?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
s390x/pci: change the device array to a list
[qemu.git] / block / iscsi.c
blob6aeeb9ec4fb37578eec8d836be2efe7dbee29507
1 /*
2 * QEMU Block driver for iSCSI images
3 *
4 * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
5 * Copyright (c) 2012-2016 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/scsi.h"
38 #include "qemu/iov.h"
39 #include "qemu/uuid.h"
40 #include "qmp-commands.h"
41 #include "qapi/qmp/qstring.h"
42 #include "crypto/secret.h"
43
44 #include <iscsi/iscsi.h>
45 #include <iscsi/scsi-lowlevel.h>
46
47 #ifdef __linux__
48 #include <scsi/sg.h>
49 #endif
50
51 typedef struct IscsiLun {
52 struct iscsi_context *iscsi;
53 AioContext *aio_context;
54 int lun;
55 enum scsi_inquiry_peripheral_device_type type;
56 int block_size;
57 uint64_t num_blocks;
58 int events;
59 QEMUTimer *nop_timer;
60 QEMUTimer *event_timer;
61 struct scsi_inquiry_logical_block_provisioning lbp;
62 struct scsi_inquiry_block_limits bl;
63 unsigned char *zeroblock;
64 /* The allocmap tracks which clusters (pages) on the iSCSI target are
65 * allocated and which are not. In case a target returns zeros for
66 * unallocated pages (iscsilun->lprz) we can directly return zeros instead
67 * of reading zeros over the wire if a read request falls within an
68 * unallocated block. As there are 3 possible states we need 2 bitmaps to
69 * track. allocmap_valid keeps track if QEMU's information about a page is
70 * valid. allocmap tracks if a page is allocated or not. In case QEMU has no
71 * valid information about a page the corresponding allocmap entry should be
72 * switched to unallocated as well to force a new lookup of the allocation
73 * status as lookups are generally skipped if a page is suspect to be
74 * allocated. If a iSCSI target is opened with cache.direct = on the
75 * allocmap_valid does not exist turning all cached information invalid so
76 * that a fresh lookup is made for any page even if allocmap entry returns
77 * it's unallocated. */
78 unsigned long *allocmap;
79 unsigned long *allocmap_valid;
80 long allocmap_size;
81 int cluster_sectors;
82 bool use_16_for_rw;
83 bool write_protected;
84 bool lbpme;
85 bool lbprz;
86 bool dpofua;
87 bool has_write_same;
88 bool request_timed_out;
89 } IscsiLun;
90
91 typedef struct IscsiTask {
92 int status;
93 int complete;
94 int retries;
95 int do_retry;
96 struct scsi_task *task;
97 Coroutine *co;
98 IscsiLun *iscsilun;
99 QEMUTimer retry_timer;
100 int err_code;
101 } IscsiTask;
102
103 typedef struct IscsiAIOCB {
104 BlockAIOCB common;
105 QEMUIOVector *qiov;
106 QEMUBH *bh;
107 IscsiLun *iscsilun;
108 struct scsi_task *task;
109 uint8_t *buf;
110 int status;
111 int64_t sector_num;
112 int nb_sectors;
113 int ret;
114 #ifdef __linux__
115 sg_io_hdr_t *ioh;
116 #endif
117 } IscsiAIOCB;
118
119 /* libiscsi uses time_t so its enough to process events every second */
120 #define EVENT_INTERVAL 1000
121 #define NOP_INTERVAL 5000
122 #define MAX_NOP_FAILURES 3
123 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times)
124 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768};
125
126 /* this threshold is a trade-off knob to choose between
127 * the potential additional overhead of an extra GET_LBA_STATUS request
128 * vs. unnecessarily reading a lot of zero sectors over the wire.
129 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
130 * sectors we check the allocation status of the area covered by the
131 * request first if the allocationmap indicates that the area might be
132 * unallocated. */
133 #define ISCSI_CHECKALLOC_THRES 64
134
135 static void
136 iscsi_bh_cb(void *p)
137 {
138 IscsiAIOCB *acb = p;
139
140 qemu_bh_delete(acb->bh);
141
142 g_free(acb->buf);
143 acb->buf = NULL;
144
145 acb->common.cb(acb->common.opaque, acb->status);
146
147 if (acb->task != NULL) {
148 scsi_free_scsi_task(acb->task);
149 acb->task = NULL;
150 }
151
152 qemu_aio_unref(acb);
153 }
154
155 static void
156 iscsi_schedule_bh(IscsiAIOCB *acb)
157 {
158 if (acb->bh) {
159 return;
160 }
161 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb);
162 qemu_bh_schedule(acb->bh);
163 }
164
165 static void iscsi_co_generic_bh_cb(void *opaque)
166 {
167 struct IscsiTask *iTask = opaque;
168 iTask->complete = 1;
169 qemu_coroutine_enter(iTask->co);
170 }
171
172 static void iscsi_retry_timer_expired(void *opaque)
173 {
174 struct IscsiTask *iTask = opaque;
175 iTask->complete = 1;
176 if (iTask->co) {
177 qemu_coroutine_enter(iTask->co);
178 }
179 }
180
181 static inline unsigned exp_random(double mean)
182 {
183 return -mean * log((double)rand() / RAND_MAX);
184 }
185
186 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in
187 * libiscsi 1.10.0, together with other constants we need. Use it as
188 * a hint that we have to define them ourselves if needed, to keep the
189 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for
190 * the test because SCSI_STATUS_* is an enum.
191 *
192 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes
193 * an enum, check against the LIBISCSI_API_VERSION macro, which was
194 * introduced in 1.11.0. If it is present, there is no need to define
195 * anything.
196 */
197 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \
198 !defined(LIBISCSI_API_VERSION)
199 #define SCSI_STATUS_TASK_SET_FULL 0x28
200 #define SCSI_STATUS_TIMEOUT 0x0f000002
201 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600
202 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00
203 #endif
204
205 #ifndef LIBISCSI_API_VERSION
206 #define LIBISCSI_API_VERSION 20130701
207 #endif
208
209 static int iscsi_translate_sense(struct scsi_sense *sense)
210 {
211 int ret;
212
213 switch (sense->key) {
214 case SCSI_SENSE_NOT_READY:
215 return -EBUSY;
216 case SCSI_SENSE_DATA_PROTECTION:
217 return -EACCES;
218 case SCSI_SENSE_COMMAND_ABORTED:
219 return -ECANCELED;
220 case SCSI_SENSE_ILLEGAL_REQUEST:
221 /* Parse ASCQ */
222 break;
223 default:
224 return -EIO;
225 }
226 switch (sense->ascq) {
227 case SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR:
228 case SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE:
229 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB:
230 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST:
231 ret = -EINVAL;
232 break;
233 case SCSI_SENSE_ASCQ_LBA_OUT_OF_RANGE:
234 ret = -ENOSPC;
235 break;
236 case SCSI_SENSE_ASCQ_LOGICAL_UNIT_NOT_SUPPORTED:
237 ret = -ENOTSUP;
238 break;
239 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT:
240 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_CLOSED:
241 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_OPEN:
242 ret = -ENOMEDIUM;
243 break;
244 case SCSI_SENSE_ASCQ_WRITE_PROTECTED:
245 ret = -EACCES;
246 break;
247 default:
248 ret = -EIO;
249 break;
250 }
251 return ret;
252 }
253
254 static void
255 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
256 void *command_data, void *opaque)
257 {
258 struct IscsiTask *iTask = opaque;
259 struct scsi_task *task = command_data;
260
261 iTask->status = status;
262 iTask->do_retry = 0;
263 iTask->task = task;
264
265 if (status != SCSI_STATUS_GOOD) {
266 if (iTask->retries++ < ISCSI_CMD_RETRIES) {
267 if (status == SCSI_STATUS_CHECK_CONDITION
268 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
269 error_report("iSCSI CheckCondition: %s",
270 iscsi_get_error(iscsi));
271 iTask->do_retry = 1;
272 goto out;
273 }
274 if (status == SCSI_STATUS_BUSY ||
275 status == SCSI_STATUS_TIMEOUT ||
276 status == SCSI_STATUS_TASK_SET_FULL) {
277 unsigned retry_time =
278 exp_random(iscsi_retry_times[iTask->retries - 1]);
279 if (status == SCSI_STATUS_TIMEOUT) {
280 /* make sure the request is rescheduled AFTER the
281 * reconnect is initiated */
282 retry_time = EVENT_INTERVAL * 2;
283 iTask->iscsilun->request_timed_out = true;
284 }
285 error_report("iSCSI Busy/TaskSetFull/TimeOut"
286 " (retry #%u in %u ms): %s",
287 iTask->retries, retry_time,
288 iscsi_get_error(iscsi));
289 aio_timer_init(iTask->iscsilun->aio_context,
290 &iTask->retry_timer, QEMU_CLOCK_REALTIME,
291 SCALE_MS, iscsi_retry_timer_expired, iTask);
292 timer_mod(&iTask->retry_timer,
293 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
294 iTask->do_retry = 1;
295 return;
296 }
297 }
298 iTask->err_code = iscsi_translate_sense(&task->sense);
299 error_report("iSCSI Failure: %s", iscsi_get_error(iscsi));
300 }
301
302 out:
303 if (iTask->co) {
304 aio_bh_schedule_oneshot(iTask->iscsilun->aio_context,
305 iscsi_co_generic_bh_cb, iTask);
306 } else {
307 iTask->complete = 1;
308 }
309 }
310
311 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
312 {
313 *iTask = (struct IscsiTask) {
314 .co = qemu_coroutine_self(),
315 .iscsilun = iscsilun,
316 };
317 }
318
319 static void
320 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
321 void *private_data)
322 {
323 IscsiAIOCB *acb = private_data;
324
325 acb->status = -ECANCELED;
326 iscsi_schedule_bh(acb);
327 }
328
329 static void
330 iscsi_aio_cancel(BlockAIOCB *blockacb)
331 {
332 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
333 IscsiLun *iscsilun = acb->iscsilun;
334
335 if (acb->status != -EINPROGRESS) {
336 return;
337 }
338
339 /* send a task mgmt call to the target to cancel the task on the target */
340 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
341 iscsi_abort_task_cb, acb);
342
343 }
344
345 static const AIOCBInfo iscsi_aiocb_info = {
346 .aiocb_size = sizeof(IscsiAIOCB),
347 .cancel_async = iscsi_aio_cancel,
348 };
349
350
351 static void iscsi_process_read(void *arg);
352 static void iscsi_process_write(void *arg);
353
354 static void
355 iscsi_set_events(IscsiLun *iscsilun)
356 {
357 struct iscsi_context *iscsi = iscsilun->iscsi;
358 int ev = iscsi_which_events(iscsi);
359
360 if (ev != iscsilun->events) {
361 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi),
362 false,
363 (ev & POLLIN) ? iscsi_process_read : NULL,
364 (ev & POLLOUT) ? iscsi_process_write : NULL,
365 NULL,
366 iscsilun);
367 iscsilun->events = ev;
368 }
369 }
370
371 static void iscsi_timed_check_events(void *opaque)
372 {
373 IscsiLun *iscsilun = opaque;
374
375 /* check for timed out requests */
376 iscsi_service(iscsilun->iscsi, 0);
377
378 if (iscsilun->request_timed_out) {
379 iscsilun->request_timed_out = false;
380 iscsi_reconnect(iscsilun->iscsi);
381 }
382
383 /* newer versions of libiscsi may return zero events. Ensure we are able
384 * to return to service once this situation changes. */
385 iscsi_set_events(iscsilun);
386
387 timer_mod(iscsilun->event_timer,
388 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
389 }
390
391 static void
392 iscsi_process_read(void *arg)
393 {
394 IscsiLun *iscsilun = arg;
395 struct iscsi_context *iscsi = iscsilun->iscsi;
396
397 iscsi_service(iscsi, POLLIN);
398 iscsi_set_events(iscsilun);
399 }
400
401 static void
402 iscsi_process_write(void *arg)
403 {
404 IscsiLun *iscsilun = arg;
405 struct iscsi_context *iscsi = iscsilun->iscsi;
406
407 iscsi_service(iscsi, POLLOUT);
408 iscsi_set_events(iscsilun);
409 }
410
411 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
412 {
413 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
414 }
415
416 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
417 {
418 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
419 }
420
421 static bool is_byte_request_lun_aligned(int64_t offset, int count,
422 IscsiLun *iscsilun)
423 {
424 if (offset % iscsilun->block_size || count % iscsilun->block_size) {
425 error_report("iSCSI misaligned request: "
426 "iscsilun->block_size %u, offset %" PRIi64
427 ", count %d",
428 iscsilun->block_size, offset, count);
429 return false;
430 }
431 return true;
432 }
433
434 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors,
435 IscsiLun *iscsilun)
436 {
437 assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS);
438 return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS,
439 nb_sectors << BDRV_SECTOR_BITS,
440 iscsilun);
441 }
442
443 static void iscsi_allocmap_free(IscsiLun *iscsilun)
444 {
445 g_free(iscsilun->allocmap);
446 g_free(iscsilun->allocmap_valid);
447 iscsilun->allocmap = NULL;
448 iscsilun->allocmap_valid = NULL;
449 }
450
451
452 static int iscsi_allocmap_init(IscsiLun *iscsilun, int open_flags)
453 {
454 iscsi_allocmap_free(iscsilun);
455
456 iscsilun->allocmap_size =
457 DIV_ROUND_UP(sector_lun2qemu(iscsilun->num_blocks, iscsilun),
458 iscsilun->cluster_sectors);
459
460 iscsilun->allocmap = bitmap_try_new(iscsilun->allocmap_size);
461 if (!iscsilun->allocmap) {
462 return -ENOMEM;
463 }
464
465 if (open_flags & BDRV_O_NOCACHE) {
466 /* in case that cache.direct = on all allocmap entries are
467 * treated as invalid to force a relookup of the block
468 * status on every read request */
469 return 0;
470 }
471
472 iscsilun->allocmap_valid = bitmap_try_new(iscsilun->allocmap_size);
473 if (!iscsilun->allocmap_valid) {
474 /* if we are under memory pressure free the allocmap as well */
475 iscsi_allocmap_free(iscsilun);
476 return -ENOMEM;
477 }
478
479 return 0;
480 }
481
482 static void
483 iscsi_allocmap_update(IscsiLun *iscsilun, int64_t sector_num,
484 int nb_sectors, bool allocated, bool valid)
485 {
486 int64_t cl_num_expanded, nb_cls_expanded, cl_num_shrunk, nb_cls_shrunk;
487
488 if (iscsilun->allocmap == NULL) {
489 return;
490 }
491 /* expand to entirely contain all affected clusters */
492 cl_num_expanded = sector_num / iscsilun->cluster_sectors;
493 nb_cls_expanded = DIV_ROUND_UP(sector_num + nb_sectors,
494 iscsilun->cluster_sectors) - cl_num_expanded;
495 /* shrink to touch only completely contained clusters */
496 cl_num_shrunk = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors);
497 nb_cls_shrunk = (sector_num + nb_sectors) / iscsilun->cluster_sectors
498 - cl_num_shrunk;
499 if (allocated) {
500 bitmap_set(iscsilun->allocmap, cl_num_expanded, nb_cls_expanded);
501 } else {
502 bitmap_clear(iscsilun->allocmap, cl_num_shrunk, nb_cls_shrunk);
503 }
504
505 if (iscsilun->allocmap_valid == NULL) {
506 return;
507 }
508 if (valid) {
509 bitmap_set(iscsilun->allocmap_valid, cl_num_shrunk, nb_cls_shrunk);
510 } else {
511 bitmap_clear(iscsilun->allocmap_valid, cl_num_expanded,
512 nb_cls_expanded);
513 }
514 }
515
516 static void
517 iscsi_allocmap_set_allocated(IscsiLun *iscsilun, int64_t sector_num,
518 int nb_sectors)
519 {
520 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, true, true);
521 }
522
523 static void
524 iscsi_allocmap_set_unallocated(IscsiLun *iscsilun, int64_t sector_num,
525 int nb_sectors)
526 {
527 /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update
528 * is ignored, so this will in effect be an iscsi_allocmap_set_invalid.
529 */
530 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, false, true);
531 }
532
533 static void iscsi_allocmap_set_invalid(IscsiLun *iscsilun, int64_t sector_num,
534 int nb_sectors)
535 {
536 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, false, false);
537 }
538
539 static void iscsi_allocmap_invalidate(IscsiLun *iscsilun)
540 {
541 if (iscsilun->allocmap) {
542 bitmap_zero(iscsilun->allocmap, iscsilun->allocmap_size);
543 }
544 if (iscsilun->allocmap_valid) {
545 bitmap_zero(iscsilun->allocmap_valid, iscsilun->allocmap_size);
546 }
547 }
548
549 static inline bool
550 iscsi_allocmap_is_allocated(IscsiLun *iscsilun, int64_t sector_num,
551 int nb_sectors)
552 {
553 unsigned long size;
554 if (iscsilun->allocmap == NULL) {
555 return true;
556 }
557 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
558 return !(find_next_bit(iscsilun->allocmap, size,
559 sector_num / iscsilun->cluster_sectors) == size);
560 }
561
562 static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun,
563 int64_t sector_num, int nb_sectors)
564 {
565 unsigned long size;
566 if (iscsilun->allocmap_valid == NULL) {
567 return false;
568 }
569 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
570 return (find_next_zero_bit(iscsilun->allocmap_valid, size,
571 sector_num / iscsilun->cluster_sectors) == size);
572 }
573
574 static int coroutine_fn
575 iscsi_co_writev_flags(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
576 QEMUIOVector *iov, int flags)
577 {
578 IscsiLun *iscsilun = bs->opaque;
579 struct IscsiTask iTask;
580 uint64_t lba;
581 uint32_t num_sectors;
582 bool fua = flags & BDRV_REQ_FUA;
583
584 if (fua) {
585 assert(iscsilun->dpofua);
586 }
587 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
588 return -EINVAL;
589 }
590
591 if (bs->bl.max_transfer) {
592 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
593 }
594
595 lba = sector_qemu2lun(sector_num, iscsilun);
596 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
597 iscsi_co_init_iscsitask(iscsilun, &iTask);
598 retry:
599 if (iscsilun->use_16_for_rw) {
600 #if LIBISCSI_API_VERSION >= (20160603)
601 iTask.task = iscsi_write16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
602 NULL, num_sectors * iscsilun->block_size,
603 iscsilun->block_size, 0, 0, fua, 0, 0,
604 iscsi_co_generic_cb, &iTask,
605 (struct scsi_iovec *)iov->iov, iov->niov);
606 } else {
607 iTask.task = iscsi_write10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
608 NULL, num_sectors * iscsilun->block_size,
609 iscsilun->block_size, 0, 0, fua, 0, 0,
610 iscsi_co_generic_cb, &iTask,
611 (struct scsi_iovec *)iov->iov, iov->niov);
612 }
613 #else
614 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
615 NULL, num_sectors * iscsilun->block_size,
616 iscsilun->block_size, 0, 0, fua, 0, 0,
617 iscsi_co_generic_cb, &iTask);
618 } else {
619 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
620 NULL, num_sectors * iscsilun->block_size,
621 iscsilun->block_size, 0, 0, fua, 0, 0,
622 iscsi_co_generic_cb, &iTask);
623 }
624 #endif
625 if (iTask.task == NULL) {
626 return -ENOMEM;
627 }
628 #if LIBISCSI_API_VERSION < (20160603)
629 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
630 iov->niov);
631 #endif
632 while (!iTask.complete) {
633 iscsi_set_events(iscsilun);
634 qemu_coroutine_yield();
635 }
636
637 if (iTask.task != NULL) {
638 scsi_free_scsi_task(iTask.task);
639 iTask.task = NULL;
640 }
641
642 if (iTask.do_retry) {
643 iTask.complete = 0;
644 goto retry;
645 }
646
647 if (iTask.status != SCSI_STATUS_GOOD) {
648 iscsi_allocmap_set_invalid(iscsilun, sector_num, nb_sectors);
649 return iTask.err_code;
650 }
651
652 iscsi_allocmap_set_allocated(iscsilun, sector_num, nb_sectors);
653
654 return 0;
655 }
656
657
658
659 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
660 int64_t sector_num,
661 int nb_sectors, int *pnum,
662 BlockDriverState **file)
663 {
664 IscsiLun *iscsilun = bs->opaque;
665 struct scsi_get_lba_status *lbas = NULL;
666 struct scsi_lba_status_descriptor *lbasd = NULL;
667 struct IscsiTask iTask;
668 int64_t ret;
669
670 iscsi_co_init_iscsitask(iscsilun, &iTask);
671
672 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
673 ret = -EINVAL;
674 goto out;
675 }
676
677 /* default to all sectors allocated */
678 ret = BDRV_BLOCK_DATA;
679 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
680 *pnum = nb_sectors;
681
682 /* LUN does not support logical block provisioning */
683 if (!iscsilun->lbpme) {
684 goto out;
685 }
686
687 retry:
688 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
689 sector_qemu2lun(sector_num, iscsilun),
690 8 + 16, iscsi_co_generic_cb,
691 &iTask) == NULL) {
692 ret = -ENOMEM;
693 goto out;
694 }
695
696 while (!iTask.complete) {
697 iscsi_set_events(iscsilun);
698 qemu_coroutine_yield();
699 }
700
701 if (iTask.do_retry) {
702 if (iTask.task != NULL) {
703 scsi_free_scsi_task(iTask.task);
704 iTask.task = NULL;
705 }
706 iTask.complete = 0;
707 goto retry;
708 }
709
710 if (iTask.status != SCSI_STATUS_GOOD) {
711 /* in case the get_lba_status_callout fails (i.e.
712 * because the device is busy or the cmd is not
713 * supported) we pretend all blocks are allocated
714 * for backwards compatibility */
715 goto out;
716 }
717
718 lbas = scsi_datain_unmarshall(iTask.task);
719 if (lbas == NULL) {
720 ret = -EIO;
721 goto out;
722 }
723
724 lbasd = &lbas->descriptors[0];
725
726 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
727 ret = -EIO;
728 goto out;
729 }
730
731 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
732
733 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
734 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
735 ret &= ~BDRV_BLOCK_DATA;
736 if (iscsilun->lbprz) {
737 ret |= BDRV_BLOCK_ZERO;
738 }
739 }
740
741 if (ret & BDRV_BLOCK_ZERO) {
742 iscsi_allocmap_set_unallocated(iscsilun, sector_num, *pnum);
743 } else {
744 iscsi_allocmap_set_allocated(iscsilun, sector_num, *pnum);
745 }
746
747 if (*pnum > nb_sectors) {
748 *pnum = nb_sectors;
749 }
750 out:
751 if (iTask.task != NULL) {
752 scsi_free_scsi_task(iTask.task);
753 }
754 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) {
755 *file = bs;
756 }
757 return ret;
758 }
759
760 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
761 int64_t sector_num, int nb_sectors,
762 QEMUIOVector *iov)
763 {
764 IscsiLun *iscsilun = bs->opaque;
765 struct IscsiTask iTask;
766 uint64_t lba;
767 uint32_t num_sectors;
768
769 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
770 return -EINVAL;
771 }
772
773 if (bs->bl.max_transfer) {
774 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
775 }
776
777 /* if cache.direct is off and we have a valid entry in our allocation map
778 * we can skip checking the block status and directly return zeroes if
779 * the request falls within an unallocated area */
780 if (iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) &&
781 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
782 qemu_iovec_memset(iov, 0, 0x00, iov->size);
783 return 0;
784 }
785
786 if (nb_sectors >= ISCSI_CHECKALLOC_THRES &&
787 !iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) &&
788 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
789 int pnum;
790 BlockDriverState *file;
791 /* check the block status from the beginning of the cluster
792 * containing the start sector */
793 int64_t ret = iscsi_co_get_block_status(bs,
794 sector_num - sector_num % iscsilun->cluster_sectors,
795 BDRV_REQUEST_MAX_SECTORS, &pnum, &file);
796 if (ret < 0) {
797 return ret;
798 }
799 /* if the whole request falls into an unallocated area we can avoid
800 * to read and directly return zeroes instead */
801 if (ret & BDRV_BLOCK_ZERO &&
802 pnum >= nb_sectors + sector_num % iscsilun->cluster_sectors) {
803 qemu_iovec_memset(iov, 0, 0x00, iov->size);
804 return 0;
805 }
806 }
807
808 lba = sector_qemu2lun(sector_num, iscsilun);
809 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
810
811 iscsi_co_init_iscsitask(iscsilun, &iTask);
812 retry:
813 if (iscsilun->use_16_for_rw) {
814 #if LIBISCSI_API_VERSION >= (20160603)
815 iTask.task = iscsi_read16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
816 num_sectors * iscsilun->block_size,
817 iscsilun->block_size, 0, 0, 0, 0, 0,
818 iscsi_co_generic_cb, &iTask,
819 (struct scsi_iovec *)iov->iov, iov->niov);
820 } else {
821 iTask.task = iscsi_read10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
822 num_sectors * iscsilun->block_size,
823 iscsilun->block_size,
824 0, 0, 0, 0, 0,
825 iscsi_co_generic_cb, &iTask,
826 (struct scsi_iovec *)iov->iov, iov->niov);
827 }
828 #else
829 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
830 num_sectors * iscsilun->block_size,
831 iscsilun->block_size, 0, 0, 0, 0, 0,
832 iscsi_co_generic_cb, &iTask);
833 } else {
834 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
835 num_sectors * iscsilun->block_size,
836 iscsilun->block_size,
837 0, 0, 0, 0, 0,
838 iscsi_co_generic_cb, &iTask);
839 }
840 #endif
841 if (iTask.task == NULL) {
842 return -ENOMEM;
843 }
844 #if LIBISCSI_API_VERSION < (20160603)
845 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
846 #endif
847 while (!iTask.complete) {
848 iscsi_set_events(iscsilun);
849 qemu_coroutine_yield();
850 }
851
852 if (iTask.task != NULL) {
853 scsi_free_scsi_task(iTask.task);
854 iTask.task = NULL;
855 }
856
857 if (iTask.do_retry) {
858 iTask.complete = 0;
859 goto retry;
860 }
861
862 if (iTask.status != SCSI_STATUS_GOOD) {
863 return iTask.err_code;
864 }
865
866 return 0;
867 }
868
869 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
870 {
871 IscsiLun *iscsilun = bs->opaque;
872 struct IscsiTask iTask;
873
874 iscsi_co_init_iscsitask(iscsilun, &iTask);
875 retry:
876 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
877 0, iscsi_co_generic_cb, &iTask) == NULL) {
878 return -ENOMEM;
879 }
880
881 while (!iTask.complete) {
882 iscsi_set_events(iscsilun);
883 qemu_coroutine_yield();
884 }
885
886 if (iTask.task != NULL) {
887 scsi_free_scsi_task(iTask.task);
888 iTask.task = NULL;
889 }
890
891 if (iTask.do_retry) {
892 iTask.complete = 0;
893 goto retry;
894 }
895
896 if (iTask.status != SCSI_STATUS_GOOD) {
897 return iTask.err_code;
898 }
899
900 return 0;
901 }
902
903 #ifdef __linux__
904 static void
905 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
906 void *command_data, void *opaque)
907 {
908 IscsiAIOCB *acb = opaque;
909
910 g_free(acb->buf);
911 acb->buf = NULL;
912
913 acb->status = 0;
914 if (status < 0) {
915 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
916 iscsi_get_error(iscsi));
917 acb->status = iscsi_translate_sense(&acb->task->sense);
918 }
919
920 acb->ioh->driver_status = 0;
921 acb->ioh->host_status = 0;
922 acb->ioh->resid = 0;
923 acb->ioh->status = status;
924
925 #define SG_ERR_DRIVER_SENSE 0x08
926
927 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
928 int ss;
929
930 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
931
932 acb->ioh->sb_len_wr = acb->task->datain.size - 2;
933 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
934 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
935 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
936 }
937
938 iscsi_schedule_bh(acb);
939 }
940
941 static void iscsi_ioctl_bh_completion(void *opaque)
942 {
943 IscsiAIOCB *acb = opaque;
944
945 qemu_bh_delete(acb->bh);
946 acb->common.cb(acb->common.opaque, acb->ret);
947 qemu_aio_unref(acb);
948 }
949
950 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
951 {
952 BlockDriverState *bs = acb->common.bs;
953 IscsiLun *iscsilun = bs->opaque;
954 int ret = 0;
955
956 switch (req) {
957 case SG_GET_VERSION_NUM:
958 *(int *)buf = 30000;
959 break;
960 case SG_GET_SCSI_ID:
961 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
962 break;
963 default:
964 ret = -EINVAL;
965 }
966 assert(!acb->bh);
967 acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
968 iscsi_ioctl_bh_completion, acb);
969 acb->ret = ret;
970 qemu_bh_schedule(acb->bh);
971 }
972
973 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
974 unsigned long int req, void *buf,
975 BlockCompletionFunc *cb, void *opaque)
976 {
977 IscsiLun *iscsilun = bs->opaque;
978 struct iscsi_context *iscsi = iscsilun->iscsi;
979 struct iscsi_data data;
980 IscsiAIOCB *acb;
981
982 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
983
984 acb->iscsilun = iscsilun;
985 acb->bh = NULL;
986 acb->status = -EINPROGRESS;
987 acb->buf = NULL;
988 acb->ioh = buf;
989
990 if (req != SG_IO) {
991 iscsi_ioctl_handle_emulated(acb, req, buf);
992 return &acb->common;
993 }
994
995 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) {
996 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)",
997 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE);
998 qemu_aio_unref(acb);
999 return NULL;
1000 }
1001
1002 acb->task = malloc(sizeof(struct scsi_task));
1003 if (acb->task == NULL) {
1004 error_report("iSCSI: Failed to allocate task for scsi command. %s",
1005 iscsi_get_error(iscsi));
1006 qemu_aio_unref(acb);
1007 return NULL;
1008 }
1009 memset(acb->task, 0, sizeof(struct scsi_task));
1010
1011 switch (acb->ioh->dxfer_direction) {
1012 case SG_DXFER_TO_DEV:
1013 acb->task->xfer_dir = SCSI_XFER_WRITE;
1014 break;
1015 case SG_DXFER_FROM_DEV:
1016 acb->task->xfer_dir = SCSI_XFER_READ;
1017 break;
1018 default:
1019 acb->task->xfer_dir = SCSI_XFER_NONE;
1020 break;
1021 }
1022
1023 acb->task->cdb_size = acb->ioh->cmd_len;
1024 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
1025 acb->task->expxferlen = acb->ioh->dxfer_len;
1026
1027 data.size = 0;
1028 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
1029 if (acb->ioh->iovec_count == 0) {
1030 data.data = acb->ioh->dxferp;
1031 data.size = acb->ioh->dxfer_len;
1032 } else {
1033 scsi_task_set_iov_out(acb->task,
1034 (struct scsi_iovec *) acb->ioh->dxferp,
1035 acb->ioh->iovec_count);
1036 }
1037 }
1038
1039 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
1040 iscsi_aio_ioctl_cb,
1041 (data.size > 0) ? &data : NULL,
1042 acb) != 0) {
1043 scsi_free_scsi_task(acb->task);
1044 qemu_aio_unref(acb);
1045 return NULL;
1046 }
1047
1048 /* tell libiscsi to read straight into the buffer we got from ioctl */
1049 if (acb->task->xfer_dir == SCSI_XFER_READ) {
1050 if (acb->ioh->iovec_count == 0) {
1051 scsi_task_add_data_in_buffer(acb->task,
1052 acb->ioh->dxfer_len,
1053 acb->ioh->dxferp);
1054 } else {
1055 scsi_task_set_iov_in(acb->task,
1056 (struct scsi_iovec *) acb->ioh->dxferp,
1057 acb->ioh->iovec_count);
1058 }
1059 }
1060
1061 iscsi_set_events(iscsilun);
1062
1063 return &acb->common;
1064 }
1065
1066 #endif
1067
1068 static int64_t
1069 iscsi_getlength(BlockDriverState *bs)
1070 {
1071 IscsiLun *iscsilun = bs->opaque;
1072 int64_t len;
1073
1074 len = iscsilun->num_blocks;
1075 len *= iscsilun->block_size;
1076
1077 return len;
1078 }
1079
1080 static int
1081 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, int count)
1082 {
1083 IscsiLun *iscsilun = bs->opaque;
1084 struct IscsiTask iTask;
1085 struct unmap_list list;
1086
1087 if (!is_byte_request_lun_aligned(offset, count, iscsilun)) {
1088 return -ENOTSUP;
1089 }
1090
1091 if (!iscsilun->lbp.lbpu) {
1092 /* UNMAP is not supported by the target */
1093 return 0;
1094 }
1095
1096 list.lba = offset / iscsilun->block_size;
1097 list.num = count / iscsilun->block_size;
1098
1099 iscsi_co_init_iscsitask(iscsilun, &iTask);
1100 retry:
1101 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
1102 iscsi_co_generic_cb, &iTask) == NULL) {
1103 return -ENOMEM;
1104 }
1105
1106 while (!iTask.complete) {
1107 iscsi_set_events(iscsilun);
1108 qemu_coroutine_yield();
1109 }
1110
1111 if (iTask.task != NULL) {
1112 scsi_free_scsi_task(iTask.task);
1113 iTask.task = NULL;
1114 }
1115
1116 if (iTask.do_retry) {
1117 iTask.complete = 0;
1118 goto retry;
1119 }
1120
1121 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
1122 /* the target might fail with a check condition if it
1123 is not happy with the alignment of the UNMAP request
1124 we silently fail in this case */
1125 return 0;
1126 }
1127
1128 if (iTask.status != SCSI_STATUS_GOOD) {
1129 return iTask.err_code;
1130 }
1131
1132 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1133 count >> BDRV_SECTOR_BITS);
1134
1135 return 0;
1136 }
1137
1138 static int
1139 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1140 int count, BdrvRequestFlags flags)
1141 {
1142 IscsiLun *iscsilun = bs->opaque;
1143 struct IscsiTask iTask;
1144 uint64_t lba;
1145 uint32_t nb_blocks;
1146 bool use_16_for_ws = iscsilun->use_16_for_rw;
1147
1148 if (!is_byte_request_lun_aligned(offset, count, iscsilun)) {
1149 return -ENOTSUP;
1150 }
1151
1152 if (flags & BDRV_REQ_MAY_UNMAP) {
1153 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
1154 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
1155 use_16_for_ws = true;
1156 }
1157 if (use_16_for_ws && !iscsilun->lbp.lbpws) {
1158 /* WRITESAME16 with UNMAP is not supported by the target,
1159 * fall back and try WRITESAME10/16 without UNMAP */
1160 flags &= ~BDRV_REQ_MAY_UNMAP;
1161 use_16_for_ws = iscsilun->use_16_for_rw;
1162 }
1163 }
1164
1165 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1166 /* WRITESAME without UNMAP is not supported by the target */
1167 return -ENOTSUP;
1168 }
1169
1170 lba = offset / iscsilun->block_size;
1171 nb_blocks = count / iscsilun->block_size;
1172
1173 if (iscsilun->zeroblock == NULL) {
1174 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1175 if (iscsilun->zeroblock == NULL) {
1176 return -ENOMEM;
1177 }
1178 }
1179
1180 iscsi_co_init_iscsitask(iscsilun, &iTask);
1181 retry:
1182 if (use_16_for_ws) {
1183 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1184 iscsilun->zeroblock, iscsilun->block_size,
1185 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1186 0, 0, iscsi_co_generic_cb, &iTask);
1187 } else {
1188 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1189 iscsilun->zeroblock, iscsilun->block_size,
1190 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1191 0, 0, iscsi_co_generic_cb, &iTask);
1192 }
1193 if (iTask.task == NULL) {
1194 return -ENOMEM;
1195 }
1196
1197 while (!iTask.complete) {
1198 iscsi_set_events(iscsilun);
1199 qemu_coroutine_yield();
1200 }
1201
1202 if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1203 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1204 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1205 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1206 /* WRITE SAME is not supported by the target */
1207 iscsilun->has_write_same = false;
1208 scsi_free_scsi_task(iTask.task);
1209 return -ENOTSUP;
1210 }
1211
1212 if (iTask.task != NULL) {
1213 scsi_free_scsi_task(iTask.task);
1214 iTask.task = NULL;
1215 }
1216
1217 if (iTask.do_retry) {
1218 iTask.complete = 0;
1219 goto retry;
1220 }
1221
1222 if (iTask.status != SCSI_STATUS_GOOD) {
1223 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1224 count >> BDRV_SECTOR_BITS);
1225 return iTask.err_code;
1226 }
1227
1228 if (flags & BDRV_REQ_MAY_UNMAP) {
1229 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1230 count >> BDRV_SECTOR_BITS);
1231 } else {
1232 iscsi_allocmap_set_allocated(iscsilun, offset >> BDRV_SECTOR_BITS,
1233 count >> BDRV_SECTOR_BITS);
1234 }
1235
1236 return 0;
1237 }
1238
1239 static void parse_chap(struct iscsi_context *iscsi, const char *target,
1240 Error **errp)
1241 {
1242 QemuOptsList *list;
1243 QemuOpts *opts;
1244 const char *user = NULL;
1245 const char *password = NULL;
1246 const char *secretid;
1247 char *secret = NULL;
1248
1249 list = qemu_find_opts("iscsi");
1250 if (!list) {
1251 return;
1252 }
1253
1254 opts = qemu_opts_find(list, target);
1255 if (opts == NULL) {
1256 opts = QTAILQ_FIRST(&list->head);
1257 if (!opts) {
1258 return;
1259 }
1260 }
1261
1262 user = qemu_opt_get(opts, "user");
1263 if (!user) {
1264 return;
1265 }
1266
1267 secretid = qemu_opt_get(opts, "password-secret");
1268 password = qemu_opt_get(opts, "password");
1269 if (secretid && password) {
1270 error_setg(errp, "'password' and 'password-secret' properties are "
1271 "mutually exclusive");
1272 return;
1273 }
1274 if (secretid) {
1275 secret = qcrypto_secret_lookup_as_utf8(secretid, errp);
1276 if (!secret) {
1277 return;
1278 }
1279 password = secret;
1280 } else if (!password) {
1281 error_setg(errp, "CHAP username specified but no password was given");
1282 return;
1283 }
1284
1285 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1286 error_setg(errp, "Failed to set initiator username and password");
1287 }
1288
1289 g_free(secret);
1290 }
1291
1292 static void parse_header_digest(struct iscsi_context *iscsi, const char *target,
1293 Error **errp)
1294 {
1295 QemuOptsList *list;
1296 QemuOpts *opts;
1297 const char *digest = NULL;
1298
1299 list = qemu_find_opts("iscsi");
1300 if (!list) {
1301 return;
1302 }
1303
1304 opts = qemu_opts_find(list, target);
1305 if (opts == NULL) {
1306 opts = QTAILQ_FIRST(&list->head);
1307 if (!opts) {
1308 return;
1309 }
1310 }
1311
1312 digest = qemu_opt_get(opts, "header-digest");
1313 if (!digest) {
1314 return;
1315 }
1316
1317 if (!strcmp(digest, "CRC32C")) {
1318 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1319 } else if (!strcmp(digest, "NONE")) {
1320 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1321 } else if (!strcmp(digest, "CRC32C-NONE")) {
1322 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1323 } else if (!strcmp(digest, "NONE-CRC32C")) {
1324 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1325 } else {
1326 error_setg(errp, "Invalid header-digest setting : %s", digest);
1327 }
1328 }
1329
1330 static char *parse_initiator_name(const char *target)
1331 {
1332 QemuOptsList *list;
1333 QemuOpts *opts;
1334 const char *name;
1335 char *iscsi_name;
1336 UuidInfo *uuid_info;
1337
1338 list = qemu_find_opts("iscsi");
1339 if (list) {
1340 opts = qemu_opts_find(list, target);
1341 if (!opts) {
1342 opts = QTAILQ_FIRST(&list->head);
1343 }
1344 if (opts) {
1345 name = qemu_opt_get(opts, "initiator-name");
1346 if (name) {
1347 return g_strdup(name);
1348 }
1349 }
1350 }
1351
1352 uuid_info = qmp_query_uuid(NULL);
1353 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1354 name = qemu_get_vm_name();
1355 } else {
1356 name = uuid_info->UUID;
1357 }
1358 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1359 name ? ":" : "", name ? name : "");
1360 qapi_free_UuidInfo(uuid_info);
1361 return iscsi_name;
1362 }
1363
1364 static int parse_timeout(const char *target)
1365 {
1366 QemuOptsList *list;
1367 QemuOpts *opts;
1368 const char *timeout;
1369
1370 list = qemu_find_opts("iscsi");
1371 if (list) {
1372 opts = qemu_opts_find(list, target);
1373 if (!opts) {
1374 opts = QTAILQ_FIRST(&list->head);
1375 }
1376 if (opts) {
1377 timeout = qemu_opt_get(opts, "timeout");
1378 if (timeout) {
1379 return atoi(timeout);
1380 }
1381 }
1382 }
1383
1384 return 0;
1385 }
1386
1387 static void iscsi_nop_timed_event(void *opaque)
1388 {
1389 IscsiLun *iscsilun = opaque;
1390
1391 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1392 error_report("iSCSI: NOP timeout. Reconnecting...");
1393 iscsilun->request_timed_out = true;
1394 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1395 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1396 return;
1397 }
1398
1399 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1400 iscsi_set_events(iscsilun);
1401 }
1402
1403 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1404 {
1405 struct scsi_task *task = NULL;
1406 struct scsi_readcapacity10 *rc10 = NULL;
1407 struct scsi_readcapacity16 *rc16 = NULL;
1408 int retries = ISCSI_CMD_RETRIES;
1409
1410 do {
1411 if (task != NULL) {
1412 scsi_free_scsi_task(task);
1413 task = NULL;
1414 }
1415
1416 switch (iscsilun->type) {
1417 case TYPE_DISK:
1418 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1419 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1420 rc16 = scsi_datain_unmarshall(task);
1421 if (rc16 == NULL) {
1422 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1423 } else {
1424 iscsilun->block_size = rc16->block_length;
1425 iscsilun->num_blocks = rc16->returned_lba + 1;
1426 iscsilun->lbpme = !!rc16->lbpme;
1427 iscsilun->lbprz = !!rc16->lbprz;
1428 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1429 }
1430 break;
1431 }
1432 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1433 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1434 break;
1435 }
1436 /* Fall through and try READ CAPACITY(10) instead. */
1437 case TYPE_ROM:
1438 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1439 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1440 rc10 = scsi_datain_unmarshall(task);
1441 if (rc10 == NULL) {
1442 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1443 } else {
1444 iscsilun->block_size = rc10->block_size;
1445 if (rc10->lba == 0) {
1446 /* blank disk loaded */
1447 iscsilun->num_blocks = 0;
1448 } else {
1449 iscsilun->num_blocks = rc10->lba + 1;
1450 }
1451 }
1452 }
1453 break;
1454 default:
1455 return;
1456 }
1457 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1458 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1459 && retries-- > 0);
1460
1461 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1462 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1463 } else if (!iscsilun->block_size ||
1464 iscsilun->block_size % BDRV_SECTOR_SIZE) {
1465 error_setg(errp, "iSCSI: the target returned an invalid "
1466 "block size of %d.", iscsilun->block_size);
1467 }
1468 if (task) {
1469 scsi_free_scsi_task(task);
1470 }
1471 }
1472
1473 /* TODO Convert to fine grained options */
1474 static QemuOptsList runtime_opts = {
1475 .name = "iscsi",
1476 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1477 .desc = {
1478 {
1479 .name = "filename",
1480 .type = QEMU_OPT_STRING,
1481 .help = "URL to the iscsi image",
1482 },
1483 { /* end of list */ }
1484 },
1485 };
1486
1487 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1488 int evpd, int pc, void **inq, Error **errp)
1489 {
1490 int full_size;
1491 struct scsi_task *task = NULL;
1492 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1493 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1494 goto fail;
1495 }
1496 full_size = scsi_datain_getfullsize(task);
1497 if (full_size > task->datain.size) {
1498 scsi_free_scsi_task(task);
1499
1500 /* we need more data for the full list */
1501 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1502 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1503 goto fail;
1504 }
1505 }
1506
1507 *inq = scsi_datain_unmarshall(task);
1508 if (*inq == NULL) {
1509 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1510 goto fail_with_err;
1511 }
1512
1513 return task;
1514
1515 fail:
1516 error_setg(errp, "iSCSI: Inquiry command failed : %s",
1517 iscsi_get_error(iscsi));
1518 fail_with_err:
1519 if (task != NULL) {
1520 scsi_free_scsi_task(task);
1521 }
1522 return NULL;
1523 }
1524
1525 static void iscsi_detach_aio_context(BlockDriverState *bs)
1526 {
1527 IscsiLun *iscsilun = bs->opaque;
1528
1529 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1530 false, NULL, NULL, NULL, NULL);
1531 iscsilun->events = 0;
1532
1533 if (iscsilun->nop_timer) {
1534 timer_del(iscsilun->nop_timer);
1535 timer_free(iscsilun->nop_timer);
1536 iscsilun->nop_timer = NULL;
1537 }
1538 if (iscsilun->event_timer) {
1539 timer_del(iscsilun->event_timer);
1540 timer_free(iscsilun->event_timer);
1541 iscsilun->event_timer = NULL;
1542 }
1543 }
1544
1545 static void iscsi_attach_aio_context(BlockDriverState *bs,
1546 AioContext *new_context)
1547 {
1548 IscsiLun *iscsilun = bs->opaque;
1549
1550 iscsilun->aio_context = new_context;
1551 iscsi_set_events(iscsilun);
1552
1553 /* Set up a timer for sending out iSCSI NOPs */
1554 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1555 QEMU_CLOCK_REALTIME, SCALE_MS,
1556 iscsi_nop_timed_event, iscsilun);
1557 timer_mod(iscsilun->nop_timer,
1558 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1559
1560 /* Set up a timer for periodic calls to iscsi_set_events and to
1561 * scan for command timeout */
1562 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1563 QEMU_CLOCK_REALTIME, SCALE_MS,
1564 iscsi_timed_check_events, iscsilun);
1565 timer_mod(iscsilun->event_timer,
1566 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1567 }
1568
1569 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1570 {
1571 struct scsi_task *task;
1572 struct scsi_mode_sense *ms = NULL;
1573 iscsilun->write_protected = false;
1574 iscsilun->dpofua = false;
1575
1576 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1577 1, SCSI_MODESENSE_PC_CURRENT,
1578 0x3F, 0, 255);
1579 if (task == NULL) {
1580 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1581 iscsi_get_error(iscsilun->iscsi));
1582 goto out;
1583 }
1584
1585 if (task->status != SCSI_STATUS_GOOD) {
1586 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1587 goto out;
1588 }
1589 ms = scsi_datain_unmarshall(task);
1590 if (!ms) {
1591 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1592 iscsi_get_error(iscsilun->iscsi));
1593 goto out;
1594 }
1595 iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1596 iscsilun->dpofua = ms->device_specific_parameter & 0x10;
1597
1598 out:
1599 if (task) {
1600 scsi_free_scsi_task(task);
1601 }
1602 }
1603
1604 /*
1605 * We support iscsi url's on the form
1606 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1607 */
1608 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1609 Error **errp)
1610 {
1611 IscsiLun *iscsilun = bs->opaque;
1612 struct iscsi_context *iscsi = NULL;
1613 struct iscsi_url *iscsi_url = NULL;
1614 struct scsi_task *task = NULL;
1615 struct scsi_inquiry_standard *inq = NULL;
1616 struct scsi_inquiry_supported_pages *inq_vpd;
1617 char *initiator_name = NULL;
1618 QemuOpts *opts;
1619 Error *local_err = NULL;
1620 const char *filename;
1621 int i, ret = 0, timeout = 0;
1622
1623 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1624 qemu_opts_absorb_qdict(opts, options, &local_err);
1625 if (local_err) {
1626 error_propagate(errp, local_err);
1627 ret = -EINVAL;
1628 goto out;
1629 }
1630
1631 filename = qemu_opt_get(opts, "filename");
1632
1633 iscsi_url = iscsi_parse_full_url(iscsi, filename);
1634 if (iscsi_url == NULL) {
1635 error_setg(errp, "Failed to parse URL : %s", filename);
1636 ret = -EINVAL;
1637 goto out;
1638 }
1639
1640 memset(iscsilun, 0, sizeof(IscsiLun));
1641
1642 initiator_name = parse_initiator_name(iscsi_url->target);
1643
1644 iscsi = iscsi_create_context(initiator_name);
1645 if (iscsi == NULL) {
1646 error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1647 ret = -ENOMEM;
1648 goto out;
1649 }
1650 #if LIBISCSI_API_VERSION >= (20160603)
1651 if (iscsi_init_transport(iscsi, iscsi_url->transport)) {
1652 error_setg(errp, ("Error initializing transport."));
1653 ret = -EINVAL;
1654 goto out;
1655 }
1656 #endif
1657 if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1658 error_setg(errp, "iSCSI: Failed to set target name.");
1659 ret = -EINVAL;
1660 goto out;
1661 }
1662
1663 if (iscsi_url->user[0] != '\0') {
1664 ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1665 iscsi_url->passwd);
1666 if (ret != 0) {
1667 error_setg(errp, "Failed to set initiator username and password");
1668 ret = -EINVAL;
1669 goto out;
1670 }
1671 }
1672
1673 /* check if we got CHAP username/password via the options */
1674 parse_chap(iscsi, iscsi_url->target, &local_err);
1675 if (local_err != NULL) {
1676 error_propagate(errp, local_err);
1677 ret = -EINVAL;
1678 goto out;
1679 }
1680
1681 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1682 error_setg(errp, "iSCSI: Failed to set session type to normal.");
1683 ret = -EINVAL;
1684 goto out;
1685 }
1686
1687 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1688
1689 /* check if we got HEADER_DIGEST via the options */
1690 parse_header_digest(iscsi, iscsi_url->target, &local_err);
1691 if (local_err != NULL) {
1692 error_propagate(errp, local_err);
1693 ret = -EINVAL;
1694 goto out;
1695 }
1696
1697 /* timeout handling is broken in libiscsi before 1.15.0 */
1698 timeout = parse_timeout(iscsi_url->target);
1699 #if LIBISCSI_API_VERSION >= 20150621
1700 iscsi_set_timeout(iscsi, timeout);
1701 #else
1702 if (timeout) {
1703 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1704 }
1705 #endif
1706
1707 if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1708 error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1709 iscsi_get_error(iscsi));
1710 ret = -EINVAL;
1711 goto out;
1712 }
1713
1714 iscsilun->iscsi = iscsi;
1715 iscsilun->aio_context = bdrv_get_aio_context(bs);
1716 iscsilun->lun = iscsi_url->lun;
1717 iscsilun->has_write_same = true;
1718
1719 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1720 (void **) &inq, errp);
1721 if (task == NULL) {
1722 ret = -EINVAL;
1723 goto out;
1724 }
1725 iscsilun->type = inq->periperal_device_type;
1726 scsi_free_scsi_task(task);
1727 task = NULL;
1728
1729 iscsi_modesense_sync(iscsilun);
1730 if (iscsilun->dpofua) {
1731 bs->supported_write_flags = BDRV_REQ_FUA;
1732 }
1733 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
1734
1735 /* Check the write protect flag of the LUN if we want to write */
1736 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1737 iscsilun->write_protected) {
1738 error_setg(errp, "Cannot open a write protected LUN as read-write");
1739 ret = -EACCES;
1740 goto out;
1741 }
1742
1743 iscsi_readcapacity_sync(iscsilun, &local_err);
1744 if (local_err != NULL) {
1745 error_propagate(errp, local_err);
1746 ret = -EINVAL;
1747 goto out;
1748 }
1749 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1750
1751 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1752 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1753 * will try to read from the device to guess the image format.
1754 */
1755 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1756 bs->sg = true;
1757 }
1758
1759 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1760 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1761 (void **) &inq_vpd, errp);
1762 if (task == NULL) {
1763 ret = -EINVAL;
1764 goto out;
1765 }
1766 for (i = 0; i < inq_vpd->num_pages; i++) {
1767 struct scsi_task *inq_task;
1768 struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1769 struct scsi_inquiry_block_limits *inq_bl;
1770 switch (inq_vpd->pages[i]) {
1771 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1772 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1773 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1774 (void **) &inq_lbp, errp);
1775 if (inq_task == NULL) {
1776 ret = -EINVAL;
1777 goto out;
1778 }
1779 memcpy(&iscsilun->lbp, inq_lbp,
1780 sizeof(struct scsi_inquiry_logical_block_provisioning));
1781 scsi_free_scsi_task(inq_task);
1782 break;
1783 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1784 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1785 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1786 (void **) &inq_bl, errp);
1787 if (inq_task == NULL) {
1788 ret = -EINVAL;
1789 goto out;
1790 }
1791 memcpy(&iscsilun->bl, inq_bl,
1792 sizeof(struct scsi_inquiry_block_limits));
1793 scsi_free_scsi_task(inq_task);
1794 break;
1795 default:
1796 break;
1797 }
1798 }
1799 scsi_free_scsi_task(task);
1800 task = NULL;
1801
1802 iscsi_attach_aio_context(bs, iscsilun->aio_context);
1803
1804 /* Guess the internal cluster (page) size of the iscsi target by the means
1805 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1806 * reasonable size */
1807 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
1808 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1809 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran *
1810 iscsilun->block_size) >> BDRV_SECTOR_BITS;
1811 if (iscsilun->lbprz) {
1812 ret = iscsi_allocmap_init(iscsilun, bs->open_flags);
1813 }
1814 }
1815
1816 out:
1817 qemu_opts_del(opts);
1818 g_free(initiator_name);
1819 if (iscsi_url != NULL) {
1820 iscsi_destroy_url(iscsi_url);
1821 }
1822 if (task != NULL) {
1823 scsi_free_scsi_task(task);
1824 }
1825
1826 if (ret) {
1827 if (iscsi != NULL) {
1828 if (iscsi_is_logged_in(iscsi)) {
1829 iscsi_logout_sync(iscsi);
1830 }
1831 iscsi_destroy_context(iscsi);
1832 }
1833 memset(iscsilun, 0, sizeof(IscsiLun));
1834 }
1835 return ret;
1836 }
1837
1838 static void iscsi_close(BlockDriverState *bs)
1839 {
1840 IscsiLun *iscsilun = bs->opaque;
1841 struct iscsi_context *iscsi = iscsilun->iscsi;
1842
1843 iscsi_detach_aio_context(bs);
1844 if (iscsi_is_logged_in(iscsi)) {
1845 iscsi_logout_sync(iscsi);
1846 }
1847 iscsi_destroy_context(iscsi);
1848 g_free(iscsilun->zeroblock);
1849 iscsi_allocmap_free(iscsilun);
1850 memset(iscsilun, 0, sizeof(IscsiLun));
1851 }
1852
1853 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
1854 {
1855 /* We don't actually refresh here, but just return data queried in
1856 * iscsi_open(): iscsi targets don't change their limits. */
1857
1858 IscsiLun *iscsilun = bs->opaque;
1859 uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
1860 unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size);
1861
1862 assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bs->sg);
1863
1864 bs->bl.request_alignment = block_size;
1865
1866 if (iscsilun->bl.max_xfer_len) {
1867 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
1868 }
1869
1870 if (max_xfer_len * block_size < INT_MAX) {
1871 bs->bl.max_transfer = max_xfer_len * iscsilun->block_size;
1872 }
1873
1874 if (iscsilun->lbp.lbpu) {
1875 if (iscsilun->bl.max_unmap < 0xffffffff / block_size) {
1876 bs->bl.max_pdiscard =
1877 iscsilun->bl.max_unmap * iscsilun->block_size;
1878 }
1879 bs->bl.pdiscard_alignment =
1880 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
1881 } else {
1882 bs->bl.pdiscard_alignment = iscsilun->block_size;
1883 }
1884
1885 if (iscsilun->bl.max_ws_len < 0xffffffff / block_size) {
1886 bs->bl.max_pwrite_zeroes =
1887 iscsilun->bl.max_ws_len * iscsilun->block_size;
1888 }
1889 if (iscsilun->lbp.lbpws) {
1890 bs->bl.pwrite_zeroes_alignment =
1891 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
1892 } else {
1893 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size;
1894 }
1895 if (iscsilun->bl.opt_xfer_len &&
1896 iscsilun->bl.opt_xfer_len < INT_MAX / block_size) {
1897 bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len *
1898 iscsilun->block_size);
1899 }
1900 }
1901
1902 /* Note that this will not re-establish a connection with an iSCSI target - it
1903 * is effectively a NOP. */
1904 static int iscsi_reopen_prepare(BDRVReopenState *state,
1905 BlockReopenQueue *queue, Error **errp)
1906 {
1907 IscsiLun *iscsilun = state->bs->opaque;
1908
1909 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
1910 error_setg(errp, "Cannot open a write protected LUN as read-write");
1911 return -EACCES;
1912 }
1913 return 0;
1914 }
1915
1916 static void iscsi_reopen_commit(BDRVReopenState *reopen_state)
1917 {
1918 IscsiLun *iscsilun = reopen_state->bs->opaque;
1919
1920 /* the cache.direct status might have changed */
1921 if (iscsilun->allocmap != NULL) {
1922 iscsi_allocmap_init(iscsilun, reopen_state->flags);
1923 }
1924 }
1925
1926 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1927 {
1928 IscsiLun *iscsilun = bs->opaque;
1929 Error *local_err = NULL;
1930
1931 if (iscsilun->type != TYPE_DISK) {
1932 return -ENOTSUP;
1933 }
1934
1935 iscsi_readcapacity_sync(iscsilun, &local_err);
1936 if (local_err != NULL) {
1937 error_free(local_err);
1938 return -EIO;
1939 }
1940
1941 if (offset > iscsi_getlength(bs)) {
1942 return -EINVAL;
1943 }
1944
1945 if (iscsilun->allocmap != NULL) {
1946 iscsi_allocmap_init(iscsilun, bs->open_flags);
1947 }
1948
1949 return 0;
1950 }
1951
1952 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp)
1953 {
1954 int ret = 0;
1955 int64_t total_size = 0;
1956 BlockDriverState *bs;
1957 IscsiLun *iscsilun = NULL;
1958 QDict *bs_options;
1959
1960 bs = bdrv_new();
1961
1962 /* Read out options */
1963 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1964 BDRV_SECTOR_SIZE);
1965 bs->opaque = g_new0(struct IscsiLun, 1);
1966 iscsilun = bs->opaque;
1967
1968 bs_options = qdict_new();
1969 qdict_put(bs_options, "filename", qstring_from_str(filename));
1970 ret = iscsi_open(bs, bs_options, 0, NULL);
1971 QDECREF(bs_options);
1972
1973 if (ret != 0) {
1974 goto out;
1975 }
1976 iscsi_detach_aio_context(bs);
1977 if (iscsilun->type != TYPE_DISK) {
1978 ret = -ENODEV;
1979 goto out;
1980 }
1981 if (bs->total_sectors < total_size) {
1982 ret = -ENOSPC;
1983 goto out;
1984 }
1985
1986 ret = 0;
1987 out:
1988 if (iscsilun->iscsi != NULL) {
1989 iscsi_destroy_context(iscsilun->iscsi);
1990 }
1991 g_free(bs->opaque);
1992 bs->opaque = NULL;
1993 bdrv_unref(bs);
1994 return ret;
1995 }
1996
1997 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1998 {
1999 IscsiLun *iscsilun = bs->opaque;
2000 bdi->unallocated_blocks_are_zero = iscsilun->lbprz;
2001 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
2002 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE;
2003 return 0;
2004 }
2005
2006 static void iscsi_invalidate_cache(BlockDriverState *bs,
2007 Error **errp)
2008 {
2009 IscsiLun *iscsilun = bs->opaque;
2010 iscsi_allocmap_invalidate(iscsilun);
2011 }
2012
2013 static QemuOptsList iscsi_create_opts = {
2014 .name = "iscsi-create-opts",
2015 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
2016 .desc = {
2017 {
2018 .name = BLOCK_OPT_SIZE,
2019 .type = QEMU_OPT_SIZE,
2020 .help = "Virtual disk size"
2021 },
2022 { /* end of list */ }
2023 }
2024 };
2025
2026 static BlockDriver bdrv_iscsi = {
2027 .format_name = "iscsi",
2028 .protocol_name = "iscsi",
2029
2030 .instance_size = sizeof(IscsiLun),
2031 .bdrv_needs_filename = true,
2032 .bdrv_file_open = iscsi_open,
2033 .bdrv_close = iscsi_close,
2034 .bdrv_create = iscsi_create,
2035 .create_opts = &iscsi_create_opts,
2036 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2037 .bdrv_reopen_commit = iscsi_reopen_commit,
2038 .bdrv_invalidate_cache = iscsi_invalidate_cache,
2039
2040 .bdrv_getlength = iscsi_getlength,
2041 .bdrv_get_info = iscsi_get_info,
2042 .bdrv_truncate = iscsi_truncate,
2043 .bdrv_refresh_limits = iscsi_refresh_limits,
2044
2045 .bdrv_co_get_block_status = iscsi_co_get_block_status,
2046 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2047 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2048 .bdrv_co_readv = iscsi_co_readv,
2049 .bdrv_co_writev_flags = iscsi_co_writev_flags,
2050 .bdrv_co_flush_to_disk = iscsi_co_flush,
2051
2052 #ifdef __linux__
2053 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2054 #endif
2055
2056 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2057 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2058 };
2059
2060 #if LIBISCSI_API_VERSION >= (20160603)
2061 static BlockDriver bdrv_iser = {
2062 .format_name = "iser",
2063 .protocol_name = "iser",
2064
2065 .instance_size = sizeof(IscsiLun),
2066 .bdrv_needs_filename = true,
2067 .bdrv_file_open = iscsi_open,
2068 .bdrv_close = iscsi_close,
2069 .bdrv_create = iscsi_create,
2070 .create_opts = &iscsi_create_opts,
2071 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2072 .bdrv_reopen_commit = iscsi_reopen_commit,
2073 .bdrv_invalidate_cache = iscsi_invalidate_cache,
2074
2075 .bdrv_getlength = iscsi_getlength,
2076 .bdrv_get_info = iscsi_get_info,
2077 .bdrv_truncate = iscsi_truncate,
2078 .bdrv_refresh_limits = iscsi_refresh_limits,
2079
2080 .bdrv_co_get_block_status = iscsi_co_get_block_status,
2081 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2082 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2083 .bdrv_co_readv = iscsi_co_readv,
2084 .bdrv_co_writev_flags = iscsi_co_writev_flags,
2085 .bdrv_co_flush_to_disk = iscsi_co_flush,
2086
2087 #ifdef __linux__
2088 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2089 #endif
2090
2091 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2092 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2093 };
2094 #endif
2095
2096 static void iscsi_block_init(void)
2097 {
2098 bdrv_register(&bdrv_iscsi);
2099 #if LIBISCSI_API_VERSION >= (20160603)
2100 bdrv_register(&bdrv_iser);
2101 #endif
2102 }
2103
2104 block_init(iscsi_block_init);
QEmu