xref: /qemu/block/iscsi.c (revision e7b3af81)
1 /*
2  * QEMU Block driver for iSCSI images
3  *
4  * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
5  * Copyright (c) 2012-2017 Peter Lieven <pl@kamp.de>
6  *
7  * Permission is hereby granted, free of charge, to any person obtaining a copy
8  * of this software and associated documentation files (the "Software"), to deal
9  * in the Software without restriction, including without limitation the rights
10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11  * copies of the Software, and to permit persons to whom the Software is
12  * furnished to do so, subject to the following conditions:
13  *
14  * The above copyright notice and this permission notice shall be included in
15  * all copies or substantial portions of the Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23  * THE SOFTWARE.
24  */
25 
26 #include "qemu/osdep.h"
27 
28 #include <poll.h>
29 #include <math.h>
30 #include <arpa/inet.h>
31 #include "qemu/config-file.h"
32 #include "qemu/error-report.h"
33 #include "qemu/bitops.h"
34 #include "qemu/bitmap.h"
35 #include "block/block_int.h"
36 #include "block/qdict.h"
37 #include "scsi/constants.h"
38 #include "qemu/iov.h"
39 #include "qemu/option.h"
40 #include "qemu/uuid.h"
41 #include "qapi/error.h"
42 #include "qapi/qapi-commands-misc.h"
43 #include "qapi/qmp/qdict.h"
44 #include "qapi/qmp/qstring.h"
45 #include "crypto/secret.h"
46 #include "scsi/utils.h"
47 
48 /* Conflict between scsi/utils.h and libiscsi! :( */
49 #define SCSI_XFER_NONE ISCSI_XFER_NONE
50 #include <iscsi/iscsi.h>
51 #include <iscsi/scsi-lowlevel.h>
52 #undef SCSI_XFER_NONE
53 QEMU_BUILD_BUG_ON((int)SCSI_XFER_NONE != (int)ISCSI_XFER_NONE);
54 
55 #ifdef __linux__
56 #include <scsi/sg.h>
57 #endif
58 
59 typedef struct IscsiLun {
60     struct iscsi_context *iscsi;
61     AioContext *aio_context;
62     int lun;
63     enum scsi_inquiry_peripheral_device_type type;
64     int block_size;
65     uint64_t num_blocks;
66     int events;
67     QEMUTimer *nop_timer;
68     QEMUTimer *event_timer;
69     QemuMutex mutex;
70     struct scsi_inquiry_logical_block_provisioning lbp;
71     struct scsi_inquiry_block_limits bl;
72     struct scsi_inquiry_device_designator *dd;
73     unsigned char *zeroblock;
74     /* The allocmap tracks which clusters (pages) on the iSCSI target are
75      * allocated and which are not. In case a target returns zeros for
76      * unallocated pages (iscsilun->lprz) we can directly return zeros instead
77      * of reading zeros over the wire if a read request falls within an
78      * unallocated block. As there are 3 possible states we need 2 bitmaps to
79      * track. allocmap_valid keeps track if QEMU's information about a page is
80      * valid. allocmap tracks if a page is allocated or not. In case QEMU has no
81      * valid information about a page the corresponding allocmap entry should be
82      * switched to unallocated as well to force a new lookup of the allocation
83      * status as lookups are generally skipped if a page is suspect to be
84      * allocated. If a iSCSI target is opened with cache.direct = on the
85      * allocmap_valid does not exist turning all cached information invalid so
86      * that a fresh lookup is made for any page even if allocmap entry returns
87      * it's unallocated. */
88     unsigned long *allocmap;
89     unsigned long *allocmap_valid;
90     long allocmap_size;
91     int cluster_size;
92     bool use_16_for_rw;
93     bool write_protected;
94     bool lbpme;
95     bool lbprz;
96     bool dpofua;
97     bool has_write_same;
98     bool request_timed_out;
99 } IscsiLun;
100 
101 typedef struct IscsiTask {
102     int status;
103     int complete;
104     int retries;
105     int do_retry;
106     struct scsi_task *task;
107     Coroutine *co;
108     IscsiLun *iscsilun;
109     QEMUTimer retry_timer;
110     int err_code;
111     char *err_str;
112 } IscsiTask;
113 
114 typedef struct IscsiAIOCB {
115     BlockAIOCB common;
116     QEMUBH *bh;
117     IscsiLun *iscsilun;
118     struct scsi_task *task;
119     uint8_t *buf;
120     int status;
121     int64_t sector_num;
122     int nb_sectors;
123     int ret;
124 #ifdef __linux__
125     sg_io_hdr_t *ioh;
126 #endif
127 } IscsiAIOCB;
128 
129 /* libiscsi uses time_t so its enough to process events every second */
130 #define EVENT_INTERVAL 1000
131 #define NOP_INTERVAL 5000
132 #define MAX_NOP_FAILURES 3
133 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times)
134 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768};
135 
136 /* this threshold is a trade-off knob to choose between
137  * the potential additional overhead of an extra GET_LBA_STATUS request
138  * vs. unnecessarily reading a lot of zero sectors over the wire.
139  * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
140  * sectors we check the allocation status of the area covered by the
141  * request first if the allocationmap indicates that the area might be
142  * unallocated. */
143 #define ISCSI_CHECKALLOC_THRES 64
144 
145 static void
146 iscsi_bh_cb(void *p)
147 {
148     IscsiAIOCB *acb = p;
149 
150     qemu_bh_delete(acb->bh);
151 
152     g_free(acb->buf);
153     acb->buf = NULL;
154 
155     acb->common.cb(acb->common.opaque, acb->status);
156 
157     if (acb->task != NULL) {
158         scsi_free_scsi_task(acb->task);
159         acb->task = NULL;
160     }
161 
162     qemu_aio_unref(acb);
163 }
164 
165 static void
166 iscsi_schedule_bh(IscsiAIOCB *acb)
167 {
168     if (acb->bh) {
169         return;
170     }
171     acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb);
172     qemu_bh_schedule(acb->bh);
173 }
174 
175 static void iscsi_co_generic_bh_cb(void *opaque)
176 {
177     struct IscsiTask *iTask = opaque;
178 
179     iTask->complete = 1;
180     aio_co_wake(iTask->co);
181 }
182 
183 static void iscsi_retry_timer_expired(void *opaque)
184 {
185     struct IscsiTask *iTask = opaque;
186     iTask->complete = 1;
187     if (iTask->co) {
188         aio_co_wake(iTask->co);
189     }
190 }
191 
192 static inline unsigned exp_random(double mean)
193 {
194     return -mean * log((double)rand() / RAND_MAX);
195 }
196 
197 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in
198  * libiscsi 1.10.0, together with other constants we need.  Use it as
199  * a hint that we have to define them ourselves if needed, to keep the
200  * minimum required libiscsi version at 1.9.0.  We use an ASCQ macro for
201  * the test because SCSI_STATUS_* is an enum.
202  *
203  * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes
204  * an enum, check against the LIBISCSI_API_VERSION macro, which was
205  * introduced in 1.11.0.  If it is present, there is no need to define
206  * anything.
207  */
208 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \
209     !defined(LIBISCSI_API_VERSION)
210 #define SCSI_STATUS_TASK_SET_FULL                          0x28
211 #define SCSI_STATUS_TIMEOUT                                0x0f000002
212 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST    0x2600
213 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR        0x1a00
214 #endif
215 
216 #ifndef LIBISCSI_API_VERSION
217 #define LIBISCSI_API_VERSION 20130701
218 #endif
219 
220 static int iscsi_translate_sense(struct scsi_sense *sense)
221 {
222     return - scsi_sense_to_errno(sense->key,
223                                  (sense->ascq & 0xFF00) >> 8,
224                                  sense->ascq & 0xFF);
225 }
226 
227 /* Called (via iscsi_service) with QemuMutex held.  */
228 static void
229 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
230                         void *command_data, void *opaque)
231 {
232     struct IscsiTask *iTask = opaque;
233     struct scsi_task *task = command_data;
234 
235     iTask->status = status;
236     iTask->do_retry = 0;
237     iTask->task = task;
238 
239     if (status != SCSI_STATUS_GOOD) {
240         if (iTask->retries++ < ISCSI_CMD_RETRIES) {
241             if (status == SCSI_STATUS_CHECK_CONDITION
242                 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
243                 error_report("iSCSI CheckCondition: %s",
244                              iscsi_get_error(iscsi));
245                 iTask->do_retry = 1;
246                 goto out;
247             }
248             if (status == SCSI_STATUS_BUSY ||
249                 status == SCSI_STATUS_TIMEOUT ||
250                 status == SCSI_STATUS_TASK_SET_FULL) {
251                 unsigned retry_time =
252                     exp_random(iscsi_retry_times[iTask->retries - 1]);
253                 if (status == SCSI_STATUS_TIMEOUT) {
254                     /* make sure the request is rescheduled AFTER the
255                      * reconnect is initiated */
256                     retry_time = EVENT_INTERVAL * 2;
257                     iTask->iscsilun->request_timed_out = true;
258                 }
259                 error_report("iSCSI Busy/TaskSetFull/TimeOut"
260                              " (retry #%u in %u ms): %s",
261                              iTask->retries, retry_time,
262                              iscsi_get_error(iscsi));
263                 aio_timer_init(iTask->iscsilun->aio_context,
264                                &iTask->retry_timer, QEMU_CLOCK_REALTIME,
265                                SCALE_MS, iscsi_retry_timer_expired, iTask);
266                 timer_mod(&iTask->retry_timer,
267                           qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
268                 iTask->do_retry = 1;
269                 return;
270             }
271         }
272         iTask->err_code = iscsi_translate_sense(&task->sense);
273         iTask->err_str = g_strdup(iscsi_get_error(iscsi));
274     }
275 
276 out:
277     if (iTask->co) {
278         aio_bh_schedule_oneshot(iTask->iscsilun->aio_context,
279                                  iscsi_co_generic_bh_cb, iTask);
280     } else {
281         iTask->complete = 1;
282     }
283 }
284 
285 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
286 {
287     *iTask = (struct IscsiTask) {
288         .co         = qemu_coroutine_self(),
289         .iscsilun   = iscsilun,
290     };
291 }
292 
293 static void
294 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
295                     void *private_data)
296 {
297     IscsiAIOCB *acb = private_data;
298 
299     acb->status = -ECANCELED;
300     iscsi_schedule_bh(acb);
301 }
302 
303 static void
304 iscsi_aio_cancel(BlockAIOCB *blockacb)
305 {
306     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
307     IscsiLun *iscsilun = acb->iscsilun;
308 
309     if (acb->status != -EINPROGRESS) {
310         return;
311     }
312 
313     /* send a task mgmt call to the target to cancel the task on the target */
314     iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
315                                      iscsi_abort_task_cb, acb);
316 
317 }
318 
319 static const AIOCBInfo iscsi_aiocb_info = {
320     .aiocb_size         = sizeof(IscsiAIOCB),
321     .cancel_async       = iscsi_aio_cancel,
322 };
323 
324 
325 static void iscsi_process_read(void *arg);
326 static void iscsi_process_write(void *arg);
327 
328 /* Called with QemuMutex held.  */
329 static void
330 iscsi_set_events(IscsiLun *iscsilun)
331 {
332     struct iscsi_context *iscsi = iscsilun->iscsi;
333     int ev = iscsi_which_events(iscsi);
334 
335     if (ev != iscsilun->events) {
336         aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi),
337                            false,
338                            (ev & POLLIN) ? iscsi_process_read : NULL,
339                            (ev & POLLOUT) ? iscsi_process_write : NULL,
340                            NULL,
341                            iscsilun);
342         iscsilun->events = ev;
343     }
344 }
345 
346 static void iscsi_timed_check_events(void *opaque)
347 {
348     IscsiLun *iscsilun = opaque;
349 
350     /* check for timed out requests */
351     iscsi_service(iscsilun->iscsi, 0);
352 
353     if (iscsilun->request_timed_out) {
354         iscsilun->request_timed_out = false;
355         iscsi_reconnect(iscsilun->iscsi);
356     }
357 
358     /* newer versions of libiscsi may return zero events. Ensure we are able
359      * to return to service once this situation changes. */
360     iscsi_set_events(iscsilun);
361 
362     timer_mod(iscsilun->event_timer,
363               qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
364 }
365 
366 static void
367 iscsi_process_read(void *arg)
368 {
369     IscsiLun *iscsilun = arg;
370     struct iscsi_context *iscsi = iscsilun->iscsi;
371 
372     qemu_mutex_lock(&iscsilun->mutex);
373     iscsi_service(iscsi, POLLIN);
374     iscsi_set_events(iscsilun);
375     qemu_mutex_unlock(&iscsilun->mutex);
376 }
377 
378 static void
379 iscsi_process_write(void *arg)
380 {
381     IscsiLun *iscsilun = arg;
382     struct iscsi_context *iscsi = iscsilun->iscsi;
383 
384     qemu_mutex_lock(&iscsilun->mutex);
385     iscsi_service(iscsi, POLLOUT);
386     iscsi_set_events(iscsilun);
387     qemu_mutex_unlock(&iscsilun->mutex);
388 }
389 
390 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
391 {
392     return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
393 }
394 
395 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
396 {
397     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
398 }
399 
400 static bool is_byte_request_lun_aligned(int64_t offset, int count,
401                                         IscsiLun *iscsilun)
402 {
403     if (offset % iscsilun->block_size || count % iscsilun->block_size) {
404         error_report("iSCSI misaligned request: "
405                      "iscsilun->block_size %u, offset %" PRIi64
406                      ", count %d",
407                      iscsilun->block_size, offset, count);
408         return false;
409     }
410     return true;
411 }
412 
413 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors,
414                                           IscsiLun *iscsilun)
415 {
416     assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS);
417     return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS,
418                                        nb_sectors << BDRV_SECTOR_BITS,
419                                        iscsilun);
420 }
421 
422 static void iscsi_allocmap_free(IscsiLun *iscsilun)
423 {
424     g_free(iscsilun->allocmap);
425     g_free(iscsilun->allocmap_valid);
426     iscsilun->allocmap = NULL;
427     iscsilun->allocmap_valid = NULL;
428 }
429 
430 
431 static int iscsi_allocmap_init(IscsiLun *iscsilun, int open_flags)
432 {
433     iscsi_allocmap_free(iscsilun);
434 
435     assert(iscsilun->cluster_size);
436     iscsilun->allocmap_size =
437         DIV_ROUND_UP(iscsilun->num_blocks * iscsilun->block_size,
438                      iscsilun->cluster_size);
439 
440     iscsilun->allocmap = bitmap_try_new(iscsilun->allocmap_size);
441     if (!iscsilun->allocmap) {
442         return -ENOMEM;
443     }
444 
445     if (open_flags & BDRV_O_NOCACHE) {
446         /* when cache.direct = on all allocmap entries are
447          * treated as invalid to force a relookup of the block
448          * status on every read request */
449         return 0;
450     }
451 
452     iscsilun->allocmap_valid = bitmap_try_new(iscsilun->allocmap_size);
453     if (!iscsilun->allocmap_valid) {
454         /* if we are under memory pressure free the allocmap as well */
455         iscsi_allocmap_free(iscsilun);
456         return -ENOMEM;
457     }
458 
459     return 0;
460 }
461 
462 static void
463 iscsi_allocmap_update(IscsiLun *iscsilun, int64_t offset,
464                       int64_t bytes, bool allocated, bool valid)
465 {
466     int64_t cl_num_expanded, nb_cls_expanded, cl_num_shrunk, nb_cls_shrunk;
467 
468     if (iscsilun->allocmap == NULL) {
469         return;
470     }
471     /* expand to entirely contain all affected clusters */
472     assert(iscsilun->cluster_size);
473     cl_num_expanded = offset / iscsilun->cluster_size;
474     nb_cls_expanded = DIV_ROUND_UP(offset + bytes,
475                                    iscsilun->cluster_size) - cl_num_expanded;
476     /* shrink to touch only completely contained clusters */
477     cl_num_shrunk = DIV_ROUND_UP(offset, iscsilun->cluster_size);
478     nb_cls_shrunk = (offset + bytes) / iscsilun->cluster_size - cl_num_shrunk;
479     if (allocated) {
480         bitmap_set(iscsilun->allocmap, cl_num_expanded, nb_cls_expanded);
481     } else {
482         if (nb_cls_shrunk > 0) {
483             bitmap_clear(iscsilun->allocmap, cl_num_shrunk, nb_cls_shrunk);
484         }
485     }
486 
487     if (iscsilun->allocmap_valid == NULL) {
488         return;
489     }
490     if (valid) {
491         if (nb_cls_shrunk > 0) {
492             bitmap_set(iscsilun->allocmap_valid, cl_num_shrunk, nb_cls_shrunk);
493         }
494     } else {
495         bitmap_clear(iscsilun->allocmap_valid, cl_num_expanded,
496                      nb_cls_expanded);
497     }
498 }
499 
500 static void
501 iscsi_allocmap_set_allocated(IscsiLun *iscsilun, int64_t offset,
502                              int64_t bytes)
503 {
504     iscsi_allocmap_update(iscsilun, offset, bytes, true, true);
505 }
506 
507 static void
508 iscsi_allocmap_set_unallocated(IscsiLun *iscsilun, int64_t offset,
509                                int64_t bytes)
510 {
511     /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update
512      * is ignored, so this will in effect be an iscsi_allocmap_set_invalid.
513      */
514     iscsi_allocmap_update(iscsilun, offset, bytes, false, true);
515 }
516 
517 static void iscsi_allocmap_set_invalid(IscsiLun *iscsilun, int64_t offset,
518                                        int64_t bytes)
519 {
520     iscsi_allocmap_update(iscsilun, offset, bytes, false, false);
521 }
522 
523 static void iscsi_allocmap_invalidate(IscsiLun *iscsilun)
524 {
525     if (iscsilun->allocmap) {
526         bitmap_zero(iscsilun->allocmap, iscsilun->allocmap_size);
527     }
528     if (iscsilun->allocmap_valid) {
529         bitmap_zero(iscsilun->allocmap_valid, iscsilun->allocmap_size);
530     }
531 }
532 
533 static inline bool
534 iscsi_allocmap_is_allocated(IscsiLun *iscsilun, int64_t offset,
535                             int64_t bytes)
536 {
537     unsigned long size;
538     if (iscsilun->allocmap == NULL) {
539         return true;
540     }
541     assert(iscsilun->cluster_size);
542     size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size);
543     return !(find_next_bit(iscsilun->allocmap, size,
544                            offset / iscsilun->cluster_size) == size);
545 }
546 
547 static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun,
548                                            int64_t offset, int64_t bytes)
549 {
550     unsigned long size;
551     if (iscsilun->allocmap_valid == NULL) {
552         return false;
553     }
554     assert(iscsilun->cluster_size);
555     size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size);
556     return (find_next_zero_bit(iscsilun->allocmap_valid, size,
557                                offset / iscsilun->cluster_size) == size);
558 }
559 
560 static void coroutine_fn iscsi_co_wait_for_task(IscsiTask *iTask,
561                                                 IscsiLun *iscsilun)
562 {
563     while (!iTask->complete) {
564         iscsi_set_events(iscsilun);
565         qemu_mutex_unlock(&iscsilun->mutex);
566         qemu_coroutine_yield();
567         qemu_mutex_lock(&iscsilun->mutex);
568     }
569 }
570 
571 static int coroutine_fn
572 iscsi_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
573                 QEMUIOVector *iov, int flags)
574 {
575     IscsiLun *iscsilun = bs->opaque;
576     struct IscsiTask iTask;
577     uint64_t lba;
578     uint32_t num_sectors;
579     bool fua = flags & BDRV_REQ_FUA;
580     int r = 0;
581 
582     if (fua) {
583         assert(iscsilun->dpofua);
584     }
585     if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
586         return -EINVAL;
587     }
588 
589     if (bs->bl.max_transfer) {
590         assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
591     }
592 
593     lba = sector_qemu2lun(sector_num, iscsilun);
594     num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
595     iscsi_co_init_iscsitask(iscsilun, &iTask);
596     qemu_mutex_lock(&iscsilun->mutex);
597 retry:
598     if (iscsilun->use_16_for_rw) {
599 #if LIBISCSI_API_VERSION >= (20160603)
600         iTask.task = iscsi_write16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
601                                             NULL, num_sectors * iscsilun->block_size,
602                                             iscsilun->block_size, 0, 0, fua, 0, 0,
603                                             iscsi_co_generic_cb, &iTask,
604                                             (struct scsi_iovec *)iov->iov, iov->niov);
605     } else {
606         iTask.task = iscsi_write10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
607                                             NULL, num_sectors * iscsilun->block_size,
608                                             iscsilun->block_size, 0, 0, fua, 0, 0,
609                                             iscsi_co_generic_cb, &iTask,
610                                             (struct scsi_iovec *)iov->iov, iov->niov);
611     }
612 #else
613         iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
614                                         NULL, num_sectors * iscsilun->block_size,
615                                         iscsilun->block_size, 0, 0, fua, 0, 0,
616                                         iscsi_co_generic_cb, &iTask);
617     } else {
618         iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
619                                         NULL, num_sectors * iscsilun->block_size,
620                                         iscsilun->block_size, 0, 0, fua, 0, 0,
621                                         iscsi_co_generic_cb, &iTask);
622     }
623 #endif
624     if (iTask.task == NULL) {
625         qemu_mutex_unlock(&iscsilun->mutex);
626         return -ENOMEM;
627     }
628 #if LIBISCSI_API_VERSION < (20160603)
629     scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
630                           iov->niov);
631 #endif
632     iscsi_co_wait_for_task(&iTask, iscsilun);
633 
634     if (iTask.task != NULL) {
635         scsi_free_scsi_task(iTask.task);
636         iTask.task = NULL;
637     }
638 
639     if (iTask.do_retry) {
640         iTask.complete = 0;
641         goto retry;
642     }
643 
644     if (iTask.status != SCSI_STATUS_GOOD) {
645         iscsi_allocmap_set_invalid(iscsilun, sector_num * BDRV_SECTOR_SIZE,
646                                    nb_sectors * BDRV_SECTOR_SIZE);
647         error_report("iSCSI WRITE10/16 failed at lba %" PRIu64 ": %s", lba,
648                      iTask.err_str);
649         r = iTask.err_code;
650         goto out_unlock;
651     }
652 
653     iscsi_allocmap_set_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE,
654                                  nb_sectors * BDRV_SECTOR_SIZE);
655 
656 out_unlock:
657     qemu_mutex_unlock(&iscsilun->mutex);
658     g_free(iTask.err_str);
659     return r;
660 }
661 
662 
663 
664 static int coroutine_fn iscsi_co_block_status(BlockDriverState *bs,
665                                               bool want_zero, int64_t offset,
666                                               int64_t bytes, int64_t *pnum,
667                                               int64_t *map,
668                                               BlockDriverState **file)
669 {
670     IscsiLun *iscsilun = bs->opaque;
671     struct scsi_get_lba_status *lbas = NULL;
672     struct scsi_lba_status_descriptor *lbasd = NULL;
673     struct IscsiTask iTask;
674     uint64_t lba;
675     int ret;
676 
677     iscsi_co_init_iscsitask(iscsilun, &iTask);
678 
679     assert(QEMU_IS_ALIGNED(offset | bytes, iscsilun->block_size));
680 
681     /* default to all sectors allocated */
682     ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
683     if (map) {
684         *map = offset;
685     }
686     *pnum = bytes;
687 
688     /* LUN does not support logical block provisioning */
689     if (!iscsilun->lbpme) {
690         goto out;
691     }
692 
693     lba = offset / iscsilun->block_size;
694 
695     qemu_mutex_lock(&iscsilun->mutex);
696 retry:
697     if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
698                                   lba, 8 + 16, iscsi_co_generic_cb,
699                                   &iTask) == NULL) {
700         ret = -ENOMEM;
701         goto out_unlock;
702     }
703     iscsi_co_wait_for_task(&iTask, iscsilun);
704 
705     if (iTask.do_retry) {
706         if (iTask.task != NULL) {
707             scsi_free_scsi_task(iTask.task);
708             iTask.task = NULL;
709         }
710         iTask.complete = 0;
711         goto retry;
712     }
713 
714     if (iTask.status != SCSI_STATUS_GOOD) {
715         /* in case the get_lba_status_callout fails (i.e.
716          * because the device is busy or the cmd is not
717          * supported) we pretend all blocks are allocated
718          * for backwards compatibility */
719         error_report("iSCSI GET_LBA_STATUS failed at lba %" PRIu64 ": %s",
720                      lba, iTask.err_str);
721         goto out_unlock;
722     }
723 
724     lbas = scsi_datain_unmarshall(iTask.task);
725     if (lbas == NULL) {
726         ret = -EIO;
727         goto out_unlock;
728     }
729 
730     lbasd = &lbas->descriptors[0];
731 
732     if (lba != lbasd->lba) {
733         ret = -EIO;
734         goto out_unlock;
735     }
736 
737     *pnum = lbasd->num_blocks * iscsilun->block_size;
738 
739     if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
740         lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
741         ret &= ~BDRV_BLOCK_DATA;
742         if (iscsilun->lbprz) {
743             ret |= BDRV_BLOCK_ZERO;
744         }
745     }
746 
747     if (ret & BDRV_BLOCK_ZERO) {
748         iscsi_allocmap_set_unallocated(iscsilun, offset, *pnum);
749     } else {
750         iscsi_allocmap_set_allocated(iscsilun, offset, *pnum);
751     }
752 
753     if (*pnum > bytes) {
754         *pnum = bytes;
755     }
756 out_unlock:
757     qemu_mutex_unlock(&iscsilun->mutex);
758     g_free(iTask.err_str);
759 out:
760     if (iTask.task != NULL) {
761         scsi_free_scsi_task(iTask.task);
762     }
763     if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID && file) {
764         *file = bs;
765     }
766     return ret;
767 }
768 
769 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
770                                        int64_t sector_num, int nb_sectors,
771                                        QEMUIOVector *iov)
772 {
773     IscsiLun *iscsilun = bs->opaque;
774     struct IscsiTask iTask;
775     uint64_t lba;
776     uint32_t num_sectors;
777     int r = 0;
778 
779     if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
780         return -EINVAL;
781     }
782 
783     if (bs->bl.max_transfer) {
784         assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
785     }
786 
787     /* if cache.direct is off and we have a valid entry in our allocation map
788      * we can skip checking the block status and directly return zeroes if
789      * the request falls within an unallocated area */
790     if (iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE,
791                                 nb_sectors * BDRV_SECTOR_SIZE) &&
792         !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE,
793                                      nb_sectors * BDRV_SECTOR_SIZE)) {
794             qemu_iovec_memset(iov, 0, 0x00, iov->size);
795             return 0;
796     }
797 
798     if (nb_sectors >= ISCSI_CHECKALLOC_THRES &&
799         !iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE,
800                                  nb_sectors * BDRV_SECTOR_SIZE) &&
801         !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE,
802                                      nb_sectors * BDRV_SECTOR_SIZE)) {
803         int64_t pnum;
804         /* check the block status from the beginning of the cluster
805          * containing the start sector */
806         int64_t head;
807         int ret;
808 
809         assert(iscsilun->cluster_size);
810         head = (sector_num * BDRV_SECTOR_SIZE) % iscsilun->cluster_size;
811         ret = iscsi_co_block_status(bs, true,
812                                     sector_num * BDRV_SECTOR_SIZE - head,
813                                     BDRV_REQUEST_MAX_BYTES, &pnum, NULL, NULL);
814         if (ret < 0) {
815             return ret;
816         }
817         /* if the whole request falls into an unallocated area we can avoid
818          * reading and directly return zeroes instead */
819         if (ret & BDRV_BLOCK_ZERO &&
820             pnum >= nb_sectors * BDRV_SECTOR_SIZE + head) {
821             qemu_iovec_memset(iov, 0, 0x00, iov->size);
822             return 0;
823         }
824     }
825 
826     lba = sector_qemu2lun(sector_num, iscsilun);
827     num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
828 
829     iscsi_co_init_iscsitask(iscsilun, &iTask);
830     qemu_mutex_lock(&iscsilun->mutex);
831 retry:
832     if (iscsilun->use_16_for_rw) {
833 #if LIBISCSI_API_VERSION >= (20160603)
834         iTask.task = iscsi_read16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
835                                            num_sectors * iscsilun->block_size,
836                                            iscsilun->block_size, 0, 0, 0, 0, 0,
837                                            iscsi_co_generic_cb, &iTask,
838                                            (struct scsi_iovec *)iov->iov, iov->niov);
839     } else {
840         iTask.task = iscsi_read10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
841                                            num_sectors * iscsilun->block_size,
842                                            iscsilun->block_size,
843                                            0, 0, 0, 0, 0,
844                                            iscsi_co_generic_cb, &iTask,
845                                            (struct scsi_iovec *)iov->iov, iov->niov);
846     }
847 #else
848         iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
849                                        num_sectors * iscsilun->block_size,
850                                        iscsilun->block_size, 0, 0, 0, 0, 0,
851                                        iscsi_co_generic_cb, &iTask);
852     } else {
853         iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
854                                        num_sectors * iscsilun->block_size,
855                                        iscsilun->block_size,
856                                        0, 0, 0, 0, 0,
857                                        iscsi_co_generic_cb, &iTask);
858     }
859 #endif
860     if (iTask.task == NULL) {
861         qemu_mutex_unlock(&iscsilun->mutex);
862         return -ENOMEM;
863     }
864 #if LIBISCSI_API_VERSION < (20160603)
865     scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
866 #endif
867 
868     iscsi_co_wait_for_task(&iTask, iscsilun);
869     if (iTask.task != NULL) {
870         scsi_free_scsi_task(iTask.task);
871         iTask.task = NULL;
872     }
873 
874     if (iTask.do_retry) {
875         iTask.complete = 0;
876         goto retry;
877     }
878 
879     if (iTask.status != SCSI_STATUS_GOOD) {
880         error_report("iSCSI READ10/16 failed at lba %" PRIu64 ": %s",
881                      lba, iTask.err_str);
882         r = iTask.err_code;
883     }
884 
885     qemu_mutex_unlock(&iscsilun->mutex);
886     g_free(iTask.err_str);
887     return r;
888 }
889 
890 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
891 {
892     IscsiLun *iscsilun = bs->opaque;
893     struct IscsiTask iTask;
894     int r = 0;
895 
896     iscsi_co_init_iscsitask(iscsilun, &iTask);
897     qemu_mutex_lock(&iscsilun->mutex);
898 retry:
899     if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
900                                       0, iscsi_co_generic_cb, &iTask) == NULL) {
901         qemu_mutex_unlock(&iscsilun->mutex);
902         return -ENOMEM;
903     }
904 
905     iscsi_co_wait_for_task(&iTask, iscsilun);
906 
907     if (iTask.task != NULL) {
908         scsi_free_scsi_task(iTask.task);
909         iTask.task = NULL;
910     }
911 
912     if (iTask.do_retry) {
913         iTask.complete = 0;
914         goto retry;
915     }
916 
917     if (iTask.status != SCSI_STATUS_GOOD) {
918         error_report("iSCSI SYNCHRONIZECACHE10 failed: %s", iTask.err_str);
919         r = iTask.err_code;
920     }
921 
922     qemu_mutex_unlock(&iscsilun->mutex);
923     g_free(iTask.err_str);
924     return r;
925 }
926 
927 #ifdef __linux__
928 /* Called (via iscsi_service) with QemuMutex held.  */
929 static void
930 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
931                      void *command_data, void *opaque)
932 {
933     IscsiAIOCB *acb = opaque;
934 
935     g_free(acb->buf);
936     acb->buf = NULL;
937 
938     acb->status = 0;
939     if (status < 0) {
940         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
941                      iscsi_get_error(iscsi));
942         acb->status = iscsi_translate_sense(&acb->task->sense);
943     }
944 
945     acb->ioh->driver_status = 0;
946     acb->ioh->host_status   = 0;
947     acb->ioh->resid         = 0;
948     acb->ioh->status        = status;
949 
950 #define SG_ERR_DRIVER_SENSE    0x08
951 
952     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
953         int ss;
954 
955         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
956 
957         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
958         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
959              acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
960         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
961     }
962 
963     iscsi_schedule_bh(acb);
964 }
965 
966 static void iscsi_ioctl_bh_completion(void *opaque)
967 {
968     IscsiAIOCB *acb = opaque;
969 
970     qemu_bh_delete(acb->bh);
971     acb->common.cb(acb->common.opaque, acb->ret);
972     qemu_aio_unref(acb);
973 }
974 
975 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
976 {
977     BlockDriverState *bs = acb->common.bs;
978     IscsiLun *iscsilun = bs->opaque;
979     int ret = 0;
980 
981     switch (req) {
982     case SG_GET_VERSION_NUM:
983         *(int *)buf = 30000;
984         break;
985     case SG_GET_SCSI_ID:
986         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
987         break;
988     default:
989         ret = -EINVAL;
990     }
991     assert(!acb->bh);
992     acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
993                          iscsi_ioctl_bh_completion, acb);
994     acb->ret = ret;
995     qemu_bh_schedule(acb->bh);
996 }
997 
998 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
999         unsigned long int req, void *buf,
1000         BlockCompletionFunc *cb, void *opaque)
1001 {
1002     IscsiLun *iscsilun = bs->opaque;
1003     struct iscsi_context *iscsi = iscsilun->iscsi;
1004     struct iscsi_data data;
1005     IscsiAIOCB *acb;
1006 
1007     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
1008 
1009     acb->iscsilun = iscsilun;
1010     acb->bh          = NULL;
1011     acb->status      = -EINPROGRESS;
1012     acb->buf         = NULL;
1013     acb->ioh         = buf;
1014 
1015     if (req != SG_IO) {
1016         iscsi_ioctl_handle_emulated(acb, req, buf);
1017         return &acb->common;
1018     }
1019 
1020     if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) {
1021         error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)",
1022                      acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE);
1023         qemu_aio_unref(acb);
1024         return NULL;
1025     }
1026 
1027     acb->task = malloc(sizeof(struct scsi_task));
1028     if (acb->task == NULL) {
1029         error_report("iSCSI: Failed to allocate task for scsi command. %s",
1030                      iscsi_get_error(iscsi));
1031         qemu_aio_unref(acb);
1032         return NULL;
1033     }
1034     memset(acb->task, 0, sizeof(struct scsi_task));
1035 
1036     switch (acb->ioh->dxfer_direction) {
1037     case SG_DXFER_TO_DEV:
1038         acb->task->xfer_dir = SCSI_XFER_WRITE;
1039         break;
1040     case SG_DXFER_FROM_DEV:
1041         acb->task->xfer_dir = SCSI_XFER_READ;
1042         break;
1043     default:
1044         acb->task->xfer_dir = SCSI_XFER_NONE;
1045         break;
1046     }
1047 
1048     acb->task->cdb_size = acb->ioh->cmd_len;
1049     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
1050     acb->task->expxferlen = acb->ioh->dxfer_len;
1051 
1052     data.size = 0;
1053     qemu_mutex_lock(&iscsilun->mutex);
1054     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
1055         if (acb->ioh->iovec_count == 0) {
1056             data.data = acb->ioh->dxferp;
1057             data.size = acb->ioh->dxfer_len;
1058         } else {
1059             scsi_task_set_iov_out(acb->task,
1060                                  (struct scsi_iovec *) acb->ioh->dxferp,
1061                                  acb->ioh->iovec_count);
1062         }
1063     }
1064 
1065     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
1066                                  iscsi_aio_ioctl_cb,
1067                                  (data.size > 0) ? &data : NULL,
1068                                  acb) != 0) {
1069         qemu_mutex_unlock(&iscsilun->mutex);
1070         scsi_free_scsi_task(acb->task);
1071         qemu_aio_unref(acb);
1072         return NULL;
1073     }
1074 
1075     /* tell libiscsi to read straight into the buffer we got from ioctl */
1076     if (acb->task->xfer_dir == SCSI_XFER_READ) {
1077         if (acb->ioh->iovec_count == 0) {
1078             scsi_task_add_data_in_buffer(acb->task,
1079                                          acb->ioh->dxfer_len,
1080                                          acb->ioh->dxferp);
1081         } else {
1082             scsi_task_set_iov_in(acb->task,
1083                                  (struct scsi_iovec *) acb->ioh->dxferp,
1084                                  acb->ioh->iovec_count);
1085         }
1086     }
1087 
1088     iscsi_set_events(iscsilun);
1089     qemu_mutex_unlock(&iscsilun->mutex);
1090 
1091     return &acb->common;
1092 }
1093 
1094 #endif
1095 
1096 static int64_t
1097 iscsi_getlength(BlockDriverState *bs)
1098 {
1099     IscsiLun *iscsilun = bs->opaque;
1100     int64_t len;
1101 
1102     len  = iscsilun->num_blocks;
1103     len *= iscsilun->block_size;
1104 
1105     return len;
1106 }
1107 
1108 static int
1109 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
1110 {
1111     IscsiLun *iscsilun = bs->opaque;
1112     struct IscsiTask iTask;
1113     struct unmap_list list;
1114     int r = 0;
1115 
1116     if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) {
1117         return -ENOTSUP;
1118     }
1119 
1120     if (!iscsilun->lbp.lbpu) {
1121         /* UNMAP is not supported by the target */
1122         return 0;
1123     }
1124 
1125     list.lba = offset / iscsilun->block_size;
1126     list.num = bytes / iscsilun->block_size;
1127 
1128     iscsi_co_init_iscsitask(iscsilun, &iTask);
1129     qemu_mutex_lock(&iscsilun->mutex);
1130 retry:
1131     if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
1132                          iscsi_co_generic_cb, &iTask) == NULL) {
1133         r = -ENOMEM;
1134         goto out_unlock;
1135     }
1136 
1137     iscsi_co_wait_for_task(&iTask, iscsilun);
1138 
1139     if (iTask.task != NULL) {
1140         scsi_free_scsi_task(iTask.task);
1141         iTask.task = NULL;
1142     }
1143 
1144     if (iTask.do_retry) {
1145         iTask.complete = 0;
1146         goto retry;
1147     }
1148 
1149     iscsi_allocmap_set_invalid(iscsilun, offset, bytes);
1150 
1151     if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
1152         /* the target might fail with a check condition if it
1153            is not happy with the alignment of the UNMAP request
1154            we silently fail in this case */
1155         goto out_unlock;
1156     }
1157 
1158     if (iTask.status != SCSI_STATUS_GOOD) {
1159         error_report("iSCSI UNMAP failed at lba %" PRIu64 ": %s",
1160                      list.lba, iTask.err_str);
1161         r = iTask.err_code;
1162         goto out_unlock;
1163     }
1164 
1165 out_unlock:
1166     qemu_mutex_unlock(&iscsilun->mutex);
1167     g_free(iTask.err_str);
1168     return r;
1169 }
1170 
1171 static int
1172 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1173                                     int bytes, BdrvRequestFlags flags)
1174 {
1175     IscsiLun *iscsilun = bs->opaque;
1176     struct IscsiTask iTask;
1177     uint64_t lba;
1178     uint32_t nb_blocks;
1179     bool use_16_for_ws = iscsilun->use_16_for_rw;
1180     int r = 0;
1181 
1182     if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) {
1183         return -ENOTSUP;
1184     }
1185 
1186     if (flags & BDRV_REQ_MAY_UNMAP) {
1187         if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
1188             /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
1189             use_16_for_ws = true;
1190         }
1191         if (use_16_for_ws && !iscsilun->lbp.lbpws) {
1192             /* WRITESAME16 with UNMAP is not supported by the target,
1193              * fall back and try WRITESAME10/16 without UNMAP */
1194             flags &= ~BDRV_REQ_MAY_UNMAP;
1195             use_16_for_ws = iscsilun->use_16_for_rw;
1196         }
1197     }
1198 
1199     if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1200         /* WRITESAME without UNMAP is not supported by the target */
1201         return -ENOTSUP;
1202     }
1203 
1204     lba = offset / iscsilun->block_size;
1205     nb_blocks = bytes / iscsilun->block_size;
1206 
1207     if (iscsilun->zeroblock == NULL) {
1208         iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1209         if (iscsilun->zeroblock == NULL) {
1210             return -ENOMEM;
1211         }
1212     }
1213 
1214     qemu_mutex_lock(&iscsilun->mutex);
1215     iscsi_co_init_iscsitask(iscsilun, &iTask);
1216 retry:
1217     if (use_16_for_ws) {
1218         iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1219                                             iscsilun->zeroblock, iscsilun->block_size,
1220                                             nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1221                                             0, 0, iscsi_co_generic_cb, &iTask);
1222     } else {
1223         iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1224                                             iscsilun->zeroblock, iscsilun->block_size,
1225                                             nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1226                                             0, 0, iscsi_co_generic_cb, &iTask);
1227     }
1228     if (iTask.task == NULL) {
1229         qemu_mutex_unlock(&iscsilun->mutex);
1230         return -ENOMEM;
1231     }
1232 
1233     iscsi_co_wait_for_task(&iTask, iscsilun);
1234 
1235     if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1236         iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1237         (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1238          iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1239         /* WRITE SAME is not supported by the target */
1240         iscsilun->has_write_same = false;
1241         scsi_free_scsi_task(iTask.task);
1242         r = -ENOTSUP;
1243         goto out_unlock;
1244     }
1245 
1246     if (iTask.task != NULL) {
1247         scsi_free_scsi_task(iTask.task);
1248         iTask.task = NULL;
1249     }
1250 
1251     if (iTask.do_retry) {
1252         iTask.complete = 0;
1253         goto retry;
1254     }
1255 
1256     if (iTask.status != SCSI_STATUS_GOOD) {
1257         iscsi_allocmap_set_invalid(iscsilun, offset, bytes);
1258         error_report("iSCSI WRITESAME10/16 failed at lba %" PRIu64 ": %s",
1259                      lba, iTask.err_str);
1260         r = iTask.err_code;
1261         goto out_unlock;
1262     }
1263 
1264     if (flags & BDRV_REQ_MAY_UNMAP) {
1265         iscsi_allocmap_set_invalid(iscsilun, offset, bytes);
1266     } else {
1267         iscsi_allocmap_set_allocated(iscsilun, offset, bytes);
1268     }
1269 
1270 out_unlock:
1271     qemu_mutex_unlock(&iscsilun->mutex);
1272     g_free(iTask.err_str);
1273     return r;
1274 }
1275 
1276 static void apply_chap(struct iscsi_context *iscsi, QemuOpts *opts,
1277                        Error **errp)
1278 {
1279     const char *user = NULL;
1280     const char *password = NULL;
1281     const char *secretid;
1282     char *secret = NULL;
1283 
1284     user = qemu_opt_get(opts, "user");
1285     if (!user) {
1286         return;
1287     }
1288 
1289     secretid = qemu_opt_get(opts, "password-secret");
1290     password = qemu_opt_get(opts, "password");
1291     if (secretid && password) {
1292         error_setg(errp, "'password' and 'password-secret' properties are "
1293                    "mutually exclusive");
1294         return;
1295     }
1296     if (secretid) {
1297         secret = qcrypto_secret_lookup_as_utf8(secretid, errp);
1298         if (!secret) {
1299             return;
1300         }
1301         password = secret;
1302     } else if (!password) {
1303         error_setg(errp, "CHAP username specified but no password was given");
1304         return;
1305     }
1306 
1307     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1308         error_setg(errp, "Failed to set initiator username and password");
1309     }
1310 
1311     g_free(secret);
1312 }
1313 
1314 static void apply_header_digest(struct iscsi_context *iscsi, QemuOpts *opts,
1315                                 Error **errp)
1316 {
1317     const char *digest = NULL;
1318 
1319     digest = qemu_opt_get(opts, "header-digest");
1320     if (!digest) {
1321         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1322     } else if (!strcmp(digest, "crc32c")) {
1323         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1324     } else if (!strcmp(digest, "none")) {
1325         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1326     } else if (!strcmp(digest, "crc32c-none")) {
1327         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1328     } else if (!strcmp(digest, "none-crc32c")) {
1329         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1330     } else {
1331         error_setg(errp, "Invalid header-digest setting : %s", digest);
1332     }
1333 }
1334 
1335 static char *get_initiator_name(QemuOpts *opts)
1336 {
1337     const char *name;
1338     char *iscsi_name;
1339     UuidInfo *uuid_info;
1340 
1341     name = qemu_opt_get(opts, "initiator-name");
1342     if (name) {
1343         return g_strdup(name);
1344     }
1345 
1346     uuid_info = qmp_query_uuid(NULL);
1347     if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1348         name = qemu_get_vm_name();
1349     } else {
1350         name = uuid_info->UUID;
1351     }
1352     iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1353                                  name ? ":" : "", name ? name : "");
1354     qapi_free_UuidInfo(uuid_info);
1355     return iscsi_name;
1356 }
1357 
1358 static void iscsi_nop_timed_event(void *opaque)
1359 {
1360     IscsiLun *iscsilun = opaque;
1361 
1362     qemu_mutex_lock(&iscsilun->mutex);
1363     if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1364         error_report("iSCSI: NOP timeout. Reconnecting...");
1365         iscsilun->request_timed_out = true;
1366     } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1367         error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1368         goto out;
1369     }
1370 
1371     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1372     iscsi_set_events(iscsilun);
1373 
1374 out:
1375     qemu_mutex_unlock(&iscsilun->mutex);
1376 }
1377 
1378 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1379 {
1380     struct scsi_task *task = NULL;
1381     struct scsi_readcapacity10 *rc10 = NULL;
1382     struct scsi_readcapacity16 *rc16 = NULL;
1383     int retries = ISCSI_CMD_RETRIES;
1384 
1385     do {
1386         if (task != NULL) {
1387             scsi_free_scsi_task(task);
1388             task = NULL;
1389         }
1390 
1391         switch (iscsilun->type) {
1392         case TYPE_DISK:
1393             task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1394             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1395                 rc16 = scsi_datain_unmarshall(task);
1396                 if (rc16 == NULL) {
1397                     error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1398                 } else {
1399                     iscsilun->block_size = rc16->block_length;
1400                     iscsilun->num_blocks = rc16->returned_lba + 1;
1401                     iscsilun->lbpme = !!rc16->lbpme;
1402                     iscsilun->lbprz = !!rc16->lbprz;
1403                     iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1404                 }
1405                 break;
1406             }
1407             if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1408                 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1409                 break;
1410             }
1411             /* Fall through and try READ CAPACITY(10) instead.  */
1412         case TYPE_ROM:
1413             task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1414             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1415                 rc10 = scsi_datain_unmarshall(task);
1416                 if (rc10 == NULL) {
1417                     error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1418                 } else {
1419                     iscsilun->block_size = rc10->block_size;
1420                     if (rc10->lba == 0) {
1421                         /* blank disk loaded */
1422                         iscsilun->num_blocks = 0;
1423                     } else {
1424                         iscsilun->num_blocks = rc10->lba + 1;
1425                     }
1426                 }
1427             }
1428             break;
1429         default:
1430             return;
1431         }
1432     } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1433              && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1434              && retries-- > 0);
1435 
1436     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1437         error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1438     } else if (!iscsilun->block_size ||
1439                iscsilun->block_size % BDRV_SECTOR_SIZE) {
1440         error_setg(errp, "iSCSI: the target returned an invalid "
1441                    "block size of %d.", iscsilun->block_size);
1442     }
1443     if (task) {
1444         scsi_free_scsi_task(task);
1445     }
1446 }
1447 
1448 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1449                                           int evpd, int pc, void **inq, Error **errp)
1450 {
1451     int full_size;
1452     struct scsi_task *task = NULL;
1453     task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1454     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1455         goto fail;
1456     }
1457     full_size = scsi_datain_getfullsize(task);
1458     if (full_size > task->datain.size) {
1459         scsi_free_scsi_task(task);
1460 
1461         /* we need more data for the full list */
1462         task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1463         if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1464             goto fail;
1465         }
1466     }
1467 
1468     *inq = scsi_datain_unmarshall(task);
1469     if (*inq == NULL) {
1470         error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1471         goto fail_with_err;
1472     }
1473 
1474     return task;
1475 
1476 fail:
1477     error_setg(errp, "iSCSI: Inquiry command failed : %s",
1478                iscsi_get_error(iscsi));
1479 fail_with_err:
1480     if (task != NULL) {
1481         scsi_free_scsi_task(task);
1482     }
1483     return NULL;
1484 }
1485 
1486 static void iscsi_detach_aio_context(BlockDriverState *bs)
1487 {
1488     IscsiLun *iscsilun = bs->opaque;
1489 
1490     aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1491                        false, NULL, NULL, NULL, NULL);
1492     iscsilun->events = 0;
1493 
1494     if (iscsilun->nop_timer) {
1495         timer_del(iscsilun->nop_timer);
1496         timer_free(iscsilun->nop_timer);
1497         iscsilun->nop_timer = NULL;
1498     }
1499     if (iscsilun->event_timer) {
1500         timer_del(iscsilun->event_timer);
1501         timer_free(iscsilun->event_timer);
1502         iscsilun->event_timer = NULL;
1503     }
1504 }
1505 
1506 static void iscsi_attach_aio_context(BlockDriverState *bs,
1507                                      AioContext *new_context)
1508 {
1509     IscsiLun *iscsilun = bs->opaque;
1510 
1511     iscsilun->aio_context = new_context;
1512     iscsi_set_events(iscsilun);
1513 
1514     /* Set up a timer for sending out iSCSI NOPs */
1515     iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1516                                         QEMU_CLOCK_REALTIME, SCALE_MS,
1517                                         iscsi_nop_timed_event, iscsilun);
1518     timer_mod(iscsilun->nop_timer,
1519               qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1520 
1521     /* Set up a timer for periodic calls to iscsi_set_events and to
1522      * scan for command timeout */
1523     iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1524                                           QEMU_CLOCK_REALTIME, SCALE_MS,
1525                                           iscsi_timed_check_events, iscsilun);
1526     timer_mod(iscsilun->event_timer,
1527               qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1528 }
1529 
1530 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1531 {
1532     struct scsi_task *task;
1533     struct scsi_mode_sense *ms = NULL;
1534     iscsilun->write_protected = false;
1535     iscsilun->dpofua = false;
1536 
1537     task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1538                                  1, SCSI_MODESENSE_PC_CURRENT,
1539                                  0x3F, 0, 255);
1540     if (task == NULL) {
1541         error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1542                      iscsi_get_error(iscsilun->iscsi));
1543         goto out;
1544     }
1545 
1546     if (task->status != SCSI_STATUS_GOOD) {
1547         error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1548         goto out;
1549     }
1550     ms = scsi_datain_unmarshall(task);
1551     if (!ms) {
1552         error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1553                      iscsi_get_error(iscsilun->iscsi));
1554         goto out;
1555     }
1556     iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1557     iscsilun->dpofua          = ms->device_specific_parameter & 0x10;
1558 
1559 out:
1560     if (task) {
1561         scsi_free_scsi_task(task);
1562     }
1563 }
1564 
1565 static void iscsi_parse_iscsi_option(const char *target, QDict *options)
1566 {
1567     QemuOptsList *list;
1568     QemuOpts *opts;
1569     const char *user, *password, *password_secret, *initiator_name,
1570                *header_digest, *timeout;
1571 
1572     list = qemu_find_opts("iscsi");
1573     if (!list) {
1574         return;
1575     }
1576 
1577     opts = qemu_opts_find(list, target);
1578     if (opts == NULL) {
1579         opts = QTAILQ_FIRST(&list->head);
1580         if (!opts) {
1581             return;
1582         }
1583     }
1584 
1585     user = qemu_opt_get(opts, "user");
1586     if (user) {
1587         qdict_set_default_str(options, "user", user);
1588     }
1589 
1590     password = qemu_opt_get(opts, "password");
1591     if (password) {
1592         qdict_set_default_str(options, "password", password);
1593     }
1594 
1595     password_secret = qemu_opt_get(opts, "password-secret");
1596     if (password_secret) {
1597         qdict_set_default_str(options, "password-secret", password_secret);
1598     }
1599 
1600     initiator_name = qemu_opt_get(opts, "initiator-name");
1601     if (initiator_name) {
1602         qdict_set_default_str(options, "initiator-name", initiator_name);
1603     }
1604 
1605     header_digest = qemu_opt_get(opts, "header-digest");
1606     if (header_digest) {
1607         /* -iscsi takes upper case values, but QAPI only supports lower case
1608          * enum constant names, so we have to convert here. */
1609         char *qapi_value = g_ascii_strdown(header_digest, -1);
1610         qdict_set_default_str(options, "header-digest", qapi_value);
1611         g_free(qapi_value);
1612     }
1613 
1614     timeout = qemu_opt_get(opts, "timeout");
1615     if (timeout) {
1616         qdict_set_default_str(options, "timeout", timeout);
1617     }
1618 }
1619 
1620 /*
1621  * We support iscsi url's on the form
1622  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1623  */
1624 static void iscsi_parse_filename(const char *filename, QDict *options,
1625                                  Error **errp)
1626 {
1627     struct iscsi_url *iscsi_url;
1628     const char *transport_name;
1629     char *lun_str;
1630 
1631     iscsi_url = iscsi_parse_full_url(NULL, filename);
1632     if (iscsi_url == NULL) {
1633         error_setg(errp, "Failed to parse URL : %s", filename);
1634         return;
1635     }
1636 
1637 #if LIBISCSI_API_VERSION >= (20160603)
1638     switch (iscsi_url->transport) {
1639     case TCP_TRANSPORT:
1640         transport_name = "tcp";
1641         break;
1642     case ISER_TRANSPORT:
1643         transport_name = "iser";
1644         break;
1645     default:
1646         error_setg(errp, "Unknown transport type (%d)",
1647                    iscsi_url->transport);
1648         return;
1649     }
1650 #else
1651     transport_name = "tcp";
1652 #endif
1653 
1654     qdict_set_default_str(options, "transport", transport_name);
1655     qdict_set_default_str(options, "portal", iscsi_url->portal);
1656     qdict_set_default_str(options, "target", iscsi_url->target);
1657 
1658     lun_str = g_strdup_printf("%d", iscsi_url->lun);
1659     qdict_set_default_str(options, "lun", lun_str);
1660     g_free(lun_str);
1661 
1662     /* User/password from -iscsi take precedence over those from the URL */
1663     iscsi_parse_iscsi_option(iscsi_url->target, options);
1664 
1665     if (iscsi_url->user[0] != '\0') {
1666         qdict_set_default_str(options, "user", iscsi_url->user);
1667         qdict_set_default_str(options, "password", iscsi_url->passwd);
1668     }
1669 
1670     iscsi_destroy_url(iscsi_url);
1671 }
1672 
1673 static QemuOptsList runtime_opts = {
1674     .name = "iscsi",
1675     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1676     .desc = {
1677         {
1678             .name = "transport",
1679             .type = QEMU_OPT_STRING,
1680         },
1681         {
1682             .name = "portal",
1683             .type = QEMU_OPT_STRING,
1684         },
1685         {
1686             .name = "target",
1687             .type = QEMU_OPT_STRING,
1688         },
1689         {
1690             .name = "user",
1691             .type = QEMU_OPT_STRING,
1692         },
1693         {
1694             .name = "password",
1695             .type = QEMU_OPT_STRING,
1696         },
1697         {
1698             .name = "password-secret",
1699             .type = QEMU_OPT_STRING,
1700         },
1701         {
1702             .name = "lun",
1703             .type = QEMU_OPT_NUMBER,
1704         },
1705         {
1706             .name = "initiator-name",
1707             .type = QEMU_OPT_STRING,
1708         },
1709         {
1710             .name = "header-digest",
1711             .type = QEMU_OPT_STRING,
1712         },
1713         {
1714             .name = "timeout",
1715             .type = QEMU_OPT_NUMBER,
1716         },
1717         { /* end of list */ }
1718     },
1719 };
1720 
1721 static void iscsi_save_designator(IscsiLun *lun,
1722                                   struct scsi_inquiry_device_identification *inq_di)
1723 {
1724     struct scsi_inquiry_device_designator *desig, *copy = NULL;
1725 
1726     for (desig = inq_di->designators; desig; desig = desig->next) {
1727         if (desig->association ||
1728             desig->designator_type > SCSI_DESIGNATOR_TYPE_NAA) {
1729             continue;
1730         }
1731         /* NAA works better than T10 vendor ID based designator. */
1732         if (!copy || copy->designator_type < desig->designator_type) {
1733             copy = desig;
1734         }
1735     }
1736     if (copy) {
1737         lun->dd = g_new(struct scsi_inquiry_device_designator, 1);
1738         *lun->dd = *copy;
1739         lun->dd->next = NULL;
1740         lun->dd->designator = g_malloc(copy->designator_length);
1741         memcpy(lun->dd->designator, copy->designator, copy->designator_length);
1742     }
1743 }
1744 
1745 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1746                       Error **errp)
1747 {
1748     IscsiLun *iscsilun = bs->opaque;
1749     struct iscsi_context *iscsi = NULL;
1750     struct scsi_task *task = NULL;
1751     struct scsi_inquiry_standard *inq = NULL;
1752     struct scsi_inquiry_supported_pages *inq_vpd;
1753     char *initiator_name = NULL;
1754     QemuOpts *opts;
1755     Error *local_err = NULL;
1756     const char *transport_name, *portal, *target;
1757 #if LIBISCSI_API_VERSION >= (20160603)
1758     enum iscsi_transport_type transport;
1759 #endif
1760     int i, ret = 0, timeout = 0, lun;
1761 
1762     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1763     qemu_opts_absorb_qdict(opts, options, &local_err);
1764     if (local_err) {
1765         error_propagate(errp, local_err);
1766         ret = -EINVAL;
1767         goto out;
1768     }
1769 
1770     transport_name = qemu_opt_get(opts, "transport");
1771     portal = qemu_opt_get(opts, "portal");
1772     target = qemu_opt_get(opts, "target");
1773     lun = qemu_opt_get_number(opts, "lun", 0);
1774 
1775     if (!transport_name || !portal || !target) {
1776         error_setg(errp, "Need all of transport, portal and target options");
1777         ret = -EINVAL;
1778         goto out;
1779     }
1780 
1781     if (!strcmp(transport_name, "tcp")) {
1782 #if LIBISCSI_API_VERSION >= (20160603)
1783         transport = TCP_TRANSPORT;
1784     } else if (!strcmp(transport_name, "iser")) {
1785         transport = ISER_TRANSPORT;
1786 #else
1787         /* TCP is what older libiscsi versions always use */
1788 #endif
1789     } else {
1790         error_setg(errp, "Unknown transport: %s", transport_name);
1791         ret = -EINVAL;
1792         goto out;
1793     }
1794 
1795     memset(iscsilun, 0, sizeof(IscsiLun));
1796 
1797     initiator_name = get_initiator_name(opts);
1798 
1799     iscsi = iscsi_create_context(initiator_name);
1800     if (iscsi == NULL) {
1801         error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1802         ret = -ENOMEM;
1803         goto out;
1804     }
1805 #if LIBISCSI_API_VERSION >= (20160603)
1806     if (iscsi_init_transport(iscsi, transport)) {
1807         error_setg(errp, ("Error initializing transport."));
1808         ret = -EINVAL;
1809         goto out;
1810     }
1811 #endif
1812     if (iscsi_set_targetname(iscsi, target)) {
1813         error_setg(errp, "iSCSI: Failed to set target name.");
1814         ret = -EINVAL;
1815         goto out;
1816     }
1817 
1818     /* check if we got CHAP username/password via the options */
1819     apply_chap(iscsi, opts, &local_err);
1820     if (local_err != NULL) {
1821         error_propagate(errp, local_err);
1822         ret = -EINVAL;
1823         goto out;
1824     }
1825 
1826     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1827         error_setg(errp, "iSCSI: Failed to set session type to normal.");
1828         ret = -EINVAL;
1829         goto out;
1830     }
1831 
1832     /* check if we got HEADER_DIGEST via the options */
1833     apply_header_digest(iscsi, opts, &local_err);
1834     if (local_err != NULL) {
1835         error_propagate(errp, local_err);
1836         ret = -EINVAL;
1837         goto out;
1838     }
1839 
1840     /* timeout handling is broken in libiscsi before 1.15.0 */
1841     timeout = qemu_opt_get_number(opts, "timeout", 0);
1842 #if LIBISCSI_API_VERSION >= 20150621
1843     iscsi_set_timeout(iscsi, timeout);
1844 #else
1845     if (timeout) {
1846         error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1847     }
1848 #endif
1849 
1850     if (iscsi_full_connect_sync(iscsi, portal, lun) != 0) {
1851         error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1852             iscsi_get_error(iscsi));
1853         ret = -EINVAL;
1854         goto out;
1855     }
1856 
1857     iscsilun->iscsi = iscsi;
1858     iscsilun->aio_context = bdrv_get_aio_context(bs);
1859     iscsilun->lun = lun;
1860     iscsilun->has_write_same = true;
1861 
1862     task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1863                             (void **) &inq, errp);
1864     if (task == NULL) {
1865         ret = -EINVAL;
1866         goto out;
1867     }
1868     iscsilun->type = inq->periperal_device_type;
1869     scsi_free_scsi_task(task);
1870     task = NULL;
1871 
1872     iscsi_modesense_sync(iscsilun);
1873     if (iscsilun->dpofua) {
1874         bs->supported_write_flags = BDRV_REQ_FUA;
1875     }
1876 
1877     /* Check the write protect flag of the LUN if we want to write */
1878     if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1879         iscsilun->write_protected) {
1880         error_setg(errp, "Cannot open a write protected LUN as read-write");
1881         ret = -EACCES;
1882         goto out;
1883     }
1884 
1885     iscsi_readcapacity_sync(iscsilun, &local_err);
1886     if (local_err != NULL) {
1887         error_propagate(errp, local_err);
1888         ret = -EINVAL;
1889         goto out;
1890     }
1891     bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1892 
1893     /* We don't have any emulation for devices other than disks and CD-ROMs, so
1894      * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1895      * will try to read from the device to guess the image format.
1896      */
1897     if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1898         bs->sg = true;
1899     }
1900 
1901     task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1902                             SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1903                             (void **) &inq_vpd, errp);
1904     if (task == NULL) {
1905         ret = -EINVAL;
1906         goto out;
1907     }
1908     for (i = 0; i < inq_vpd->num_pages; i++) {
1909         struct scsi_task *inq_task;
1910         struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1911         struct scsi_inquiry_block_limits *inq_bl;
1912         struct scsi_inquiry_device_identification *inq_di;
1913         switch (inq_vpd->pages[i]) {
1914         case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1915             inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1916                                         SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1917                                         (void **) &inq_lbp, errp);
1918             if (inq_task == NULL) {
1919                 ret = -EINVAL;
1920                 goto out;
1921             }
1922             memcpy(&iscsilun->lbp, inq_lbp,
1923                    sizeof(struct scsi_inquiry_logical_block_provisioning));
1924             scsi_free_scsi_task(inq_task);
1925             break;
1926         case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1927             inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1928                                     SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1929                                     (void **) &inq_bl, errp);
1930             if (inq_task == NULL) {
1931                 ret = -EINVAL;
1932                 goto out;
1933             }
1934             memcpy(&iscsilun->bl, inq_bl,
1935                    sizeof(struct scsi_inquiry_block_limits));
1936             scsi_free_scsi_task(inq_task);
1937             break;
1938         case SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION:
1939             inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1940                                     SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION,
1941                                     (void **) &inq_di, errp);
1942             if (inq_task == NULL) {
1943                 ret = -EINVAL;
1944                 goto out;
1945             }
1946             iscsi_save_designator(iscsilun, inq_di);
1947             scsi_free_scsi_task(inq_task);
1948             break;
1949         default:
1950             break;
1951         }
1952     }
1953     scsi_free_scsi_task(task);
1954     task = NULL;
1955 
1956     qemu_mutex_init(&iscsilun->mutex);
1957     iscsi_attach_aio_context(bs, iscsilun->aio_context);
1958 
1959     /* Guess the internal cluster (page) size of the iscsi target by the means
1960      * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1961      * reasonable size */
1962     if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
1963         iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1964         iscsilun->cluster_size = iscsilun->bl.opt_unmap_gran *
1965             iscsilun->block_size;
1966         if (iscsilun->lbprz) {
1967             ret = iscsi_allocmap_init(iscsilun, bs->open_flags);
1968         }
1969     }
1970 
1971     if (iscsilun->lbprz && iscsilun->lbp.lbpws) {
1972         bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
1973     }
1974 
1975 out:
1976     qemu_opts_del(opts);
1977     g_free(initiator_name);
1978     if (task != NULL) {
1979         scsi_free_scsi_task(task);
1980     }
1981 
1982     if (ret) {
1983         if (iscsi != NULL) {
1984             if (iscsi_is_logged_in(iscsi)) {
1985                 iscsi_logout_sync(iscsi);
1986             }
1987             iscsi_destroy_context(iscsi);
1988         }
1989         memset(iscsilun, 0, sizeof(IscsiLun));
1990     }
1991 
1992     return ret;
1993 }
1994 
1995 static void iscsi_close(BlockDriverState *bs)
1996 {
1997     IscsiLun *iscsilun = bs->opaque;
1998     struct iscsi_context *iscsi = iscsilun->iscsi;
1999 
2000     iscsi_detach_aio_context(bs);
2001     if (iscsi_is_logged_in(iscsi)) {
2002         iscsi_logout_sync(iscsi);
2003     }
2004     iscsi_destroy_context(iscsi);
2005     if (iscsilun->dd) {
2006         g_free(iscsilun->dd->designator);
2007         g_free(iscsilun->dd);
2008     }
2009     g_free(iscsilun->zeroblock);
2010     iscsi_allocmap_free(iscsilun);
2011     qemu_mutex_destroy(&iscsilun->mutex);
2012     memset(iscsilun, 0, sizeof(IscsiLun));
2013 }
2014 
2015 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
2016 {
2017     /* We don't actually refresh here, but just return data queried in
2018      * iscsi_open(): iscsi targets don't change their limits. */
2019 
2020     IscsiLun *iscsilun = bs->opaque;
2021     uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
2022     unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size);
2023 
2024     assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bs->sg);
2025 
2026     bs->bl.request_alignment = block_size;
2027 
2028     if (iscsilun->bl.max_xfer_len) {
2029         max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
2030     }
2031 
2032     if (max_xfer_len * block_size < INT_MAX) {
2033         bs->bl.max_transfer = max_xfer_len * iscsilun->block_size;
2034     }
2035 
2036     if (iscsilun->lbp.lbpu) {
2037         if (iscsilun->bl.max_unmap < 0xffffffff / block_size) {
2038             bs->bl.max_pdiscard =
2039                 iscsilun->bl.max_unmap * iscsilun->block_size;
2040         }
2041         bs->bl.pdiscard_alignment =
2042             iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2043     } else {
2044         bs->bl.pdiscard_alignment = iscsilun->block_size;
2045     }
2046 
2047     if (iscsilun->bl.max_ws_len < 0xffffffff / block_size) {
2048         bs->bl.max_pwrite_zeroes =
2049             iscsilun->bl.max_ws_len * iscsilun->block_size;
2050     }
2051     if (iscsilun->lbp.lbpws) {
2052         bs->bl.pwrite_zeroes_alignment =
2053             iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2054     } else {
2055         bs->bl.pwrite_zeroes_alignment = iscsilun->block_size;
2056     }
2057     if (iscsilun->bl.opt_xfer_len &&
2058         iscsilun->bl.opt_xfer_len < INT_MAX / block_size) {
2059         bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len *
2060                                         iscsilun->block_size);
2061     }
2062 }
2063 
2064 /* Note that this will not re-establish a connection with an iSCSI target - it
2065  * is effectively a NOP.  */
2066 static int iscsi_reopen_prepare(BDRVReopenState *state,
2067                                 BlockReopenQueue *queue, Error **errp)
2068 {
2069     IscsiLun *iscsilun = state->bs->opaque;
2070 
2071     if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
2072         error_setg(errp, "Cannot open a write protected LUN as read-write");
2073         return -EACCES;
2074     }
2075     return 0;
2076 }
2077 
2078 static void iscsi_reopen_commit(BDRVReopenState *reopen_state)
2079 {
2080     IscsiLun *iscsilun = reopen_state->bs->opaque;
2081 
2082     /* the cache.direct status might have changed */
2083     if (iscsilun->allocmap != NULL) {
2084         iscsi_allocmap_init(iscsilun, reopen_state->flags);
2085     }
2086 }
2087 
2088 static int iscsi_truncate(BlockDriverState *bs, int64_t offset,
2089                           PreallocMode prealloc, Error **errp)
2090 {
2091     IscsiLun *iscsilun = bs->opaque;
2092     Error *local_err = NULL;
2093 
2094     if (prealloc != PREALLOC_MODE_OFF) {
2095         error_setg(errp, "Unsupported preallocation mode '%s'",
2096                    PreallocMode_str(prealloc));
2097         return -ENOTSUP;
2098     }
2099 
2100     if (iscsilun->type != TYPE_DISK) {
2101         error_setg(errp, "Cannot resize non-disk iSCSI devices");
2102         return -ENOTSUP;
2103     }
2104 
2105     iscsi_readcapacity_sync(iscsilun, &local_err);
2106     if (local_err != NULL) {
2107         error_propagate(errp, local_err);
2108         return -EIO;
2109     }
2110 
2111     if (offset > iscsi_getlength(bs)) {
2112         error_setg(errp, "Cannot grow iSCSI devices");
2113         return -EINVAL;
2114     }
2115 
2116     if (iscsilun->allocmap != NULL) {
2117         iscsi_allocmap_init(iscsilun, bs->open_flags);
2118     }
2119 
2120     return 0;
2121 }
2122 
2123 static int coroutine_fn iscsi_co_create_opts(const char *filename, QemuOpts *opts,
2124                                              Error **errp)
2125 {
2126     int ret = 0;
2127     int64_t total_size = 0;
2128     BlockDriverState *bs;
2129     IscsiLun *iscsilun = NULL;
2130     QDict *bs_options;
2131     Error *local_err = NULL;
2132 
2133     bs = bdrv_new();
2134 
2135     /* Read out options */
2136     total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2137                               BDRV_SECTOR_SIZE);
2138     bs->opaque = g_new0(struct IscsiLun, 1);
2139     iscsilun = bs->opaque;
2140 
2141     bs_options = qdict_new();
2142     iscsi_parse_filename(filename, bs_options, &local_err);
2143     if (local_err) {
2144         error_propagate(errp, local_err);
2145         ret = -EINVAL;
2146     } else {
2147         ret = iscsi_open(bs, bs_options, 0, NULL);
2148     }
2149     qobject_unref(bs_options);
2150 
2151     if (ret != 0) {
2152         goto out;
2153     }
2154     iscsi_detach_aio_context(bs);
2155     if (iscsilun->type != TYPE_DISK) {
2156         ret = -ENODEV;
2157         goto out;
2158     }
2159     if (bs->total_sectors < total_size) {
2160         ret = -ENOSPC;
2161         goto out;
2162     }
2163 
2164     ret = 0;
2165 out:
2166     if (iscsilun->iscsi != NULL) {
2167         iscsi_destroy_context(iscsilun->iscsi);
2168     }
2169     g_free(bs->opaque);
2170     bs->opaque = NULL;
2171     bdrv_unref(bs);
2172     return ret;
2173 }
2174 
2175 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2176 {
2177     IscsiLun *iscsilun = bs->opaque;
2178     bdi->unallocated_blocks_are_zero = iscsilun->lbprz;
2179     bdi->cluster_size = iscsilun->cluster_size;
2180     return 0;
2181 }
2182 
2183 static void coroutine_fn iscsi_co_invalidate_cache(BlockDriverState *bs,
2184                                                    Error **errp)
2185 {
2186     IscsiLun *iscsilun = bs->opaque;
2187     iscsi_allocmap_invalidate(iscsilun);
2188 }
2189 
2190 static int coroutine_fn iscsi_co_copy_range_from(BlockDriverState *bs,
2191                                                  BdrvChild *src,
2192                                                  uint64_t src_offset,
2193                                                  BdrvChild *dst,
2194                                                  uint64_t dst_offset,
2195                                                  uint64_t bytes,
2196                                                  BdrvRequestFlags flags)
2197 {
2198     return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, flags);
2199 }
2200 
2201 static struct scsi_task *iscsi_xcopy_task(int param_len)
2202 {
2203     struct scsi_task *task;
2204 
2205     task = g_new0(struct scsi_task, 1);
2206 
2207     task->cdb[0]     = EXTENDED_COPY;
2208     task->cdb[10]    = (param_len >> 24) & 0xFF;
2209     task->cdb[11]    = (param_len >> 16) & 0xFF;
2210     task->cdb[12]    = (param_len >> 8) & 0xFF;
2211     task->cdb[13]    = param_len & 0xFF;
2212     task->cdb_size   = 16;
2213     task->xfer_dir   = SCSI_XFER_WRITE;
2214     task->expxferlen = param_len;
2215 
2216     return task;
2217 }
2218 
2219 static void iscsi_populate_target_desc(unsigned char *desc, IscsiLun *lun)
2220 {
2221     struct scsi_inquiry_device_designator *dd = lun->dd;
2222 
2223     memset(desc, 0, 32);
2224     desc[0] = 0xE4; /* IDENT_DESCR_TGT_DESCR */
2225     desc[4] = dd->code_set;
2226     desc[5] = (dd->designator_type & 0xF)
2227         | ((dd->association & 3) << 4);
2228     desc[7] = dd->designator_length;
2229     memcpy(desc + 8, dd->designator, dd->designator_length);
2230 
2231     desc[28] = 0;
2232     desc[29] = (lun->block_size >> 16) & 0xFF;
2233     desc[30] = (lun->block_size >> 8) & 0xFF;
2234     desc[31] = lun->block_size & 0xFF;
2235 }
2236 
2237 static void iscsi_xcopy_desc_hdr(uint8_t *hdr, int dc, int cat, int src_index,
2238                                  int dst_index)
2239 {
2240     hdr[0] = 0x02; /* BLK_TO_BLK_SEG_DESCR */
2241     hdr[1] = ((dc << 1) | cat) & 0xFF;
2242     hdr[2] = (XCOPY_BLK2BLK_SEG_DESC_SIZE >> 8) & 0xFF;
2243     /* don't account for the first 4 bytes in descriptor header*/
2244     hdr[3] = (XCOPY_BLK2BLK_SEG_DESC_SIZE - 4 /* SEG_DESC_SRC_INDEX_OFFSET */) & 0xFF;
2245     hdr[4] = (src_index >> 8) & 0xFF;
2246     hdr[5] = src_index & 0xFF;
2247     hdr[6] = (dst_index >> 8) & 0xFF;
2248     hdr[7] = dst_index & 0xFF;
2249 }
2250 
2251 static void iscsi_xcopy_populate_desc(uint8_t *desc, int dc, int cat,
2252                                       int src_index, int dst_index, int num_blks,
2253                                       uint64_t src_lba, uint64_t dst_lba)
2254 {
2255     iscsi_xcopy_desc_hdr(desc, dc, cat, src_index, dst_index);
2256 
2257     /* The caller should verify the request size */
2258     assert(num_blks < 65536);
2259     desc[10] = (num_blks >> 8) & 0xFF;
2260     desc[11] = num_blks & 0xFF;
2261     desc[12] = (src_lba >> 56) & 0xFF;
2262     desc[13] = (src_lba >> 48) & 0xFF;
2263     desc[14] = (src_lba >> 40) & 0xFF;
2264     desc[15] = (src_lba >> 32) & 0xFF;
2265     desc[16] = (src_lba >> 24) & 0xFF;
2266     desc[17] = (src_lba >> 16) & 0xFF;
2267     desc[18] = (src_lba >> 8) & 0xFF;
2268     desc[19] = src_lba & 0xFF;
2269     desc[20] = (dst_lba >> 56) & 0xFF;
2270     desc[21] = (dst_lba >> 48) & 0xFF;
2271     desc[22] = (dst_lba >> 40) & 0xFF;
2272     desc[23] = (dst_lba >> 32) & 0xFF;
2273     desc[24] = (dst_lba >> 24) & 0xFF;
2274     desc[25] = (dst_lba >> 16) & 0xFF;
2275     desc[26] = (dst_lba >> 8) & 0xFF;
2276     desc[27] = dst_lba & 0xFF;
2277 }
2278 
2279 static void iscsi_xcopy_populate_header(unsigned char *buf, int list_id, int str,
2280                                         int list_id_usage, int prio,
2281                                         int tgt_desc_len,
2282                                         int seg_desc_len, int inline_data_len)
2283 {
2284     buf[0] = list_id;
2285     buf[1] = ((str & 1) << 5) | ((list_id_usage & 3) << 3) | (prio & 7);
2286     buf[2] = (tgt_desc_len >> 8) & 0xFF;
2287     buf[3] = tgt_desc_len & 0xFF;
2288     buf[8] = (seg_desc_len >> 24) & 0xFF;
2289     buf[9] = (seg_desc_len >> 16) & 0xFF;
2290     buf[10] = (seg_desc_len >> 8) & 0xFF;
2291     buf[11] = seg_desc_len & 0xFF;
2292     buf[12] = (inline_data_len >> 24) & 0xFF;
2293     buf[13] = (inline_data_len >> 16) & 0xFF;
2294     buf[14] = (inline_data_len >> 8) & 0xFF;
2295     buf[15] = inline_data_len & 0xFF;
2296 }
2297 
2298 static void iscsi_xcopy_data(struct iscsi_data *data,
2299                              IscsiLun *src, int64_t src_lba,
2300                              IscsiLun *dst, int64_t dst_lba,
2301                              uint16_t num_blocks)
2302 {
2303     uint8_t *buf;
2304     const int src_offset = XCOPY_DESC_OFFSET;
2305     const int dst_offset = XCOPY_DESC_OFFSET + IDENT_DESCR_TGT_DESCR_SIZE;
2306     const int seg_offset = dst_offset + IDENT_DESCR_TGT_DESCR_SIZE;
2307 
2308     data->size = XCOPY_DESC_OFFSET +
2309                  IDENT_DESCR_TGT_DESCR_SIZE * 2 +
2310                  XCOPY_BLK2BLK_SEG_DESC_SIZE;
2311     data->data = g_malloc0(data->size);
2312     buf = data->data;
2313 
2314     /* Initialise the parameter list header */
2315     iscsi_xcopy_populate_header(buf, 1, 0, 2 /* LIST_ID_USAGE_DISCARD */,
2316                                 0, 2 * IDENT_DESCR_TGT_DESCR_SIZE,
2317                                 XCOPY_BLK2BLK_SEG_DESC_SIZE,
2318                                 0);
2319 
2320     /* Initialise CSCD list with one src + one dst descriptor */
2321     iscsi_populate_target_desc(&buf[src_offset], src);
2322     iscsi_populate_target_desc(&buf[dst_offset], dst);
2323 
2324     /* Initialise one segment descriptor */
2325     iscsi_xcopy_populate_desc(&buf[seg_offset], 0, 0, 0, 1, num_blocks,
2326                               src_lba, dst_lba);
2327 }
2328 
2329 static int coroutine_fn iscsi_co_copy_range_to(BlockDriverState *bs,
2330                                                BdrvChild *src,
2331                                                uint64_t src_offset,
2332                                                BdrvChild *dst,
2333                                                uint64_t dst_offset,
2334                                                uint64_t bytes,
2335                                                BdrvRequestFlags flags)
2336 {
2337     IscsiLun *dst_lun = dst->bs->opaque;
2338     IscsiLun *src_lun;
2339     struct IscsiTask iscsi_task;
2340     struct iscsi_data data;
2341     int r = 0;
2342     int block_size;
2343 
2344     if (src->bs->drv->bdrv_co_copy_range_to != iscsi_co_copy_range_to) {
2345         return -ENOTSUP;
2346     }
2347     src_lun = src->bs->opaque;
2348 
2349     if (!src_lun->dd || !dst_lun->dd) {
2350         return -ENOTSUP;
2351     }
2352     if (!is_byte_request_lun_aligned(dst_offset, bytes, dst_lun)) {
2353         return -ENOTSUP;
2354     }
2355     if (!is_byte_request_lun_aligned(src_offset, bytes, src_lun)) {
2356         return -ENOTSUP;
2357     }
2358     if (dst_lun->block_size != src_lun->block_size ||
2359         !dst_lun->block_size) {
2360         return -ENOTSUP;
2361     }
2362 
2363     block_size = dst_lun->block_size;
2364     if (bytes / block_size > 65535) {
2365         return -ENOTSUP;
2366     }
2367 
2368     iscsi_xcopy_data(&data,
2369                      src_lun, src_offset / block_size,
2370                      dst_lun, dst_offset / block_size,
2371                      bytes / block_size);
2372 
2373     iscsi_co_init_iscsitask(dst_lun, &iscsi_task);
2374 
2375     qemu_mutex_lock(&dst_lun->mutex);
2376     iscsi_task.task = iscsi_xcopy_task(data.size);
2377 retry:
2378     if (iscsi_scsi_command_async(dst_lun->iscsi, dst_lun->lun,
2379                                  iscsi_task.task, iscsi_co_generic_cb,
2380                                  &data,
2381                                  &iscsi_task) != 0) {
2382         r = -EIO;
2383         goto out_unlock;
2384     }
2385 
2386     iscsi_co_wait_for_task(&iscsi_task, dst_lun);
2387 
2388     if (iscsi_task.do_retry) {
2389         iscsi_task.complete = 0;
2390         goto retry;
2391     }
2392 
2393     if (iscsi_task.status != SCSI_STATUS_GOOD) {
2394         r = iscsi_task.err_code;
2395         goto out_unlock;
2396     }
2397 
2398 out_unlock:
2399     g_free(iscsi_task.task);
2400     qemu_mutex_unlock(&dst_lun->mutex);
2401     g_free(iscsi_task.err_str);
2402     return r;
2403 }
2404 
2405 static QemuOptsList iscsi_create_opts = {
2406     .name = "iscsi-create-opts",
2407     .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
2408     .desc = {
2409         {
2410             .name = BLOCK_OPT_SIZE,
2411             .type = QEMU_OPT_SIZE,
2412             .help = "Virtual disk size"
2413         },
2414         { /* end of list */ }
2415     }
2416 };
2417 
2418 static BlockDriver bdrv_iscsi = {
2419     .format_name     = "iscsi",
2420     .protocol_name   = "iscsi",
2421 
2422     .instance_size          = sizeof(IscsiLun),
2423     .bdrv_parse_filename    = iscsi_parse_filename,
2424     .bdrv_file_open         = iscsi_open,
2425     .bdrv_close             = iscsi_close,
2426     .bdrv_co_create_opts    = iscsi_co_create_opts,
2427     .create_opts            = &iscsi_create_opts,
2428     .bdrv_reopen_prepare    = iscsi_reopen_prepare,
2429     .bdrv_reopen_commit     = iscsi_reopen_commit,
2430     .bdrv_co_invalidate_cache = iscsi_co_invalidate_cache,
2431 
2432     .bdrv_getlength  = iscsi_getlength,
2433     .bdrv_get_info   = iscsi_get_info,
2434     .bdrv_truncate   = iscsi_truncate,
2435     .bdrv_refresh_limits = iscsi_refresh_limits,
2436 
2437     .bdrv_co_block_status  = iscsi_co_block_status,
2438     .bdrv_co_pdiscard      = iscsi_co_pdiscard,
2439     .bdrv_co_copy_range_from = iscsi_co_copy_range_from,
2440     .bdrv_co_copy_range_to  = iscsi_co_copy_range_to,
2441     .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2442     .bdrv_co_readv         = iscsi_co_readv,
2443     .bdrv_co_writev        = iscsi_co_writev,
2444     .bdrv_co_flush_to_disk = iscsi_co_flush,
2445 
2446 #ifdef __linux__
2447     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
2448 #endif
2449 
2450     .bdrv_detach_aio_context = iscsi_detach_aio_context,
2451     .bdrv_attach_aio_context = iscsi_attach_aio_context,
2452 };
2453 
2454 #if LIBISCSI_API_VERSION >= (20160603)
2455 static BlockDriver bdrv_iser = {
2456     .format_name     = "iser",
2457     .protocol_name   = "iser",
2458 
2459     .instance_size          = sizeof(IscsiLun),
2460     .bdrv_parse_filename    = iscsi_parse_filename,
2461     .bdrv_file_open         = iscsi_open,
2462     .bdrv_close             = iscsi_close,
2463     .bdrv_co_create_opts    = iscsi_co_create_opts,
2464     .create_opts            = &iscsi_create_opts,
2465     .bdrv_reopen_prepare    = iscsi_reopen_prepare,
2466     .bdrv_reopen_commit     = iscsi_reopen_commit,
2467     .bdrv_co_invalidate_cache  = iscsi_co_invalidate_cache,
2468 
2469     .bdrv_getlength  = iscsi_getlength,
2470     .bdrv_get_info   = iscsi_get_info,
2471     .bdrv_truncate   = iscsi_truncate,
2472     .bdrv_refresh_limits = iscsi_refresh_limits,
2473 
2474     .bdrv_co_block_status  = iscsi_co_block_status,
2475     .bdrv_co_pdiscard      = iscsi_co_pdiscard,
2476     .bdrv_co_copy_range_from = iscsi_co_copy_range_from,
2477     .bdrv_co_copy_range_to  = iscsi_co_copy_range_to,
2478     .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2479     .bdrv_co_readv         = iscsi_co_readv,
2480     .bdrv_co_writev        = iscsi_co_writev,
2481     .bdrv_co_flush_to_disk = iscsi_co_flush,
2482 
2483 #ifdef __linux__
2484     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
2485 #endif
2486 
2487     .bdrv_detach_aio_context = iscsi_detach_aio_context,
2488     .bdrv_attach_aio_context = iscsi_attach_aio_context,
2489 };
2490 #endif
2491 
2492 static void iscsi_block_init(void)
2493 {
2494     bdrv_register(&bdrv_iscsi);
2495 #if LIBISCSI_API_VERSION >= (20160603)
2496     bdrv_register(&bdrv_iser);
2497 #endif
2498 }
2499 
2500 block_init(iscsi_block_init);
2501