2 * QEMU Block driver for iSCSI images
4 * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
5 * Copyright (c) 2012-2017 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 "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"
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"
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>
60 QEMU_BUILD_BUG_ON((int)SCSI_XFER_NONE
!= (int)ISCSI_XFER_NONE
);
66 typedef struct IscsiLun
{
67 struct iscsi_context
*iscsi
;
68 AioContext
*aio_context
;
70 enum scsi_inquiry_peripheral_device_type type
;
75 QEMUTimer
*event_timer
;
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
;
100 bool write_protected
;
105 bool request_timed_out
;
108 typedef struct IscsiTask
{
113 struct scsi_task
*task
;
116 QEMUTimer retry_timer
;
121 typedef struct IscsiAIOCB
{
125 struct scsi_task
*task
;
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};
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
150 #define ISCSI_CHECKALLOC_THRES 64
159 qemu_bh_delete(acb
->bh
);
161 acb
->common
.cb(acb
->common
.opaque
, acb
->status
);
163 if (acb
->task
!= NULL
) {
164 scsi_free_scsi_task(acb
->task
);
172 iscsi_schedule_bh(IscsiAIOCB
*acb
)
177 acb
->bh
= aio_bh_new(acb
->iscsilun
->aio_context
, iscsi_bh_cb
, acb
);
178 qemu_bh_schedule(acb
->bh
);
183 static void iscsi_co_generic_bh_cb(void *opaque
)
185 struct IscsiTask
*iTask
= opaque
;
188 aio_co_wake(iTask
->co
);
191 static void iscsi_retry_timer_expired(void *opaque
)
193 struct IscsiTask
*iTask
= opaque
;
196 aio_co_wake(iTask
->co
);
200 static inline unsigned exp_random(double mean
)
202 return -mean
* log((double)rand() / RAND_MAX
);
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.
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
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
224 #ifndef LIBISCSI_API_VERSION
225 #define LIBISCSI_API_VERSION 20130701
228 static int iscsi_translate_sense(struct scsi_sense
*sense
)
230 return scsi_sense_to_errno(sense
->key
,
231 (sense
->ascq
& 0xFF00) >> 8,
235 /* Called (via iscsi_service) with QemuMutex held. */
237 iscsi_co_generic_cb(struct iscsi_context
*iscsi
, int status
,
238 void *command_data
, void *opaque
)
240 struct IscsiTask
*iTask
= opaque
;
241 struct scsi_task
*task
= command_data
;
243 iTask
->status
= status
;
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;
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
);
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
));
280 iTask
->err_code
= -error
;
281 iTask
->err_str
= g_strdup(iscsi_get_error(iscsi
));
288 replay_bh_schedule_oneshot_event(iTask
->iscsilun
->aio_context
,
289 iscsi_co_generic_bh_cb
, iTask
);
295 static void coroutine_fn
296 iscsi_co_init_iscsitask(IscsiLun
*iscsilun
, struct IscsiTask
*iTask
)
298 *iTask
= (struct IscsiTask
) {
299 .co
= qemu_coroutine_self(),
300 .iscsilun
= iscsilun
,
306 /* Called (via iscsi_service) with QemuMutex held. */
308 iscsi_abort_task_cb(struct iscsi_context
*iscsi
, int status
, void *command_data
,
311 IscsiAIOCB
*acb
= private_data
;
313 /* If the command callback hasn't been called yet, drop the task */
315 /* Call iscsi_aio_ioctl_cb() with SCSI_STATUS_CANCELLED */
316 iscsi_scsi_cancel_task(iscsi
, acb
->task
);
319 qemu_aio_unref(acb
); /* acquired in iscsi_aio_cancel() */
323 iscsi_aio_cancel(BlockAIOCB
*blockacb
)
325 IscsiAIOCB
*acb
= (IscsiAIOCB
*)blockacb
;
326 IscsiLun
*iscsilun
= acb
->iscsilun
;
328 WITH_QEMU_LOCK_GUARD(&iscsilun
->mutex
) {
330 /* If it was cancelled or completed already, our work is done here */
331 if (acb
->cancelled
|| acb
->status
!= -EINPROGRESS
) {
335 acb
->cancelled
= true;
337 qemu_aio_ref(acb
); /* released in iscsi_abort_task_cb() */
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 */
347 static const AIOCBInfo iscsi_aiocb_info
= {
348 .aiocb_size
= sizeof(IscsiAIOCB
),
349 .cancel_async
= iscsi_aio_cancel
,
354 static void iscsi_process_read(void *arg
);
355 static void iscsi_process_write(void *arg
);
357 /* Called with QemuMutex held. */
359 iscsi_set_events(IscsiLun
*iscsilun
)
361 struct iscsi_context
*iscsi
= iscsilun
->iscsi
;
362 int ev
= iscsi_which_events(iscsi
);
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
,
370 iscsilun
->events
= ev
;
374 static void iscsi_timed_check_events(void *opaque
)
376 IscsiLun
*iscsilun
= opaque
;
378 WITH_QEMU_LOCK_GUARD(&iscsilun
->mutex
) {
379 /* check for timed out requests */
380 iscsi_service(iscsilun
->iscsi
, 0);
382 if (iscsilun
->request_timed_out
) {
383 iscsilun
->request_timed_out
= false;
384 iscsi_reconnect(iscsilun
->iscsi
);
388 * newer versions of libiscsi may return zero events. Ensure we are
389 * able to return to service once this situation changes.
391 iscsi_set_events(iscsilun
);
394 timer_mod(iscsilun
->event_timer
,
395 qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) + EVENT_INTERVAL
);
399 iscsi_process_read(void *arg
)
401 IscsiLun
*iscsilun
= arg
;
402 struct iscsi_context
*iscsi
= iscsilun
->iscsi
;
404 qemu_mutex_lock(&iscsilun
->mutex
);
405 iscsi_service(iscsi
, POLLIN
);
406 iscsi_set_events(iscsilun
);
407 qemu_mutex_unlock(&iscsilun
->mutex
);
411 iscsi_process_write(void *arg
)
413 IscsiLun
*iscsilun
= arg
;
414 struct iscsi_context
*iscsi
= iscsilun
->iscsi
;
416 qemu_mutex_lock(&iscsilun
->mutex
);
417 iscsi_service(iscsi
, POLLOUT
);
418 iscsi_set_events(iscsilun
);
419 qemu_mutex_unlock(&iscsilun
->mutex
);
422 static int64_t sector_lun2qemu(int64_t sector
, IscsiLun
*iscsilun
)
424 return sector
* iscsilun
->block_size
/ BDRV_SECTOR_SIZE
;
427 static int64_t sector_qemu2lun(int64_t sector
, IscsiLun
*iscsilun
)
429 return sector
* BDRV_SECTOR_SIZE
/ iscsilun
->block_size
;
432 static bool is_byte_request_lun_aligned(int64_t offset
, int64_t bytes
,
435 if (offset
% iscsilun
->block_size
|| bytes
% iscsilun
->block_size
) {
436 error_report("iSCSI misaligned request: "
437 "iscsilun->block_size %u, offset %" PRIi64
439 iscsilun
->block_size
, offset
, bytes
);
445 static bool is_sector_request_lun_aligned(int64_t sector_num
, int nb_sectors
,
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
,
454 static void iscsi_allocmap_free(IscsiLun
*iscsilun
)
456 g_free(iscsilun
->allocmap
);
457 g_free(iscsilun
->allocmap_valid
);
458 iscsilun
->allocmap
= NULL
;
459 iscsilun
->allocmap_valid
= NULL
;
463 static int iscsi_allocmap_init(IscsiLun
*iscsilun
, int open_flags
)
465 iscsi_allocmap_free(iscsilun
);
467 assert(iscsilun
->cluster_size
);
468 iscsilun
->allocmap_size
=
469 DIV_ROUND_UP(iscsilun
->num_blocks
* iscsilun
->block_size
,
470 iscsilun
->cluster_size
);
472 iscsilun
->allocmap
= bitmap_try_new(iscsilun
->allocmap_size
);
473 if (!iscsilun
->allocmap
) {
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 */
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
);
495 iscsi_allocmap_update(IscsiLun
*iscsilun
, int64_t offset
,
496 int64_t bytes
, bool allocated
, bool valid
)
498 int64_t cl_num_expanded
, nb_cls_expanded
, cl_num_shrunk
, nb_cls_shrunk
;
500 if (iscsilun
->allocmap
== NULL
) {
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
;
512 bitmap_set(iscsilun
->allocmap
, cl_num_expanded
, nb_cls_expanded
);
514 if (nb_cls_shrunk
> 0) {
515 bitmap_clear(iscsilun
->allocmap
, cl_num_shrunk
, nb_cls_shrunk
);
519 if (iscsilun
->allocmap_valid
== NULL
) {
523 if (nb_cls_shrunk
> 0) {
524 bitmap_set(iscsilun
->allocmap_valid
, cl_num_shrunk
, nb_cls_shrunk
);
527 bitmap_clear(iscsilun
->allocmap_valid
, cl_num_expanded
,
533 iscsi_allocmap_set_allocated(IscsiLun
*iscsilun
, int64_t offset
,
536 iscsi_allocmap_update(iscsilun
, offset
, bytes
, true, true);
540 iscsi_allocmap_set_unallocated(IscsiLun
*iscsilun
, int64_t offset
,
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.
546 iscsi_allocmap_update(iscsilun
, offset
, bytes
, false, true);
549 static void iscsi_allocmap_set_invalid(IscsiLun
*iscsilun
, int64_t offset
,
552 iscsi_allocmap_update(iscsilun
, offset
, bytes
, false, false);
555 static void iscsi_allocmap_invalidate(IscsiLun
*iscsilun
)
557 if (iscsilun
->allocmap
) {
558 bitmap_zero(iscsilun
->allocmap
, iscsilun
->allocmap_size
);
560 if (iscsilun
->allocmap_valid
) {
561 bitmap_zero(iscsilun
->allocmap_valid
, iscsilun
->allocmap_size
);
566 iscsi_allocmap_is_allocated(IscsiLun
*iscsilun
, int64_t offset
,
570 if (iscsilun
->allocmap
== NULL
) {
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
);
579 static inline bool iscsi_allocmap_is_valid(IscsiLun
*iscsilun
,
580 int64_t offset
, int64_t bytes
)
583 if (iscsilun
->allocmap_valid
== NULL
) {
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
);
592 static void coroutine_fn
iscsi_co_wait_for_task(IscsiTask
*iTask
,
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
);
603 static int coroutine_fn
604 iscsi_co_writev(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
,
605 QEMUIOVector
*iov
, int flags
)
607 IscsiLun
*iscsilun
= bs
->opaque
;
608 struct IscsiTask iTask
;
610 uint32_t num_sectors
;
611 bool fua
= flags
& BDRV_REQ_FUA
;
615 assert(iscsilun
->dpofua
);
617 if (!is_sector_request_lun_aligned(sector_num
, nb_sectors
, iscsilun
)) {
621 if (bs
->bl
.max_transfer
) {
622 assert(nb_sectors
<< BDRV_SECTOR_BITS
<= bs
->bl
.max_transfer
);
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
);
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
);
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
);
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
);
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
);
656 if (iTask
.task
== NULL
) {
657 qemu_mutex_unlock(&iscsilun
->mutex
);
660 #if LIBISCSI_API_VERSION < (20160603)
661 scsi_task_set_iov_out(iTask
.task
, (struct scsi_iovec
*) iov
->iov
,
664 iscsi_co_wait_for_task(&iTask
, iscsilun
);
666 if (iTask
.task
!= NULL
) {
667 scsi_free_scsi_task(iTask
.task
);
671 if (iTask
.do_retry
) {
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
,
685 iscsi_allocmap_set_allocated(iscsilun
, sector_num
* BDRV_SECTOR_SIZE
,
686 nb_sectors
* BDRV_SECTOR_SIZE
);
689 qemu_mutex_unlock(&iscsilun
->mutex
);
690 g_free(iTask
.err_str
);
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
,
700 BlockDriverState
**file
)
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
;
709 iscsi_co_init_iscsitask(iscsilun
, &iTask
);
711 assert(QEMU_IS_ALIGNED(offset
| bytes
, iscsilun
->block_size
));
713 /* default to all sectors allocated */
714 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
720 /* LUN does not support logical block provisioning */
721 if (!iscsilun
->lbpme
) {
725 lba
= offset
/ iscsilun
->block_size
;
726 max_bytes
= (iscsilun
->num_blocks
- lba
) * iscsilun
->block_size
;
728 qemu_mutex_lock(&iscsilun
->mutex
);
730 if (iscsi_get_lba_status_task(iscsilun
->iscsi
, iscsilun
->lun
,
731 lba
, 8 + 16, iscsi_co_generic_cb
,
736 iscsi_co_wait_for_task(&iTask
, iscsilun
);
738 if (iTask
.do_retry
) {
739 if (iTask
.task
!= NULL
) {
740 scsi_free_scsi_task(iTask
.task
);
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",
757 lbas
= scsi_datain_unmarshall(iTask
.task
);
758 if (lbas
== NULL
|| lbas
->num_descriptors
== 0) {
763 lbasd
= &lbas
->descriptors
[0];
765 if (lba
!= lbasd
->lba
) {
770 *pnum
= MIN((int64_t) lbasd
->num_blocks
* iscsilun
->block_size
, max_bytes
);
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
;
780 if (ret
& BDRV_BLOCK_ZERO
) {
781 iscsi_allocmap_set_unallocated(iscsilun
, offset
, *pnum
);
783 iscsi_allocmap_set_allocated(iscsilun
, offset
, *pnum
);
787 qemu_mutex_unlock(&iscsilun
->mutex
);
788 g_free(iTask
.err_str
);
790 if (iTask
.task
!= NULL
) {
791 scsi_free_scsi_task(iTask
.task
);
793 if (ret
> 0 && ret
& BDRV_BLOCK_OFFSET_VALID
&& file
) {
799 static int coroutine_fn
iscsi_co_readv(BlockDriverState
*bs
,
800 int64_t sector_num
, int nb_sectors
,
803 IscsiLun
*iscsilun
= bs
->opaque
;
804 struct IscsiTask iTask
;
806 uint32_t num_sectors
;
809 if (!is_sector_request_lun_aligned(sector_num
, nb_sectors
, iscsilun
)) {
813 if (bs
->bl
.max_transfer
) {
814 assert(nb_sectors
<< BDRV_SECTOR_BITS
<= bs
->bl
.max_transfer
);
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
);
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
)) {
834 /* check the block status from the beginning of the cluster
835 * containing the start sector */
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
);
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
);
856 lba
= sector_qemu2lun(sector_num
, iscsilun
);
857 num_sectors
= sector_qemu2lun(nb_sectors
, iscsilun
);
859 iscsi_co_init_iscsitask(iscsilun
, &iTask
);
860 qemu_mutex_lock(&iscsilun
->mutex
);
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
);
870 iTask
.task
= iscsi_read10_iov_task(iscsilun
->iscsi
, iscsilun
->lun
, lba
,
871 num_sectors
* iscsilun
->block_size
,
872 iscsilun
->block_size
,
874 iscsi_co_generic_cb
, &iTask
,
875 (struct scsi_iovec
*)iov
->iov
, iov
->niov
);
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
);
883 iTask
.task
= iscsi_read10_task(iscsilun
->iscsi
, iscsilun
->lun
, lba
,
884 num_sectors
* iscsilun
->block_size
,
885 iscsilun
->block_size
,
887 iscsi_co_generic_cb
, &iTask
);
890 if (iTask
.task
== NULL
) {
891 qemu_mutex_unlock(&iscsilun
->mutex
);
894 #if LIBISCSI_API_VERSION < (20160603)
895 scsi_task_set_iov_in(iTask
.task
, (struct scsi_iovec
*) iov
->iov
, iov
->niov
);
898 iscsi_co_wait_for_task(&iTask
, iscsilun
);
899 if (iTask
.task
!= NULL
) {
900 scsi_free_scsi_task(iTask
.task
);
904 if (iTask
.do_retry
) {
909 if (iTask
.status
!= SCSI_STATUS_GOOD
) {
910 error_report("iSCSI READ10/16 failed at lba %" PRIu64
": %s",
915 qemu_mutex_unlock(&iscsilun
->mutex
);
916 g_free(iTask
.err_str
);
920 static int coroutine_fn
iscsi_co_flush(BlockDriverState
*bs
)
922 IscsiLun
*iscsilun
= bs
->opaque
;
923 struct IscsiTask iTask
;
926 iscsi_co_init_iscsitask(iscsilun
, &iTask
);
927 qemu_mutex_lock(&iscsilun
->mutex
);
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
);
935 iscsi_co_wait_for_task(&iTask
, iscsilun
);
937 if (iTask
.task
!= NULL
) {
938 scsi_free_scsi_task(iTask
.task
);
942 if (iTask
.do_retry
) {
947 if (iTask
.status
!= SCSI_STATUS_GOOD
) {
948 error_report("iSCSI SYNCHRONIZECACHE10 failed: %s", iTask
.err_str
);
952 qemu_mutex_unlock(&iscsilun
->mutex
);
953 g_free(iTask
.err_str
);
958 /* Called (via iscsi_service) with QemuMutex held. */
960 iscsi_aio_ioctl_cb(struct iscsi_context
*iscsi
, int status
,
961 void *command_data
, void *opaque
)
963 IscsiAIOCB
*acb
= opaque
;
965 if (status
== SCSI_STATUS_CANCELLED
) {
967 acb
->status
= -ECANCELED
;
968 iscsi_schedule_bh(acb
);
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
);
980 acb
->ioh
->driver_status
= 0;
981 acb
->ioh
->host_status
= 0;
983 acb
->ioh
->status
= status
;
985 #define SG_ERR_DRIVER_SENSE 0x08
987 if (status
== SCSI_STATUS_CHECK_CONDITION
&& acb
->task
->datain
.size
>= 2) {
990 acb
->ioh
->driver_status
|= SG_ERR_DRIVER_SENSE
;
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
);
997 iscsi_schedule_bh(acb
);
1000 static void iscsi_ioctl_bh_completion(void *opaque
)
1002 IscsiAIOCB
*acb
= opaque
;
1004 qemu_bh_delete(acb
->bh
);
1005 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
1006 qemu_aio_unref(acb
);
1009 static void iscsi_ioctl_handle_emulated(IscsiAIOCB
*acb
, int req
, void *buf
)
1011 BlockDriverState
*bs
= acb
->common
.bs
;
1012 IscsiLun
*iscsilun
= bs
->opaque
;
1016 case SG_GET_VERSION_NUM
:
1017 *(int *)buf
= 30000;
1019 case SG_GET_SCSI_ID
:
1020 ((struct sg_scsi_id
*)buf
)->scsi_type
= iscsilun
->type
;
1026 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
),
1027 iscsi_ioctl_bh_completion
, acb
);
1029 qemu_bh_schedule(acb
->bh
);
1032 static BlockAIOCB
*iscsi_aio_ioctl(BlockDriverState
*bs
,
1033 unsigned long int req
, void *buf
,
1034 BlockCompletionFunc
*cb
, void *opaque
)
1036 IscsiLun
*iscsilun
= bs
->opaque
;
1037 struct iscsi_context
*iscsi
= iscsilun
->iscsi
;
1038 struct iscsi_data data
;
1041 acb
= qemu_aio_get(&iscsi_aiocb_info
, bs
, cb
, opaque
);
1043 acb
->iscsilun
= iscsilun
;
1045 acb
->status
= -EINPROGRESS
;
1047 acb
->cancelled
= false;
1050 iscsi_ioctl_handle_emulated(acb
, req
, buf
);
1051 return &acb
->common
;
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
);
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
);
1069 memset(acb
->task
, 0, sizeof(struct scsi_task
));
1071 switch (acb
->ioh
->dxfer_direction
) {
1072 case SG_DXFER_TO_DEV
:
1073 acb
->task
->xfer_dir
= SCSI_XFER_WRITE
;
1075 case SG_DXFER_FROM_DEV
:
1076 acb
->task
->xfer_dir
= SCSI_XFER_READ
;
1079 acb
->task
->xfer_dir
= SCSI_XFER_NONE
;
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
;
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
;
1094 scsi_task_set_iov_out(acb
->task
,
1095 (struct scsi_iovec
*) acb
->ioh
->dxferp
,
1096 acb
->ioh
->iovec_count
);
1100 if (iscsi_scsi_command_async(iscsi
, iscsilun
->lun
, acb
->task
,
1102 (data
.size
> 0) ? &data
: NULL
,
1104 qemu_mutex_unlock(&iscsilun
->mutex
);
1105 scsi_free_scsi_task(acb
->task
);
1106 qemu_aio_unref(acb
);
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
,
1117 scsi_task_set_iov_in(acb
->task
,
1118 (struct scsi_iovec
*) acb
->ioh
->dxferp
,
1119 acb
->ioh
->iovec_count
);
1123 iscsi_set_events(iscsilun
);
1124 qemu_mutex_unlock(&iscsilun
->mutex
);
1126 return &acb
->common
;
1131 static int64_t coroutine_fn
1132 iscsi_co_getlength(BlockDriverState
*bs
)
1134 IscsiLun
*iscsilun
= bs
->opaque
;
1137 len
= iscsilun
->num_blocks
;
1138 len
*= iscsilun
->block_size
;
1144 coroutine_fn
iscsi_co_pdiscard(BlockDriverState
*bs
, int64_t offset
,
1147 IscsiLun
*iscsilun
= bs
->opaque
;
1148 struct IscsiTask iTask
;
1149 struct unmap_list list
;
1152 if (!is_byte_request_lun_aligned(offset
, bytes
, iscsilun
)) {
1156 if (!iscsilun
->lbp
.lbpu
) {
1157 /* UNMAP is not supported by the target */
1162 * We don't want to overflow list.num which is uint32_t.
1163 * We rely on our max_pdiscard.
1165 assert(bytes
/ iscsilun
->block_size
<= UINT32_MAX
);
1167 list
.lba
= offset
/ iscsilun
->block_size
;
1168 list
.num
= bytes
/ iscsilun
->block_size
;
1170 iscsi_co_init_iscsitask(iscsilun
, &iTask
);
1171 qemu_mutex_lock(&iscsilun
->mutex
);
1173 if (iscsi_unmap_task(iscsilun
->iscsi
, iscsilun
->lun
, 0, 0, &list
, 1,
1174 iscsi_co_generic_cb
, &iTask
) == NULL
) {
1179 iscsi_co_wait_for_task(&iTask
, iscsilun
);
1181 if (iTask
.task
!= NULL
) {
1182 scsi_free_scsi_task(iTask
.task
);
1186 if (iTask
.do_retry
) {
1191 iscsi_allocmap_set_invalid(iscsilun
, offset
, bytes
);
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 */
1200 if (iTask
.status
!= SCSI_STATUS_GOOD
) {
1201 error_report("iSCSI UNMAP failed at lba %" PRIu64
": %s",
1202 list
.lba
, iTask
.err_str
);
1208 qemu_mutex_unlock(&iscsilun
->mutex
);
1209 g_free(iTask
.err_str
);
1214 coroutine_fn
iscsi_co_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
,
1215 int64_t bytes
, BdrvRequestFlags flags
)
1217 IscsiLun
*iscsilun
= bs
->opaque
;
1218 struct IscsiTask iTask
;
1221 bool use_16_for_ws
= iscsilun
->use_16_for_rw
;
1224 if (!is_byte_request_lun_aligned(offset
, bytes
, iscsilun
)) {
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;
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
;
1241 if (!(flags
& BDRV_REQ_MAY_UNMAP
) && !iscsilun
->has_write_same
) {
1242 /* WRITESAME without UNMAP is not supported by the target */
1246 lba
= offset
/ iscsilun
->block_size
;
1247 nb_blocks
= bytes
/ iscsilun
->block_size
;
1249 if (iscsilun
->zeroblock
== NULL
) {
1250 iscsilun
->zeroblock
= g_try_malloc0(iscsilun
->block_size
);
1251 if (iscsilun
->zeroblock
== NULL
) {
1256 qemu_mutex_lock(&iscsilun
->mutex
);
1257 iscsi_co_init_iscsitask(iscsilun
, &iTask
);
1259 if (use_16_for_ws
) {
1261 * iscsi_writesame16_task num_blocks argument is uint32_t. We rely here
1262 * on our max_pwrite_zeroes limit.
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
);
1271 * iscsi_writesame10_task num_blocks argument is uint16_t. We rely here
1272 * on our max_pwrite_zeroes limit.
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
);
1280 if (iTask
.task
== NULL
) {
1281 qemu_mutex_unlock(&iscsilun
->mutex
);
1285 iscsi_co_wait_for_task(&iTask
, iscsilun
);
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
);
1298 if (iTask
.task
!= NULL
) {
1299 scsi_free_scsi_task(iTask
.task
);
1303 if (iTask
.do_retry
) {
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
);
1316 if (flags
& BDRV_REQ_MAY_UNMAP
) {
1317 iscsi_allocmap_set_invalid(iscsilun
, offset
, bytes
);
1319 iscsi_allocmap_set_allocated(iscsilun
, offset
, bytes
);
1323 qemu_mutex_unlock(&iscsilun
->mutex
);
1324 g_free(iTask
.err_str
);
1328 static void apply_chap(struct iscsi_context
*iscsi
, QemuOpts
*opts
,
1331 const char *user
= NULL
;
1332 const char *password
= NULL
;
1333 const char *secretid
;
1334 char *secret
= NULL
;
1336 user
= qemu_opt_get(opts
, "user");
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");
1349 secret
= qcrypto_secret_lookup_as_utf8(secretid
, errp
);
1354 } else if (!password
) {
1355 error_setg(errp
, "CHAP username specified but no password was given");
1358 warn_report("iSCSI block driver 'password' option is deprecated, "
1359 "use 'password-secret' instead");
1362 if (iscsi_set_initiator_username_pwd(iscsi
, user
, password
)) {
1363 error_setg(errp
, "Failed to set initiator username and password");
1369 static void apply_header_digest(struct iscsi_context
*iscsi
, QemuOpts
*opts
,
1372 const char *digest
= NULL
;
1374 digest
= qemu_opt_get(opts
, "header-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
);
1386 error_setg(errp
, "Invalid header-digest setting : %s", digest
);
1390 static char *get_initiator_name(QemuOpts
*opts
)
1394 UuidInfo
*uuid_info
;
1396 name
= qemu_opt_get(opts
, "initiator-name");
1398 return g_strdup(name
);
1401 uuid_info
= qmp_query_uuid(NULL
);
1402 if (strcmp(uuid_info
->UUID
, UUID_NONE
) == 0) {
1403 name
= qemu_get_vm_name();
1405 name
= uuid_info
->UUID
;
1407 iscsi_name
= g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1408 name
? ":" : "", name
? name
: "");
1409 qapi_free_UuidInfo(uuid_info
);
1413 static void iscsi_nop_timed_event(void *opaque
)
1415 IscsiLun
*iscsilun
= opaque
;
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.");
1426 timer_mod(iscsilun
->nop_timer
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) + NOP_INTERVAL
);
1427 iscsi_set_events(iscsilun
);
1430 static void iscsi_readcapacity_sync(IscsiLun
*iscsilun
, Error
**errp
)
1432 struct scsi_task
*task
= NULL
;
1433 struct scsi_readcapacity10
*rc10
= NULL
;
1434 struct scsi_readcapacity16
*rc16
= NULL
;
1435 int retries
= ISCSI_CMD_RETRIES
;
1439 scsi_free_scsi_task(task
);
1443 switch (iscsilun
->type
) {
1445 task
= iscsi_readcapacity16_sync(iscsilun
->iscsi
, iscsilun
->lun
);
1446 if (task
!= NULL
&& task
->status
== SCSI_STATUS_GOOD
) {
1447 rc16
= scsi_datain_unmarshall(task
);
1449 error_setg(errp
, "iSCSI: Failed to unmarshall readcapacity16 data.");
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);
1459 if (task
!= NULL
&& task
->status
== SCSI_STATUS_CHECK_CONDITION
1460 && task
->sense
.key
== SCSI_SENSE_UNIT_ATTENTION
) {
1463 /* Fall through and try READ CAPACITY(10) instead. */
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
);
1469 error_setg(errp
, "iSCSI: Failed to unmarshall readcapacity10 data.");
1471 iscsilun
->block_size
= rc10
->block_size
;
1472 if (rc10
->lba
== 0) {
1473 /* blank disk loaded */
1474 iscsilun
->num_blocks
= 0;
1476 iscsilun
->num_blocks
= rc10
->lba
+ 1;
1484 } while (task
!= NULL
&& task
->status
== SCSI_STATUS_CHECK_CONDITION
1485 && task
->sense
.key
== SCSI_SENSE_UNIT_ATTENTION
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
);
1496 scsi_free_scsi_task(task
);
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 NULL
, NULL
, NULL
, NULL
, NULL
);
1544 iscsilun
->events
= 0;
1546 if (iscsilun
->nop_timer
) {
1547 timer_free(iscsilun
->nop_timer
);
1548 iscsilun
->nop_timer
= NULL
;
1550 if (iscsilun
->event_timer
) {
1551 timer_free(iscsilun
->event_timer
);
1552 iscsilun
->event_timer
= NULL
;
1556 static void iscsi_attach_aio_context(BlockDriverState
*bs
,
1557 AioContext
*new_context
)
1559 IscsiLun
*iscsilun
= bs
->opaque
;
1561 iscsilun
->aio_context
= new_context
;
1562 iscsi_set_events(iscsilun
);
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
);
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
);
1580 static void iscsi_modesense_sync(IscsiLun
*iscsilun
)
1582 struct scsi_task
*task
;
1583 struct scsi_mode_sense
*ms
= NULL
;
1584 iscsilun
->write_protected
= false;
1585 iscsilun
->dpofua
= false;
1587 task
= iscsi_modesense6_sync(iscsilun
->iscsi
, iscsilun
->lun
,
1588 1, SCSI_MODESENSE_PC_CURRENT
,
1591 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1592 iscsi_get_error(iscsilun
->iscsi
));
1596 if (task
->status
!= SCSI_STATUS_GOOD
) {
1597 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1600 ms
= scsi_datain_unmarshall(task
);
1602 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1603 iscsi_get_error(iscsilun
->iscsi
));
1606 iscsilun
->write_protected
= ms
->device_specific_parameter
& 0x80;
1607 iscsilun
->dpofua
= ms
->device_specific_parameter
& 0x10;
1611 scsi_free_scsi_task(task
);
1615 static void iscsi_parse_iscsi_option(const char *target
, QDict
*options
)
1619 const char *user
, *password
, *password_secret
, *initiator_name
,
1620 *header_digest
, *timeout
;
1622 list
= qemu_find_opts("iscsi");
1627 opts
= qemu_opts_find(list
, target
);
1629 opts
= QTAILQ_FIRST(&list
->head
);
1635 user
= qemu_opt_get(opts
, "user");
1637 qdict_set_default_str(options
, "user", user
);
1640 password
= qemu_opt_get(opts
, "password");
1642 qdict_set_default_str(options
, "password", password
);
1645 password_secret
= qemu_opt_get(opts
, "password-secret");
1646 if (password_secret
) {
1647 qdict_set_default_str(options
, "password-secret", password_secret
);
1650 initiator_name
= qemu_opt_get(opts
, "initiator-name");
1651 if (initiator_name
) {
1652 qdict_set_default_str(options
, "initiator-name", initiator_name
);
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
);
1664 timeout
= qemu_opt_get(opts
, "timeout");
1666 qdict_set_default_str(options
, "timeout", timeout
);
1671 * We support iscsi url's on the form
1672 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1674 static void iscsi_parse_filename(const char *filename
, QDict
*options
,
1677 struct iscsi_url
*iscsi_url
;
1678 const char *transport_name
;
1681 iscsi_url
= iscsi_parse_full_url(NULL
, filename
);
1682 if (iscsi_url
== NULL
) {
1683 error_setg(errp
, "Failed to parse URL : %s", filename
);
1687 #if LIBISCSI_API_VERSION >= (20160603)
1688 switch (iscsi_url
->transport
) {
1690 transport_name
= "tcp";
1692 case ISER_TRANSPORT
:
1693 transport_name
= "iser";
1696 error_setg(errp
, "Unknown transport type (%d)",
1697 iscsi_url
->transport
);
1701 transport_name
= "tcp";
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
);
1708 lun_str
= g_strdup_printf("%d", iscsi_url
->lun
);
1709 qdict_set_default_str(options
, "lun", lun_str
);
1712 /* User/password from -iscsi take precedence over those from the URL */
1713 iscsi_parse_iscsi_option(iscsi_url
->target
, options
);
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
);
1720 iscsi_destroy_url(iscsi_url
);
1723 static QemuOptsList runtime_opts
= {
1725 .head
= QTAILQ_HEAD_INITIALIZER(runtime_opts
.head
),
1728 .name
= "transport",
1729 .type
= QEMU_OPT_STRING
,
1733 .type
= QEMU_OPT_STRING
,
1737 .type
= QEMU_OPT_STRING
,
1741 .type
= QEMU_OPT_STRING
,
1745 .type
= QEMU_OPT_STRING
,
1748 .name
= "password-secret",
1749 .type
= QEMU_OPT_STRING
,
1753 .type
= QEMU_OPT_NUMBER
,
1756 .name
= "initiator-name",
1757 .type
= QEMU_OPT_STRING
,
1760 .name
= "header-digest",
1761 .type
= QEMU_OPT_STRING
,
1765 .type
= QEMU_OPT_NUMBER
,
1767 { /* end of list */ }
1771 static void iscsi_save_designator(IscsiLun
*lun
,
1772 struct scsi_inquiry_device_identification
*inq_di
)
1774 struct scsi_inquiry_device_designator
*desig
, *copy
= NULL
;
1776 for (desig
= inq_di
->designators
; desig
; desig
= desig
->next
) {
1777 if (desig
->association
||
1778 desig
->designator_type
> SCSI_DESIGNATOR_TYPE_NAA
) {
1781 /* NAA works better than T10 vendor ID based designator. */
1782 if (!copy
|| copy
->designator_type
< desig
->designator_type
) {
1787 lun
->dd
= g_new(struct scsi_inquiry_device_designator
, 1);
1789 lun
->dd
->next
= NULL
;
1790 lun
->dd
->designator
= g_malloc(copy
->designator_length
);
1791 memcpy(lun
->dd
->designator
, copy
->designator
, copy
->designator_length
);
1795 static int iscsi_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
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
;
1805 Error
*local_err
= NULL
;
1806 const char *transport_name
, *portal
, *target
;
1807 #if LIBISCSI_API_VERSION >= (20160603)
1808 enum iscsi_transport_type transport
;
1810 int i
, ret
= 0, timeout
= 0, lun
;
1812 opts
= qemu_opts_create(&runtime_opts
, NULL
, 0, &error_abort
);
1813 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
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);
1823 if (!transport_name
|| !portal
|| !target
) {
1824 error_setg(errp
, "Need all of transport, portal and target options");
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
;
1835 /* TCP is what older libiscsi versions always use */
1838 error_setg(errp
, "Unknown transport: %s", transport_name
);
1843 memset(iscsilun
, 0, sizeof(IscsiLun
));
1845 initiator_name
= get_initiator_name(opts
);
1847 iscsi
= iscsi_create_context(initiator_name
);
1848 if (iscsi
== NULL
) {
1849 error_setg(errp
, "iSCSI: Failed to create iSCSI context.");
1853 #if LIBISCSI_API_VERSION >= (20160603)
1854 if (iscsi_init_transport(iscsi
, transport
)) {
1855 error_setg(errp
, ("Error initializing transport."));
1860 if (iscsi_set_targetname(iscsi
, target
)) {
1861 error_setg(errp
, "iSCSI: Failed to set target name.");
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
);
1874 if (iscsi_set_session_type(iscsi
, ISCSI_SESSION_NORMAL
) != 0) {
1875 error_setg(errp
, "iSCSI: Failed to set session type to normal.");
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
);
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
);
1894 warn_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
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
));
1905 iscsilun
->iscsi
= iscsi
;
1906 iscsilun
->aio_context
= bdrv_get_aio_context(bs
);
1907 iscsilun
->lun
= lun
;
1908 iscsilun
->has_write_same
= true;
1910 task
= iscsi_do_inquiry(iscsilun
->iscsi
, iscsilun
->lun
, 0, 0,
1911 (void **) &inq
, errp
);
1916 iscsilun
->type
= inq
->periperal_device_type
;
1917 scsi_free_scsi_task(task
);
1920 iscsi_modesense_sync(iscsilun
);
1921 if (iscsilun
->dpofua
) {
1922 bs
->supported_write_flags
= BDRV_REQ_FUA
;
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 ret
= bdrv_apply_auto_read_only(bs
, "LUN is write protected", errp
);
1932 flags
&= ~BDRV_O_RDWR
;
1935 iscsi_readcapacity_sync(iscsilun
, &local_err
);
1936 if (local_err
!= NULL
) {
1937 error_propagate(errp
, local_err
);
1941 bs
->total_sectors
= sector_lun2qemu(iscsilun
->num_blocks
, iscsilun
);
1943 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1944 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1945 * will try to read from the device to guess the image format.
1947 if (iscsilun
->type
!= TYPE_DISK
&& iscsilun
->type
!= TYPE_ROM
) {
1951 task
= iscsi_do_inquiry(iscsilun
->iscsi
, iscsilun
->lun
, 1,
1952 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES
,
1953 (void **) &inq_vpd
, errp
);
1958 for (i
= 0; i
< inq_vpd
->num_pages
; i
++) {
1959 struct scsi_task
*inq_task
;
1960 struct scsi_inquiry_logical_block_provisioning
*inq_lbp
;
1961 struct scsi_inquiry_block_limits
*inq_bl
;
1962 struct scsi_inquiry_device_identification
*inq_di
;
1963 switch (inq_vpd
->pages
[i
]) {
1964 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING
:
1965 inq_task
= iscsi_do_inquiry(iscsilun
->iscsi
, iscsilun
->lun
, 1,
1966 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING
,
1967 (void **) &inq_lbp
, errp
);
1968 if (inq_task
== NULL
) {
1972 memcpy(&iscsilun
->lbp
, inq_lbp
,
1973 sizeof(struct scsi_inquiry_logical_block_provisioning
));
1974 scsi_free_scsi_task(inq_task
);
1976 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS
:
1977 inq_task
= iscsi_do_inquiry(iscsilun
->iscsi
, iscsilun
->lun
, 1,
1978 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS
,
1979 (void **) &inq_bl
, errp
);
1980 if (inq_task
== NULL
) {
1984 memcpy(&iscsilun
->bl
, inq_bl
,
1985 sizeof(struct scsi_inquiry_block_limits
));
1986 scsi_free_scsi_task(inq_task
);
1988 case SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION
:
1989 inq_task
= iscsi_do_inquiry(iscsilun
->iscsi
, iscsilun
->lun
, 1,
1990 SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION
,
1991 (void **) &inq_di
, errp
);
1992 if (inq_task
== NULL
) {
1996 iscsi_save_designator(iscsilun
, inq_di
);
1997 scsi_free_scsi_task(inq_task
);
2003 scsi_free_scsi_task(task
);
2006 qemu_mutex_init(&iscsilun
->mutex
);
2007 iscsi_attach_aio_context(bs
, iscsilun
->aio_context
);
2009 /* Guess the internal cluster (page) size of the iscsi target by the means
2010 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
2011 * reasonable size */
2012 if (iscsilun
->bl
.opt_unmap_gran
* iscsilun
->block_size
>= 4 * 1024 &&
2013 iscsilun
->bl
.opt_unmap_gran
* iscsilun
->block_size
<= 16 * 1024 * 1024) {
2014 iscsilun
->cluster_size
= iscsilun
->bl
.opt_unmap_gran
*
2015 iscsilun
->block_size
;
2016 if (iscsilun
->lbprz
) {
2017 ret
= iscsi_allocmap_init(iscsilun
, flags
);
2021 if (iscsilun
->lbprz
&& iscsilun
->lbp
.lbpws
) {
2022 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
;
2026 qemu_opts_del(opts
);
2027 g_free(initiator_name
);
2029 scsi_free_scsi_task(task
);
2033 if (iscsi
!= NULL
) {
2034 if (iscsi_is_logged_in(iscsi
)) {
2035 iscsi_logout_sync(iscsi
);
2037 iscsi_destroy_context(iscsi
);
2039 memset(iscsilun
, 0, sizeof(IscsiLun
));
2045 static void iscsi_close(BlockDriverState
*bs
)
2047 IscsiLun
*iscsilun
= bs
->opaque
;
2048 struct iscsi_context
*iscsi
= iscsilun
->iscsi
;
2050 iscsi_detach_aio_context(bs
);
2051 if (iscsi_is_logged_in(iscsi
)) {
2052 iscsi_logout_sync(iscsi
);
2054 iscsi_destroy_context(iscsi
);
2056 g_free(iscsilun
->dd
->designator
);
2057 g_free(iscsilun
->dd
);
2059 g_free(iscsilun
->zeroblock
);
2060 iscsi_allocmap_free(iscsilun
);
2061 qemu_mutex_destroy(&iscsilun
->mutex
);
2062 memset(iscsilun
, 0, sizeof(IscsiLun
));
2065 static void iscsi_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
2067 /* We don't actually refresh here, but just return data queried in
2068 * iscsi_open(): iscsi targets don't change their limits. */
2070 IscsiLun
*iscsilun
= bs
->opaque
;
2071 uint64_t max_xfer_len
= iscsilun
->use_16_for_rw
? 0xffffffff : 0xffff;
2072 unsigned int block_size
= MAX(BDRV_SECTOR_SIZE
, iscsilun
->block_size
);
2074 assert(iscsilun
->block_size
>= BDRV_SECTOR_SIZE
|| bdrv_is_sg(bs
));
2076 bs
->bl
.request_alignment
= block_size
;
2078 if (iscsilun
->bl
.max_xfer_len
) {
2079 max_xfer_len
= MIN(max_xfer_len
, iscsilun
->bl
.max_xfer_len
);
2082 if (max_xfer_len
* block_size
< INT_MAX
) {
2083 bs
->bl
.max_transfer
= max_xfer_len
* iscsilun
->block_size
;
2086 if (iscsilun
->lbp
.lbpu
) {
2087 bs
->bl
.max_pdiscard
=
2088 MIN_NON_ZERO(iscsilun
->bl
.max_unmap
* iscsilun
->block_size
,
2089 (uint64_t)UINT32_MAX
* iscsilun
->block_size
);
2090 bs
->bl
.pdiscard_alignment
=
2091 iscsilun
->bl
.opt_unmap_gran
* iscsilun
->block_size
;
2093 bs
->bl
.pdiscard_alignment
= iscsilun
->block_size
;
2096 bs
->bl
.max_pwrite_zeroes
=
2097 MIN_NON_ZERO(iscsilun
->bl
.max_ws_len
* iscsilun
->block_size
,
2098 max_xfer_len
* iscsilun
->block_size
);
2100 if (iscsilun
->lbp
.lbpws
) {
2101 bs
->bl
.pwrite_zeroes_alignment
=
2102 iscsilun
->bl
.opt_unmap_gran
* iscsilun
->block_size
;
2104 bs
->bl
.pwrite_zeroes_alignment
= iscsilun
->block_size
;
2106 if (iscsilun
->bl
.opt_xfer_len
&&
2107 iscsilun
->bl
.opt_xfer_len
< INT_MAX
/ block_size
) {
2108 bs
->bl
.opt_transfer
= pow2floor(iscsilun
->bl
.opt_xfer_len
*
2109 iscsilun
->block_size
);
2113 /* Note that this will not re-establish a connection with an iSCSI target - it
2114 * is effectively a NOP. */
2115 static int iscsi_reopen_prepare(BDRVReopenState
*state
,
2116 BlockReopenQueue
*queue
, Error
**errp
)
2118 IscsiLun
*iscsilun
= state
->bs
->opaque
;
2120 if (state
->flags
& BDRV_O_RDWR
&& iscsilun
->write_protected
) {
2121 error_setg(errp
, "Cannot open a write protected LUN as read-write");
2127 static void iscsi_reopen_commit(BDRVReopenState
*reopen_state
)
2129 IscsiLun
*iscsilun
= reopen_state
->bs
->opaque
;
2131 /* the cache.direct status might have changed */
2132 if (iscsilun
->allocmap
!= NULL
) {
2133 iscsi_allocmap_init(iscsilun
, reopen_state
->flags
);
2137 static int coroutine_fn
iscsi_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2138 bool exact
, PreallocMode prealloc
,
2139 BdrvRequestFlags flags
, Error
**errp
)
2141 IscsiLun
*iscsilun
= bs
->opaque
;
2143 Error
*local_err
= NULL
;
2145 if (prealloc
!= PREALLOC_MODE_OFF
) {
2146 error_setg(errp
, "Unsupported preallocation mode '%s'",
2147 PreallocMode_str(prealloc
));
2151 if (iscsilun
->type
!= TYPE_DISK
) {
2152 error_setg(errp
, "Cannot resize non-disk iSCSI devices");
2156 iscsi_readcapacity_sync(iscsilun
, &local_err
);
2157 if (local_err
!= NULL
) {
2158 error_propagate(errp
, local_err
);
2162 cur_length
= iscsi_co_getlength(bs
);
2163 if (offset
!= cur_length
&& exact
) {
2164 error_setg(errp
, "Cannot resize iSCSI devices");
2166 } else if (offset
> cur_length
) {
2167 error_setg(errp
, "Cannot grow iSCSI devices");
2171 if (iscsilun
->allocmap
!= NULL
) {
2172 iscsi_allocmap_init(iscsilun
, bs
->open_flags
);
2178 static int coroutine_fn
2179 iscsi_co_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2181 IscsiLun
*iscsilun
= bs
->opaque
;
2182 bdi
->cluster_size
= iscsilun
->cluster_size
;
2186 static void coroutine_fn
iscsi_co_invalidate_cache(BlockDriverState
*bs
,
2189 IscsiLun
*iscsilun
= bs
->opaque
;
2190 iscsi_allocmap_invalidate(iscsilun
);
2193 static int coroutine_fn GRAPH_RDLOCK
2194 iscsi_co_copy_range_from(BlockDriverState
*bs
,
2195 BdrvChild
*src
, int64_t src_offset
,
2196 BdrvChild
*dst
, int64_t dst_offset
,
2197 int64_t bytes
, BdrvRequestFlags read_flags
,
2198 BdrvRequestFlags write_flags
)
2200 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
2201 read_flags
, write_flags
);
2204 static struct scsi_task
*iscsi_xcopy_task(int param_len
)
2206 struct scsi_task
*task
;
2208 task
= g_new0(struct scsi_task
, 1);
2210 task
->cdb
[0] = EXTENDED_COPY
;
2211 task
->cdb
[10] = (param_len
>> 24) & 0xFF;
2212 task
->cdb
[11] = (param_len
>> 16) & 0xFF;
2213 task
->cdb
[12] = (param_len
>> 8) & 0xFF;
2214 task
->cdb
[13] = param_len
& 0xFF;
2215 task
->cdb_size
= 16;
2216 task
->xfer_dir
= SCSI_XFER_WRITE
;
2217 task
->expxferlen
= param_len
;
2222 static void iscsi_populate_target_desc(unsigned char *desc
, IscsiLun
*lun
)
2224 struct scsi_inquiry_device_designator
*dd
= lun
->dd
;
2226 memset(desc
, 0, 32);
2227 desc
[0] = 0xE4; /* IDENT_DESCR_TGT_DESCR */
2228 desc
[4] = dd
->code_set
;
2229 desc
[5] = (dd
->designator_type
& 0xF)
2230 | ((dd
->association
& 3) << 4);
2231 desc
[7] = dd
->designator_length
;
2232 memcpy(desc
+ 8, dd
->designator
, MIN(dd
->designator_length
, 20));
2235 desc
[29] = (lun
->block_size
>> 16) & 0xFF;
2236 desc
[30] = (lun
->block_size
>> 8) & 0xFF;
2237 desc
[31] = lun
->block_size
& 0xFF;
2240 static void iscsi_xcopy_desc_hdr(uint8_t *hdr
, int dc
, int cat
, int src_index
,
2243 hdr
[0] = 0x02; /* BLK_TO_BLK_SEG_DESCR */
2244 hdr
[1] = ((dc
<< 1) | cat
) & 0xFF;
2245 hdr
[2] = (XCOPY_BLK2BLK_SEG_DESC_SIZE
>> 8) & 0xFF;
2246 /* don't account for the first 4 bytes in descriptor header*/
2247 hdr
[3] = (XCOPY_BLK2BLK_SEG_DESC_SIZE
- 4 /* SEG_DESC_SRC_INDEX_OFFSET */) & 0xFF;
2248 hdr
[4] = (src_index
>> 8) & 0xFF;
2249 hdr
[5] = src_index
& 0xFF;
2250 hdr
[6] = (dst_index
>> 8) & 0xFF;
2251 hdr
[7] = dst_index
& 0xFF;
2254 static void iscsi_xcopy_populate_desc(uint8_t *desc
, int dc
, int cat
,
2255 int src_index
, int dst_index
, int num_blks
,
2256 uint64_t src_lba
, uint64_t dst_lba
)
2258 iscsi_xcopy_desc_hdr(desc
, dc
, cat
, src_index
, dst_index
);
2260 /* The caller should verify the request size */
2261 assert(num_blks
< 65536);
2262 desc
[10] = (num_blks
>> 8) & 0xFF;
2263 desc
[11] = num_blks
& 0xFF;
2264 desc
[12] = (src_lba
>> 56) & 0xFF;
2265 desc
[13] = (src_lba
>> 48) & 0xFF;
2266 desc
[14] = (src_lba
>> 40) & 0xFF;
2267 desc
[15] = (src_lba
>> 32) & 0xFF;
2268 desc
[16] = (src_lba
>> 24) & 0xFF;
2269 desc
[17] = (src_lba
>> 16) & 0xFF;
2270 desc
[18] = (src_lba
>> 8) & 0xFF;
2271 desc
[19] = src_lba
& 0xFF;
2272 desc
[20] = (dst_lba
>> 56) & 0xFF;
2273 desc
[21] = (dst_lba
>> 48) & 0xFF;
2274 desc
[22] = (dst_lba
>> 40) & 0xFF;
2275 desc
[23] = (dst_lba
>> 32) & 0xFF;
2276 desc
[24] = (dst_lba
>> 24) & 0xFF;
2277 desc
[25] = (dst_lba
>> 16) & 0xFF;
2278 desc
[26] = (dst_lba
>> 8) & 0xFF;
2279 desc
[27] = dst_lba
& 0xFF;
2282 static void iscsi_xcopy_populate_header(unsigned char *buf
, int list_id
, int str
,
2283 int list_id_usage
, int prio
,
2285 int seg_desc_len
, int inline_data_len
)
2288 buf
[1] = ((str
& 1) << 5) | ((list_id_usage
& 3) << 3) | (prio
& 7);
2289 buf
[2] = (tgt_desc_len
>> 8) & 0xFF;
2290 buf
[3] = tgt_desc_len
& 0xFF;
2291 buf
[8] = (seg_desc_len
>> 24) & 0xFF;
2292 buf
[9] = (seg_desc_len
>> 16) & 0xFF;
2293 buf
[10] = (seg_desc_len
>> 8) & 0xFF;
2294 buf
[11] = seg_desc_len
& 0xFF;
2295 buf
[12] = (inline_data_len
>> 24) & 0xFF;
2296 buf
[13] = (inline_data_len
>> 16) & 0xFF;
2297 buf
[14] = (inline_data_len
>> 8) & 0xFF;
2298 buf
[15] = inline_data_len
& 0xFF;
2301 static void iscsi_xcopy_data(struct iscsi_data
*data
,
2302 IscsiLun
*src
, int64_t src_lba
,
2303 IscsiLun
*dst
, int64_t dst_lba
,
2304 uint16_t num_blocks
)
2307 const int src_offset
= XCOPY_DESC_OFFSET
;
2308 const int dst_offset
= XCOPY_DESC_OFFSET
+ IDENT_DESCR_TGT_DESCR_SIZE
;
2309 const int seg_offset
= dst_offset
+ IDENT_DESCR_TGT_DESCR_SIZE
;
2311 data
->size
= XCOPY_DESC_OFFSET
+
2312 IDENT_DESCR_TGT_DESCR_SIZE
* 2 +
2313 XCOPY_BLK2BLK_SEG_DESC_SIZE
;
2314 data
->data
= g_malloc0(data
->size
);
2317 /* Initialise the parameter list header */
2318 iscsi_xcopy_populate_header(buf
, 1, 0, 2 /* LIST_ID_USAGE_DISCARD */,
2319 0, 2 * IDENT_DESCR_TGT_DESCR_SIZE
,
2320 XCOPY_BLK2BLK_SEG_DESC_SIZE
,
2323 /* Initialise CSCD list with one src + one dst descriptor */
2324 iscsi_populate_target_desc(&buf
[src_offset
], src
);
2325 iscsi_populate_target_desc(&buf
[dst_offset
], dst
);
2327 /* Initialise one segment descriptor */
2328 iscsi_xcopy_populate_desc(&buf
[seg_offset
], 0, 0, 0, 1, num_blocks
,
2332 static int coroutine_fn GRAPH_RDLOCK
2333 iscsi_co_copy_range_to(BlockDriverState
*bs
,
2334 BdrvChild
*src
, int64_t src_offset
,
2335 BdrvChild
*dst
, int64_t dst_offset
,
2336 int64_t bytes
, BdrvRequestFlags read_flags
,
2337 BdrvRequestFlags write_flags
)
2339 IscsiLun
*dst_lun
= dst
->bs
->opaque
;
2341 struct IscsiTask iscsi_task
;
2342 struct iscsi_data data
;
2346 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= iscsi_co_copy_range_to
) {
2349 src_lun
= src
->bs
->opaque
;
2351 if (!src_lun
->dd
|| !dst_lun
->dd
) {
2354 if (!is_byte_request_lun_aligned(dst_offset
, bytes
, dst_lun
)) {
2357 if (!is_byte_request_lun_aligned(src_offset
, bytes
, src_lun
)) {
2360 if (dst_lun
->block_size
!= src_lun
->block_size
||
2361 !dst_lun
->block_size
) {
2365 block_size
= dst_lun
->block_size
;
2366 if (bytes
/ block_size
> 65535) {
2370 iscsi_xcopy_data(&data
,
2371 src_lun
, src_offset
/ block_size
,
2372 dst_lun
, dst_offset
/ block_size
,
2373 bytes
/ block_size
);
2375 iscsi_co_init_iscsitask(dst_lun
, &iscsi_task
);
2377 qemu_mutex_lock(&dst_lun
->mutex
);
2378 iscsi_task
.task
= iscsi_xcopy_task(data
.size
);
2380 if (iscsi_scsi_command_async(dst_lun
->iscsi
, dst_lun
->lun
,
2381 iscsi_task
.task
, iscsi_co_generic_cb
,
2383 &iscsi_task
) != 0) {
2388 iscsi_co_wait_for_task(&iscsi_task
, dst_lun
);
2390 if (iscsi_task
.do_retry
) {
2391 iscsi_task
.complete
= 0;
2395 if (iscsi_task
.status
!= SCSI_STATUS_GOOD
) {
2396 r
= iscsi_task
.err_code
;
2402 trace_iscsi_xcopy(src_lun
, src_offset
, dst_lun
, dst_offset
, bytes
, r
);
2403 g_free(iscsi_task
.task
);
2404 qemu_mutex_unlock(&dst_lun
->mutex
);
2405 g_free(iscsi_task
.err_str
);
2410 static const char *const iscsi_strong_runtime_opts
[] = {
2424 static BlockDriver bdrv_iscsi
= {
2425 .format_name
= "iscsi",
2426 .protocol_name
= "iscsi",
2428 .instance_size
= sizeof(IscsiLun
),
2429 .bdrv_parse_filename
= iscsi_parse_filename
,
2430 .bdrv_file_open
= iscsi_open
,
2431 .bdrv_close
= iscsi_close
,
2432 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
2433 .create_opts
= &bdrv_create_opts_simple
,
2434 .bdrv_reopen_prepare
= iscsi_reopen_prepare
,
2435 .bdrv_reopen_commit
= iscsi_reopen_commit
,
2436 .bdrv_co_invalidate_cache
= iscsi_co_invalidate_cache
,
2438 .bdrv_co_getlength
= iscsi_co_getlength
,
2439 .bdrv_co_get_info
= iscsi_co_get_info
,
2440 .bdrv_co_truncate
= iscsi_co_truncate
,
2441 .bdrv_refresh_limits
= iscsi_refresh_limits
,
2443 .bdrv_co_block_status
= iscsi_co_block_status
,
2444 .bdrv_co_pdiscard
= iscsi_co_pdiscard
,
2445 .bdrv_co_copy_range_from
= iscsi_co_copy_range_from
,
2446 .bdrv_co_copy_range_to
= iscsi_co_copy_range_to
,
2447 .bdrv_co_pwrite_zeroes
= iscsi_co_pwrite_zeroes
,
2448 .bdrv_co_readv
= iscsi_co_readv
,
2449 .bdrv_co_writev
= iscsi_co_writev
,
2450 .bdrv_co_flush_to_disk
= iscsi_co_flush
,
2453 .bdrv_aio_ioctl
= iscsi_aio_ioctl
,
2456 .bdrv_detach_aio_context
= iscsi_detach_aio_context
,
2457 .bdrv_attach_aio_context
= iscsi_attach_aio_context
,
2459 .strong_runtime_opts
= iscsi_strong_runtime_opts
,
2462 #if LIBISCSI_API_VERSION >= (20160603)
2463 static BlockDriver bdrv_iser
= {
2464 .format_name
= "iser",
2465 .protocol_name
= "iser",
2467 .instance_size
= sizeof(IscsiLun
),
2468 .bdrv_parse_filename
= iscsi_parse_filename
,
2469 .bdrv_file_open
= iscsi_open
,
2470 .bdrv_close
= iscsi_close
,
2471 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
2472 .create_opts
= &bdrv_create_opts_simple
,
2473 .bdrv_reopen_prepare
= iscsi_reopen_prepare
,
2474 .bdrv_reopen_commit
= iscsi_reopen_commit
,
2475 .bdrv_co_invalidate_cache
= iscsi_co_invalidate_cache
,
2477 .bdrv_co_getlength
= iscsi_co_getlength
,
2478 .bdrv_co_get_info
= iscsi_co_get_info
,
2479 .bdrv_co_truncate
= iscsi_co_truncate
,
2480 .bdrv_refresh_limits
= iscsi_refresh_limits
,
2482 .bdrv_co_block_status
= iscsi_co_block_status
,
2483 .bdrv_co_pdiscard
= iscsi_co_pdiscard
,
2484 .bdrv_co_copy_range_from
= iscsi_co_copy_range_from
,
2485 .bdrv_co_copy_range_to
= iscsi_co_copy_range_to
,
2486 .bdrv_co_pwrite_zeroes
= iscsi_co_pwrite_zeroes
,
2487 .bdrv_co_readv
= iscsi_co_readv
,
2488 .bdrv_co_writev
= iscsi_co_writev
,
2489 .bdrv_co_flush_to_disk
= iscsi_co_flush
,
2492 .bdrv_aio_ioctl
= iscsi_aio_ioctl
,
2495 .bdrv_detach_aio_context
= iscsi_detach_aio_context
,
2496 .bdrv_attach_aio_context
= iscsi_attach_aio_context
,
2498 .strong_runtime_opts
= iscsi_strong_runtime_opts
,
2502 static void iscsi_block_init(void)
2504 bdrv_register(&bdrv_iscsi
);
2505 #if LIBISCSI_API_VERSION >= (20160603)
2506 bdrv_register(&bdrv_iser
);
2510 block_init(iscsi_block_init
);