?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
pcihp: isolate rule whether slot should be described in DSDT
[qemu/armbru.git] / block / iscsi.c
blobc16c592042f2b6fd31d0f67b3930ab788f376880
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 "sysemu/sysemu.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-io.h"
37 #include "block/block_int.h"
38 #include "block/qdict.h"
39 #include "scsi/constants.h"
40 #include "qemu/iov.h"
41 #include "qemu/module.h"
42 #include "qemu/option.h"
43 #include "qemu/uuid.h"
44 #include "sysemu/replay.h"
45 #include "qapi/error.h"
46 #include "qapi/qapi-commands-machine.h"
47 #include "qapi/qmp/qdict.h"
48 #include "qapi/qmp/qstring.h"
49 #include "crypto/secret.h"
50 #include "scsi/utils.h"
51 #include "trace.h"
52
53 /* Conflict between scsi/utils.h and libiscsi! :( */
54 #define SCSI_XFER_NONE ISCSI_XFER_NONE
55 #include <iscsi/iscsi.h>
56 #define inline __attribute__((gnu_inline)) /* required for libiscsi v1.9.0 */
57 #include <iscsi/scsi-lowlevel.h>
58 #undef inline
59 #undef SCSI_XFER_NONE
60 QEMU_BUILD_BUG_ON((int)SCSI_XFER_NONE != (int)ISCSI_XFER_NONE);
61
62 #ifdef __linux__
63 #include <scsi/sg.h>
64 #endif
65
66 typedef struct IscsiLun {
67 struct iscsi_context *iscsi;
68 AioContext *aio_context;
69 int lun;
70 enum scsi_inquiry_peripheral_device_type type;
71 int block_size;
72 uint64_t num_blocks;
73 int events;
74 QEMUTimer *nop_timer;
75 QEMUTimer *event_timer;
76 QemuMutex mutex;
77 struct scsi_inquiry_logical_block_provisioning lbp;
78 struct scsi_inquiry_block_limits bl;
79 struct scsi_inquiry_device_designator *dd;
80 unsigned char *zeroblock;
81 /* The allocmap tracks which clusters (pages) on the iSCSI target are
82 * allocated and which are not. In case a target returns zeros for
83 * unallocated pages (iscsilun->lprz) we can directly return zeros instead
84 * of reading zeros over the wire if a read request falls within an
85 * unallocated block. As there are 3 possible states we need 2 bitmaps to
86 * track. allocmap_valid keeps track if QEMU's information about a page is
87 * valid. allocmap tracks if a page is allocated or not. In case QEMU has no
88 * valid information about a page the corresponding allocmap entry should be
89 * switched to unallocated as well to force a new lookup of the allocation
90 * status as lookups are generally skipped if a page is suspect to be
91 * allocated. If a iSCSI target is opened with cache.direct = on the
92 * allocmap_valid does not exist turning all cached information invalid so
93 * that a fresh lookup is made for any page even if allocmap entry returns
94 * it's unallocated. */
95 unsigned long *allocmap;
96 unsigned long *allocmap_valid;
97 long allocmap_size;
98 int cluster_size;
99 bool use_16_for_rw;
100 bool write_protected;
101 bool lbpme;
102 bool lbprz;
103 bool dpofua;
104 bool has_write_same;
105 bool request_timed_out;
106 } IscsiLun;
107
108 typedef struct IscsiTask {
109 int status;
110 int complete;
111 int retries;
112 int do_retry;
113 struct scsi_task *task;
114 Coroutine *co;
115 IscsiLun *iscsilun;
116 QEMUTimer retry_timer;
117 int err_code;
118 char *err_str;
119 } IscsiTask;
120
121 typedef struct IscsiAIOCB {
122 BlockAIOCB common;
123 QEMUBH *bh;
124 IscsiLun *iscsilun;
125 struct scsi_task *task;
126 int status;
127 int64_t sector_num;
128 int nb_sectors;
129 int ret;
130 #ifdef __linux__
131 sg_io_hdr_t *ioh;
132 #endif
133 bool cancelled;
134 } IscsiAIOCB;
135
136 /* libiscsi uses time_t so its enough to process events every second */
137 #define EVENT_INTERVAL 1000
138 #define NOP_INTERVAL 5000
139 #define MAX_NOP_FAILURES 3
140 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times)
141 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768};
142
143 /* this threshold is a trade-off knob to choose between
144 * the potential additional overhead of an extra GET_LBA_STATUS request
145 * vs. unnecessarily reading a lot of zero sectors over the wire.
146 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
147 * sectors we check the allocation status of the area covered by the
148 * request first if the allocationmap indicates that the area might be
149 * unallocated. */
150 #define ISCSI_CHECKALLOC_THRES 64
151
152 #ifdef __linux__
153
154 static void
155 iscsi_bh_cb(void *p)
156 {
157 IscsiAIOCB *acb = p;
158
159 qemu_bh_delete(acb->bh);
160
161 acb->common.cb(acb->common.opaque, acb->status);
162
163 if (acb->task != NULL) {
164 scsi_free_scsi_task(acb->task);
165 acb->task = NULL;
166 }
167
168 qemu_aio_unref(acb);
169 }
170
171 static void
172 iscsi_schedule_bh(IscsiAIOCB *acb)
173 {
174 if (acb->bh) {
175 return;
176 }
177 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb);
178 qemu_bh_schedule(acb->bh);
179 }
180
181 #endif
182
183 static void iscsi_co_generic_bh_cb(void *opaque)
184 {
185 struct IscsiTask *iTask = opaque;
186
187 iTask->complete = 1;
188 aio_co_wake(iTask->co);
189 }
190
191 static void iscsi_retry_timer_expired(void *opaque)
192 {
193 struct IscsiTask *iTask = opaque;
194 iTask->complete = 1;
195 if (iTask->co) {
196 aio_co_wake(iTask->co);
197 }
198 }
199
200 static inline unsigned exp_random(double mean)
201 {
202 return -mean * log((double)rand() / RAND_MAX);
203 }
204
205 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in
206 * libiscsi 1.10.0, together with other constants we need. Use it as
207 * a hint that we have to define them ourselves if needed, to keep the
208 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for
209 * the test because SCSI_STATUS_* is an enum.
210 *
211 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes
212 * an enum, check against the LIBISCSI_API_VERSION macro, which was
213 * introduced in 1.11.0. If it is present, there is no need to define
214 * anything.
215 */
216 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \
217 !defined(LIBISCSI_API_VERSION)
218 #define SCSI_STATUS_TASK_SET_FULL 0x28
219 #define SCSI_STATUS_TIMEOUT 0x0f000002
220 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600
221 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00
222 #endif
223
224 #ifndef LIBISCSI_API_VERSION
225 #define LIBISCSI_API_VERSION 20130701
226 #endif
227
228 static int iscsi_translate_sense(struct scsi_sense *sense)
229 {
230 return scsi_sense_to_errno(sense->key,
231 (sense->ascq & 0xFF00) >> 8,
232 sense->ascq & 0xFF);
233 }
234
235 /* Called (via iscsi_service) with QemuMutex held. */
236 static void
237 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
238 void *command_data, void *opaque)
239 {
240 struct IscsiTask *iTask = opaque;
241 struct scsi_task *task = command_data;
242
243 iTask->status = status;
244 iTask->do_retry = 0;
245 iTask->err_code = 0;
246 iTask->task = task;
247
248 if (status != SCSI_STATUS_GOOD) {
249 iTask->err_code = -EIO;
250 if (iTask->retries++ < ISCSI_CMD_RETRIES) {
251 if (status == SCSI_STATUS_BUSY ||
252 status == SCSI_STATUS_TIMEOUT ||
253 status == SCSI_STATUS_TASK_SET_FULL) {
254 unsigned retry_time =
255 exp_random(iscsi_retry_times[iTask->retries - 1]);
256 if (status == SCSI_STATUS_TIMEOUT) {
257 /* make sure the request is rescheduled AFTER the
258 * reconnect is initiated */
259 retry_time = EVENT_INTERVAL * 2;
260 iTask->iscsilun->request_timed_out = true;
261 }
262 error_report("iSCSI Busy/TaskSetFull/TimeOut"
263 " (retry #%u in %u ms): %s",
264 iTask->retries, retry_time,
265 iscsi_get_error(iscsi));
266 aio_timer_init(iTask->iscsilun->aio_context,
267 &iTask->retry_timer, QEMU_CLOCK_REALTIME,
268 SCALE_MS, iscsi_retry_timer_expired, iTask);
269 timer_mod(&iTask->retry_timer,
270 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
271 iTask->do_retry = 1;
272 } else if (status == SCSI_STATUS_CHECK_CONDITION) {
273 int error = iscsi_translate_sense(&task->sense);
274 if (error == EAGAIN) {
275 error_report("iSCSI CheckCondition: %s",
276 iscsi_get_error(iscsi));
277 iTask->do_retry = 1;
278 } else {
279 iTask->err_code = -error;
280 iTask->err_str = g_strdup(iscsi_get_error(iscsi));
281 }
282 }
283 }
284 }
285
286 if (iTask->co) {
287 replay_bh_schedule_oneshot_event(iTask->iscsilun->aio_context,
288 iscsi_co_generic_bh_cb, iTask);
289 } else {
290 iTask->complete = 1;
291 }
292 }
293
294 static void coroutine_fn
295 iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
296 {
297 *iTask = (struct IscsiTask) {
298 .co = qemu_coroutine_self(),
299 .iscsilun = iscsilun,
300 };
301 }
302
303 #ifdef __linux__
304
305 /* Called (via iscsi_service) with QemuMutex held. */
306 static void
307 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
308 void *private_data)
309 {
310 IscsiAIOCB *acb = private_data;
311
312 /* If the command callback hasn't been called yet, drop the task */
313 if (!acb->bh) {
314 /* Call iscsi_aio_ioctl_cb() with SCSI_STATUS_CANCELLED */
315 iscsi_scsi_cancel_task(iscsi, acb->task);
316 }
317
318 qemu_aio_unref(acb); /* acquired in iscsi_aio_cancel() */
319 }
320
321 static void
322 iscsi_aio_cancel(BlockAIOCB *blockacb)
323 {
324 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
325 IscsiLun *iscsilun = acb->iscsilun;
326
327 WITH_QEMU_LOCK_GUARD(&iscsilun->mutex) {
328
329 /* If it was cancelled or completed already, our work is done here */
330 if (acb->cancelled || acb->status != -EINPROGRESS) {
331 return;
332 }
333
334 acb->cancelled = true;
335
336 qemu_aio_ref(acb); /* released in iscsi_abort_task_cb() */
337
338 /* send a task mgmt call to the target to cancel the task on the target */
339 if (iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
340 iscsi_abort_task_cb, acb) < 0) {
341 qemu_aio_unref(acb); /* since iscsi_abort_task_cb() won't be called */
342 }
343 }
344 }
345
346 static const AIOCBInfo iscsi_aiocb_info = {
347 .aiocb_size = sizeof(IscsiAIOCB),
348 .cancel_async = iscsi_aio_cancel,
349 };
350
351 #endif
352
353 static void iscsi_process_read(void *arg);
354 static void iscsi_process_write(void *arg);
355
356 /* Called with QemuMutex held. */
357 static void
358 iscsi_set_events(IscsiLun *iscsilun)
359 {
360 struct iscsi_context *iscsi = iscsilun->iscsi;
361 int ev = iscsi_which_events(iscsi);
362
363 if (ev != iscsilun->events) {
364 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi),
365 false,
366 (ev & POLLIN) ? iscsi_process_read : NULL,
367 (ev & POLLOUT) ? iscsi_process_write : NULL,
368 NULL, NULL,
369 iscsilun);
370 iscsilun->events = ev;
371 }
372 }
373
374 static void iscsi_timed_check_events(void *opaque)
375 {
376 IscsiLun *iscsilun = opaque;
377
378 WITH_QEMU_LOCK_GUARD(&iscsilun->mutex) {
379 /* check for timed out requests */
380 iscsi_service(iscsilun->iscsi, 0);
381
382 if (iscsilun->request_timed_out) {
383 iscsilun->request_timed_out = false;
384 iscsi_reconnect(iscsilun->iscsi);
385 }
386
387 /*
388 * newer versions of libiscsi may return zero events. Ensure we are
389 * able to return to service once this situation changes.
390 */
391 iscsi_set_events(iscsilun);
392 }
393
394 timer_mod(iscsilun->event_timer,
395 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
396 }
397
398 static void
399 iscsi_process_read(void *arg)
400 {
401 IscsiLun *iscsilun = arg;
402 struct iscsi_context *iscsi = iscsilun->iscsi;
403
404 qemu_mutex_lock(&iscsilun->mutex);
405 iscsi_service(iscsi, POLLIN);
406 iscsi_set_events(iscsilun);
407 qemu_mutex_unlock(&iscsilun->mutex);
408 }
409
410 static void
411 iscsi_process_write(void *arg)
412 {
413 IscsiLun *iscsilun = arg;
414 struct iscsi_context *iscsi = iscsilun->iscsi;
415
416 qemu_mutex_lock(&iscsilun->mutex);
417 iscsi_service(iscsi, POLLOUT);
418 iscsi_set_events(iscsilun);
419 qemu_mutex_unlock(&iscsilun->mutex);
420 }
421
422 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
423 {
424 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
425 }
426
427 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
428 {
429 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
430 }
431
432 static bool is_byte_request_lun_aligned(int64_t offset, int64_t bytes,
433 IscsiLun *iscsilun)
434 {
435 if (offset % iscsilun->block_size || bytes % iscsilun->block_size) {
436 error_report("iSCSI misaligned request: "
437 "iscsilun->block_size %u, offset %" PRIi64
438 ", bytes %" PRIi64,
439 iscsilun->block_size, offset, bytes);
440 return false;
441 }
442 return true;
443 }
444
445 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors,
446 IscsiLun *iscsilun)
447 {
448 assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS);
449 return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS,
450 nb_sectors << BDRV_SECTOR_BITS,
451 iscsilun);
452 }
453
454 static void iscsi_allocmap_free(IscsiLun *iscsilun)
455 {
456 g_free(iscsilun->allocmap);
457 g_free(iscsilun->allocmap_valid);
458 iscsilun->allocmap = NULL;
459 iscsilun->allocmap_valid = NULL;
460 }
461
462
463 static int iscsi_allocmap_init(IscsiLun *iscsilun, int open_flags)
464 {
465 iscsi_allocmap_free(iscsilun);
466
467 assert(iscsilun->cluster_size);
468 iscsilun->allocmap_size =
469 DIV_ROUND_UP(iscsilun->num_blocks * iscsilun->block_size,
470 iscsilun->cluster_size);
471
472 iscsilun->allocmap = bitmap_try_new(iscsilun->allocmap_size);
473 if (!iscsilun->allocmap) {
474 return -ENOMEM;
475 }
476
477 if (open_flags & BDRV_O_NOCACHE) {
478 /* when cache.direct = on all allocmap entries are
479 * treated as invalid to force a relookup of the block
480 * status on every read request */
481 return 0;
482 }
483
484 iscsilun->allocmap_valid = bitmap_try_new(iscsilun->allocmap_size);
485 if (!iscsilun->allocmap_valid) {
486 /* if we are under memory pressure free the allocmap as well */
487 iscsi_allocmap_free(iscsilun);
488 return -ENOMEM;
489 }
490
491 return 0;
492 }
493
494 static void
495 iscsi_allocmap_update(IscsiLun *iscsilun, int64_t offset,
496 int64_t bytes, bool allocated, bool valid)
497 {
498 int64_t cl_num_expanded, nb_cls_expanded, cl_num_shrunk, nb_cls_shrunk;
499
500 if (iscsilun->allocmap == NULL) {
501 return;
502 }
503 /* expand to entirely contain all affected clusters */
504 assert(iscsilun->cluster_size);
505 cl_num_expanded = offset / iscsilun->cluster_size;
506 nb_cls_expanded = DIV_ROUND_UP(offset + bytes,
507 iscsilun->cluster_size) - cl_num_expanded;
508 /* shrink to touch only completely contained clusters */
509 cl_num_shrunk = DIV_ROUND_UP(offset, iscsilun->cluster_size);
510 nb_cls_shrunk = (offset + bytes) / iscsilun->cluster_size - cl_num_shrunk;
511 if (allocated) {
512 bitmap_set(iscsilun->allocmap, cl_num_expanded, nb_cls_expanded);
513 } else {
514 if (nb_cls_shrunk > 0) {
515 bitmap_clear(iscsilun->allocmap, cl_num_shrunk, nb_cls_shrunk);
516 }
517 }
518
519 if (iscsilun->allocmap_valid == NULL) {
520 return;
521 }
522 if (valid) {
523 if (nb_cls_shrunk > 0) {
524 bitmap_set(iscsilun->allocmap_valid, cl_num_shrunk, nb_cls_shrunk);
525 }
526 } else {
527 bitmap_clear(iscsilun->allocmap_valid, cl_num_expanded,
528 nb_cls_expanded);
529 }
530 }
531
532 static void
533 iscsi_allocmap_set_allocated(IscsiLun *iscsilun, int64_t offset,
534 int64_t bytes)
535 {
536 iscsi_allocmap_update(iscsilun, offset, bytes, true, true);
537 }
538
539 static void
540 iscsi_allocmap_set_unallocated(IscsiLun *iscsilun, int64_t offset,
541 int64_t bytes)
542 {
543 /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update
544 * is ignored, so this will in effect be an iscsi_allocmap_set_invalid.
545 */
546 iscsi_allocmap_update(iscsilun, offset, bytes, false, true);
547 }
548
549 static void iscsi_allocmap_set_invalid(IscsiLun *iscsilun, int64_t offset,
550 int64_t bytes)
551 {
552 iscsi_allocmap_update(iscsilun, offset, bytes, false, false);
553 }
554
555 static void iscsi_allocmap_invalidate(IscsiLun *iscsilun)
556 {
557 if (iscsilun->allocmap) {
558 bitmap_zero(iscsilun->allocmap, iscsilun->allocmap_size);
559 }
560 if (iscsilun->allocmap_valid) {
561 bitmap_zero(iscsilun->allocmap_valid, iscsilun->allocmap_size);
562 }
563 }
564
565 static inline bool
566 iscsi_allocmap_is_allocated(IscsiLun *iscsilun, int64_t offset,
567 int64_t bytes)
568 {
569 unsigned long size;
570 if (iscsilun->allocmap == NULL) {
571 return true;
572 }
573 assert(iscsilun->cluster_size);
574 size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size);
575 return !(find_next_bit(iscsilun->allocmap, size,
576 offset / iscsilun->cluster_size) == size);
577 }
578
579 static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun,
580 int64_t offset, int64_t bytes)
581 {
582 unsigned long size;
583 if (iscsilun->allocmap_valid == NULL) {
584 return false;
585 }
586 assert(iscsilun->cluster_size);
587 size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size);
588 return (find_next_zero_bit(iscsilun->allocmap_valid, size,
589 offset / iscsilun->cluster_size) == size);
590 }
591
592 static void coroutine_fn iscsi_co_wait_for_task(IscsiTask *iTask,
593 IscsiLun *iscsilun)
594 {
595 while (!iTask->complete) {
596 iscsi_set_events(iscsilun);
597 qemu_mutex_unlock(&iscsilun->mutex);
598 qemu_coroutine_yield();
599 qemu_mutex_lock(&iscsilun->mutex);
600 }
601 }
602
603 static int coroutine_fn
604 iscsi_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
605 QEMUIOVector *iov, int flags)
606 {
607 IscsiLun *iscsilun = bs->opaque;
608 struct IscsiTask iTask;
609 uint64_t lba;
610 uint32_t num_sectors;
611 bool fua = flags & BDRV_REQ_FUA;
612 int r = 0;
613
614 if (fua) {
615 assert(iscsilun->dpofua);
616 }
617 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
618 return -EINVAL;
619 }
620
621 if (bs->bl.max_transfer) {
622 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
623 }
624
625 lba = sector_qemu2lun(sector_num, iscsilun);
626 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
627 iscsi_co_init_iscsitask(iscsilun, &iTask);
628 qemu_mutex_lock(&iscsilun->mutex);
629 retry:
630 if (iscsilun->use_16_for_rw) {
631 #if LIBISCSI_API_VERSION >= (20160603)
632 iTask.task = iscsi_write16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
633 NULL, num_sectors * iscsilun->block_size,
634 iscsilun->block_size, 0, 0, fua, 0, 0,
635 iscsi_co_generic_cb, &iTask,
636 (struct scsi_iovec *)iov->iov, iov->niov);
637 } else {
638 iTask.task = iscsi_write10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
639 NULL, num_sectors * iscsilun->block_size,
640 iscsilun->block_size, 0, 0, fua, 0, 0,
641 iscsi_co_generic_cb, &iTask,
642 (struct scsi_iovec *)iov->iov, iov->niov);
643 }
644 #else
645 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
646 NULL, num_sectors * iscsilun->block_size,
647 iscsilun->block_size, 0, 0, fua, 0, 0,
648 iscsi_co_generic_cb, &iTask);
649 } else {
650 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
651 NULL, num_sectors * iscsilun->block_size,
652 iscsilun->block_size, 0, 0, fua, 0, 0,
653 iscsi_co_generic_cb, &iTask);
654 }
655 #endif
656 if (iTask.task == NULL) {
657 qemu_mutex_unlock(&iscsilun->mutex);
658 return -ENOMEM;
659 }
660 #if LIBISCSI_API_VERSION < (20160603)
661 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
662 iov->niov);
663 #endif
664 iscsi_co_wait_for_task(&iTask, iscsilun);
665
666 if (iTask.task != NULL) {
667 scsi_free_scsi_task(iTask.task);
668 iTask.task = NULL;
669 }
670
671 if (iTask.do_retry) {
672 iTask.complete = 0;
673 goto retry;
674 }
675
676 if (iTask.status != SCSI_STATUS_GOOD) {
677 iscsi_allocmap_set_invalid(iscsilun, sector_num * BDRV_SECTOR_SIZE,
678 nb_sectors * BDRV_SECTOR_SIZE);
679 error_report("iSCSI WRITE10/16 failed at lba %" PRIu64 ": %s", lba,
680 iTask.err_str);
681 r = iTask.err_code;
682 goto out_unlock;
683 }
684
685 iscsi_allocmap_set_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE,
686 nb_sectors * BDRV_SECTOR_SIZE);
687
688 out_unlock:
689 qemu_mutex_unlock(&iscsilun->mutex);
690 g_free(iTask.err_str);
691 return r;
692 }
693
694
695
696 static int coroutine_fn iscsi_co_block_status(BlockDriverState *bs,
697 bool want_zero, int64_t offset,
698 int64_t bytes, int64_t *pnum,
699 int64_t *map,
700 BlockDriverState **file)
701 {
702 IscsiLun *iscsilun = bs->opaque;
703 struct scsi_get_lba_status *lbas = NULL;
704 struct scsi_lba_status_descriptor *lbasd = NULL;
705 struct IscsiTask iTask;
706 uint64_t lba, max_bytes;
707 int ret;
708
709 iscsi_co_init_iscsitask(iscsilun, &iTask);
710
711 assert(QEMU_IS_ALIGNED(offset | bytes, iscsilun->block_size));
712
713 /* default to all sectors allocated */
714 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
715 if (map) {
716 *map = offset;
717 }
718 *pnum = bytes;
719
720 /* LUN does not support logical block provisioning */
721 if (!iscsilun->lbpme) {
722 goto out;
723 }
724
725 lba = offset / iscsilun->block_size;
726 max_bytes = (iscsilun->num_blocks - lba) * iscsilun->block_size;
727
728 qemu_mutex_lock(&iscsilun->mutex);
729 retry:
730 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
731 lba, 8 + 16, iscsi_co_generic_cb,
732 &iTask) == NULL) {
733 ret = -ENOMEM;
734 goto out_unlock;
735 }
736 iscsi_co_wait_for_task(&iTask, iscsilun);
737
738 if (iTask.do_retry) {
739 if (iTask.task != NULL) {
740 scsi_free_scsi_task(iTask.task);
741 iTask.task = NULL;
742 }
743 iTask.complete = 0;
744 goto retry;
745 }
746
747 if (iTask.status != SCSI_STATUS_GOOD) {
748 /* in case the get_lba_status_callout fails (i.e.
749 * because the device is busy or the cmd is not
750 * supported) we pretend all blocks are allocated
751 * for backwards compatibility */
752 error_report("iSCSI GET_LBA_STATUS failed at lba %" PRIu64 ": %s",
753 lba, iTask.err_str);
754 goto out_unlock;
755 }
756
757 lbas = scsi_datain_unmarshall(iTask.task);
758 if (lbas == NULL || lbas->num_descriptors == 0) {
759 ret = -EIO;
760 goto out_unlock;
761 }
762
763 lbasd = &lbas->descriptors[0];
764
765 if (lba != lbasd->lba) {
766 ret = -EIO;
767 goto out_unlock;
768 }
769
770 *pnum = MIN((int64_t) lbasd->num_blocks * iscsilun->block_size, max_bytes);
771
772 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
773 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
774 ret &= ~BDRV_BLOCK_DATA;
775 if (iscsilun->lbprz) {
776 ret |= BDRV_BLOCK_ZERO;
777 }
778 }
779
780 if (ret & BDRV_BLOCK_ZERO) {
781 iscsi_allocmap_set_unallocated(iscsilun, offset, *pnum);
782 } else {
783 iscsi_allocmap_set_allocated(iscsilun, offset, *pnum);
784 }
785
786 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 = MIN(acb->ioh->mx_sb_len, acb->ioh->sb_len_wr);
994 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
995 }
996
997 iscsi_schedule_bh(acb);
998 }
999
1000 static void iscsi_ioctl_bh_completion(void *opaque)
1001 {
1002 IscsiAIOCB *acb = opaque;
1003
1004 qemu_bh_delete(acb->bh);
1005 acb->common.cb(acb->common.opaque, acb->ret);
1006 qemu_aio_unref(acb);
1007 }
1008
1009 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
1010 {
1011 BlockDriverState *bs = acb->common.bs;
1012 IscsiLun *iscsilun = bs->opaque;
1013 int ret = 0;
1014
1015 switch (req) {
1016 case SG_GET_VERSION_NUM:
1017 *(int *)buf = 30000;
1018 break;
1019 case SG_GET_SCSI_ID:
1020 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
1021 break;
1022 default:
1023 ret = -EINVAL;
1024 }
1025 assert(!acb->bh);
1026 acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
1027 iscsi_ioctl_bh_completion, acb);
1028 acb->ret = ret;
1029 qemu_bh_schedule(acb->bh);
1030 }
1031
1032 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
1033 unsigned long int req, void *buf,
1034 BlockCompletionFunc *cb, void *opaque)
1035 {
1036 IscsiLun *iscsilun = bs->opaque;
1037 struct iscsi_context *iscsi = iscsilun->iscsi;
1038 struct iscsi_data data;
1039 IscsiAIOCB *acb;
1040
1041 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
1042
1043 acb->iscsilun = iscsilun;
1044 acb->bh = NULL;
1045 acb->status = -EINPROGRESS;
1046 acb->ioh = buf;
1047 acb->cancelled = false;
1048
1049 if (req != SG_IO) {
1050 iscsi_ioctl_handle_emulated(acb, req, buf);
1051 return &acb->common;
1052 }
1053
1054 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) {
1055 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)",
1056 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE);
1057 qemu_aio_unref(acb);
1058 return NULL;
1059 }
1060
1061 acb->task = malloc(sizeof(struct scsi_task));
1062 if (acb->task == NULL) {
1063 error_report("iSCSI: Failed to allocate task for scsi command. %s",
1064 iscsi_get_error(iscsi));
1065 qemu_aio_unref(acb);
1066 return NULL;
1067 }
1068 memset(acb->task, 0, sizeof(struct scsi_task));
1069
1070 switch (acb->ioh->dxfer_direction) {
1071 case SG_DXFER_TO_DEV:
1072 acb->task->xfer_dir = SCSI_XFER_WRITE;
1073 break;
1074 case SG_DXFER_FROM_DEV:
1075 acb->task->xfer_dir = SCSI_XFER_READ;
1076 break;
1077 default:
1078 acb->task->xfer_dir = SCSI_XFER_NONE;
1079 break;
1080 }
1081
1082 acb->task->cdb_size = acb->ioh->cmd_len;
1083 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
1084 acb->task->expxferlen = acb->ioh->dxfer_len;
1085
1086 data.size = 0;
1087 qemu_mutex_lock(&iscsilun->mutex);
1088 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
1089 if (acb->ioh->iovec_count == 0) {
1090 data.data = acb->ioh->dxferp;
1091 data.size = acb->ioh->dxfer_len;
1092 } else {
1093 scsi_task_set_iov_out(acb->task,
1094 (struct scsi_iovec *) acb->ioh->dxferp,
1095 acb->ioh->iovec_count);
1096 }
1097 }
1098
1099 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
1100 iscsi_aio_ioctl_cb,
1101 (data.size > 0) ? &data : NULL,
1102 acb) != 0) {
1103 qemu_mutex_unlock(&iscsilun->mutex);
1104 scsi_free_scsi_task(acb->task);
1105 qemu_aio_unref(acb);
1106 return NULL;
1107 }
1108
1109 /* tell libiscsi to read straight into the buffer we got from ioctl */
1110 if (acb->task->xfer_dir == SCSI_XFER_READ) {
1111 if (acb->ioh->iovec_count == 0) {
1112 scsi_task_add_data_in_buffer(acb->task,
1113 acb->ioh->dxfer_len,
1114 acb->ioh->dxferp);
1115 } else {
1116 scsi_task_set_iov_in(acb->task,
1117 (struct scsi_iovec *) acb->ioh->dxferp,
1118 acb->ioh->iovec_count);
1119 }
1120 }
1121
1122 iscsi_set_events(iscsilun);
1123 qemu_mutex_unlock(&iscsilun->mutex);
1124
1125 return &acb->common;
1126 }
1127
1128 #endif
1129
1130 static int64_t
1131 iscsi_getlength(BlockDriverState *bs)
1132 {
1133 IscsiLun *iscsilun = bs->opaque;
1134 int64_t len;
1135
1136 len = iscsilun->num_blocks;
1137 len *= iscsilun->block_size;
1138
1139 return len;
1140 }
1141
1142 static int
1143 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset,
1144 int64_t 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 /*
1161 * We don't want to overflow list.num which is uint32_t.
1162 * We rely on our max_pdiscard.
1163 */
1164 assert(bytes / iscsilun->block_size <= UINT32_MAX);
1165
1166 list.lba = offset / iscsilun->block_size;
1167 list.num = bytes / iscsilun->block_size;
1168
1169 iscsi_co_init_iscsitask(iscsilun, &iTask);
1170 qemu_mutex_lock(&iscsilun->mutex);
1171 retry:
1172 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
1173 iscsi_co_generic_cb, &iTask) == NULL) {
1174 r = -ENOMEM;
1175 goto out_unlock;
1176 }
1177
1178 iscsi_co_wait_for_task(&iTask, iscsilun);
1179
1180 if (iTask.task != NULL) {
1181 scsi_free_scsi_task(iTask.task);
1182 iTask.task = NULL;
1183 }
1184
1185 if (iTask.do_retry) {
1186 iTask.complete = 0;
1187 goto retry;
1188 }
1189
1190 iscsi_allocmap_set_invalid(iscsilun, offset, bytes);
1191
1192 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
1193 /* the target might fail with a check condition if it
1194 is not happy with the alignment of the UNMAP request
1195 we silently fail in this case */
1196 goto out_unlock;
1197 }
1198
1199 if (iTask.status != SCSI_STATUS_GOOD) {
1200 error_report("iSCSI UNMAP failed at lba %" PRIu64 ": %s",
1201 list.lba, iTask.err_str);
1202 r = iTask.err_code;
1203 goto out_unlock;
1204 }
1205
1206 out_unlock:
1207 qemu_mutex_unlock(&iscsilun->mutex);
1208 g_free(iTask.err_str);
1209 return r;
1210 }
1211
1212 static int
1213 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1214 int64_t bytes, BdrvRequestFlags flags)
1215 {
1216 IscsiLun *iscsilun = bs->opaque;
1217 struct IscsiTask iTask;
1218 uint64_t lba;
1219 uint64_t nb_blocks;
1220 bool use_16_for_ws = iscsilun->use_16_for_rw;
1221 int r = 0;
1222
1223 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) {
1224 return -ENOTSUP;
1225 }
1226
1227 if (flags & BDRV_REQ_MAY_UNMAP) {
1228 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
1229 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
1230 use_16_for_ws = true;
1231 }
1232 if (use_16_for_ws && !iscsilun->lbp.lbpws) {
1233 /* WRITESAME16 with UNMAP is not supported by the target,
1234 * fall back and try WRITESAME10/16 without UNMAP */
1235 flags &= ~BDRV_REQ_MAY_UNMAP;
1236 use_16_for_ws = iscsilun->use_16_for_rw;
1237 }
1238 }
1239
1240 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1241 /* WRITESAME without UNMAP is not supported by the target */
1242 return -ENOTSUP;
1243 }
1244
1245 lba = offset / iscsilun->block_size;
1246 nb_blocks = bytes / iscsilun->block_size;
1247
1248 if (iscsilun->zeroblock == NULL) {
1249 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1250 if (iscsilun->zeroblock == NULL) {
1251 return -ENOMEM;
1252 }
1253 }
1254
1255 qemu_mutex_lock(&iscsilun->mutex);
1256 iscsi_co_init_iscsitask(iscsilun, &iTask);
1257 retry:
1258 if (use_16_for_ws) {
1259 /*
1260 * iscsi_writesame16_task num_blocks argument is uint32_t. We rely here
1261 * on our max_pwrite_zeroes limit.
1262 */
1263 assert(nb_blocks <= UINT32_MAX);
1264 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1265 iscsilun->zeroblock, iscsilun->block_size,
1266 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1267 0, 0, iscsi_co_generic_cb, &iTask);
1268 } else {
1269 /*
1270 * iscsi_writesame10_task num_blocks argument is uint16_t. We rely here
1271 * on our max_pwrite_zeroes limit.
1272 */
1273 assert(nb_blocks <= UINT16_MAX);
1274 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1275 iscsilun->zeroblock, iscsilun->block_size,
1276 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1277 0, 0, iscsi_co_generic_cb, &iTask);
1278 }
1279 if (iTask.task == NULL) {
1280 qemu_mutex_unlock(&iscsilun->mutex);
1281 return -ENOMEM;
1282 }
1283
1284 iscsi_co_wait_for_task(&iTask, iscsilun);
1285
1286 if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1287 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1288 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1289 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1290 /* WRITE SAME is not supported by the target */
1291 iscsilun->has_write_same = false;
1292 scsi_free_scsi_task(iTask.task);
1293 r = -ENOTSUP;
1294 goto out_unlock;
1295 }
1296
1297 if (iTask.task != NULL) {
1298 scsi_free_scsi_task(iTask.task);
1299 iTask.task = NULL;
1300 }
1301
1302 if (iTask.do_retry) {
1303 iTask.complete = 0;
1304 goto retry;
1305 }
1306
1307 if (iTask.status != SCSI_STATUS_GOOD) {
1308 iscsi_allocmap_set_invalid(iscsilun, offset, bytes);
1309 error_report("iSCSI WRITESAME10/16 failed at lba %" PRIu64 ": %s",
1310 lba, iTask.err_str);
1311 r = iTask.err_code;
1312 goto out_unlock;
1313 }
1314
1315 if (flags & BDRV_REQ_MAY_UNMAP) {
1316 iscsi_allocmap_set_invalid(iscsilun, offset, bytes);
1317 } else {
1318 iscsi_allocmap_set_allocated(iscsilun, offset, bytes);
1319 }
1320
1321 out_unlock:
1322 qemu_mutex_unlock(&iscsilun->mutex);
1323 g_free(iTask.err_str);
1324 return r;
1325 }
1326
1327 static void apply_chap(struct iscsi_context *iscsi, QemuOpts *opts,
1328 Error **errp)
1329 {
1330 const char *user = NULL;
1331 const char *password = NULL;
1332 const char *secretid;
1333 char *secret = NULL;
1334
1335 user = qemu_opt_get(opts, "user");
1336 if (!user) {
1337 return;
1338 }
1339
1340 secretid = qemu_opt_get(opts, "password-secret");
1341 password = qemu_opt_get(opts, "password");
1342 if (secretid && password) {
1343 error_setg(errp, "'password' and 'password-secret' properties are "
1344 "mutually exclusive");
1345 return;
1346 }
1347 if (secretid) {
1348 secret = qcrypto_secret_lookup_as_utf8(secretid, errp);
1349 if (!secret) {
1350 return;
1351 }
1352 password = secret;
1353 } else if (!password) {
1354 error_setg(errp, "CHAP username specified but no password was given");
1355 return;
1356 }
1357
1358 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1359 error_setg(errp, "Failed to set initiator username and password");
1360 }
1361
1362 g_free(secret);
1363 }
1364
1365 static void apply_header_digest(struct iscsi_context *iscsi, QemuOpts *opts,
1366 Error **errp)
1367 {
1368 const char *digest = NULL;
1369
1370 digest = qemu_opt_get(opts, "header-digest");
1371 if (!digest) {
1372 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1373 } else if (!strcmp(digest, "crc32c")) {
1374 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1375 } else if (!strcmp(digest, "none")) {
1376 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1377 } else if (!strcmp(digest, "crc32c-none")) {
1378 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1379 } else if (!strcmp(digest, "none-crc32c")) {
1380 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1381 } else {
1382 error_setg(errp, "Invalid header-digest setting : %s", digest);
1383 }
1384 }
1385
1386 static char *get_initiator_name(QemuOpts *opts)
1387 {
1388 const char *name;
1389 char *iscsi_name;
1390 UuidInfo *uuid_info;
1391
1392 name = qemu_opt_get(opts, "initiator-name");
1393 if (name) {
1394 return g_strdup(name);
1395 }
1396
1397 uuid_info = qmp_query_uuid(NULL);
1398 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1399 name = qemu_get_vm_name();
1400 } else {
1401 name = uuid_info->UUID;
1402 }
1403 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1404 name ? ":" : "", name ? name : "");
1405 qapi_free_UuidInfo(uuid_info);
1406 return iscsi_name;
1407 }
1408
1409 static void iscsi_nop_timed_event(void *opaque)
1410 {
1411 IscsiLun *iscsilun = opaque;
1412
1413 QEMU_LOCK_GUARD(&iscsilun->mutex);
1414 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1415 error_report("iSCSI: NOP timeout. Reconnecting...");
1416 iscsilun->request_timed_out = true;
1417 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1418 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1419 return;
1420 }
1421
1422 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1423 iscsi_set_events(iscsilun);
1424 }
1425
1426 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1427 {
1428 struct scsi_task *task = NULL;
1429 struct scsi_readcapacity10 *rc10 = NULL;
1430 struct scsi_readcapacity16 *rc16 = NULL;
1431 int retries = ISCSI_CMD_RETRIES;
1432
1433 do {
1434 if (task != NULL) {
1435 scsi_free_scsi_task(task);
1436 task = NULL;
1437 }
1438
1439 switch (iscsilun->type) {
1440 case TYPE_DISK:
1441 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1442 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1443 rc16 = scsi_datain_unmarshall(task);
1444 if (rc16 == NULL) {
1445 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1446 } else {
1447 iscsilun->block_size = rc16->block_length;
1448 iscsilun->num_blocks = rc16->returned_lba + 1;
1449 iscsilun->lbpme = !!rc16->lbpme;
1450 iscsilun->lbprz = !!rc16->lbprz;
1451 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1452 }
1453 break;
1454 }
1455 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1456 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1457 break;
1458 }
1459 /* Fall through and try READ CAPACITY(10) instead. */
1460 case TYPE_ROM:
1461 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1462 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1463 rc10 = scsi_datain_unmarshall(task);
1464 if (rc10 == NULL) {
1465 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1466 } else {
1467 iscsilun->block_size = rc10->block_size;
1468 if (rc10->lba == 0) {
1469 /* blank disk loaded */
1470 iscsilun->num_blocks = 0;
1471 } else {
1472 iscsilun->num_blocks = rc10->lba + 1;
1473 }
1474 }
1475 }
1476 break;
1477 default:
1478 return;
1479 }
1480 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1481 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1482 && retries-- > 0);
1483
1484 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1485 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1486 } else if (!iscsilun->block_size ||
1487 iscsilun->block_size % BDRV_SECTOR_SIZE) {
1488 error_setg(errp, "iSCSI: the target returned an invalid "
1489 "block size of %d.", iscsilun->block_size);
1490 }
1491 if (task) {
1492 scsi_free_scsi_task(task);
1493 }
1494 }
1495
1496 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1497 int evpd, int pc, void **inq, Error **errp)
1498 {
1499 int full_size;
1500 struct scsi_task *task = NULL;
1501 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1502 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1503 goto fail;
1504 }
1505 full_size = scsi_datain_getfullsize(task);
1506 if (full_size > task->datain.size) {
1507 scsi_free_scsi_task(task);
1508
1509 /* we need more data for the full list */
1510 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1511 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1512 goto fail;
1513 }
1514 }
1515
1516 *inq = scsi_datain_unmarshall(task);
1517 if (*inq == NULL) {
1518 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1519 goto fail_with_err;
1520 }
1521
1522 return task;
1523
1524 fail:
1525 error_setg(errp, "iSCSI: Inquiry command failed : %s",
1526 iscsi_get_error(iscsi));
1527 fail_with_err:
1528 if (task != NULL) {
1529 scsi_free_scsi_task(task);
1530 }
1531 return NULL;
1532 }
1533
1534 static void iscsi_detach_aio_context(BlockDriverState *bs)
1535 {
1536 IscsiLun *iscsilun = bs->opaque;
1537
1538 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1539 false, NULL, NULL, NULL, NULL, NULL);
1540 iscsilun->events = 0;
1541
1542 if (iscsilun->nop_timer) {
1543 timer_free(iscsilun->nop_timer);
1544 iscsilun->nop_timer = NULL;
1545 }
1546 if (iscsilun->event_timer) {
1547 timer_free(iscsilun->event_timer);
1548 iscsilun->event_timer = NULL;
1549 }
1550 }
1551
1552 static void iscsi_attach_aio_context(BlockDriverState *bs,
1553 AioContext *new_context)
1554 {
1555 IscsiLun *iscsilun = bs->opaque;
1556
1557 iscsilun->aio_context = new_context;
1558 iscsi_set_events(iscsilun);
1559
1560 /* Set up a timer for sending out iSCSI NOPs */
1561 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1562 QEMU_CLOCK_REALTIME, SCALE_MS,
1563 iscsi_nop_timed_event, iscsilun);
1564 timer_mod(iscsilun->nop_timer,
1565 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1566
1567 /* Set up a timer for periodic calls to iscsi_set_events and to
1568 * scan for command timeout */
1569 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1570 QEMU_CLOCK_REALTIME, SCALE_MS,
1571 iscsi_timed_check_events, iscsilun);
1572 timer_mod(iscsilun->event_timer,
1573 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1574 }
1575
1576 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1577 {
1578 struct scsi_task *task;
1579 struct scsi_mode_sense *ms = NULL;
1580 iscsilun->write_protected = false;
1581 iscsilun->dpofua = false;
1582
1583 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1584 1, SCSI_MODESENSE_PC_CURRENT,
1585 0x3F, 0, 255);
1586 if (task == NULL) {
1587 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1588 iscsi_get_error(iscsilun->iscsi));
1589 goto out;
1590 }
1591
1592 if (task->status != SCSI_STATUS_GOOD) {
1593 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1594 goto out;
1595 }
1596 ms = scsi_datain_unmarshall(task);
1597 if (!ms) {
1598 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1599 iscsi_get_error(iscsilun->iscsi));
1600 goto out;
1601 }
1602 iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1603 iscsilun->dpofua = ms->device_specific_parameter & 0x10;
1604
1605 out:
1606 if (task) {
1607 scsi_free_scsi_task(task);
1608 }
1609 }
1610
1611 static void iscsi_parse_iscsi_option(const char *target, QDict *options)
1612 {
1613 QemuOptsList *list;
1614 QemuOpts *opts;
1615 const char *user, *password, *password_secret, *initiator_name,
1616 *header_digest, *timeout;
1617
1618 list = qemu_find_opts("iscsi");
1619 if (!list) {
1620 return;
1621 }
1622
1623 opts = qemu_opts_find(list, target);
1624 if (opts == NULL) {
1625 opts = QTAILQ_FIRST(&list->head);
1626 if (!opts) {
1627 return;
1628 }
1629 }
1630
1631 user = qemu_opt_get(opts, "user");
1632 if (user) {
1633 qdict_set_default_str(options, "user", user);
1634 }
1635
1636 password = qemu_opt_get(opts, "password");
1637 if (password) {
1638 qdict_set_default_str(options, "password", password);
1639 }
1640
1641 password_secret = qemu_opt_get(opts, "password-secret");
1642 if (password_secret) {
1643 qdict_set_default_str(options, "password-secret", password_secret);
1644 }
1645
1646 initiator_name = qemu_opt_get(opts, "initiator-name");
1647 if (initiator_name) {
1648 qdict_set_default_str(options, "initiator-name", initiator_name);
1649 }
1650
1651 header_digest = qemu_opt_get(opts, "header-digest");
1652 if (header_digest) {
1653 /* -iscsi takes upper case values, but QAPI only supports lower case
1654 * enum constant names, so we have to convert here. */
1655 char *qapi_value = g_ascii_strdown(header_digest, -1);
1656 qdict_set_default_str(options, "header-digest", qapi_value);
1657 g_free(qapi_value);
1658 }
1659
1660 timeout = qemu_opt_get(opts, "timeout");
1661 if (timeout) {
1662 qdict_set_default_str(options, "timeout", timeout);
1663 }
1664 }
1665
1666 /*
1667 * We support iscsi url's on the form
1668 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1669 */
1670 static void iscsi_parse_filename(const char *filename, QDict *options,
1671 Error **errp)
1672 {
1673 struct iscsi_url *iscsi_url;
1674 const char *transport_name;
1675 char *lun_str;
1676
1677 iscsi_url = iscsi_parse_full_url(NULL, filename);
1678 if (iscsi_url == NULL) {
1679 error_setg(errp, "Failed to parse URL : %s", filename);
1680 return;
1681 }
1682
1683 #if LIBISCSI_API_VERSION >= (20160603)
1684 switch (iscsi_url->transport) {
1685 case TCP_TRANSPORT:
1686 transport_name = "tcp";
1687 break;
1688 case ISER_TRANSPORT:
1689 transport_name = "iser";
1690 break;
1691 default:
1692 error_setg(errp, "Unknown transport type (%d)",
1693 iscsi_url->transport);
1694 return;
1695 }
1696 #else
1697 transport_name = "tcp";
1698 #endif
1699
1700 qdict_set_default_str(options, "transport", transport_name);
1701 qdict_set_default_str(options, "portal", iscsi_url->portal);
1702 qdict_set_default_str(options, "target", iscsi_url->target);
1703
1704 lun_str = g_strdup_printf("%d", iscsi_url->lun);
1705 qdict_set_default_str(options, "lun", lun_str);
1706 g_free(lun_str);
1707
1708 /* User/password from -iscsi take precedence over those from the URL */
1709 iscsi_parse_iscsi_option(iscsi_url->target, options);
1710
1711 if (iscsi_url->user[0] != '\0') {
1712 qdict_set_default_str(options, "user", iscsi_url->user);
1713 qdict_set_default_str(options, "password", iscsi_url->passwd);
1714 }
1715
1716 iscsi_destroy_url(iscsi_url);
1717 }
1718
1719 static QemuOptsList runtime_opts = {
1720 .name = "iscsi",
1721 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1722 .desc = {
1723 {
1724 .name = "transport",
1725 .type = QEMU_OPT_STRING,
1726 },
1727 {
1728 .name = "portal",
1729 .type = QEMU_OPT_STRING,
1730 },
1731 {
1732 .name = "target",
1733 .type = QEMU_OPT_STRING,
1734 },
1735 {
1736 .name = "user",
1737 .type = QEMU_OPT_STRING,
1738 },
1739 {
1740 .name = "password",
1741 .type = QEMU_OPT_STRING,
1742 },
1743 {
1744 .name = "password-secret",
1745 .type = QEMU_OPT_STRING,
1746 },
1747 {
1748 .name = "lun",
1749 .type = QEMU_OPT_NUMBER,
1750 },
1751 {
1752 .name = "initiator-name",
1753 .type = QEMU_OPT_STRING,
1754 },
1755 {
1756 .name = "header-digest",
1757 .type = QEMU_OPT_STRING,
1758 },
1759 {
1760 .name = "timeout",
1761 .type = QEMU_OPT_NUMBER,
1762 },
1763 { /* end of list */ }
1764 },
1765 };
1766
1767 static void iscsi_save_designator(IscsiLun *lun,
1768 struct scsi_inquiry_device_identification *inq_di)
1769 {
1770 struct scsi_inquiry_device_designator *desig, *copy = NULL;
1771
1772 for (desig = inq_di->designators; desig; desig = desig->next) {
1773 if (desig->association ||
1774 desig->designator_type > SCSI_DESIGNATOR_TYPE_NAA) {
1775 continue;
1776 }
1777 /* NAA works better than T10 vendor ID based designator. */
1778 if (!copy || copy->designator_type < desig->designator_type) {
1779 copy = desig;
1780 }
1781 }
1782 if (copy) {
1783 lun->dd = g_new(struct scsi_inquiry_device_designator, 1);
1784 *lun->dd = *copy;
1785 lun->dd->next = NULL;
1786 lun->dd->designator = g_malloc(copy->designator_length);
1787 memcpy(lun->dd->designator, copy->designator, copy->designator_length);
1788 }
1789 }
1790
1791 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1792 Error **errp)
1793 {
1794 IscsiLun *iscsilun = bs->opaque;
1795 struct iscsi_context *iscsi = NULL;
1796 struct scsi_task *task = NULL;
1797 struct scsi_inquiry_standard *inq = NULL;
1798 struct scsi_inquiry_supported_pages *inq_vpd;
1799 char *initiator_name = NULL;
1800 QemuOpts *opts;
1801 Error *local_err = NULL;
1802 const char *transport_name, *portal, *target;
1803 #if LIBISCSI_API_VERSION >= (20160603)
1804 enum iscsi_transport_type transport;
1805 #endif
1806 int i, ret = 0, timeout = 0, lun;
1807
1808 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1809 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1810 ret = -EINVAL;
1811 goto out;
1812 }
1813
1814 transport_name = qemu_opt_get(opts, "transport");
1815 portal = qemu_opt_get(opts, "portal");
1816 target = qemu_opt_get(opts, "target");
1817 lun = qemu_opt_get_number(opts, "lun", 0);
1818
1819 if (!transport_name || !portal || !target) {
1820 error_setg(errp, "Need all of transport, portal and target options");
1821 ret = -EINVAL;
1822 goto out;
1823 }
1824
1825 if (!strcmp(transport_name, "tcp")) {
1826 #if LIBISCSI_API_VERSION >= (20160603)
1827 transport = TCP_TRANSPORT;
1828 } else if (!strcmp(transport_name, "iser")) {
1829 transport = ISER_TRANSPORT;
1830 #else
1831 /* TCP is what older libiscsi versions always use */
1832 #endif
1833 } else {
1834 error_setg(errp, "Unknown transport: %s", transport_name);
1835 ret = -EINVAL;
1836 goto out;
1837 }
1838
1839 memset(iscsilun, 0, sizeof(IscsiLun));
1840
1841 initiator_name = get_initiator_name(opts);
1842
1843 iscsi = iscsi_create_context(initiator_name);
1844 if (iscsi == NULL) {
1845 error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1846 ret = -ENOMEM;
1847 goto out;
1848 }
1849 #if LIBISCSI_API_VERSION >= (20160603)
1850 if (iscsi_init_transport(iscsi, transport)) {
1851 error_setg(errp, ("Error initializing transport."));
1852 ret = -EINVAL;
1853 goto out;
1854 }
1855 #endif
1856 if (iscsi_set_targetname(iscsi, target)) {
1857 error_setg(errp, "iSCSI: Failed to set target name.");
1858 ret = -EINVAL;
1859 goto out;
1860 }
1861
1862 /* check if we got CHAP username/password via the options */
1863 apply_chap(iscsi, opts, &local_err);
1864 if (local_err != NULL) {
1865 error_propagate(errp, local_err);
1866 ret = -EINVAL;
1867 goto out;
1868 }
1869
1870 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1871 error_setg(errp, "iSCSI: Failed to set session type to normal.");
1872 ret = -EINVAL;
1873 goto out;
1874 }
1875
1876 /* check if we got HEADER_DIGEST via the options */
1877 apply_header_digest(iscsi, opts, &local_err);
1878 if (local_err != NULL) {
1879 error_propagate(errp, local_err);
1880 ret = -EINVAL;
1881 goto out;
1882 }
1883
1884 /* timeout handling is broken in libiscsi before 1.15.0 */
1885 timeout = qemu_opt_get_number(opts, "timeout", 0);
1886 #if LIBISCSI_API_VERSION >= 20150621
1887 iscsi_set_timeout(iscsi, timeout);
1888 #else
1889 if (timeout) {
1890 warn_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1891 }
1892 #endif
1893
1894 if (iscsi_full_connect_sync(iscsi, portal, lun) != 0) {
1895 error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1896 iscsi_get_error(iscsi));
1897 ret = -EINVAL;
1898 goto out;
1899 }
1900
1901 iscsilun->iscsi = iscsi;
1902 iscsilun->aio_context = bdrv_get_aio_context(bs);
1903 iscsilun->lun = lun;
1904 iscsilun->has_write_same = true;
1905
1906 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1907 (void **) &inq, errp);
1908 if (task == NULL) {
1909 ret = -EINVAL;
1910 goto out;
1911 }
1912 iscsilun->type = inq->periperal_device_type;
1913 scsi_free_scsi_task(task);
1914 task = NULL;
1915
1916 iscsi_modesense_sync(iscsilun);
1917 if (iscsilun->dpofua) {
1918 bs->supported_write_flags = BDRV_REQ_FUA;
1919 }
1920
1921 /* Check the write protect flag of the LUN if we want to write */
1922 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1923 iscsilun->write_protected) {
1924 ret = bdrv_apply_auto_read_only(bs, "LUN is write protected", errp);
1925 if (ret < 0) {
1926 goto out;
1927 }
1928 flags &= ~BDRV_O_RDWR;
1929 }
1930
1931 iscsi_readcapacity_sync(iscsilun, &local_err);
1932 if (local_err != NULL) {
1933 error_propagate(errp, local_err);
1934 ret = -EINVAL;
1935 goto out;
1936 }
1937 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1938
1939 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1940 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1941 * will try to read from the device to guess the image format.
1942 */
1943 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1944 bs->sg = true;
1945 }
1946
1947 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1948 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1949 (void **) &inq_vpd, errp);
1950 if (task == NULL) {
1951 ret = -EINVAL;
1952 goto out;
1953 }
1954 for (i = 0; i < inq_vpd->num_pages; i++) {
1955 struct scsi_task *inq_task;
1956 struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1957 struct scsi_inquiry_block_limits *inq_bl;
1958 struct scsi_inquiry_device_identification *inq_di;
1959 switch (inq_vpd->pages[i]) {
1960 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1961 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1962 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1963 (void **) &inq_lbp, errp);
1964 if (inq_task == NULL) {
1965 ret = -EINVAL;
1966 goto out;
1967 }
1968 memcpy(&iscsilun->lbp, inq_lbp,
1969 sizeof(struct scsi_inquiry_logical_block_provisioning));
1970 scsi_free_scsi_task(inq_task);
1971 break;
1972 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1973 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1974 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1975 (void **) &inq_bl, errp);
1976 if (inq_task == NULL) {
1977 ret = -EINVAL;
1978 goto out;
1979 }
1980 memcpy(&iscsilun->bl, inq_bl,
1981 sizeof(struct scsi_inquiry_block_limits));
1982 scsi_free_scsi_task(inq_task);
1983 break;
1984 case SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION:
1985 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1986 SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION,
1987 (void **) &inq_di, errp);
1988 if (inq_task == NULL) {
1989 ret = -EINVAL;
1990 goto out;
1991 }
1992 iscsi_save_designator(iscsilun, inq_di);
1993 scsi_free_scsi_task(inq_task);
1994 break;
1995 default:
1996 break;
1997 }
1998 }
1999 scsi_free_scsi_task(task);
2000 task = NULL;
2001
2002 qemu_mutex_init(&iscsilun->mutex);
2003 iscsi_attach_aio_context(bs, iscsilun->aio_context);
2004
2005 /* Guess the internal cluster (page) size of the iscsi target by the means
2006 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
2007 * reasonable size */
2008 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
2009 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
2010 iscsilun->cluster_size = iscsilun->bl.opt_unmap_gran *
2011 iscsilun->block_size;
2012 if (iscsilun->lbprz) {
2013 ret = iscsi_allocmap_init(iscsilun, flags);
2014 }
2015 }
2016
2017 if (iscsilun->lbprz && iscsilun->lbp.lbpws) {
2018 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
2019 }
2020
2021 out:
2022 qemu_opts_del(opts);
2023 g_free(initiator_name);
2024 if (task != NULL) {
2025 scsi_free_scsi_task(task);
2026 }
2027
2028 if (ret) {
2029 if (iscsi != NULL) {
2030 if (iscsi_is_logged_in(iscsi)) {
2031 iscsi_logout_sync(iscsi);
2032 }
2033 iscsi_destroy_context(iscsi);
2034 }
2035 memset(iscsilun, 0, sizeof(IscsiLun));
2036 }
2037
2038 return ret;
2039 }
2040
2041 static void iscsi_close(BlockDriverState *bs)
2042 {
2043 IscsiLun *iscsilun = bs->opaque;
2044 struct iscsi_context *iscsi = iscsilun->iscsi;
2045
2046 iscsi_detach_aio_context(bs);
2047 if (iscsi_is_logged_in(iscsi)) {
2048 iscsi_logout_sync(iscsi);
2049 }
2050 iscsi_destroy_context(iscsi);
2051 if (iscsilun->dd) {
2052 g_free(iscsilun->dd->designator);
2053 g_free(iscsilun->dd);
2054 }
2055 g_free(iscsilun->zeroblock);
2056 iscsi_allocmap_free(iscsilun);
2057 qemu_mutex_destroy(&iscsilun->mutex);
2058 memset(iscsilun, 0, sizeof(IscsiLun));
2059 }
2060
2061 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
2062 {
2063 /* We don't actually refresh here, but just return data queried in
2064 * iscsi_open(): iscsi targets don't change their limits. */
2065
2066 IscsiLun *iscsilun = bs->opaque;
2067 uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
2068 unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size);
2069
2070 assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bdrv_is_sg(bs));
2071
2072 bs->bl.request_alignment = block_size;
2073
2074 if (iscsilun->bl.max_xfer_len) {
2075 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
2076 }
2077
2078 if (max_xfer_len * block_size < INT_MAX) {
2079 bs->bl.max_transfer = max_xfer_len * iscsilun->block_size;
2080 }
2081
2082 if (iscsilun->lbp.lbpu) {
2083 bs->bl.max_pdiscard =
2084 MIN_NON_ZERO(iscsilun->bl.max_unmap * iscsilun->block_size,
2085 (uint64_t)UINT32_MAX * iscsilun->block_size);
2086 bs->bl.pdiscard_alignment =
2087 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2088 } else {
2089 bs->bl.pdiscard_alignment = iscsilun->block_size;
2090 }
2091
2092 bs->bl.max_pwrite_zeroes =
2093 MIN_NON_ZERO(iscsilun->bl.max_ws_len * iscsilun->block_size,
2094 max_xfer_len * iscsilun->block_size);
2095
2096 if (iscsilun->lbp.lbpws) {
2097 bs->bl.pwrite_zeroes_alignment =
2098 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2099 } else {
2100 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size;
2101 }
2102 if (iscsilun->bl.opt_xfer_len &&
2103 iscsilun->bl.opt_xfer_len < INT_MAX / block_size) {
2104 bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len *
2105 iscsilun->block_size);
2106 }
2107 }
2108
2109 /* Note that this will not re-establish a connection with an iSCSI target - it
2110 * is effectively a NOP. */
2111 static int iscsi_reopen_prepare(BDRVReopenState *state,
2112 BlockReopenQueue *queue, Error **errp)
2113 {
2114 IscsiLun *iscsilun = state->bs->opaque;
2115
2116 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
2117 error_setg(errp, "Cannot open a write protected LUN as read-write");
2118 return -EACCES;
2119 }
2120 return 0;
2121 }
2122
2123 static void iscsi_reopen_commit(BDRVReopenState *reopen_state)
2124 {
2125 IscsiLun *iscsilun = reopen_state->bs->opaque;
2126
2127 /* the cache.direct status might have changed */
2128 if (iscsilun->allocmap != NULL) {
2129 iscsi_allocmap_init(iscsilun, reopen_state->flags);
2130 }
2131 }
2132
2133 static int coroutine_fn iscsi_co_truncate(BlockDriverState *bs, int64_t offset,
2134 bool exact, PreallocMode prealloc,
2135 BdrvRequestFlags flags, Error **errp)
2136 {
2137 IscsiLun *iscsilun = bs->opaque;
2138 int64_t cur_length;
2139 Error *local_err = NULL;
2140
2141 if (prealloc != PREALLOC_MODE_OFF) {
2142 error_setg(errp, "Unsupported preallocation mode '%s'",
2143 PreallocMode_str(prealloc));
2144 return -ENOTSUP;
2145 }
2146
2147 if (iscsilun->type != TYPE_DISK) {
2148 error_setg(errp, "Cannot resize non-disk iSCSI devices");
2149 return -ENOTSUP;
2150 }
2151
2152 iscsi_readcapacity_sync(iscsilun, &local_err);
2153 if (local_err != NULL) {
2154 error_propagate(errp, local_err);
2155 return -EIO;
2156 }
2157
2158 cur_length = iscsi_getlength(bs);
2159 if (offset != cur_length && exact) {
2160 error_setg(errp, "Cannot resize iSCSI devices");
2161 return -ENOTSUP;
2162 } else if (offset > cur_length) {
2163 error_setg(errp, "Cannot grow iSCSI devices");
2164 return -EINVAL;
2165 }
2166
2167 if (iscsilun->allocmap != NULL) {
2168 iscsi_allocmap_init(iscsilun, bs->open_flags);
2169 }
2170
2171 return 0;
2172 }
2173
2174 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2175 {
2176 IscsiLun *iscsilun = bs->opaque;
2177 bdi->cluster_size = iscsilun->cluster_size;
2178 return 0;
2179 }
2180
2181 static void coroutine_fn iscsi_co_invalidate_cache(BlockDriverState *bs,
2182 Error **errp)
2183 {
2184 IscsiLun *iscsilun = bs->opaque;
2185 iscsi_allocmap_invalidate(iscsilun);
2186 }
2187
2188 static int coroutine_fn iscsi_co_copy_range_from(BlockDriverState *bs,
2189 BdrvChild *src,
2190 int64_t src_offset,
2191 BdrvChild *dst,
2192 int64_t dst_offset,
2193 int64_t bytes,
2194 BdrvRequestFlags read_flags,
2195 BdrvRequestFlags write_flags)
2196 {
2197 return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes,
2198 read_flags, write_flags);
2199 }
2200
2201 static struct scsi_task *iscsi_xcopy_task(int param_len)
2202 {
2203 struct scsi_task *task;
2204
2205 task = g_new0(struct scsi_task, 1);
2206
2207 task->cdb[0] = EXTENDED_COPY;
2208 task->cdb[10] = (param_len >> 24) & 0xFF;
2209 task->cdb[11] = (param_len >> 16) & 0xFF;
2210 task->cdb[12] = (param_len >> 8) & 0xFF;
2211 task->cdb[13] = param_len & 0xFF;
2212 task->cdb_size = 16;
2213 task->xfer_dir = SCSI_XFER_WRITE;
2214 task->expxferlen = param_len;
2215
2216 return task;
2217 }
2218
2219 static void iscsi_populate_target_desc(unsigned char *desc, IscsiLun *lun)
2220 {
2221 struct scsi_inquiry_device_designator *dd = lun->dd;
2222
2223 memset(desc, 0, 32);
2224 desc[0] = 0xE4; /* IDENT_DESCR_TGT_DESCR */
2225 desc[4] = dd->code_set;
2226 desc[5] = (dd->designator_type & 0xF)
2227 | ((dd->association & 3) << 4);
2228 desc[7] = dd->designator_length;
2229 memcpy(desc + 8, dd->designator, MIN(dd->designator_length, 20));
2230
2231 desc[28] = 0;
2232 desc[29] = (lun->block_size >> 16) & 0xFF;
2233 desc[30] = (lun->block_size >> 8) & 0xFF;
2234 desc[31] = lun->block_size & 0xFF;
2235 }
2236
2237 static void iscsi_xcopy_desc_hdr(uint8_t *hdr, int dc, int cat, int src_index,
2238 int dst_index)
2239 {
2240 hdr[0] = 0x02; /* BLK_TO_BLK_SEG_DESCR */
2241 hdr[1] = ((dc << 1) | cat) & 0xFF;
2242 hdr[2] = (XCOPY_BLK2BLK_SEG_DESC_SIZE >> 8) & 0xFF;
2243 /* don't account for the first 4 bytes in descriptor header*/
2244 hdr[3] = (XCOPY_BLK2BLK_SEG_DESC_SIZE - 4 /* SEG_DESC_SRC_INDEX_OFFSET */) & 0xFF;
2245 hdr[4] = (src_index >> 8) & 0xFF;
2246 hdr[5] = src_index & 0xFF;
2247 hdr[6] = (dst_index >> 8) & 0xFF;
2248 hdr[7] = dst_index & 0xFF;
2249 }
2250
2251 static void iscsi_xcopy_populate_desc(uint8_t *desc, int dc, int cat,
2252 int src_index, int dst_index, int num_blks,
2253 uint64_t src_lba, uint64_t dst_lba)
2254 {
2255 iscsi_xcopy_desc_hdr(desc, dc, cat, src_index, dst_index);
2256
2257 /* The caller should verify the request size */
2258 assert(num_blks < 65536);
2259 desc[10] = (num_blks >> 8) & 0xFF;
2260 desc[11] = num_blks & 0xFF;
2261 desc[12] = (src_lba >> 56) & 0xFF;
2262 desc[13] = (src_lba >> 48) & 0xFF;
2263 desc[14] = (src_lba >> 40) & 0xFF;
2264 desc[15] = (src_lba >> 32) & 0xFF;
2265 desc[16] = (src_lba >> 24) & 0xFF;
2266 desc[17] = (src_lba >> 16) & 0xFF;
2267 desc[18] = (src_lba >> 8) & 0xFF;
2268 desc[19] = src_lba & 0xFF;
2269 desc[20] = (dst_lba >> 56) & 0xFF;
2270 desc[21] = (dst_lba >> 48) & 0xFF;
2271 desc[22] = (dst_lba >> 40) & 0xFF;
2272 desc[23] = (dst_lba >> 32) & 0xFF;
2273 desc[24] = (dst_lba >> 24) & 0xFF;
2274 desc[25] = (dst_lba >> 16) & 0xFF;
2275 desc[26] = (dst_lba >> 8) & 0xFF;
2276 desc[27] = dst_lba & 0xFF;
2277 }
2278
2279 static void iscsi_xcopy_populate_header(unsigned char *buf, int list_id, int str,
2280 int list_id_usage, int prio,
2281 int tgt_desc_len,
2282 int seg_desc_len, int inline_data_len)
2283 {
2284 buf[0] = list_id;
2285 buf[1] = ((str & 1) << 5) | ((list_id_usage & 3) << 3) | (prio & 7);
2286 buf[2] = (tgt_desc_len >> 8) & 0xFF;
2287 buf[3] = tgt_desc_len & 0xFF;
2288 buf[8] = (seg_desc_len >> 24) & 0xFF;
2289 buf[9] = (seg_desc_len >> 16) & 0xFF;
2290 buf[10] = (seg_desc_len >> 8) & 0xFF;
2291 buf[11] = seg_desc_len & 0xFF;
2292 buf[12] = (inline_data_len >> 24) & 0xFF;
2293 buf[13] = (inline_data_len >> 16) & 0xFF;
2294 buf[14] = (inline_data_len >> 8) & 0xFF;
2295 buf[15] = inline_data_len & 0xFF;
2296 }
2297
2298 static void iscsi_xcopy_data(struct iscsi_data *data,
2299 IscsiLun *src, int64_t src_lba,
2300 IscsiLun *dst, int64_t dst_lba,
2301 uint16_t num_blocks)
2302 {
2303 uint8_t *buf;
2304 const int src_offset = XCOPY_DESC_OFFSET;
2305 const int dst_offset = XCOPY_DESC_OFFSET + IDENT_DESCR_TGT_DESCR_SIZE;
2306 const int seg_offset = dst_offset + IDENT_DESCR_TGT_DESCR_SIZE;
2307
2308 data->size = XCOPY_DESC_OFFSET +
2309 IDENT_DESCR_TGT_DESCR_SIZE * 2 +
2310 XCOPY_BLK2BLK_SEG_DESC_SIZE;
2311 data->data = g_malloc0(data->size);
2312 buf = data->data;
2313
2314 /* Initialise the parameter list header */
2315 iscsi_xcopy_populate_header(buf, 1, 0, 2 /* LIST_ID_USAGE_DISCARD */,
2316 0, 2 * IDENT_DESCR_TGT_DESCR_SIZE,
2317 XCOPY_BLK2BLK_SEG_DESC_SIZE,
2318 0);
2319
2320 /* Initialise CSCD list with one src + one dst descriptor */
2321 iscsi_populate_target_desc(&buf[src_offset], src);
2322 iscsi_populate_target_desc(&buf[dst_offset], dst);
2323
2324 /* Initialise one segment descriptor */
2325 iscsi_xcopy_populate_desc(&buf[seg_offset], 0, 0, 0, 1, num_blocks,
2326 src_lba, dst_lba);
2327 }
2328
2329 static int coroutine_fn iscsi_co_copy_range_to(BlockDriverState *bs,
2330 BdrvChild *src,
2331 int64_t src_offset,
2332 BdrvChild *dst,
2333 int64_t dst_offset,
2334 int64_t bytes,
2335 BdrvRequestFlags read_flags,
2336 BdrvRequestFlags write_flags)
2337 {
2338 IscsiLun *dst_lun = dst->bs->opaque;
2339 IscsiLun *src_lun;
2340 struct IscsiTask iscsi_task;
2341 struct iscsi_data data;
2342 int r = 0;
2343 int block_size;
2344
2345 if (src->bs->drv->bdrv_co_copy_range_to != iscsi_co_copy_range_to) {
2346 return -ENOTSUP;
2347 }
2348 src_lun = src->bs->opaque;
2349
2350 if (!src_lun->dd || !dst_lun->dd) {
2351 return -ENOTSUP;
2352 }
2353 if (!is_byte_request_lun_aligned(dst_offset, bytes, dst_lun)) {
2354 return -ENOTSUP;
2355 }
2356 if (!is_byte_request_lun_aligned(src_offset, bytes, src_lun)) {
2357 return -ENOTSUP;
2358 }
2359 if (dst_lun->block_size != src_lun->block_size ||
2360 !dst_lun->block_size) {
2361 return -ENOTSUP;
2362 }
2363
2364 block_size = dst_lun->block_size;
2365 if (bytes / block_size > 65535) {
2366 return -ENOTSUP;
2367 }
2368
2369 iscsi_xcopy_data(&data,
2370 src_lun, src_offset / block_size,
2371 dst_lun, dst_offset / block_size,
2372 bytes / block_size);
2373
2374 iscsi_co_init_iscsitask(dst_lun, &iscsi_task);
2375
2376 qemu_mutex_lock(&dst_lun->mutex);
2377 iscsi_task.task = iscsi_xcopy_task(data.size);
2378 retry:
2379 if (iscsi_scsi_command_async(dst_lun->iscsi, dst_lun->lun,
2380 iscsi_task.task, iscsi_co_generic_cb,
2381 &data,
2382 &iscsi_task) != 0) {
2383 r = -EIO;
2384 goto out_unlock;
2385 }
2386
2387 iscsi_co_wait_for_task(&iscsi_task, dst_lun);
2388
2389 if (iscsi_task.do_retry) {
2390 iscsi_task.complete = 0;
2391 goto retry;
2392 }
2393
2394 if (iscsi_task.status != SCSI_STATUS_GOOD) {
2395 r = iscsi_task.err_code;
2396 goto out_unlock;
2397 }
2398
2399 out_unlock:
2400
2401 trace_iscsi_xcopy(src_lun, src_offset, dst_lun, dst_offset, bytes, r);
2402 g_free(iscsi_task.task);
2403 qemu_mutex_unlock(&dst_lun->mutex);
2404 g_free(iscsi_task.err_str);
2405 return r;
2406 }
2407
2408
2409 static const char *const iscsi_strong_runtime_opts[] = {
2410 "transport",
2411 "portal",
2412 "target",
2413 "user",
2414 "password",
2415 "password-secret",
2416 "lun",
2417 "initiator-name",
2418 "header-digest",
2419
2420 NULL
2421 };
2422
2423 static BlockDriver bdrv_iscsi = {
2424 .format_name = "iscsi",
2425 .protocol_name = "iscsi",
2426
2427 .instance_size = sizeof(IscsiLun),
2428 .bdrv_parse_filename = iscsi_parse_filename,
2429 .bdrv_file_open = iscsi_open,
2430 .bdrv_close = iscsi_close,
2431 .bdrv_co_create_opts = bdrv_co_create_opts_simple,
2432 .create_opts = &bdrv_create_opts_simple,
2433 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2434 .bdrv_reopen_commit = iscsi_reopen_commit,
2435 .bdrv_co_invalidate_cache = iscsi_co_invalidate_cache,
2436
2437 .bdrv_getlength = iscsi_getlength,
2438 .bdrv_get_info = iscsi_get_info,
2439 .bdrv_co_truncate = iscsi_co_truncate,
2440 .bdrv_refresh_limits = iscsi_refresh_limits,
2441
2442 .bdrv_co_block_status = iscsi_co_block_status,
2443 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2444 .bdrv_co_copy_range_from = iscsi_co_copy_range_from,
2445 .bdrv_co_copy_range_to = iscsi_co_copy_range_to,
2446 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2447 .bdrv_co_readv = iscsi_co_readv,
2448 .bdrv_co_writev = iscsi_co_writev,
2449 .bdrv_co_flush_to_disk = iscsi_co_flush,
2450
2451 #ifdef __linux__
2452 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2453 #endif
2454
2455 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2456 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2457
2458 .strong_runtime_opts = iscsi_strong_runtime_opts,
2459 };
2460
2461 #if LIBISCSI_API_VERSION >= (20160603)
2462 static BlockDriver bdrv_iser = {
2463 .format_name = "iser",
2464 .protocol_name = "iser",
2465
2466 .instance_size = sizeof(IscsiLun),
2467 .bdrv_parse_filename = iscsi_parse_filename,
2468 .bdrv_file_open = iscsi_open,
2469 .bdrv_close = iscsi_close,
2470 .bdrv_co_create_opts = bdrv_co_create_opts_simple,
2471 .create_opts = &bdrv_create_opts_simple,
2472 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2473 .bdrv_reopen_commit = iscsi_reopen_commit,
2474 .bdrv_co_invalidate_cache = iscsi_co_invalidate_cache,
2475
2476 .bdrv_getlength = iscsi_getlength,
2477 .bdrv_get_info = iscsi_get_info,
2478 .bdrv_co_truncate = iscsi_co_truncate,
2479 .bdrv_refresh_limits = iscsi_refresh_limits,
2480
2481 .bdrv_co_block_status = iscsi_co_block_status,
2482 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2483 .bdrv_co_copy_range_from = iscsi_co_copy_range_from,
2484 .bdrv_co_copy_range_to = iscsi_co_copy_range_to,
2485 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2486 .bdrv_co_readv = iscsi_co_readv,
2487 .bdrv_co_writev = iscsi_co_writev,
2488 .bdrv_co_flush_to_disk = iscsi_co_flush,
2489
2490 #ifdef __linux__
2491 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2492 #endif
2493
2494 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2495 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2496
2497 .strong_runtime_opts = iscsi_strong_runtime_opts,
2498 };
2499 #endif
2500
2501 static void iscsi_block_init(void)
2502 {
2503 bdrv_register(&bdrv_iscsi);
2504 #if LIBISCSI_API_VERSION >= (20160603)
2505 bdrv_register(&bdrv_iser);
2506 #endif
2507 }
2508
2509 block_init(iscsi_block_init);
armbru's QEMU hackery