alignment of the UNMAP request
1133 we silently fail in this case */
1137 if (iTask
.status
!= SCSI_STATUS_GOOD
) {
1138 return iTask
.err_code
;
1141 iscsi_allocmap_set_invalid(iscsilun
, offset
>> BDRV_SECTOR_BITS
,
1142 count
>> BDRV_SECTOR_BITS
);
1148 coroutine_fn
iscsi_co_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
,
1149 int count
, BdrvRequestFlags flags
)
1151 IscsiLun
*iscsilun
= bs
->opaque
;
1152 struct IscsiTask iTask
;
1155 bool use_16_for_ws
= iscsilun
->use_16_for_rw
;
1157 if (!is_byte_request_lun_aligned(offset
, count
, iscsilun
)) {
1161 if (flags
& BDRV_REQ_MAY_UNMAP
) {
1162 if (!use_16_for_ws
&& !iscsilun
->lbp
.lbpws10
) {
1163 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
1164 use_16_for_ws
= true;
1166 if (use_16_for_ws
&& !iscsilun
->lbp
.lbpws
) {
1167 /* WRITESAME16 with UNMAP is not supported by the target,
1168 * fall back and try WRITESAME10/16 without UNMAP */
1169 flags
&= ~BDRV_REQ_MAY_UNMAP
;
1170 use_16_for_ws
= iscsilun
->use_16_for_rw
;
1174 if (!(flags
& BDRV_REQ_MAY_UNMAP
) && !iscsilun
->has_write_same
) {
1175 /* WRITESAME without UNMAP is not supported by the target */
1179 lba
= offset
/ iscsilun
->block_size
;
1180 nb_blocks
= count
/ iscsilun
->block_size
;
1182 if (iscsilun
->zeroblock
== NULL
) {
1183 iscsilun
->zeroblock
= g_try_malloc0(iscsilun
->block_size
);
1184 if (iscsilun
->zeroblock
== NULL
) {
1189 iscsi_co_init_iscsitask(iscsilun
, &iTask
);
1191 if (use_16_for_ws
) {
1192 iTask
.task
= iscsi_writesame16_task(iscsilun
->iscsi
, iscsilun
->lun
, lba
,
1193 iscsilun
->zeroblock
, iscsilun
->block_size
,
1194 nb_blocks
, 0, !!(flags
& BDRV_REQ_MAY_UNMAP
),
1195 0, 0, iscsi_co_generic_cb
, &iTask
);
1197 iTask
.task
= iscsi_writesame10_task(iscsilun
->iscsi
, iscsilun
->lun
, lba
,
1198 iscsilun
->zeroblock
, iscsilun
->block_size
,
1199 nb_blocks
, 0, !!(flags
& BDRV_REQ_MAY_UNMAP
),
1200 0, 0, iscsi_co_generic_cb
, &iTask
);
1202 if (iTask
.task
== NULL
) {
1206 while (!iTask
.complete
) {
1207 iscsi_set_events(iscsilun
);
1208 qemu_coroutine_yield();
1211 if (iTask
.status
== SCSI_STATUS_CHECK_CONDITION
&&
1212 iTask
.task
->sense
.key
== SCSI_SENSE_ILLEGAL_REQUEST
&&
1213 (iTask
.task
->sense
.ascq
== SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE
||
1214 iTask
.task
->sense
.ascq
== SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB
)) {
1215 /* WRITE SAME is not supported by the target */
1216 iscsilun
->has_write_same
= false;
1217 scsi_free_scsi_task(iTask
.task
);
1221 if (iTask
.task
!= NULL
) {
1222 scsi_free_scsi_task(iTask
.task
);
1226 if (iTask
.do_retry
) {
1231 if (iTask
.status
!= SCSI_STATUS_GOOD
) {
1232 iscsi_allocmap_set_invalid(iscsilun
, offset
>> BDRV_SECTOR_BITS
,
1233 count
>> BDRV_SECTOR_BITS
);
1234 return iTask
.err_code
;
1237 if (flags
& BDRV_REQ_MAY_UNMAP
) {
1238 iscsi_allocmap_set_invalid(iscsilun
, offset
>> BDRV_SECTOR_BITS
,
1239 count
>> BDRV_SECTOR_BITS
);
1241 iscsi_allocmap_set_allocated(iscsilun
, offset
>> BDRV_SECTOR_BITS
,
1242 count
>> BDRV_SECTOR_BITS
);
1248 static void parse_chap(struct iscsi_context
*iscsi
, const char *target
,
1253 const char *user
= NULL
;
1254 const char *password
= NULL
;
1255 const char *secretid
;
1256 char *secret
= NULL
;
1258 list
= qemu_find_opts("iscsi");
1263 opts
= qemu_opts_find(list
, target
);
1265 opts
= QTAILQ_FIRST(&list
->head
);
1271 user
= qemu_opt_get(opts
, "user");
1276 secretid
= qemu_opt_get(opts
, "password-secret");
1277 password
= qemu_opt_get(opts
, "password");
1278 if (secretid
&& password
) {
1279 error_setg(errp
, "'password' and 'password-secret' properties are "
1280 "mutually exclusive");
1284 secret
= qcrypto_secret_lookup_as_utf8(secretid
, errp
);
1289 } else if (!password
) {
1290 error_setg(errp
, "CHAP username specified but no password was given");
1294 if (iscsi_set_initiator_username_pwd(iscsi
, user
, password
)) {
1295 error_setg(errp
, "Failed to set initiator username and password");
1301 static void parse_header_digest(struct iscsi_context
*iscsi
, const char *target
,
1306 const char *digest
= NULL
;
1308 list
= qemu_find_opts("iscsi");
1313 opts
= qemu_opts_find(list
, target
);
1315 opts
= QTAILQ_FIRST(&list
->head
);
1321 digest
= qemu_opt_get(opts
, "header-digest");
1326 if (!strcmp(digest
, "CRC32C")) {
1327 iscsi_set_header_digest(iscsi
, ISCSI_HEADER_DIGEST_CRC32C
);
1328 } else if (!strcmp(digest
, "NONE")) {
1329 iscsi_set_header_digest(iscsi
, ISCSI_HEADER_DIGEST_NONE
);
1330 } else if (!strcmp(digest
, "CRC32C-NONE")) {
1331 iscsi_set_header_digest(iscsi
, ISCSI_HEADER_DIGEST_CRC32C_NONE
);
1332 } else if (!strcmp(digest
, "NONE-CRC32C")) {
1333 iscsi_set_header_digest(iscsi
, ISCSI_HEADER_DIGEST_NONE_CRC32C
);
1335 error_setg(errp
, "Invalid header-digest setting : %s", digest
);
1339 static char *parse_initiator_name(const char *target
)
1345 UuidInfo
*uuid_info
;
1347 list
= qemu_find_opts("iscsi");
1349 opts
= qemu_opts_find(list
, target
);
1351 opts
= QTAILQ_FIRST(&list
->head
);
1354 name
= qemu_opt_get(opts
, "initiator-name");
1356 return g_strdup(name
);
1361 uuid_info
= qmp_query_uuid(NULL
);
1362 if (strcmp(uuid_info
->UUID
, UUID_NONE
) == 0) {
1363 name
= qemu_get_vm_name();
1365 name
= uuid_info
->UUID
;
1367 iscsi_name
= g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1368 name
? ":" : "", name
? name
: "");
1369 qapi_free_UuidInfo(uuid_info
);
1373 static int parse_timeout(const char *target
)
1377 const char *timeout
;
1379 list
= qemu_find_opts("iscsi");
1381 opts
= qemu_opts_find(list
, target
);
1383 opts
= QTAILQ_FIRST(&list
->head
);
1386 timeout
= qemu_opt_get(opts
, "timeout");
1388 return atoi(timeout
);
1396 static void iscsi_nop_timed_event(void *opaque
)
1398 IscsiLun
*iscsilun
= opaque
;
1400 aio_context_acquire(iscsilun
->aio_context
);
1401 if (iscsi_get_nops_in_flight(iscsilun
->iscsi
) >= MAX_NOP_FAILURES
) {
1402 error_report("iSCSI: NOP timeout. Reconnecting...");
1403 iscsilun
->request_timed_out
= true;
1404 } else if (iscsi_nop_out_async(iscsilun
->iscsi
, NULL
, NULL
, 0, NULL
) != 0) {
1405 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1409 timer_mod(iscsilun
->nop_timer
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) + NOP_INTERVAL
);
1410 iscsi_set_events(iscsilun
);
1413 aio_context_release(iscsilun
->aio_context
);
1416 static void iscsi_readcapacity_sync(IscsiLun
*iscsilun
, Error
**errp
)
1418 struct scsi_task
*task
= NULL
;
1419 struct scsi_readcapacity10
*rc10
= NULL
;
1420 struct scsi_readcapacity16
*rc16
= NULL
;
1421 int retries
= ISCSI_CMD_RETRIES
;
1425 scsi_free_scsi_task(task
);
1429 switch (iscsilun
->type
) {
1431 task
= iscsi_readcapacity16_sync(iscsilun
->iscsi
, iscsilun
->lun
);
1432 if (task
!= NULL
&& task
->status
== SCSI_STATUS_GOOD
) {
1433 rc16
= scsi_datain_unmarshall(task
);
1435 error_setg(errp
, "iSCSI: Failed to unmarshall readcapacity16 data.");
1437 iscsilun
->block_size
= rc16
->block_length
;
1438 iscsilun
->num_blocks
= rc16
->returned_lba
+ 1;
1439 iscsilun
->lbpme
= !!rc16
->lbpme
;
1440 iscsilun
->lbprz
= !!rc16
->lbprz
;
1441 iscsilun
->use_16_for_rw
= (rc16
->returned_lba
> 0xffffffff);
1445 if (task
!= NULL
&& task
->status
== SCSI_STATUS_CHECK_CONDITION
1446 && task
->sense
.key
== SCSI_SENSE_UNIT_ATTENTION
) {
1449 /* Fall through and try READ CAPACITY(10) instead. */
1451 task
= iscsi_readcapacity10_sync(iscsilun
->iscsi
, iscsilun
->lun
, 0, 0);
1452 if (task
!= NULL
&& task
->status
== SCSI_STATUS_GOOD
) {
1453 rc10
= scsi_datain_unmarshall(task
);
1455 error_setg(errp
, "iSCSI: Failed to unmarshall readcapacity10 data.");
1457 iscsilun
->block_size
= rc10
->block_size
;
1458 if (rc10
->lba
== 0) {
1459 /* blank disk loaded */
1460 iscsilun
->num_blocks
= 0;
1462 iscsilun
->num_blocks
= rc10
->lba
+ 1;
1470 } while (task
!= NULL
&& task
->status
== SCSI_STATUS_CHECK_CONDITION
1471 && task
->sense
.key
== SCSI_SENSE_UNIT_ATTENTION
1474 if (task
== NULL
|| task
->status
!= SCSI_STATUS_GOOD
) {
1475 error_setg(errp
, "iSCSI: failed to send readcapacity10/16 command");
1476 } else if (!iscsilun
->block_size
||
1477 iscsilun
->block_size
% BDRV_SECTOR_SIZE
) {
1478 error_setg(errp
, "iSCSI: the target returned an invalid "
1479 "block size of %d.", iscsilun
->block_size
);
1482 scsi_free_scsi_task(task
);
1486 /* TODO Convert to fine grained options */
1487 static QemuOptsList runtime_opts
= {
1489 .head
= QTAILQ_HEAD_INITIALIZER(runtime_opts
.head
),
1493 .type
= QEMU_OPT_STRING
,
1494 .help
= "URL to the iscsi image",
1496 { /* end of list */ }
1500 static struct scsi_task
*iscsi_do_inquiry(struct iscsi_context
*iscsi
, int lun
,
1501 int evpd
, int pc
, void **inq
, Error
**errp
)
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
) {
1509 full_size
= scsi_datain_getfullsize(task
);
1510 if (full_size
> task
->datain
.size
) {
1511 scsi_free_scsi_task(task
);
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
) {
1520 *inq
= scsi_datain_unmarshall(task
);
1522 error_setg(errp
, "iSCSI: failed to unmarshall inquiry datain blob");
1529 error_setg(errp
, "iSCSI: Inquiry command failed : %s",
1530 iscsi_get_error(iscsi
));
1533 scsi_free_scsi_task(task
);
1538 static void iscsi_detach_aio_context(BlockDriverState
*bs
)
1540 IscsiLun
*iscsilun
= bs
->opaque
;
1542 aio_set_fd_handler(iscsilun
->aio_context
, iscsi_get_fd(iscsilun
->iscsi
),
1543 false, NULL
, NULL
, NULL
, NULL
);
1544 iscsilun
->events
= 0;
1546 if (iscsilun
->nop_timer
) {
1547 timer_del(iscsilun
->nop_timer
);
1548 timer_free(iscsilun
->nop_timer
);
1549 iscsilun
->nop_timer
= NULL
;
1551 if (iscsilun
->event_timer
) {
1552 timer_del(iscsilun
->event_timer
);
1553 timer_free(iscsilun
->event_timer
);
1554 iscsilun
->event_timer
= NULL
;
1558 static void iscsi_attach_aio_context(BlockDriverState
*bs
,
1559 AioContext
*new_context
)
1561 IscsiLun
*iscsilun
= bs
->opaque
;
1563 iscsilun
->aio_context
= new_context
;
1564 iscsi_set_events(iscsilun
);
1566 /* Set up a timer for sending out iSCSI NOPs */
1567 iscsilun
->nop_timer
= aio_timer_new(iscsilun
->aio_context
,
1568 QEMU_CLOCK_REALTIME
, SCALE_MS
,
1569 iscsi_nop_timed_event
, iscsilun
);
1570 timer_mod(iscsilun
->nop_timer
,
1571 qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) + NOP_INTERVAL
);
1573 /* Set up a timer for periodic calls to iscsi_set_events and to
1574 * scan for command timeout */
1575 iscsilun
->event_timer
= aio_timer_new(iscsilun
->aio_context
,
1576 QEMU_CLOCK_REALTIME
, SCALE_MS
,
1577 iscsi_timed_check_events
, iscsilun
);
1578 timer_mod(iscsilun
->event_timer
,
1579 qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) + EVENT_INTERVAL
);
1582 static void iscsi_modesense_sync(IscsiLun
*iscsilun
)
1584 struct scsi_task
*task
;
1585 struct scsi_mode_sense
*ms
= NULL
;
1586 iscsilun
->write_protected
= false;
1587 iscsilun
->dpofua
= false;
1589 task
= iscsi_modesense6_sync(iscsilun
->iscsi
, iscsilun
->lun
,
1590 1, SCSI_MODESENSE_PC_CURRENT
,
1593 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1594 iscsi_get_error(iscsilun
->iscsi
));
1598 if (task
->status
!= SCSI_STATUS_GOOD
) {
1599 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1602 ms
= scsi_datain_unmarshall(task
);
1604 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1605 iscsi_get_error(iscsilun
->iscsi
));
1608 iscsilun
->write_protected
= ms
->device_specific_parameter
& 0x80;
1609 iscsilun
->dpofua
= ms
->device_specific_parameter
& 0x10;
1613 scsi_free_scsi_task(task
);
1618 * We support iscsi url's on the form
1619 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1621 static int iscsi_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
1624 IscsiLun
*iscsilun
= bs
->opaque
;
1625 struct iscsi_context
*iscsi
= NULL
;
1626 struct iscsi_url
*iscsi_url
= NULL
;
1627 struct scsi_task
*task
= NULL
;
1628 struct scsi_inquiry_standard
*inq
= NULL
;
1629 struct scsi_inquiry_supported_pages
*inq_vpd
;
1630 char *initiator_name
= NULL
;
1632 Error
*local_err
= NULL
;
1633 const char *filename
;
1634 int i
, ret
= 0, timeout
= 0;
1636 opts
= qemu_opts_create(&runtime_opts
, NULL
, 0, &error_abort
);
1637 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1639 error_propagate(errp
, local_err
);
1644 filename
= qemu_opt_get(opts
, "filename");
1646 iscsi_url
= iscsi_parse_full_url(iscsi
, filename
);
1647 if (iscsi_url
== NULL
) {
1648 error_setg(errp
, "Failed to parse URL : %s", filename
);
1653 memset(iscsilun
, 0, sizeof(IscsiLun
));
1655 initiator_name
= parse_initiator_name(iscsi_url
->target
);
1657 iscsi
= iscsi_create_context(initiator_name
);
1658 if (iscsi
== NULL
) {
1659 error_setg(errp
, "iSCSI: Failed to create iSCSI context.");
1663 #if LIBISCSI_API_VERSION >= (20160603)
1664 if (iscsi_init_transport(iscsi
, iscsi_url
->transport
)) {
1665 error_setg(errp
, ("Error initializing transport."));
1670 if (iscsi_set_targetname(iscsi
, iscsi_url
->target
)) {
1671 error_setg(errp
, "iSCSI: Failed to set target name.");
1676 if (iscsi_url
->user
[0] != '\0') {
1677 ret
= iscsi_set_initiator_username_pwd(iscsi
, iscsi_url
->user
,
1680 error_setg(errp
, "Failed to set initiator username and password");
1686 /* check if we got CHAP username/password via the options */
1687 parse_chap(iscsi
, iscsi_url
->target
, &local_err
);
1688 if (local_err
!= NULL
) {
1689 error_propagate(errp
, local_err
);
1694 if (iscsi_set_session_type(iscsi
, ISCSI_SESSION_NORMAL
) != 0) {
1695 error_setg(errp
, "iSCSI: Failed to set session type to normal.");
1700 iscsi_set_header_digest(iscsi
, ISCSI_HEADER_DIGEST_NONE_CRC32C
);
1702 /* check if we got HEADER_DIGEST via the options */
1703 parse_header_digest(iscsi
, iscsi_url
->target
, &local_err
);
1704 if (local_err
!= NULL
) {
1705 error_propagate(errp
, local_err
);
1710 /* timeout handling is broken in libiscsi before 1.15.0 */
1711 timeout
= parse_timeout(iscsi_url
->target
);
1712 #if LIBISCSI_API_VERSION >= 20150621
1713 iscsi_set_timeout(iscsi
, timeout
);
1716 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1720 if (iscsi_full_connect_sync(iscsi
, iscsi_url
->portal
, iscsi_url
->lun
) != 0) {
1721 error_setg(errp
, "iSCSI: Failed to connect to LUN : %s",
1722 iscsi_get_error(iscsi
));
1727 iscsilun
->iscsi
= iscsi
;
1728 iscsilun
->aio_context
= bdrv_get_aio_context(bs
);
1729 iscsilun
->lun
= iscsi_url
->lun
;
1730 iscsilun
->has_write_same
= true;
1732 task
= iscsi_do_inquiry(iscsilun
->iscsi
, iscsilun
->lun
, 0, 0,
1733 (void **) &inq
, errp
);
1738 iscsilun
->type
= inq
->periperal_device_type
;
1739 scsi_free_scsi_task(task
);
1742 iscsi_modesense_sync(iscsilun
);
1743 if (iscsilun
->dpofua
) {
1744 bs
->supported_write_flags
= BDRV_REQ_FUA
;
1746 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
;
1748 /* Check the write protect flag of the LUN if we want to write */
1749 if (iscsilun
->type
== TYPE_DISK
&& (flags
& BDRV_O_RDWR
) &&
1750 iscsilun
->write_protected
) {
1751 error_setg(errp
, "Cannot open a write protected LUN as read-write");
1756 iscsi_readcapacity_sync(iscsilun
, &local_err
);
1757 if (local_err
!= NULL
) {
1758 error_propagate(errp
, local_err
);
1762 bs
->total_sectors
= sector_lun2qemu(iscsilun
->num_blocks
, iscsilun
);
1764 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1765 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1766 * will try to read from the device to guess the image format.
1768 if (iscsilun
->type
!= TYPE_DISK
&& iscsilun
->type
!= TYPE_ROM
) {
1772 task
= iscsi_do_inquiry(iscsilun
->iscsi
, iscsilun
->lun
, 1,
1773 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES
,
1774 (void **) &inq_vpd
, errp
);
1779 for (i
= 0; i
< inq_vpd
->num_pages
; i
++) {
1780 struct scsi_task
*inq_task
;
1781 struct scsi_inquiry_logical_block_provisioning
*inq_lbp
;
1782 struct scsi_inquiry_block_limits
*inq_bl
;
1783 switch (inq_vpd
->pages
[i
]) {
1784 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING
:
1785 inq_task
= iscsi_do_inquiry(iscsilun
->iscsi
, iscsilun
->lun
, 1,
1786 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING
,
1787 (void **) &inq_lbp
, errp
);
1788 if (inq_task
== NULL
) {
1792 memcpy(&iscsilun
->lbp
, inq_lbp
,
1793 sizeof(struct scsi_inquiry_logical_block_provisioning
));
1794 scsi_free_scsi_task(inq_task
);
1796 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS
:
1797 inq_task
= iscsi_do_inquiry(iscsilun
->iscsi
, iscsilun
->lun
, 1,
1798 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS
,
1799 (void **) &inq_bl
, errp
);
1800 if (inq_task
== NULL
) {
1804 memcpy(&iscsilun
->bl
, inq_bl
,
1805 sizeof(struct scsi_inquiry_block_limits
));
1806 scsi_free_scsi_task(inq_task
);
1812 scsi_free_scsi_task(task
);
1815 iscsi_attach_aio_context(bs
, iscsilun
->aio_context
);
1817 /* Guess the internal cluster (page) size of the iscsi target by the means
1818 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1819 * reasonable size */
1820 if (iscsilun
->bl
.opt_unmap_gran
* iscsilun
->block_size
>= 4 * 1024 &&
1821 iscsilun
->bl
.opt_unmap_gran
* iscsilun
->block_size
<= 16 * 1024 * 1024) {
1822 iscsilun
->cluster_sectors
= (iscsilun
->bl
.opt_unmap_gran
*
1823 iscsilun
->block_size
) >> BDRV_SECTOR_BITS
;
1824 if (iscsilun
->lbprz
) {
1825 ret
= iscsi_allocmap_init(iscsilun
, bs
->open_flags
);
1830 qemu_opts_del(opts
);
1831 g_free(initiator_name
);
1832 if (iscsi_url
!= NULL
) {
1833 iscsi_destroy_url(iscsi_url
);
1836 scsi_free_scsi_task(task
);
1840 if (iscsi
!= NULL
) {
1841 if (iscsi_is_logged_in(iscsi
)) {
1842 iscsi_logout_sync(iscsi
);
1844 iscsi_destroy_context(iscsi
);
1846 memset(iscsilun
, 0, sizeof(IscsiLun
));
1851 static void iscsi_close(BlockDriverState
*bs
)
1853 IscsiLun
*iscsilun
= bs
->opaque
;
1854 struct iscsi_context
*iscsi
= iscsilun
->iscsi
;
1856 iscsi_detach_aio_context(bs
);
1857 if (iscsi_is_logged_in(iscsi
)) {
1858 iscsi_logout_sync(iscsi
);
1860 iscsi_destroy_context(iscsi
);
1861 g_free(iscsilun
->zeroblock
);
1862 iscsi_allocmap_free(iscsilun
);
1863 memset(iscsilun
, 0, sizeof(IscsiLun
));
1866 static void iscsi_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1868 /* We don't actually refresh here, but just return data queried in
1869 * iscsi_open(): iscsi targets don't change their limits. */
1871 IscsiLun
*iscsilun
= bs
->opaque
;
1872 uint64_t max_xfer_len
= iscsilun
->use_16_for_rw
? 0xffffffff : 0xffff;
1873 unsigned int block_size
= MAX(BDRV_SECTOR_SIZE
, iscsilun
->block_size
);
1875 assert(iscsilun
->block_size
>= BDRV_SECTOR_SIZE
|| bs
->sg
);
1877 bs
->bl
.request_alignment
= block_size
;
1879 if (iscsilun
->bl
.max_xfer_len
) {
1880 max_xfer_len
= MIN(max_xfer_len
, iscsilun
->bl
.max_xfer_len
);
1883 if (max_xfer_len
* block_size
< INT_MAX
) {
1884 bs
->bl
.max_transfer
= max_xfer_len
* iscsilun
->block_size
;
1887 if (iscsilun
->lbp
.lbpu
) {
1888 if (iscsilun
->bl
.max_unmap
< 0xffffffff / block_size
) {
1889 bs
->bl
.max_pdiscard
=
1890 iscsilun
->bl
.max_unmap
* iscsilun
->block_size
;
1892 bs
->bl
.pdiscard_alignment
=
1893 iscsilun
->bl
.opt_unmap_gran
* iscsilun
->block_size
;
1895 bs
->bl
.pdiscard_alignment
= iscsilun
->block_size
;
1898 if (iscsilun
->bl
.max_ws_len
< 0xffffffff / block_size
) {
1899 bs
->bl
.max_pwrite_zeroes
=
1900 iscsilun
->bl
.max_ws_len
* iscsilun
->block_size
;
1902 if (iscsilun
->lbp
.lbpws
) {
1903 bs
->bl
.pwrite_zeroes_alignment
=
1904 iscsilun
->bl
.opt_unmap_gran
* iscsilun
->block_size
;
1906 bs
->bl
.pwrite_zeroes_alignment
= iscsilun
->block_size
;
1908 if (iscsilun
->bl
.opt_xfer_len
&&
1909 iscsilun
->bl
.opt_xfer_len
< INT_MAX
/ block_size
) {
1910 bs
->bl
.opt_transfer
= pow2floor(iscsilun
->bl
.opt_xfer_len
*
1911 iscsilun
->block_size
);
1915 /* Note that this will not re-establish a connection with an iSCSI target - it
1916 * is effectively a NOP. */
1917 static int iscsi_reopen_prepare(BDRVReopenState
*state
,
1918 BlockReopenQueue
*queue
, Error
**errp
)
1920 IscsiLun
*iscsilun
= state
->bs
->opaque
;
1922 if (state
->flags
& BDRV_O_RDWR
&& iscsilun
->write_protected
) {
1923 error_setg(errp
, "Cannot open a write protected LUN as read-write");
1929 static void iscsi_reopen_commit(BDRVReopenState
*reopen_state
)
1931 IscsiLun
*iscsilun
= reopen_state
->bs
->opaque
;
1933 /* the cache.direct status might have changed */
1934 if (iscsilun
->allocmap
!= NULL
) {
1935 iscsi_allocmap_init(iscsilun
, reopen_state
->flags
);
1939 static int iscsi_truncate(BlockDriverState
*bs
, int64_t offset
)
1941 IscsiLun
*iscsilun
= bs
->opaque
;
1942 Error
*local_err
= NULL
;
1944 if (iscsilun
->type
!= TYPE_DISK
) {
1948 iscsi_readcapacity_sync(iscsilun
, &local_err
);
1949 if (local_err
!= NULL
) {
1950 error_free(local_err
);
1954 if (offset
> iscsi_getlength(bs
)) {
1958 if (iscsilun
->allocmap
!= NULL
) {
1959 iscsi_allocmap_init(iscsilun
, bs
->open_flags
);
1965 static int iscsi_create(const char *filename
, QemuOpts
*opts
, Error
**errp
)
1968 int64_t total_size
= 0;
1969 BlockDriverState
*bs
;
1970 IscsiLun
*iscsilun
= NULL
;
1975 /* Read out options */
1976 total_size
= DIV_ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
1978 bs
->opaque
= g_new0(struct IscsiLun
, 1);
1979 iscsilun
= bs
->opaque
;
1981 bs_options
= qdict_new();
1982 qdict_put(bs_options
, "filename", qstring_from_str(filename
));
1983 ret
= iscsi_open(bs
, bs_options
, 0, NULL
);
1984 QDECREF(bs_options
);
1989 iscsi_detach_aio_context(bs
);
1990 if (iscsilun
->type
!= TYPE_DISK
) {
1994 if (bs
->total_sectors
< total_size
) {
2001 if (iscsilun
->iscsi
!= NULL
) {
2002 iscsi_destroy_context(iscsilun
->iscsi
);
2010 static int iscsi_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2012 IscsiLun
*iscsilun
= bs
->opaque
;
2013 bdi
->unallocated_blocks_are_zero
= iscsilun
->lbprz
;
2014 bdi
->can_write_zeroes_with_unmap
= iscsilun
->lbprz
&& iscsilun
->lbp
.lbpws
;
2015 bdi
->cluster_size
= iscsilun
->cluster_sectors
* BDRV_SECTOR_SIZE
;
2019 static void iscsi_invalidate_cache(BlockDriverState
*bs
,
2022 IscsiLun
*iscsilun
= bs
->opaque
;
2023 iscsi_allocmap_invalidate(iscsilun
);
2026 static QemuOptsList iscsi_create_opts
= {
2027 .name
= "iscsi-create-opts",
2028 .head
= QTAILQ_HEAD_INITIALIZER(iscsi_create_opts
.head
),
2031 .name
= BLOCK_OPT_SIZE
,
2032 .type
= QEMU_OPT_SIZE
,
2033 .help
= "Virtual disk size"
2035 { /* end of list */ }
2039 static BlockDriver bdrv_iscsi
= {
2040 .format_name
= "iscsi",
2041 .protocol_name
= "iscsi",
2043 .instance_size
= sizeof(IscsiLun
),
2044 .bdrv_needs_filename
= true,
2045 .bdrv_file_open
= iscsi_open
,
2046 .bdrv_close
= iscsi_close
,
2047 .bdrv_create
= iscsi_create
,
2048 .create_opts
= &iscsi_create_opts
,
2049 .bdrv_reopen_prepare
= iscsi_reopen_prepare
,
2050 .bdrv_reopen_commit
= iscsi_reopen_commit
,
2051 .bdrv_invalidate_cache
= iscsi_invalidate_cache
,
2053 .bdrv_getlength
= iscsi_getlength
,
2054 .bdrv_get_info
= iscsi_get_info
,
2055 .bdrv_truncate
= iscsi_truncate
,
2056 .bdrv_refresh_limits
= iscsi_refresh_limits
,
2058 .bdrv_co_get_block_status
= iscsi_co_get_block_status
,
2059 .bdrv_co_pdiscard
= iscsi_co_pdiscard
,
2060 .bdrv_co_pwrite_zeroes
= iscsi_co_pwrite_zeroes
,
2061 .bdrv_co_readv
= iscsi_co_readv
,
2062 .bdrv_co_writev_flags
= iscsi_co_writev_flags
,
2063 .bdrv_co_flush_to_disk
= iscsi_co_flush
,
2066 .bdrv_aio_ioctl
= iscsi_aio_ioctl
,
2069 .bdrv_detach_aio_context
= iscsi_detach_aio_context
,
2070 .bdrv_attach_aio_context
= iscsi_attach_aio_context
,
2073 #if LIBISCSI_API_VERSION >= (20160603)
2074 static BlockDriver bdrv_iser
= {
2075 .format_name
= "iser",
2076 .protocol_name
= "iser",
2078 .instance_size
= sizeof(IscsiLun
),
2079 .bdrv_needs_filename
= true,
2080 .bdrv_file_open
= iscsi_open
,
2081 .bdrv_close
= iscsi_close
,
2082 .bdrv_create
= iscsi_create
,
2083 .create_opts
= &iscsi_create_opts
,
2084 .bdrv_reopen_prepare
= iscsi_reopen_prepare
,
2085 .bdrv_reopen_commit
= iscsi_reopen_commit
,
2086 .bdrv_invalidate_cache
= iscsi_invalidate_cache
,
2088 .bdrv_getlength
= iscsi_getlength
,
2089 .bdrv_get_info
= iscsi_get_info
,
2090 .bdrv_truncate
= iscsi_truncate
,
2091 .bdrv_refresh_limits
= iscsi_refresh_limits
,
2093 .bdrv_co_get_block_status
= iscsi_co_get_block_status
,
2094 .bdrv_co_pdiscard
= iscsi_co_pdiscard
,
2095 .bdrv_co_pwrite_zeroes
= iscsi_co_pwrite_zeroes
,
2096 .bdrv_co_readv
= iscsi_co_readv
,
2097 .bdrv_co_writev_flags
= iscsi_co_writev_flags
,
2098 .bdrv_co_flush_to_disk
= iscsi_co_flush
,
2101 .bdrv_aio_ioctl
= iscsi_aio_ioctl
,
2104 .bdrv_detach_aio_context
= iscsi_detach_aio_context
,
2105 .bdrv_attach_aio_context
= iscsi_attach_aio_context
,
2109 static void iscsi_block_init(void)
2111 bdrv_register(&bdrv_iscsi
);
2112 #if LIBISCSI_API_VERSION >= (20160603)
2113 bdrv_register(&bdrv_iser
);
2117 block_init(iscsi_block_init
);