Merge remote-tracking branch 'remotes/armbru/tags/pull-qapi-2017-03-16' into staging
[qemu.git] / block / iscsi.c
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 QemuMutex mutex;
62 struct scsi_inquiry_logical_block_provisioning lbp;
63 struct scsi_inquiry_block_limits bl;
64 unsigned char *zeroblock;
65 /* The allocmap tracks which clusters (pages) on the iSCSI target are
66 * allocated and which are not. In case a target returns zeros for
67 * unallocated pages (iscsilun->lprz) we can directly return zeros instead
68 * of reading zeros over the wire if a read request falls within an
69 * unallocated block. As there are 3 possible states we need 2 bitmaps to
70 * track. allocmap_valid keeps track if QEMU's information about a page is
71 * valid. allocmap tracks if a page is allocated or not. In case QEMU has no
72 * valid information about a page the corresponding allocmap entry should be
73 * switched to unallocated as well to force a new lookup of the allocation
74 * status as lookups are generally skipped if a page is suspect to be
75 * allocated. If a iSCSI target is opened with cache.direct = on the
76 * allocmap_valid does not exist turning all cached information invalid so
77 * that a fresh lookup is made for any page even if allocmap entry returns
78 * it's unallocated. */
79 unsigned long *allocmap;
80 unsigned long *allocmap_valid;
81 long allocmap_size;
82 int cluster_sectors;
83 bool use_16_for_rw;
84 bool write_protected;
85 bool lbpme;
86 bool lbprz;
87 bool dpofua;
88 bool has_write_same;
89 bool request_timed_out;
90 } IscsiLun;
91
92 typedef struct IscsiTask {
93 int status;
94 int complete;
95 int retries;
96 int do_retry;
97 struct scsi_task *task;
98 Coroutine *co;
99 IscsiLun *iscsilun;
100 QEMUTimer retry_timer;
101 int err_code;
102 } IscsiTask;
103
104 typedef struct IscsiAIOCB {
105 BlockAIOCB common;
106 QEMUIOVector *qiov;
107 QEMUBH *bh;
108 IscsiLun *iscsilun;
109 struct scsi_task *task;
110 uint8_t *buf;
111 int status;
112 int64_t sector_num;
113 int nb_sectors;
114 int ret;
115 #ifdef __linux__
116 sg_io_hdr_t *ioh;
117 #endif
118 } IscsiAIOCB;
119
120 /* libiscsi uses time_t so its enough to process events every second */
121 #define EVENT_INTERVAL 1000
122 #define NOP_INTERVAL 5000
123 #define MAX_NOP_FAILURES 3
124 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times)
125 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768};
126
127 /* this threshold is a trade-off knob to choose between
128 * the potential additional overhead of an extra GET_LBA_STATUS request
129 * vs. unnecessarily reading a lot of zero sectors over the wire.
130 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
131 * sectors we check the allocation status of the area covered by the
132 * request first if the allocationmap indicates that the area might be
133 * unallocated. */
134 #define ISCSI_CHECKALLOC_THRES 64
135
136 static void
137 iscsi_bh_cb(void *p)
138 {
139 IscsiAIOCB *acb = p;
140
141 qemu_bh_delete(acb->bh);
142
143 g_free(acb->buf);
144 acb->buf = NULL;
145
146 acb->common.cb(acb->common.opaque, acb->status);
147
148 if (acb->task != NULL) {
149 scsi_free_scsi_task(acb->task);
150 acb->task = NULL;
151 }
152
153 qemu_aio_unref(acb);
154 }
155
156 static void
157 iscsi_schedule_bh(IscsiAIOCB *acb)
158 {
159 if (acb->bh) {
160 return;
161 }
162 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb);
163 qemu_bh_schedule(acb->bh);
164 }
165
166 static void iscsi_co_generic_bh_cb(void *opaque)
167 {
168 struct IscsiTask *iTask = opaque;
169
170 iTask->complete = 1;
171 aio_co_wake(iTask->co);
172 }
173
174 static void iscsi_retry_timer_expired(void *opaque)
175 {
176 struct IscsiTask *iTask = opaque;
177 iTask->complete = 1;
178 if (iTask->co) {
179 aio_co_wake(iTask->co);
180 }
181 }
182
183 static inline unsigned exp_random(double mean)
184 {
185 return -mean * log((double)rand() / RAND_MAX);
186 }
187
188 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in
189 * libiscsi 1.10.0, together with other constants we need. Use it as
190 * a hint that we have to define them ourselves if needed, to keep the
191 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for
192 * the test because SCSI_STATUS_* is an enum.
193 *
194 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes
195 * an enum, check against the LIBISCSI_API_VERSION macro, which was
196 * introduced in 1.11.0. If it is present, there is no need to define
197 * anything.
198 */
199 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \
200 !defined(LIBISCSI_API_VERSION)
201 #define SCSI_STATUS_TASK_SET_FULL 0x28
202 #define SCSI_STATUS_TIMEOUT 0x0f000002
203 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600
204 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00
205 #endif
206
207 #ifndef LIBISCSI_API_VERSION
208 #define LIBISCSI_API_VERSION 20130701
209 #endif
210
211 static int iscsi_translate_sense(struct scsi_sense *sense)
212 {
213 int ret;
214
215 switch (sense->key) {
216 case SCSI_SENSE_NOT_READY:
217 return -EBUSY;
218 case SCSI_SENSE_DATA_PROTECTION:
219 return -EACCES;
220 case SCSI_SENSE_COMMAND_ABORTED:
221 return -ECANCELED;
222 case SCSI_SENSE_ILLEGAL_REQUEST:
223 /* Parse ASCQ */
224 break;
225 default:
226 return -EIO;
227 }
228 switch (sense->ascq) {
229 case SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR:
230 case SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE:
231 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB:
232 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST:
233 ret = -EINVAL;
234 break;
235 case SCSI_SENSE_ASCQ_LBA_OUT_OF_RANGE:
236 ret = -ENOSPC;
237 break;
238 case SCSI_SENSE_ASCQ_LOGICAL_UNIT_NOT_SUPPORTED:
239 ret = -ENOTSUP;
240 break;
241 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT:
242 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_CLOSED:
243 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_OPEN:
244 ret = -ENOMEDIUM;
245 break;
246 case SCSI_SENSE_ASCQ_WRITE_PROTECTED:
247 ret = -EACCES;
248 break;
249 default:
250 ret = -EIO;
251 break;
252 }
253 return ret;
254 }
255
256 /* Called (via iscsi_service) with QemuMutex held. */
257 static void
258 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
259 void *command_data, void *opaque)
260 {
261 struct IscsiTask *iTask = opaque;
262 struct scsi_task *task = command_data;
263
264 iTask->status = status;
265 iTask->do_retry = 0;
266 iTask->task = task;
267
268 if (status != SCSI_STATUS_GOOD) {
269 if (iTask->retries++ < ISCSI_CMD_RETRIES) {
270 if (status == SCSI_STATUS_CHECK_CONDITION
271 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
272 error_report("iSCSI CheckCondition: %s",
273 iscsi_get_error(iscsi));
274 iTask->do_retry = 1;
275 goto out;
276 }
277 if (status == SCSI_STATUS_BUSY ||
278 status == SCSI_STATUS_TIMEOUT ||
279 status == SCSI_STATUS_TASK_SET_FULL) {
280 unsigned retry_time =
281 exp_random(iscsi_retry_times[iTask->retries - 1]);
282 if (status == SCSI_STATUS_TIMEOUT) {
283 /* make sure the request is rescheduled AFTER the
284 * reconnect is initiated */
285 retry_time = EVENT_INTERVAL * 2;
286 iTask->iscsilun->request_timed_out = true;
287 }
288 error_report("iSCSI Busy/TaskSetFull/TimeOut"
289 " (retry #%u in %u ms): %s",
290 iTask->retries, retry_time,
291 iscsi_get_error(iscsi));
292 aio_timer_init(iTask->iscsilun->aio_context,
293 &iTask->retry_timer, QEMU_CLOCK_REALTIME,
294 SCALE_MS, iscsi_retry_timer_expired, iTask);
295 timer_mod(&iTask->retry_timer,
296 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
297 iTask->do_retry = 1;
298 return;
299 }
300 }
301 iTask->err_code = iscsi_translate_sense(&task->sense);
302 error_report("iSCSI Failure: %s", iscsi_get_error(iscsi));
303 }
304
305 out:
306 if (iTask->co) {
307 aio_bh_schedule_oneshot(iTask->iscsilun->aio_context,
308 iscsi_co_generic_bh_cb, iTask);
309 } else {
310 iTask->complete = 1;
311 }
312 }
313
314 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
315 {
316 *iTask = (struct IscsiTask) {
317 .co = qemu_coroutine_self(),
318 .iscsilun = iscsilun,
319 };
320 }
321
322 static void
323 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
324 void *private_data)
325 {
326 IscsiAIOCB *acb = private_data;
327
328 acb->status = -ECANCELED;
329 iscsi_schedule_bh(acb);
330 }
331
332 static void
333 iscsi_aio_cancel(BlockAIOCB *blockacb)
334 {
335 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
336 IscsiLun *iscsilun = acb->iscsilun;
337
338 if (acb->status != -EINPROGRESS) {
339 return;
340 }
341
342 /* send a task mgmt call to the target to cancel the task on the target */
343 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
344 iscsi_abort_task_cb, acb);
345
346 }
347
348 static const AIOCBInfo iscsi_aiocb_info = {
349 .aiocb_size = sizeof(IscsiAIOCB),
350 .cancel_async = iscsi_aio_cancel,
351 };
352
353
354 static void iscsi_process_read(void *arg);
355 static void iscsi_process_write(void *arg);
356
357 /* Called with QemuMutex held. */
358 static void
359 iscsi_set_events(IscsiLun *iscsilun)
360 {
361 struct iscsi_context *iscsi = iscsilun->iscsi;
362 int ev = iscsi_which_events(iscsi);
363
364 if (ev != iscsilun->events) {
365 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi),
366 false,
367 (ev & POLLIN) ? iscsi_process_read : NULL,
368 (ev & POLLOUT) ? iscsi_process_write : NULL,
369 NULL,
370 iscsilun);
371 iscsilun->events = ev;
372 }
373 }
374
375 static void iscsi_timed_check_events(void *opaque)
376 {
377 IscsiLun *iscsilun = opaque;
378
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 /* newer versions of libiscsi may return zero events. Ensure we are able
388 * to return to service once this situation changes. */
389 iscsi_set_events(iscsilun);
390
391 timer_mod(iscsilun->event_timer,
392 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
393 }
394
395 static void
396 iscsi_process_read(void *arg)
397 {
398 IscsiLun *iscsilun = arg;
399 struct iscsi_context *iscsi = iscsilun->iscsi;
400
401 qemu_mutex_lock(&iscsilun->mutex);
402 iscsi_service(iscsi, POLLIN);
403 iscsi_set_events(iscsilun);
404 qemu_mutex_unlock(&iscsilun->mutex);
405 }
406
407 static void
408 iscsi_process_write(void *arg)
409 {
410 IscsiLun *iscsilun = arg;
411 struct iscsi_context *iscsi = iscsilun->iscsi;
412
413 qemu_mutex_lock(&iscsilun->mutex);
414 iscsi_service(iscsi, POLLOUT);
415 iscsi_set_events(iscsilun);
416 qemu_mutex_unlock(&iscsilun->mutex);
417 }
418
419 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
420 {
421 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
422 }
423
424 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
425 {
426 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
427 }
428
429 static bool is_byte_request_lun_aligned(int64_t offset, int count,
430 IscsiLun *iscsilun)
431 {
432 if (offset % iscsilun->block_size || count % iscsilun->block_size) {
433 error_report("iSCSI misaligned request: "
434 "iscsilun->block_size %u, offset %" PRIi64
435 ", count %d",
436 iscsilun->block_size, offset, count);
437 return false;
438 }
439 return true;
440 }
441
442 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors,
443 IscsiLun *iscsilun)
444 {
445 assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS);
446 return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS,
447 nb_sectors << BDRV_SECTOR_BITS,
448 iscsilun);
449 }
450
451 static void iscsi_allocmap_free(IscsiLun *iscsilun)
452 {
453 g_free(iscsilun->allocmap);
454 g_free(iscsilun->allocmap_valid);
455 iscsilun->allocmap = NULL;
456 iscsilun->allocmap_valid = NULL;
457 }
458
459
460 static int iscsi_allocmap_init(IscsiLun *iscsilun, int open_flags)
461 {
462 iscsi_allocmap_free(iscsilun);
463
464 iscsilun->allocmap_size =
465 DIV_ROUND_UP(sector_lun2qemu(iscsilun->num_blocks, iscsilun),
466 iscsilun->cluster_sectors);
467
468 iscsilun->allocmap = bitmap_try_new(iscsilun->allocmap_size);
469 if (!iscsilun->allocmap) {
470 return -ENOMEM;
471 }
472
473 if (open_flags & BDRV_O_NOCACHE) {
474 /* in case that cache.direct = on all allocmap entries are
475 * treated as invalid to force a relookup of the block
476 * status on every read request */
477 return 0;
478 }
479
480 iscsilun->allocmap_valid = bitmap_try_new(iscsilun->allocmap_size);
481 if (!iscsilun->allocmap_valid) {
482 /* if we are under memory pressure free the allocmap as well */
483 iscsi_allocmap_free(iscsilun);
484 return -ENOMEM;
485 }
486
487 return 0;
488 }
489
490 static void
491 iscsi_allocmap_update(IscsiLun *iscsilun, int64_t sector_num,
492 int nb_sectors, bool allocated, bool valid)
493 {
494 int64_t cl_num_expanded, nb_cls_expanded, cl_num_shrunk, nb_cls_shrunk;
495
496 if (iscsilun->allocmap == NULL) {
497 return;
498 }
499 /* expand to entirely contain all affected clusters */
500 cl_num_expanded = sector_num / iscsilun->cluster_sectors;
501 nb_cls_expanded = DIV_ROUND_UP(sector_num + nb_sectors,
502 iscsilun->cluster_sectors) - cl_num_expanded;
503 /* shrink to touch only completely contained clusters */
504 cl_num_shrunk = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors);
505 nb_cls_shrunk = (sector_num + nb_sectors) / iscsilun->cluster_sectors
506 - cl_num_shrunk;
507 if (allocated) {
508 bitmap_set(iscsilun->allocmap, cl_num_expanded, nb_cls_expanded);
509 } else {
510 if (nb_cls_shrunk > 0) {
511 bitmap_clear(iscsilun->allocmap, cl_num_shrunk, nb_cls_shrunk);
512 }
513 }
514
515 if (iscsilun->allocmap_valid == NULL) {
516 return;
517 }
518 if (valid) {
519 if (nb_cls_shrunk > 0) {
520 bitmap_set(iscsilun->allocmap_valid, cl_num_shrunk, nb_cls_shrunk);
521 }
522 } else {
523 bitmap_clear(iscsilun->allocmap_valid, cl_num_expanded,
524 nb_cls_expanded);
525 }
526 }
527
528 static void
529 iscsi_allocmap_set_allocated(IscsiLun *iscsilun, int64_t sector_num,
530 int nb_sectors)
531 {
532 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, true, true);
533 }
534
535 static void
536 iscsi_allocmap_set_unallocated(IscsiLun *iscsilun, int64_t sector_num,
537 int nb_sectors)
538 {
539 /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update
540 * is ignored, so this will in effect be an iscsi_allocmap_set_invalid.
541 */
542 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, false, true);
543 }
544
545 static void iscsi_allocmap_set_invalid(IscsiLun *iscsilun, int64_t sector_num,
546 int nb_sectors)
547 {
548 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, false, false);
549 }
550
551 static void iscsi_allocmap_invalidate(IscsiLun *iscsilun)
552 {
553 if (iscsilun->allocmap) {
554 bitmap_zero(iscsilun->allocmap, iscsilun->allocmap_size);
555 }
556 if (iscsilun->allocmap_valid) {
557 bitmap_zero(iscsilun->allocmap_valid, iscsilun->allocmap_size);
558 }
559 }
560
561 static inline bool
562 iscsi_allocmap_is_allocated(IscsiLun *iscsilun, int64_t sector_num,
563 int nb_sectors)
564 {
565 unsigned long size;
566 if (iscsilun->allocmap == NULL) {
567 return true;
568 }
569 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
570 return !(find_next_bit(iscsilun->allocmap, size,
571 sector_num / iscsilun->cluster_sectors) == size);
572 }
573
574 static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun,
575 int64_t sector_num, int nb_sectors)
576 {
577 unsigned long size;
578 if (iscsilun->allocmap_valid == NULL) {
579 return false;
580 }
581 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
582 return (find_next_zero_bit(iscsilun->allocmap_valid, size,
583 sector_num / iscsilun->cluster_sectors) == size);
584 }
585
586 static int coroutine_fn
587 iscsi_co_writev_flags(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
588 QEMUIOVector *iov, int flags)
589 {
590 IscsiLun *iscsilun = bs->opaque;
591 struct IscsiTask iTask;
592 uint64_t lba;
593 uint32_t num_sectors;
594 bool fua = flags & BDRV_REQ_FUA;
595 int r = 0;
596
597 if (fua) {
598 assert(iscsilun->dpofua);
599 }
600 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
601 return -EINVAL;
602 }
603
604 if (bs->bl.max_transfer) {
605 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
606 }
607
608 lba = sector_qemu2lun(sector_num, iscsilun);
609 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
610 iscsi_co_init_iscsitask(iscsilun, &iTask);
611 qemu_mutex_lock(&iscsilun->mutex);
612 retry:
613 if (iscsilun->use_16_for_rw) {
614 #if LIBISCSI_API_VERSION >= (20160603)
615 iTask.task = iscsi_write16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
616 NULL, num_sectors * iscsilun->block_size,
617 iscsilun->block_size, 0, 0, fua, 0, 0,
618 iscsi_co_generic_cb, &iTask,
619 (struct scsi_iovec *)iov->iov, iov->niov);
620 } else {
621 iTask.task = iscsi_write10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
622 NULL, num_sectors * iscsilun->block_size,
623 iscsilun->block_size, 0, 0, fua, 0, 0,
624 iscsi_co_generic_cb, &iTask,
625 (struct scsi_iovec *)iov->iov, iov->niov);
626 }
627 #else
628 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
629 NULL, num_sectors * iscsilun->block_size,
630 iscsilun->block_size, 0, 0, fua, 0, 0,
631 iscsi_co_generic_cb, &iTask);
632 } else {
633 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
634 NULL, num_sectors * iscsilun->block_size,
635 iscsilun->block_size, 0, 0, fua, 0, 0,
636 iscsi_co_generic_cb, &iTask);
637 }
638 #endif
639 if (iTask.task == NULL) {
640 qemu_mutex_unlock(&iscsilun->mutex);
641 return -ENOMEM;
642 }
643 #if LIBISCSI_API_VERSION < (20160603)
644 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
645 iov->niov);
646 #endif
647 while (!iTask.complete) {
648 iscsi_set_events(iscsilun);
649 qemu_mutex_unlock(&iscsilun->mutex);
650 qemu_coroutine_yield();
651 qemu_mutex_lock(&iscsilun->mutex);
652 }
653
654 if (iTask.task != NULL) {
655 scsi_free_scsi_task(iTask.task);
656 iTask.task = NULL;
657 }
658
659 if (iTask.do_retry) {
660 iTask.complete = 0;
661 goto retry;
662 }
663
664 if (iTask.status != SCSI_STATUS_GOOD) {
665 iscsi_allocmap_set_invalid(iscsilun, sector_num, nb_sectors);
666 r = iTask.err_code;
667 goto out_unlock;
668 }
669
670 iscsi_allocmap_set_allocated(iscsilun, sector_num, nb_sectors);
671
672 out_unlock:
673 qemu_mutex_unlock(&iscsilun->mutex);
674 return r;
675 }
676
677
678
679 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
680 int64_t sector_num,
681 int nb_sectors, int *pnum,
682 BlockDriverState **file)
683 {
684 IscsiLun *iscsilun = bs->opaque;
685 struct scsi_get_lba_status *lbas = NULL;
686 struct scsi_lba_status_descriptor *lbasd = NULL;
687 struct IscsiTask iTask;
688 int64_t ret;
689
690 iscsi_co_init_iscsitask(iscsilun, &iTask);
691
692 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
693 ret = -EINVAL;
694 goto out;
695 }
696
697 /* default to all sectors allocated */
698 ret = BDRV_BLOCK_DATA;
699 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
700 *pnum = nb_sectors;
701
702 /* LUN does not support logical block provisioning */
703 if (!iscsilun->lbpme) {
704 goto out;
705 }
706
707 qemu_mutex_lock(&iscsilun->mutex);
708 retry:
709 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
710 sector_qemu2lun(sector_num, iscsilun),
711 8 + 16, iscsi_co_generic_cb,
712 &iTask) == NULL) {
713 ret = -ENOMEM;
714 goto out_unlock;
715 }
716
717 while (!iTask.complete) {
718 iscsi_set_events(iscsilun);
719 qemu_mutex_unlock(&iscsilun->mutex);
720 qemu_coroutine_yield();
721 qemu_mutex_lock(&iscsilun->mutex);
722 }
723
724 if (iTask.do_retry) {
725 if (iTask.task != NULL) {
726 scsi_free_scsi_task(iTask.task);
727 iTask.task = NULL;
728 }
729 iTask.complete = 0;
730 goto retry;
731 }
732
733 if (iTask.status != SCSI_STATUS_GOOD) {
734 /* in case the get_lba_status_callout fails (i.e.
735 * because the device is busy or the cmd is not
736 * supported) we pretend all blocks are allocated
737 * for backwards compatibility */
738 goto out_unlock;
739 }
740
741 lbas = scsi_datain_unmarshall(iTask.task);
742 if (lbas == NULL) {
743 ret = -EIO;
744 goto out_unlock;
745 }
746
747 lbasd = &lbas->descriptors[0];
748
749 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
750 ret = -EIO;
751 goto out_unlock;
752 }
753
754 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
755
756 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
757 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
758 ret &= ~BDRV_BLOCK_DATA;
759 if (iscsilun->lbprz) {
760 ret |= BDRV_BLOCK_ZERO;
761 }
762 }
763
764 if (ret & BDRV_BLOCK_ZERO) {
765 iscsi_allocmap_set_unallocated(iscsilun, sector_num, *pnum);
766 } else {
767 iscsi_allocmap_set_allocated(iscsilun, sector_num, *pnum);
768 }
769
770 if (*pnum > nb_sectors) {
771 *pnum = nb_sectors;
772 }
773 out_unlock:
774 qemu_mutex_unlock(&iscsilun->mutex);
775 out:
776 if (iTask.task != NULL) {
777 scsi_free_scsi_task(iTask.task);
778 }
779 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) {
780 *file = bs;
781 }
782 return ret;
783 }
784
785 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
786 int64_t sector_num, int nb_sectors,
787 QEMUIOVector *iov)
788 {
789 IscsiLun *iscsilun = bs->opaque;
790 struct IscsiTask iTask;
791 uint64_t lba;
792 uint32_t num_sectors;
793
794 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
795 return -EINVAL;
796 }
797
798 if (bs->bl.max_transfer) {
799 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
800 }
801
802 /* if cache.direct is off and we have a valid entry in our allocation map
803 * we can skip checking the block status and directly return zeroes if
804 * the request falls within an unallocated area */
805 if (iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) &&
806 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
807 qemu_iovec_memset(iov, 0, 0x00, iov->size);
808 return 0;
809 }
810
811 if (nb_sectors >= ISCSI_CHECKALLOC_THRES &&
812 !iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) &&
813 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
814 int pnum;
815 BlockDriverState *file;
816 /* check the block status from the beginning of the cluster
817 * containing the start sector */
818 int64_t ret = iscsi_co_get_block_status(bs,
819 sector_num - sector_num % iscsilun->cluster_sectors,
820 BDRV_REQUEST_MAX_SECTORS, &pnum, &file);
821 if (ret < 0) {
822 return ret;
823 }
824 /* if the whole request falls into an unallocated area we can avoid
825 * to read and directly return zeroes instead */
826 if (ret & BDRV_BLOCK_ZERO &&
827 pnum >= nb_sectors + sector_num % iscsilun->cluster_sectors) {
828 qemu_iovec_memset(iov, 0, 0x00, iov->size);
829 return 0;
830 }
831 }
832
833 lba = sector_qemu2lun(sector_num, iscsilun);
834 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
835
836 iscsi_co_init_iscsitask(iscsilun, &iTask);
837 qemu_mutex_lock(&iscsilun->mutex);
838 retry:
839 if (iscsilun->use_16_for_rw) {
840 #if LIBISCSI_API_VERSION >= (20160603)
841 iTask.task = iscsi_read16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
842 num_sectors * iscsilun->block_size,
843 iscsilun->block_size, 0, 0, 0, 0, 0,
844 iscsi_co_generic_cb, &iTask,
845 (struct scsi_iovec *)iov->iov, iov->niov);
846 } else {
847 iTask.task = iscsi_read10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
848 num_sectors * iscsilun->block_size,
849 iscsilun->block_size,
850 0, 0, 0, 0, 0,
851 iscsi_co_generic_cb, &iTask,
852 (struct scsi_iovec *)iov->iov, iov->niov);
853 }
854 #else
855 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
856 num_sectors * iscsilun->block_size,
857 iscsilun->block_size, 0, 0, 0, 0, 0,
858 iscsi_co_generic_cb, &iTask);
859 } else {
860 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
861 num_sectors * iscsilun->block_size,
862 iscsilun->block_size,
863 0, 0, 0, 0, 0,
864 iscsi_co_generic_cb, &iTask);
865 }
866 #endif
867 if (iTask.task == NULL) {
868 qemu_mutex_unlock(&iscsilun->mutex);
869 return -ENOMEM;
870 }
871 #if LIBISCSI_API_VERSION < (20160603)
872 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
873 #endif
874 while (!iTask.complete) {
875 iscsi_set_events(iscsilun);
876 qemu_mutex_unlock(&iscsilun->mutex);
877 qemu_coroutine_yield();
878 qemu_mutex_lock(&iscsilun->mutex);
879 }
880
881 if (iTask.task != NULL) {
882 scsi_free_scsi_task(iTask.task);
883 iTask.task = NULL;
884 }
885
886 if (iTask.do_retry) {
887 iTask.complete = 0;
888 goto retry;
889 }
890 qemu_mutex_unlock(&iscsilun->mutex);
891
892 if (iTask.status != SCSI_STATUS_GOOD) {
893 return iTask.err_code;
894 }
895
896 return 0;
897 }
898
899 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
900 {
901 IscsiLun *iscsilun = bs->opaque;
902 struct IscsiTask iTask;
903
904 iscsi_co_init_iscsitask(iscsilun, &iTask);
905 qemu_mutex_lock(&iscsilun->mutex);
906 retry:
907 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
908 0, iscsi_co_generic_cb, &iTask) == NULL) {
909 qemu_mutex_unlock(&iscsilun->mutex);
910 return -ENOMEM;
911 }
912
913 while (!iTask.complete) {
914 iscsi_set_events(iscsilun);
915 qemu_mutex_unlock(&iscsilun->mutex);
916 qemu_coroutine_yield();
917 qemu_mutex_lock(&iscsilun->mutex);
918 }
919
920 if (iTask.task != NULL) {
921 scsi_free_scsi_task(iTask.task);
922 iTask.task = NULL;
923 }
924
925 if (iTask.do_retry) {
926 iTask.complete = 0;
927 goto retry;
928 }
929 qemu_mutex_unlock(&iscsilun->mutex);
930
931 if (iTask.status != SCSI_STATUS_GOOD) {
932 return iTask.err_code;
933 }
934
935 return 0;
936 }
937
938 #ifdef __linux__
939 /* Called (via iscsi_service) with QemuMutex held. */
940 static void
941 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
942 void *command_data, void *opaque)
943 {
944 IscsiAIOCB *acb = opaque;
945
946 g_free(acb->buf);
947 acb->buf = NULL;
948
949 acb->status = 0;
950 if (status < 0) {
951 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
952 iscsi_get_error(iscsi));
953 acb->status = iscsi_translate_sense(&acb->task->sense);
954 }
955
956 acb->ioh->driver_status = 0;
957 acb->ioh->host_status = 0;
958 acb->ioh->resid = 0;
959 acb->ioh->status = status;
960
961 #define SG_ERR_DRIVER_SENSE 0x08
962
963 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
964 int ss;
965
966 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
967
968 acb->ioh->sb_len_wr = acb->task->datain.size - 2;
969 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
970 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
971 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
972 }
973
974 iscsi_schedule_bh(acb);
975 }
976
977 static void iscsi_ioctl_bh_completion(void *opaque)
978 {
979 IscsiAIOCB *acb = opaque;
980
981 qemu_bh_delete(acb->bh);
982 acb->common.cb(acb->common.opaque, acb->ret);
983 qemu_aio_unref(acb);
984 }
985
986 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
987 {
988 BlockDriverState *bs = acb->common.bs;
989 IscsiLun *iscsilun = bs->opaque;
990 int ret = 0;
991
992 switch (req) {
993 case SG_GET_VERSION_NUM:
994 *(int *)buf = 30000;
995 break;
996 case SG_GET_SCSI_ID:
997 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
998 break;
999 default:
1000 ret = -EINVAL;
1001 }
1002 assert(!acb->bh);
1003 acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
1004 iscsi_ioctl_bh_completion, acb);
1005 acb->ret = ret;
1006 qemu_bh_schedule(acb->bh);
1007 }
1008
1009 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
1010 unsigned long int req, void *buf,
1011 BlockCompletionFunc *cb, void *opaque)
1012 {
1013 IscsiLun *iscsilun = bs->opaque;
1014 struct iscsi_context *iscsi = iscsilun->iscsi;
1015 struct iscsi_data data;
1016 IscsiAIOCB *acb;
1017
1018 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
1019
1020 acb->iscsilun = iscsilun;
1021 acb->bh = NULL;
1022 acb->status = -EINPROGRESS;
1023 acb->buf = NULL;
1024 acb->ioh = buf;
1025
1026 if (req != SG_IO) {
1027 iscsi_ioctl_handle_emulated(acb, req, buf);
1028 return &acb->common;
1029 }
1030
1031 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) {
1032 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)",
1033 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE);
1034 qemu_aio_unref(acb);
1035 return NULL;
1036 }
1037
1038 acb->task = malloc(sizeof(struct scsi_task));
1039 if (acb->task == NULL) {
1040 error_report("iSCSI: Failed to allocate task for scsi command. %s",
1041 iscsi_get_error(iscsi));
1042 qemu_aio_unref(acb);
1043 return NULL;
1044 }
1045 memset(acb->task, 0, sizeof(struct scsi_task));
1046
1047 switch (acb->ioh->dxfer_direction) {
1048 case SG_DXFER_TO_DEV:
1049 acb->task->xfer_dir = SCSI_XFER_WRITE;
1050 break;
1051 case SG_DXFER_FROM_DEV:
1052 acb->task->xfer_dir = SCSI_XFER_READ;
1053 break;
1054 default:
1055 acb->task->xfer_dir = SCSI_XFER_NONE;
1056 break;
1057 }
1058
1059 acb->task->cdb_size = acb->ioh->cmd_len;
1060 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
1061 acb->task->expxferlen = acb->ioh->dxfer_len;
1062
1063 data.size = 0;
1064 qemu_mutex_lock(&iscsilun->mutex);
1065 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
1066 if (acb->ioh->iovec_count == 0) {
1067 data.data = acb->ioh->dxferp;
1068 data.size = acb->ioh->dxfer_len;
1069 } else {
1070 scsi_task_set_iov_out(acb->task,
1071 (struct scsi_iovec *) acb->ioh->dxferp,
1072 acb->ioh->iovec_count);
1073 }
1074 }
1075
1076 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
1077 iscsi_aio_ioctl_cb,
1078 (data.size > 0) ? &data : NULL,
1079 acb) != 0) {
1080 qemu_mutex_unlock(&iscsilun->mutex);
1081 scsi_free_scsi_task(acb->task);
1082 qemu_aio_unref(acb);
1083 return NULL;
1084 }
1085
1086 /* tell libiscsi to read straight into the buffer we got from ioctl */
1087 if (acb->task->xfer_dir == SCSI_XFER_READ) {
1088 if (acb->ioh->iovec_count == 0) {
1089 scsi_task_add_data_in_buffer(acb->task,
1090 acb->ioh->dxfer_len,
1091 acb->ioh->dxferp);
1092 } else {
1093 scsi_task_set_iov_in(acb->task,
1094 (struct scsi_iovec *) acb->ioh->dxferp,
1095 acb->ioh->iovec_count);
1096 }
1097 }
1098
1099 iscsi_set_events(iscsilun);
1100 qemu_mutex_unlock(&iscsilun->mutex);
1101
1102 return &acb->common;
1103 }
1104
1105 #endif
1106
1107 static int64_t
1108 iscsi_getlength(BlockDriverState *bs)
1109 {
1110 IscsiLun *iscsilun = bs->opaque;
1111 int64_t len;
1112
1113 len = iscsilun->num_blocks;
1114 len *= iscsilun->block_size;
1115
1116 return len;
1117 }
1118
1119 static int
1120 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, int count)
1121 {
1122 IscsiLun *iscsilun = bs->opaque;
1123 struct IscsiTask iTask;
1124 struct unmap_list list;
1125 int r = 0;
1126
1127 if (!is_byte_request_lun_aligned(offset, count, iscsilun)) {
1128 return -ENOTSUP;
1129 }
1130
1131 if (!iscsilun->lbp.lbpu) {
1132 /* UNMAP is not supported by the target */
1133 return 0;
1134 }
1135
1136 list.lba = offset / iscsilun->block_size;
1137 list.num = count / iscsilun->block_size;
1138
1139 iscsi_co_init_iscsitask(iscsilun, &iTask);
1140 qemu_mutex_lock(&iscsilun->mutex);
1141 retry:
1142 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
1143 iscsi_co_generic_cb, &iTask) == NULL) {
1144 r = -ENOMEM;
1145 goto out_unlock;
1146 }
1147
1148 while (!iTask.complete) {
1149 iscsi_set_events(iscsilun);
1150 qemu_mutex_unlock(&iscsilun->mutex);
1151 qemu_coroutine_yield();
1152 qemu_mutex_lock(&iscsilun->mutex);
1153 }
1154
1155 if (iTask.task != NULL) {
1156 scsi_free_scsi_task(iTask.task);
1157 iTask.task = NULL;
1158 }
1159
1160 if (iTask.do_retry) {
1161 iTask.complete = 0;
1162 goto retry;
1163 }
1164
1165 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
1166 /* the target might fail with a check condition if it
1167 is not happy with the alignment of the UNMAP request
1168 we silently fail in this case */
1169 goto out_unlock;
1170 }
1171
1172 if (iTask.status != SCSI_STATUS_GOOD) {
1173 r = iTask.err_code;
1174 goto out_unlock;
1175 }
1176
1177 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1178 count >> BDRV_SECTOR_BITS);
1179
1180 out_unlock:
1181 qemu_mutex_unlock(&iscsilun->mutex);
1182 return r;
1183 }
1184
1185 static int
1186 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1187 int count, BdrvRequestFlags flags)
1188 {
1189 IscsiLun *iscsilun = bs->opaque;
1190 struct IscsiTask iTask;
1191 uint64_t lba;
1192 uint32_t nb_blocks;
1193 bool use_16_for_ws = iscsilun->use_16_for_rw;
1194 int r = 0;
1195
1196 if (!is_byte_request_lun_aligned(offset, count, iscsilun)) {
1197 return -ENOTSUP;
1198 }
1199
1200 if (flags & BDRV_REQ_MAY_UNMAP) {
1201 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
1202 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
1203 use_16_for_ws = true;
1204 }
1205 if (use_16_for_ws && !iscsilun->lbp.lbpws) {
1206 /* WRITESAME16 with UNMAP is not supported by the target,
1207 * fall back and try WRITESAME10/16 without UNMAP */
1208 flags &= ~BDRV_REQ_MAY_UNMAP;
1209 use_16_for_ws = iscsilun->use_16_for_rw;
1210 }
1211 }
1212
1213 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1214 /* WRITESAME without UNMAP is not supported by the target */
1215 return -ENOTSUP;
1216 }
1217
1218 lba = offset / iscsilun->block_size;
1219 nb_blocks = count / iscsilun->block_size;
1220
1221 if (iscsilun->zeroblock == NULL) {
1222 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1223 if (iscsilun->zeroblock == NULL) {
1224 return -ENOMEM;
1225 }
1226 }
1227
1228 qemu_mutex_lock(&iscsilun->mutex);
1229 iscsi_co_init_iscsitask(iscsilun, &iTask);
1230 retry:
1231 if (use_16_for_ws) {
1232 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1233 iscsilun->zeroblock, iscsilun->block_size,
1234 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1235 0, 0, iscsi_co_generic_cb, &iTask);
1236 } else {
1237 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1238 iscsilun->zeroblock, iscsilun->block_size,
1239 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1240 0, 0, iscsi_co_generic_cb, &iTask);
1241 }
1242 if (iTask.task == NULL) {
1243 qemu_mutex_unlock(&iscsilun->mutex);
1244 return -ENOMEM;
1245 }
1246
1247 while (!iTask.complete) {
1248 iscsi_set_events(iscsilun);
1249 qemu_mutex_unlock(&iscsilun->mutex);
1250 qemu_coroutine_yield();
1251 qemu_mutex_lock(&iscsilun->mutex);
1252 }
1253
1254 if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1255 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1256 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1257 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1258 /* WRITE SAME is not supported by the target */
1259 iscsilun->has_write_same = false;
1260 scsi_free_scsi_task(iTask.task);
1261 r = -ENOTSUP;
1262 goto out_unlock;
1263 }
1264
1265 if (iTask.task != NULL) {
1266 scsi_free_scsi_task(iTask.task);
1267 iTask.task = NULL;
1268 }
1269
1270 if (iTask.do_retry) {
1271 iTask.complete = 0;
1272 goto retry;
1273 }
1274
1275 if (iTask.status != SCSI_STATUS_GOOD) {
1276 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1277 count >> BDRV_SECTOR_BITS);
1278 r = iTask.err_code;
1279 goto out_unlock;
1280 }
1281
1282 if (flags & BDRV_REQ_MAY_UNMAP) {
1283 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1284 count >> BDRV_SECTOR_BITS);
1285 } else {
1286 iscsi_allocmap_set_allocated(iscsilun, offset >> BDRV_SECTOR_BITS,
1287 count >> BDRV_SECTOR_BITS);
1288 }
1289
1290 out_unlock:
1291 qemu_mutex_unlock(&iscsilun->mutex);
1292 return r;
1293 }
1294
1295 static void apply_chap(struct iscsi_context *iscsi, QemuOpts *opts,
1296 Error **errp)
1297 {
1298 const char *user = NULL;
1299 const char *password = NULL;
1300 const char *secretid;
1301 char *secret = NULL;
1302
1303 user = qemu_opt_get(opts, "user");
1304 if (!user) {
1305 return;
1306 }
1307
1308 secretid = qemu_opt_get(opts, "password-secret");
1309 password = qemu_opt_get(opts, "password");
1310 if (secretid && password) {
1311 error_setg(errp, "'password' and 'password-secret' properties are "
1312 "mutually exclusive");
1313 return;
1314 }
1315 if (secretid) {
1316 secret = qcrypto_secret_lookup_as_utf8(secretid, errp);
1317 if (!secret) {
1318 return;
1319 }
1320 password = secret;
1321 } else if (!password) {
1322 error_setg(errp, "CHAP username specified but no password was given");
1323 return;
1324 }
1325
1326 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1327 error_setg(errp, "Failed to set initiator username and password");
1328 }
1329
1330 g_free(secret);
1331 }
1332
1333 static void apply_header_digest(struct iscsi_context *iscsi, QemuOpts *opts,
1334 Error **errp)
1335 {
1336 const char *digest = NULL;
1337
1338 digest = qemu_opt_get(opts, "header-digest");
1339 if (!digest) {
1340 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1341 } else if (!strcmp(digest, "crc32c")) {
1342 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1343 } else if (!strcmp(digest, "none")) {
1344 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1345 } else if (!strcmp(digest, "crc32c-none")) {
1346 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1347 } else if (!strcmp(digest, "none-crc32c")) {
1348 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1349 } else {
1350 error_setg(errp, "Invalid header-digest setting : %s", digest);
1351 }
1352 }
1353
1354 static char *get_initiator_name(QemuOpts *opts)
1355 {
1356 const char *name;
1357 char *iscsi_name;
1358 UuidInfo *uuid_info;
1359
1360 name = qemu_opt_get(opts, "initiator-name");
1361 if (name) {
1362 return g_strdup(name);
1363 }
1364
1365 uuid_info = qmp_query_uuid(NULL);
1366 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1367 name = qemu_get_vm_name();
1368 } else {
1369 name = uuid_info->UUID;
1370 }
1371 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1372 name ? ":" : "", name ? name : "");
1373 qapi_free_UuidInfo(uuid_info);
1374 return iscsi_name;
1375 }
1376
1377 static void iscsi_nop_timed_event(void *opaque)
1378 {
1379 IscsiLun *iscsilun = opaque;
1380
1381 qemu_mutex_lock(&iscsilun->mutex);
1382 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1383 error_report("iSCSI: NOP timeout. Reconnecting...");
1384 iscsilun->request_timed_out = true;
1385 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1386 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1387 goto out;
1388 }
1389
1390 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1391 iscsi_set_events(iscsilun);
1392
1393 out:
1394 qemu_mutex_unlock(&iscsilun->mutex);
1395 }
1396
1397 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1398 {
1399 struct scsi_task *task = NULL;
1400 struct scsi_readcapacity10 *rc10 = NULL;
1401 struct scsi_readcapacity16 *rc16 = NULL;
1402 int retries = ISCSI_CMD_RETRIES;
1403
1404 do {
1405 if (task != NULL) {
1406 scsi_free_scsi_task(task);
1407 task = NULL;
1408 }
1409
1410 switch (iscsilun->type) {
1411 case TYPE_DISK:
1412 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1413 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1414 rc16 = scsi_datain_unmarshall(task);
1415 if (rc16 == NULL) {
1416 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1417 } else {
1418 iscsilun->block_size = rc16->block_length;
1419 iscsilun->num_blocks = rc16->returned_lba + 1;
1420 iscsilun->lbpme = !!rc16->lbpme;
1421 iscsilun->lbprz = !!rc16->lbprz;
1422 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1423 }
1424 break;
1425 }
1426 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1427 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1428 break;
1429 }
1430 /* Fall through and try READ CAPACITY(10) instead. */
1431 case TYPE_ROM:
1432 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1433 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1434 rc10 = scsi_datain_unmarshall(task);
1435 if (rc10 == NULL) {
1436 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1437 } else {
1438 iscsilun->block_size = rc10->block_size;
1439 if (rc10->lba == 0) {
1440 /* blank disk loaded */
1441 iscsilun->num_blocks = 0;
1442 } else {
1443 iscsilun->num_blocks = rc10->lba + 1;
1444 }
1445 }
1446 }
1447 break;
1448 default:
1449 return;
1450 }
1451 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1452 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1453 && retries-- > 0);
1454
1455 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1456 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1457 } else if (!iscsilun->block_size ||
1458 iscsilun->block_size % BDRV_SECTOR_SIZE) {
1459 error_setg(errp, "iSCSI: the target returned an invalid "
1460 "block size of %d.", iscsilun->block_size);
1461 }
1462 if (task) {
1463 scsi_free_scsi_task(task);
1464 }
1465 }
1466
1467 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1468 int evpd, int pc, void **inq, Error **errp)
1469 {
1470 int full_size;
1471 struct scsi_task *task = NULL;
1472 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1473 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1474 goto fail;
1475 }
1476 full_size = scsi_datain_getfullsize(task);
1477 if (full_size > task->datain.size) {
1478 scsi_free_scsi_task(task);
1479
1480 /* we need more data for the full list */
1481 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1482 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1483 goto fail;
1484 }
1485 }
1486
1487 *inq = scsi_datain_unmarshall(task);
1488 if (*inq == NULL) {
1489 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1490 goto fail_with_err;
1491 }
1492
1493 return task;
1494
1495 fail:
1496 error_setg(errp, "iSCSI: Inquiry command failed : %s",
1497 iscsi_get_error(iscsi));
1498 fail_with_err:
1499 if (task != NULL) {
1500 scsi_free_scsi_task(task);
1501 }
1502 return NULL;
1503 }
1504
1505 static void iscsi_detach_aio_context(BlockDriverState *bs)
1506 {
1507 IscsiLun *iscsilun = bs->opaque;
1508
1509 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1510 false, NULL, NULL, NULL, NULL);
1511 iscsilun->events = 0;
1512
1513 if (iscsilun->nop_timer) {
1514 timer_del(iscsilun->nop_timer);
1515 timer_free(iscsilun->nop_timer);
1516 iscsilun->nop_timer = NULL;
1517 }
1518 if (iscsilun->event_timer) {
1519 timer_del(iscsilun->event_timer);
1520 timer_free(iscsilun->event_timer);
1521 iscsilun->event_timer = NULL;
1522 }
1523 }
1524
1525 static void iscsi_attach_aio_context(BlockDriverState *bs,
1526 AioContext *new_context)
1527 {
1528 IscsiLun *iscsilun = bs->opaque;
1529
1530 iscsilun->aio_context = new_context;
1531 iscsi_set_events(iscsilun);
1532
1533 /* Set up a timer for sending out iSCSI NOPs */
1534 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1535 QEMU_CLOCK_REALTIME, SCALE_MS,
1536 iscsi_nop_timed_event, iscsilun);
1537 timer_mod(iscsilun->nop_timer,
1538 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1539
1540 /* Set up a timer for periodic calls to iscsi_set_events and to
1541 * scan for command timeout */
1542 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1543 QEMU_CLOCK_REALTIME, SCALE_MS,
1544 iscsi_timed_check_events, iscsilun);
1545 timer_mod(iscsilun->event_timer,
1546 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1547 }
1548
1549 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1550 {
1551 struct scsi_task *task;
1552 struct scsi_mode_sense *ms = NULL;
1553 iscsilun->write_protected = false;
1554 iscsilun->dpofua = false;
1555
1556 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1557 1, SCSI_MODESENSE_PC_CURRENT,
1558 0x3F, 0, 255);
1559 if (task == NULL) {
1560 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1561 iscsi_get_error(iscsilun->iscsi));
1562 goto out;
1563 }
1564
1565 if (task->status != SCSI_STATUS_GOOD) {
1566 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1567 goto out;
1568 }
1569 ms = scsi_datain_unmarshall(task);
1570 if (!ms) {
1571 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1572 iscsi_get_error(iscsilun->iscsi));
1573 goto out;
1574 }
1575 iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1576 iscsilun->dpofua = ms->device_specific_parameter & 0x10;
1577
1578 out:
1579 if (task) {
1580 scsi_free_scsi_task(task);
1581 }
1582 }
1583
1584 static void iscsi_parse_iscsi_option(const char *target, QDict *options)
1585 {
1586 QemuOptsList *list;
1587 QemuOpts *opts;
1588 const char *user, *password, *password_secret, *initiator_name,
1589 *header_digest, *timeout;
1590
1591 list = qemu_find_opts("iscsi");
1592 if (!list) {
1593 return;
1594 }
1595
1596 opts = qemu_opts_find(list, target);
1597 if (opts == NULL) {
1598 opts = QTAILQ_FIRST(&list->head);
1599 if (!opts) {
1600 return;
1601 }
1602 }
1603
1604 user = qemu_opt_get(opts, "user");
1605 if (user) {
1606 qdict_set_default_str(options, "user", user);
1607 }
1608
1609 password = qemu_opt_get(opts, "password");
1610 if (password) {
1611 qdict_set_default_str(options, "password", password);
1612 }
1613
1614 password_secret = qemu_opt_get(opts, "password-secret");
1615 if (password_secret) {
1616 qdict_set_default_str(options, "password-secret", password_secret);
1617 }
1618
1619 initiator_name = qemu_opt_get(opts, "initiator-name");
1620 if (initiator_name) {
1621 qdict_set_default_str(options, "initiator-name", initiator_name);
1622 }
1623
1624 header_digest = qemu_opt_get(opts, "header-digest");
1625 if (header_digest) {
1626 /* -iscsi takes upper case values, but QAPI only supports lower case
1627 * enum constant names, so we have to convert here. */
1628 char *qapi_value = g_ascii_strdown(header_digest, -1);
1629 qdict_set_default_str(options, "header-digest", qapi_value);
1630 g_free(qapi_value);
1631 }
1632
1633 timeout = qemu_opt_get(opts, "timeout");
1634 if (timeout) {
1635 qdict_set_default_str(options, "timeout", timeout);
1636 }
1637 }
1638
1639 /*
1640 * We support iscsi url's on the form
1641 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1642 */
1643 static void iscsi_parse_filename(const char *filename, QDict *options,
1644 Error **errp)
1645 {
1646 struct iscsi_url *iscsi_url;
1647 const char *transport_name;
1648 char *lun_str;
1649
1650 iscsi_url = iscsi_parse_full_url(NULL, filename);
1651 if (iscsi_url == NULL) {
1652 error_setg(errp, "Failed to parse URL : %s", filename);
1653 return;
1654 }
1655
1656 #if LIBISCSI_API_VERSION >= (20160603)
1657 switch (iscsi_url->transport) {
1658 case TCP_TRANSPORT:
1659 transport_name = "tcp";
1660 break;
1661 case ISER_TRANSPORT:
1662 transport_name = "iser";
1663 break;
1664 default:
1665 error_setg(errp, "Unknown transport type (%d)",
1666 iscsi_url->transport);
1667 return;
1668 }
1669 #else
1670 transport_name = "tcp";
1671 #endif
1672
1673 qdict_set_default_str(options, "transport", transport_name);
1674 qdict_set_default_str(options, "portal", iscsi_url->portal);
1675 qdict_set_default_str(options, "target", iscsi_url->target);
1676
1677 lun_str = g_strdup_printf("%d", iscsi_url->lun);
1678 qdict_set_default_str(options, "lun", lun_str);
1679 g_free(lun_str);
1680
1681 /* User/password from -iscsi take precedence over those from the URL */
1682 iscsi_parse_iscsi_option(iscsi_url->target, options);
1683
1684 if (iscsi_url->user[0] != '\0') {
1685 qdict_set_default_str(options, "user", iscsi_url->user);
1686 qdict_set_default_str(options, "password", iscsi_url->passwd);
1687 }
1688
1689 iscsi_destroy_url(iscsi_url);
1690 }
1691
1692 static QemuOptsList runtime_opts = {
1693 .name = "iscsi",
1694 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1695 .desc = {
1696 {
1697 .name = "transport",
1698 .type = QEMU_OPT_STRING,
1699 },
1700 {
1701 .name = "portal",
1702 .type = QEMU_OPT_STRING,
1703 },
1704 {
1705 .name = "target",
1706 .type = QEMU_OPT_STRING,
1707 },
1708 {
1709 .name = "user",
1710 .type = QEMU_OPT_STRING,
1711 },
1712 {
1713 .name = "password",
1714 .type = QEMU_OPT_STRING,
1715 },
1716 {
1717 .name = "password-secret",
1718 .type = QEMU_OPT_STRING,
1719 },
1720 {
1721 .name = "lun",
1722 .type = QEMU_OPT_NUMBER,
1723 },
1724 {
1725 .name = "initiator-name",
1726 .type = QEMU_OPT_STRING,
1727 },
1728 {
1729 .name = "header-digest",
1730 .type = QEMU_OPT_STRING,
1731 },
1732 {
1733 .name = "timeout",
1734 .type = QEMU_OPT_NUMBER,
1735 },
1736 { /* end of list */ }
1737 },
1738 };
1739
1740 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1741 Error **errp)
1742 {
1743 IscsiLun *iscsilun = bs->opaque;
1744 struct iscsi_context *iscsi = NULL;
1745 struct scsi_task *task = NULL;
1746 struct scsi_inquiry_standard *inq = NULL;
1747 struct scsi_inquiry_supported_pages *inq_vpd;
1748 char *initiator_name = NULL;
1749 QemuOpts *opts;
1750 Error *local_err = NULL;
1751 const char *transport_name, *portal, *target;
1752 #if LIBISCSI_API_VERSION >= (20160603)
1753 enum iscsi_transport_type transport;
1754 #endif
1755 int i, ret = 0, timeout = 0, lun;
1756
1757 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1758 qemu_opts_absorb_qdict(opts, options, &local_err);
1759 if (local_err) {
1760 error_propagate(errp, local_err);
1761 ret = -EINVAL;
1762 goto out;
1763 }
1764
1765 transport_name = qemu_opt_get(opts, "transport");
1766 portal = qemu_opt_get(opts, "portal");
1767 target = qemu_opt_get(opts, "target");
1768 lun = qemu_opt_get_number(opts, "lun", 0);
1769
1770 if (!transport_name || !portal || !target) {
1771 error_setg(errp, "Need all of transport, portal and target options");
1772 ret = -EINVAL;
1773 goto out;
1774 }
1775
1776 if (!strcmp(transport_name, "tcp")) {
1777 #if LIBISCSI_API_VERSION >= (20160603)
1778 transport = TCP_TRANSPORT;
1779 } else if (!strcmp(transport_name, "iser")) {
1780 transport = ISER_TRANSPORT;
1781 #else
1782 /* TCP is what older libiscsi versions always use */
1783 #endif
1784 } else {
1785 error_setg(errp, "Unknown transport: %s", transport_name);
1786 ret = -EINVAL;
1787 goto out;
1788 }
1789
1790 memset(iscsilun, 0, sizeof(IscsiLun));
1791
1792 initiator_name = get_initiator_name(opts);
1793
1794 iscsi = iscsi_create_context(initiator_name);
1795 if (iscsi == NULL) {
1796 error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1797 ret = -ENOMEM;
1798 goto out;
1799 }
1800 #if LIBISCSI_API_VERSION >= (20160603)
1801 if (iscsi_init_transport(iscsi, transport)) {
1802 error_setg(errp, ("Error initializing transport."));
1803 ret = -EINVAL;
1804 goto out;
1805 }
1806 #endif
1807 if (iscsi_set_targetname(iscsi, target)) {
1808 error_setg(errp, "iSCSI: Failed to set target name.");
1809 ret = -EINVAL;
1810 goto out;
1811 }
1812
1813 /* check if we got CHAP username/password via the options */
1814 apply_chap(iscsi, opts, &local_err);
1815 if (local_err != NULL) {
1816 error_propagate(errp, local_err);
1817 ret = -EINVAL;
1818 goto out;
1819 }
1820
1821 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1822 error_setg(errp, "iSCSI: Failed to set session type to normal.");
1823 ret = -EINVAL;
1824 goto out;
1825 }
1826
1827 /* check if we got HEADER_DIGEST via the options */
1828 apply_header_digest(iscsi, opts, &local_err);
1829 if (local_err != NULL) {
1830 error_propagate(errp, local_err);
1831 ret = -EINVAL;
1832 goto out;
1833 }
1834
1835 /* timeout handling is broken in libiscsi before 1.15.0 */
1836 timeout = qemu_opt_get_number(opts, "timeout", 0);
1837 #if LIBISCSI_API_VERSION >= 20150621
1838 iscsi_set_timeout(iscsi, timeout);
1839 #else
1840 if (timeout) {
1841 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1842 }
1843 #endif
1844
1845 if (iscsi_full_connect_sync(iscsi, portal, lun) != 0) {
1846 error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1847 iscsi_get_error(iscsi));
1848 ret = -EINVAL;
1849 goto out;
1850 }
1851
1852 iscsilun->iscsi = iscsi;
1853 iscsilun->aio_context = bdrv_get_aio_context(bs);
1854 iscsilun->lun = lun;
1855 iscsilun->has_write_same = true;
1856
1857 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1858 (void **) &inq, errp);
1859 if (task == NULL) {
1860 ret = -EINVAL;
1861 goto out;
1862 }
1863 iscsilun->type = inq->periperal_device_type;
1864 scsi_free_scsi_task(task);
1865 task = NULL;
1866
1867 iscsi_modesense_sync(iscsilun);
1868 if (iscsilun->dpofua) {
1869 bs->supported_write_flags = BDRV_REQ_FUA;
1870 }
1871 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
1872
1873 /* Check the write protect flag of the LUN if we want to write */
1874 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1875 iscsilun->write_protected) {
1876 error_setg(errp, "Cannot open a write protected LUN as read-write");
1877 ret = -EACCES;
1878 goto out;
1879 }
1880
1881 iscsi_readcapacity_sync(iscsilun, &local_err);
1882 if (local_err != NULL) {
1883 error_propagate(errp, local_err);
1884 ret = -EINVAL;
1885 goto out;
1886 }
1887 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1888
1889 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1890 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1891 * will try to read from the device to guess the image format.
1892 */
1893 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1894 bs->sg = true;
1895 }
1896
1897 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1898 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1899 (void **) &inq_vpd, errp);
1900 if (task == NULL) {
1901 ret = -EINVAL;
1902 goto out;
1903 }
1904 for (i = 0; i < inq_vpd->num_pages; i++) {
1905 struct scsi_task *inq_task;
1906 struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1907 struct scsi_inquiry_block_limits *inq_bl;
1908 switch (inq_vpd->pages[i]) {
1909 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1910 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1911 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1912 (void **) &inq_lbp, errp);
1913 if (inq_task == NULL) {
1914 ret = -EINVAL;
1915 goto out;
1916 }
1917 memcpy(&iscsilun->lbp, inq_lbp,
1918 sizeof(struct scsi_inquiry_logical_block_provisioning));
1919 scsi_free_scsi_task(inq_task);
1920 break;
1921 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1922 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1923 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1924 (void **) &inq_bl, errp);
1925 if (inq_task == NULL) {
1926 ret = -EINVAL;
1927 goto out;
1928 }
1929 memcpy(&iscsilun->bl, inq_bl,
1930 sizeof(struct scsi_inquiry_block_limits));
1931 scsi_free_scsi_task(inq_task);
1932 break;
1933 default:
1934 break;
1935 }
1936 }
1937 scsi_free_scsi_task(task);
1938 task = NULL;
1939
1940 qemu_mutex_init(&iscsilun->mutex);
1941 iscsi_attach_aio_context(bs, iscsilun->aio_context);
1942
1943 /* Guess the internal cluster (page) size of the iscsi target by the means
1944 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1945 * reasonable size */
1946 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
1947 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1948 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran *
1949 iscsilun->block_size) >> BDRV_SECTOR_BITS;
1950 if (iscsilun->lbprz) {
1951 ret = iscsi_allocmap_init(iscsilun, bs->open_flags);
1952 }
1953 }
1954
1955 out:
1956 qemu_opts_del(opts);
1957 g_free(initiator_name);
1958 if (task != NULL) {
1959 scsi_free_scsi_task(task);
1960 }
1961
1962 if (ret) {
1963 if (iscsi != NULL) {
1964 if (iscsi_is_logged_in(iscsi)) {
1965 iscsi_logout_sync(iscsi);
1966 }
1967 iscsi_destroy_context(iscsi);
1968 }
1969 memset(iscsilun, 0, sizeof(IscsiLun));
1970 }
1971 return ret;
1972 }
1973
1974 static void iscsi_close(BlockDriverState *bs)
1975 {
1976 IscsiLun *iscsilun = bs->opaque;
1977 struct iscsi_context *iscsi = iscsilun->iscsi;
1978
1979 iscsi_detach_aio_context(bs);
1980 if (iscsi_is_logged_in(iscsi)) {
1981 iscsi_logout_sync(iscsi);
1982 }
1983 iscsi_destroy_context(iscsi);
1984 g_free(iscsilun->zeroblock);
1985 iscsi_allocmap_free(iscsilun);
1986 qemu_mutex_destroy(&iscsilun->mutex);
1987 memset(iscsilun, 0, sizeof(IscsiLun));
1988 }
1989
1990 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
1991 {
1992 /* We don't actually refresh here, but just return data queried in
1993 * iscsi_open(): iscsi targets don't change their limits. */
1994
1995 IscsiLun *iscsilun = bs->opaque;
1996 uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
1997 unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size);
1998
1999 assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bs->sg);
2000
2001 bs->bl.request_alignment = block_size;
2002
2003 if (iscsilun->bl.max_xfer_len) {
2004 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
2005 }
2006
2007 if (max_xfer_len * block_size < INT_MAX) {
2008 bs->bl.max_transfer = max_xfer_len * iscsilun->block_size;
2009 }
2010
2011 if (iscsilun->lbp.lbpu) {
2012 if (iscsilun->bl.max_unmap < 0xffffffff / block_size) {
2013 bs->bl.max_pdiscard =
2014 iscsilun->bl.max_unmap * iscsilun->block_size;
2015 }
2016 bs->bl.pdiscard_alignment =
2017 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2018 } else {
2019 bs->bl.pdiscard_alignment = iscsilun->block_size;
2020 }
2021
2022 if (iscsilun->bl.max_ws_len < 0xffffffff / block_size) {
2023 bs->bl.max_pwrite_zeroes =
2024 iscsilun->bl.max_ws_len * iscsilun->block_size;
2025 }
2026 if (iscsilun->lbp.lbpws) {
2027 bs->bl.pwrite_zeroes_alignment =
2028 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2029 } else {
2030 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size;
2031 }
2032 if (iscsilun->bl.opt_xfer_len &&
2033 iscsilun->bl.opt_xfer_len < INT_MAX / block_size) {
2034 bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len *
2035 iscsilun->block_size);
2036 }
2037 }
2038
2039 /* Note that this will not re-establish a connection with an iSCSI target - it
2040 * is effectively a NOP. */
2041 static int iscsi_reopen_prepare(BDRVReopenState *state,
2042 BlockReopenQueue *queue, Error **errp)
2043 {
2044 IscsiLun *iscsilun = state->bs->opaque;
2045
2046 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
2047 error_setg(errp, "Cannot open a write protected LUN as read-write");
2048 return -EACCES;
2049 }
2050 return 0;
2051 }
2052
2053 static void iscsi_reopen_commit(BDRVReopenState *reopen_state)
2054 {
2055 IscsiLun *iscsilun = reopen_state->bs->opaque;
2056
2057 /* the cache.direct status might have changed */
2058 if (iscsilun->allocmap != NULL) {
2059 iscsi_allocmap_init(iscsilun, reopen_state->flags);
2060 }
2061 }
2062
2063 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
2064 {
2065 IscsiLun *iscsilun = bs->opaque;
2066 Error *local_err = NULL;
2067
2068 if (iscsilun->type != TYPE_DISK) {
2069 return -ENOTSUP;
2070 }
2071
2072 iscsi_readcapacity_sync(iscsilun, &local_err);
2073 if (local_err != NULL) {
2074 error_free(local_err);
2075 return -EIO;
2076 }
2077
2078 if (offset > iscsi_getlength(bs)) {
2079 return -EINVAL;
2080 }
2081
2082 if (iscsilun->allocmap != NULL) {
2083 iscsi_allocmap_init(iscsilun, bs->open_flags);
2084 }
2085
2086 return 0;
2087 }
2088
2089 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp)
2090 {
2091 int ret = 0;
2092 int64_t total_size = 0;
2093 BlockDriverState *bs;
2094 IscsiLun *iscsilun = NULL;
2095 QDict *bs_options;
2096
2097 bs = bdrv_new();
2098
2099 /* Read out options */
2100 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2101 BDRV_SECTOR_SIZE);
2102 bs->opaque = g_new0(struct IscsiLun, 1);
2103 iscsilun = bs->opaque;
2104
2105 bs_options = qdict_new();
2106 qdict_put(bs_options, "filename", qstring_from_str(filename));
2107 ret = iscsi_open(bs, bs_options, 0, NULL);
2108 QDECREF(bs_options);
2109
2110 if (ret != 0) {
2111 goto out;
2112 }
2113 iscsi_detach_aio_context(bs);
2114 if (iscsilun->type != TYPE_DISK) {
2115 ret = -ENODEV;
2116 goto out;
2117 }
2118 if (bs->total_sectors < total_size) {
2119 ret = -ENOSPC;
2120 goto out;
2121 }
2122
2123 ret = 0;
2124 out:
2125 if (iscsilun->iscsi != NULL) {
2126 iscsi_destroy_context(iscsilun->iscsi);
2127 }
2128 g_free(bs->opaque);
2129 bs->opaque = NULL;
2130 bdrv_unref(bs);
2131 return ret;
2132 }
2133
2134 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2135 {
2136 IscsiLun *iscsilun = bs->opaque;
2137 bdi->unallocated_blocks_are_zero = iscsilun->lbprz;
2138 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
2139 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE;
2140 return 0;
2141 }
2142
2143 static void iscsi_invalidate_cache(BlockDriverState *bs,
2144 Error **errp)
2145 {
2146 IscsiLun *iscsilun = bs->opaque;
2147 iscsi_allocmap_invalidate(iscsilun);
2148 }
2149
2150 static QemuOptsList iscsi_create_opts = {
2151 .name = "iscsi-create-opts",
2152 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
2153 .desc = {
2154 {
2155 .name = BLOCK_OPT_SIZE,
2156 .type = QEMU_OPT_SIZE,
2157 .help = "Virtual disk size"
2158 },
2159 { /* end of list */ }
2160 }
2161 };
2162
2163 static BlockDriver bdrv_iscsi = {
2164 .format_name = "iscsi",
2165 .protocol_name = "iscsi",
2166
2167 .instance_size = sizeof(IscsiLun),
2168 .bdrv_parse_filename = iscsi_parse_filename,
2169 .bdrv_file_open = iscsi_open,
2170 .bdrv_close = iscsi_close,
2171 .bdrv_create = iscsi_create,
2172 .create_opts = &iscsi_create_opts,
2173 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2174 .bdrv_reopen_commit = iscsi_reopen_commit,
2175 .bdrv_invalidate_cache = iscsi_invalidate_cache,
2176
2177 .bdrv_getlength = iscsi_getlength,
2178 .bdrv_get_info = iscsi_get_info,
2179 .bdrv_truncate = iscsi_truncate,
2180 .bdrv_refresh_limits = iscsi_refresh_limits,
2181
2182 .bdrv_co_get_block_status = iscsi_co_get_block_status,
2183 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2184 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2185 .bdrv_co_readv = iscsi_co_readv,
2186 .bdrv_co_writev_flags = iscsi_co_writev_flags,
2187 .bdrv_co_flush_to_disk = iscsi_co_flush,
2188
2189 #ifdef __linux__
2190 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2191 #endif
2192
2193 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2194 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2195 };
2196
2197 #if LIBISCSI_API_VERSION >= (20160603)
2198 static BlockDriver bdrv_iser = {
2199 .format_name = "iser",
2200 .protocol_name = "iser",
2201
2202 .instance_size = sizeof(IscsiLun),
2203 .bdrv_parse_filename = iscsi_parse_filename,
2204 .bdrv_file_open = iscsi_open,
2205 .bdrv_close = iscsi_close,
2206 .bdrv_create = iscsi_create,
2207 .create_opts = &iscsi_create_opts,
2208 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2209 .bdrv_reopen_commit = iscsi_reopen_commit,
2210 .bdrv_invalidate_cache = iscsi_invalidate_cache,
2211
2212 .bdrv_getlength = iscsi_getlength,
2213 .bdrv_get_info = iscsi_get_info,
2214 .bdrv_truncate = iscsi_truncate,
2215 .bdrv_refresh_limits = iscsi_refresh_limits,
2216
2217 .bdrv_co_get_block_status = iscsi_co_get_block_status,
2218 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2219 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2220 .bdrv_co_readv = iscsi_co_readv,
2221 .bdrv_co_writev_flags = iscsi_co_writev_flags,
2222 .bdrv_co_flush_to_disk = iscsi_co_flush,
2223
2224 #ifdef __linux__
2225 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2226 #endif
2227
2228 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2229 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2230 };
2231 #endif
2232
2233 static void iscsi_block_init(void)
2234 {
2235 bdrv_register(&bdrv_iscsi);
2236 #if LIBISCSI_API_VERSION >= (20160603)
2237 bdrv_register(&bdrv_iser);
2238 #endif
2239 }
2240
2241 block_init(iscsi_block_init);