?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
block: Introduce bdrv_replace_child()
[qemu.git] / block / iscsi.c
blob2ca8e72967f8b9016ed322ad76621cb8d3b4fb3c
1 /*
2 * QEMU Block driver for iSCSI images
3 *
4 * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
5 * Copyright (c) 2012-2015 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 "sysemu/sysemu.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 #include <block/scsi.h>
50 #endif
51
52 typedef struct IscsiLun {
53 struct iscsi_context *iscsi;
54 AioContext *aio_context;
55 int lun;
56 enum scsi_inquiry_peripheral_device_type type;
57 int block_size;
58 uint64_t num_blocks;
59 int events;
60 QEMUTimer *nop_timer;
61 QEMUTimer *event_timer;
62 struct scsi_inquiry_logical_block_provisioning lbp;
63 struct scsi_inquiry_block_limits bl;
64 unsigned char *zeroblock;
65 unsigned long *allocationmap;
66 int cluster_sectors;
67 bool use_16_for_rw;
68 bool write_protected;
69 bool lbpme;
70 bool lbprz;
71 bool dpofua;
72 bool has_write_same;
73 bool request_timed_out;
74 } IscsiLun;
75
76 typedef struct IscsiTask {
77 int status;
78 int complete;
79 int retries;
80 int do_retry;
81 struct scsi_task *task;
82 Coroutine *co;
83 QEMUBH *bh;
84 IscsiLun *iscsilun;
85 QEMUTimer retry_timer;
86 int err_code;
87 } IscsiTask;
88
89 typedef struct IscsiAIOCB {
90 BlockAIOCB common;
91 QEMUIOVector *qiov;
92 QEMUBH *bh;
93 IscsiLun *iscsilun;
94 struct scsi_task *task;
95 uint8_t *buf;
96 int status;
97 int64_t sector_num;
98 int nb_sectors;
99 int ret;
100 #ifdef __linux__
101 sg_io_hdr_t *ioh;
102 #endif
103 } IscsiAIOCB;
104
105 /* libiscsi uses time_t so its enough to process events every second */
106 #define EVENT_INTERVAL 1000
107 #define NOP_INTERVAL 5000
108 #define MAX_NOP_FAILURES 3
109 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times)
110 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768};
111
112 /* this threshold is a trade-off knob to choose between
113 * the potential additional overhead of an extra GET_LBA_STATUS request
114 * vs. unnecessarily reading a lot of zero sectors over the wire.
115 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
116 * sectors we check the allocation status of the area covered by the
117 * request first if the allocationmap indicates that the area might be
118 * unallocated. */
119 #define ISCSI_CHECKALLOC_THRES 64
120
121 static void
122 iscsi_bh_cb(void *p)
123 {
124 IscsiAIOCB *acb = p;
125
126 qemu_bh_delete(acb->bh);
127
128 g_free(acb->buf);
129 acb->buf = NULL;
130
131 acb->common.cb(acb->common.opaque, acb->status);
132
133 if (acb->task != NULL) {
134 scsi_free_scsi_task(acb->task);
135 acb->task = NULL;
136 }
137
138 qemu_aio_unref(acb);
139 }
140
141 static void
142 iscsi_schedule_bh(IscsiAIOCB *acb)
143 {
144 if (acb->bh) {
145 return;
146 }
147 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb);
148 qemu_bh_schedule(acb->bh);
149 }
150
151 static void iscsi_co_generic_bh_cb(void *opaque)
152 {
153 struct IscsiTask *iTask = opaque;
154 iTask->complete = 1;
155 qemu_bh_delete(iTask->bh);
156 qemu_coroutine_enter(iTask->co, NULL);
157 }
158
159 static void iscsi_retry_timer_expired(void *opaque)
160 {
161 struct IscsiTask *iTask = opaque;
162 iTask->complete = 1;
163 if (iTask->co) {
164 qemu_coroutine_enter(iTask->co, NULL);
165 }
166 }
167
168 static inline unsigned exp_random(double mean)
169 {
170 return -mean * log((double)rand() / RAND_MAX);
171 }
172
173 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in
174 * libiscsi 1.10.0, together with other constants we need. Use it as
175 * a hint that we have to define them ourselves if needed, to keep the
176 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for
177 * the test because SCSI_STATUS_* is an enum.
178 *
179 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes
180 * an enum, check against the LIBISCSI_API_VERSION macro, which was
181 * introduced in 1.11.0. If it is present, there is no need to define
182 * anything.
183 */
184 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \
185 !defined(LIBISCSI_API_VERSION)
186 #define SCSI_STATUS_TASK_SET_FULL 0x28
187 #define SCSI_STATUS_TIMEOUT 0x0f000002
188 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600
189 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00
190 #endif
191
192 static int iscsi_translate_sense(struct scsi_sense *sense)
193 {
194 int ret;
195
196 switch (sense->key) {
197 case SCSI_SENSE_NOT_READY:
198 return -EBUSY;
199 case SCSI_SENSE_DATA_PROTECTION:
200 return -EACCES;
201 case SCSI_SENSE_COMMAND_ABORTED:
202 return -ECANCELED;
203 case SCSI_SENSE_ILLEGAL_REQUEST:
204 /* Parse ASCQ */
205 break;
206 default:
207 return -EIO;
208 }
209 switch (sense->ascq) {
210 case SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR:
211 case SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE:
212 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB:
213 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST:
214 ret = -EINVAL;
215 break;
216 case SCSI_SENSE_ASCQ_LBA_OUT_OF_RANGE:
217 ret = -ENOSPC;
218 break;
219 case SCSI_SENSE_ASCQ_LOGICAL_UNIT_NOT_SUPPORTED:
220 ret = -ENOTSUP;
221 break;
222 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT:
223 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_CLOSED:
224 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_OPEN:
225 ret = -ENOMEDIUM;
226 break;
227 case SCSI_SENSE_ASCQ_WRITE_PROTECTED:
228 ret = -EACCES;
229 break;
230 default:
231 ret = -EIO;
232 break;
233 }
234 return ret;
235 }
236
237 static void
238 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
239 void *command_data, void *opaque)
240 {
241 struct IscsiTask *iTask = opaque;
242 struct scsi_task *task = command_data;
243
244 iTask->status = status;
245 iTask->do_retry = 0;
246 iTask->task = task;
247
248 if (status != SCSI_STATUS_GOOD) {
249 if (iTask->retries++ < ISCSI_CMD_RETRIES) {
250 if (status == SCSI_STATUS_CHECK_CONDITION
251 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
252 error_report("iSCSI CheckCondition: %s",
253 iscsi_get_error(iscsi));
254 iTask->do_retry = 1;
255 goto out;
256 }
257 if (status == SCSI_STATUS_BUSY ||
258 status == SCSI_STATUS_TIMEOUT ||
259 status == SCSI_STATUS_TASK_SET_FULL) {
260 unsigned retry_time =
261 exp_random(iscsi_retry_times[iTask->retries - 1]);
262 if (status == SCSI_STATUS_TIMEOUT) {
263 /* make sure the request is rescheduled AFTER the
264 * reconnect is initiated */
265 retry_time = EVENT_INTERVAL * 2;
266 iTask->iscsilun->request_timed_out = true;
267 }
268 error_report("iSCSI Busy/TaskSetFull/TimeOut"
269 " (retry #%u in %u ms): %s",
270 iTask->retries, retry_time,
271 iscsi_get_error(iscsi));
272 aio_timer_init(iTask->iscsilun->aio_context,
273 &iTask->retry_timer, QEMU_CLOCK_REALTIME,
274 SCALE_MS, iscsi_retry_timer_expired, iTask);
275 timer_mod(&iTask->retry_timer,
276 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
277 iTask->do_retry = 1;
278 return;
279 }
280 }
281 iTask->err_code = iscsi_translate_sense(&task->sense);
282 error_report("iSCSI Failure: %s", iscsi_get_error(iscsi));
283 }
284
285 out:
286 if (iTask->co) {
287 iTask->bh = aio_bh_new(iTask->iscsilun->aio_context,
288 iscsi_co_generic_bh_cb, iTask);
289 qemu_bh_schedule(iTask->bh);
290 } else {
291 iTask->complete = 1;
292 }
293 }
294
295 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
296 {
297 *iTask = (struct IscsiTask) {
298 .co = qemu_coroutine_self(),
299 .iscsilun = iscsilun,
300 };
301 }
302
303 static void
304 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
305 void *private_data)
306 {
307 IscsiAIOCB *acb = private_data;
308
309 acb->status = -ECANCELED;
310 iscsi_schedule_bh(acb);
311 }
312
313 static void
314 iscsi_aio_cancel(BlockAIOCB *blockacb)
315 {
316 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
317 IscsiLun *iscsilun = acb->iscsilun;
318
319 if (acb->status != -EINPROGRESS) {
320 return;
321 }
322
323 /* send a task mgmt call to the target to cancel the task on the target */
324 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
325 iscsi_abort_task_cb, acb);
326
327 }
328
329 static const AIOCBInfo iscsi_aiocb_info = {
330 .aiocb_size = sizeof(IscsiAIOCB),
331 .cancel_async = iscsi_aio_cancel,
332 };
333
334
335 static void iscsi_process_read(void *arg);
336 static void iscsi_process_write(void *arg);
337
338 static void
339 iscsi_set_events(IscsiLun *iscsilun)
340 {
341 struct iscsi_context *iscsi = iscsilun->iscsi;
342 int ev = iscsi_which_events(iscsi);
343
344 if (ev != iscsilun->events) {
345 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi),
346 false,
347 (ev & POLLIN) ? iscsi_process_read : NULL,
348 (ev & POLLOUT) ? iscsi_process_write : NULL,
349 iscsilun);
350 iscsilun->events = ev;
351 }
352 }
353
354 static void iscsi_timed_check_events(void *opaque)
355 {
356 IscsiLun *iscsilun = opaque;
357
358 /* check for timed out requests */
359 iscsi_service(iscsilun->iscsi, 0);
360
361 if (iscsilun->request_timed_out) {
362 iscsilun->request_timed_out = false;
363 iscsi_reconnect(iscsilun->iscsi);
364 }
365
366 /* newer versions of libiscsi may return zero events. Ensure we are able
367 * to return to service once this situation changes. */
368 iscsi_set_events(iscsilun);
369
370 timer_mod(iscsilun->event_timer,
371 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
372 }
373
374 static void
375 iscsi_process_read(void *arg)
376 {
377 IscsiLun *iscsilun = arg;
378 struct iscsi_context *iscsi = iscsilun->iscsi;
379
380 iscsi_service(iscsi, POLLIN);
381 iscsi_set_events(iscsilun);
382 }
383
384 static void
385 iscsi_process_write(void *arg)
386 {
387 IscsiLun *iscsilun = arg;
388 struct iscsi_context *iscsi = iscsilun->iscsi;
389
390 iscsi_service(iscsi, POLLOUT);
391 iscsi_set_events(iscsilun);
392 }
393
394 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
395 {
396 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
397 }
398
399 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
400 {
401 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
402 }
403
404 static bool is_request_lun_aligned(int64_t sector_num, int nb_sectors,
405 IscsiLun *iscsilun)
406 {
407 if ((sector_num * BDRV_SECTOR_SIZE) % iscsilun->block_size ||
408 (nb_sectors * BDRV_SECTOR_SIZE) % iscsilun->block_size) {
409 error_report("iSCSI misaligned request: "
410 "iscsilun->block_size %u, sector_num %" PRIi64
411 ", nb_sectors %d",
412 iscsilun->block_size, sector_num, nb_sectors);
413 return 0;
414 }
415 return 1;
416 }
417
418 static unsigned long *iscsi_allocationmap_init(IscsiLun *iscsilun)
419 {
420 return bitmap_try_new(DIV_ROUND_UP(sector_lun2qemu(iscsilun->num_blocks,
421 iscsilun),
422 iscsilun->cluster_sectors));
423 }
424
425 static void iscsi_allocationmap_set(IscsiLun *iscsilun, int64_t sector_num,
426 int nb_sectors)
427 {
428 if (iscsilun->allocationmap == NULL) {
429 return;
430 }
431 bitmap_set(iscsilun->allocationmap,
432 sector_num / iscsilun->cluster_sectors,
433 DIV_ROUND_UP(nb_sectors, iscsilun->cluster_sectors));
434 }
435
436 static void iscsi_allocationmap_clear(IscsiLun *iscsilun, int64_t sector_num,
437 int nb_sectors)
438 {
439 int64_t cluster_num, nb_clusters;
440 if (iscsilun->allocationmap == NULL) {
441 return;
442 }
443 cluster_num = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors);
444 nb_clusters = (sector_num + nb_sectors) / iscsilun->cluster_sectors
445 - cluster_num;
446 if (nb_clusters > 0) {
447 bitmap_clear(iscsilun->allocationmap, cluster_num, nb_clusters);
448 }
449 }
450
451 static int coroutine_fn
452 iscsi_co_writev_flags(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
453 QEMUIOVector *iov, int flags)
454 {
455 IscsiLun *iscsilun = bs->opaque;
456 struct IscsiTask iTask;
457 uint64_t lba;
458 uint32_t num_sectors;
459 bool fua = flags & BDRV_REQ_FUA;
460
461 if (fua) {
462 assert(iscsilun->dpofua);
463 }
464 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
465 return -EINVAL;
466 }
467
468 if (bs->bl.max_transfer_length && nb_sectors > bs->bl.max_transfer_length) {
469 error_report("iSCSI Error: Write of %d sectors exceeds max_xfer_len "
470 "of %d sectors", nb_sectors, bs->bl.max_transfer_length);
471 return -EINVAL;
472 }
473
474 lba = sector_qemu2lun(sector_num, iscsilun);
475 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
476 iscsi_co_init_iscsitask(iscsilun, &iTask);
477 retry:
478 if (iscsilun->use_16_for_rw) {
479 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
480 NULL, num_sectors * iscsilun->block_size,
481 iscsilun->block_size, 0, 0, fua, 0, 0,
482 iscsi_co_generic_cb, &iTask);
483 } else {
484 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
485 NULL, num_sectors * iscsilun->block_size,
486 iscsilun->block_size, 0, 0, fua, 0, 0,
487 iscsi_co_generic_cb, &iTask);
488 }
489 if (iTask.task == NULL) {
490 return -ENOMEM;
491 }
492 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
493 iov->niov);
494 while (!iTask.complete) {
495 iscsi_set_events(iscsilun);
496 qemu_coroutine_yield();
497 }
498
499 if (iTask.task != NULL) {
500 scsi_free_scsi_task(iTask.task);
501 iTask.task = NULL;
502 }
503
504 if (iTask.do_retry) {
505 iTask.complete = 0;
506 goto retry;
507 }
508
509 if (iTask.status != SCSI_STATUS_GOOD) {
510 return iTask.err_code;
511 }
512
513 iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors);
514
515 return 0;
516 }
517
518
519 static bool iscsi_allocationmap_is_allocated(IscsiLun *iscsilun,
520 int64_t sector_num, int nb_sectors)
521 {
522 unsigned long size;
523 if (iscsilun->allocationmap == NULL) {
524 return true;
525 }
526 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
527 return !(find_next_bit(iscsilun->allocationmap, size,
528 sector_num / iscsilun->cluster_sectors) == size);
529 }
530
531 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
532 int64_t sector_num,
533 int nb_sectors, int *pnum,
534 BlockDriverState **file)
535 {
536 IscsiLun *iscsilun = bs->opaque;
537 struct scsi_get_lba_status *lbas = NULL;
538 struct scsi_lba_status_descriptor *lbasd = NULL;
539 struct IscsiTask iTask;
540 int64_t ret;
541
542 iscsi_co_init_iscsitask(iscsilun, &iTask);
543
544 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
545 ret = -EINVAL;
546 goto out;
547 }
548
549 /* default to all sectors allocated */
550 ret = BDRV_BLOCK_DATA;
551 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
552 *pnum = nb_sectors;
553
554 /* LUN does not support logical block provisioning */
555 if (!iscsilun->lbpme) {
556 goto out;
557 }
558
559 retry:
560 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
561 sector_qemu2lun(sector_num, iscsilun),
562 8 + 16, iscsi_co_generic_cb,
563 &iTask) == NULL) {
564 ret = -ENOMEM;
565 goto out;
566 }
567
568 while (!iTask.complete) {
569 iscsi_set_events(iscsilun);
570 qemu_coroutine_yield();
571 }
572
573 if (iTask.do_retry) {
574 if (iTask.task != NULL) {
575 scsi_free_scsi_task(iTask.task);
576 iTask.task = NULL;
577 }
578 iTask.complete = 0;
579 goto retry;
580 }
581
582 if (iTask.status != SCSI_STATUS_GOOD) {
583 /* in case the get_lba_status_callout fails (i.e.
584 * because the device is busy or the cmd is not
585 * supported) we pretend all blocks are allocated
586 * for backwards compatibility */
587 goto out;
588 }
589
590 lbas = scsi_datain_unmarshall(iTask.task);
591 if (lbas == NULL) {
592 ret = -EIO;
593 goto out;
594 }
595
596 lbasd = &lbas->descriptors[0];
597
598 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
599 ret = -EIO;
600 goto out;
601 }
602
603 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
604
605 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
606 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
607 ret &= ~BDRV_BLOCK_DATA;
608 if (iscsilun->lbprz) {
609 ret |= BDRV_BLOCK_ZERO;
610 }
611 }
612
613 if (ret & BDRV_BLOCK_ZERO) {
614 iscsi_allocationmap_clear(iscsilun, sector_num, *pnum);
615 } else {
616 iscsi_allocationmap_set(iscsilun, sector_num, *pnum);
617 }
618
619 if (*pnum > nb_sectors) {
620 *pnum = nb_sectors;
621 }
622 out:
623 if (iTask.task != NULL) {
624 scsi_free_scsi_task(iTask.task);
625 }
626 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) {
627 *file = bs;
628 }
629 return ret;
630 }
631
632 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
633 int64_t sector_num, int nb_sectors,
634 QEMUIOVector *iov)
635 {
636 IscsiLun *iscsilun = bs->opaque;
637 struct IscsiTask iTask;
638 uint64_t lba;
639 uint32_t num_sectors;
640
641 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
642 return -EINVAL;
643 }
644
645 if (bs->bl.max_transfer_length && nb_sectors > bs->bl.max_transfer_length) {
646 error_report("iSCSI Error: Read of %d sectors exceeds max_xfer_len "
647 "of %d sectors", nb_sectors, bs->bl.max_transfer_length);
648 return -EINVAL;
649 }
650
651 if (iscsilun->lbprz && nb_sectors >= ISCSI_CHECKALLOC_THRES &&
652 !iscsi_allocationmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
653 int64_t ret;
654 int pnum;
655 BlockDriverState *file;
656 ret = iscsi_co_get_block_status(bs, sector_num, INT_MAX, &pnum, &file);
657 if (ret < 0) {
658 return ret;
659 }
660 if (ret & BDRV_BLOCK_ZERO && pnum >= nb_sectors) {
661 qemu_iovec_memset(iov, 0, 0x00, iov->size);
662 return 0;
663 }
664 }
665
666 lba = sector_qemu2lun(sector_num, iscsilun);
667 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
668
669 iscsi_co_init_iscsitask(iscsilun, &iTask);
670 retry:
671 if (iscsilun->use_16_for_rw) {
672 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
673 num_sectors * iscsilun->block_size,
674 iscsilun->block_size, 0, 0, 0, 0, 0,
675 iscsi_co_generic_cb, &iTask);
676 } else {
677 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
678 num_sectors * iscsilun->block_size,
679 iscsilun->block_size,
680 0, 0, 0, 0, 0,
681 iscsi_co_generic_cb, &iTask);
682 }
683 if (iTask.task == NULL) {
684 return -ENOMEM;
685 }
686 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
687
688 while (!iTask.complete) {
689 iscsi_set_events(iscsilun);
690 qemu_coroutine_yield();
691 }
692
693 if (iTask.task != NULL) {
694 scsi_free_scsi_task(iTask.task);
695 iTask.task = NULL;
696 }
697
698 if (iTask.do_retry) {
699 iTask.complete = 0;
700 goto retry;
701 }
702
703 if (iTask.status != SCSI_STATUS_GOOD) {
704 return iTask.err_code;
705 }
706
707 return 0;
708 }
709
710 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
711 {
712 IscsiLun *iscsilun = bs->opaque;
713 struct IscsiTask iTask;
714
715 iscsi_co_init_iscsitask(iscsilun, &iTask);
716 retry:
717 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
718 0, iscsi_co_generic_cb, &iTask) == NULL) {
719 return -ENOMEM;
720 }
721
722 while (!iTask.complete) {
723 iscsi_set_events(iscsilun);
724 qemu_coroutine_yield();
725 }
726
727 if (iTask.task != NULL) {
728 scsi_free_scsi_task(iTask.task);
729 iTask.task = NULL;
730 }
731
732 if (iTask.do_retry) {
733 iTask.complete = 0;
734 goto retry;
735 }
736
737 if (iTask.status != SCSI_STATUS_GOOD) {
738 return iTask.err_code;
739 }
740
741 return 0;
742 }
743
744 #ifdef __linux__
745 static void
746 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
747 void *command_data, void *opaque)
748 {
749 IscsiAIOCB *acb = opaque;
750
751 g_free(acb->buf);
752 acb->buf = NULL;
753
754 acb->status = 0;
755 if (status < 0) {
756 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
757 iscsi_get_error(iscsi));
758 acb->status = iscsi_translate_sense(&acb->task->sense);
759 }
760
761 acb->ioh->driver_status = 0;
762 acb->ioh->host_status = 0;
763 acb->ioh->resid = 0;
764 acb->ioh->status = status;
765
766 #define SG_ERR_DRIVER_SENSE 0x08
767
768 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
769 int ss;
770
771 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
772
773 acb->ioh->sb_len_wr = acb->task->datain.size - 2;
774 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
775 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
776 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
777 }
778
779 iscsi_schedule_bh(acb);
780 }
781
782 static void iscsi_ioctl_bh_completion(void *opaque)
783 {
784 IscsiAIOCB *acb = opaque;
785
786 qemu_bh_delete(acb->bh);
787 acb->common.cb(acb->common.opaque, acb->ret);
788 qemu_aio_unref(acb);
789 }
790
791 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
792 {
793 BlockDriverState *bs = acb->common.bs;
794 IscsiLun *iscsilun = bs->opaque;
795 int ret = 0;
796
797 switch (req) {
798 case SG_GET_VERSION_NUM:
799 *(int *)buf = 30000;
800 break;
801 case SG_GET_SCSI_ID:
802 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
803 break;
804 default:
805 ret = -EINVAL;
806 }
807 assert(!acb->bh);
808 acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
809 iscsi_ioctl_bh_completion, acb);
810 acb->ret = ret;
811 qemu_bh_schedule(acb->bh);
812 }
813
814 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
815 unsigned long int req, void *buf,
816 BlockCompletionFunc *cb, void *opaque)
817 {
818 IscsiLun *iscsilun = bs->opaque;
819 struct iscsi_context *iscsi = iscsilun->iscsi;
820 struct iscsi_data data;
821 IscsiAIOCB *acb;
822
823 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
824
825 acb->iscsilun = iscsilun;
826 acb->bh = NULL;
827 acb->status = -EINPROGRESS;
828 acb->buf = NULL;
829 acb->ioh = buf;
830
831 if (req != SG_IO) {
832 iscsi_ioctl_handle_emulated(acb, req, buf);
833 return &acb->common;
834 }
835
836 acb->task = malloc(sizeof(struct scsi_task));
837 if (acb->task == NULL) {
838 error_report("iSCSI: Failed to allocate task for scsi command. %s",
839 iscsi_get_error(iscsi));
840 qemu_aio_unref(acb);
841 return NULL;
842 }
843 memset(acb->task, 0, sizeof(struct scsi_task));
844
845 switch (acb->ioh->dxfer_direction) {
846 case SG_DXFER_TO_DEV:
847 acb->task->xfer_dir = SCSI_XFER_WRITE;
848 break;
849 case SG_DXFER_FROM_DEV:
850 acb->task->xfer_dir = SCSI_XFER_READ;
851 break;
852 default:
853 acb->task->xfer_dir = SCSI_XFER_NONE;
854 break;
855 }
856
857 acb->task->cdb_size = acb->ioh->cmd_len;
858 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
859 acb->task->expxferlen = acb->ioh->dxfer_len;
860
861 data.size = 0;
862 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
863 if (acb->ioh->iovec_count == 0) {
864 data.data = acb->ioh->dxferp;
865 data.size = acb->ioh->dxfer_len;
866 } else {
867 scsi_task_set_iov_out(acb->task,
868 (struct scsi_iovec *) acb->ioh->dxferp,
869 acb->ioh->iovec_count);
870 }
871 }
872
873 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
874 iscsi_aio_ioctl_cb,
875 (data.size > 0) ? &data : NULL,
876 acb) != 0) {
877 scsi_free_scsi_task(acb->task);
878 qemu_aio_unref(acb);
879 return NULL;
880 }
881
882 /* tell libiscsi to read straight into the buffer we got from ioctl */
883 if (acb->task->xfer_dir == SCSI_XFER_READ) {
884 if (acb->ioh->iovec_count == 0) {
885 scsi_task_add_data_in_buffer(acb->task,
886 acb->ioh->dxfer_len,
887 acb->ioh->dxferp);
888 } else {
889 scsi_task_set_iov_in(acb->task,
890 (struct scsi_iovec *) acb->ioh->dxferp,
891 acb->ioh->iovec_count);
892 }
893 }
894
895 iscsi_set_events(iscsilun);
896
897 return &acb->common;
898 }
899
900 #endif
901
902 static int64_t
903 iscsi_getlength(BlockDriverState *bs)
904 {
905 IscsiLun *iscsilun = bs->opaque;
906 int64_t len;
907
908 len = iscsilun->num_blocks;
909 len *= iscsilun->block_size;
910
911 return len;
912 }
913
914 static int
915 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num,
916 int nb_sectors)
917 {
918 IscsiLun *iscsilun = bs->opaque;
919 struct IscsiTask iTask;
920 struct unmap_list list;
921
922 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
923 return -EINVAL;
924 }
925
926 if (!iscsilun->lbp.lbpu) {
927 /* UNMAP is not supported by the target */
928 return 0;
929 }
930
931 list.lba = sector_qemu2lun(sector_num, iscsilun);
932 list.num = sector_qemu2lun(nb_sectors, iscsilun);
933
934 iscsi_co_init_iscsitask(iscsilun, &iTask);
935 retry:
936 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
937 iscsi_co_generic_cb, &iTask) == NULL) {
938 return -ENOMEM;
939 }
940
941 while (!iTask.complete) {
942 iscsi_set_events(iscsilun);
943 qemu_coroutine_yield();
944 }
945
946 if (iTask.task != NULL) {
947 scsi_free_scsi_task(iTask.task);
948 iTask.task = NULL;
949 }
950
951 if (iTask.do_retry) {
952 iTask.complete = 0;
953 goto retry;
954 }
955
956 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
957 /* the target might fail with a check condition if it
958 is not happy with the alignment of the UNMAP request
959 we silently fail in this case */
960 return 0;
961 }
962
963 if (iTask.status != SCSI_STATUS_GOOD) {
964 return iTask.err_code;
965 }
966
967 iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors);
968
969 return 0;
970 }
971
972 static int
973 coroutine_fn iscsi_co_write_zeroes(BlockDriverState *bs, int64_t sector_num,
974 int nb_sectors, BdrvRequestFlags flags)
975 {
976 IscsiLun *iscsilun = bs->opaque;
977 struct IscsiTask iTask;
978 uint64_t lba;
979 uint32_t nb_blocks;
980 bool use_16_for_ws = iscsilun->use_16_for_rw;
981
982 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
983 return -EINVAL;
984 }
985
986 if (flags & BDRV_REQ_MAY_UNMAP) {
987 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
988 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
989 use_16_for_ws = true;
990 }
991 if (use_16_for_ws && !iscsilun->lbp.lbpws) {
992 /* WRITESAME16 with UNMAP is not supported by the target,
993 * fall back and try WRITESAME10/16 without UNMAP */
994 flags &= ~BDRV_REQ_MAY_UNMAP;
995 use_16_for_ws = iscsilun->use_16_for_rw;
996 }
997 }
998
999 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1000 /* WRITESAME without UNMAP is not supported by the target */
1001 return -ENOTSUP;
1002 }
1003
1004 lba = sector_qemu2lun(sector_num, iscsilun);
1005 nb_blocks = sector_qemu2lun(nb_sectors, iscsilun);
1006
1007 if (iscsilun->zeroblock == NULL) {
1008 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1009 if (iscsilun->zeroblock == NULL) {
1010 return -ENOMEM;
1011 }
1012 }
1013
1014 iscsi_co_init_iscsitask(iscsilun, &iTask);
1015 retry:
1016 if (use_16_for_ws) {
1017 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1018 iscsilun->zeroblock, iscsilun->block_size,
1019 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1020 0, 0, iscsi_co_generic_cb, &iTask);
1021 } else {
1022 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1023 iscsilun->zeroblock, iscsilun->block_size,
1024 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1025 0, 0, iscsi_co_generic_cb, &iTask);
1026 }
1027 if (iTask.task == NULL) {
1028 return -ENOMEM;
1029 }
1030
1031 while (!iTask.complete) {
1032 iscsi_set_events(iscsilun);
1033 qemu_coroutine_yield();
1034 }
1035
1036 if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1037 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1038 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1039 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1040 /* WRITE SAME is not supported by the target */
1041 iscsilun->has_write_same = false;
1042 scsi_free_scsi_task(iTask.task);
1043 return -ENOTSUP;
1044 }
1045
1046 if (iTask.task != NULL) {
1047 scsi_free_scsi_task(iTask.task);
1048 iTask.task = NULL;
1049 }
1050
1051 if (iTask.do_retry) {
1052 iTask.complete = 0;
1053 goto retry;
1054 }
1055
1056 if (iTask.status != SCSI_STATUS_GOOD) {
1057 return iTask.err_code;
1058 }
1059
1060 if (flags & BDRV_REQ_MAY_UNMAP) {
1061 iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors);
1062 } else {
1063 iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors);
1064 }
1065
1066 return 0;
1067 }
1068
1069 static void parse_chap(struct iscsi_context *iscsi, const char *target,
1070 Error **errp)
1071 {
1072 QemuOptsList *list;
1073 QemuOpts *opts;
1074 const char *user = NULL;
1075 const char *password = NULL;
1076 const char *secretid;
1077 char *secret = NULL;
1078
1079 list = qemu_find_opts("iscsi");
1080 if (!list) {
1081 return;
1082 }
1083
1084 opts = qemu_opts_find(list, target);
1085 if (opts == NULL) {
1086 opts = QTAILQ_FIRST(&list->head);
1087 if (!opts) {
1088 return;
1089 }
1090 }
1091
1092 user = qemu_opt_get(opts, "user");
1093 if (!user) {
1094 return;
1095 }
1096
1097 secretid = qemu_opt_get(opts, "password-secret");
1098 password = qemu_opt_get(opts, "password");
1099 if (secretid && password) {
1100 error_setg(errp, "'password' and 'password-secret' properties are "
1101 "mutually exclusive");
1102 return;
1103 }
1104 if (secretid) {
1105 secret = qcrypto_secret_lookup_as_utf8(secretid, errp);
1106 if (!secret) {
1107 return;
1108 }
1109 password = secret;
1110 } else if (!password) {
1111 error_setg(errp, "CHAP username specified but no password was given");
1112 return;
1113 }
1114
1115 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1116 error_setg(errp, "Failed to set initiator username and password");
1117 }
1118
1119 g_free(secret);
1120 }
1121
1122 static void parse_header_digest(struct iscsi_context *iscsi, const char *target,
1123 Error **errp)
1124 {
1125 QemuOptsList *list;
1126 QemuOpts *opts;
1127 const char *digest = NULL;
1128
1129 list = qemu_find_opts("iscsi");
1130 if (!list) {
1131 return;
1132 }
1133
1134 opts = qemu_opts_find(list, target);
1135 if (opts == NULL) {
1136 opts = QTAILQ_FIRST(&list->head);
1137 if (!opts) {
1138 return;
1139 }
1140 }
1141
1142 digest = qemu_opt_get(opts, "header-digest");
1143 if (!digest) {
1144 return;
1145 }
1146
1147 if (!strcmp(digest, "CRC32C")) {
1148 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1149 } else if (!strcmp(digest, "NONE")) {
1150 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1151 } else if (!strcmp(digest, "CRC32C-NONE")) {
1152 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1153 } else if (!strcmp(digest, "NONE-CRC32C")) {
1154 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1155 } else {
1156 error_setg(errp, "Invalid header-digest setting : %s", digest);
1157 }
1158 }
1159
1160 static char *parse_initiator_name(const char *target)
1161 {
1162 QemuOptsList *list;
1163 QemuOpts *opts;
1164 const char *name;
1165 char *iscsi_name;
1166 UuidInfo *uuid_info;
1167
1168 list = qemu_find_opts("iscsi");
1169 if (list) {
1170 opts = qemu_opts_find(list, target);
1171 if (!opts) {
1172 opts = QTAILQ_FIRST(&list->head);
1173 }
1174 if (opts) {
1175 name = qemu_opt_get(opts, "initiator-name");
1176 if (name) {
1177 return g_strdup(name);
1178 }
1179 }
1180 }
1181
1182 uuid_info = qmp_query_uuid(NULL);
1183 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1184 name = qemu_get_vm_name();
1185 } else {
1186 name = uuid_info->UUID;
1187 }
1188 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1189 name ? ":" : "", name ? name : "");
1190 qapi_free_UuidInfo(uuid_info);
1191 return iscsi_name;
1192 }
1193
1194 static int parse_timeout(const char *target)
1195 {
1196 QemuOptsList *list;
1197 QemuOpts *opts;
1198 const char *timeout;
1199
1200 list = qemu_find_opts("iscsi");
1201 if (list) {
1202 opts = qemu_opts_find(list, target);
1203 if (!opts) {
1204 opts = QTAILQ_FIRST(&list->head);
1205 }
1206 if (opts) {
1207 timeout = qemu_opt_get(opts, "timeout");
1208 if (timeout) {
1209 return atoi(timeout);
1210 }
1211 }
1212 }
1213
1214 return 0;
1215 }
1216
1217 static void iscsi_nop_timed_event(void *opaque)
1218 {
1219 IscsiLun *iscsilun = opaque;
1220
1221 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1222 error_report("iSCSI: NOP timeout. Reconnecting...");
1223 iscsilun->request_timed_out = true;
1224 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1225 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1226 return;
1227 }
1228
1229 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1230 iscsi_set_events(iscsilun);
1231 }
1232
1233 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1234 {
1235 struct scsi_task *task = NULL;
1236 struct scsi_readcapacity10 *rc10 = NULL;
1237 struct scsi_readcapacity16 *rc16 = NULL;
1238 int retries = ISCSI_CMD_RETRIES;
1239
1240 do {
1241 if (task != NULL) {
1242 scsi_free_scsi_task(task);
1243 task = NULL;
1244 }
1245
1246 switch (iscsilun->type) {
1247 case TYPE_DISK:
1248 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1249 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1250 rc16 = scsi_datain_unmarshall(task);
1251 if (rc16 == NULL) {
1252 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1253 } else {
1254 iscsilun->block_size = rc16->block_length;
1255 iscsilun->num_blocks = rc16->returned_lba + 1;
1256 iscsilun->lbpme = !!rc16->lbpme;
1257 iscsilun->lbprz = !!rc16->lbprz;
1258 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1259 }
1260 break;
1261 }
1262 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1263 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1264 break;
1265 }
1266 /* Fall through and try READ CAPACITY(10) instead. */
1267 case TYPE_ROM:
1268 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1269 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1270 rc10 = scsi_datain_unmarshall(task);
1271 if (rc10 == NULL) {
1272 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1273 } else {
1274 iscsilun->block_size = rc10->block_size;
1275 if (rc10->lba == 0) {
1276 /* blank disk loaded */
1277 iscsilun->num_blocks = 0;
1278 } else {
1279 iscsilun->num_blocks = rc10->lba + 1;
1280 }
1281 }
1282 }
1283 break;
1284 default:
1285 return;
1286 }
1287 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1288 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1289 && retries-- > 0);
1290
1291 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1292 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1293 } else if (!iscsilun->block_size ||
1294 iscsilun->block_size % BDRV_SECTOR_SIZE) {
1295 error_setg(errp, "iSCSI: the target returned an invalid "
1296 "block size of %d.", iscsilun->block_size);
1297 }
1298 if (task) {
1299 scsi_free_scsi_task(task);
1300 }
1301 }
1302
1303 /* TODO Convert to fine grained options */
1304 static QemuOptsList runtime_opts = {
1305 .name = "iscsi",
1306 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1307 .desc = {
1308 {
1309 .name = "filename",
1310 .type = QEMU_OPT_STRING,
1311 .help = "URL to the iscsi image",
1312 },
1313 { /* end of list */ }
1314 },
1315 };
1316
1317 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1318 int evpd, int pc, void **inq, Error **errp)
1319 {
1320 int full_size;
1321 struct scsi_task *task = NULL;
1322 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1323 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1324 goto fail;
1325 }
1326 full_size = scsi_datain_getfullsize(task);
1327 if (full_size > task->datain.size) {
1328 scsi_free_scsi_task(task);
1329
1330 /* we need more data for the full list */
1331 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1332 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1333 goto fail;
1334 }
1335 }
1336
1337 *inq = scsi_datain_unmarshall(task);
1338 if (*inq == NULL) {
1339 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1340 goto fail_with_err;
1341 }
1342
1343 return task;
1344
1345 fail:
1346 error_setg(errp, "iSCSI: Inquiry command failed : %s",
1347 iscsi_get_error(iscsi));
1348 fail_with_err:
1349 if (task != NULL) {
1350 scsi_free_scsi_task(task);
1351 }
1352 return NULL;
1353 }
1354
1355 static void iscsi_detach_aio_context(BlockDriverState *bs)
1356 {
1357 IscsiLun *iscsilun = bs->opaque;
1358
1359 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1360 false, NULL, NULL, NULL);
1361 iscsilun->events = 0;
1362
1363 if (iscsilun->nop_timer) {
1364 timer_del(iscsilun->nop_timer);
1365 timer_free(iscsilun->nop_timer);
1366 iscsilun->nop_timer = NULL;
1367 }
1368 if (iscsilun->event_timer) {
1369 timer_del(iscsilun->event_timer);
1370 timer_free(iscsilun->event_timer);
1371 iscsilun->event_timer = NULL;
1372 }
1373 }
1374
1375 static void iscsi_attach_aio_context(BlockDriverState *bs,
1376 AioContext *new_context)
1377 {
1378 IscsiLun *iscsilun = bs->opaque;
1379
1380 iscsilun->aio_context = new_context;
1381 iscsi_set_events(iscsilun);
1382
1383 /* Set up a timer for sending out iSCSI NOPs */
1384 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1385 QEMU_CLOCK_REALTIME, SCALE_MS,
1386 iscsi_nop_timed_event, iscsilun);
1387 timer_mod(iscsilun->nop_timer,
1388 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1389
1390 /* Set up a timer for periodic calls to iscsi_set_events and to
1391 * scan for command timeout */
1392 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1393 QEMU_CLOCK_REALTIME, SCALE_MS,
1394 iscsi_timed_check_events, iscsilun);
1395 timer_mod(iscsilun->event_timer,
1396 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1397 }
1398
1399 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1400 {
1401 struct scsi_task *task;
1402 struct scsi_mode_sense *ms = NULL;
1403 iscsilun->write_protected = false;
1404 iscsilun->dpofua = false;
1405
1406 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1407 1, SCSI_MODESENSE_PC_CURRENT,
1408 0x3F, 0, 255);
1409 if (task == NULL) {
1410 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1411 iscsi_get_error(iscsilun->iscsi));
1412 goto out;
1413 }
1414
1415 if (task->status != SCSI_STATUS_GOOD) {
1416 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1417 goto out;
1418 }
1419 ms = scsi_datain_unmarshall(task);
1420 if (!ms) {
1421 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1422 iscsi_get_error(iscsilun->iscsi));
1423 goto out;
1424 }
1425 iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1426 iscsilun->dpofua = ms->device_specific_parameter & 0x10;
1427
1428 out:
1429 if (task) {
1430 scsi_free_scsi_task(task);
1431 }
1432 }
1433
1434 /*
1435 * We support iscsi url's on the form
1436 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1437 */
1438 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1439 Error **errp)
1440 {
1441 IscsiLun *iscsilun = bs->opaque;
1442 struct iscsi_context *iscsi = NULL;
1443 struct iscsi_url *iscsi_url = NULL;
1444 struct scsi_task *task = NULL;
1445 struct scsi_inquiry_standard *inq = NULL;
1446 struct scsi_inquiry_supported_pages *inq_vpd;
1447 char *initiator_name = NULL;
1448 QemuOpts *opts;
1449 Error *local_err = NULL;
1450 const char *filename;
1451 int i, ret = 0, timeout = 0;
1452
1453 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1454 qemu_opts_absorb_qdict(opts, options, &local_err);
1455 if (local_err) {
1456 error_propagate(errp, local_err);
1457 ret = -EINVAL;
1458 goto out;
1459 }
1460
1461 filename = qemu_opt_get(opts, "filename");
1462
1463 iscsi_url = iscsi_parse_full_url(iscsi, filename);
1464 if (iscsi_url == NULL) {
1465 error_setg(errp, "Failed to parse URL : %s", filename);
1466 ret = -EINVAL;
1467 goto out;
1468 }
1469
1470 memset(iscsilun, 0, sizeof(IscsiLun));
1471
1472 initiator_name = parse_initiator_name(iscsi_url->target);
1473
1474 iscsi = iscsi_create_context(initiator_name);
1475 if (iscsi == NULL) {
1476 error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1477 ret = -ENOMEM;
1478 goto out;
1479 }
1480
1481 if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1482 error_setg(errp, "iSCSI: Failed to set target name.");
1483 ret = -EINVAL;
1484 goto out;
1485 }
1486
1487 if (iscsi_url->user[0] != '\0') {
1488 ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1489 iscsi_url->passwd);
1490 if (ret != 0) {
1491 error_setg(errp, "Failed to set initiator username and password");
1492 ret = -EINVAL;
1493 goto out;
1494 }
1495 }
1496
1497 /* check if we got CHAP username/password via the options */
1498 parse_chap(iscsi, iscsi_url->target, &local_err);
1499 if (local_err != NULL) {
1500 error_propagate(errp, local_err);
1501 ret = -EINVAL;
1502 goto out;
1503 }
1504
1505 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1506 error_setg(errp, "iSCSI: Failed to set session type to normal.");
1507 ret = -EINVAL;
1508 goto out;
1509 }
1510
1511 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1512
1513 /* check if we got HEADER_DIGEST via the options */
1514 parse_header_digest(iscsi, iscsi_url->target, &local_err);
1515 if (local_err != NULL) {
1516 error_propagate(errp, local_err);
1517 ret = -EINVAL;
1518 goto out;
1519 }
1520
1521 /* timeout handling is broken in libiscsi before 1.15.0 */
1522 timeout = parse_timeout(iscsi_url->target);
1523 #if defined(LIBISCSI_API_VERSION) && LIBISCSI_API_VERSION >= 20150621
1524 iscsi_set_timeout(iscsi, timeout);
1525 #else
1526 if (timeout) {
1527 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1528 }
1529 #endif
1530
1531 if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1532 error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1533 iscsi_get_error(iscsi));
1534 ret = -EINVAL;
1535 goto out;
1536 }
1537
1538 iscsilun->iscsi = iscsi;
1539 iscsilun->aio_context = bdrv_get_aio_context(bs);
1540 iscsilun->lun = iscsi_url->lun;
1541 iscsilun->has_write_same = true;
1542
1543 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1544 (void **) &inq, errp);
1545 if (task == NULL) {
1546 ret = -EINVAL;
1547 goto out;
1548 }
1549 iscsilun->type = inq->periperal_device_type;
1550 scsi_free_scsi_task(task);
1551 task = NULL;
1552
1553 iscsi_modesense_sync(iscsilun);
1554 if (iscsilun->dpofua) {
1555 bs->supported_write_flags = BDRV_REQ_FUA;
1556 }
1557 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
1558
1559 /* Check the write protect flag of the LUN if we want to write */
1560 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1561 iscsilun->write_protected) {
1562 error_setg(errp, "Cannot open a write protected LUN as read-write");
1563 ret = -EACCES;
1564 goto out;
1565 }
1566
1567 iscsi_readcapacity_sync(iscsilun, &local_err);
1568 if (local_err != NULL) {
1569 error_propagate(errp, local_err);
1570 ret = -EINVAL;
1571 goto out;
1572 }
1573 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1574 bs->request_alignment = iscsilun->block_size;
1575
1576 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1577 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1578 * will try to read from the device to guess the image format.
1579 */
1580 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1581 bs->sg = 1;
1582 }
1583
1584 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1585 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1586 (void **) &inq_vpd, errp);
1587 if (task == NULL) {
1588 ret = -EINVAL;
1589 goto out;
1590 }
1591 for (i = 0; i < inq_vpd->num_pages; i++) {
1592 struct scsi_task *inq_task;
1593 struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1594 struct scsi_inquiry_block_limits *inq_bl;
1595 switch (inq_vpd->pages[i]) {
1596 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1597 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1598 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1599 (void **) &inq_lbp, errp);
1600 if (inq_task == NULL) {
1601 ret = -EINVAL;
1602 goto out;
1603 }
1604 memcpy(&iscsilun->lbp, inq_lbp,
1605 sizeof(struct scsi_inquiry_logical_block_provisioning));
1606 scsi_free_scsi_task(inq_task);
1607 break;
1608 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1609 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1610 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1611 (void **) &inq_bl, errp);
1612 if (inq_task == NULL) {
1613 ret = -EINVAL;
1614 goto out;
1615 }
1616 memcpy(&iscsilun->bl, inq_bl,
1617 sizeof(struct scsi_inquiry_block_limits));
1618 scsi_free_scsi_task(inq_task);
1619 break;
1620 default:
1621 break;
1622 }
1623 }
1624 scsi_free_scsi_task(task);
1625 task = NULL;
1626
1627 iscsi_attach_aio_context(bs, iscsilun->aio_context);
1628
1629 /* Guess the internal cluster (page) size of the iscsi target by the means
1630 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1631 * reasonable size */
1632 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
1633 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1634 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran *
1635 iscsilun->block_size) >> BDRV_SECTOR_BITS;
1636 if (iscsilun->lbprz) {
1637 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun);
1638 if (iscsilun->allocationmap == NULL) {
1639 ret = -ENOMEM;
1640 }
1641 }
1642 }
1643
1644 out:
1645 qemu_opts_del(opts);
1646 g_free(initiator_name);
1647 if (iscsi_url != NULL) {
1648 iscsi_destroy_url(iscsi_url);
1649 }
1650 if (task != NULL) {
1651 scsi_free_scsi_task(task);
1652 }
1653
1654 if (ret) {
1655 if (iscsi != NULL) {
1656 if (iscsi_is_logged_in(iscsi)) {
1657 iscsi_logout_sync(iscsi);
1658 }
1659 iscsi_destroy_context(iscsi);
1660 }
1661 memset(iscsilun, 0, sizeof(IscsiLun));
1662 }
1663 return ret;
1664 }
1665
1666 static void iscsi_close(BlockDriverState *bs)
1667 {
1668 IscsiLun *iscsilun = bs->opaque;
1669 struct iscsi_context *iscsi = iscsilun->iscsi;
1670
1671 iscsi_detach_aio_context(bs);
1672 if (iscsi_is_logged_in(iscsi)) {
1673 iscsi_logout_sync(iscsi);
1674 }
1675 iscsi_destroy_context(iscsi);
1676 g_free(iscsilun->zeroblock);
1677 g_free(iscsilun->allocationmap);
1678 memset(iscsilun, 0, sizeof(IscsiLun));
1679 }
1680
1681 static int sector_limits_lun2qemu(int64_t sector, IscsiLun *iscsilun)
1682 {
1683 return MIN(sector_lun2qemu(sector, iscsilun), INT_MAX / 2 + 1);
1684 }
1685
1686 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
1687 {
1688 /* We don't actually refresh here, but just return data queried in
1689 * iscsi_open(): iscsi targets don't change their limits. */
1690
1691 IscsiLun *iscsilun = bs->opaque;
1692 uint32_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
1693
1694 if (iscsilun->bl.max_xfer_len) {
1695 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
1696 }
1697
1698 bs->bl.max_transfer_length = sector_limits_lun2qemu(max_xfer_len, iscsilun);
1699
1700 if (iscsilun->lbp.lbpu) {
1701 if (iscsilun->bl.max_unmap < 0xffffffff) {
1702 bs->bl.max_discard =
1703 sector_limits_lun2qemu(iscsilun->bl.max_unmap, iscsilun);
1704 }
1705 bs->bl.discard_alignment =
1706 sector_limits_lun2qemu(iscsilun->bl.opt_unmap_gran, iscsilun);
1707 }
1708
1709 if (iscsilun->bl.max_ws_len < 0xffffffff) {
1710 bs->bl.max_write_zeroes =
1711 sector_limits_lun2qemu(iscsilun->bl.max_ws_len, iscsilun);
1712 }
1713 if (iscsilun->lbp.lbpws) {
1714 bs->bl.write_zeroes_alignment =
1715 sector_limits_lun2qemu(iscsilun->bl.opt_unmap_gran, iscsilun);
1716 }
1717 bs->bl.opt_transfer_length =
1718 sector_limits_lun2qemu(iscsilun->bl.opt_xfer_len, iscsilun);
1719 }
1720
1721 /* Note that this will not re-establish a connection with an iSCSI target - it
1722 * is effectively a NOP. */
1723 static int iscsi_reopen_prepare(BDRVReopenState *state,
1724 BlockReopenQueue *queue, Error **errp)
1725 {
1726 IscsiLun *iscsilun = state->bs->opaque;
1727
1728 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
1729 error_setg(errp, "Cannot open a write protected LUN as read-write");
1730 return -EACCES;
1731 }
1732 return 0;
1733 }
1734
1735 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1736 {
1737 IscsiLun *iscsilun = bs->opaque;
1738 Error *local_err = NULL;
1739
1740 if (iscsilun->type != TYPE_DISK) {
1741 return -ENOTSUP;
1742 }
1743
1744 iscsi_readcapacity_sync(iscsilun, &local_err);
1745 if (local_err != NULL) {
1746 error_free(local_err);
1747 return -EIO;
1748 }
1749
1750 if (offset > iscsi_getlength(bs)) {
1751 return -EINVAL;
1752 }
1753
1754 if (iscsilun->allocationmap != NULL) {
1755 g_free(iscsilun->allocationmap);
1756 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun);
1757 }
1758
1759 return 0;
1760 }
1761
1762 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp)
1763 {
1764 int ret = 0;
1765 int64_t total_size = 0;
1766 BlockDriverState *bs;
1767 IscsiLun *iscsilun = NULL;
1768 QDict *bs_options;
1769
1770 bs = bdrv_new();
1771
1772 /* Read out options */
1773 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1774 BDRV_SECTOR_SIZE);
1775 bs->opaque = g_new0(struct IscsiLun, 1);
1776 iscsilun = bs->opaque;
1777
1778 bs_options = qdict_new();
1779 qdict_put(bs_options, "filename", qstring_from_str(filename));
1780 ret = iscsi_open(bs, bs_options, 0, NULL);
1781 QDECREF(bs_options);
1782
1783 if (ret != 0) {
1784 goto out;
1785 }
1786 iscsi_detach_aio_context(bs);
1787 if (iscsilun->type != TYPE_DISK) {
1788 ret = -ENODEV;
1789 goto out;
1790 }
1791 if (bs->total_sectors < total_size) {
1792 ret = -ENOSPC;
1793 goto out;
1794 }
1795
1796 ret = 0;
1797 out:
1798 if (iscsilun->iscsi != NULL) {
1799 iscsi_destroy_context(iscsilun->iscsi);
1800 }
1801 g_free(bs->opaque);
1802 bs->opaque = NULL;
1803 bdrv_unref(bs);
1804 return ret;
1805 }
1806
1807 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1808 {
1809 IscsiLun *iscsilun = bs->opaque;
1810 bdi->unallocated_blocks_are_zero = iscsilun->lbprz;
1811 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
1812 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE;
1813 return 0;
1814 }
1815
1816 static QemuOptsList iscsi_create_opts = {
1817 .name = "iscsi-create-opts",
1818 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
1819 .desc = {
1820 {
1821 .name = BLOCK_OPT_SIZE,
1822 .type = QEMU_OPT_SIZE,
1823 .help = "Virtual disk size"
1824 },
1825 { /* end of list */ }
1826 }
1827 };
1828
1829 static BlockDriver bdrv_iscsi = {
1830 .format_name = "iscsi",
1831 .protocol_name = "iscsi",
1832
1833 .instance_size = sizeof(IscsiLun),
1834 .bdrv_needs_filename = true,
1835 .bdrv_file_open = iscsi_open,
1836 .bdrv_close = iscsi_close,
1837 .bdrv_create = iscsi_create,
1838 .create_opts = &iscsi_create_opts,
1839 .bdrv_reopen_prepare = iscsi_reopen_prepare,
1840
1841 .bdrv_getlength = iscsi_getlength,
1842 .bdrv_get_info = iscsi_get_info,
1843 .bdrv_truncate = iscsi_truncate,
1844 .bdrv_refresh_limits = iscsi_refresh_limits,
1845
1846 .bdrv_co_get_block_status = iscsi_co_get_block_status,
1847 .bdrv_co_discard = iscsi_co_discard,
1848 .bdrv_co_write_zeroes = iscsi_co_write_zeroes,
1849 .bdrv_co_readv = iscsi_co_readv,
1850 .bdrv_co_writev_flags = iscsi_co_writev_flags,
1851 .bdrv_co_flush_to_disk = iscsi_co_flush,
1852
1853 #ifdef __linux__
1854 .bdrv_aio_ioctl = iscsi_aio_ioctl,
1855 #endif
1856
1857 .bdrv_detach_aio_context = iscsi_detach_aio_context,
1858 .bdrv_attach_aio_context = iscsi_attach_aio_context,
1859 };
1860
1861 static QemuOptsList qemu_iscsi_opts = {
1862 .name = "iscsi",
1863 .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1864 .desc = {
1865 {
1866 .name = "user",
1867 .type = QEMU_OPT_STRING,
1868 .help = "username for CHAP authentication to target",
1869 },{
1870 .name = "password",
1871 .type = QEMU_OPT_STRING,
1872 .help = "password for CHAP authentication to target",
1873 },{
1874 .name = "password-secret",
1875 .type = QEMU_OPT_STRING,
1876 .help = "ID of the secret providing password for CHAP "
1877 "authentication to target",
1878 },{
1879 .name = "header-digest",
1880 .type = QEMU_OPT_STRING,
1881 .help = "HeaderDigest setting. "
1882 "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1883 },{
1884 .name = "initiator-name",
1885 .type = QEMU_OPT_STRING,
1886 .help = "Initiator iqn name to use when connecting",
1887 },{
1888 .name = "timeout",
1889 .type = QEMU_OPT_NUMBER,
1890 .help = "Request timeout in seconds (default 0 = no timeout)",
1891 },
1892 { /* end of list */ }
1893 },
1894 };
1895
1896 static void iscsi_block_init(void)
1897 {
1898 bdrv_register(&bdrv_iscsi);
1899 qemu_add_opts(&qemu_iscsi_opts);
1900 }
1901
1902 block_init(iscsi_block_init);
QEmu