hw/virtio/balloon: Replace TARGET_PAGE_SIZE with BALLOON_PAGE_SIZE
[qemu/cris-port.git] / block / iscsi.c
blob302baf84c1c71131ebb784910ed50570f78c2432
1 /*
2 * QEMU Block driver for iSCSI images
4 * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
5 * Copyright (c) 2012-2015 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
23 * THE SOFTWARE.
26 #include "qemu/osdep.h"
28 #include <poll.h>
29 #include <math.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"
38 #include "qemu/iov.h"
39 #include "sysemu/sysemu.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>
47 #ifdef __linux__
48 #include <scsi/sg.h>
49 #include <block/scsi.h>
50 #endif
52 typedef struct IscsiLun {
53 struct iscsi_context *iscsi;
54 AioContext *aio_context;
55 int lun;
56 enum scsi_inquiry_peripheral_device_type type;
57 int block_size;
58 uint64_t num_blocks;
59 int events;
60 QEMUTimer *nop_timer;
61 QEMUTimer *event_timer;
62 struct scsi_inquiry_logical_block_provisioning lbp;
63 struct scsi_inquiry_block_limits bl;
64 unsigned char *zeroblock;
65 unsigned long *allocationmap;
66 int cluster_sectors;
67 bool use_16_for_rw;
68 bool write_protected;
69 bool lbpme;
70 bool lbprz;
71 bool dpofua;
72 bool has_write_same;
73 bool request_timed_out;
74 } IscsiLun;
76 typedef struct IscsiTask {
77 int status;
78 int complete;
79 int retries;
80 int do_retry;
81 struct scsi_task *task;
82 Coroutine *co;
83 QEMUBH *bh;
84 IscsiLun *iscsilun;
85 QEMUTimer retry_timer;
86 int err_code;
87 } IscsiTask;
89 typedef struct IscsiAIOCB {
90 BlockAIOCB common;
91 QEMUIOVector *qiov;
92 QEMUBH *bh;
93 IscsiLun *iscsilun;
94 struct scsi_task *task;
95 uint8_t *buf;
96 int status;
97 int64_t sector_num;
98 int nb_sectors;
99 int ret;
100 #ifdef __linux__
101 sg_io_hdr_t *ioh;
102 #endif
103 } IscsiAIOCB;
105 /* libiscsi uses time_t so its enough to process events every second */
106 #define EVENT_INTERVAL 1000
107 #define NOP_INTERVAL 5000
108 #define MAX_NOP_FAILURES 3
109 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times)
110 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768};
112 /* this threshold is a trade-off knob to choose between
113 * the potential additional overhead of an extra GET_LBA_STATUS request
114 * vs. unnecessarily reading a lot of zero sectors over the wire.
115 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
116 * sectors we check the allocation status of the area covered by the
117 * request first if the allocationmap indicates that the area might be
118 * unallocated. */
119 #define ISCSI_CHECKALLOC_THRES 64
121 static void
122 iscsi_bh_cb(void *p)
124 IscsiAIOCB *acb = p;
126 qemu_bh_delete(acb->bh);
128 g_free(acb->buf);
129 acb->buf = NULL;
131 acb->common.cb(acb->common.opaque, acb->status);
133 if (acb->task != NULL) {
134 scsi_free_scsi_task(acb->task);
135 acb->task = NULL;
138 qemu_aio_unref(acb);
141 static void
142 iscsi_schedule_bh(IscsiAIOCB *acb)
144 if (acb->bh) {
145 return;
147 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb);
148 qemu_bh_schedule(acb->bh);
151 static void iscsi_co_generic_bh_cb(void *opaque)
153 struct IscsiTask *iTask = opaque;
154 iTask->complete = 1;
155 qemu_bh_delete(iTask->bh);
156 qemu_coroutine_enter(iTask->co, NULL);
159 static void iscsi_retry_timer_expired(void *opaque)
161 struct IscsiTask *iTask = opaque;
162 iTask->complete = 1;
163 if (iTask->co) {
164 qemu_coroutine_enter(iTask->co, NULL);
168 static inline unsigned exp_random(double mean)
170 return -mean * log((double)rand() / RAND_MAX);
173 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in
174 * libiscsi 1.10.0, together with other constants we need. Use it as
175 * a hint that we have to define them ourselves if needed, to keep the
176 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for
177 * the test because SCSI_STATUS_* is an enum.
179 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes
180 * an enum, check against the LIBISCSI_API_VERSION macro, which was
181 * introduced in 1.11.0. If it is present, there is no need to define
182 * anything.
184 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \
185 !defined(LIBISCSI_API_VERSION)
186 #define SCSI_STATUS_TASK_SET_FULL 0x28
187 #define SCSI_STATUS_TIMEOUT 0x0f000002
188 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600
189 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00
190 #endif
192 static int iscsi_translate_sense(struct scsi_sense *sense)
194 int ret;
196 switch (sense->key) {
197 case SCSI_SENSE_NOT_READY:
198 return -EBUSY;
199 case SCSI_SENSE_DATA_PROTECTION:
200 return -EACCES;
201 case SCSI_SENSE_COMMAND_ABORTED:
202 return -ECANCELED;
203 case SCSI_SENSE_ILLEGAL_REQUEST:
204 /* Parse ASCQ */
205 break;
206 default:
207 return -EIO;
209 switch (sense->ascq) {
210 case SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR:
211 case SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE:
212 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB:
213 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST:
214 ret = -EINVAL;
215 break;
216 case SCSI_SENSE_ASCQ_LBA_OUT_OF_RANGE:
217 ret = -ENOSPC;
218 break;
219 case SCSI_SENSE_ASCQ_LOGICAL_UNIT_NOT_SUPPORTED:
220 ret = -ENOTSUP;
221 break;
222 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT:
223 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_CLOSED:
224 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_OPEN:
225 ret = -ENOMEDIUM;
226 break;
227 case SCSI_SENSE_ASCQ_WRITE_PROTECTED:
228 ret = -EACCES;
229 break;
230 default:
231 ret = -EIO;
232 break;
234 return ret;
237 static void
238 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
239 void *command_data, void *opaque)
241 struct IscsiTask *iTask = opaque;
242 struct scsi_task *task = command_data;
244 iTask->status = status;
245 iTask->do_retry = 0;
246 iTask->task = task;
248 if (status != SCSI_STATUS_GOOD) {
249 if (iTask->retries++ < ISCSI_CMD_RETRIES) {
250 if (status == SCSI_STATUS_CHECK_CONDITION
251 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
252 error_report("iSCSI CheckCondition: %s",
253 iscsi_get_error(iscsi));
254 iTask->do_retry = 1;
255 goto out;
257 if (status == SCSI_STATUS_BUSY ||
258 status == SCSI_STATUS_TIMEOUT ||
259 status == SCSI_STATUS_TASK_SET_FULL) {
260 unsigned retry_time =
261 exp_random(iscsi_retry_times[iTask->retries - 1]);
262 if (status == SCSI_STATUS_TIMEOUT) {
263 /* make sure the request is rescheduled AFTER the
264 * reconnect is initiated */
265 retry_time = EVENT_INTERVAL * 2;
266 iTask->iscsilun->request_timed_out = true;
268 error_report("iSCSI Busy/TaskSetFull/TimeOut"
269 " (retry #%u in %u ms): %s",
270 iTask->retries, retry_time,
271 iscsi_get_error(iscsi));
272 aio_timer_init(iTask->iscsilun->aio_context,
273 &iTask->retry_timer, QEMU_CLOCK_REALTIME,
274 SCALE_MS, iscsi_retry_timer_expired, iTask);
275 timer_mod(&iTask->retry_timer,
276 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
277 iTask->do_retry = 1;
278 return;
281 iTask->err_code = iscsi_translate_sense(&task->sense);
282 error_report("iSCSI Failure: %s", iscsi_get_error(iscsi));
285 out:
286 if (iTask->co) {
287 iTask->bh = aio_bh_new(iTask->iscsilun->aio_context,
288 iscsi_co_generic_bh_cb, iTask);
289 qemu_bh_schedule(iTask->bh);
290 } else {
291 iTask->complete = 1;
295 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
297 *iTask = (struct IscsiTask) {
298 .co = qemu_coroutine_self(),
299 .iscsilun = iscsilun,
303 static void
304 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
305 void *private_data)
307 IscsiAIOCB *acb = private_data;
309 acb->status = -ECANCELED;
310 iscsi_schedule_bh(acb);
313 static void
314 iscsi_aio_cancel(BlockAIOCB *blockacb)
316 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
317 IscsiLun *iscsilun = acb->iscsilun;
319 if (acb->status != -EINPROGRESS) {
320 return;
323 /* send a task mgmt call to the target to cancel the task on the target */
324 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
325 iscsi_abort_task_cb, acb);
329 static const AIOCBInfo iscsi_aiocb_info = {
330 .aiocb_size = sizeof(IscsiAIOCB),
331 .cancel_async = iscsi_aio_cancel,
335 static void iscsi_process_read(void *arg);
336 static void iscsi_process_write(void *arg);
338 static void
339 iscsi_set_events(IscsiLun *iscsilun)
341 struct iscsi_context *iscsi = iscsilun->iscsi;
342 int ev = iscsi_which_events(iscsi);
344 if (ev != iscsilun->events) {
345 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi),
346 false,
347 (ev & POLLIN) ? iscsi_process_read : NULL,
348 (ev & POLLOUT) ? iscsi_process_write : NULL,
349 iscsilun);
350 iscsilun->events = ev;
354 static void iscsi_timed_check_events(void *opaque)
356 IscsiLun *iscsilun = opaque;
358 /* check for timed out requests */
359 iscsi_service(iscsilun->iscsi, 0);
361 if (iscsilun->request_timed_out) {
362 iscsilun->request_timed_out = false;
363 iscsi_reconnect(iscsilun->iscsi);
366 /* newer versions of libiscsi may return zero events. Ensure we are able
367 * to return to service once this situation changes. */
368 iscsi_set_events(iscsilun);
370 timer_mod(iscsilun->event_timer,
371 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
374 static void
375 iscsi_process_read(void *arg)
377 IscsiLun *iscsilun = arg;
378 struct iscsi_context *iscsi = iscsilun->iscsi;
380 iscsi_service(iscsi, POLLIN);
381 iscsi_set_events(iscsilun);
384 static void
385 iscsi_process_write(void *arg)
387 IscsiLun *iscsilun = arg;
388 struct iscsi_context *iscsi = iscsilun->iscsi;
390 iscsi_service(iscsi, POLLOUT);
391 iscsi_set_events(iscsilun);
394 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
396 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
399 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
401 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
404 static bool is_request_lun_aligned(int64_t sector_num, int nb_sectors,
405 IscsiLun *iscsilun)
407 if ((sector_num * BDRV_SECTOR_SIZE) % iscsilun->block_size ||
408 (nb_sectors * BDRV_SECTOR_SIZE) % iscsilun->block_size) {
409 error_report("iSCSI misaligned request: "
410 "iscsilun->block_size %u, sector_num %" PRIi64
411 ", nb_sectors %d",
412 iscsilun->block_size, sector_num, nb_sectors);
413 return 0;
415 return 1;
418 static unsigned long *iscsi_allocationmap_init(IscsiLun *iscsilun)
420 return bitmap_try_new(DIV_ROUND_UP(sector_lun2qemu(iscsilun->num_blocks,
421 iscsilun),
422 iscsilun->cluster_sectors));
425 static void iscsi_allocationmap_set(IscsiLun *iscsilun, int64_t sector_num,
426 int nb_sectors)
428 if (iscsilun->allocationmap == NULL) {
429 return;
431 bitmap_set(iscsilun->allocationmap,
432 sector_num / iscsilun->cluster_sectors,
433 DIV_ROUND_UP(nb_sectors, iscsilun->cluster_sectors));
436 static void iscsi_allocationmap_clear(IscsiLun *iscsilun, int64_t sector_num,
437 int nb_sectors)
439 int64_t cluster_num, nb_clusters;
440 if (iscsilun->allocationmap == NULL) {
441 return;
443 cluster_num = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors);
444 nb_clusters = (sector_num + nb_sectors) / iscsilun->cluster_sectors
445 - cluster_num;
446 if (nb_clusters > 0) {
447 bitmap_clear(iscsilun->allocationmap, cluster_num, nb_clusters);
451 static int coroutine_fn
452 iscsi_co_writev_flags(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
453 QEMUIOVector *iov, int flags)
455 IscsiLun *iscsilun = bs->opaque;
456 struct IscsiTask iTask;
457 uint64_t lba;
458 uint32_t num_sectors;
459 bool fua;
461 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
462 return -EINVAL;
465 if (bs->bl.max_transfer_length && nb_sectors > bs->bl.max_transfer_length) {
466 error_report("iSCSI Error: Write of %d sectors exceeds max_xfer_len "
467 "of %d sectors", nb_sectors, bs->bl.max_transfer_length);
468 return -EINVAL;
471 lba = sector_qemu2lun(sector_num, iscsilun);
472 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
473 iscsi_co_init_iscsitask(iscsilun, &iTask);
474 retry:
475 fua = iscsilun->dpofua && (flags & BDRV_REQ_FUA);
476 if (iscsilun->use_16_for_rw) {
477 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
478 NULL, num_sectors * iscsilun->block_size,
479 iscsilun->block_size, 0, 0, fua, 0, 0,
480 iscsi_co_generic_cb, &iTask);
481 } else {
482 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
483 NULL, num_sectors * iscsilun->block_size,
484 iscsilun->block_size, 0, 0, fua, 0, 0,
485 iscsi_co_generic_cb, &iTask);
487 if (iTask.task == NULL) {
488 return -ENOMEM;
490 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
491 iov->niov);
492 while (!iTask.complete) {
493 iscsi_set_events(iscsilun);
494 qemu_coroutine_yield();
497 if (iTask.task != NULL) {
498 scsi_free_scsi_task(iTask.task);
499 iTask.task = NULL;
502 if (iTask.do_retry) {
503 iTask.complete = 0;
504 goto retry;
507 if (iTask.status != SCSI_STATUS_GOOD) {
508 return iTask.err_code;
511 iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors);
513 return 0;
516 static int coroutine_fn
517 iscsi_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
518 QEMUIOVector *iov)
520 return iscsi_co_writev_flags(bs, sector_num, nb_sectors, iov, 0);
524 static bool iscsi_allocationmap_is_allocated(IscsiLun *iscsilun,
525 int64_t sector_num, int nb_sectors)
527 unsigned long size;
528 if (iscsilun->allocationmap == NULL) {
529 return true;
531 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
532 return !(find_next_bit(iscsilun->allocationmap, size,
533 sector_num / iscsilun->cluster_sectors) == size);
536 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
537 int64_t sector_num,
538 int nb_sectors, int *pnum,
539 BlockDriverState **file)
541 IscsiLun *iscsilun = bs->opaque;
542 struct scsi_get_lba_status *lbas = NULL;
543 struct scsi_lba_status_descriptor *lbasd = NULL;
544 struct IscsiTask iTask;
545 int64_t ret;
547 iscsi_co_init_iscsitask(iscsilun, &iTask);
549 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
550 ret = -EINVAL;
551 goto out;
554 /* default to all sectors allocated */
555 ret = BDRV_BLOCK_DATA;
556 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
557 *pnum = nb_sectors;
559 /* LUN does not support logical block provisioning */
560 if (!iscsilun->lbpme) {
561 goto out;
564 retry:
565 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
566 sector_qemu2lun(sector_num, iscsilun),
567 8 + 16, iscsi_co_generic_cb,
568 &iTask) == NULL) {
569 ret = -ENOMEM;
570 goto out;
573 while (!iTask.complete) {
574 iscsi_set_events(iscsilun);
575 qemu_coroutine_yield();
578 if (iTask.do_retry) {
579 if (iTask.task != NULL) {
580 scsi_free_scsi_task(iTask.task);
581 iTask.task = NULL;
583 iTask.complete = 0;
584 goto retry;
587 if (iTask.status != SCSI_STATUS_GOOD) {
588 /* in case the get_lba_status_callout fails (i.e.
589 * because the device is busy or the cmd is not
590 * supported) we pretend all blocks are allocated
591 * for backwards compatibility */
592 goto out;
595 lbas = scsi_datain_unmarshall(iTask.task);
596 if (lbas == NULL) {
597 ret = -EIO;
598 goto out;
601 lbasd = &lbas->descriptors[0];
603 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
604 ret = -EIO;
605 goto out;
608 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
610 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
611 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
612 ret &= ~BDRV_BLOCK_DATA;
613 if (iscsilun->lbprz) {
614 ret |= BDRV_BLOCK_ZERO;
618 if (ret & BDRV_BLOCK_ZERO) {
619 iscsi_allocationmap_clear(iscsilun, sector_num, *pnum);
620 } else {
621 iscsi_allocationmap_set(iscsilun, sector_num, *pnum);
624 if (*pnum > nb_sectors) {
625 *pnum = nb_sectors;
627 out:
628 if (iTask.task != NULL) {
629 scsi_free_scsi_task(iTask.task);
631 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) {
632 *file = bs;
634 return ret;
637 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
638 int64_t sector_num, int nb_sectors,
639 QEMUIOVector *iov)
641 IscsiLun *iscsilun = bs->opaque;
642 struct IscsiTask iTask;
643 uint64_t lba;
644 uint32_t num_sectors;
646 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
647 return -EINVAL;
650 if (bs->bl.max_transfer_length && nb_sectors > bs->bl.max_transfer_length) {
651 error_report("iSCSI Error: Read of %d sectors exceeds max_xfer_len "
652 "of %d sectors", nb_sectors, bs->bl.max_transfer_length);
653 return -EINVAL;
656 if (iscsilun->lbprz && nb_sectors >= ISCSI_CHECKALLOC_THRES &&
657 !iscsi_allocationmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
658 int64_t ret;
659 int pnum;
660 BlockDriverState *file;
661 ret = iscsi_co_get_block_status(bs, sector_num, INT_MAX, &pnum, &file);
662 if (ret < 0) {
663 return ret;
665 if (ret & BDRV_BLOCK_ZERO && pnum >= nb_sectors) {
666 qemu_iovec_memset(iov, 0, 0x00, iov->size);
667 return 0;
671 lba = sector_qemu2lun(sector_num, iscsilun);
672 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
674 iscsi_co_init_iscsitask(iscsilun, &iTask);
675 retry:
676 if (iscsilun->use_16_for_rw) {
677 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
678 num_sectors * iscsilun->block_size,
679 iscsilun->block_size, 0, 0, 0, 0, 0,
680 iscsi_co_generic_cb, &iTask);
681 } else {
682 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
683 num_sectors * iscsilun->block_size,
684 iscsilun->block_size,
685 0, 0, 0, 0, 0,
686 iscsi_co_generic_cb, &iTask);
688 if (iTask.task == NULL) {
689 return -ENOMEM;
691 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
693 while (!iTask.complete) {
694 iscsi_set_events(iscsilun);
695 qemu_coroutine_yield();
698 if (iTask.task != NULL) {
699 scsi_free_scsi_task(iTask.task);
700 iTask.task = NULL;
703 if (iTask.do_retry) {
704 iTask.complete = 0;
705 goto retry;
708 if (iTask.status != SCSI_STATUS_GOOD) {
709 return iTask.err_code;
712 return 0;
715 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
717 IscsiLun *iscsilun = bs->opaque;
718 struct IscsiTask iTask;
720 iscsi_co_init_iscsitask(iscsilun, &iTask);
721 retry:
722 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
723 0, iscsi_co_generic_cb, &iTask) == NULL) {
724 return -ENOMEM;
727 while (!iTask.complete) {
728 iscsi_set_events(iscsilun);
729 qemu_coroutine_yield();
732 if (iTask.task != NULL) {
733 scsi_free_scsi_task(iTask.task);
734 iTask.task = NULL;
737 if (iTask.do_retry) {
738 iTask.complete = 0;
739 goto retry;
742 if (iTask.status != SCSI_STATUS_GOOD) {
743 return iTask.err_code;
746 return 0;
749 #ifdef __linux__
750 static void
751 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
752 void *command_data, void *opaque)
754 IscsiAIOCB *acb = opaque;
756 g_free(acb->buf);
757 acb->buf = NULL;
759 acb->status = 0;
760 if (status < 0) {
761 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
762 iscsi_get_error(iscsi));
763 acb->status = iscsi_translate_sense(&acb->task->sense);
766 acb->ioh->driver_status = 0;
767 acb->ioh->host_status = 0;
768 acb->ioh->resid = 0;
770 #define SG_ERR_DRIVER_SENSE 0x08
772 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
773 int ss;
775 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
777 acb->ioh->sb_len_wr = acb->task->datain.size - 2;
778 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
779 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
780 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
783 iscsi_schedule_bh(acb);
786 static void iscsi_ioctl_bh_completion(void *opaque)
788 IscsiAIOCB *acb = opaque;
790 qemu_bh_delete(acb->bh);
791 acb->common.cb(acb->common.opaque, acb->ret);
792 qemu_aio_unref(acb);
795 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
797 BlockDriverState *bs = acb->common.bs;
798 IscsiLun *iscsilun = bs->opaque;
799 int ret = 0;
801 switch (req) {
802 case SG_GET_VERSION_NUM:
803 *(int *)buf = 30000;
804 break;
805 case SG_GET_SCSI_ID:
806 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
807 break;
808 default:
809 ret = -EINVAL;
811 assert(!acb->bh);
812 acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
813 iscsi_ioctl_bh_completion, acb);
814 acb->ret = ret;
815 qemu_bh_schedule(acb->bh);
818 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
819 unsigned long int req, void *buf,
820 BlockCompletionFunc *cb, void *opaque)
822 IscsiLun *iscsilun = bs->opaque;
823 struct iscsi_context *iscsi = iscsilun->iscsi;
824 struct iscsi_data data;
825 IscsiAIOCB *acb;
827 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
829 acb->iscsilun = iscsilun;
830 acb->bh = NULL;
831 acb->status = -EINPROGRESS;
832 acb->buf = NULL;
833 acb->ioh = buf;
835 if (req != SG_IO) {
836 iscsi_ioctl_handle_emulated(acb, req, buf);
837 return &acb->common;
840 acb->task = malloc(sizeof(struct scsi_task));
841 if (acb->task == NULL) {
842 error_report("iSCSI: Failed to allocate task for scsi command. %s",
843 iscsi_get_error(iscsi));
844 qemu_aio_unref(acb);
845 return NULL;
847 memset(acb->task, 0, sizeof(struct scsi_task));
849 switch (acb->ioh->dxfer_direction) {
850 case SG_DXFER_TO_DEV:
851 acb->task->xfer_dir = SCSI_XFER_WRITE;
852 break;
853 case SG_DXFER_FROM_DEV:
854 acb->task->xfer_dir = SCSI_XFER_READ;
855 break;
856 default:
857 acb->task->xfer_dir = SCSI_XFER_NONE;
858 break;
861 acb->task->cdb_size = acb->ioh->cmd_len;
862 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
863 acb->task->expxferlen = acb->ioh->dxfer_len;
865 data.size = 0;
866 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
867 if (acb->ioh->iovec_count == 0) {
868 data.data = acb->ioh->dxferp;
869 data.size = acb->ioh->dxfer_len;
870 } else {
871 scsi_task_set_iov_out(acb->task,
872 (struct scsi_iovec *) acb->ioh->dxferp,
873 acb->ioh->iovec_count);
877 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
878 iscsi_aio_ioctl_cb,
879 (data.size > 0) ? &data : NULL,
880 acb) != 0) {
881 scsi_free_scsi_task(acb->task);
882 qemu_aio_unref(acb);
883 return NULL;
886 /* tell libiscsi to read straight into the buffer we got from ioctl */
887 if (acb->task->xfer_dir == SCSI_XFER_READ) {
888 if (acb->ioh->iovec_count == 0) {
889 scsi_task_add_data_in_buffer(acb->task,
890 acb->ioh->dxfer_len,
891 acb->ioh->dxferp);
892 } else {
893 scsi_task_set_iov_in(acb->task,
894 (struct scsi_iovec *) acb->ioh->dxferp,
895 acb->ioh->iovec_count);
899 iscsi_set_events(iscsilun);
901 return &acb->common;
904 #endif
906 static int64_t
907 iscsi_getlength(BlockDriverState *bs)
909 IscsiLun *iscsilun = bs->opaque;
910 int64_t len;
912 len = iscsilun->num_blocks;
913 len *= iscsilun->block_size;
915 return len;
918 static int
919 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num,
920 int nb_sectors)
922 IscsiLun *iscsilun = bs->opaque;
923 struct IscsiTask iTask;
924 struct unmap_list list;
926 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
927 return -EINVAL;
930 if (!iscsilun->lbp.lbpu) {
931 /* UNMAP is not supported by the target */
932 return 0;
935 list.lba = sector_qemu2lun(sector_num, iscsilun);
936 list.num = sector_qemu2lun(nb_sectors, iscsilun);
938 iscsi_co_init_iscsitask(iscsilun, &iTask);
939 retry:
940 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
941 iscsi_co_generic_cb, &iTask) == NULL) {
942 return -ENOMEM;
945 while (!iTask.complete) {
946 iscsi_set_events(iscsilun);
947 qemu_coroutine_yield();
950 if (iTask.task != NULL) {
951 scsi_free_scsi_task(iTask.task);
952 iTask.task = NULL;
955 if (iTask.do_retry) {
956 iTask.complete = 0;
957 goto retry;
960 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
961 /* the target might fail with a check condition if it
962 is not happy with the alignment of the UNMAP request
963 we silently fail in this case */
964 return 0;
967 if (iTask.status != SCSI_STATUS_GOOD) {
968 return iTask.err_code;
971 iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors);
973 return 0;
976 static int
977 coroutine_fn iscsi_co_write_zeroes(BlockDriverState *bs, int64_t sector_num,
978 int nb_sectors, BdrvRequestFlags flags)
980 IscsiLun *iscsilun = bs->opaque;
981 struct IscsiTask iTask;
982 uint64_t lba;
983 uint32_t nb_blocks;
984 bool use_16_for_ws = iscsilun->use_16_for_rw;
986 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
987 return -EINVAL;
990 if (flags & BDRV_REQ_MAY_UNMAP) {
991 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
992 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
993 use_16_for_ws = true;
995 if (use_16_for_ws && !iscsilun->lbp.lbpws) {
996 /* WRITESAME16 with UNMAP is not supported by the target,
997 * fall back and try WRITESAME10/16 without UNMAP */
998 flags &= ~BDRV_REQ_MAY_UNMAP;
999 use_16_for_ws = iscsilun->use_16_for_rw;
1003 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1004 /* WRITESAME without UNMAP is not supported by the target */
1005 return -ENOTSUP;
1008 lba = sector_qemu2lun(sector_num, iscsilun);
1009 nb_blocks = sector_qemu2lun(nb_sectors, iscsilun);
1011 if (iscsilun->zeroblock == NULL) {
1012 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1013 if (iscsilun->zeroblock == NULL) {
1014 return -ENOMEM;
1018 iscsi_co_init_iscsitask(iscsilun, &iTask);
1019 retry:
1020 if (use_16_for_ws) {
1021 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1022 iscsilun->zeroblock, iscsilun->block_size,
1023 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1024 0, 0, iscsi_co_generic_cb, &iTask);
1025 } else {
1026 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1027 iscsilun->zeroblock, iscsilun->block_size,
1028 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1029 0, 0, iscsi_co_generic_cb, &iTask);
1031 if (iTask.task == NULL) {
1032 return -ENOMEM;
1035 while (!iTask.complete) {
1036 iscsi_set_events(iscsilun);
1037 qemu_coroutine_yield();
1040 if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1041 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1042 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1043 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1044 /* WRITE SAME is not supported by the target */
1045 iscsilun->has_write_same = false;
1046 scsi_free_scsi_task(iTask.task);
1047 return -ENOTSUP;
1050 if (iTask.task != NULL) {
1051 scsi_free_scsi_task(iTask.task);
1052 iTask.task = NULL;
1055 if (iTask.do_retry) {
1056 iTask.complete = 0;
1057 goto retry;
1060 if (iTask.status != SCSI_STATUS_GOOD) {
1061 return iTask.err_code;
1064 if (flags & BDRV_REQ_MAY_UNMAP) {
1065 iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors);
1066 } else {
1067 iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors);
1070 return 0;
1073 static void parse_chap(struct iscsi_context *iscsi, const char *target,
1074 Error **errp)
1076 QemuOptsList *list;
1077 QemuOpts *opts;
1078 const char *user = NULL;
1079 const char *password = NULL;
1080 const char *secretid;
1081 char *secret = NULL;
1083 list = qemu_find_opts("iscsi");
1084 if (!list) {
1085 return;
1088 opts = qemu_opts_find(list, target);
1089 if (opts == NULL) {
1090 opts = QTAILQ_FIRST(&list->head);
1091 if (!opts) {
1092 return;
1096 user = qemu_opt_get(opts, "user");
1097 if (!user) {
1098 return;
1101 secretid = qemu_opt_get(opts, "password-secret");
1102 password = qemu_opt_get(opts, "password");
1103 if (secretid && password) {
1104 error_setg(errp, "'password' and 'password-secret' properties are "
1105 "mutually exclusive");
1106 return;
1108 if (secretid) {
1109 secret = qcrypto_secret_lookup_as_utf8(secretid, errp);
1110 if (!secret) {
1111 return;
1113 password = secret;
1114 } else if (!password) {
1115 error_setg(errp, "CHAP username specified but no password was given");
1116 return;
1119 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1120 error_setg(errp, "Failed to set initiator username and password");
1123 g_free(secret);
1126 static void parse_header_digest(struct iscsi_context *iscsi, const char *target,
1127 Error **errp)
1129 QemuOptsList *list;
1130 QemuOpts *opts;
1131 const char *digest = NULL;
1133 list = qemu_find_opts("iscsi");
1134 if (!list) {
1135 return;
1138 opts = qemu_opts_find(list, target);
1139 if (opts == NULL) {
1140 opts = QTAILQ_FIRST(&list->head);
1141 if (!opts) {
1142 return;
1146 digest = qemu_opt_get(opts, "header-digest");
1147 if (!digest) {
1148 return;
1151 if (!strcmp(digest, "CRC32C")) {
1152 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1153 } else if (!strcmp(digest, "NONE")) {
1154 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1155 } else if (!strcmp(digest, "CRC32C-NONE")) {
1156 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1157 } else if (!strcmp(digest, "NONE-CRC32C")) {
1158 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1159 } else {
1160 error_setg(errp, "Invalid header-digest setting : %s", digest);
1164 static char *parse_initiator_name(const char *target)
1166 QemuOptsList *list;
1167 QemuOpts *opts;
1168 const char *name;
1169 char *iscsi_name;
1170 UuidInfo *uuid_info;
1172 list = qemu_find_opts("iscsi");
1173 if (list) {
1174 opts = qemu_opts_find(list, target);
1175 if (!opts) {
1176 opts = QTAILQ_FIRST(&list->head);
1178 if (opts) {
1179 name = qemu_opt_get(opts, "initiator-name");
1180 if (name) {
1181 return g_strdup(name);
1186 uuid_info = qmp_query_uuid(NULL);
1187 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1188 name = qemu_get_vm_name();
1189 } else {
1190 name = uuid_info->UUID;
1192 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1193 name ? ":" : "", name ? name : "");
1194 qapi_free_UuidInfo(uuid_info);
1195 return iscsi_name;
1198 static int parse_timeout(const char *target)
1200 QemuOptsList *list;
1201 QemuOpts *opts;
1202 const char *timeout;
1204 list = qemu_find_opts("iscsi");
1205 if (list) {
1206 opts = qemu_opts_find(list, target);
1207 if (!opts) {
1208 opts = QTAILQ_FIRST(&list->head);
1210 if (opts) {
1211 timeout = qemu_opt_get(opts, "timeout");
1212 if (timeout) {
1213 return atoi(timeout);
1218 return 0;
1221 static void iscsi_nop_timed_event(void *opaque)
1223 IscsiLun *iscsilun = opaque;
1225 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1226 error_report("iSCSI: NOP timeout. Reconnecting...");
1227 iscsilun->request_timed_out = true;
1228 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1229 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1230 return;
1233 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1234 iscsi_set_events(iscsilun);
1237 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1239 struct scsi_task *task = NULL;
1240 struct scsi_readcapacity10 *rc10 = NULL;
1241 struct scsi_readcapacity16 *rc16 = NULL;
1242 int retries = ISCSI_CMD_RETRIES;
1244 do {
1245 if (task != NULL) {
1246 scsi_free_scsi_task(task);
1247 task = NULL;
1250 switch (iscsilun->type) {
1251 case TYPE_DISK:
1252 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1253 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1254 rc16 = scsi_datain_unmarshall(task);
1255 if (rc16 == NULL) {
1256 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1257 } else {
1258 iscsilun->block_size = rc16->block_length;
1259 iscsilun->num_blocks = rc16->returned_lba + 1;
1260 iscsilun->lbpme = !!rc16->lbpme;
1261 iscsilun->lbprz = !!rc16->lbprz;
1262 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1264 break;
1266 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1267 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1268 break;
1270 /* Fall through and try READ CAPACITY(10) instead. */
1271 case TYPE_ROM:
1272 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1273 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1274 rc10 = scsi_datain_unmarshall(task);
1275 if (rc10 == NULL) {
1276 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1277 } else {
1278 iscsilun->block_size = rc10->block_size;
1279 if (rc10->lba == 0) {
1280 /* blank disk loaded */
1281 iscsilun->num_blocks = 0;
1282 } else {
1283 iscsilun->num_blocks = rc10->lba + 1;
1287 break;
1288 default:
1289 return;
1291 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1292 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1293 && retries-- > 0);
1295 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1296 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1297 } else if (!iscsilun->block_size ||
1298 iscsilun->block_size % BDRV_SECTOR_SIZE) {
1299 error_setg(errp, "iSCSI: the target returned an invalid "
1300 "block size of %d.", iscsilun->block_size);
1302 if (task) {
1303 scsi_free_scsi_task(task);
1307 /* TODO Convert to fine grained options */
1308 static QemuOptsList runtime_opts = {
1309 .name = "iscsi",
1310 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1311 .desc = {
1313 .name = "filename",
1314 .type = QEMU_OPT_STRING,
1315 .help = "URL to the iscsi image",
1317 { /* end of list */ }
1321 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1322 int evpd, int pc, void **inq, Error **errp)
1324 int full_size;
1325 struct scsi_task *task = NULL;
1326 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1327 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1328 goto fail;
1330 full_size = scsi_datain_getfullsize(task);
1331 if (full_size > task->datain.size) {
1332 scsi_free_scsi_task(task);
1334 /* we need more data for the full list */
1335 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1336 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1337 goto fail;
1341 *inq = scsi_datain_unmarshall(task);
1342 if (*inq == NULL) {
1343 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1344 goto fail_with_err;
1347 return task;
1349 fail:
1350 error_setg(errp, "iSCSI: Inquiry command failed : %s",
1351 iscsi_get_error(iscsi));
1352 fail_with_err:
1353 if (task != NULL) {
1354 scsi_free_scsi_task(task);
1356 return NULL;
1359 static void iscsi_detach_aio_context(BlockDriverState *bs)
1361 IscsiLun *iscsilun = bs->opaque;
1363 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1364 false, NULL, NULL, NULL);
1365 iscsilun->events = 0;
1367 if (iscsilun->nop_timer) {
1368 timer_del(iscsilun->nop_timer);
1369 timer_free(iscsilun->nop_timer);
1370 iscsilun->nop_timer = NULL;
1372 if (iscsilun->event_timer) {
1373 timer_del(iscsilun->event_timer);
1374 timer_free(iscsilun->event_timer);
1375 iscsilun->event_timer = NULL;
1379 static void iscsi_attach_aio_context(BlockDriverState *bs,
1380 AioContext *new_context)
1382 IscsiLun *iscsilun = bs->opaque;
1384 iscsilun->aio_context = new_context;
1385 iscsi_set_events(iscsilun);
1387 /* Set up a timer for sending out iSCSI NOPs */
1388 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1389 QEMU_CLOCK_REALTIME, SCALE_MS,
1390 iscsi_nop_timed_event, iscsilun);
1391 timer_mod(iscsilun->nop_timer,
1392 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1394 /* Set up a timer for periodic calls to iscsi_set_events and to
1395 * scan for command timeout */
1396 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1397 QEMU_CLOCK_REALTIME, SCALE_MS,
1398 iscsi_timed_check_events, iscsilun);
1399 timer_mod(iscsilun->event_timer,
1400 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1403 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1405 struct scsi_task *task;
1406 struct scsi_mode_sense *ms = NULL;
1407 iscsilun->write_protected = false;
1408 iscsilun->dpofua = false;
1410 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1411 1, SCSI_MODESENSE_PC_CURRENT,
1412 0x3F, 0, 255);
1413 if (task == NULL) {
1414 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1415 iscsi_get_error(iscsilun->iscsi));
1416 goto out;
1419 if (task->status != SCSI_STATUS_GOOD) {
1420 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1421 goto out;
1423 ms = scsi_datain_unmarshall(task);
1424 if (!ms) {
1425 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1426 iscsi_get_error(iscsilun->iscsi));
1427 goto out;
1429 iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1430 iscsilun->dpofua = ms->device_specific_parameter & 0x10;
1432 out:
1433 if (task) {
1434 scsi_free_scsi_task(task);
1439 * We support iscsi url's on the form
1440 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1442 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1443 Error **errp)
1445 IscsiLun *iscsilun = bs->opaque;
1446 struct iscsi_context *iscsi = NULL;
1447 struct iscsi_url *iscsi_url = NULL;
1448 struct scsi_task *task = NULL;
1449 struct scsi_inquiry_standard *inq = NULL;
1450 struct scsi_inquiry_supported_pages *inq_vpd;
1451 char *initiator_name = NULL;
1452 QemuOpts *opts;
1453 Error *local_err = NULL;
1454 const char *filename;
1455 int i, ret = 0, timeout = 0;
1457 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1458 qemu_opts_absorb_qdict(opts, options, &local_err);
1459 if (local_err) {
1460 error_propagate(errp, local_err);
1461 ret = -EINVAL;
1462 goto out;
1465 filename = qemu_opt_get(opts, "filename");
1467 iscsi_url = iscsi_parse_full_url(iscsi, filename);
1468 if (iscsi_url == NULL) {
1469 error_setg(errp, "Failed to parse URL : %s", filename);
1470 ret = -EINVAL;
1471 goto out;
1474 memset(iscsilun, 0, sizeof(IscsiLun));
1476 initiator_name = parse_initiator_name(iscsi_url->target);
1478 iscsi = iscsi_create_context(initiator_name);
1479 if (iscsi == NULL) {
1480 error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1481 ret = -ENOMEM;
1482 goto out;
1485 if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1486 error_setg(errp, "iSCSI: Failed to set target name.");
1487 ret = -EINVAL;
1488 goto out;
1491 if (iscsi_url->user[0] != '\0') {
1492 ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1493 iscsi_url->passwd);
1494 if (ret != 0) {
1495 error_setg(errp, "Failed to set initiator username and password");
1496 ret = -EINVAL;
1497 goto out;
1501 /* check if we got CHAP username/password via the options */
1502 parse_chap(iscsi, iscsi_url->target, &local_err);
1503 if (local_err != NULL) {
1504 error_propagate(errp, local_err);
1505 ret = -EINVAL;
1506 goto out;
1509 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1510 error_setg(errp, "iSCSI: Failed to set session type to normal.");
1511 ret = -EINVAL;
1512 goto out;
1515 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1517 /* check if we got HEADER_DIGEST via the options */
1518 parse_header_digest(iscsi, iscsi_url->target, &local_err);
1519 if (local_err != NULL) {
1520 error_propagate(errp, local_err);
1521 ret = -EINVAL;
1522 goto out;
1525 /* timeout handling is broken in libiscsi before 1.15.0 */
1526 timeout = parse_timeout(iscsi_url->target);
1527 #if defined(LIBISCSI_API_VERSION) && LIBISCSI_API_VERSION >= 20150621
1528 iscsi_set_timeout(iscsi, timeout);
1529 #else
1530 if (timeout) {
1531 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1533 #endif
1535 if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1536 error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1537 iscsi_get_error(iscsi));
1538 ret = -EINVAL;
1539 goto out;
1542 iscsilun->iscsi = iscsi;
1543 iscsilun->aio_context = bdrv_get_aio_context(bs);
1544 iscsilun->lun = iscsi_url->lun;
1545 iscsilun->has_write_same = true;
1547 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1548 (void **) &inq, errp);
1549 if (task == NULL) {
1550 ret = -EINVAL;
1551 goto out;
1553 iscsilun->type = inq->periperal_device_type;
1554 scsi_free_scsi_task(task);
1555 task = NULL;
1557 iscsi_modesense_sync(iscsilun);
1559 /* Check the write protect flag of the LUN if we want to write */
1560 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1561 iscsilun->write_protected) {
1562 error_setg(errp, "Cannot open a write protected LUN as read-write");
1563 ret = -EACCES;
1564 goto out;
1567 iscsi_readcapacity_sync(iscsilun, &local_err);
1568 if (local_err != NULL) {
1569 error_propagate(errp, local_err);
1570 ret = -EINVAL;
1571 goto out;
1573 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1574 bs->request_alignment = iscsilun->block_size;
1576 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1577 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1578 * will try to read from the device to guess the image format.
1580 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1581 bs->sg = 1;
1584 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1585 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1586 (void **) &inq_vpd, errp);
1587 if (task == NULL) {
1588 ret = -EINVAL;
1589 goto out;
1591 for (i = 0; i < inq_vpd->num_pages; i++) {
1592 struct scsi_task *inq_task;
1593 struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1594 struct scsi_inquiry_block_limits *inq_bl;
1595 switch (inq_vpd->pages[i]) {
1596 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1597 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1598 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1599 (void **) &inq_lbp, errp);
1600 if (inq_task == NULL) {
1601 ret = -EINVAL;
1602 goto out;
1604 memcpy(&iscsilun->lbp, inq_lbp,
1605 sizeof(struct scsi_inquiry_logical_block_provisioning));
1606 scsi_free_scsi_task(inq_task);
1607 break;
1608 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1609 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1610 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1611 (void **) &inq_bl, errp);
1612 if (inq_task == NULL) {
1613 ret = -EINVAL;
1614 goto out;
1616 memcpy(&iscsilun->bl, inq_bl,
1617 sizeof(struct scsi_inquiry_block_limits));
1618 scsi_free_scsi_task(inq_task);
1619 break;
1620 default:
1621 break;
1624 scsi_free_scsi_task(task);
1625 task = NULL;
1627 iscsi_attach_aio_context(bs, iscsilun->aio_context);
1629 /* Guess the internal cluster (page) size of the iscsi target by the means
1630 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1631 * reasonable size */
1632 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
1633 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1634 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran *
1635 iscsilun->block_size) >> BDRV_SECTOR_BITS;
1636 if (iscsilun->lbprz) {
1637 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun);
1638 if (iscsilun->allocationmap == NULL) {
1639 ret = -ENOMEM;
1644 out:
1645 qemu_opts_del(opts);
1646 g_free(initiator_name);
1647 if (iscsi_url != NULL) {
1648 iscsi_destroy_url(iscsi_url);
1650 if (task != NULL) {
1651 scsi_free_scsi_task(task);
1654 if (ret) {
1655 if (iscsi != NULL) {
1656 if (iscsi_is_logged_in(iscsi)) {
1657 iscsi_logout_sync(iscsi);
1659 iscsi_destroy_context(iscsi);
1661 memset(iscsilun, 0, sizeof(IscsiLun));
1663 return ret;
1666 static void iscsi_close(BlockDriverState *bs)
1668 IscsiLun *iscsilun = bs->opaque;
1669 struct iscsi_context *iscsi = iscsilun->iscsi;
1671 iscsi_detach_aio_context(bs);
1672 if (iscsi_is_logged_in(iscsi)) {
1673 iscsi_logout_sync(iscsi);
1675 iscsi_destroy_context(iscsi);
1676 g_free(iscsilun->zeroblock);
1677 g_free(iscsilun->allocationmap);
1678 memset(iscsilun, 0, sizeof(IscsiLun));
1681 static int sector_limits_lun2qemu(int64_t sector, IscsiLun *iscsilun)
1683 return MIN(sector_lun2qemu(sector, iscsilun), INT_MAX / 2 + 1);
1686 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
1688 /* We don't actually refresh here, but just return data queried in
1689 * iscsi_open(): iscsi targets don't change their limits. */
1691 IscsiLun *iscsilun = bs->opaque;
1692 uint32_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
1694 if (iscsilun->bl.max_xfer_len) {
1695 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
1698 bs->bl.max_transfer_length = sector_limits_lun2qemu(max_xfer_len, iscsilun);
1700 if (iscsilun->lbp.lbpu) {
1701 if (iscsilun->bl.max_unmap < 0xffffffff) {
1702 bs->bl.max_discard =
1703 sector_limits_lun2qemu(iscsilun->bl.max_unmap, iscsilun);
1705 bs->bl.discard_alignment =
1706 sector_limits_lun2qemu(iscsilun->bl.opt_unmap_gran, iscsilun);
1709 if (iscsilun->bl.max_ws_len < 0xffffffff) {
1710 bs->bl.max_write_zeroes =
1711 sector_limits_lun2qemu(iscsilun->bl.max_ws_len, iscsilun);
1713 if (iscsilun->lbp.lbpws) {
1714 bs->bl.write_zeroes_alignment =
1715 sector_limits_lun2qemu(iscsilun->bl.opt_unmap_gran, iscsilun);
1717 bs->bl.opt_transfer_length =
1718 sector_limits_lun2qemu(iscsilun->bl.opt_xfer_len, iscsilun);
1721 /* Note that this will not re-establish a connection with an iSCSI target - it
1722 * is effectively a NOP. */
1723 static int iscsi_reopen_prepare(BDRVReopenState *state,
1724 BlockReopenQueue *queue, Error **errp)
1726 IscsiLun *iscsilun = state->bs->opaque;
1728 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
1729 error_setg(errp, "Cannot open a write protected LUN as read-write");
1730 return -EACCES;
1732 return 0;
1735 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1737 IscsiLun *iscsilun = bs->opaque;
1738 Error *local_err = NULL;
1740 if (iscsilun->type != TYPE_DISK) {
1741 return -ENOTSUP;
1744 iscsi_readcapacity_sync(iscsilun, &local_err);
1745 if (local_err != NULL) {
1746 error_free(local_err);
1747 return -EIO;
1750 if (offset > iscsi_getlength(bs)) {
1751 return -EINVAL;
1754 if (iscsilun->allocationmap != NULL) {
1755 g_free(iscsilun->allocationmap);
1756 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun);
1759 return 0;
1762 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp)
1764 int ret = 0;
1765 int64_t total_size = 0;
1766 BlockDriverState *bs;
1767 IscsiLun *iscsilun = NULL;
1768 QDict *bs_options;
1770 bs = bdrv_new();
1772 /* Read out options */
1773 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1774 BDRV_SECTOR_SIZE);
1775 bs->opaque = g_new0(struct IscsiLun, 1);
1776 iscsilun = bs->opaque;
1778 bs_options = qdict_new();
1779 qdict_put(bs_options, "filename", qstring_from_str(filename));
1780 ret = iscsi_open(bs, bs_options, 0, NULL);
1781 QDECREF(bs_options);
1783 if (ret != 0) {
1784 goto out;
1786 iscsi_detach_aio_context(bs);
1787 if (iscsilun->type != TYPE_DISK) {
1788 ret = -ENODEV;
1789 goto out;
1791 if (bs->total_sectors < total_size) {
1792 ret = -ENOSPC;
1793 goto out;
1796 ret = 0;
1797 out:
1798 if (iscsilun->iscsi != NULL) {
1799 iscsi_destroy_context(iscsilun->iscsi);
1801 g_free(bs->opaque);
1802 bs->opaque = NULL;
1803 bdrv_unref(bs);
1804 return ret;
1807 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1809 IscsiLun *iscsilun = bs->opaque;
1810 bdi->unallocated_blocks_are_zero = iscsilun->lbprz;
1811 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
1812 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE;
1813 return 0;
1816 static QemuOptsList iscsi_create_opts = {
1817 .name = "iscsi-create-opts",
1818 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
1819 .desc = {
1821 .name = BLOCK_OPT_SIZE,
1822 .type = QEMU_OPT_SIZE,
1823 .help = "Virtual disk size"
1825 { /* end of list */ }
1829 static BlockDriver bdrv_iscsi = {
1830 .format_name = "iscsi",
1831 .protocol_name = "iscsi",
1833 .instance_size = sizeof(IscsiLun),
1834 .bdrv_needs_filename = true,
1835 .bdrv_file_open = iscsi_open,
1836 .bdrv_close = iscsi_close,
1837 .bdrv_create = iscsi_create,
1838 .create_opts = &iscsi_create_opts,
1839 .bdrv_reopen_prepare = iscsi_reopen_prepare,
1841 .bdrv_getlength = iscsi_getlength,
1842 .bdrv_get_info = iscsi_get_info,
1843 .bdrv_truncate = iscsi_truncate,
1844 .bdrv_refresh_limits = iscsi_refresh_limits,
1846 .bdrv_co_get_block_status = iscsi_co_get_block_status,
1847 .bdrv_co_discard = iscsi_co_discard,
1848 .bdrv_co_write_zeroes = iscsi_co_write_zeroes,
1849 .bdrv_co_readv = iscsi_co_readv,
1850 .bdrv_co_writev = iscsi_co_writev,
1851 .bdrv_co_writev_flags = iscsi_co_writev_flags,
1852 .supported_write_flags = BDRV_REQ_FUA,
1853 .bdrv_co_flush_to_disk = iscsi_co_flush,
1855 #ifdef __linux__
1856 .bdrv_aio_ioctl = iscsi_aio_ioctl,
1857 #endif
1859 .bdrv_detach_aio_context = iscsi_detach_aio_context,
1860 .bdrv_attach_aio_context = iscsi_attach_aio_context,
1863 static QemuOptsList qemu_iscsi_opts = {
1864 .name = "iscsi",
1865 .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1866 .desc = {
1868 .name = "user",
1869 .type = QEMU_OPT_STRING,
1870 .help = "username for CHAP authentication to target",
1872 .name = "password",
1873 .type = QEMU_OPT_STRING,
1874 .help = "password for CHAP authentication to target",
1876 .name = "password-secret",
1877 .type = QEMU_OPT_STRING,
1878 .help = "ID of the secret providing password for CHAP "
1879 "authentication to target",
1881 .name = "header-digest",
1882 .type = QEMU_OPT_STRING,
1883 .help = "HeaderDigest setting. "
1884 "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1886 .name = "initiator-name",
1887 .type = QEMU_OPT_STRING,
1888 .help = "Initiator iqn name to use when connecting",
1890 .name = "timeout",
1891 .type = QEMU_OPT_NUMBER,
1892 .help = "Request timeout in seconds (default 0 = no timeout)",
1894 { /* end of list */ }
1898 static void iscsi_block_init(void)
1900 bdrv_register(&bdrv_iscsi);
1901 qemu_add_opts(&qemu_iscsi_opts);
1904 block_init(iscsi_block_init);