xref: /openbmc/qemu/block/iscsi.c (revision 75c5bb0b)
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;
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 
725     qemu_mutex_lock(&iscsilun->mutex);
726 retry:
727     if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
728                                   lba, 8 + 16, iscsi_co_generic_cb,
729                                   &iTask) == NULL) {
730         ret = -ENOMEM;
731         goto out_unlock;
732     }
733     iscsi_co_wait_for_task(&iTask, iscsilun);
734 
735     if (iTask.do_retry) {
736         if (iTask.task != NULL) {
737             scsi_free_scsi_task(iTask.task);
738             iTask.task = NULL;
739         }
740         iTask.complete = 0;
741         goto retry;
742     }
743 
744     if (iTask.status != SCSI_STATUS_GOOD) {
745         /* in case the get_lba_status_callout fails (i.e.
746          * because the device is busy or the cmd is not
747          * supported) we pretend all blocks are allocated
748          * for backwards compatibility */
749         error_report("iSCSI GET_LBA_STATUS failed at lba %" PRIu64 ": %s",
750                      lba, iTask.err_str);
751         goto out_unlock;
752     }
753 
754     lbas = scsi_datain_unmarshall(iTask.task);
755     if (lbas == NULL) {
756         ret = -EIO;
757         goto out_unlock;
758     }
759 
760     lbasd = &lbas->descriptors[0];
761 
762     if (lba != lbasd->lba) {
763         ret = -EIO;
764         goto out_unlock;
765     }
766 
767     *pnum = (int64_t) lbasd->num_blocks * iscsilun->block_size;
768 
769     if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
770         lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
771         ret &= ~BDRV_BLOCK_DATA;
772         if (iscsilun->lbprz) {
773             ret |= BDRV_BLOCK_ZERO;
774         }
775     }
776 
777     if (ret & BDRV_BLOCK_ZERO) {
778         iscsi_allocmap_set_unallocated(iscsilun, offset, *pnum);
779     } else {
780         iscsi_allocmap_set_allocated(iscsilun, offset, *pnum);
781     }
782 
783     if (*pnum > bytes) {
784         *pnum = bytes;
785     }
786 out_unlock:
787     qemu_mutex_unlock(&iscsilun->mutex);
788     g_free(iTask.err_str);
789 out:
790     if (iTask.task != NULL) {
791         scsi_free_scsi_task(iTask.task);
792     }
793     if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID && file) {
794         *file = bs;
795     }
796     return ret;
797 }
798 
799 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
800                                        int64_t sector_num, int nb_sectors,
801                                        QEMUIOVector *iov)
802 {
803     IscsiLun *iscsilun = bs->opaque;
804     struct IscsiTask iTask;
805     uint64_t lba;
806     uint32_t num_sectors;
807     int r = 0;
808 
809     if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
810         return -EINVAL;
811     }
812 
813     if (bs->bl.max_transfer) {
814         assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
815     }
816 
817     /* if cache.direct is off and we have a valid entry in our allocation map
818      * we can skip checking the block status and directly return zeroes if
819      * the request falls within an unallocated area */
820     if (iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE,
821                                 nb_sectors * BDRV_SECTOR_SIZE) &&
822         !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE,
823                                      nb_sectors * BDRV_SECTOR_SIZE)) {
824             qemu_iovec_memset(iov, 0, 0x00, iov->size);
825             return 0;
826     }
827 
828     if (nb_sectors >= ISCSI_CHECKALLOC_THRES &&
829         !iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE,
830                                  nb_sectors * BDRV_SECTOR_SIZE) &&
831         !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE,
832                                      nb_sectors * BDRV_SECTOR_SIZE)) {
833         int64_t pnum;
834         /* check the block status from the beginning of the cluster
835          * containing the start sector */
836         int64_t head;
837         int ret;
838 
839         assert(iscsilun->cluster_size);
840         head = (sector_num * BDRV_SECTOR_SIZE) % iscsilun->cluster_size;
841         ret = iscsi_co_block_status(bs, true,
842                                     sector_num * BDRV_SECTOR_SIZE - head,
843                                     BDRV_REQUEST_MAX_BYTES, &pnum, NULL, NULL);
844         if (ret < 0) {
845             return ret;
846         }
847         /* if the whole request falls into an unallocated area we can avoid
848          * reading and directly return zeroes instead */
849         if (ret & BDRV_BLOCK_ZERO &&
850             pnum >= nb_sectors * BDRV_SECTOR_SIZE + head) {
851             qemu_iovec_memset(iov, 0, 0x00, iov->size);
852             return 0;
853         }
854     }
855 
856     lba = sector_qemu2lun(sector_num, iscsilun);
857     num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
858 
859     iscsi_co_init_iscsitask(iscsilun, &iTask);
860     qemu_mutex_lock(&iscsilun->mutex);
861 retry:
862     if (iscsilun->use_16_for_rw) {
863 #if LIBISCSI_API_VERSION >= (20160603)
864         iTask.task = iscsi_read16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
865                                            num_sectors * iscsilun->block_size,
866                                            iscsilun->block_size, 0, 0, 0, 0, 0,
867                                            iscsi_co_generic_cb, &iTask,
868                                            (struct scsi_iovec *)iov->iov, iov->niov);
869     } else {
870         iTask.task = iscsi_read10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
871                                            num_sectors * iscsilun->block_size,
872                                            iscsilun->block_size,
873                                            0, 0, 0, 0, 0,
874                                            iscsi_co_generic_cb, &iTask,
875                                            (struct scsi_iovec *)iov->iov, iov->niov);
876     }
877 #else
878         iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
879                                        num_sectors * iscsilun->block_size,
880                                        iscsilun->block_size, 0, 0, 0, 0, 0,
881                                        iscsi_co_generic_cb, &iTask);
882     } else {
883         iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
884                                        num_sectors * iscsilun->block_size,
885                                        iscsilun->block_size,
886                                        0, 0, 0, 0, 0,
887                                        iscsi_co_generic_cb, &iTask);
888     }
889 #endif
890     if (iTask.task == NULL) {
891         qemu_mutex_unlock(&iscsilun->mutex);
892         return -ENOMEM;
893     }
894 #if LIBISCSI_API_VERSION < (20160603)
895     scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
896 #endif
897 
898     iscsi_co_wait_for_task(&iTask, iscsilun);
899     if (iTask.task != NULL) {
900         scsi_free_scsi_task(iTask.task);
901         iTask.task = NULL;
902     }
903 
904     if (iTask.do_retry) {
905         iTask.complete = 0;
906         goto retry;
907     }
908 
909     if (iTask.status != SCSI_STATUS_GOOD) {
910         error_report("iSCSI READ10/16 failed at lba %" PRIu64 ": %s",
911                      lba, iTask.err_str);
912         r = iTask.err_code;
913     }
914 
915     qemu_mutex_unlock(&iscsilun->mutex);
916     g_free(iTask.err_str);
917     return r;
918 }
919 
920 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
921 {
922     IscsiLun *iscsilun = bs->opaque;
923     struct IscsiTask iTask;
924     int r = 0;
925 
926     iscsi_co_init_iscsitask(iscsilun, &iTask);
927     qemu_mutex_lock(&iscsilun->mutex);
928 retry:
929     if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
930                                       0, iscsi_co_generic_cb, &iTask) == NULL) {
931         qemu_mutex_unlock(&iscsilun->mutex);
932         return -ENOMEM;
933     }
934 
935     iscsi_co_wait_for_task(&iTask, iscsilun);
936 
937     if (iTask.task != NULL) {
938         scsi_free_scsi_task(iTask.task);
939         iTask.task = NULL;
940     }
941 
942     if (iTask.do_retry) {
943         iTask.complete = 0;
944         goto retry;
945     }
946 
947     if (iTask.status != SCSI_STATUS_GOOD) {
948         error_report("iSCSI SYNCHRONIZECACHE10 failed: %s", iTask.err_str);
949         r = iTask.err_code;
950     }
951 
952     qemu_mutex_unlock(&iscsilun->mutex);
953     g_free(iTask.err_str);
954     return r;
955 }
956 
957 #ifdef __linux__
958 /* Called (via iscsi_service) with QemuMutex held.  */
959 static void
960 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
961                      void *command_data, void *opaque)
962 {
963     IscsiAIOCB *acb = opaque;
964 
965     if (status == SCSI_STATUS_CANCELLED) {
966         if (!acb->bh) {
967             acb->status = -ECANCELED;
968             iscsi_schedule_bh(acb);
969         }
970         return;
971     }
972 
973     acb->status = 0;
974     if (status < 0) {
975         error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
976                      iscsi_get_error(iscsi));
977         acb->status = -iscsi_translate_sense(&acb->task->sense);
978     }
979 
980     acb->ioh->driver_status = 0;
981     acb->ioh->host_status   = 0;
982     acb->ioh->resid         = 0;
983     acb->ioh->status        = status;
984 
985 #define SG_ERR_DRIVER_SENSE    0x08
986 
987     if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
988         int ss;
989 
990         acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
991 
992         acb->ioh->sb_len_wr = acb->task->datain.size - 2;
993         ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
994              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_mutex_lock(&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         goto out;
1404     }
1405 
1406     timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1407     iscsi_set_events(iscsilun);
1408 
1409 out:
1410     qemu_mutex_unlock(&iscsilun->mutex);
1411 }
1412 
1413 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1414 {
1415     struct scsi_task *task = NULL;
1416     struct scsi_readcapacity10 *rc10 = NULL;
1417     struct scsi_readcapacity16 *rc16 = NULL;
1418     int retries = ISCSI_CMD_RETRIES;
1419 
1420     do {
1421         if (task != NULL) {
1422             scsi_free_scsi_task(task);
1423             task = NULL;
1424         }
1425 
1426         switch (iscsilun->type) {
1427         case TYPE_DISK:
1428             task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1429             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1430                 rc16 = scsi_datain_unmarshall(task);
1431                 if (rc16 == NULL) {
1432                     error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1433                 } else {
1434                     iscsilun->block_size = rc16->block_length;
1435                     iscsilun->num_blocks = rc16->returned_lba + 1;
1436                     iscsilun->lbpme = !!rc16->lbpme;
1437                     iscsilun->lbprz = !!rc16->lbprz;
1438                     iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1439                 }
1440                 break;
1441             }
1442             if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1443                 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1444                 break;
1445             }
1446             /* Fall through and try READ CAPACITY(10) instead.  */
1447         case TYPE_ROM:
1448             task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1449             if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1450                 rc10 = scsi_datain_unmarshall(task);
1451                 if (rc10 == NULL) {
1452                     error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1453                 } else {
1454                     iscsilun->block_size = rc10->block_size;
1455                     if (rc10->lba == 0) {
1456                         /* blank disk loaded */
1457                         iscsilun->num_blocks = 0;
1458                     } else {
1459                         iscsilun->num_blocks = rc10->lba + 1;
1460                     }
1461                 }
1462             }
1463             break;
1464         default:
1465             return;
1466         }
1467     } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1468              && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1469              && retries-- > 0);
1470 
1471     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1472         error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1473     } else if (!iscsilun->block_size ||
1474                iscsilun->block_size % BDRV_SECTOR_SIZE) {
1475         error_setg(errp, "iSCSI: the target returned an invalid "
1476                    "block size of %d.", iscsilun->block_size);
1477     }
1478     if (task) {
1479         scsi_free_scsi_task(task);
1480     }
1481 }
1482 
1483 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1484                                           int evpd, int pc, void **inq, Error **errp)
1485 {
1486     int full_size;
1487     struct scsi_task *task = NULL;
1488     task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1489     if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1490         goto fail;
1491     }
1492     full_size = scsi_datain_getfullsize(task);
1493     if (full_size > task->datain.size) {
1494         scsi_free_scsi_task(task);
1495 
1496         /* we need more data for the full list */
1497         task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1498         if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1499             goto fail;
1500         }
1501     }
1502 
1503     *inq = scsi_datain_unmarshall(task);
1504     if (*inq == NULL) {
1505         error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1506         goto fail_with_err;
1507     }
1508 
1509     return task;
1510 
1511 fail:
1512     error_setg(errp, "iSCSI: Inquiry command failed : %s",
1513                iscsi_get_error(iscsi));
1514 fail_with_err:
1515     if (task != NULL) {
1516         scsi_free_scsi_task(task);
1517     }
1518     return NULL;
1519 }
1520 
1521 static void iscsi_detach_aio_context(BlockDriverState *bs)
1522 {
1523     IscsiLun *iscsilun = bs->opaque;
1524 
1525     aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1526                        false, NULL, NULL, NULL, NULL);
1527     iscsilun->events = 0;
1528 
1529     if (iscsilun->nop_timer) {
1530         timer_del(iscsilun->nop_timer);
1531         timer_free(iscsilun->nop_timer);
1532         iscsilun->nop_timer = NULL;
1533     }
1534     if (iscsilun->event_timer) {
1535         timer_del(iscsilun->event_timer);
1536         timer_free(iscsilun->event_timer);
1537         iscsilun->event_timer = NULL;
1538     }
1539 }
1540 
1541 static void iscsi_attach_aio_context(BlockDriverState *bs,
1542                                      AioContext *new_context)
1543 {
1544     IscsiLun *iscsilun = bs->opaque;
1545 
1546     iscsilun->aio_context = new_context;
1547     iscsi_set_events(iscsilun);
1548 
1549     /* Set up a timer for sending out iSCSI NOPs */
1550     iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1551                                         QEMU_CLOCK_REALTIME, SCALE_MS,
1552                                         iscsi_nop_timed_event, iscsilun);
1553     timer_mod(iscsilun->nop_timer,
1554               qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1555 
1556     /* Set up a timer for periodic calls to iscsi_set_events and to
1557      * scan for command timeout */
1558     iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1559                                           QEMU_CLOCK_REALTIME, SCALE_MS,
1560                                           iscsi_timed_check_events, iscsilun);
1561     timer_mod(iscsilun->event_timer,
1562               qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1563 }
1564 
1565 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1566 {
1567     struct scsi_task *task;
1568     struct scsi_mode_sense *ms = NULL;
1569     iscsilun->write_protected = false;
1570     iscsilun->dpofua = false;
1571 
1572     task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1573                                  1, SCSI_MODESENSE_PC_CURRENT,
1574                                  0x3F, 0, 255);
1575     if (task == NULL) {
1576         error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1577                      iscsi_get_error(iscsilun->iscsi));
1578         goto out;
1579     }
1580 
1581     if (task->status != SCSI_STATUS_GOOD) {
1582         error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1583         goto out;
1584     }
1585     ms = scsi_datain_unmarshall(task);
1586     if (!ms) {
1587         error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1588                      iscsi_get_error(iscsilun->iscsi));
1589         goto out;
1590     }
1591     iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1592     iscsilun->dpofua          = ms->device_specific_parameter & 0x10;
1593 
1594 out:
1595     if (task) {
1596         scsi_free_scsi_task(task);
1597     }
1598 }
1599 
1600 static void iscsi_parse_iscsi_option(const char *target, QDict *options)
1601 {
1602     QemuOptsList *list;
1603     QemuOpts *opts;
1604     const char *user, *password, *password_secret, *initiator_name,
1605                *header_digest, *timeout;
1606 
1607     list = qemu_find_opts("iscsi");
1608     if (!list) {
1609         return;
1610     }
1611 
1612     opts = qemu_opts_find(list, target);
1613     if (opts == NULL) {
1614         opts = QTAILQ_FIRST(&list->head);
1615         if (!opts) {
1616             return;
1617         }
1618     }
1619 
1620     user = qemu_opt_get(opts, "user");
1621     if (user) {
1622         qdict_set_default_str(options, "user", user);
1623     }
1624 
1625     password = qemu_opt_get(opts, "password");
1626     if (password) {
1627         qdict_set_default_str(options, "password", password);
1628     }
1629 
1630     password_secret = qemu_opt_get(opts, "password-secret");
1631     if (password_secret) {
1632         qdict_set_default_str(options, "password-secret", password_secret);
1633     }
1634 
1635     initiator_name = qemu_opt_get(opts, "initiator-name");
1636     if (initiator_name) {
1637         qdict_set_default_str(options, "initiator-name", initiator_name);
1638     }
1639 
1640     header_digest = qemu_opt_get(opts, "header-digest");
1641     if (header_digest) {
1642         /* -iscsi takes upper case values, but QAPI only supports lower case
1643          * enum constant names, so we have to convert here. */
1644         char *qapi_value = g_ascii_strdown(header_digest, -1);
1645         qdict_set_default_str(options, "header-digest", qapi_value);
1646         g_free(qapi_value);
1647     }
1648 
1649     timeout = qemu_opt_get(opts, "timeout");
1650     if (timeout) {
1651         qdict_set_default_str(options, "timeout", timeout);
1652     }
1653 }
1654 
1655 /*
1656  * We support iscsi url's on the form
1657  * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1658  */
1659 static void iscsi_parse_filename(const char *filename, QDict *options,
1660                                  Error **errp)
1661 {
1662     struct iscsi_url *iscsi_url;
1663     const char *transport_name;
1664     char *lun_str;
1665 
1666     iscsi_url = iscsi_parse_full_url(NULL, filename);
1667     if (iscsi_url == NULL) {
1668         error_setg(errp, "Failed to parse URL : %s", filename);
1669         return;
1670     }
1671 
1672 #if LIBISCSI_API_VERSION >= (20160603)
1673     switch (iscsi_url->transport) {
1674     case TCP_TRANSPORT:
1675         transport_name = "tcp";
1676         break;
1677     case ISER_TRANSPORT:
1678         transport_name = "iser";
1679         break;
1680     default:
1681         error_setg(errp, "Unknown transport type (%d)",
1682                    iscsi_url->transport);
1683         return;
1684     }
1685 #else
1686     transport_name = "tcp";
1687 #endif
1688 
1689     qdict_set_default_str(options, "transport", transport_name);
1690     qdict_set_default_str(options, "portal", iscsi_url->portal);
1691     qdict_set_default_str(options, "target", iscsi_url->target);
1692 
1693     lun_str = g_strdup_printf("%d", iscsi_url->lun);
1694     qdict_set_default_str(options, "lun", lun_str);
1695     g_free(lun_str);
1696 
1697     /* User/password from -iscsi take precedence over those from the URL */
1698     iscsi_parse_iscsi_option(iscsi_url->target, options);
1699 
1700     if (iscsi_url->user[0] != '\0') {
1701         qdict_set_default_str(options, "user", iscsi_url->user);
1702         qdict_set_default_str(options, "password", iscsi_url->passwd);
1703     }
1704 
1705     iscsi_destroy_url(iscsi_url);
1706 }
1707 
1708 static QemuOptsList runtime_opts = {
1709     .name = "iscsi",
1710     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1711     .desc = {
1712         {
1713             .name = "transport",
1714             .type = QEMU_OPT_STRING,
1715         },
1716         {
1717             .name = "portal",
1718             .type = QEMU_OPT_STRING,
1719         },
1720         {
1721             .name = "target",
1722             .type = QEMU_OPT_STRING,
1723         },
1724         {
1725             .name = "user",
1726             .type = QEMU_OPT_STRING,
1727         },
1728         {
1729             .name = "password",
1730             .type = QEMU_OPT_STRING,
1731         },
1732         {
1733             .name = "password-secret",
1734             .type = QEMU_OPT_STRING,
1735         },
1736         {
1737             .name = "lun",
1738             .type = QEMU_OPT_NUMBER,
1739         },
1740         {
1741             .name = "initiator-name",
1742             .type = QEMU_OPT_STRING,
1743         },
1744         {
1745             .name = "header-digest",
1746             .type = QEMU_OPT_STRING,
1747         },
1748         {
1749             .name = "timeout",
1750             .type = QEMU_OPT_NUMBER,
1751         },
1752         { /* end of list */ }
1753     },
1754 };
1755 
1756 static void iscsi_save_designator(IscsiLun *lun,
1757                                   struct scsi_inquiry_device_identification *inq_di)
1758 {
1759     struct scsi_inquiry_device_designator *desig, *copy = NULL;
1760 
1761     for (desig = inq_di->designators; desig; desig = desig->next) {
1762         if (desig->association ||
1763             desig->designator_type > SCSI_DESIGNATOR_TYPE_NAA) {
1764             continue;
1765         }
1766         /* NAA works better than T10 vendor ID based designator. */
1767         if (!copy || copy->designator_type < desig->designator_type) {
1768             copy = desig;
1769         }
1770     }
1771     if (copy) {
1772         lun->dd = g_new(struct scsi_inquiry_device_designator, 1);
1773         *lun->dd = *copy;
1774         lun->dd->next = NULL;
1775         lun->dd->designator = g_malloc(copy->designator_length);
1776         memcpy(lun->dd->designator, copy->designator, copy->designator_length);
1777     }
1778 }
1779 
1780 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1781                       Error **errp)
1782 {
1783     IscsiLun *iscsilun = bs->opaque;
1784     struct iscsi_context *iscsi = NULL;
1785     struct scsi_task *task = NULL;
1786     struct scsi_inquiry_standard *inq = NULL;
1787     struct scsi_inquiry_supported_pages *inq_vpd;
1788     char *initiator_name = NULL;
1789     QemuOpts *opts;
1790     Error *local_err = NULL;
1791     const char *transport_name, *portal, *target;
1792 #if LIBISCSI_API_VERSION >= (20160603)
1793     enum iscsi_transport_type transport;
1794 #endif
1795     int i, ret = 0, timeout = 0, lun;
1796 
1797     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1798     qemu_opts_absorb_qdict(opts, options, &local_err);
1799     if (local_err) {
1800         error_propagate(errp, local_err);
1801         ret = -EINVAL;
1802         goto out;
1803     }
1804 
1805     transport_name = qemu_opt_get(opts, "transport");
1806     portal = qemu_opt_get(opts, "portal");
1807     target = qemu_opt_get(opts, "target");
1808     lun = qemu_opt_get_number(opts, "lun", 0);
1809 
1810     if (!transport_name || !portal || !target) {
1811         error_setg(errp, "Need all of transport, portal and target options");
1812         ret = -EINVAL;
1813         goto out;
1814     }
1815 
1816     if (!strcmp(transport_name, "tcp")) {
1817 #if LIBISCSI_API_VERSION >= (20160603)
1818         transport = TCP_TRANSPORT;
1819     } else if (!strcmp(transport_name, "iser")) {
1820         transport = ISER_TRANSPORT;
1821 #else
1822         /* TCP is what older libiscsi versions always use */
1823 #endif
1824     } else {
1825         error_setg(errp, "Unknown transport: %s", transport_name);
1826         ret = -EINVAL;
1827         goto out;
1828     }
1829 
1830     memset(iscsilun, 0, sizeof(IscsiLun));
1831 
1832     initiator_name = get_initiator_name(opts);
1833 
1834     iscsi = iscsi_create_context(initiator_name);
1835     if (iscsi == NULL) {
1836         error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1837         ret = -ENOMEM;
1838         goto out;
1839     }
1840 #if LIBISCSI_API_VERSION >= (20160603)
1841     if (iscsi_init_transport(iscsi, transport)) {
1842         error_setg(errp, ("Error initializing transport."));
1843         ret = -EINVAL;
1844         goto out;
1845     }
1846 #endif
1847     if (iscsi_set_targetname(iscsi, target)) {
1848         error_setg(errp, "iSCSI: Failed to set target name.");
1849         ret = -EINVAL;
1850         goto out;
1851     }
1852 
1853     /* check if we got CHAP username/password via the options */
1854     apply_chap(iscsi, opts, &local_err);
1855     if (local_err != NULL) {
1856         error_propagate(errp, local_err);
1857         ret = -EINVAL;
1858         goto out;
1859     }
1860 
1861     if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1862         error_setg(errp, "iSCSI: Failed to set session type to normal.");
1863         ret = -EINVAL;
1864         goto out;
1865     }
1866 
1867     /* check if we got HEADER_DIGEST via the options */
1868     apply_header_digest(iscsi, opts, &local_err);
1869     if (local_err != NULL) {
1870         error_propagate(errp, local_err);
1871         ret = -EINVAL;
1872         goto out;
1873     }
1874 
1875     /* timeout handling is broken in libiscsi before 1.15.0 */
1876     timeout = qemu_opt_get_number(opts, "timeout", 0);
1877 #if LIBISCSI_API_VERSION >= 20150621
1878     iscsi_set_timeout(iscsi, timeout);
1879 #else
1880     if (timeout) {
1881         warn_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1882     }
1883 #endif
1884 
1885     if (iscsi_full_connect_sync(iscsi, portal, lun) != 0) {
1886         error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1887             iscsi_get_error(iscsi));
1888         ret = -EINVAL;
1889         goto out;
1890     }
1891 
1892     iscsilun->iscsi = iscsi;
1893     iscsilun->aio_context = bdrv_get_aio_context(bs);
1894     iscsilun->lun = lun;
1895     iscsilun->has_write_same = true;
1896 
1897     task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1898                             (void **) &inq, errp);
1899     if (task == NULL) {
1900         ret = -EINVAL;
1901         goto out;
1902     }
1903     iscsilun->type = inq->periperal_device_type;
1904     scsi_free_scsi_task(task);
1905     task = NULL;
1906 
1907     iscsi_modesense_sync(iscsilun);
1908     if (iscsilun->dpofua) {
1909         bs->supported_write_flags = BDRV_REQ_FUA;
1910     }
1911 
1912     /* Check the write protect flag of the LUN if we want to write */
1913     if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1914         iscsilun->write_protected) {
1915         ret = bdrv_apply_auto_read_only(bs, "LUN is write protected", errp);
1916         if (ret < 0) {
1917             goto out;
1918         }
1919         flags &= ~BDRV_O_RDWR;
1920     }
1921 
1922     iscsi_readcapacity_sync(iscsilun, &local_err);
1923     if (local_err != NULL) {
1924         error_propagate(errp, local_err);
1925         ret = -EINVAL;
1926         goto out;
1927     }
1928     bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1929 
1930     /* We don't have any emulation for devices other than disks and CD-ROMs, so
1931      * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1932      * will try to read from the device to guess the image format.
1933      */
1934     if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1935         bs->sg = true;
1936     }
1937 
1938     task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1939                             SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1940                             (void **) &inq_vpd, errp);
1941     if (task == NULL) {
1942         ret = -EINVAL;
1943         goto out;
1944     }
1945     for (i = 0; i < inq_vpd->num_pages; i++) {
1946         struct scsi_task *inq_task;
1947         struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1948         struct scsi_inquiry_block_limits *inq_bl;
1949         struct scsi_inquiry_device_identification *inq_di;
1950         switch (inq_vpd->pages[i]) {
1951         case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1952             inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1953                                         SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1954                                         (void **) &inq_lbp, errp);
1955             if (inq_task == NULL) {
1956                 ret = -EINVAL;
1957                 goto out;
1958             }
1959             memcpy(&iscsilun->lbp, inq_lbp,
1960                    sizeof(struct scsi_inquiry_logical_block_provisioning));
1961             scsi_free_scsi_task(inq_task);
1962             break;
1963         case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1964             inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1965                                     SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1966                                     (void **) &inq_bl, errp);
1967             if (inq_task == NULL) {
1968                 ret = -EINVAL;
1969                 goto out;
1970             }
1971             memcpy(&iscsilun->bl, inq_bl,
1972                    sizeof(struct scsi_inquiry_block_limits));
1973             scsi_free_scsi_task(inq_task);
1974             break;
1975         case SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION:
1976             inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1977                                     SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION,
1978                                     (void **) &inq_di, errp);
1979             if (inq_task == NULL) {
1980                 ret = -EINVAL;
1981                 goto out;
1982             }
1983             iscsi_save_designator(iscsilun, inq_di);
1984             scsi_free_scsi_task(inq_task);
1985             break;
1986         default:
1987             break;
1988         }
1989     }
1990     scsi_free_scsi_task(task);
1991     task = NULL;
1992 
1993     qemu_mutex_init(&iscsilun->mutex);
1994     iscsi_attach_aio_context(bs, iscsilun->aio_context);
1995 
1996     /* Guess the internal cluster (page) size of the iscsi target by the means
1997      * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1998      * reasonable size */
1999     if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
2000         iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
2001         iscsilun->cluster_size = iscsilun->bl.opt_unmap_gran *
2002             iscsilun->block_size;
2003         if (iscsilun->lbprz) {
2004             ret = iscsi_allocmap_init(iscsilun, bs->open_flags);
2005         }
2006     }
2007 
2008     if (iscsilun->lbprz && iscsilun->lbp.lbpws) {
2009         bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
2010     }
2011 
2012 out:
2013     qemu_opts_del(opts);
2014     g_free(initiator_name);
2015     if (task != NULL) {
2016         scsi_free_scsi_task(task);
2017     }
2018 
2019     if (ret) {
2020         if (iscsi != NULL) {
2021             if (iscsi_is_logged_in(iscsi)) {
2022                 iscsi_logout_sync(iscsi);
2023             }
2024             iscsi_destroy_context(iscsi);
2025         }
2026         memset(iscsilun, 0, sizeof(IscsiLun));
2027     }
2028 
2029     return ret;
2030 }
2031 
2032 static void iscsi_close(BlockDriverState *bs)
2033 {
2034     IscsiLun *iscsilun = bs->opaque;
2035     struct iscsi_context *iscsi = iscsilun->iscsi;
2036 
2037     iscsi_detach_aio_context(bs);
2038     if (iscsi_is_logged_in(iscsi)) {
2039         iscsi_logout_sync(iscsi);
2040     }
2041     iscsi_destroy_context(iscsi);
2042     if (iscsilun->dd) {
2043         g_free(iscsilun->dd->designator);
2044         g_free(iscsilun->dd);
2045     }
2046     g_free(iscsilun->zeroblock);
2047     iscsi_allocmap_free(iscsilun);
2048     qemu_mutex_destroy(&iscsilun->mutex);
2049     memset(iscsilun, 0, sizeof(IscsiLun));
2050 }
2051 
2052 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
2053 {
2054     /* We don't actually refresh here, but just return data queried in
2055      * iscsi_open(): iscsi targets don't change their limits. */
2056 
2057     IscsiLun *iscsilun = bs->opaque;
2058     uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
2059     unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size);
2060 
2061     assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bs->sg);
2062 
2063     bs->bl.request_alignment = block_size;
2064 
2065     if (iscsilun->bl.max_xfer_len) {
2066         max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
2067     }
2068 
2069     if (max_xfer_len * block_size < INT_MAX) {
2070         bs->bl.max_transfer = max_xfer_len * iscsilun->block_size;
2071     }
2072 
2073     if (iscsilun->lbp.lbpu) {
2074         if (iscsilun->bl.max_unmap < 0xffffffff / block_size) {
2075             bs->bl.max_pdiscard =
2076                 iscsilun->bl.max_unmap * iscsilun->block_size;
2077         }
2078         bs->bl.pdiscard_alignment =
2079             iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2080     } else {
2081         bs->bl.pdiscard_alignment = iscsilun->block_size;
2082     }
2083 
2084     if (iscsilun->bl.max_ws_len < 0xffffffff / block_size) {
2085         bs->bl.max_pwrite_zeroes =
2086             iscsilun->bl.max_ws_len * iscsilun->block_size;
2087     }
2088     if (iscsilun->lbp.lbpws) {
2089         bs->bl.pwrite_zeroes_alignment =
2090             iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2091     } else {
2092         bs->bl.pwrite_zeroes_alignment = iscsilun->block_size;
2093     }
2094     if (iscsilun->bl.opt_xfer_len &&
2095         iscsilun->bl.opt_xfer_len < INT_MAX / block_size) {
2096         bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len *
2097                                         iscsilun->block_size);
2098     }
2099 }
2100 
2101 /* Note that this will not re-establish a connection with an iSCSI target - it
2102  * is effectively a NOP.  */
2103 static int iscsi_reopen_prepare(BDRVReopenState *state,
2104                                 BlockReopenQueue *queue, Error **errp)
2105 {
2106     IscsiLun *iscsilun = state->bs->opaque;
2107 
2108     if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
2109         error_setg(errp, "Cannot open a write protected LUN as read-write");
2110         return -EACCES;
2111     }
2112     return 0;
2113 }
2114 
2115 static void iscsi_reopen_commit(BDRVReopenState *reopen_state)
2116 {
2117     IscsiLun *iscsilun = reopen_state->bs->opaque;
2118 
2119     /* the cache.direct status might have changed */
2120     if (iscsilun->allocmap != NULL) {
2121         iscsi_allocmap_init(iscsilun, reopen_state->flags);
2122     }
2123 }
2124 
2125 static int coroutine_fn iscsi_co_truncate(BlockDriverState *bs, int64_t offset,
2126                                           bool exact, PreallocMode prealloc,
2127                                           Error **errp)
2128 {
2129     IscsiLun *iscsilun = bs->opaque;
2130     int64_t cur_length;
2131     Error *local_err = NULL;
2132 
2133     if (prealloc != PREALLOC_MODE_OFF) {
2134         error_setg(errp, "Unsupported preallocation mode '%s'",
2135                    PreallocMode_str(prealloc));
2136         return -ENOTSUP;
2137     }
2138 
2139     if (iscsilun->type != TYPE_DISK) {
2140         error_setg(errp, "Cannot resize non-disk iSCSI devices");
2141         return -ENOTSUP;
2142     }
2143 
2144     iscsi_readcapacity_sync(iscsilun, &local_err);
2145     if (local_err != NULL) {
2146         error_propagate(errp, local_err);
2147         return -EIO;
2148     }
2149 
2150     cur_length = iscsi_getlength(bs);
2151     if (offset != cur_length && exact) {
2152         error_setg(errp, "Cannot resize iSCSI devices");
2153         return -ENOTSUP;
2154     } else if (offset > cur_length) {
2155         error_setg(errp, "Cannot grow iSCSI devices");
2156         return -EINVAL;
2157     }
2158 
2159     if (iscsilun->allocmap != NULL) {
2160         iscsi_allocmap_init(iscsilun, bs->open_flags);
2161     }
2162 
2163     return 0;
2164 }
2165 
2166 static int coroutine_fn iscsi_co_create_opts(const char *filename, QemuOpts *opts,
2167                                              Error **errp)
2168 {
2169     int ret = 0;
2170     int64_t total_size = 0;
2171     BlockDriverState *bs;
2172     IscsiLun *iscsilun = NULL;
2173     QDict *bs_options;
2174     Error *local_err = NULL;
2175 
2176     bs = bdrv_new();
2177 
2178     /* Read out options */
2179     total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2180                               BDRV_SECTOR_SIZE);
2181     bs->opaque = g_new0(struct IscsiLun, 1);
2182     iscsilun = bs->opaque;
2183 
2184     bs_options = qdict_new();
2185     iscsi_parse_filename(filename, bs_options, &local_err);
2186     if (local_err) {
2187         error_propagate(errp, local_err);
2188         ret = -EINVAL;
2189     } else {
2190         ret = iscsi_open(bs, bs_options, 0, NULL);
2191     }
2192     qobject_unref(bs_options);
2193 
2194     if (ret != 0) {
2195         goto out;
2196     }
2197     iscsi_detach_aio_context(bs);
2198     if (iscsilun->type != TYPE_DISK) {
2199         ret = -ENODEV;
2200         goto out;
2201     }
2202     if (bs->total_sectors < total_size) {
2203         ret = -ENOSPC;
2204         goto out;
2205     }
2206 
2207     ret = 0;
2208 out:
2209     if (iscsilun->iscsi != NULL) {
2210         iscsi_destroy_context(iscsilun->iscsi);
2211     }
2212     g_free(bs->opaque);
2213     bs->opaque = NULL;
2214     bdrv_unref(bs);
2215     return ret;
2216 }
2217 
2218 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2219 {
2220     IscsiLun *iscsilun = bs->opaque;
2221     bdi->unallocated_blocks_are_zero = iscsilun->lbprz;
2222     bdi->cluster_size = iscsilun->cluster_size;
2223     return 0;
2224 }
2225 
2226 static void coroutine_fn iscsi_co_invalidate_cache(BlockDriverState *bs,
2227                                                    Error **errp)
2228 {
2229     IscsiLun *iscsilun = bs->opaque;
2230     iscsi_allocmap_invalidate(iscsilun);
2231 }
2232 
2233 static int coroutine_fn iscsi_co_copy_range_from(BlockDriverState *bs,
2234                                                  BdrvChild *src,
2235                                                  uint64_t src_offset,
2236                                                  BdrvChild *dst,
2237                                                  uint64_t dst_offset,
2238                                                  uint64_t bytes,
2239                                                  BdrvRequestFlags read_flags,
2240                                                  BdrvRequestFlags write_flags)
2241 {
2242     return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes,
2243                                  read_flags, write_flags);
2244 }
2245 
2246 static struct scsi_task *iscsi_xcopy_task(int param_len)
2247 {
2248     struct scsi_task *task;
2249 
2250     task = g_new0(struct scsi_task, 1);
2251 
2252     task->cdb[0]     = EXTENDED_COPY;
2253     task->cdb[10]    = (param_len >> 24) & 0xFF;
2254     task->cdb[11]    = (param_len >> 16) & 0xFF;
2255     task->cdb[12]    = (param_len >> 8) & 0xFF;
2256     task->cdb[13]    = param_len & 0xFF;
2257     task->cdb_size   = 16;
2258     task->xfer_dir   = SCSI_XFER_WRITE;
2259     task->expxferlen = param_len;
2260 
2261     return task;
2262 }
2263 
2264 static void iscsi_populate_target_desc(unsigned char *desc, IscsiLun *lun)
2265 {
2266     struct scsi_inquiry_device_designator *dd = lun->dd;
2267 
2268     memset(desc, 0, 32);
2269     desc[0] = 0xE4; /* IDENT_DESCR_TGT_DESCR */
2270     desc[4] = dd->code_set;
2271     desc[5] = (dd->designator_type & 0xF)
2272         | ((dd->association & 3) << 4);
2273     desc[7] = dd->designator_length;
2274     memcpy(desc + 8, dd->designator, MIN(dd->designator_length, 20));
2275 
2276     desc[28] = 0;
2277     desc[29] = (lun->block_size >> 16) & 0xFF;
2278     desc[30] = (lun->block_size >> 8) & 0xFF;
2279     desc[31] = lun->block_size & 0xFF;
2280 }
2281 
2282 static void iscsi_xcopy_desc_hdr(uint8_t *hdr, int dc, int cat, int src_index,
2283                                  int dst_index)
2284 {
2285     hdr[0] = 0x02; /* BLK_TO_BLK_SEG_DESCR */
2286     hdr[1] = ((dc << 1) | cat) & 0xFF;
2287     hdr[2] = (XCOPY_BLK2BLK_SEG_DESC_SIZE >> 8) & 0xFF;
2288     /* don't account for the first 4 bytes in descriptor header*/
2289     hdr[3] = (XCOPY_BLK2BLK_SEG_DESC_SIZE - 4 /* SEG_DESC_SRC_INDEX_OFFSET */) & 0xFF;
2290     hdr[4] = (src_index >> 8) & 0xFF;
2291     hdr[5] = src_index & 0xFF;
2292     hdr[6] = (dst_index >> 8) & 0xFF;
2293     hdr[7] = dst_index & 0xFF;
2294 }
2295 
2296 static void iscsi_xcopy_populate_desc(uint8_t *desc, int dc, int cat,
2297                                       int src_index, int dst_index, int num_blks,
2298                                       uint64_t src_lba, uint64_t dst_lba)
2299 {
2300     iscsi_xcopy_desc_hdr(desc, dc, cat, src_index, dst_index);
2301 
2302     /* The caller should verify the request size */
2303     assert(num_blks < 65536);
2304     desc[10] = (num_blks >> 8) & 0xFF;
2305     desc[11] = num_blks & 0xFF;
2306     desc[12] = (src_lba >> 56) & 0xFF;
2307     desc[13] = (src_lba >> 48) & 0xFF;
2308     desc[14] = (src_lba >> 40) & 0xFF;
2309     desc[15] = (src_lba >> 32) & 0xFF;
2310     desc[16] = (src_lba >> 24) & 0xFF;
2311     desc[17] = (src_lba >> 16) & 0xFF;
2312     desc[18] = (src_lba >> 8) & 0xFF;
2313     desc[19] = src_lba & 0xFF;
2314     desc[20] = (dst_lba >> 56) & 0xFF;
2315     desc[21] = (dst_lba >> 48) & 0xFF;
2316     desc[22] = (dst_lba >> 40) & 0xFF;
2317     desc[23] = (dst_lba >> 32) & 0xFF;
2318     desc[24] = (dst_lba >> 24) & 0xFF;
2319     desc[25] = (dst_lba >> 16) & 0xFF;
2320     desc[26] = (dst_lba >> 8) & 0xFF;
2321     desc[27] = dst_lba & 0xFF;
2322 }
2323 
2324 static void iscsi_xcopy_populate_header(unsigned char *buf, int list_id, int str,
2325                                         int list_id_usage, int prio,
2326                                         int tgt_desc_len,
2327                                         int seg_desc_len, int inline_data_len)
2328 {
2329     buf[0] = list_id;
2330     buf[1] = ((str & 1) << 5) | ((list_id_usage & 3) << 3) | (prio & 7);
2331     buf[2] = (tgt_desc_len >> 8) & 0xFF;
2332     buf[3] = tgt_desc_len & 0xFF;
2333     buf[8] = (seg_desc_len >> 24) & 0xFF;
2334     buf[9] = (seg_desc_len >> 16) & 0xFF;
2335     buf[10] = (seg_desc_len >> 8) & 0xFF;
2336     buf[11] = seg_desc_len & 0xFF;
2337     buf[12] = (inline_data_len >> 24) & 0xFF;
2338     buf[13] = (inline_data_len >> 16) & 0xFF;
2339     buf[14] = (inline_data_len >> 8) & 0xFF;
2340     buf[15] = inline_data_len & 0xFF;
2341 }
2342 
2343 static void iscsi_xcopy_data(struct iscsi_data *data,
2344                              IscsiLun *src, int64_t src_lba,
2345                              IscsiLun *dst, int64_t dst_lba,
2346                              uint16_t num_blocks)
2347 {
2348     uint8_t *buf;
2349     const int src_offset = XCOPY_DESC_OFFSET;
2350     const int dst_offset = XCOPY_DESC_OFFSET + IDENT_DESCR_TGT_DESCR_SIZE;
2351     const int seg_offset = dst_offset + IDENT_DESCR_TGT_DESCR_SIZE;
2352 
2353     data->size = XCOPY_DESC_OFFSET +
2354                  IDENT_DESCR_TGT_DESCR_SIZE * 2 +
2355                  XCOPY_BLK2BLK_SEG_DESC_SIZE;
2356     data->data = g_malloc0(data->size);
2357     buf = data->data;
2358 
2359     /* Initialise the parameter list header */
2360     iscsi_xcopy_populate_header(buf, 1, 0, 2 /* LIST_ID_USAGE_DISCARD */,
2361                                 0, 2 * IDENT_DESCR_TGT_DESCR_SIZE,
2362                                 XCOPY_BLK2BLK_SEG_DESC_SIZE,
2363                                 0);
2364 
2365     /* Initialise CSCD list with one src + one dst descriptor */
2366     iscsi_populate_target_desc(&buf[src_offset], src);
2367     iscsi_populate_target_desc(&buf[dst_offset], dst);
2368 
2369     /* Initialise one segment descriptor */
2370     iscsi_xcopy_populate_desc(&buf[seg_offset], 0, 0, 0, 1, num_blocks,
2371                               src_lba, dst_lba);
2372 }
2373 
2374 static int coroutine_fn iscsi_co_copy_range_to(BlockDriverState *bs,
2375                                                BdrvChild *src,
2376                                                uint64_t src_offset,
2377                                                BdrvChild *dst,
2378                                                uint64_t dst_offset,
2379                                                uint64_t bytes,
2380                                                BdrvRequestFlags read_flags,
2381                                                BdrvRequestFlags write_flags)
2382 {
2383     IscsiLun *dst_lun = dst->bs->opaque;
2384     IscsiLun *src_lun;
2385     struct IscsiTask iscsi_task;
2386     struct iscsi_data data;
2387     int r = 0;
2388     int block_size;
2389 
2390     if (src->bs->drv->bdrv_co_copy_range_to != iscsi_co_copy_range_to) {
2391         return -ENOTSUP;
2392     }
2393     src_lun = src->bs->opaque;
2394 
2395     if (!src_lun->dd || !dst_lun->dd) {
2396         return -ENOTSUP;
2397     }
2398     if (!is_byte_request_lun_aligned(dst_offset, bytes, dst_lun)) {
2399         return -ENOTSUP;
2400     }
2401     if (!is_byte_request_lun_aligned(src_offset, bytes, src_lun)) {
2402         return -ENOTSUP;
2403     }
2404     if (dst_lun->block_size != src_lun->block_size ||
2405         !dst_lun->block_size) {
2406         return -ENOTSUP;
2407     }
2408 
2409     block_size = dst_lun->block_size;
2410     if (bytes / block_size > 65535) {
2411         return -ENOTSUP;
2412     }
2413 
2414     iscsi_xcopy_data(&data,
2415                      src_lun, src_offset / block_size,
2416                      dst_lun, dst_offset / block_size,
2417                      bytes / block_size);
2418 
2419     iscsi_co_init_iscsitask(dst_lun, &iscsi_task);
2420 
2421     qemu_mutex_lock(&dst_lun->mutex);
2422     iscsi_task.task = iscsi_xcopy_task(data.size);
2423 retry:
2424     if (iscsi_scsi_command_async(dst_lun->iscsi, dst_lun->lun,
2425                                  iscsi_task.task, iscsi_co_generic_cb,
2426                                  &data,
2427                                  &iscsi_task) != 0) {
2428         r = -EIO;
2429         goto out_unlock;
2430     }
2431 
2432     iscsi_co_wait_for_task(&iscsi_task, dst_lun);
2433 
2434     if (iscsi_task.do_retry) {
2435         iscsi_task.complete = 0;
2436         goto retry;
2437     }
2438 
2439     if (iscsi_task.status != SCSI_STATUS_GOOD) {
2440         r = iscsi_task.err_code;
2441         goto out_unlock;
2442     }
2443 
2444 out_unlock:
2445 
2446     trace_iscsi_xcopy(src_lun, src_offset, dst_lun, dst_offset, bytes, r);
2447     g_free(iscsi_task.task);
2448     qemu_mutex_unlock(&dst_lun->mutex);
2449     g_free(iscsi_task.err_str);
2450     return r;
2451 }
2452 
2453 static QemuOptsList iscsi_create_opts = {
2454     .name = "iscsi-create-opts",
2455     .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
2456     .desc = {
2457         {
2458             .name = BLOCK_OPT_SIZE,
2459             .type = QEMU_OPT_SIZE,
2460             .help = "Virtual disk size"
2461         },
2462         { /* end of list */ }
2463     }
2464 };
2465 
2466 static const char *const iscsi_strong_runtime_opts[] = {
2467     "transport",
2468     "portal",
2469     "target",
2470     "user",
2471     "password",
2472     "password-secret",
2473     "lun",
2474     "initiator-name",
2475     "header-digest",
2476 
2477     NULL
2478 };
2479 
2480 static BlockDriver bdrv_iscsi = {
2481     .format_name     = "iscsi",
2482     .protocol_name   = "iscsi",
2483 
2484     .instance_size          = sizeof(IscsiLun),
2485     .bdrv_parse_filename    = iscsi_parse_filename,
2486     .bdrv_file_open         = iscsi_open,
2487     .bdrv_close             = iscsi_close,
2488     .bdrv_co_create_opts    = iscsi_co_create_opts,
2489     .create_opts            = &iscsi_create_opts,
2490     .bdrv_reopen_prepare    = iscsi_reopen_prepare,
2491     .bdrv_reopen_commit     = iscsi_reopen_commit,
2492     .bdrv_co_invalidate_cache = iscsi_co_invalidate_cache,
2493 
2494     .bdrv_getlength  = iscsi_getlength,
2495     .bdrv_get_info   = iscsi_get_info,
2496     .bdrv_co_truncate    = iscsi_co_truncate,
2497     .bdrv_refresh_limits = iscsi_refresh_limits,
2498 
2499     .bdrv_co_block_status  = iscsi_co_block_status,
2500     .bdrv_co_pdiscard      = iscsi_co_pdiscard,
2501     .bdrv_co_copy_range_from = iscsi_co_copy_range_from,
2502     .bdrv_co_copy_range_to  = iscsi_co_copy_range_to,
2503     .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2504     .bdrv_co_readv         = iscsi_co_readv,
2505     .bdrv_co_writev        = iscsi_co_writev,
2506     .bdrv_co_flush_to_disk = iscsi_co_flush,
2507 
2508 #ifdef __linux__
2509     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
2510 #endif
2511 
2512     .bdrv_detach_aio_context = iscsi_detach_aio_context,
2513     .bdrv_attach_aio_context = iscsi_attach_aio_context,
2514 
2515     .strong_runtime_opts = iscsi_strong_runtime_opts,
2516 };
2517 
2518 #if LIBISCSI_API_VERSION >= (20160603)
2519 static BlockDriver bdrv_iser = {
2520     .format_name     = "iser",
2521     .protocol_name   = "iser",
2522 
2523     .instance_size          = sizeof(IscsiLun),
2524     .bdrv_parse_filename    = iscsi_parse_filename,
2525     .bdrv_file_open         = iscsi_open,
2526     .bdrv_close             = iscsi_close,
2527     .bdrv_co_create_opts    = iscsi_co_create_opts,
2528     .create_opts            = &iscsi_create_opts,
2529     .bdrv_reopen_prepare    = iscsi_reopen_prepare,
2530     .bdrv_reopen_commit     = iscsi_reopen_commit,
2531     .bdrv_co_invalidate_cache  = iscsi_co_invalidate_cache,
2532 
2533     .bdrv_getlength  = iscsi_getlength,
2534     .bdrv_get_info   = iscsi_get_info,
2535     .bdrv_co_truncate    = iscsi_co_truncate,
2536     .bdrv_refresh_limits = iscsi_refresh_limits,
2537 
2538     .bdrv_co_block_status  = iscsi_co_block_status,
2539     .bdrv_co_pdiscard      = iscsi_co_pdiscard,
2540     .bdrv_co_copy_range_from = iscsi_co_copy_range_from,
2541     .bdrv_co_copy_range_to  = iscsi_co_copy_range_to,
2542     .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2543     .bdrv_co_readv         = iscsi_co_readv,
2544     .bdrv_co_writev        = iscsi_co_writev,
2545     .bdrv_co_flush_to_disk = iscsi_co_flush,
2546 
2547 #ifdef __linux__
2548     .bdrv_aio_ioctl   = iscsi_aio_ioctl,
2549 #endif
2550 
2551     .bdrv_detach_aio_context = iscsi_detach_aio_context,
2552     .bdrv_attach_aio_context = iscsi_attach_aio_context,
2553 
2554     .strong_runtime_opts = iscsi_strong_runtime_opts,
2555 };
2556 #endif
2557 
2558 static void iscsi_block_init(void)
2559 {
2560     bdrv_register(&bdrv_iscsi);
2561 #if LIBISCSI_API_VERSION >= (20160603)
2562     bdrv_register(&bdrv_iser);
2563 #endif
2564 }
2565 
2566 block_init(iscsi_block_init);
2567