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