xref: /qemu/block/iscsi.c (revision a976ed3f)
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->task = task;
245 
246     if (status != SCSI_STATUS_GOOD) {
247         if (iTask->retries++ < ISCSI_CMD_RETRIES) {
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             }
270         } else if (status == SCSI_STATUS_CHECK_CONDITION) {
271             int error = iscsi_translate_sense(&task->sense);
272             if (error == EAGAIN) {
273                 error_report("iSCSI CheckCondition: %s",
274                              iscsi_get_error(iscsi));
275                 iTask->do_retry = 1;
276             } else {
277                 iTask->err_code = -error;
278                 iTask->err_str = g_strdup(iscsi_get_error(iscsi));
279             }
280         }
281     }
282 
283     if (iTask->co) {
284         replay_bh_schedule_oneshot_event(iTask->iscsilun->aio_context,
285                                          iscsi_co_generic_bh_cb, iTask);
286     } else {
287         iTask->complete = 1;
288     }
289 }
290 
291 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
292 {
293     *iTask = (struct IscsiTask) {
294         .co         = qemu_coroutine_self(),
295         .iscsilun   = iscsilun,
296     };
297 }
298 
299 #ifdef __linux__
300 
301 /* Called (via iscsi_service) with QemuMutex held. */
302 static void
303 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
304                     void *private_data)
305 {
306     IscsiAIOCB *acb = private_data;
307 
308     /* If the command callback hasn't been called yet, drop the task */
309     if (!acb->bh) {
310         /* Call iscsi_aio_ioctl_cb() with SCSI_STATUS_CANCELLED */
311         iscsi_scsi_cancel_task(iscsi, acb->task);
312     }
313 
314     qemu_aio_unref(acb); /* acquired in iscsi_aio_cancel() */
315 }
316 
317 static void
318 iscsi_aio_cancel(BlockAIOCB *blockacb)
319 {
320     IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
321     IscsiLun *iscsilun = acb->iscsilun;
322 
323     qemu_mutex_lock(&iscsilun->mutex);
324 
325     /* If it was cancelled or completed already, our work is done here */
326     if (acb->cancelled || acb->status != -EINPROGRESS) {
327         qemu_mutex_unlock(&iscsilun->mutex);
328         return;
329     }
330 
331     acb->cancelled = true;
332 
333     qemu_aio_ref(acb); /* released in iscsi_abort_task_cb() */
334 
335     /* send a task mgmt call to the target to cancel the task on the target */
336     if (iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
337                                          iscsi_abort_task_cb, acb) < 0) {
338         qemu_aio_unref(acb); /* since iscsi_abort_task_cb() won't be called */
339     }
340 
341     qemu_mutex_unlock(&iscsilun->mutex);
342 }
343 
344 static const AIOCBInfo iscsi_aiocb_info = {
345     .aiocb_size         = sizeof(IscsiAIOCB),
346     .cancel_async       = iscsi_aio_cancel,
347 };
348 
349 #endif
350 
351 static void iscsi_process_read(void *arg);
352 static void iscsi_process_write(void *arg);
353 
354 /* Called with QemuMutex held.  */
355 static void
356 iscsi_set_events(IscsiLun *iscsilun)
357 {
358     struct iscsi_context *iscsi = iscsilun->iscsi;
359     int ev = iscsi_which_events(iscsi);
360 
361     if (ev != iscsilun->events) {
362         aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi),
363                            false,
364                            (ev & POLLIN) ? iscsi_process_read : NULL,
365                            (ev & POLLOUT) ? iscsi_process_write : NULL,
366                            NULL,
367                            iscsilun);
368         iscsilun->events = ev;
369     }
370 }
371 
372 static void iscsi_timed_check_events(void *opaque)
373 {
374     IscsiLun *iscsilun = opaque;
375 
376     qemu_mutex_lock(&iscsilun->mutex);
377 
378     /* check for timed out requests */
379     iscsi_service(iscsilun->iscsi, 0);
380 
381     if (iscsilun->request_timed_out) {
382         iscsilun->request_timed_out = false;
383         iscsi_reconnect(iscsilun->iscsi);
384     }
385 
386     /* newer versions of libiscsi may return zero events. Ensure we are able
387      * to return to service once this situation changes. */
388     iscsi_set_events(iscsilun);
389 
390     qemu_mutex_unlock(&iscsilun->mutex);
391 
392     timer_mod(iscsilun->event_timer,
393               qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
394 }
395 
396 static void
397 iscsi_process_read(void *arg)
398 {
399     IscsiLun *iscsilun = arg;
400     struct iscsi_context *iscsi = iscsilun->iscsi;
401 
402     qemu_mutex_lock(&iscsilun->mutex);
403     iscsi_service(iscsi, POLLIN);
404     iscsi_set_events(iscsilun);
405     qemu_mutex_unlock(&iscsilun->mutex);
406 }
407 
408 static void
409 iscsi_process_write(void *arg)
410 {
411     IscsiLun *iscsilun = arg;
412     struct iscsi_context *iscsi = iscsilun->iscsi;
413 
414     qemu_mutex_lock(&iscsilun->mutex);
415     iscsi_service(iscsi, POLLOUT);
416     iscsi_set_events(iscsilun);
417     qemu_mutex_unlock(&iscsilun->mutex);
418 }
419 
420 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
421 {
422     return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
423 }
424 
425 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
426 {
427     return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
428 }
429 
430 static bool is_byte_request_lun_aligned(int64_t offset, int count,
431                                         IscsiLun *iscsilun)
432 {
433     if (offset % iscsilun->block_size || count % iscsilun->block_size) {
434         error_report("iSCSI misaligned request: "
435                      "iscsilun->block_size %u, offset %" PRIi64
436                      ", count %d",
437                      iscsilun->block_size, offset, count);
438         return false;
439     }
440     return true;
441 }
442 
443 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors,
444                                           IscsiLun *iscsilun)
445 {
446     assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS);
447     return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS,
448                                        nb_sectors << BDRV_SECTOR_BITS,
449                                        iscsilun);
450 }
451 
452 static void iscsi_allocmap_free(IscsiLun *iscsilun)
453 {
454     g_free(iscsilun->allocmap);
455     g_free(iscsilun->allocmap_valid);
456     iscsilun->allocmap = NULL;
457     iscsilun->allocmap_valid = NULL;
458 }
459 
460 
461 static int iscsi_allocmap_init(IscsiLun *iscsilun, int open_flags)
462 {
463     iscsi_allocmap_free(iscsilun);
464 
465     assert(iscsilun->cluster_size);
466     iscsilun->allocmap_size =
467         DIV_ROUND_UP(iscsilun->num_blocks * iscsilun->block_size,
468                      iscsilun->cluster_size);
469 
470     iscsilun->allocmap = bitmap_try_new(iscsilun->allocmap_size);
471     if (!iscsilun->allocmap) {
472         return -ENOMEM;
473     }
474 
475     if (open_flags & BDRV_O_NOCACHE) {
476         /* when cache.direct = on all allocmap entries are
477          * treated as invalid to force a relookup of the block
478          * status on every read request */
479         return 0;
480     }
481 
482     iscsilun->allocmap_valid = bitmap_try_new(iscsilun->allocmap_size);
483     if (!iscsilun->allocmap_valid) {
484         /* if we are under memory pressure free the allocmap as well */
485         iscsi_allocmap_free(iscsilun);
486         return -ENOMEM;
487     }
488 
489     return 0;
490 }
491 
492 static void
493 iscsi_allocmap_update(IscsiLun *iscsilun, int64_t offset,
494                       int64_t bytes, bool allocated, bool valid)
495 {
496     int64_t cl_num_expanded, nb_cls_expanded, cl_num_shrunk, nb_cls_shrunk;
497 
498     if (iscsilun->allocmap == NULL) {
499         return;
500     }
501     /* expand to entirely contain all affected clusters */
502     assert(iscsilun->cluster_size);
503     cl_num_expanded = offset / iscsilun->cluster_size;
504     nb_cls_expanded = DIV_ROUND_UP(offset + bytes,
505                                    iscsilun->cluster_size) - cl_num_expanded;
506     /* shrink to touch only completely contained clusters */
507     cl_num_shrunk = DIV_ROUND_UP(offset, iscsilun->cluster_size);
508     nb_cls_shrunk = (offset + bytes) / iscsilun->cluster_size - cl_num_shrunk;
509     if (allocated) {
510         bitmap_set(iscsilun->allocmap, cl_num_expanded, nb_cls_expanded);
511     } else {
512         if (nb_cls_shrunk > 0) {
513             bitmap_clear(iscsilun->allocmap, cl_num_shrunk, nb_cls_shrunk);
514         }
515     }
516 
517     if (iscsilun->allocmap_valid == NULL) {
518         return;
519     }
520     if (valid) {
521         if (nb_cls_shrunk > 0) {
522             bitmap_set(iscsilun->allocmap_valid, cl_num_shrunk, nb_cls_shrunk);
523         }
524     } else {
525         bitmap_clear(iscsilun->allocmap_valid, cl_num_expanded,
526                      nb_cls_expanded);
527     }
528 }
529 
530 static void
531 iscsi_allocmap_set_allocated(IscsiLun *iscsilun, int64_t offset,
532                              int64_t bytes)
533 {
534     iscsi_allocmap_update(iscsilun, offset, bytes, true, true);
535 }
536 
537 static void
538 iscsi_allocmap_set_unallocated(IscsiLun *iscsilun, int64_t offset,
539                                int64_t bytes)
540 {
541     /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update
542      * is ignored, so this will in effect be an iscsi_allocmap_set_invalid.
543      */
544     iscsi_allocmap_update(iscsilun, offset, bytes, false, true);
545 }
546 
547 static void iscsi_allocmap_set_invalid(IscsiLun *iscsilun, int64_t offset,
548                                        int64_t bytes)
549 {
550     iscsi_allocmap_update(iscsilun, offset, bytes, false, false);
551 }
552 
553 static void iscsi_allocmap_invalidate(IscsiLun *iscsilun)
554 {
555     if (iscsilun->allocmap) {
556         bitmap_zero(iscsilun->allocmap, iscsilun->allocmap_size);
557     }
558     if (iscsilun->allocmap_valid) {
559         bitmap_zero(iscsilun->allocmap_valid, iscsilun->allocmap_size);
560     }
561 }
562 
563 static inline bool
564 iscsi_allocmap_is_allocated(IscsiLun *iscsilun, int64_t offset,
565                             int64_t bytes)
566 {
567     unsigned long size;
568     if (iscsilun->allocmap == NULL) {
569         return true;
570     }
571     assert(iscsilun->cluster_size);
572     size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size);
573     return !(find_next_bit(iscsilun->allocmap, size,
574                            offset / iscsilun->cluster_size) == size);
575 }
576 
577 static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun,
578                                            int64_t offset, int64_t bytes)
579 {
580     unsigned long size;
581     if (iscsilun->allocmap_valid == NULL) {
582         return false;
583     }
584     assert(iscsilun->cluster_size);
585     size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size);
586     return (find_next_zero_bit(iscsilun->allocmap_valid, size,
587                                offset / iscsilun->cluster_size) == size);
588 }
589 
590 static void coroutine_fn iscsi_co_wait_for_task(IscsiTask *iTask,
591                                                 IscsiLun *iscsilun)
592 {
593     while (!iTask->complete) {
594         iscsi_set_events(iscsilun);
595         qemu_mutex_unlock(&iscsilun->mutex);
596         qemu_coroutine_yield();
597         qemu_mutex_lock(&iscsilun->mutex);
598     }
599 }
600 
601 static int coroutine_fn
602 iscsi_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
603                 QEMUIOVector *iov, int flags)
604 {
605     IscsiLun *iscsilun = bs->opaque;
606     struct IscsiTask iTask;
607     uint64_t lba;
608     uint32_t num_sectors;
609     bool fua = flags & BDRV_REQ_FUA;
610     int r = 0;
611 
612     if (fua) {
613         assert(iscsilun->dpofua);
614     }
615     if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
616         return -EINVAL;
617     }
618 
619     if (bs->bl.max_transfer) {
620         assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
621     }
622 
623     lba = sector_qemu2lun(sector_num, iscsilun);
624     num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
625     iscsi_co_init_iscsitask(iscsilun, &iTask);
626     qemu_mutex_lock(&iscsilun->mutex);
627 retry:
628     if (iscsilun->use_16_for_rw) {
629 #if LIBISCSI_API_VERSION >= (20160603)
630         iTask.task = iscsi_write16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
631                                             NULL, num_sectors * iscsilun->block_size,
632                                             iscsilun->block_size, 0, 0, fua, 0, 0,
633                                             iscsi_co_generic_cb, &iTask,
634                                             (struct scsi_iovec *)iov->iov, iov->niov);
635     } else {
636         iTask.task = iscsi_write10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
637                                             NULL, num_sectors * iscsilun->block_size,
638                                             iscsilun->block_size, 0, 0, fua, 0, 0,
639                                             iscsi_co_generic_cb, &iTask,
640                                             (struct scsi_iovec *)iov->iov, iov->niov);
641     }
642 #else
643         iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
644                                         NULL, num_sectors * iscsilun->block_size,
645                                         iscsilun->block_size, 0, 0, fua, 0, 0,
646                                         iscsi_co_generic_cb, &iTask);
647     } else {
648         iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
649                                         NULL, num_sectors * iscsilun->block_size,
650                                         iscsilun->block_size, 0, 0, fua, 0, 0,
651                                         iscsi_co_generic_cb, &iTask);
652     }
653 #endif
654     if (iTask.task == NULL) {
655         qemu_mutex_unlock(&iscsilun->mutex);
656         return -ENOMEM;
657     }
658 #if LIBISCSI_API_VERSION < (20160603)
659     scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
660                           iov->niov);
661 #endif
662     iscsi_co_wait_for_task(&iTask, iscsilun);
663 
664     if (iTask.task != NULL) {
665         scsi_free_scsi_task(iTask.task);
666         iTask.task = NULL;
667     }
668 
669     if (iTask.do_retry) {
670         iTask.complete = 0;
671         goto retry;
672     }
673 
674     if (iTask.status != SCSI_STATUS_GOOD) {
675         iscsi_allocmap_set_invalid(iscsilun, sector_num * BDRV_SECTOR_SIZE,
676                                    nb_sectors * BDRV_SECTOR_SIZE);
677         error_report("iSCSI WRITE10/16 failed at lba %" PRIu64 ": %s", lba,
678                      iTask.err_str);
679         r = iTask.err_code;
680         goto out_unlock;
681     }
682 
683     iscsi_allocmap_set_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE,
684                                  nb_sectors * BDRV_SECTOR_SIZE);
685 
686 out_unlock:
687     qemu_mutex_unlock(&iscsilun->mutex);
688     g_free(iTask.err_str);
689     return r;
690 }
691 
692 
693 
694 static int coroutine_fn iscsi_co_block_status(BlockDriverState *bs,
695                                               bool want_zero, int64_t offset,
696                                               int64_t bytes, int64_t *pnum,
697                                               int64_t *map,
698                                               BlockDriverState **file)
699 {
700     IscsiLun *iscsilun = bs->opaque;
701     struct scsi_get_lba_status *lbas = NULL;
702     struct scsi_lba_status_descriptor *lbasd = NULL;
703     struct IscsiTask iTask;
704     uint64_t lba, max_bytes;
705     int ret;
706 
707     iscsi_co_init_iscsitask(iscsilun, &iTask);
708 
709     assert(QEMU_IS_ALIGNED(offset | bytes, iscsilun->block_size));
710 
711     /* default to all sectors allocated */
712     ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
713     if (map) {
714         *map = offset;
715     }
716     *pnum = bytes;
717 
718     /* LUN does not support logical block provisioning */
719     if (!iscsilun->lbpme) {
720         goto out;
721     }
722 
723     lba = offset / iscsilun->block_size;
724     max_bytes = (iscsilun->num_blocks - lba) * iscsilun->block_size;
725 
726     qemu_mutex_lock(&iscsilun->mutex);
727 retry:
728     if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
729                                   lba, 8 + 16, iscsi_co_generic_cb,
730                                   &iTask) == NULL) {
731         ret = -ENOMEM;
732         goto out_unlock;
733     }
734     iscsi_co_wait_for_task(&iTask, iscsilun);
735 
736     if (iTask.do_retry) {
737         if (iTask.task != NULL) {
738             scsi_free_scsi_task(iTask.task);
739             iTask.task = NULL;
740         }
741         iTask.complete = 0;
742         goto retry;
743     }
744 
745     if (iTask.status != SCSI_STATUS_GOOD) {
746         /* in case the get_lba_status_callout fails (i.e.
747          * because the device is busy or the cmd is not
748          * supported) we pretend all blocks are allocated
749          * for backwards compatibility */
750         error_report("iSCSI GET_LBA_STATUS failed at lba %" PRIu64 ": %s",
751                      lba, iTask.err_str);
752         goto out_unlock;
753     }
754 
755     lbas = scsi_datain_unmarshall(iTask.task);
756     if (lbas == NULL || lbas->num_descriptors == 0) {
757         ret = -EIO;
758         goto out_unlock;
759     }
760 
761     lbasd = &lbas->descriptors[0];
762 
763     if (lba != lbasd->lba) {
764         ret = -EIO;
765         goto out_unlock;
766     }
767 
768     *pnum = MIN((int64_t) lbasd->num_blocks * iscsilun->block_size, max_bytes);
769 
770     if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
771         lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
772         ret &= ~BDRV_BLOCK_DATA;
773         if (iscsilun->lbprz) {
774             ret |= BDRV_BLOCK_ZERO;
775         }
776     }
777 
778     if (ret & BDRV_BLOCK_ZERO) {
779         iscsi_allocmap_set_unallocated(iscsilun, offset, *pnum);
780     } else {
781         iscsi_allocmap_set_allocated(iscsilun, offset, *pnum);
782     }
783 
784     if (*pnum > bytes) {
785         *pnum = bytes;
786     }
787 out_unlock:
788     qemu_mutex_unlock(&iscsilun->mutex);
789     g_free(iTask.err_str);
790 out:
791     if (iTask.task != NULL) {
792         scsi_free_scsi_task(iTask.task);
793     }
794     if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID && file) {
795         *file = bs;
796     }
797     return ret;
798 }
799 
800 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
801                                        int64_t sector_num, int nb_sectors,
802                                        QEMUIOVector *iov)
803 {
804     IscsiLun *iscsilun = bs->opaque;
805     struct IscsiTask iTask;
806     uint64_t lba;
807     uint32_t num_sectors;
808     int r = 0;
809 
810     if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
811         return -EINVAL;
812     }
813 
814     if (bs->bl.max_transfer) {
815         assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
816     }
817 
818     /* if cache.direct is off and we have a valid entry in our allocation map
819      * we can skip checking the block status and directly return zeroes if
820      * the request falls within an unallocated area */
821     if (iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE,
822                                 nb_sectors * BDRV_SECTOR_SIZE) &&
823         !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE,
824                                      nb_sectors * BDRV_SECTOR_SIZE)) {
825             qemu_iovec_memset(iov, 0, 0x00, iov->size);
826             return 0;
827     }
828 
829     if (nb_sectors >= ISCSI_CHECKALLOC_THRES &&
830         !iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE,
831                                  nb_sectors * BDRV_SECTOR_SIZE) &&
832         !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE,
833                                      nb_sectors * BDRV_SECTOR_SIZE)) {
834         int64_t pnum;
835         /* check the block status from the beginning of the cluster
836          * containing the start sector */
837         int64_t head;
838         int ret;
839 
840         assert(iscsilun->cluster_size);
841         head = (sector_num * BDRV_SECTOR_SIZE) % iscsilun->cluster_size;
842         ret = iscsi_co_block_status(bs, true,
843                                     sector_num * BDRV_SECTOR_SIZE - head,
844                                     BDRV_REQUEST_MAX_BYTES, &pnum, NULL, NULL);
845         if (ret < 0) {
846             return ret;
847         }
848         /* if the whole request falls into an unallocated area we can avoid
849          * reading and directly return zeroes instead */
850         if (ret & BDRV_BLOCK_ZERO &&
851             pnum >= nb_sectors * BDRV_SECTOR_SIZE + head) {
852             qemu_iovec_memset(iov, 0, 0x00, iov->size);
853             return 0;
854         }
855     }
856 
857     lba = sector_qemu2lun(sector_num, iscsilun);
858     num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
859 
860     iscsi_co_init_iscsitask(iscsilun, &iTask);
861     qemu_mutex_lock(&iscsilun->mutex);
862 retry:
863     if (iscsilun->use_16_for_rw) {
864 #if LIBISCSI_API_VERSION >= (20160603)
865         iTask.task = iscsi_read16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
866                                            num_sectors * iscsilun->block_size,
867                                            iscsilun->block_size, 0, 0, 0, 0, 0,
868                                            iscsi_co_generic_cb, &iTask,
869                                            (struct scsi_iovec *)iov->iov, iov->niov);
870     } else {
871         iTask.task = iscsi_read10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
872                                            num_sectors * iscsilun->block_size,
873                                            iscsilun->block_size,
874                                            0, 0, 0, 0, 0,
875                                            iscsi_co_generic_cb, &iTask,
876                                            (struct scsi_iovec *)iov->iov, iov->niov);
877     }
878 #else
879         iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
880                                        num_sectors * iscsilun->block_size,
881                                        iscsilun->block_size, 0, 0, 0, 0, 0,
882                                        iscsi_co_generic_cb, &iTask);
883     } else {
884         iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
885                                        num_sectors * iscsilun->block_size,
886                                        iscsilun->block_size,
887                                        0, 0, 0, 0, 0,
888                                        iscsi_co_generic_cb, &iTask);
889     }
890 #endif
891     if (iTask.task == NULL) {
892         qemu_mutex_unlock(&iscsilun->mutex);
893         return -ENOMEM;
894     }
895 #if LIBISCSI_API_VERSION < (20160603)
896     scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
897 #endif
898 
899     iscsi_co_wait_for_task(&iTask, iscsilun);
900     if (iTask.task != NULL) {
901         scsi_free_scsi_task(iTask.task);
902         iTask.task = NULL;
903     }
904 
905     if (iTask.do_retry) {
906         iTask.complete = 0;
907         goto retry;
908     }
909 
910     if (iTask.status != SCSI_STATUS_GOOD) {
911         error_report("iSCSI READ10/16 failed at lba %" PRIu64 ": %s",
912                      lba, iTask.err_str);
913         r = iTask.err_code;
914     }
915 
916     qemu_mutex_unlock(&iscsilun->mutex);
917     g_free(iTask.err_str);
918     return r;
919 }
920 
921 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
922 {
923     IscsiLun *iscsilun = bs->opaque;
924     struct IscsiTask iTask;
925     int r = 0;
926 
927     iscsi_co_init_iscsitask(iscsilun, &iTask);
928     qemu_mutex_lock(&iscsilun->mutex);
929 retry:
930     if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
931                                       0, iscsi_co_generic_cb, &iTask) == NULL) {
932         qemu_mutex_unlock(&iscsilun->mutex);
933         return -ENOMEM;
934     }
935 
936     iscsi_co_wait_for_task(&iTask, iscsilun);
937 
938     if (iTask.task != NULL) {
939         scsi_free_scsi_task(iTask.task);
940         iTask.task = NULL;
941     }
942 
943     if (iTask.do_retry) {
944         iTask.complete = 0;
945         goto retry;
946     }
947 
948     if (iTask.status != SCSI_STATUS_GOOD) {
949         error_report("iSCSI SYNCHRONIZECACHE10 failed: %s", iTask.err_str);
950         r = iTask.err_code;
951     }
952 
953     qemu_mutex_unlock(&iscsilun->mutex);
954     g_free(iTask.err_str);
955     return r;
956 }
957 
958 #ifdef __linux__
959 /* Called (via iscsi_service) with QemuMutex held.  */
960 static void
961 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
962                      void *command_data, void *opaque)
963 {
964     IscsiAIOCB *acb = opaque;
965 
966     if (status == SCSI_STATUS_CANCELLED) {
967         if (!acb->bh) {
968             acb->status = -ECANCELED;
969             iscsi_schedule_bh(acb);
970         }
971         return;
972     }
973 
974     acb->status = 0;
975     if (status < 0) {
976         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
977                      iscsi_get_error(iscsi));
978         acb->status = -iscsi_translate_sense(&acb->task->sense);
979     }
980 
981     acb->ioh->driver_status = 0;
982     acb->ioh->host_status   = 0;
983     acb->ioh->resid         = 0;
984     acb->ioh->status        = status;
985 
986 #define SG_ERR_DRIVER_SENSE    0x08
987 
988     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
989         int ss;
990 
991         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
992 
993         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
994         ss = MIN(acb->ioh->mx_sb_len, acb->ioh->sb_len_wr);
995         memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
996     }
997 
998     iscsi_schedule_bh(acb);
999 }
1000 
1001 static void iscsi_ioctl_bh_completion(void *opaque)
1002 {
1003     IscsiAIOCB *acb = opaque;
1004 
1005     qemu_bh_delete(acb->bh);
1006     acb->common.cb(acb->common.opaque, acb->ret);
1007     qemu_aio_unref(acb);
1008 }
1009 
1010 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
1011 {
1012     BlockDriverState *bs = acb->common.bs;
1013     IscsiLun *iscsilun = bs->opaque;
1014     int ret = 0;
1015 
1016     switch (req) {
1017     case SG_GET_VERSION_NUM:
1018         *(int *)buf = 30000;
1019         break;
1020     case SG_GET_SCSI_ID:
1021         ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
1022         break;
1023     default:
1024         ret = -EINVAL;
1025     }
1026     assert(!acb->bh);
1027     acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
1028                          iscsi_ioctl_bh_completion, acb);
1029     acb->ret = ret;
1030     qemu_bh_schedule(acb->bh);
1031 }
1032 
1033 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
1034         unsigned long int req, void *buf,
1035         BlockCompletionFunc *cb, void *opaque)
1036 {
1037     IscsiLun *iscsilun = bs->opaque;
1038     struct iscsi_context *iscsi = iscsilun->iscsi;
1039     struct iscsi_data data;
1040     IscsiAIOCB *acb;
1041 
1042     acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
1043 
1044     acb->iscsilun = iscsilun;
1045     acb->bh          = NULL;
1046     acb->status      = -EINPROGRESS;
1047     acb->ioh         = buf;
1048     acb->cancelled   = false;
1049 
1050     if (req != SG_IO) {
1051         iscsi_ioctl_handle_emulated(acb, req, buf);
1052         return &acb->common;
1053     }
1054 
1055     if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) {
1056         error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)",
1057                      acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE);
1058         qemu_aio_unref(acb);
1059         return NULL;
1060     }
1061 
1062     acb->task = malloc(sizeof(struct scsi_task));
1063     if (acb->task == NULL) {
1064         error_report("iSCSI: Failed to allocate task for scsi command. %s",
1065                      iscsi_get_error(iscsi));
1066         qemu_aio_unref(acb);
1067         return NULL;
1068     }
1069     memset(acb->task, 0, sizeof(struct scsi_task));
1070 
1071     switch (acb->ioh->dxfer_direction) {
1072     case SG_DXFER_TO_DEV:
1073         acb->task->xfer_dir = SCSI_XFER_WRITE;
1074         break;
1075     case SG_DXFER_FROM_DEV:
1076         acb->task->xfer_dir = SCSI_XFER_READ;
1077         break;
1078     default:
1079         acb->task->xfer_dir = SCSI_XFER_NONE;
1080         break;
1081     }
1082 
1083     acb->task->cdb_size = acb->ioh->cmd_len;
1084     memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
1085     acb->task->expxferlen = acb->ioh->dxfer_len;
1086 
1087     data.size = 0;
1088     qemu_mutex_lock(&iscsilun->mutex);
1089     if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
1090         if (acb->ioh->iovec_count == 0) {
1091             data.data = acb->ioh->dxferp;
1092             data.size = acb->ioh->dxfer_len;
1093         } else {
1094             scsi_task_set_iov_out(acb->task,
1095                                  (struct scsi_iovec *) acb->ioh->dxferp,
1096                                  acb->ioh->iovec_count);
1097         }
1098     }
1099 
1100     if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
1101                                  iscsi_aio_ioctl_cb,
1102                                  (data.size > 0) ? &data : NULL,
1103                                  acb) != 0) {
1104         qemu_mutex_unlock(&iscsilun->mutex);
1105         scsi_free_scsi_task(acb->task);
1106         qemu_aio_unref(acb);
1107         return NULL;
1108     }
1109 
1110     /* tell libiscsi to read straight into the buffer we got from ioctl */
1111     if (acb->task->xfer_dir == SCSI_XFER_READ) {
1112         if (acb->ioh->iovec_count == 0) {
1113             scsi_task_add_data_in_buffer(acb->task,
1114                                          acb->ioh->dxfer_len,
1115                                          acb->ioh->dxferp);
1116         } else {
1117             scsi_task_set_iov_in(acb->task,
1118                                  (struct scsi_iovec *) acb->ioh->dxferp,
1119                                  acb->ioh->iovec_count);
1120         }
1121     }
1122 
1123     iscsi_set_events(iscsilun);
1124     qemu_mutex_unlock(&iscsilun->mutex);
1125 
1126     return &acb->common;
1127 }
1128 
1129 #endif
1130 
1131 static int64_t
1132 iscsi_getlength(BlockDriverState *bs)
1133 {
1134     IscsiLun *iscsilun = bs->opaque;
1135     int64_t len;
1136 
1137     len  = iscsilun->num_blocks;
1138     len *= iscsilun->block_size;
1139 
1140     return len;
1141 }
1142 
1143 static int
1144 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
1145 {
1146     IscsiLun *iscsilun = bs->opaque;
1147     struct IscsiTask iTask;
1148     struct unmap_list list;
1149     int r = 0;
1150 
1151     if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) {
1152         return -ENOTSUP;
1153     }
1154 
1155     if (!iscsilun->lbp.lbpu) {
1156         /* UNMAP is not supported by the target */
1157         return 0;
1158     }
1159 
1160     list.lba = offset / iscsilun->block_size;
1161     list.num = bytes / iscsilun->block_size;
1162 
1163     iscsi_co_init_iscsitask(iscsilun, &iTask);
1164     qemu_mutex_lock(&iscsilun->mutex);
1165 retry:
1166     if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
1167                          iscsi_co_generic_cb, &iTask) == NULL) {
1168         r = -ENOMEM;
1169         goto out_unlock;
1170     }
1171 
1172     iscsi_co_wait_for_task(&iTask, iscsilun);
1173 
1174     if (iTask.task != NULL) {
1175         scsi_free_scsi_task(iTask.task);
1176         iTask.task = NULL;
1177     }
1178 
1179     if (iTask.do_retry) {
1180         iTask.complete = 0;
1181         goto retry;
1182     }
1183 
1184     iscsi_allocmap_set_invalid(iscsilun, offset, bytes);
1185 
1186     if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
1187         /* the target might fail with a check condition if it
1188            is not happy with the alignment of the UNMAP request
1189            we silently fail in this case */
1190         goto out_unlock;
1191     }
1192 
1193     if (iTask.status != SCSI_STATUS_GOOD) {
1194         error_report("iSCSI UNMAP failed at lba %" PRIu64 ": %s",
1195                      list.lba, iTask.err_str);
1196         r = iTask.err_code;
1197         goto out_unlock;
1198     }
1199 
1200 out_unlock:
1201     qemu_mutex_unlock(&iscsilun->mutex);
1202     g_free(iTask.err_str);
1203     return r;
1204 }
1205 
1206 static int
1207 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1208                                     int bytes, BdrvRequestFlags flags)
1209 {
1210     IscsiLun *iscsilun = bs->opaque;
1211     struct IscsiTask iTask;
1212     uint64_t lba;
1213     uint32_t nb_blocks;
1214     bool use_16_for_ws = iscsilun->use_16_for_rw;
1215     int r = 0;
1216 
1217     if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) {
1218         return -ENOTSUP;
1219     }
1220 
1221     if (flags & BDRV_REQ_MAY_UNMAP) {
1222         if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
1223             /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
1224             use_16_for_ws = true;
1225         }
1226         if (use_16_for_ws && !iscsilun->lbp.lbpws) {
1227             /* WRITESAME16 with UNMAP is not supported by the target,
1228              * fall back and try WRITESAME10/16 without UNMAP */
1229             flags &= ~BDRV_REQ_MAY_UNMAP;
1230             use_16_for_ws = iscsilun->use_16_for_rw;
1231         }
1232     }
1233 
1234     if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1235         /* WRITESAME without UNMAP is not supported by the target */
1236         return -ENOTSUP;
1237     }
1238 
1239     lba = offset / iscsilun->block_size;
1240     nb_blocks = bytes / iscsilun->block_size;
1241 
1242     if (iscsilun->zeroblock == NULL) {
1243         iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1244         if (iscsilun->zeroblock == NULL) {
1245             return -ENOMEM;
1246         }
1247     }
1248 
1249     qemu_mutex_lock(&iscsilun->mutex);
1250     iscsi_co_init_iscsitask(iscsilun, &iTask);
1251 retry:
1252     if (use_16_for_ws) {
1253         iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1254                                             iscsilun->zeroblock, iscsilun->block_size,
1255                                             nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1256                                             0, 0, iscsi_co_generic_cb, &iTask);
1257     } else {
1258         iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1259                                             iscsilun->zeroblock, iscsilun->block_size,
1260                                             nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1261                                             0, 0, iscsi_co_generic_cb, &iTask);
1262     }
1263     if (iTask.task == NULL) {
1264         qemu_mutex_unlock(&iscsilun->mutex);
1265         return -ENOMEM;
1266     }
1267 
1268     iscsi_co_wait_for_task(&iTask, iscsilun);
1269 
1270     if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1271         iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1272         (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1273          iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1274         /* WRITE SAME is not supported by the target */
1275         iscsilun->has_write_same = false;
1276         scsi_free_scsi_task(iTask.task);
1277         r = -ENOTSUP;
1278         goto out_unlock;
1279     }
1280 
1281     if (iTask.task != NULL) {
1282         scsi_free_scsi_task(iTask.task);
1283         iTask.task = NULL;
1284     }
1285 
1286     if (iTask.do_retry) {
1287         iTask.complete = 0;
1288         goto retry;
1289     }
1290 
1291     if (iTask.status != SCSI_STATUS_GOOD) {
1292         iscsi_allocmap_set_invalid(iscsilun, offset, bytes);
1293         error_report("iSCSI WRITESAME10/16 failed at lba %" PRIu64 ": %s",
1294                      lba, iTask.err_str);
1295         r = iTask.err_code;
1296         goto out_unlock;
1297     }
1298 
1299     if (flags & BDRV_REQ_MAY_UNMAP) {
1300         iscsi_allocmap_set_invalid(iscsilun, offset, bytes);
1301     } else {
1302         iscsi_allocmap_set_allocated(iscsilun, offset, bytes);
1303     }
1304 
1305 out_unlock:
1306     qemu_mutex_unlock(&iscsilun->mutex);
1307     g_free(iTask.err_str);
1308     return r;
1309 }
1310 
1311 static void apply_chap(struct iscsi_context *iscsi, QemuOpts *opts,
1312                        Error **errp)
1313 {
1314     const char *user = NULL;
1315     const char *password = NULL;
1316     const char *secretid;
1317     char *secret = NULL;
1318 
1319     user = qemu_opt_get(opts, "user");
1320     if (!user) {
1321         return;
1322     }
1323 
1324     secretid = qemu_opt_get(opts, "password-secret");
1325     password = qemu_opt_get(opts, "password");
1326     if (secretid && password) {
1327         error_setg(errp, "'password' and 'password-secret' properties are "
1328                    "mutually exclusive");
1329         return;
1330     }
1331     if (secretid) {
1332         secret = qcrypto_secret_lookup_as_utf8(secretid, errp);
1333         if (!secret) {
1334             return;
1335         }
1336         password = secret;
1337     } else if (!password) {
1338         error_setg(errp, "CHAP username specified but no password was given");
1339         return;
1340     }
1341 
1342     if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1343         error_setg(errp, "Failed to set initiator username and password");
1344     }
1345 
1346     g_free(secret);
1347 }
1348 
1349 static void apply_header_digest(struct iscsi_context *iscsi, QemuOpts *opts,
1350                                 Error **errp)
1351 {
1352     const char *digest = NULL;
1353 
1354     digest = qemu_opt_get(opts, "header-digest");
1355     if (!digest) {
1356         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1357     } else if (!strcmp(digest, "crc32c")) {
1358         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1359     } else if (!strcmp(digest, "none")) {
1360         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1361     } else if (!strcmp(digest, "crc32c-none")) {
1362         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1363     } else if (!strcmp(digest, "none-crc32c")) {
1364         iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1365     } else {
1366         error_setg(errp, "Invalid header-digest setting : %s", digest);
1367     }
1368 }
1369 
1370 static char *get_initiator_name(QemuOpts *opts)
1371 {
1372     const char *name;
1373     char *iscsi_name;
1374     UuidInfo *uuid_info;
1375 
1376     name = qemu_opt_get(opts, "initiator-name");
1377     if (name) {
1378         return g_strdup(name);
1379     }
1380 
1381     uuid_info = qmp_query_uuid(NULL);
1382     if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1383         name = qemu_get_vm_name();
1384     } else {
1385         name = uuid_info->UUID;
1386     }
1387     iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1388                                  name ? ":" : "", name ? name : "");
1389     qapi_free_UuidInfo(uuid_info);
1390     return iscsi_name;
1391 }
1392 
1393 static void iscsi_nop_timed_event(void *opaque)
1394 {
1395     IscsiLun *iscsilun = opaque;
1396 
1397     QEMU_LOCK_GUARD(&iscsilun->mutex);
1398     if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1399         error_report("iSCSI: NOP timeout. Reconnecting...");
1400         iscsilun->request_timed_out = true;
1401     } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1402         error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1403         return;
1404     }
1405 
1406     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1407     iscsi_set_events(iscsilun);
1408 }
1409 
1410 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1411 {
1412     struct scsi_task *task = NULL;
1413     struct scsi_readcapacity10 *rc10 = NULL;
1414     struct scsi_readcapacity16 *rc16 = NULL;
1415     int retries = ISCSI_CMD_RETRIES;
1416 
1417     do {
1418         if (task != NULL) {
1419             scsi_free_scsi_task(task);
1420             task = NULL;
1421         }
1422 
1423         switch (iscsilun->type) {
1424         case TYPE_DISK:
1425             task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1426             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1427                 rc16 = scsi_datain_unmarshall(task);
1428                 if (rc16 == NULL) {
1429                     error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1430                 } else {
1431                     iscsilun->block_size = rc16->block_length;
1432                     iscsilun->num_blocks = rc16->returned_lba + 1;
1433                     iscsilun->lbpme = !!rc16->lbpme;
1434                     iscsilun->lbprz = !!rc16->lbprz;
1435                     iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1436                 }
1437                 break;
1438             }
1439             if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1440                 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1441                 break;
1442             }
1443             /* Fall through and try READ CAPACITY(10) instead.  */
1444         case TYPE_ROM:
1445             task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1446             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1447                 rc10 = scsi_datain_unmarshall(task);
1448                 if (rc10 == NULL) {
1449                     error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1450                 } else {
1451                     iscsilun->block_size = rc10->block_size;
1452                     if (rc10->lba == 0) {
1453                         /* blank disk loaded */
1454                         iscsilun->num_blocks = 0;
1455                     } else {
1456                         iscsilun->num_blocks = rc10->lba + 1;
1457                     }
1458                 }
1459             }
1460             break;
1461         default:
1462             return;
1463         }
1464     } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1465              && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1466              && retries-- > 0);
1467 
1468     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1469         error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1470     } else if (!iscsilun->block_size ||
1471                iscsilun->block_size % BDRV_SECTOR_SIZE) {
1472         error_setg(errp, "iSCSI: the target returned an invalid "
1473                    "block size of %d.", iscsilun->block_size);
1474     }
1475     if (task) {
1476         scsi_free_scsi_task(task);
1477     }
1478 }
1479 
1480 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1481                                           int evpd, int pc, void **inq, Error **errp)
1482 {
1483     int full_size;
1484     struct scsi_task *task = NULL;
1485     task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1486     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1487         goto fail;
1488     }
1489     full_size = scsi_datain_getfullsize(task);
1490     if (full_size > task->datain.size) {
1491         scsi_free_scsi_task(task);
1492 
1493         /* we need more data for the full list */
1494         task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1495         if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1496             goto fail;
1497         }
1498     }
1499 
1500     *inq = scsi_datain_unmarshall(task);
1501     if (*inq == NULL) {
1502         error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1503         goto fail_with_err;
1504     }
1505 
1506     return task;
1507 
1508 fail:
1509     error_setg(errp, "iSCSI: Inquiry command failed : %s",
1510                iscsi_get_error(iscsi));
1511 fail_with_err:
1512     if (task != NULL) {
1513         scsi_free_scsi_task(task);
1514     }
1515     return NULL;
1516 }
1517 
1518 static void iscsi_detach_aio_context(BlockDriverState *bs)
1519 {
1520     IscsiLun *iscsilun = bs->opaque;
1521 
1522     aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1523                        false, NULL, NULL, NULL, NULL);
1524     iscsilun->events = 0;
1525 
1526     if (iscsilun->nop_timer) {
1527         timer_del(iscsilun->nop_timer);
1528         timer_free(iscsilun->nop_timer);
1529         iscsilun->nop_timer = NULL;
1530     }
1531     if (iscsilun->event_timer) {
1532         timer_del(iscsilun->event_timer);
1533         timer_free(iscsilun->event_timer);
1534         iscsilun->event_timer = NULL;
1535     }
1536 }
1537 
1538 static void iscsi_attach_aio_context(BlockDriverState *bs,
1539                                      AioContext *new_context)
1540 {
1541     IscsiLun *iscsilun = bs->opaque;
1542 
1543     iscsilun->aio_context = new_context;
1544     iscsi_set_events(iscsilun);
1545 
1546     /* Set up a timer for sending out iSCSI NOPs */
1547     iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1548                                         QEMU_CLOCK_REALTIME, SCALE_MS,
1549                                         iscsi_nop_timed_event, iscsilun);
1550     timer_mod(iscsilun->nop_timer,
1551               qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1552 
1553     /* Set up a timer for periodic calls to iscsi_set_events and to
1554      * scan for command timeout */
1555     iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1556                                           QEMU_CLOCK_REALTIME, SCALE_MS,
1557                                           iscsi_timed_check_events, iscsilun);
1558     timer_mod(iscsilun->event_timer,
1559               qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1560 }
1561 
1562 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1563 {
1564     struct scsi_task *task;
1565     struct scsi_mode_sense *ms = NULL;
1566     iscsilun->write_protected = false;
1567     iscsilun->dpofua = false;
1568 
1569     task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1570                                  1, SCSI_MODESENSE_PC_CURRENT,
1571                                  0x3F, 0, 255);
1572     if (task == NULL) {
1573         error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1574                      iscsi_get_error(iscsilun->iscsi));
1575         goto out;
1576     }
1577 
1578     if (task->status != SCSI_STATUS_GOOD) {
1579         error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1580         goto out;
1581     }
1582     ms = scsi_datain_unmarshall(task);
1583     if (!ms) {
1584         error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1585                      iscsi_get_error(iscsilun->iscsi));
1586         goto out;
1587     }
1588     iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1589     iscsilun->dpofua          = ms->device_specific_parameter & 0x10;
1590 
1591 out:
1592     if (task) {
1593         scsi_free_scsi_task(task);
1594     }
1595 }
1596 
1597 static void iscsi_parse_iscsi_option(const char *target, QDict *options)
1598 {
1599     QemuOptsList *list;
1600     QemuOpts *opts;
1601     const char *user, *password, *password_secret, *initiator_name,
1602                *header_digest, *timeout;
1603 
1604     list = qemu_find_opts("iscsi");
1605     if (!list) {
1606         return;
1607     }
1608 
1609     opts = qemu_opts_find(list, target);
1610     if (opts == NULL) {
1611         opts = QTAILQ_FIRST(&list->head);
1612         if (!opts) {
1613             return;
1614         }
1615     }
1616 
1617     user = qemu_opt_get(opts, "user");
1618     if (user) {
1619         qdict_set_default_str(options, "user", user);
1620     }
1621 
1622     password = qemu_opt_get(opts, "password");
1623     if (password) {
1624         qdict_set_default_str(options, "password", password);
1625     }
1626 
1627     password_secret = qemu_opt_get(opts, "password-secret");
1628     if (password_secret) {
1629         qdict_set_default_str(options, "password-secret", password_secret);
1630     }
1631 
1632     initiator_name = qemu_opt_get(opts, "initiator-name");
1633     if (initiator_name) {
1634         qdict_set_default_str(options, "initiator-name", initiator_name);
1635     }
1636 
1637     header_digest = qemu_opt_get(opts, "header-digest");
1638     if (header_digest) {
1639         /* -iscsi takes upper case values, but QAPI only supports lower case
1640          * enum constant names, so we have to convert here. */
1641         char *qapi_value = g_ascii_strdown(header_digest, -1);
1642         qdict_set_default_str(options, "header-digest", qapi_value);
1643         g_free(qapi_value);
1644     }
1645 
1646     timeout = qemu_opt_get(opts, "timeout");
1647     if (timeout) {
1648         qdict_set_default_str(options, "timeout", timeout);
1649     }
1650 }
1651 
1652 /*
1653  * We support iscsi url's on the form
1654  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1655  */
1656 static void iscsi_parse_filename(const char *filename, QDict *options,
1657                                  Error **errp)
1658 {
1659     struct iscsi_url *iscsi_url;
1660     const char *transport_name;
1661     char *lun_str;
1662 
1663     iscsi_url = iscsi_parse_full_url(NULL, filename);
1664     if (iscsi_url == NULL) {
1665         error_setg(errp, "Failed to parse URL : %s", filename);
1666         return;
1667     }
1668 
1669 #if LIBISCSI_API_VERSION >= (20160603)
1670     switch (iscsi_url->transport) {
1671     case TCP_TRANSPORT:
1672         transport_name = "tcp";
1673         break;
1674     case ISER_TRANSPORT:
1675         transport_name = "iser";
1676         break;
1677     default:
1678         error_setg(errp, "Unknown transport type (%d)",
1679                    iscsi_url->transport);
1680         return;
1681     }
1682 #else
1683     transport_name = "tcp";
1684 #endif
1685 
1686     qdict_set_default_str(options, "transport", transport_name);
1687     qdict_set_default_str(options, "portal", iscsi_url->portal);
1688     qdict_set_default_str(options, "target", iscsi_url->target);
1689 
1690     lun_str = g_strdup_printf("%d", iscsi_url->lun);
1691     qdict_set_default_str(options, "lun", lun_str);
1692     g_free(lun_str);
1693 
1694     /* User/password from -iscsi take precedence over those from the URL */
1695     iscsi_parse_iscsi_option(iscsi_url->target, options);
1696 
1697     if (iscsi_url->user[0] != '\0') {
1698         qdict_set_default_str(options, "user", iscsi_url->user);
1699         qdict_set_default_str(options, "password", iscsi_url->passwd);
1700     }
1701 
1702     iscsi_destroy_url(iscsi_url);
1703 }
1704 
1705 static QemuOptsList runtime_opts = {
1706     .name = "iscsi",
1707     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1708     .desc = {
1709         {
1710             .name = "transport",
1711             .type = QEMU_OPT_STRING,
1712         },
1713         {
1714             .name = "portal",
1715             .type = QEMU_OPT_STRING,
1716         },
1717         {
1718             .name = "target",
1719             .type = QEMU_OPT_STRING,
1720         },
1721         {
1722             .name = "user",
1723             .type = QEMU_OPT_STRING,
1724         },
1725         {
1726             .name = "password",
1727             .type = QEMU_OPT_STRING,
1728         },
1729         {
1730             .name = "password-secret",
1731             .type = QEMU_OPT_STRING,
1732         },
1733         {
1734             .name = "lun",
1735             .type = QEMU_OPT_NUMBER,
1736         },
1737         {
1738             .name = "initiator-name",
1739             .type = QEMU_OPT_STRING,
1740         },
1741         {
1742             .name = "header-digest",
1743             .type = QEMU_OPT_STRING,
1744         },
1745         {
1746             .name = "timeout",
1747             .type = QEMU_OPT_NUMBER,
1748         },
1749         { /* end of list */ }
1750     },
1751 };
1752 
1753 static void iscsi_save_designator(IscsiLun *lun,
1754                                   struct scsi_inquiry_device_identification *inq_di)
1755 {
1756     struct scsi_inquiry_device_designator *desig, *copy = NULL;
1757 
1758     for (desig = inq_di->designators; desig; desig = desig->next) {
1759         if (desig->association ||
1760             desig->designator_type > SCSI_DESIGNATOR_TYPE_NAA) {
1761             continue;
1762         }
1763         /* NAA works better than T10 vendor ID based designator. */
1764         if (!copy || copy->designator_type < desig->designator_type) {
1765             copy = desig;
1766         }
1767     }
1768     if (copy) {
1769         lun->dd = g_new(struct scsi_inquiry_device_designator, 1);
1770         *lun->dd = *copy;
1771         lun->dd->next = NULL;
1772         lun->dd->designator = g_malloc(copy->designator_length);
1773         memcpy(lun->dd->designator, copy->designator, copy->designator_length);
1774     }
1775 }
1776 
1777 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1778                       Error **errp)
1779 {
1780     IscsiLun *iscsilun = bs->opaque;
1781     struct iscsi_context *iscsi = NULL;
1782     struct scsi_task *task = NULL;
1783     struct scsi_inquiry_standard *inq = NULL;
1784     struct scsi_inquiry_supported_pages *inq_vpd;
1785     char *initiator_name = NULL;
1786     QemuOpts *opts;
1787     Error *local_err = NULL;
1788     const char *transport_name, *portal, *target;
1789 #if LIBISCSI_API_VERSION >= (20160603)
1790     enum iscsi_transport_type transport;
1791 #endif
1792     int i, ret = 0, timeout = 0, lun;
1793 
1794     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1795     qemu_opts_absorb_qdict(opts, options, &local_err);
1796     if (local_err) {
1797         error_propagate(errp, local_err);
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->unallocated_blocks_are_zero = iscsilun->lbprz;
2167     bdi->cluster_size = iscsilun->cluster_size;
2168     return 0;
2169 }
2170 
2171 static void coroutine_fn iscsi_co_invalidate_cache(BlockDriverState *bs,
2172                                                    Error **errp)
2173 {
2174     IscsiLun *iscsilun = bs->opaque;
2175     iscsi_allocmap_invalidate(iscsilun);
2176 }
2177 
2178 static int coroutine_fn iscsi_co_copy_range_from(BlockDriverState *bs,
2179                                                  BdrvChild *src,
2180                                                  uint64_t src_offset,
2181                                                  BdrvChild *dst,
2182                                                  uint64_t dst_offset,
2183                                                  uint64_t bytes,
2184                                                  BdrvRequestFlags read_flags,
2185                                                  BdrvRequestFlags write_flags)
2186 {
2187     return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes,
2188                                  read_flags, write_flags);
2189 }
2190 
2191 static struct scsi_task *iscsi_xcopy_task(int param_len)
2192 {
2193     struct scsi_task *task;
2194 
2195     task = g_new0(struct scsi_task, 1);
2196 
2197     task->cdb[0]     = EXTENDED_COPY;
2198     task->cdb[10]    = (param_len >> 24) & 0xFF;
2199     task->cdb[11]    = (param_len >> 16) & 0xFF;
2200     task->cdb[12]    = (param_len >> 8) & 0xFF;
2201     task->cdb[13]    = param_len & 0xFF;
2202     task->cdb_size   = 16;
2203     task->xfer_dir   = SCSI_XFER_WRITE;
2204     task->expxferlen = param_len;
2205 
2206     return task;
2207 }
2208 
2209 static void iscsi_populate_target_desc(unsigned char *desc, IscsiLun *lun)
2210 {
2211     struct scsi_inquiry_device_designator *dd = lun->dd;
2212 
2213     memset(desc, 0, 32);
2214     desc[0] = 0xE4; /* IDENT_DESCR_TGT_DESCR */
2215     desc[4] = dd->code_set;
2216     desc[5] = (dd->designator_type & 0xF)
2217         | ((dd->association & 3) << 4);
2218     desc[7] = dd->designator_length;
2219     memcpy(desc + 8, dd->designator, MIN(dd->designator_length, 20));
2220 
2221     desc[28] = 0;
2222     desc[29] = (lun->block_size >> 16) & 0xFF;
2223     desc[30] = (lun->block_size >> 8) & 0xFF;
2224     desc[31] = lun->block_size & 0xFF;
2225 }
2226 
2227 static void iscsi_xcopy_desc_hdr(uint8_t *hdr, int dc, int cat, int src_index,
2228                                  int dst_index)
2229 {
2230     hdr[0] = 0x02; /* BLK_TO_BLK_SEG_DESCR */
2231     hdr[1] = ((dc << 1) | cat) & 0xFF;
2232     hdr[2] = (XCOPY_BLK2BLK_SEG_DESC_SIZE >> 8) & 0xFF;
2233     /* don't account for the first 4 bytes in descriptor header*/
2234     hdr[3] = (XCOPY_BLK2BLK_SEG_DESC_SIZE - 4 /* SEG_DESC_SRC_INDEX_OFFSET */) & 0xFF;
2235     hdr[4] = (src_index >> 8) & 0xFF;
2236     hdr[5] = src_index & 0xFF;
2237     hdr[6] = (dst_index >> 8) & 0xFF;
2238     hdr[7] = dst_index & 0xFF;
2239 }
2240 
2241 static void iscsi_xcopy_populate_desc(uint8_t *desc, int dc, int cat,
2242                                       int src_index, int dst_index, int num_blks,
2243                                       uint64_t src_lba, uint64_t dst_lba)
2244 {
2245     iscsi_xcopy_desc_hdr(desc, dc, cat, src_index, dst_index);
2246 
2247     /* The caller should verify the request size */
2248     assert(num_blks < 65536);
2249     desc[10] = (num_blks >> 8) & 0xFF;
2250     desc[11] = num_blks & 0xFF;
2251     desc[12] = (src_lba >> 56) & 0xFF;
2252     desc[13] = (src_lba >> 48) & 0xFF;
2253     desc[14] = (src_lba >> 40) & 0xFF;
2254     desc[15] = (src_lba >> 32) & 0xFF;
2255     desc[16] = (src_lba >> 24) & 0xFF;
2256     desc[17] = (src_lba >> 16) & 0xFF;
2257     desc[18] = (src_lba >> 8) & 0xFF;
2258     desc[19] = src_lba & 0xFF;
2259     desc[20] = (dst_lba >> 56) & 0xFF;
2260     desc[21] = (dst_lba >> 48) & 0xFF;
2261     desc[22] = (dst_lba >> 40) & 0xFF;
2262     desc[23] = (dst_lba >> 32) & 0xFF;
2263     desc[24] = (dst_lba >> 24) & 0xFF;
2264     desc[25] = (dst_lba >> 16) & 0xFF;
2265     desc[26] = (dst_lba >> 8) & 0xFF;
2266     desc[27] = dst_lba & 0xFF;
2267 }
2268 
2269 static void iscsi_xcopy_populate_header(unsigned char *buf, int list_id, int str,
2270                                         int list_id_usage, int prio,
2271                                         int tgt_desc_len,
2272                                         int seg_desc_len, int inline_data_len)
2273 {
2274     buf[0] = list_id;
2275     buf[1] = ((str & 1) << 5) | ((list_id_usage & 3) << 3) | (prio & 7);
2276     buf[2] = (tgt_desc_len >> 8) & 0xFF;
2277     buf[3] = tgt_desc_len & 0xFF;
2278     buf[8] = (seg_desc_len >> 24) & 0xFF;
2279     buf[9] = (seg_desc_len >> 16) & 0xFF;
2280     buf[10] = (seg_desc_len >> 8) & 0xFF;
2281     buf[11] = seg_desc_len & 0xFF;
2282     buf[12] = (inline_data_len >> 24) & 0xFF;
2283     buf[13] = (inline_data_len >> 16) & 0xFF;
2284     buf[14] = (inline_data_len >> 8) & 0xFF;
2285     buf[15] = inline_data_len & 0xFF;
2286 }
2287 
2288 static void iscsi_xcopy_data(struct iscsi_data *data,
2289                              IscsiLun *src, int64_t src_lba,
2290                              IscsiLun *dst, int64_t dst_lba,
2291                              uint16_t num_blocks)
2292 {
2293     uint8_t *buf;
2294     const int src_offset = XCOPY_DESC_OFFSET;
2295     const int dst_offset = XCOPY_DESC_OFFSET + IDENT_DESCR_TGT_DESCR_SIZE;
2296     const int seg_offset = dst_offset + IDENT_DESCR_TGT_DESCR_SIZE;
2297 
2298     data->size = XCOPY_DESC_OFFSET +
2299                  IDENT_DESCR_TGT_DESCR_SIZE * 2 +
2300                  XCOPY_BLK2BLK_SEG_DESC_SIZE;
2301     data->data = g_malloc0(data->size);
2302     buf = data->data;
2303 
2304     /* Initialise the parameter list header */
2305     iscsi_xcopy_populate_header(buf, 1, 0, 2 /* LIST_ID_USAGE_DISCARD */,
2306                                 0, 2 * IDENT_DESCR_TGT_DESCR_SIZE,
2307                                 XCOPY_BLK2BLK_SEG_DESC_SIZE,
2308                                 0);
2309 
2310     /* Initialise CSCD list with one src + one dst descriptor */
2311     iscsi_populate_target_desc(&buf[src_offset], src);
2312     iscsi_populate_target_desc(&buf[dst_offset], dst);
2313 
2314     /* Initialise one segment descriptor */
2315     iscsi_xcopy_populate_desc(&buf[seg_offset], 0, 0, 0, 1, num_blocks,
2316                               src_lba, dst_lba);
2317 }
2318 
2319 static int coroutine_fn iscsi_co_copy_range_to(BlockDriverState *bs,
2320                                                BdrvChild *src,
2321                                                uint64_t src_offset,
2322                                                BdrvChild *dst,
2323                                                uint64_t dst_offset,
2324                                                uint64_t bytes,
2325                                                BdrvRequestFlags read_flags,
2326                                                BdrvRequestFlags write_flags)
2327 {
2328     IscsiLun *dst_lun = dst->bs->opaque;
2329     IscsiLun *src_lun;
2330     struct IscsiTask iscsi_task;
2331     struct iscsi_data data;
2332     int r = 0;
2333     int block_size;
2334 
2335     if (src->bs->drv->bdrv_co_copy_range_to != iscsi_co_copy_range_to) {
2336         return -ENOTSUP;
2337     }
2338     src_lun = src->bs->opaque;
2339 
2340     if (!src_lun->dd || !dst_lun->dd) {
2341         return -ENOTSUP;
2342     }
2343     if (!is_byte_request_lun_aligned(dst_offset, bytes, dst_lun)) {
2344         return -ENOTSUP;
2345     }
2346     if (!is_byte_request_lun_aligned(src_offset, bytes, src_lun)) {
2347         return -ENOTSUP;
2348     }
2349     if (dst_lun->block_size != src_lun->block_size ||
2350         !dst_lun->block_size) {
2351         return -ENOTSUP;
2352     }
2353 
2354     block_size = dst_lun->block_size;
2355     if (bytes / block_size > 65535) {
2356         return -ENOTSUP;
2357     }
2358 
2359     iscsi_xcopy_data(&data,
2360                      src_lun, src_offset / block_size,
2361                      dst_lun, dst_offset / block_size,
2362                      bytes / block_size);
2363 
2364     iscsi_co_init_iscsitask(dst_lun, &iscsi_task);
2365 
2366     qemu_mutex_lock(&dst_lun->mutex);
2367     iscsi_task.task = iscsi_xcopy_task(data.size);
2368 retry:
2369     if (iscsi_scsi_command_async(dst_lun->iscsi, dst_lun->lun,
2370                                  iscsi_task.task, iscsi_co_generic_cb,
2371                                  &data,
2372                                  &iscsi_task) != 0) {
2373         r = -EIO;
2374         goto out_unlock;
2375     }
2376 
2377     iscsi_co_wait_for_task(&iscsi_task, dst_lun);
2378 
2379     if (iscsi_task.do_retry) {
2380         iscsi_task.complete = 0;
2381         goto retry;
2382     }
2383 
2384     if (iscsi_task.status != SCSI_STATUS_GOOD) {
2385         r = iscsi_task.err_code;
2386         goto out_unlock;
2387     }
2388 
2389 out_unlock:
2390 
2391     trace_iscsi_xcopy(src_lun, src_offset, dst_lun, dst_offset, bytes, r);
2392     g_free(iscsi_task.task);
2393     qemu_mutex_unlock(&dst_lun->mutex);
2394     g_free(iscsi_task.err_str);
2395     return r;
2396 }
2397 
2398 
2399 static const char *const iscsi_strong_runtime_opts[] = {
2400     "transport",
2401     "portal",
2402     "target",
2403     "user",
2404     "password",
2405     "password-secret",
2406     "lun",
2407     "initiator-name",
2408     "header-digest",
2409 
2410     NULL
2411 };
2412 
2413 static BlockDriver bdrv_iscsi = {
2414     .format_name     = "iscsi",
2415     .protocol_name   = "iscsi",
2416 
2417     .instance_size          = sizeof(IscsiLun),
2418     .bdrv_parse_filename    = iscsi_parse_filename,
2419     .bdrv_file_open         = iscsi_open,
2420     .bdrv_close             = iscsi_close,
2421     .bdrv_co_create_opts    = bdrv_co_create_opts_simple,
2422     .create_opts            = &bdrv_create_opts_simple,
2423     .bdrv_reopen_prepare    = iscsi_reopen_prepare,
2424     .bdrv_reopen_commit     = iscsi_reopen_commit,
2425     .bdrv_co_invalidate_cache = iscsi_co_invalidate_cache,
2426 
2427     .bdrv_getlength  = iscsi_getlength,
2428     .bdrv_get_info   = iscsi_get_info,
2429     .bdrv_co_truncate    = iscsi_co_truncate,
2430     .bdrv_refresh_limits = iscsi_refresh_limits,
2431 
2432     .bdrv_co_block_status  = iscsi_co_block_status,
2433     .bdrv_co_pdiscard      = iscsi_co_pdiscard,
2434     .bdrv_co_copy_range_from = iscsi_co_copy_range_from,
2435     .bdrv_co_copy_range_to  = iscsi_co_copy_range_to,
2436     .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2437     .bdrv_co_readv         = iscsi_co_readv,
2438     .bdrv_co_writev        = iscsi_co_writev,
2439     .bdrv_co_flush_to_disk = iscsi_co_flush,
2440 
2441 #ifdef __linux__
2442     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
2443 #endif
2444 
2445     .bdrv_detach_aio_context = iscsi_detach_aio_context,
2446     .bdrv_attach_aio_context = iscsi_attach_aio_context,
2447 
2448     .strong_runtime_opts = iscsi_strong_runtime_opts,
2449 };
2450 
2451 #if LIBISCSI_API_VERSION >= (20160603)
2452 static BlockDriver bdrv_iser = {
2453     .format_name     = "iser",
2454     .protocol_name   = "iser",
2455 
2456     .instance_size          = sizeof(IscsiLun),
2457     .bdrv_parse_filename    = iscsi_parse_filename,
2458     .bdrv_file_open         = iscsi_open,
2459     .bdrv_close             = iscsi_close,
2460     .bdrv_co_create_opts    = bdrv_co_create_opts_simple,
2461     .create_opts            = &bdrv_create_opts_simple,
2462     .bdrv_reopen_prepare    = iscsi_reopen_prepare,
2463     .bdrv_reopen_commit     = iscsi_reopen_commit,
2464     .bdrv_co_invalidate_cache  = iscsi_co_invalidate_cache,
2465 
2466     .bdrv_getlength  = iscsi_getlength,
2467     .bdrv_get_info   = iscsi_get_info,
2468     .bdrv_co_truncate    = iscsi_co_truncate,
2469     .bdrv_refresh_limits = iscsi_refresh_limits,
2470 
2471     .bdrv_co_block_status  = iscsi_co_block_status,
2472     .bdrv_co_pdiscard      = iscsi_co_pdiscard,
2473     .bdrv_co_copy_range_from = iscsi_co_copy_range_from,
2474     .bdrv_co_copy_range_to  = iscsi_co_copy_range_to,
2475     .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2476     .bdrv_co_readv         = iscsi_co_readv,
2477     .bdrv_co_writev        = iscsi_co_writev,
2478     .bdrv_co_flush_to_disk = iscsi_co_flush,
2479 
2480 #ifdef __linux__
2481     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
2482 #endif
2483 
2484     .bdrv_detach_aio_context = iscsi_detach_aio_context,
2485     .bdrv_attach_aio_context = iscsi_attach_aio_context,
2486 
2487     .strong_runtime_opts = iscsi_strong_runtime_opts,
2488 };
2489 #endif
2490 
2491 static void iscsi_block_init(void)
2492 {
2493     bdrv_register(&bdrv_iscsi);
2494 #if LIBISCSI_API_VERSION >= (20160603)
2495     bdrv_register(&bdrv_iser);
2496 #endif
2497 }
2498 
2499 block_init(iscsi_block_init);
2500