block: honor BDRV_O_ALLOW_RDWR when clearing bs->read_only
[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 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
169 iTask->complete = 1;
170 aio_co_wake(iTask->co);
171 }
172
173 static void iscsi_retry_timer_expired(void *opaque)
174 {
175 struct IscsiTask *iTask = opaque;
176 iTask->complete = 1;
177 if (iTask->co) {
178 aio_co_wake(iTask->co);
179 }
180 }
181
182 static inline unsigned exp_random(double mean)
183 {
184 return -mean * log((double)rand() / RAND_MAX);
185 }
186
187 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in
188 * libiscsi 1.10.0, together with other constants we need. Use it as
189 * a hint that we have to define them ourselves if needed, to keep the
190 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for
191 * the test because SCSI_STATUS_* is an enum.
192 *
193 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes
194 * an enum, check against the LIBISCSI_API_VERSION macro, which was
195 * introduced in 1.11.0. If it is present, there is no need to define
196 * anything.
197 */
198 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \
199 !defined(LIBISCSI_API_VERSION)
200 #define SCSI_STATUS_TASK_SET_FULL 0x28
201 #define SCSI_STATUS_TIMEOUT 0x0f000002
202 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600
203 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00
204 #endif
205
206 #ifndef LIBISCSI_API_VERSION
207 #define LIBISCSI_API_VERSION 20130701
208 #endif
209
210 static int iscsi_translate_sense(struct scsi_sense *sense)
211 {
212 int ret;
213
214 switch (sense->key) {
215 case SCSI_SENSE_NOT_READY:
216 return -EBUSY;
217 case SCSI_SENSE_DATA_PROTECTION:
218 return -EACCES;
219 case SCSI_SENSE_COMMAND_ABORTED:
220 return -ECANCELED;
221 case SCSI_SENSE_ILLEGAL_REQUEST:
222 /* Parse ASCQ */
223 break;
224 default:
225 return -EIO;
226 }
227 switch (sense->ascq) {
228 case SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR:
229 case SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE:
230 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB:
231 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST:
232 ret = -EINVAL;
233 break;
234 case SCSI_SENSE_ASCQ_LBA_OUT_OF_RANGE:
235 ret = -ENOSPC;
236 break;
237 case SCSI_SENSE_ASCQ_LOGICAL_UNIT_NOT_SUPPORTED:
238 ret = -ENOTSUP;
239 break;
240 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT:
241 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_CLOSED:
242 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_OPEN:
243 ret = -ENOMEDIUM;
244 break;
245 case SCSI_SENSE_ASCQ_WRITE_PROTECTED:
246 ret = -EACCES;
247 break;
248 default:
249 ret = -EIO;
250 break;
251 }
252 return ret;
253 }
254
255 /* Called (via iscsi_service) with QemuMutex held. */
256 static void
257 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
258 void *command_data, void *opaque)
259 {
260 struct IscsiTask *iTask = opaque;
261 struct scsi_task *task = command_data;
262
263 iTask->status = status;
264 iTask->do_retry = 0;
265 iTask->task = task;
266
267 if (status != SCSI_STATUS_GOOD) {
268 if (iTask->retries++ < ISCSI_CMD_RETRIES) {
269 if (status == SCSI_STATUS_CHECK_CONDITION
270 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
271 error_report("iSCSI CheckCondition: %s",
272 iscsi_get_error(iscsi));
273 iTask->do_retry = 1;
274 goto out;
275 }
276 if (status == SCSI_STATUS_BUSY ||
277 status == SCSI_STATUS_TIMEOUT ||
278 status == SCSI_STATUS_TASK_SET_FULL) {
279 unsigned retry_time =
280 exp_random(iscsi_retry_times[iTask->retries - 1]);
281 if (status == SCSI_STATUS_TIMEOUT) {
282 /* make sure the request is rescheduled AFTER the
283 * reconnect is initiated */
284 retry_time = EVENT_INTERVAL * 2;
285 iTask->iscsilun->request_timed_out = true;
286 }
287 error_report("iSCSI Busy/TaskSetFull/TimeOut"
288 " (retry #%u in %u ms): %s",
289 iTask->retries, retry_time,
290 iscsi_get_error(iscsi));
291 aio_timer_init(iTask->iscsilun->aio_context,
292 &iTask->retry_timer, QEMU_CLOCK_REALTIME,
293 SCALE_MS, iscsi_retry_timer_expired, iTask);
294 timer_mod(&iTask->retry_timer,
295 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
296 iTask->do_retry = 1;
297 return;
298 }
299 }
300 iTask->err_code = iscsi_translate_sense(&task->sense);
301 error_report("iSCSI Failure: %s", iscsi_get_error(iscsi));
302 }
303
304 out:
305 if (iTask->co) {
306 aio_bh_schedule_oneshot(iTask->iscsilun->aio_context,
307 iscsi_co_generic_bh_cb, iTask);
308 } else {
309 iTask->complete = 1;
310 }
311 }
312
313 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
314 {
315 *iTask = (struct IscsiTask) {
316 .co = qemu_coroutine_self(),
317 .iscsilun = iscsilun,
318 };
319 }
320
321 static void
322 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
323 void *private_data)
324 {
325 IscsiAIOCB *acb = private_data;
326
327 acb->status = -ECANCELED;
328 iscsi_schedule_bh(acb);
329 }
330
331 static void
332 iscsi_aio_cancel(BlockAIOCB *blockacb)
333 {
334 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
335 IscsiLun *iscsilun = acb->iscsilun;
336
337 if (acb->status != -EINPROGRESS) {
338 return;
339 }
340
341 /* send a task mgmt call to the target to cancel the task on the target */
342 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
343 iscsi_abort_task_cb, acb);
344
345 }
346
347 static const AIOCBInfo iscsi_aiocb_info = {
348 .aiocb_size = sizeof(IscsiAIOCB),
349 .cancel_async = iscsi_aio_cancel,
350 };
351
352
353 static void iscsi_process_read(void *arg);
354 static void iscsi_process_write(void *arg);
355
356 /* Called with QemuMutex held. */
357 static void
358 iscsi_set_events(IscsiLun *iscsilun)
359 {
360 struct iscsi_context *iscsi = iscsilun->iscsi;
361 int ev = iscsi_which_events(iscsi);
362
363 if (ev != iscsilun->events) {
364 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi),
365 false,
366 (ev & POLLIN) ? iscsi_process_read : NULL,
367 (ev & POLLOUT) ? iscsi_process_write : NULL,
368 NULL,
369 iscsilun);
370 iscsilun->events = ev;
371 }
372 }
373
374 static void iscsi_timed_check_events(void *opaque)
375 {
376 IscsiLun *iscsilun = opaque;
377
378 /* check for timed out requests */
379 iscsi_service(iscsilun->iscsi, 0);
380
381 if (iscsilun->request_timed_out) {
382 iscsilun->request_timed_out = false;
383 iscsi_reconnect(iscsilun->iscsi);
384 }
385
386 /* newer versions of libiscsi may return zero events. Ensure we are able
387 * to return to service once this situation changes. */
388 iscsi_set_events(iscsilun);
389
390 timer_mod(iscsilun->event_timer,
391 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
392 }
393
394 static void
395 iscsi_process_read(void *arg)
396 {
397 IscsiLun *iscsilun = arg;
398 struct iscsi_context *iscsi = iscsilun->iscsi;
399
400 qemu_mutex_lock(&iscsilun->mutex);
401 iscsi_service(iscsi, POLLIN);
402 iscsi_set_events(iscsilun);
403 qemu_mutex_unlock(&iscsilun->mutex);
404 }
405
406 static void
407 iscsi_process_write(void *arg)
408 {
409 IscsiLun *iscsilun = arg;
410 struct iscsi_context *iscsi = iscsilun->iscsi;
411
412 qemu_mutex_lock(&iscsilun->mutex);
413 iscsi_service(iscsi, POLLOUT);
414 iscsi_set_events(iscsilun);
415 qemu_mutex_unlock(&iscsilun->mutex);
416 }
417
418 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
419 {
420 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
421 }
422
423 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
424 {
425 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
426 }
427
428 static bool is_byte_request_lun_aligned(int64_t offset, int count,
429 IscsiLun *iscsilun)
430 {
431 if (offset % iscsilun->block_size || count % iscsilun->block_size) {
432 error_report("iSCSI misaligned request: "
433 "iscsilun->block_size %u, offset %" PRIi64
434 ", count %d",
435 iscsilun->block_size, offset, count);
436 return false;
437 }
438 return true;
439 }
440
441 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors,
442 IscsiLun *iscsilun)
443 {
444 assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS);
445 return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS,
446 nb_sectors << BDRV_SECTOR_BITS,
447 iscsilun);
448 }
449
450 static void iscsi_allocmap_free(IscsiLun *iscsilun)
451 {
452 g_free(iscsilun->allocmap);
453 g_free(iscsilun->allocmap_valid);
454 iscsilun->allocmap = NULL;
455 iscsilun->allocmap_valid = NULL;
456 }
457
458
459 static int iscsi_allocmap_init(IscsiLun *iscsilun, int open_flags)
460 {
461 iscsi_allocmap_free(iscsilun);
462
463 iscsilun->allocmap_size =
464 DIV_ROUND_UP(sector_lun2qemu(iscsilun->num_blocks, iscsilun),
465 iscsilun->cluster_sectors);
466
467 iscsilun->allocmap = bitmap_try_new(iscsilun->allocmap_size);
468 if (!iscsilun->allocmap) {
469 return -ENOMEM;
470 }
471
472 if (open_flags & BDRV_O_NOCACHE) {
473 /* in case that cache.direct = on all allocmap entries are
474 * treated as invalid to force a relookup of the block
475 * status on every read request */
476 return 0;
477 }
478
479 iscsilun->allocmap_valid = bitmap_try_new(iscsilun->allocmap_size);
480 if (!iscsilun->allocmap_valid) {
481 /* if we are under memory pressure free the allocmap as well */
482 iscsi_allocmap_free(iscsilun);
483 return -ENOMEM;
484 }
485
486 return 0;
487 }
488
489 static void
490 iscsi_allocmap_update(IscsiLun *iscsilun, int64_t sector_num,
491 int nb_sectors, bool allocated, bool valid)
492 {
493 int64_t cl_num_expanded, nb_cls_expanded, cl_num_shrunk, nb_cls_shrunk;
494
495 if (iscsilun->allocmap == NULL) {
496 return;
497 }
498 /* expand to entirely contain all affected clusters */
499 cl_num_expanded = sector_num / iscsilun->cluster_sectors;
500 nb_cls_expanded = DIV_ROUND_UP(sector_num + nb_sectors,
501 iscsilun->cluster_sectors) - cl_num_expanded;
502 /* shrink to touch only completely contained clusters */
503 cl_num_shrunk = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors);
504 nb_cls_shrunk = (sector_num + nb_sectors) / iscsilun->cluster_sectors
505 - cl_num_shrunk;
506 if (allocated) {
507 bitmap_set(iscsilun->allocmap, cl_num_expanded, nb_cls_expanded);
508 } else {
509 if (nb_cls_shrunk > 0) {
510 bitmap_clear(iscsilun->allocmap, cl_num_shrunk, nb_cls_shrunk);
511 }
512 }
513
514 if (iscsilun->allocmap_valid == NULL) {
515 return;
516 }
517 if (valid) {
518 if (nb_cls_shrunk > 0) {
519 bitmap_set(iscsilun->allocmap_valid, cl_num_shrunk, nb_cls_shrunk);
520 }
521 } else {
522 bitmap_clear(iscsilun->allocmap_valid, cl_num_expanded,
523 nb_cls_expanded);
524 }
525 }
526
527 static void
528 iscsi_allocmap_set_allocated(IscsiLun *iscsilun, int64_t sector_num,
529 int nb_sectors)
530 {
531 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, true, true);
532 }
533
534 static void
535 iscsi_allocmap_set_unallocated(IscsiLun *iscsilun, int64_t sector_num,
536 int nb_sectors)
537 {
538 /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update
539 * is ignored, so this will in effect be an iscsi_allocmap_set_invalid.
540 */
541 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, false, true);
542 }
543
544 static void iscsi_allocmap_set_invalid(IscsiLun *iscsilun, int64_t sector_num,
545 int nb_sectors)
546 {
547 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, false, false);
548 }
549
550 static void iscsi_allocmap_invalidate(IscsiLun *iscsilun)
551 {
552 if (iscsilun->allocmap) {
553 bitmap_zero(iscsilun->allocmap, iscsilun->allocmap_size);
554 }
555 if (iscsilun->allocmap_valid) {
556 bitmap_zero(iscsilun->allocmap_valid, iscsilun->allocmap_size);
557 }
558 }
559
560 static inline bool
561 iscsi_allocmap_is_allocated(IscsiLun *iscsilun, int64_t sector_num,
562 int nb_sectors)
563 {
564 unsigned long size;
565 if (iscsilun->allocmap == NULL) {
566 return true;
567 }
568 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
569 return !(find_next_bit(iscsilun->allocmap, size,
570 sector_num / iscsilun->cluster_sectors) == size);
571 }
572
573 static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun,
574 int64_t sector_num, int nb_sectors)
575 {
576 unsigned long size;
577 if (iscsilun->allocmap_valid == NULL) {
578 return false;
579 }
580 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
581 return (find_next_zero_bit(iscsilun->allocmap_valid, size,
582 sector_num / iscsilun->cluster_sectors) == size);
583 }
584
585 static int coroutine_fn
586 iscsi_co_writev_flags(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
587 QEMUIOVector *iov, int flags)
588 {
589 IscsiLun *iscsilun = bs->opaque;
590 struct IscsiTask iTask;
591 uint64_t lba;
592 uint32_t num_sectors;
593 bool fua = flags & BDRV_REQ_FUA;
594 int r = 0;
595
596 if (fua) {
597 assert(iscsilun->dpofua);
598 }
599 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
600 return -EINVAL;
601 }
602
603 if (bs->bl.max_transfer) {
604 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
605 }
606
607 lba = sector_qemu2lun(sector_num, iscsilun);
608 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
609 iscsi_co_init_iscsitask(iscsilun, &iTask);
610 qemu_mutex_lock(&iscsilun->mutex);
611 retry:
612 if (iscsilun->use_16_for_rw) {
613 #if LIBISCSI_API_VERSION >= (20160603)
614 iTask.task = iscsi_write16_iov_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 (struct scsi_iovec *)iov->iov, iov->niov);
619 } else {
620 iTask.task = iscsi_write10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
621 NULL, num_sectors * iscsilun->block_size,
622 iscsilun->block_size, 0, 0, fua, 0, 0,
623 iscsi_co_generic_cb, &iTask,
624 (struct scsi_iovec *)iov->iov, iov->niov);
625 }
626 #else
627 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
628 NULL, num_sectors * iscsilun->block_size,
629 iscsilun->block_size, 0, 0, fua, 0, 0,
630 iscsi_co_generic_cb, &iTask);
631 } else {
632 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
633 NULL, num_sectors * iscsilun->block_size,
634 iscsilun->block_size, 0, 0, fua, 0, 0,
635 iscsi_co_generic_cb, &iTask);
636 }
637 #endif
638 if (iTask.task == NULL) {
639 qemu_mutex_unlock(&iscsilun->mutex);
640 return -ENOMEM;
641 }
642 #if LIBISCSI_API_VERSION < (20160603)
643 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
644 iov->niov);
645 #endif
646 while (!iTask.complete) {
647 iscsi_set_events(iscsilun);
648 qemu_mutex_unlock(&iscsilun->mutex);
649 qemu_coroutine_yield();
650 qemu_mutex_lock(&iscsilun->mutex);
651 }
652
653 if (iTask.task != NULL) {
654 scsi_free_scsi_task(iTask.task);
655 iTask.task = NULL;
656 }
657
658 if (iTask.do_retry) {
659 iTask.complete = 0;
660 goto retry;
661 }
662
663 if (iTask.status != SCSI_STATUS_GOOD) {
664 iscsi_allocmap_set_invalid(iscsilun, sector_num, nb_sectors);
665 r = iTask.err_code;
666 goto out_unlock;
667 }
668
669 iscsi_allocmap_set_allocated(iscsilun, sector_num, nb_sectors);
670
671 out_unlock:
672 qemu_mutex_unlock(&iscsilun->mutex);
673 return r;
674 }
675
676
677
678 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
679 int64_t sector_num,
680 int nb_sectors, int *pnum,
681 BlockDriverState **file)
682 {
683 IscsiLun *iscsilun = bs->opaque;
684 struct scsi_get_lba_status *lbas = NULL;
685 struct scsi_lba_status_descriptor *lbasd = NULL;
686 struct IscsiTask iTask;
687 int64_t ret;
688
689 iscsi_co_init_iscsitask(iscsilun, &iTask);
690
691 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
692 ret = -EINVAL;
693 goto out;
694 }
695
696 /* default to all sectors allocated */
697 ret = BDRV_BLOCK_DATA;
698 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
699 *pnum = nb_sectors;
700
701 /* LUN does not support logical block provisioning */
702 if (!iscsilun->lbpme) {
703 goto out;
704 }
705
706 qemu_mutex_lock(&iscsilun->mutex);
707 retry:
708 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
709 sector_qemu2lun(sector_num, iscsilun),
710 8 + 16, iscsi_co_generic_cb,
711 &iTask) == NULL) {
712 ret = -ENOMEM;
713 goto out_unlock;
714 }
715
716 while (!iTask.complete) {
717 iscsi_set_events(iscsilun);
718 qemu_mutex_unlock(&iscsilun->mutex);
719 qemu_coroutine_yield();
720 qemu_mutex_lock(&iscsilun->mutex);
721 }
722
723 if (iTask.do_retry) {
724 if (iTask.task != NULL) {
725 scsi_free_scsi_task(iTask.task);
726 iTask.task = NULL;
727 }
728 iTask.complete = 0;
729 goto retry;
730 }
731
732 if (iTask.status != SCSI_STATUS_GOOD) {
733 /* in case the get_lba_status_callout fails (i.e.
734 * because the device is busy or the cmd is not
735 * supported) we pretend all blocks are allocated
736 * for backwards compatibility */
737 goto out_unlock;
738 }
739
740 lbas = scsi_datain_unmarshall(iTask.task);
741 if (lbas == NULL) {
742 ret = -EIO;
743 goto out_unlock;
744 }
745
746 lbasd = &lbas->descriptors[0];
747
748 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
749 ret = -EIO;
750 goto out_unlock;
751 }
752
753 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
754
755 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
756 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
757 ret &= ~BDRV_BLOCK_DATA;
758 if (iscsilun->lbprz) {
759 ret |= BDRV_BLOCK_ZERO;
760 }
761 }
762
763 if (ret & BDRV_BLOCK_ZERO) {
764 iscsi_allocmap_set_unallocated(iscsilun, sector_num, *pnum);
765 } else {
766 iscsi_allocmap_set_allocated(iscsilun, sector_num, *pnum);
767 }
768
769 if (*pnum > nb_sectors) {
770 *pnum = nb_sectors;
771 }
772 out_unlock:
773 qemu_mutex_unlock(&iscsilun->mutex);
774 out:
775 if (iTask.task != NULL) {
776 scsi_free_scsi_task(iTask.task);
777 }
778 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) {
779 *file = bs;
780 }
781 return ret;
782 }
783
784 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
785 int64_t sector_num, int nb_sectors,
786 QEMUIOVector *iov)
787 {
788 IscsiLun *iscsilun = bs->opaque;
789 struct IscsiTask iTask;
790 uint64_t lba;
791 uint32_t num_sectors;
792
793 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
794 return -EINVAL;
795 }
796
797 if (bs->bl.max_transfer) {
798 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
799 }
800
801 /* if cache.direct is off and we have a valid entry in our allocation map
802 * we can skip checking the block status and directly return zeroes if
803 * the request falls within an unallocated area */
804 if (iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) &&
805 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
806 qemu_iovec_memset(iov, 0, 0x00, iov->size);
807 return 0;
808 }
809
810 if (nb_sectors >= ISCSI_CHECKALLOC_THRES &&
811 !iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) &&
812 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
813 int pnum;
814 BlockDriverState *file;
815 /* check the block status from the beginning of the cluster
816 * containing the start sector */
817 int64_t ret = iscsi_co_get_block_status(bs,
818 sector_num - sector_num % iscsilun->cluster_sectors,
819 BDRV_REQUEST_MAX_SECTORS, &pnum, &file);
820 if (ret < 0) {
821 return ret;
822 }
823 /* if the whole request falls into an unallocated area we can avoid
824 * to read and directly return zeroes instead */
825 if (ret & BDRV_BLOCK_ZERO &&
826 pnum >= nb_sectors + sector_num % iscsilun->cluster_sectors) {
827 qemu_iovec_memset(iov, 0, 0x00, iov->size);
828 return 0;
829 }
830 }
831
832 lba = sector_qemu2lun(sector_num, iscsilun);
833 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
834
835 iscsi_co_init_iscsitask(iscsilun, &iTask);
836 qemu_mutex_lock(&iscsilun->mutex);
837 retry:
838 if (iscsilun->use_16_for_rw) {
839 #if LIBISCSI_API_VERSION >= (20160603)
840 iTask.task = iscsi_read16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
841 num_sectors * iscsilun->block_size,
842 iscsilun->block_size, 0, 0, 0, 0, 0,
843 iscsi_co_generic_cb, &iTask,
844 (struct scsi_iovec *)iov->iov, iov->niov);
845 } else {
846 iTask.task = iscsi_read10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
847 num_sectors * iscsilun->block_size,
848 iscsilun->block_size,
849 0, 0, 0, 0, 0,
850 iscsi_co_generic_cb, &iTask,
851 (struct scsi_iovec *)iov->iov, iov->niov);
852 }
853 #else
854 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
855 num_sectors * iscsilun->block_size,
856 iscsilun->block_size, 0, 0, 0, 0, 0,
857 iscsi_co_generic_cb, &iTask);
858 } else {
859 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
860 num_sectors * iscsilun->block_size,
861 iscsilun->block_size,
862 0, 0, 0, 0, 0,
863 iscsi_co_generic_cb, &iTask);
864 }
865 #endif
866 if (iTask.task == NULL) {
867 qemu_mutex_unlock(&iscsilun->mutex);
868 return -ENOMEM;
869 }
870 #if LIBISCSI_API_VERSION < (20160603)
871 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
872 #endif
873 while (!iTask.complete) {
874 iscsi_set_events(iscsilun);
875 qemu_mutex_unlock(&iscsilun->mutex);
876 qemu_coroutine_yield();
877 qemu_mutex_lock(&iscsilun->mutex);
878 }
879
880 if (iTask.task != NULL) {
881 scsi_free_scsi_task(iTask.task);
882 iTask.task = NULL;
883 }
884
885 if (iTask.do_retry) {
886 iTask.complete = 0;
887 goto retry;
888 }
889 qemu_mutex_unlock(&iscsilun->mutex);
890
891 if (iTask.status != SCSI_STATUS_GOOD) {
892 return iTask.err_code;
893 }
894
895 return 0;
896 }
897
898 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
899 {
900 IscsiLun *iscsilun = bs->opaque;
901 struct IscsiTask iTask;
902
903 iscsi_co_init_iscsitask(iscsilun, &iTask);
904 qemu_mutex_lock(&iscsilun->mutex);
905 retry:
906 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
907 0, iscsi_co_generic_cb, &iTask) == NULL) {
908 qemu_mutex_unlock(&iscsilun->mutex);
909 return -ENOMEM;
910 }
911
912 while (!iTask.complete) {
913 iscsi_set_events(iscsilun);
914 qemu_mutex_unlock(&iscsilun->mutex);
915 qemu_coroutine_yield();
916 qemu_mutex_lock(&iscsilun->mutex);
917 }
918
919 if (iTask.task != NULL) {
920 scsi_free_scsi_task(iTask.task);
921 iTask.task = NULL;
922 }
923
924 if (iTask.do_retry) {
925 iTask.complete = 0;
926 goto retry;
927 }
928 qemu_mutex_unlock(&iscsilun->mutex);
929
930 if (iTask.status != SCSI_STATUS_GOOD) {
931 return iTask.err_code;
932 }
933
934 return 0;
935 }
936
937 #ifdef __linux__
938 /* Called (via iscsi_service) with QemuMutex held. */
939 static void
940 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
941 void *command_data, void *opaque)
942 {
943 IscsiAIOCB *acb = opaque;
944
945 g_free(acb->buf);
946 acb->buf = NULL;
947
948 acb->status = 0;
949 if (status < 0) {
950 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
951 iscsi_get_error(iscsi));
952 acb->status = iscsi_translate_sense(&acb->task->sense);
953 }
954
955 acb->ioh->driver_status = 0;
956 acb->ioh->host_status = 0;
957 acb->ioh->resid = 0;
958 acb->ioh->status = status;
959
960 #define SG_ERR_DRIVER_SENSE 0x08
961
962 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
963 int ss;
964
965 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
966
967 acb->ioh->sb_len_wr = acb->task->datain.size - 2;
968 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
969 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
970 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
971 }
972
973 iscsi_schedule_bh(acb);
974 }
975
976 static void iscsi_ioctl_bh_completion(void *opaque)
977 {
978 IscsiAIOCB *acb = opaque;
979
980 qemu_bh_delete(acb->bh);
981 acb->common.cb(acb->common.opaque, acb->ret);
982 qemu_aio_unref(acb);
983 }
984
985 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
986 {
987 BlockDriverState *bs = acb->common.bs;
988 IscsiLun *iscsilun = bs->opaque;
989 int ret = 0;
990
991 switch (req) {
992 case SG_GET_VERSION_NUM:
993 *(int *)buf = 30000;
994 break;
995 case SG_GET_SCSI_ID:
996 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
997 break;
998 default:
999 ret = -EINVAL;
1000 }
1001 assert(!acb->bh);
1002 acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
1003 iscsi_ioctl_bh_completion, acb);
1004 acb->ret = ret;
1005 qemu_bh_schedule(acb->bh);
1006 }
1007
1008 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
1009 unsigned long int req, void *buf,
1010 BlockCompletionFunc *cb, void *opaque)
1011 {
1012 IscsiLun *iscsilun = bs->opaque;
1013 struct iscsi_context *iscsi = iscsilun->iscsi;
1014 struct iscsi_data data;
1015 IscsiAIOCB *acb;
1016
1017 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
1018
1019 acb->iscsilun = iscsilun;
1020 acb->bh = NULL;
1021 acb->status = -EINPROGRESS;
1022 acb->buf = NULL;
1023 acb->ioh = buf;
1024
1025 if (req != SG_IO) {
1026 iscsi_ioctl_handle_emulated(acb, req, buf);
1027 return &acb->common;
1028 }
1029
1030 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) {
1031 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)",
1032 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE);
1033 qemu_aio_unref(acb);
1034 return NULL;
1035 }
1036
1037 acb->task = malloc(sizeof(struct scsi_task));
1038 if (acb->task == NULL) {
1039 error_report("iSCSI: Failed to allocate task for scsi command. %s",
1040 iscsi_get_error(iscsi));
1041 qemu_aio_unref(acb);
1042 return NULL;
1043 }
1044 memset(acb->task, 0, sizeof(struct scsi_task));
1045
1046 switch (acb->ioh->dxfer_direction) {
1047 case SG_DXFER_TO_DEV:
1048 acb->task->xfer_dir = SCSI_XFER_WRITE;
1049 break;
1050 case SG_DXFER_FROM_DEV:
1051 acb->task->xfer_dir = SCSI_XFER_READ;
1052 break;
1053 default:
1054 acb->task->xfer_dir = SCSI_XFER_NONE;
1055 break;
1056 }
1057
1058 acb->task->cdb_size = acb->ioh->cmd_len;
1059 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
1060 acb->task->expxferlen = acb->ioh->dxfer_len;
1061
1062 data.size = 0;
1063 qemu_mutex_lock(&iscsilun->mutex);
1064 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
1065 if (acb->ioh->iovec_count == 0) {
1066 data.data = acb->ioh->dxferp;
1067 data.size = acb->ioh->dxfer_len;
1068 } else {
1069 scsi_task_set_iov_out(acb->task,
1070 (struct scsi_iovec *) acb->ioh->dxferp,
1071 acb->ioh->iovec_count);
1072 }
1073 }
1074
1075 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
1076 iscsi_aio_ioctl_cb,
1077 (data.size > 0) ? &data : NULL,
1078 acb) != 0) {
1079 qemu_mutex_unlock(&iscsilun->mutex);
1080 scsi_free_scsi_task(acb->task);
1081 qemu_aio_unref(acb);
1082 return NULL;
1083 }
1084
1085 /* tell libiscsi to read straight into the buffer we got from ioctl */
1086 if (acb->task->xfer_dir == SCSI_XFER_READ) {
1087 if (acb->ioh->iovec_count == 0) {
1088 scsi_task_add_data_in_buffer(acb->task,
1089 acb->ioh->dxfer_len,
1090 acb->ioh->dxferp);
1091 } else {
1092 scsi_task_set_iov_in(acb->task,
1093 (struct scsi_iovec *) acb->ioh->dxferp,
1094 acb->ioh->iovec_count);
1095 }
1096 }
1097
1098 iscsi_set_events(iscsilun);
1099 qemu_mutex_unlock(&iscsilun->mutex);
1100
1101 return &acb->common;
1102 }
1103
1104 #endif
1105
1106 static int64_t
1107 iscsi_getlength(BlockDriverState *bs)
1108 {
1109 IscsiLun *iscsilun = bs->opaque;
1110 int64_t len;
1111
1112 len = iscsilun->num_blocks;
1113 len *= iscsilun->block_size;
1114
1115 return len;
1116 }
1117
1118 static int
1119 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, int count)
1120 {
1121 IscsiLun *iscsilun = bs->opaque;
1122 struct IscsiTask iTask;
1123 struct unmap_list list;
1124 int r = 0;
1125
1126 if (!is_byte_request_lun_aligned(offset, count, iscsilun)) {
1127 return -ENOTSUP;
1128 }
1129
1130 if (!iscsilun->lbp.lbpu) {
1131 /* UNMAP is not supported by the target */
1132 return 0;
1133 }
1134
1135 list.lba = offset / iscsilun->block_size;
1136 list.num = count / iscsilun->block_size;
1137
1138 iscsi_co_init_iscsitask(iscsilun, &iTask);
1139 qemu_mutex_lock(&iscsilun->mutex);
1140 retry:
1141 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
1142 iscsi_co_generic_cb, &iTask) == NULL) {
1143 r = -ENOMEM;
1144 goto out_unlock;
1145 }
1146
1147 while (!iTask.complete) {
1148 iscsi_set_events(iscsilun);
1149 qemu_mutex_unlock(&iscsilun->mutex);
1150 qemu_coroutine_yield();
1151 qemu_mutex_lock(&iscsilun->mutex);
1152 }
1153
1154 if (iTask.task != NULL) {
1155 scsi_free_scsi_task(iTask.task);
1156 iTask.task = NULL;
1157 }
1158
1159 if (iTask.do_retry) {
1160 iTask.complete = 0;
1161 goto retry;
1162 }
1163
1164 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
1165 /* the target might fail with a check condition if it
1166 is not happy with the alignment of the UNMAP request
1167 we silently fail in this case */
1168 goto out_unlock;
1169 }
1170
1171 if (iTask.status != SCSI_STATUS_GOOD) {
1172 r = iTask.err_code;
1173 goto out_unlock;
1174 }
1175
1176 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1177 count >> BDRV_SECTOR_BITS);
1178
1179 out_unlock:
1180 qemu_mutex_unlock(&iscsilun->mutex);
1181 return r;
1182 }
1183
1184 static int
1185 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1186 int count, BdrvRequestFlags flags)
1187 {
1188 IscsiLun *iscsilun = bs->opaque;
1189 struct IscsiTask iTask;
1190 uint64_t lba;
1191 uint32_t nb_blocks;
1192 bool use_16_for_ws = iscsilun->use_16_for_rw;
1193 int r = 0;
1194
1195 if (!is_byte_request_lun_aligned(offset, count, iscsilun)) {
1196 return -ENOTSUP;
1197 }
1198
1199 if (flags & BDRV_REQ_MAY_UNMAP) {
1200 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
1201 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
1202 use_16_for_ws = true;
1203 }
1204 if (use_16_for_ws && !iscsilun->lbp.lbpws) {
1205 /* WRITESAME16 with UNMAP is not supported by the target,
1206 * fall back and try WRITESAME10/16 without UNMAP */
1207 flags &= ~BDRV_REQ_MAY_UNMAP;
1208 use_16_for_ws = iscsilun->use_16_for_rw;
1209 }
1210 }
1211
1212 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1213 /* WRITESAME without UNMAP is not supported by the target */
1214 return -ENOTSUP;
1215 }
1216
1217 lba = offset / iscsilun->block_size;
1218 nb_blocks = count / iscsilun->block_size;
1219
1220 if (iscsilun->zeroblock == NULL) {
1221 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1222 if (iscsilun->zeroblock == NULL) {
1223 return -ENOMEM;
1224 }
1225 }
1226
1227 qemu_mutex_lock(&iscsilun->mutex);
1228 iscsi_co_init_iscsitask(iscsilun, &iTask);
1229 retry:
1230 if (use_16_for_ws) {
1231 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1232 iscsilun->zeroblock, iscsilun->block_size,
1233 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1234 0, 0, iscsi_co_generic_cb, &iTask);
1235 } else {
1236 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1237 iscsilun->zeroblock, iscsilun->block_size,
1238 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1239 0, 0, iscsi_co_generic_cb, &iTask);
1240 }
1241 if (iTask.task == NULL) {
1242 qemu_mutex_unlock(&iscsilun->mutex);
1243 return -ENOMEM;
1244 }
1245
1246 while (!iTask.complete) {
1247 iscsi_set_events(iscsilun);
1248 qemu_mutex_unlock(&iscsilun->mutex);
1249 qemu_coroutine_yield();
1250 qemu_mutex_lock(&iscsilun->mutex);
1251 }
1252
1253 if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1254 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1255 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1256 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1257 /* WRITE SAME is not supported by the target */
1258 iscsilun->has_write_same = false;
1259 scsi_free_scsi_task(iTask.task);
1260 r = -ENOTSUP;
1261 goto out_unlock;
1262 }
1263
1264 if (iTask.task != NULL) {
1265 scsi_free_scsi_task(iTask.task);
1266 iTask.task = NULL;
1267 }
1268
1269 if (iTask.do_retry) {
1270 iTask.complete = 0;
1271 goto retry;
1272 }
1273
1274 if (iTask.status != SCSI_STATUS_GOOD) {
1275 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1276 count >> BDRV_SECTOR_BITS);
1277 r = iTask.err_code;
1278 goto out_unlock;
1279 }
1280
1281 if (flags & BDRV_REQ_MAY_UNMAP) {
1282 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1283 count >> BDRV_SECTOR_BITS);
1284 } else {
1285 iscsi_allocmap_set_allocated(iscsilun, offset >> BDRV_SECTOR_BITS,
1286 count >> BDRV_SECTOR_BITS);
1287 }
1288
1289 out_unlock:
1290 qemu_mutex_unlock(&iscsilun->mutex);
1291 return r;
1292 }
1293
1294 static void apply_chap(struct iscsi_context *iscsi, QemuOpts *opts,
1295 Error **errp)
1296 {
1297 const char *user = NULL;
1298 const char *password = NULL;
1299 const char *secretid;
1300 char *secret = NULL;
1301
1302 user = qemu_opt_get(opts, "user");
1303 if (!user) {
1304 return;
1305 }
1306
1307 secretid = qemu_opt_get(opts, "password-secret");
1308 password = qemu_opt_get(opts, "password");
1309 if (secretid && password) {
1310 error_setg(errp, "'password' and 'password-secret' properties are "
1311 "mutually exclusive");
1312 return;
1313 }
1314 if (secretid) {
1315 secret = qcrypto_secret_lookup_as_utf8(secretid, errp);
1316 if (!secret) {
1317 return;
1318 }
1319 password = secret;
1320 } else if (!password) {
1321 error_setg(errp, "CHAP username specified but no password was given");
1322 return;
1323 }
1324
1325 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1326 error_setg(errp, "Failed to set initiator username and password");
1327 }
1328
1329 g_free(secret);
1330 }
1331
1332 static void apply_header_digest(struct iscsi_context *iscsi, QemuOpts *opts,
1333 Error **errp)
1334 {
1335 const char *digest = NULL;
1336
1337 digest = qemu_opt_get(opts, "header-digest");
1338 if (!digest) {
1339 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1340 } else if (!strcmp(digest, "crc32c")) {
1341 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1342 } else if (!strcmp(digest, "none")) {
1343 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1344 } else if (!strcmp(digest, "crc32c-none")) {
1345 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1346 } else if (!strcmp(digest, "none-crc32c")) {
1347 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1348 } else {
1349 error_setg(errp, "Invalid header-digest setting : %s", digest);
1350 }
1351 }
1352
1353 static char *get_initiator_name(QemuOpts *opts)
1354 {
1355 const char *name;
1356 char *iscsi_name;
1357 UuidInfo *uuid_info;
1358
1359 name = qemu_opt_get(opts, "initiator-name");
1360 if (name) {
1361 return g_strdup(name);
1362 }
1363
1364 uuid_info = qmp_query_uuid(NULL);
1365 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1366 name = qemu_get_vm_name();
1367 } else {
1368 name = uuid_info->UUID;
1369 }
1370 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1371 name ? ":" : "", name ? name : "");
1372 qapi_free_UuidInfo(uuid_info);
1373 return iscsi_name;
1374 }
1375
1376 static void iscsi_nop_timed_event(void *opaque)
1377 {
1378 IscsiLun *iscsilun = opaque;
1379
1380 qemu_mutex_lock(&iscsilun->mutex);
1381 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1382 error_report("iSCSI: NOP timeout. Reconnecting...");
1383 iscsilun->request_timed_out = true;
1384 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1385 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1386 goto out;
1387 }
1388
1389 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1390 iscsi_set_events(iscsilun);
1391
1392 out:
1393 qemu_mutex_unlock(&iscsilun->mutex);
1394 }
1395
1396 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1397 {
1398 struct scsi_task *task = NULL;
1399 struct scsi_readcapacity10 *rc10 = NULL;
1400 struct scsi_readcapacity16 *rc16 = NULL;
1401 int retries = ISCSI_CMD_RETRIES;
1402
1403 do {
1404 if (task != NULL) {
1405 scsi_free_scsi_task(task);
1406 task = NULL;
1407 }
1408
1409 switch (iscsilun->type) {
1410 case TYPE_DISK:
1411 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1412 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1413 rc16 = scsi_datain_unmarshall(task);
1414 if (rc16 == NULL) {
1415 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1416 } else {
1417 iscsilun->block_size = rc16->block_length;
1418 iscsilun->num_blocks = rc16->returned_lba + 1;
1419 iscsilun->lbpme = !!rc16->lbpme;
1420 iscsilun->lbprz = !!rc16->lbprz;
1421 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1422 }
1423 break;
1424 }
1425 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1426 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1427 break;
1428 }
1429 /* Fall through and try READ CAPACITY(10) instead. */
1430 case TYPE_ROM:
1431 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1432 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1433 rc10 = scsi_datain_unmarshall(task);
1434 if (rc10 == NULL) {
1435 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1436 } else {
1437 iscsilun->block_size = rc10->block_size;
1438 if (rc10->lba == 0) {
1439 /* blank disk loaded */
1440 iscsilun->num_blocks = 0;
1441 } else {
1442 iscsilun->num_blocks = rc10->lba + 1;
1443 }
1444 }
1445 }
1446 break;
1447 default:
1448 return;
1449 }
1450 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1451 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1452 && retries-- > 0);
1453
1454 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1455 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1456 } else if (!iscsilun->block_size ||
1457 iscsilun->block_size % BDRV_SECTOR_SIZE) {
1458 error_setg(errp, "iSCSI: the target returned an invalid "
1459 "block size of %d.", iscsilun->block_size);
1460 }
1461 if (task) {
1462 scsi_free_scsi_task(task);
1463 }
1464 }
1465
1466 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1467 int evpd, int pc, void **inq, Error **errp)
1468 {
1469 int full_size;
1470 struct scsi_task *task = NULL;
1471 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1472 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1473 goto fail;
1474 }
1475 full_size = scsi_datain_getfullsize(task);
1476 if (full_size > task->datain.size) {
1477 scsi_free_scsi_task(task);
1478
1479 /* we need more data for the full list */
1480 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1481 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1482 goto fail;
1483 }
1484 }
1485
1486 *inq = scsi_datain_unmarshall(task);
1487 if (*inq == NULL) {
1488 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1489 goto fail_with_err;
1490 }
1491
1492 return task;
1493
1494 fail:
1495 error_setg(errp, "iSCSI: Inquiry command failed : %s",
1496 iscsi_get_error(iscsi));
1497 fail_with_err:
1498 if (task != NULL) {
1499 scsi_free_scsi_task(task);
1500 }
1501 return NULL;
1502 }
1503
1504 static void iscsi_detach_aio_context(BlockDriverState *bs)
1505 {
1506 IscsiLun *iscsilun = bs->opaque;
1507
1508 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1509 false, NULL, NULL, NULL, NULL);
1510 iscsilun->events = 0;
1511
1512 if (iscsilun->nop_timer) {
1513 timer_del(iscsilun->nop_timer);
1514 timer_free(iscsilun->nop_timer);
1515 iscsilun->nop_timer = NULL;
1516 }
1517 if (iscsilun->event_timer) {
1518 timer_del(iscsilun->event_timer);
1519 timer_free(iscsilun->event_timer);
1520 iscsilun->event_timer = NULL;
1521 }
1522 }
1523
1524 static void iscsi_attach_aio_context(BlockDriverState *bs,
1525 AioContext *new_context)
1526 {
1527 IscsiLun *iscsilun = bs->opaque;
1528
1529 iscsilun->aio_context = new_context;
1530 iscsi_set_events(iscsilun);
1531
1532 /* Set up a timer for sending out iSCSI NOPs */
1533 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1534 QEMU_CLOCK_REALTIME, SCALE_MS,
1535 iscsi_nop_timed_event, iscsilun);
1536 timer_mod(iscsilun->nop_timer,
1537 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1538
1539 /* Set up a timer for periodic calls to iscsi_set_events and to
1540 * scan for command timeout */
1541 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1542 QEMU_CLOCK_REALTIME, SCALE_MS,
1543 iscsi_timed_check_events, iscsilun);
1544 timer_mod(iscsilun->event_timer,
1545 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1546 }
1547
1548 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1549 {
1550 struct scsi_task *task;
1551 struct scsi_mode_sense *ms = NULL;
1552 iscsilun->write_protected = false;
1553 iscsilun->dpofua = false;
1554
1555 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1556 1, SCSI_MODESENSE_PC_CURRENT,
1557 0x3F, 0, 255);
1558 if (task == NULL) {
1559 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1560 iscsi_get_error(iscsilun->iscsi));
1561 goto out;
1562 }
1563
1564 if (task->status != SCSI_STATUS_GOOD) {
1565 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1566 goto out;
1567 }
1568 ms = scsi_datain_unmarshall(task);
1569 if (!ms) {
1570 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1571 iscsi_get_error(iscsilun->iscsi));
1572 goto out;
1573 }
1574 iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1575 iscsilun->dpofua = ms->device_specific_parameter & 0x10;
1576
1577 out:
1578 if (task) {
1579 scsi_free_scsi_task(task);
1580 }
1581 }
1582
1583 static void iscsi_parse_iscsi_option(const char *target, QDict *options)
1584 {
1585 QemuOptsList *list;
1586 QemuOpts *opts;
1587 const char *user, *password, *password_secret, *initiator_name,
1588 *header_digest, *timeout;
1589
1590 list = qemu_find_opts("iscsi");
1591 if (!list) {
1592 return;
1593 }
1594
1595 opts = qemu_opts_find(list, target);
1596 if (opts == NULL) {
1597 opts = QTAILQ_FIRST(&list->head);
1598 if (!opts) {
1599 return;
1600 }
1601 }
1602
1603 user = qemu_opt_get(opts, "user");
1604 if (user) {
1605 qdict_set_default_str(options, "user", user);
1606 }
1607
1608 password = qemu_opt_get(opts, "password");
1609 if (password) {
1610 qdict_set_default_str(options, "password", password);
1611 }
1612
1613 password_secret = qemu_opt_get(opts, "password-secret");
1614 if (password_secret) {
1615 qdict_set_default_str(options, "password-secret", password_secret);
1616 }
1617
1618 initiator_name = qemu_opt_get(opts, "initiator-name");
1619 if (initiator_name) {
1620 qdict_set_default_str(options, "initiator-name", initiator_name);
1621 }
1622
1623 header_digest = qemu_opt_get(opts, "header-digest");
1624 if (header_digest) {
1625 /* -iscsi takes upper case values, but QAPI only supports lower case
1626 * enum constant names, so we have to convert here. */
1627 char *qapi_value = g_ascii_strdown(header_digest, -1);
1628 qdict_set_default_str(options, "header-digest", qapi_value);
1629 g_free(qapi_value);
1630 }
1631
1632 timeout = qemu_opt_get(opts, "timeout");
1633 if (timeout) {
1634 qdict_set_default_str(options, "timeout", timeout);
1635 }
1636 }
1637
1638 /*
1639 * We support iscsi url's on the form
1640 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1641 */
1642 static void iscsi_parse_filename(const char *filename, QDict *options,
1643 Error **errp)
1644 {
1645 struct iscsi_url *iscsi_url;
1646 const char *transport_name;
1647 char *lun_str;
1648
1649 iscsi_url = iscsi_parse_full_url(NULL, filename);
1650 if (iscsi_url == NULL) {
1651 error_setg(errp, "Failed to parse URL : %s", filename);
1652 return;
1653 }
1654
1655 #if LIBISCSI_API_VERSION >= (20160603)
1656 switch (iscsi_url->transport) {
1657 case TCP_TRANSPORT:
1658 transport_name = "tcp";
1659 break;
1660 case ISER_TRANSPORT:
1661 transport_name = "iser";
1662 break;
1663 default:
1664 error_setg(errp, "Unknown transport type (%d)",
1665 iscsi_url->transport);
1666 return;
1667 }
1668 #else
1669 transport_name = "tcp";
1670 #endif
1671
1672 qdict_set_default_str(options, "transport", transport_name);
1673 qdict_set_default_str(options, "portal", iscsi_url->portal);
1674 qdict_set_default_str(options, "target", iscsi_url->target);
1675
1676 lun_str = g_strdup_printf("%d", iscsi_url->lun);
1677 qdict_set_default_str(options, "lun", lun_str);
1678 g_free(lun_str);
1679
1680 /* User/password from -iscsi take precedence over those from the URL */
1681 iscsi_parse_iscsi_option(iscsi_url->target, options);
1682
1683 if (iscsi_url->user[0] != '\0') {
1684 qdict_set_default_str(options, "user", iscsi_url->user);
1685 qdict_set_default_str(options, "password", iscsi_url->passwd);
1686 }
1687
1688 iscsi_destroy_url(iscsi_url);
1689 }
1690
1691 static QemuOptsList runtime_opts = {
1692 .name = "iscsi",
1693 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1694 .desc = {
1695 {
1696 .name = "transport",
1697 .type = QEMU_OPT_STRING,
1698 },
1699 {
1700 .name = "portal",
1701 .type = QEMU_OPT_STRING,
1702 },
1703 {
1704 .name = "target",
1705 .type = QEMU_OPT_STRING,
1706 },
1707 {
1708 .name = "user",
1709 .type = QEMU_OPT_STRING,
1710 },
1711 {
1712 .name = "password",
1713 .type = QEMU_OPT_STRING,
1714 },
1715 {
1716 .name = "password-secret",
1717 .type = QEMU_OPT_STRING,
1718 },
1719 {
1720 .name = "lun",
1721 .type = QEMU_OPT_NUMBER,
1722 },
1723 {
1724 .name = "initiator-name",
1725 .type = QEMU_OPT_STRING,
1726 },
1727 {
1728 .name = "header-digest",
1729 .type = QEMU_OPT_STRING,
1730 },
1731 {
1732 .name = "timeout",
1733 .type = QEMU_OPT_NUMBER,
1734 },
1735 { /* end of list */ }
1736 },
1737 };
1738
1739 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1740 Error **errp)
1741 {
1742 IscsiLun *iscsilun = bs->opaque;
1743 struct iscsi_context *iscsi = NULL;
1744 struct scsi_task *task = NULL;
1745 struct scsi_inquiry_standard *inq = NULL;
1746 struct scsi_inquiry_supported_pages *inq_vpd;
1747 char *initiator_name = NULL;
1748 QemuOpts *opts;
1749 Error *local_err = NULL;
1750 const char *transport_name, *portal, *target;
1751 #if LIBISCSI_API_VERSION >= (20160603)
1752 enum iscsi_transport_type transport;
1753 #endif
1754 int i, ret = 0, timeout = 0, lun;
1755
1756 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1757 qemu_opts_absorb_qdict(opts, options, &local_err);
1758 if (local_err) {
1759 error_propagate(errp, local_err);
1760 ret = -EINVAL;
1761 goto out;
1762 }
1763
1764 transport_name = qemu_opt_get(opts, "transport");
1765 portal = qemu_opt_get(opts, "portal");
1766 target = qemu_opt_get(opts, "target");
1767 lun = qemu_opt_get_number(opts, "lun", 0);
1768
1769 if (!transport_name || !portal || !target) {
1770 error_setg(errp, "Need all of transport, portal and target options");
1771 ret = -EINVAL;
1772 goto out;
1773 }
1774
1775 if (!strcmp(transport_name, "tcp")) {
1776 #if LIBISCSI_API_VERSION >= (20160603)
1777 transport = TCP_TRANSPORT;
1778 } else if (!strcmp(transport_name, "iser")) {
1779 transport = ISER_TRANSPORT;
1780 #else
1781 /* TCP is what older libiscsi versions always use */
1782 #endif
1783 } else {
1784 error_setg(errp, "Unknown transport: %s", transport_name);
1785 ret = -EINVAL;
1786 goto out;
1787 }
1788
1789 memset(iscsilun, 0, sizeof(IscsiLun));
1790
1791 initiator_name = get_initiator_name(opts);
1792
1793 iscsi = iscsi_create_context(initiator_name);
1794 if (iscsi == NULL) {
1795 error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1796 ret = -ENOMEM;
1797 goto out;
1798 }
1799 #if LIBISCSI_API_VERSION >= (20160603)
1800 if (iscsi_init_transport(iscsi, transport)) {
1801 error_setg(errp, ("Error initializing transport."));
1802 ret = -EINVAL;
1803 goto out;
1804 }
1805 #endif
1806 if (iscsi_set_targetname(iscsi, target)) {
1807 error_setg(errp, "iSCSI: Failed to set target name.");
1808 ret = -EINVAL;
1809 goto out;
1810 }
1811
1812 /* check if we got CHAP username/password via the options */
1813 apply_chap(iscsi, opts, &local_err);
1814 if (local_err != NULL) {
1815 error_propagate(errp, local_err);
1816 ret = -EINVAL;
1817 goto out;
1818 }
1819
1820 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1821 error_setg(errp, "iSCSI: Failed to set session type to normal.");
1822 ret = -EINVAL;
1823 goto out;
1824 }
1825
1826 /* check if we got HEADER_DIGEST via the options */
1827 apply_header_digest(iscsi, opts, &local_err);
1828 if (local_err != NULL) {
1829 error_propagate(errp, local_err);
1830 ret = -EINVAL;
1831 goto out;
1832 }
1833
1834 /* timeout handling is broken in libiscsi before 1.15.0 */
1835 timeout = qemu_opt_get_number(opts, "timeout", 0);
1836 #if LIBISCSI_API_VERSION >= 20150621
1837 iscsi_set_timeout(iscsi, timeout);
1838 #else
1839 if (timeout) {
1840 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1841 }
1842 #endif
1843
1844 if (iscsi_full_connect_sync(iscsi, portal, lun) != 0) {
1845 error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1846 iscsi_get_error(iscsi));
1847 ret = -EINVAL;
1848 goto out;
1849 }
1850
1851 iscsilun->iscsi = iscsi;
1852 iscsilun->aio_context = bdrv_get_aio_context(bs);
1853 iscsilun->lun = lun;
1854 iscsilun->has_write_same = true;
1855
1856 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1857 (void **) &inq, errp);
1858 if (task == NULL) {
1859 ret = -EINVAL;
1860 goto out;
1861 }
1862 iscsilun->type = inq->periperal_device_type;
1863 scsi_free_scsi_task(task);
1864 task = NULL;
1865
1866 iscsi_modesense_sync(iscsilun);
1867 if (iscsilun->dpofua) {
1868 bs->supported_write_flags = BDRV_REQ_FUA;
1869 }
1870 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
1871
1872 /* Check the write protect flag of the LUN if we want to write */
1873 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1874 iscsilun->write_protected) {
1875 error_setg(errp, "Cannot open a write protected LUN as read-write");
1876 ret = -EACCES;
1877 goto out;
1878 }
1879
1880 iscsi_readcapacity_sync(iscsilun, &local_err);
1881 if (local_err != NULL) {
1882 error_propagate(errp, local_err);
1883 ret = -EINVAL;
1884 goto out;
1885 }
1886 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1887
1888 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1889 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1890 * will try to read from the device to guess the image format.
1891 */
1892 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1893 bs->sg = true;
1894 }
1895
1896 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1897 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1898 (void **) &inq_vpd, errp);
1899 if (task == NULL) {
1900 ret = -EINVAL;
1901 goto out;
1902 }
1903 for (i = 0; i < inq_vpd->num_pages; i++) {
1904 struct scsi_task *inq_task;
1905 struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1906 struct scsi_inquiry_block_limits *inq_bl;
1907 switch (inq_vpd->pages[i]) {
1908 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1909 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1910 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1911 (void **) &inq_lbp, errp);
1912 if (inq_task == NULL) {
1913 ret = -EINVAL;
1914 goto out;
1915 }
1916 memcpy(&iscsilun->lbp, inq_lbp,
1917 sizeof(struct scsi_inquiry_logical_block_provisioning));
1918 scsi_free_scsi_task(inq_task);
1919 break;
1920 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1921 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1922 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1923 (void **) &inq_bl, errp);
1924 if (inq_task == NULL) {
1925 ret = -EINVAL;
1926 goto out;
1927 }
1928 memcpy(&iscsilun->bl, inq_bl,
1929 sizeof(struct scsi_inquiry_block_limits));
1930 scsi_free_scsi_task(inq_task);
1931 break;
1932 default:
1933 break;
1934 }
1935 }
1936 scsi_free_scsi_task(task);
1937 task = NULL;
1938
1939 qemu_mutex_init(&iscsilun->mutex);
1940 iscsi_attach_aio_context(bs, iscsilun->aio_context);
1941
1942 /* Guess the internal cluster (page) size of the iscsi target by the means
1943 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1944 * reasonable size */
1945 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
1946 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1947 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran *
1948 iscsilun->block_size) >> BDRV_SECTOR_BITS;
1949 if (iscsilun->lbprz) {
1950 ret = iscsi_allocmap_init(iscsilun, bs->open_flags);
1951 }
1952 }
1953
1954 out:
1955 qemu_opts_del(opts);
1956 g_free(initiator_name);
1957 if (task != NULL) {
1958 scsi_free_scsi_task(task);
1959 }
1960
1961 if (ret) {
1962 if (iscsi != NULL) {
1963 if (iscsi_is_logged_in(iscsi)) {
1964 iscsi_logout_sync(iscsi);
1965 }
1966 iscsi_destroy_context(iscsi);
1967 }
1968 memset(iscsilun, 0, sizeof(IscsiLun));
1969 }
1970 return ret;
1971 }
1972
1973 static void iscsi_close(BlockDriverState *bs)
1974 {
1975 IscsiLun *iscsilun = bs->opaque;
1976 struct iscsi_context *iscsi = iscsilun->iscsi;
1977
1978 iscsi_detach_aio_context(bs);
1979 if (iscsi_is_logged_in(iscsi)) {
1980 iscsi_logout_sync(iscsi);
1981 }
1982 iscsi_destroy_context(iscsi);
1983 g_free(iscsilun->zeroblock);
1984 iscsi_allocmap_free(iscsilun);
1985 qemu_mutex_destroy(&iscsilun->mutex);
1986 memset(iscsilun, 0, sizeof(IscsiLun));
1987 }
1988
1989 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
1990 {
1991 /* We don't actually refresh here, but just return data queried in
1992 * iscsi_open(): iscsi targets don't change their limits. */
1993
1994 IscsiLun *iscsilun = bs->opaque;
1995 uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
1996 unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size);
1997
1998 assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bs->sg);
1999
2000 bs->bl.request_alignment = block_size;
2001
2002 if (iscsilun->bl.max_xfer_len) {
2003 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
2004 }
2005
2006 if (max_xfer_len * block_size < INT_MAX) {
2007 bs->bl.max_transfer = max_xfer_len * iscsilun->block_size;
2008 }
2009
2010 if (iscsilun->lbp.lbpu) {
2011 if (iscsilun->bl.max_unmap < 0xffffffff / block_size) {
2012 bs->bl.max_pdiscard =
2013 iscsilun->bl.max_unmap * iscsilun->block_size;
2014 }
2015 bs->bl.pdiscard_alignment =
2016 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2017 } else {
2018 bs->bl.pdiscard_alignment = iscsilun->block_size;
2019 }
2020
2021 if (iscsilun->bl.max_ws_len < 0xffffffff / block_size) {
2022 bs->bl.max_pwrite_zeroes =
2023 iscsilun->bl.max_ws_len * iscsilun->block_size;
2024 }
2025 if (iscsilun->lbp.lbpws) {
2026 bs->bl.pwrite_zeroes_alignment =
2027 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2028 } else {
2029 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size;
2030 }
2031 if (iscsilun->bl.opt_xfer_len &&
2032 iscsilun->bl.opt_xfer_len < INT_MAX / block_size) {
2033 bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len *
2034 iscsilun->block_size);
2035 }
2036 }
2037
2038 /* Note that this will not re-establish a connection with an iSCSI target - it
2039 * is effectively a NOP. */
2040 static int iscsi_reopen_prepare(BDRVReopenState *state,
2041 BlockReopenQueue *queue, Error **errp)
2042 {
2043 IscsiLun *iscsilun = state->bs->opaque;
2044
2045 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
2046 error_setg(errp, "Cannot open a write protected LUN as read-write");
2047 return -EACCES;
2048 }
2049 return 0;
2050 }
2051
2052 static void iscsi_reopen_commit(BDRVReopenState *reopen_state)
2053 {
2054 IscsiLun *iscsilun = reopen_state->bs->opaque;
2055
2056 /* the cache.direct status might have changed */
2057 if (iscsilun->allocmap != NULL) {
2058 iscsi_allocmap_init(iscsilun, reopen_state->flags);
2059 }
2060 }
2061
2062 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
2063 {
2064 IscsiLun *iscsilun = bs->opaque;
2065 Error *local_err = NULL;
2066
2067 if (iscsilun->type != TYPE_DISK) {
2068 return -ENOTSUP;
2069 }
2070
2071 iscsi_readcapacity_sync(iscsilun, &local_err);
2072 if (local_err != NULL) {
2073 error_free(local_err);
2074 return -EIO;
2075 }
2076
2077 if (offset > iscsi_getlength(bs)) {
2078 return -EINVAL;
2079 }
2080
2081 if (iscsilun->allocmap != NULL) {
2082 iscsi_allocmap_init(iscsilun, bs->open_flags);
2083 }
2084
2085 return 0;
2086 }
2087
2088 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp)
2089 {
2090 int ret = 0;
2091 int64_t total_size = 0;
2092 BlockDriverState *bs;
2093 IscsiLun *iscsilun = NULL;
2094 QDict *bs_options;
2095 Error *local_err = NULL;
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 iscsi_parse_filename(filename, bs_options, &local_err);
2107 if (local_err) {
2108 error_propagate(errp, local_err);
2109 ret = -EINVAL;
2110 } else {
2111 ret = iscsi_open(bs, bs_options, 0, NULL);
2112 }
2113 QDECREF(bs_options);
2114
2115 if (ret != 0) {
2116 goto out;
2117 }
2118 iscsi_detach_aio_context(bs);
2119 if (iscsilun->type != TYPE_DISK) {
2120 ret = -ENODEV;
2121 goto out;
2122 }
2123 if (bs->total_sectors < total_size) {
2124 ret = -ENOSPC;
2125 goto out;
2126 }
2127
2128 ret = 0;
2129 out:
2130 if (iscsilun->iscsi != NULL) {
2131 iscsi_destroy_context(iscsilun->iscsi);
2132 }
2133 g_free(bs->opaque);
2134 bs->opaque = NULL;
2135 bdrv_unref(bs);
2136 return ret;
2137 }
2138
2139 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2140 {
2141 IscsiLun *iscsilun = bs->opaque;
2142 bdi->unallocated_blocks_are_zero = iscsilun->lbprz;
2143 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
2144 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE;
2145 return 0;
2146 }
2147
2148 static void iscsi_invalidate_cache(BlockDriverState *bs,
2149 Error **errp)
2150 {
2151 IscsiLun *iscsilun = bs->opaque;
2152 iscsi_allocmap_invalidate(iscsilun);
2153 }
2154
2155 static QemuOptsList iscsi_create_opts = {
2156 .name = "iscsi-create-opts",
2157 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
2158 .desc = {
2159 {
2160 .name = BLOCK_OPT_SIZE,
2161 .type = QEMU_OPT_SIZE,
2162 .help = "Virtual disk size"
2163 },
2164 { /* end of list */ }
2165 }
2166 };
2167
2168 static BlockDriver bdrv_iscsi = {
2169 .format_name = "iscsi",
2170 .protocol_name = "iscsi",
2171
2172 .instance_size = sizeof(IscsiLun),
2173 .bdrv_parse_filename = iscsi_parse_filename,
2174 .bdrv_file_open = iscsi_open,
2175 .bdrv_close = iscsi_close,
2176 .bdrv_create = iscsi_create,
2177 .create_opts = &iscsi_create_opts,
2178 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2179 .bdrv_reopen_commit = iscsi_reopen_commit,
2180 .bdrv_invalidate_cache = iscsi_invalidate_cache,
2181
2182 .bdrv_getlength = iscsi_getlength,
2183 .bdrv_get_info = iscsi_get_info,
2184 .bdrv_truncate = iscsi_truncate,
2185 .bdrv_refresh_limits = iscsi_refresh_limits,
2186
2187 .bdrv_co_get_block_status = iscsi_co_get_block_status,
2188 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2189 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2190 .bdrv_co_readv = iscsi_co_readv,
2191 .bdrv_co_writev_flags = iscsi_co_writev_flags,
2192 .bdrv_co_flush_to_disk = iscsi_co_flush,
2193
2194 #ifdef __linux__
2195 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2196 #endif
2197
2198 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2199 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2200 };
2201
2202 #if LIBISCSI_API_VERSION >= (20160603)
2203 static BlockDriver bdrv_iser = {
2204 .format_name = "iser",
2205 .protocol_name = "iser",
2206
2207 .instance_size = sizeof(IscsiLun),
2208 .bdrv_parse_filename = iscsi_parse_filename,
2209 .bdrv_file_open = iscsi_open,
2210 .bdrv_close = iscsi_close,
2211 .bdrv_create = iscsi_create,
2212 .create_opts = &iscsi_create_opts,
2213 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2214 .bdrv_reopen_commit = iscsi_reopen_commit,
2215 .bdrv_invalidate_cache = iscsi_invalidate_cache,
2216
2217 .bdrv_getlength = iscsi_getlength,
2218 .bdrv_get_info = iscsi_get_info,
2219 .bdrv_truncate = iscsi_truncate,
2220 .bdrv_refresh_limits = iscsi_refresh_limits,
2221
2222 .bdrv_co_get_block_status = iscsi_co_get_block_status,
2223 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2224 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2225 .bdrv_co_readv = iscsi_co_readv,
2226 .bdrv_co_writev_flags = iscsi_co_writev_flags,
2227 .bdrv_co_flush_to_disk = iscsi_co_flush,
2228
2229 #ifdef __linux__
2230 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2231 #endif
2232
2233 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2234 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2235 };
2236 #endif
2237
2238 static void iscsi_block_init(void)
2239 {
2240 bdrv_register(&bdrv_iscsi);
2241 #if LIBISCSI_API_VERSION >= (20160603)
2242 bdrv_register(&bdrv_iser);
2243 #endif
2244 }
2245
2246 block_init(iscsi_block_init);