]> git.proxmox.com Git - mirror_qemu.git/blob - block/iscsi.c
projects / mirror_qemu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Use #include "..." for our own headers, <...> for others
[mirror_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) 2012-2015 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 #endif
50
51 typedef struct IscsiLun {
52 struct iscsi_context *iscsi;
53 AioContext *aio_context;
54 int lun;
55 enum scsi_inquiry_peripheral_device_type type;
56 int block_size;
57 uint64_t num_blocks;
58 int events;
59 QEMUTimer *nop_timer;
60 QEMUTimer *event_timer;
61 struct scsi_inquiry_logical_block_provisioning lbp;
62 struct scsi_inquiry_block_limits bl;
63 unsigned char *zeroblock;
64 unsigned long *allocationmap;
65 int cluster_sectors;
66 bool use_16_for_rw;
67 bool write_protected;
68 bool lbpme;
69 bool lbprz;
70 bool dpofua;
71 bool has_write_same;
72 bool request_timed_out;
73 } IscsiLun;
74
75 typedef struct IscsiTask {
76 int status;
77 int complete;
78 int retries;
79 int do_retry;
80 struct scsi_task *task;
81 Coroutine *co;
82 QEMUBH *bh;
83 IscsiLun *iscsilun;
84 QEMUTimer retry_timer;
85 int err_code;
86 } IscsiTask;
87
88 typedef struct IscsiAIOCB {
89 BlockAIOCB common;
90 QEMUIOVector *qiov;
91 QEMUBH *bh;
92 IscsiLun *iscsilun;
93 struct scsi_task *task;
94 uint8_t *buf;
95 int status;
96 int64_t sector_num;
97 int nb_sectors;
98 int ret;
99 #ifdef __linux__
100 sg_io_hdr_t *ioh;
101 #endif
102 } IscsiAIOCB;
103
104 /* libiscsi uses time_t so its enough to process events every second */
105 #define EVENT_INTERVAL 1000
106 #define NOP_INTERVAL 5000
107 #define MAX_NOP_FAILURES 3
108 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times)
109 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768};
110
111 /* this threshold is a trade-off knob to choose between
112 * the potential additional overhead of an extra GET_LBA_STATUS request
113 * vs. unnecessarily reading a lot of zero sectors over the wire.
114 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
115 * sectors we check the allocation status of the area covered by the
116 * request first if the allocationmap indicates that the area might be
117 * unallocated. */
118 #define ISCSI_CHECKALLOC_THRES 64
119
120 static void
121 iscsi_bh_cb(void *p)
122 {
123 IscsiAIOCB *acb = p;
124
125 qemu_bh_delete(acb->bh);
126
127 g_free(acb->buf);
128 acb->buf = NULL;
129
130 acb->common.cb(acb->common.opaque, acb->status);
131
132 if (acb->task != NULL) {
133 scsi_free_scsi_task(acb->task);
134 acb->task = NULL;
135 }
136
137 qemu_aio_unref(acb);
138 }
139
140 static void
141 iscsi_schedule_bh(IscsiAIOCB *acb)
142 {
143 if (acb->bh) {
144 return;
145 }
146 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb);
147 qemu_bh_schedule(acb->bh);
148 }
149
150 static void iscsi_co_generic_bh_cb(void *opaque)
151 {
152 struct IscsiTask *iTask = opaque;
153 iTask->complete = 1;
154 qemu_bh_delete(iTask->bh);
155 qemu_coroutine_enter(iTask->co, NULL);
156 }
157
158 static void iscsi_retry_timer_expired(void *opaque)
159 {
160 struct IscsiTask *iTask = opaque;
161 iTask->complete = 1;
162 if (iTask->co) {
163 qemu_coroutine_enter(iTask->co, NULL);
164 }
165 }
166
167 static inline unsigned exp_random(double mean)
168 {
169 return -mean * log((double)rand() / RAND_MAX);
170 }
171
172 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in
173 * libiscsi 1.10.0, together with other constants we need. Use it as
174 * a hint that we have to define them ourselves if needed, to keep the
175 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for
176 * the test because SCSI_STATUS_* is an enum.
177 *
178 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes
179 * an enum, check against the LIBISCSI_API_VERSION macro, which was
180 * introduced in 1.11.0. If it is present, there is no need to define
181 * anything.
182 */
183 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \
184 !defined(LIBISCSI_API_VERSION)
185 #define SCSI_STATUS_TASK_SET_FULL 0x28
186 #define SCSI_STATUS_TIMEOUT 0x0f000002
187 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600
188 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00
189 #endif
190
191 static int iscsi_translate_sense(struct scsi_sense *sense)
192 {
193 int ret;
194
195 switch (sense->key) {
196 case SCSI_SENSE_NOT_READY:
197 return -EBUSY;
198 case SCSI_SENSE_DATA_PROTECTION:
199 return -EACCES;
200 case SCSI_SENSE_COMMAND_ABORTED:
201 return -ECANCELED;
202 case SCSI_SENSE_ILLEGAL_REQUEST:
203 /* Parse ASCQ */
204 break;
205 default:
206 return -EIO;
207 }
208 switch (sense->ascq) {
209 case SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR:
210 case SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE:
211 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB:
212 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST:
213 ret = -EINVAL;
214 break;
215 case SCSI_SENSE_ASCQ_LBA_OUT_OF_RANGE:
216 ret = -ENOSPC;
217 break;
218 case SCSI_SENSE_ASCQ_LOGICAL_UNIT_NOT_SUPPORTED:
219 ret = -ENOTSUP;
220 break;
221 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT:
222 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_CLOSED:
223 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_OPEN:
224 ret = -ENOMEDIUM;
225 break;
226 case SCSI_SENSE_ASCQ_WRITE_PROTECTED:
227 ret = -EACCES;
228 break;
229 default:
230 ret = -EIO;
231 break;
232 }
233 return ret;
234 }
235
236 static void
237 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
238 void *command_data, void *opaque)
239 {
240 struct IscsiTask *iTask = opaque;
241 struct scsi_task *task = command_data;
242
243 iTask->status = status;
244 iTask->do_retry = 0;
245 iTask->task = task;
246
247 if (status != SCSI_STATUS_GOOD) {
248 if (iTask->retries++ < ISCSI_CMD_RETRIES) {
249 if (status == SCSI_STATUS_CHECK_CONDITION
250 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
251 error_report("iSCSI CheckCondition: %s",
252 iscsi_get_error(iscsi));
253 iTask->do_retry = 1;
254 goto out;
255 }
256 if (status == SCSI_STATUS_BUSY ||
257 status == SCSI_STATUS_TIMEOUT ||
258 status == SCSI_STATUS_TASK_SET_FULL) {
259 unsigned retry_time =
260 exp_random(iscsi_retry_times[iTask->retries - 1]);
261 if (status == SCSI_STATUS_TIMEOUT) {
262 /* make sure the request is rescheduled AFTER the
263 * reconnect is initiated */
264 retry_time = EVENT_INTERVAL * 2;
265 iTask->iscsilun->request_timed_out = true;
266 }
267 error_report("iSCSI Busy/TaskSetFull/TimeOut"
268 " (retry #%u in %u ms): %s",
269 iTask->retries, retry_time,
270 iscsi_get_error(iscsi));
271 aio_timer_init(iTask->iscsilun->aio_context,
272 &iTask->retry_timer, QEMU_CLOCK_REALTIME,
273 SCALE_MS, iscsi_retry_timer_expired, iTask);
274 timer_mod(&iTask->retry_timer,
275 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
276 iTask->do_retry = 1;
277 return;
278 }
279 }
280 iTask->err_code = iscsi_translate_sense(&task->sense);
281 error_report("iSCSI Failure: %s", iscsi_get_error(iscsi));
282 }
283
284 out:
285 if (iTask->co) {
286 iTask->bh = aio_bh_new(iTask->iscsilun->aio_context,
287 iscsi_co_generic_bh_cb, iTask);
288 qemu_bh_schedule(iTask->bh);
289 } else {
290 iTask->complete = 1;
291 }
292 }
293
294 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
295 {
296 *iTask = (struct IscsiTask) {
297 .co = qemu_coroutine_self(),
298 .iscsilun = iscsilun,
299 };
300 }
301
302 static void
303 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
304 void *private_data)
305 {
306 IscsiAIOCB *acb = private_data;
307
308 acb->status = -ECANCELED;
309 iscsi_schedule_bh(acb);
310 }
311
312 static void
313 iscsi_aio_cancel(BlockAIOCB *blockacb)
314 {
315 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
316 IscsiLun *iscsilun = acb->iscsilun;
317
318 if (acb->status != -EINPROGRESS) {
319 return;
320 }
321
322 /* send a task mgmt call to the target to cancel the task on the target */
323 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
324 iscsi_abort_task_cb, acb);
325
326 }
327
328 static const AIOCBInfo iscsi_aiocb_info = {
329 .aiocb_size = sizeof(IscsiAIOCB),
330 .cancel_async = iscsi_aio_cancel,
331 };
332
333
334 static void iscsi_process_read(void *arg);
335 static void iscsi_process_write(void *arg);
336
337 static void
338 iscsi_set_events(IscsiLun *iscsilun)
339 {
340 struct iscsi_context *iscsi = iscsilun->iscsi;
341 int ev = iscsi_which_events(iscsi);
342
343 if (ev != iscsilun->events) {
344 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi),
345 false,
346 (ev & POLLIN) ? iscsi_process_read : NULL,
347 (ev & POLLOUT) ? iscsi_process_write : NULL,
348 iscsilun);
349 iscsilun->events = ev;
350 }
351 }
352
353 static void iscsi_timed_check_events(void *opaque)
354 {
355 IscsiLun *iscsilun = opaque;
356
357 /* check for timed out requests */
358 iscsi_service(iscsilun->iscsi, 0);
359
360 if (iscsilun->request_timed_out) {
361 iscsilun->request_timed_out = false;
362 iscsi_reconnect(iscsilun->iscsi);
363 }
364
365 /* newer versions of libiscsi may return zero events. Ensure we are able
366 * to return to service once this situation changes. */
367 iscsi_set_events(iscsilun);
368
369 timer_mod(iscsilun->event_timer,
370 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
371 }
372
373 static void
374 iscsi_process_read(void *arg)
375 {
376 IscsiLun *iscsilun = arg;
377 struct iscsi_context *iscsi = iscsilun->iscsi;
378
379 iscsi_service(iscsi, POLLIN);
380 iscsi_set_events(iscsilun);
381 }
382
383 static void
384 iscsi_process_write(void *arg)
385 {
386 IscsiLun *iscsilun = arg;
387 struct iscsi_context *iscsi = iscsilun->iscsi;
388
389 iscsi_service(iscsi, POLLOUT);
390 iscsi_set_events(iscsilun);
391 }
392
393 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
394 {
395 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
396 }
397
398 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
399 {
400 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
401 }
402
403 static bool is_byte_request_lun_aligned(int64_t offset, int count,
404 IscsiLun *iscsilun)
405 {
406 if (offset % iscsilun->block_size || count % iscsilun->block_size) {
407 error_report("iSCSI misaligned request: "
408 "iscsilun->block_size %u, offset %" PRIi64
409 ", count %d",
410 iscsilun->block_size, offset, count);
411 return false;
412 }
413 return true;
414 }
415
416 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors,
417 IscsiLun *iscsilun)
418 {
419 assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS);
420 return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS,
421 nb_sectors << BDRV_SECTOR_BITS,
422 iscsilun);
423 }
424
425 static unsigned long *iscsi_allocationmap_init(IscsiLun *iscsilun)
426 {
427 return bitmap_try_new(DIV_ROUND_UP(sector_lun2qemu(iscsilun->num_blocks,
428 iscsilun),
429 iscsilun->cluster_sectors));
430 }
431
432 static void iscsi_allocationmap_set(IscsiLun *iscsilun, int64_t sector_num,
433 int nb_sectors)
434 {
435 if (iscsilun->allocationmap == NULL) {
436 return;
437 }
438 bitmap_set(iscsilun->allocationmap,
439 sector_num / iscsilun->cluster_sectors,
440 DIV_ROUND_UP(nb_sectors, iscsilun->cluster_sectors));
441 }
442
443 static void iscsi_allocationmap_clear(IscsiLun *iscsilun, int64_t sector_num,
444 int nb_sectors)
445 {
446 int64_t cluster_num, nb_clusters;
447 if (iscsilun->allocationmap == NULL) {
448 return;
449 }
450 cluster_num = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors);
451 nb_clusters = (sector_num + nb_sectors) / iscsilun->cluster_sectors
452 - cluster_num;
453 if (nb_clusters > 0) {
454 bitmap_clear(iscsilun->allocationmap, cluster_num, nb_clusters);
455 }
456 }
457
458 static int coroutine_fn
459 iscsi_co_writev_flags(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
460 QEMUIOVector *iov, int flags)
461 {
462 IscsiLun *iscsilun = bs->opaque;
463 struct IscsiTask iTask;
464 uint64_t lba;
465 uint32_t num_sectors;
466 bool fua = flags & BDRV_REQ_FUA;
467
468 if (fua) {
469 assert(iscsilun->dpofua);
470 }
471 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
472 return -EINVAL;
473 }
474
475 if (bs->bl.max_transfer &&
476 nb_sectors << BDRV_SECTOR_BITS > bs->bl.max_transfer) {
477 error_report("iSCSI Error: Write of %d sectors exceeds max_xfer_len "
478 "of %" PRIu32 " bytes", nb_sectors, bs->bl.max_transfer);
479 return -EINVAL;
480 }
481
482 lba = sector_qemu2lun(sector_num, iscsilun);
483 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
484 iscsi_co_init_iscsitask(iscsilun, &iTask);
485 retry:
486 if (iscsilun->use_16_for_rw) {
487 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
488 NULL, num_sectors * iscsilun->block_size,
489 iscsilun->block_size, 0, 0, fua, 0, 0,
490 iscsi_co_generic_cb, &iTask);
491 } else {
492 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
493 NULL, num_sectors * iscsilun->block_size,
494 iscsilun->block_size, 0, 0, fua, 0, 0,
495 iscsi_co_generic_cb, &iTask);
496 }
497 if (iTask.task == NULL) {
498 return -ENOMEM;
499 }
500 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
501 iov->niov);
502 while (!iTask.complete) {
503 iscsi_set_events(iscsilun);
504 qemu_coroutine_yield();
505 }
506
507 if (iTask.task != NULL) {
508 scsi_free_scsi_task(iTask.task);
509 iTask.task = NULL;
510 }
511
512 if (iTask.do_retry) {
513 iTask.complete = 0;
514 goto retry;
515 }
516
517 if (iTask.status != SCSI_STATUS_GOOD) {
518 return iTask.err_code;
519 }
520
521 iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors);
522
523 return 0;
524 }
525
526
527 static bool iscsi_allocationmap_is_allocated(IscsiLun *iscsilun,
528 int64_t sector_num, int nb_sectors)
529 {
530 unsigned long size;
531 if (iscsilun->allocationmap == NULL) {
532 return true;
533 }
534 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
535 return !(find_next_bit(iscsilun->allocationmap, size,
536 sector_num / iscsilun->cluster_sectors) == size);
537 }
538
539 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
540 int64_t sector_num,
541 int nb_sectors, int *pnum,
542 BlockDriverState **file)
543 {
544 IscsiLun *iscsilun = bs->opaque;
545 struct scsi_get_lba_status *lbas = NULL;
546 struct scsi_lba_status_descriptor *lbasd = NULL;
547 struct IscsiTask iTask;
548 int64_t ret;
549
550 iscsi_co_init_iscsitask(iscsilun, &iTask);
551
552 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
553 ret = -EINVAL;
554 goto out;
555 }
556
557 /* default to all sectors allocated */
558 ret = BDRV_BLOCK_DATA;
559 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
560 *pnum = nb_sectors;
561
562 /* LUN does not support logical block provisioning */
563 if (!iscsilun->lbpme) {
564 goto out;
565 }
566
567 retry:
568 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
569 sector_qemu2lun(sector_num, iscsilun),
570 8 + 16, iscsi_co_generic_cb,
571 &iTask) == NULL) {
572 ret = -ENOMEM;
573 goto out;
574 }
575
576 while (!iTask.complete) {
577 iscsi_set_events(iscsilun);
578 qemu_coroutine_yield();
579 }
580
581 if (iTask.do_retry) {
582 if (iTask.task != NULL) {
583 scsi_free_scsi_task(iTask.task);
584 iTask.task = NULL;
585 }
586 iTask.complete = 0;
587 goto retry;
588 }
589
590 if (iTask.status != SCSI_STATUS_GOOD) {
591 /* in case the get_lba_status_callout fails (i.e.
592 * because the device is busy or the cmd is not
593 * supported) we pretend all blocks are allocated
594 * for backwards compatibility */
595 goto out;
596 }
597
598 lbas = scsi_datain_unmarshall(iTask.task);
599 if (lbas == NULL) {
600 ret = -EIO;
601 goto out;
602 }
603
604 lbasd = &lbas->descriptors[0];
605
606 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
607 ret = -EIO;
608 goto out;
609 }
610
611 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
612
613 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
614 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
615 ret &= ~BDRV_BLOCK_DATA;
616 if (iscsilun->lbprz) {
617 ret |= BDRV_BLOCK_ZERO;
618 }
619 }
620
621 if (ret & BDRV_BLOCK_ZERO) {
622 iscsi_allocationmap_clear(iscsilun, sector_num, *pnum);
623 } else {
624 iscsi_allocationmap_set(iscsilun, sector_num, *pnum);
625 }
626
627 if (*pnum > nb_sectors) {
628 *pnum = nb_sectors;
629 }
630 out:
631 if (iTask.task != NULL) {
632 scsi_free_scsi_task(iTask.task);
633 }
634 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) {
635 *file = bs;
636 }
637 return ret;
638 }
639
640 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
641 int64_t sector_num, int nb_sectors,
642 QEMUIOVector *iov)
643 {
644 IscsiLun *iscsilun = bs->opaque;
645 struct IscsiTask iTask;
646 uint64_t lba;
647 uint32_t num_sectors;
648
649 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
650 return -EINVAL;
651 }
652
653 if (bs->bl.max_transfer &&
654 nb_sectors << BDRV_SECTOR_BITS > bs->bl.max_transfer) {
655 error_report("iSCSI Error: Read of %d sectors exceeds max_xfer_len "
656 "of %" PRIu32 " bytes", nb_sectors, bs->bl.max_transfer);
657 return -EINVAL;
658 }
659
660 if (iscsilun->lbprz && nb_sectors >= ISCSI_CHECKALLOC_THRES &&
661 !iscsi_allocationmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
662 int64_t ret;
663 int pnum;
664 BlockDriverState *file;
665 ret = iscsi_co_get_block_status(bs, sector_num,
666 BDRV_REQUEST_MAX_SECTORS, &pnum, &file);
667 if (ret < 0) {
668 return ret;
669 }
670 if (ret & BDRV_BLOCK_ZERO && pnum >= nb_sectors) {
671 qemu_iovec_memset(iov, 0, 0x00, iov->size);
672 return 0;
673 }
674 }
675
676 lba = sector_qemu2lun(sector_num, iscsilun);
677 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
678
679 iscsi_co_init_iscsitask(iscsilun, &iTask);
680 retry:
681 if (iscsilun->use_16_for_rw) {
682 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
683 num_sectors * iscsilun->block_size,
684 iscsilun->block_size, 0, 0, 0, 0, 0,
685 iscsi_co_generic_cb, &iTask);
686 } else {
687 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
688 num_sectors * iscsilun->block_size,
689 iscsilun->block_size,
690 0, 0, 0, 0, 0,
691 iscsi_co_generic_cb, &iTask);
692 }
693 if (iTask.task == NULL) {
694 return -ENOMEM;
695 }
696 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
697
698 while (!iTask.complete) {
699 iscsi_set_events(iscsilun);
700 qemu_coroutine_yield();
701 }
702
703 if (iTask.task != NULL) {
704 scsi_free_scsi_task(iTask.task);
705 iTask.task = NULL;
706 }
707
708 if (iTask.do_retry) {
709 iTask.complete = 0;
710 goto retry;
711 }
712
713 if (iTask.status != SCSI_STATUS_GOOD) {
714 return iTask.err_code;
715 }
716
717 return 0;
718 }
719
720 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
721 {
722 IscsiLun *iscsilun = bs->opaque;
723 struct IscsiTask iTask;
724
725 iscsi_co_init_iscsitask(iscsilun, &iTask);
726 retry:
727 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
728 0, iscsi_co_generic_cb, &iTask) == NULL) {
729 return -ENOMEM;
730 }
731
732 while (!iTask.complete) {
733 iscsi_set_events(iscsilun);
734 qemu_coroutine_yield();
735 }
736
737 if (iTask.task != NULL) {
738 scsi_free_scsi_task(iTask.task);
739 iTask.task = NULL;
740 }
741
742 if (iTask.do_retry) {
743 iTask.complete = 0;
744 goto retry;
745 }
746
747 if (iTask.status != SCSI_STATUS_GOOD) {
748 return iTask.err_code;
749 }
750
751 return 0;
752 }
753
754 #ifdef __linux__
755 static void
756 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
757 void *command_data, void *opaque)
758 {
759 IscsiAIOCB *acb = opaque;
760
761 g_free(acb->buf);
762 acb->buf = NULL;
763
764 acb->status = 0;
765 if (status < 0) {
766 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
767 iscsi_get_error(iscsi));
768 acb->status = iscsi_translate_sense(&acb->task->sense);
769 }
770
771 acb->ioh->driver_status = 0;
772 acb->ioh->host_status = 0;
773 acb->ioh->resid = 0;
774 acb->ioh->status = status;
775
776 #define SG_ERR_DRIVER_SENSE 0x08
777
778 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
779 int ss;
780
781 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
782
783 acb->ioh->sb_len_wr = acb->task->datain.size - 2;
784 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
785 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
786 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
787 }
788
789 iscsi_schedule_bh(acb);
790 }
791
792 static void iscsi_ioctl_bh_completion(void *opaque)
793 {
794 IscsiAIOCB *acb = opaque;
795
796 qemu_bh_delete(acb->bh);
797 acb->common.cb(acb->common.opaque, acb->ret);
798 qemu_aio_unref(acb);
799 }
800
801 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
802 {
803 BlockDriverState *bs = acb->common.bs;
804 IscsiLun *iscsilun = bs->opaque;
805 int ret = 0;
806
807 switch (req) {
808 case SG_GET_VERSION_NUM:
809 *(int *)buf = 30000;
810 break;
811 case SG_GET_SCSI_ID:
812 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
813 break;
814 default:
815 ret = -EINVAL;
816 }
817 assert(!acb->bh);
818 acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
819 iscsi_ioctl_bh_completion, acb);
820 acb->ret = ret;
821 qemu_bh_schedule(acb->bh);
822 }
823
824 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
825 unsigned long int req, void *buf,
826 BlockCompletionFunc *cb, void *opaque)
827 {
828 IscsiLun *iscsilun = bs->opaque;
829 struct iscsi_context *iscsi = iscsilun->iscsi;
830 struct iscsi_data data;
831 IscsiAIOCB *acb;
832
833 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
834
835 acb->iscsilun = iscsilun;
836 acb->bh = NULL;
837 acb->status = -EINPROGRESS;
838 acb->buf = NULL;
839 acb->ioh = buf;
840
841 if (req != SG_IO) {
842 iscsi_ioctl_handle_emulated(acb, req, buf);
843 return &acb->common;
844 }
845
846 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) {
847 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)",
848 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE);
849 qemu_aio_unref(acb);
850 return NULL;
851 }
852
853 acb->task = malloc(sizeof(struct scsi_task));
854 if (acb->task == NULL) {
855 error_report("iSCSI: Failed to allocate task for scsi command. %s",
856 iscsi_get_error(iscsi));
857 qemu_aio_unref(acb);
858 return NULL;
859 }
860 memset(acb->task, 0, sizeof(struct scsi_task));
861
862 switch (acb->ioh->dxfer_direction) {
863 case SG_DXFER_TO_DEV:
864 acb->task->xfer_dir = SCSI_XFER_WRITE;
865 break;
866 case SG_DXFER_FROM_DEV:
867 acb->task->xfer_dir = SCSI_XFER_READ;
868 break;
869 default:
870 acb->task->xfer_dir = SCSI_XFER_NONE;
871 break;
872 }
873
874 acb->task->cdb_size = acb->ioh->cmd_len;
875 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
876 acb->task->expxferlen = acb->ioh->dxfer_len;
877
878 data.size = 0;
879 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
880 if (acb->ioh->iovec_count == 0) {
881 data.data = acb->ioh->dxferp;
882 data.size = acb->ioh->dxfer_len;
883 } else {
884 scsi_task_set_iov_out(acb->task,
885 (struct scsi_iovec *) acb->ioh->dxferp,
886 acb->ioh->iovec_count);
887 }
888 }
889
890 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
891 iscsi_aio_ioctl_cb,
892 (data.size > 0) ? &data : NULL,
893 acb) != 0) {
894 scsi_free_scsi_task(acb->task);
895 qemu_aio_unref(acb);
896 return NULL;
897 }
898
899 /* tell libiscsi to read straight into the buffer we got from ioctl */
900 if (acb->task->xfer_dir == SCSI_XFER_READ) {
901 if (acb->ioh->iovec_count == 0) {
902 scsi_task_add_data_in_buffer(acb->task,
903 acb->ioh->dxfer_len,
904 acb->ioh->dxferp);
905 } else {
906 scsi_task_set_iov_in(acb->task,
907 (struct scsi_iovec *) acb->ioh->dxferp,
908 acb->ioh->iovec_count);
909 }
910 }
911
912 iscsi_set_events(iscsilun);
913
914 return &acb->common;
915 }
916
917 #endif
918
919 static int64_t
920 iscsi_getlength(BlockDriverState *bs)
921 {
922 IscsiLun *iscsilun = bs->opaque;
923 int64_t len;
924
925 len = iscsilun->num_blocks;
926 len *= iscsilun->block_size;
927
928 return len;
929 }
930
931 static int
932 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num,
933 int nb_sectors)
934 {
935 IscsiLun *iscsilun = bs->opaque;
936 struct IscsiTask iTask;
937 struct unmap_list list;
938
939 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
940 return -EINVAL;
941 }
942
943 if (!iscsilun->lbp.lbpu) {
944 /* UNMAP is not supported by the target */
945 return 0;
946 }
947
948 list.lba = sector_qemu2lun(sector_num, iscsilun);
949 list.num = sector_qemu2lun(nb_sectors, iscsilun);
950
951 iscsi_co_init_iscsitask(iscsilun, &iTask);
952 retry:
953 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
954 iscsi_co_generic_cb, &iTask) == NULL) {
955 return -ENOMEM;
956 }
957
958 while (!iTask.complete) {
959 iscsi_set_events(iscsilun);
960 qemu_coroutine_yield();
961 }
962
963 if (iTask.task != NULL) {
964 scsi_free_scsi_task(iTask.task);
965 iTask.task = NULL;
966 }
967
968 if (iTask.do_retry) {
969 iTask.complete = 0;
970 goto retry;
971 }
972
973 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
974 /* the target might fail with a check condition if it
975 is not happy with the alignment of the UNMAP request
976 we silently fail in this case */
977 return 0;
978 }
979
980 if (iTask.status != SCSI_STATUS_GOOD) {
981 return iTask.err_code;
982 }
983
984 iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors);
985
986 return 0;
987 }
988
989 static int
990 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
991 int count, BdrvRequestFlags flags)
992 {
993 IscsiLun *iscsilun = bs->opaque;
994 struct IscsiTask iTask;
995 uint64_t lba;
996 uint32_t nb_blocks;
997 bool use_16_for_ws = iscsilun->use_16_for_rw;
998
999 if (!is_byte_request_lun_aligned(offset, count, iscsilun)) {
1000 return -ENOTSUP;
1001 }
1002
1003 if (flags & BDRV_REQ_MAY_UNMAP) {
1004 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
1005 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
1006 use_16_for_ws = true;
1007 }
1008 if (use_16_for_ws && !iscsilun->lbp.lbpws) {
1009 /* WRITESAME16 with UNMAP is not supported by the target,
1010 * fall back and try WRITESAME10/16 without UNMAP */
1011 flags &= ~BDRV_REQ_MAY_UNMAP;
1012 use_16_for_ws = iscsilun->use_16_for_rw;
1013 }
1014 }
1015
1016 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1017 /* WRITESAME without UNMAP is not supported by the target */
1018 return -ENOTSUP;
1019 }
1020
1021 lba = offset / iscsilun->block_size;
1022 nb_blocks = count / iscsilun->block_size;
1023
1024 if (iscsilun->zeroblock == NULL) {
1025 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1026 if (iscsilun->zeroblock == NULL) {
1027 return -ENOMEM;
1028 }
1029 }
1030
1031 iscsi_co_init_iscsitask(iscsilun, &iTask);
1032 retry:
1033 if (use_16_for_ws) {
1034 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1035 iscsilun->zeroblock, iscsilun->block_size,
1036 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1037 0, 0, iscsi_co_generic_cb, &iTask);
1038 } else {
1039 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1040 iscsilun->zeroblock, iscsilun->block_size,
1041 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1042 0, 0, iscsi_co_generic_cb, &iTask);
1043 }
1044 if (iTask.task == NULL) {
1045 return -ENOMEM;
1046 }
1047
1048 while (!iTask.complete) {
1049 iscsi_set_events(iscsilun);
1050 qemu_coroutine_yield();
1051 }
1052
1053 if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1054 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1055 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1056 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1057 /* WRITE SAME is not supported by the target */
1058 iscsilun->has_write_same = false;
1059 scsi_free_scsi_task(iTask.task);
1060 return -ENOTSUP;
1061 }
1062
1063 if (iTask.task != NULL) {
1064 scsi_free_scsi_task(iTask.task);
1065 iTask.task = NULL;
1066 }
1067
1068 if (iTask.do_retry) {
1069 iTask.complete = 0;
1070 goto retry;
1071 }
1072
1073 if (iTask.status != SCSI_STATUS_GOOD) {
1074 return iTask.err_code;
1075 }
1076
1077 if (flags & BDRV_REQ_MAY_UNMAP) {
1078 iscsi_allocationmap_clear(iscsilun, offset >> BDRV_SECTOR_BITS,
1079 count >> BDRV_SECTOR_BITS);
1080 } else {
1081 iscsi_allocationmap_set(iscsilun, offset >> BDRV_SECTOR_BITS,
1082 count >> BDRV_SECTOR_BITS);
1083 }
1084
1085 return 0;
1086 }
1087
1088 static void parse_chap(struct iscsi_context *iscsi, const char *target,
1089 Error **errp)
1090 {
1091 QemuOptsList *list;
1092 QemuOpts *opts;
1093 const char *user = NULL;
1094 const char *password = NULL;
1095 const char *secretid;
1096 char *secret = NULL;
1097
1098 list = qemu_find_opts("iscsi");
1099 if (!list) {
1100 return;
1101 }
1102
1103 opts = qemu_opts_find(list, target);
1104 if (opts == NULL) {
1105 opts = QTAILQ_FIRST(&list->head);
1106 if (!opts) {
1107 return;
1108 }
1109 }
1110
1111 user = qemu_opt_get(opts, "user");
1112 if (!user) {
1113 return;
1114 }
1115
1116 secretid = qemu_opt_get(opts, "password-secret");
1117 password = qemu_opt_get(opts, "password");
1118 if (secretid && password) {
1119 error_setg(errp, "'password' and 'password-secret' properties are "
1120 "mutually exclusive");
1121 return;
1122 }
1123 if (secretid) {
1124 secret = qcrypto_secret_lookup_as_utf8(secretid, errp);
1125 if (!secret) {
1126 return;
1127 }
1128 password = secret;
1129 } else if (!password) {
1130 error_setg(errp, "CHAP username specified but no password was given");
1131 return;
1132 }
1133
1134 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1135 error_setg(errp, "Failed to set initiator username and password");
1136 }
1137
1138 g_free(secret);
1139 }
1140
1141 static void parse_header_digest(struct iscsi_context *iscsi, const char *target,
1142 Error **errp)
1143 {
1144 QemuOptsList *list;
1145 QemuOpts *opts;
1146 const char *digest = NULL;
1147
1148 list = qemu_find_opts("iscsi");
1149 if (!list) {
1150 return;
1151 }
1152
1153 opts = qemu_opts_find(list, target);
1154 if (opts == NULL) {
1155 opts = QTAILQ_FIRST(&list->head);
1156 if (!opts) {
1157 return;
1158 }
1159 }
1160
1161 digest = qemu_opt_get(opts, "header-digest");
1162 if (!digest) {
1163 return;
1164 }
1165
1166 if (!strcmp(digest, "CRC32C")) {
1167 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1168 } else if (!strcmp(digest, "NONE")) {
1169 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1170 } else if (!strcmp(digest, "CRC32C-NONE")) {
1171 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1172 } else if (!strcmp(digest, "NONE-CRC32C")) {
1173 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1174 } else {
1175 error_setg(errp, "Invalid header-digest setting : %s", digest);
1176 }
1177 }
1178
1179 static char *parse_initiator_name(const char *target)
1180 {
1181 QemuOptsList *list;
1182 QemuOpts *opts;
1183 const char *name;
1184 char *iscsi_name;
1185 UuidInfo *uuid_info;
1186
1187 list = qemu_find_opts("iscsi");
1188 if (list) {
1189 opts = qemu_opts_find(list, target);
1190 if (!opts) {
1191 opts = QTAILQ_FIRST(&list->head);
1192 }
1193 if (opts) {
1194 name = qemu_opt_get(opts, "initiator-name");
1195 if (name) {
1196 return g_strdup(name);
1197 }
1198 }
1199 }
1200
1201 uuid_info = qmp_query_uuid(NULL);
1202 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1203 name = qemu_get_vm_name();
1204 } else {
1205 name = uuid_info->UUID;
1206 }
1207 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1208 name ? ":" : "", name ? name : "");
1209 qapi_free_UuidInfo(uuid_info);
1210 return iscsi_name;
1211 }
1212
1213 static int parse_timeout(const char *target)
1214 {
1215 QemuOptsList *list;
1216 QemuOpts *opts;
1217 const char *timeout;
1218
1219 list = qemu_find_opts("iscsi");
1220 if (list) {
1221 opts = qemu_opts_find(list, target);
1222 if (!opts) {
1223 opts = QTAILQ_FIRST(&list->head);
1224 }
1225 if (opts) {
1226 timeout = qemu_opt_get(opts, "timeout");
1227 if (timeout) {
1228 return atoi(timeout);
1229 }
1230 }
1231 }
1232
1233 return 0;
1234 }
1235
1236 static void iscsi_nop_timed_event(void *opaque)
1237 {
1238 IscsiLun *iscsilun = opaque;
1239
1240 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1241 error_report("iSCSI: NOP timeout. Reconnecting...");
1242 iscsilun->request_timed_out = true;
1243 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1244 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1245 return;
1246 }
1247
1248 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1249 iscsi_set_events(iscsilun);
1250 }
1251
1252 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1253 {
1254 struct scsi_task *task = NULL;
1255 struct scsi_readcapacity10 *rc10 = NULL;
1256 struct scsi_readcapacity16 *rc16 = NULL;
1257 int retries = ISCSI_CMD_RETRIES;
1258
1259 do {
1260 if (task != NULL) {
1261 scsi_free_scsi_task(task);
1262 task = NULL;
1263 }
1264
1265 switch (iscsilun->type) {
1266 case TYPE_DISK:
1267 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1268 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1269 rc16 = scsi_datain_unmarshall(task);
1270 if (rc16 == NULL) {
1271 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1272 } else {
1273 iscsilun->block_size = rc16->block_length;
1274 iscsilun->num_blocks = rc16->returned_lba + 1;
1275 iscsilun->lbpme = !!rc16->lbpme;
1276 iscsilun->lbprz = !!rc16->lbprz;
1277 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1278 }
1279 break;
1280 }
1281 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1282 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1283 break;
1284 }
1285 /* Fall through and try READ CAPACITY(10) instead. */
1286 case TYPE_ROM:
1287 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1288 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1289 rc10 = scsi_datain_unmarshall(task);
1290 if (rc10 == NULL) {
1291 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1292 } else {
1293 iscsilun->block_size = rc10->block_size;
1294 if (rc10->lba == 0) {
1295 /* blank disk loaded */
1296 iscsilun->num_blocks = 0;
1297 } else {
1298 iscsilun->num_blocks = rc10->lba + 1;
1299 }
1300 }
1301 }
1302 break;
1303 default:
1304 return;
1305 }
1306 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1307 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1308 && retries-- > 0);
1309
1310 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1311 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1312 } else if (!iscsilun->block_size ||
1313 iscsilun->block_size % BDRV_SECTOR_SIZE) {
1314 error_setg(errp, "iSCSI: the target returned an invalid "
1315 "block size of %d.", iscsilun->block_size);
1316 }
1317 if (task) {
1318 scsi_free_scsi_task(task);
1319 }
1320 }
1321
1322 /* TODO Convert to fine grained options */
1323 static QemuOptsList runtime_opts = {
1324 .name = "iscsi",
1325 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1326 .desc = {
1327 {
1328 .name = "filename",
1329 .type = QEMU_OPT_STRING,
1330 .help = "URL to the iscsi image",
1331 },
1332 { /* end of list */ }
1333 },
1334 };
1335
1336 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1337 int evpd, int pc, void **inq, Error **errp)
1338 {
1339 int full_size;
1340 struct scsi_task *task = NULL;
1341 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1342 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1343 goto fail;
1344 }
1345 full_size = scsi_datain_getfullsize(task);
1346 if (full_size > task->datain.size) {
1347 scsi_free_scsi_task(task);
1348
1349 /* we need more data for the full list */
1350 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1351 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1352 goto fail;
1353 }
1354 }
1355
1356 *inq = scsi_datain_unmarshall(task);
1357 if (*inq == NULL) {
1358 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1359 goto fail_with_err;
1360 }
1361
1362 return task;
1363
1364 fail:
1365 error_setg(errp, "iSCSI: Inquiry command failed : %s",
1366 iscsi_get_error(iscsi));
1367 fail_with_err:
1368 if (task != NULL) {
1369 scsi_free_scsi_task(task);
1370 }
1371 return NULL;
1372 }
1373
1374 static void iscsi_detach_aio_context(BlockDriverState *bs)
1375 {
1376 IscsiLun *iscsilun = bs->opaque;
1377
1378 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1379 false, NULL, NULL, NULL);
1380 iscsilun->events = 0;
1381
1382 if (iscsilun->nop_timer) {
1383 timer_del(iscsilun->nop_timer);
1384 timer_free(iscsilun->nop_timer);
1385 iscsilun->nop_timer = NULL;
1386 }
1387 if (iscsilun->event_timer) {
1388 timer_del(iscsilun->event_timer);
1389 timer_free(iscsilun->event_timer);
1390 iscsilun->event_timer = NULL;
1391 }
1392 }
1393
1394 static void iscsi_attach_aio_context(BlockDriverState *bs,
1395 AioContext *new_context)
1396 {
1397 IscsiLun *iscsilun = bs->opaque;
1398
1399 iscsilun->aio_context = new_context;
1400 iscsi_set_events(iscsilun);
1401
1402 /* Set up a timer for sending out iSCSI NOPs */
1403 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1404 QEMU_CLOCK_REALTIME, SCALE_MS,
1405 iscsi_nop_timed_event, iscsilun);
1406 timer_mod(iscsilun->nop_timer,
1407 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1408
1409 /* Set up a timer for periodic calls to iscsi_set_events and to
1410 * scan for command timeout */
1411 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1412 QEMU_CLOCK_REALTIME, SCALE_MS,
1413 iscsi_timed_check_events, iscsilun);
1414 timer_mod(iscsilun->event_timer,
1415 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1416 }
1417
1418 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1419 {
1420 struct scsi_task *task;
1421 struct scsi_mode_sense *ms = NULL;
1422 iscsilun->write_protected = false;
1423 iscsilun->dpofua = false;
1424
1425 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1426 1, SCSI_MODESENSE_PC_CURRENT,
1427 0x3F, 0, 255);
1428 if (task == NULL) {
1429 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1430 iscsi_get_error(iscsilun->iscsi));
1431 goto out;
1432 }
1433
1434 if (task->status != SCSI_STATUS_GOOD) {
1435 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1436 goto out;
1437 }
1438 ms = scsi_datain_unmarshall(task);
1439 if (!ms) {
1440 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1441 iscsi_get_error(iscsilun->iscsi));
1442 goto out;
1443 }
1444 iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1445 iscsilun->dpofua = ms->device_specific_parameter & 0x10;
1446
1447 out:
1448 if (task) {
1449 scsi_free_scsi_task(task);
1450 }
1451 }
1452
1453 /*
1454 * We support iscsi url's on the form
1455 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1456 */
1457 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1458 Error **errp)
1459 {
1460 IscsiLun *iscsilun = bs->opaque;
1461 struct iscsi_context *iscsi = NULL;
1462 struct iscsi_url *iscsi_url = NULL;
1463 struct scsi_task *task = NULL;
1464 struct scsi_inquiry_standard *inq = NULL;
1465 struct scsi_inquiry_supported_pages *inq_vpd;
1466 char *initiator_name = NULL;
1467 QemuOpts *opts;
1468 Error *local_err = NULL;
1469 const char *filename;
1470 int i, ret = 0, timeout = 0;
1471
1472 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1473 qemu_opts_absorb_qdict(opts, options, &local_err);
1474 if (local_err) {
1475 error_propagate(errp, local_err);
1476 ret = -EINVAL;
1477 goto out;
1478 }
1479
1480 filename = qemu_opt_get(opts, "filename");
1481
1482 iscsi_url = iscsi_parse_full_url(iscsi, filename);
1483 if (iscsi_url == NULL) {
1484 error_setg(errp, "Failed to parse URL : %s", filename);
1485 ret = -EINVAL;
1486 goto out;
1487 }
1488
1489 memset(iscsilun, 0, sizeof(IscsiLun));
1490
1491 initiator_name = parse_initiator_name(iscsi_url->target);
1492
1493 iscsi = iscsi_create_context(initiator_name);
1494 if (iscsi == NULL) {
1495 error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1496 ret = -ENOMEM;
1497 goto out;
1498 }
1499
1500 if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1501 error_setg(errp, "iSCSI: Failed to set target name.");
1502 ret = -EINVAL;
1503 goto out;
1504 }
1505
1506 if (iscsi_url->user[0] != '\0') {
1507 ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1508 iscsi_url->passwd);
1509 if (ret != 0) {
1510 error_setg(errp, "Failed to set initiator username and password");
1511 ret = -EINVAL;
1512 goto out;
1513 }
1514 }
1515
1516 /* check if we got CHAP username/password via the options */
1517 parse_chap(iscsi, iscsi_url->target, &local_err);
1518 if (local_err != NULL) {
1519 error_propagate(errp, local_err);
1520 ret = -EINVAL;
1521 goto out;
1522 }
1523
1524 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1525 error_setg(errp, "iSCSI: Failed to set session type to normal.");
1526 ret = -EINVAL;
1527 goto out;
1528 }
1529
1530 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1531
1532 /* check if we got HEADER_DIGEST via the options */
1533 parse_header_digest(iscsi, iscsi_url->target, &local_err);
1534 if (local_err != NULL) {
1535 error_propagate(errp, local_err);
1536 ret = -EINVAL;
1537 goto out;
1538 }
1539
1540 /* timeout handling is broken in libiscsi before 1.15.0 */
1541 timeout = parse_timeout(iscsi_url->target);
1542 #if defined(LIBISCSI_API_VERSION) && LIBISCSI_API_VERSION >= 20150621
1543 iscsi_set_timeout(iscsi, timeout);
1544 #else
1545 if (timeout) {
1546 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1547 }
1548 #endif
1549
1550 if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1551 error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1552 iscsi_get_error(iscsi));
1553 ret = -EINVAL;
1554 goto out;
1555 }
1556
1557 iscsilun->iscsi = iscsi;
1558 iscsilun->aio_context = bdrv_get_aio_context(bs);
1559 iscsilun->lun = iscsi_url->lun;
1560 iscsilun->has_write_same = true;
1561
1562 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1563 (void **) &inq, errp);
1564 if (task == NULL) {
1565 ret = -EINVAL;
1566 goto out;
1567 }
1568 iscsilun->type = inq->periperal_device_type;
1569 scsi_free_scsi_task(task);
1570 task = NULL;
1571
1572 iscsi_modesense_sync(iscsilun);
1573 if (iscsilun->dpofua) {
1574 bs->supported_write_flags = BDRV_REQ_FUA;
1575 }
1576 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
1577
1578 /* Check the write protect flag of the LUN if we want to write */
1579 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1580 iscsilun->write_protected) {
1581 error_setg(errp, "Cannot open a write protected LUN as read-write");
1582 ret = -EACCES;
1583 goto out;
1584 }
1585
1586 iscsi_readcapacity_sync(iscsilun, &local_err);
1587 if (local_err != NULL) {
1588 error_propagate(errp, local_err);
1589 ret = -EINVAL;
1590 goto out;
1591 }
1592 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1593
1594 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1595 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1596 * will try to read from the device to guess the image format.
1597 */
1598 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1599 bs->sg = true;
1600 }
1601
1602 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1603 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1604 (void **) &inq_vpd, errp);
1605 if (task == NULL) {
1606 ret = -EINVAL;
1607 goto out;
1608 }
1609 for (i = 0; i < inq_vpd->num_pages; i++) {
1610 struct scsi_task *inq_task;
1611 struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1612 struct scsi_inquiry_block_limits *inq_bl;
1613 switch (inq_vpd->pages[i]) {
1614 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1615 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1616 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1617 (void **) &inq_lbp, errp);
1618 if (inq_task == NULL) {
1619 ret = -EINVAL;
1620 goto out;
1621 }
1622 memcpy(&iscsilun->lbp, inq_lbp,
1623 sizeof(struct scsi_inquiry_logical_block_provisioning));
1624 scsi_free_scsi_task(inq_task);
1625 break;
1626 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1627 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1628 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1629 (void **) &inq_bl, errp);
1630 if (inq_task == NULL) {
1631 ret = -EINVAL;
1632 goto out;
1633 }
1634 memcpy(&iscsilun->bl, inq_bl,
1635 sizeof(struct scsi_inquiry_block_limits));
1636 scsi_free_scsi_task(inq_task);
1637 break;
1638 default:
1639 break;
1640 }
1641 }
1642 scsi_free_scsi_task(task);
1643 task = NULL;
1644
1645 iscsi_attach_aio_context(bs, iscsilun->aio_context);
1646
1647 /* Guess the internal cluster (page) size of the iscsi target by the means
1648 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1649 * reasonable size */
1650 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
1651 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1652 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran *
1653 iscsilun->block_size) >> BDRV_SECTOR_BITS;
1654 if (iscsilun->lbprz) {
1655 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun);
1656 if (iscsilun->allocationmap == NULL) {
1657 ret = -ENOMEM;
1658 }
1659 }
1660 }
1661
1662 out:
1663 qemu_opts_del(opts);
1664 g_free(initiator_name);
1665 if (iscsi_url != NULL) {
1666 iscsi_destroy_url(iscsi_url);
1667 }
1668 if (task != NULL) {
1669 scsi_free_scsi_task(task);
1670 }
1671
1672 if (ret) {
1673 if (iscsi != NULL) {
1674 if (iscsi_is_logged_in(iscsi)) {
1675 iscsi_logout_sync(iscsi);
1676 }
1677 iscsi_destroy_context(iscsi);
1678 }
1679 memset(iscsilun, 0, sizeof(IscsiLun));
1680 }
1681 return ret;
1682 }
1683
1684 static void iscsi_close(BlockDriverState *bs)
1685 {
1686 IscsiLun *iscsilun = bs->opaque;
1687 struct iscsi_context *iscsi = iscsilun->iscsi;
1688
1689 iscsi_detach_aio_context(bs);
1690 if (iscsi_is_logged_in(iscsi)) {
1691 iscsi_logout_sync(iscsi);
1692 }
1693 iscsi_destroy_context(iscsi);
1694 g_free(iscsilun->zeroblock);
1695 g_free(iscsilun->allocationmap);
1696 memset(iscsilun, 0, sizeof(IscsiLun));
1697 }
1698
1699 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
1700 {
1701 /* We don't actually refresh here, but just return data queried in
1702 * iscsi_open(): iscsi targets don't change their limits. */
1703
1704 IscsiLun *iscsilun = bs->opaque;
1705 uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
1706
1707 bs->bl.request_alignment = iscsilun->block_size;
1708
1709 if (iscsilun->bl.max_xfer_len) {
1710 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
1711 }
1712
1713 if (max_xfer_len * iscsilun->block_size < INT_MAX) {
1714 bs->bl.max_transfer = max_xfer_len * iscsilun->block_size;
1715 }
1716
1717 if (iscsilun->lbp.lbpu) {
1718 if (iscsilun->bl.max_unmap < 0xffffffff / iscsilun->block_size) {
1719 bs->bl.max_pdiscard =
1720 iscsilun->bl.max_unmap * iscsilun->block_size;
1721 }
1722 bs->bl.pdiscard_alignment =
1723 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
1724 } else {
1725 bs->bl.pdiscard_alignment = iscsilun->block_size;
1726 }
1727
1728 if (iscsilun->bl.max_ws_len < 0xffffffff / iscsilun->block_size) {
1729 bs->bl.max_pwrite_zeroes =
1730 iscsilun->bl.max_ws_len * iscsilun->block_size;
1731 }
1732 if (iscsilun->lbp.lbpws) {
1733 bs->bl.pwrite_zeroes_alignment =
1734 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
1735 } else {
1736 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size;
1737 }
1738 if (iscsilun->bl.opt_xfer_len &&
1739 iscsilun->bl.opt_xfer_len < INT_MAX / iscsilun->block_size) {
1740 bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len *
1741 iscsilun->block_size);
1742 }
1743 }
1744
1745 /* Note that this will not re-establish a connection with an iSCSI target - it
1746 * is effectively a NOP. */
1747 static int iscsi_reopen_prepare(BDRVReopenState *state,
1748 BlockReopenQueue *queue, Error **errp)
1749 {
1750 IscsiLun *iscsilun = state->bs->opaque;
1751
1752 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
1753 error_setg(errp, "Cannot open a write protected LUN as read-write");
1754 return -EACCES;
1755 }
1756 return 0;
1757 }
1758
1759 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1760 {
1761 IscsiLun *iscsilun = bs->opaque;
1762 Error *local_err = NULL;
1763
1764 if (iscsilun->type != TYPE_DISK) {
1765 return -ENOTSUP;
1766 }
1767
1768 iscsi_readcapacity_sync(iscsilun, &local_err);
1769 if (local_err != NULL) {
1770 error_free(local_err);
1771 return -EIO;
1772 }
1773
1774 if (offset > iscsi_getlength(bs)) {
1775 return -EINVAL;
1776 }
1777
1778 if (iscsilun->allocationmap != NULL) {
1779 g_free(iscsilun->allocationmap);
1780 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun);
1781 }
1782
1783 return 0;
1784 }
1785
1786 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp)
1787 {
1788 int ret = 0;
1789 int64_t total_size = 0;
1790 BlockDriverState *bs;
1791 IscsiLun *iscsilun = NULL;
1792 QDict *bs_options;
1793
1794 bs = bdrv_new();
1795
1796 /* Read out options */
1797 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1798 BDRV_SECTOR_SIZE);
1799 bs->opaque = g_new0(struct IscsiLun, 1);
1800 iscsilun = bs->opaque;
1801
1802 bs_options = qdict_new();
1803 qdict_put(bs_options, "filename", qstring_from_str(filename));
1804 ret = iscsi_open(bs, bs_options, 0, NULL);
1805 QDECREF(bs_options);
1806
1807 if (ret != 0) {
1808 goto out;
1809 }
1810 iscsi_detach_aio_context(bs);
1811 if (iscsilun->type != TYPE_DISK) {
1812 ret = -ENODEV;
1813 goto out;
1814 }
1815 if (bs->total_sectors < total_size) {
1816 ret = -ENOSPC;
1817 goto out;
1818 }
1819
1820 ret = 0;
1821 out:
1822 if (iscsilun->iscsi != NULL) {
1823 iscsi_destroy_context(iscsilun->iscsi);
1824 }
1825 g_free(bs->opaque);
1826 bs->opaque = NULL;
1827 bdrv_unref(bs);
1828 return ret;
1829 }
1830
1831 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1832 {
1833 IscsiLun *iscsilun = bs->opaque;
1834 bdi->unallocated_blocks_are_zero = iscsilun->lbprz;
1835 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
1836 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE;
1837 return 0;
1838 }
1839
1840 static QemuOptsList iscsi_create_opts = {
1841 .name = "iscsi-create-opts",
1842 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
1843 .desc = {
1844 {
1845 .name = BLOCK_OPT_SIZE,
1846 .type = QEMU_OPT_SIZE,
1847 .help = "Virtual disk size"
1848 },
1849 { /* end of list */ }
1850 }
1851 };
1852
1853 static BlockDriver bdrv_iscsi = {
1854 .format_name = "iscsi",
1855 .protocol_name = "iscsi",
1856
1857 .instance_size = sizeof(IscsiLun),
1858 .bdrv_needs_filename = true,
1859 .bdrv_file_open = iscsi_open,
1860 .bdrv_close = iscsi_close,
1861 .bdrv_create = iscsi_create,
1862 .create_opts = &iscsi_create_opts,
1863 .bdrv_reopen_prepare = iscsi_reopen_prepare,
1864
1865 .bdrv_getlength = iscsi_getlength,
1866 .bdrv_get_info = iscsi_get_info,
1867 .bdrv_truncate = iscsi_truncate,
1868 .bdrv_refresh_limits = iscsi_refresh_limits,
1869
1870 .bdrv_co_get_block_status = iscsi_co_get_block_status,
1871 .bdrv_co_discard = iscsi_co_discard,
1872 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
1873 .bdrv_co_readv = iscsi_co_readv,
1874 .bdrv_co_writev_flags = iscsi_co_writev_flags,
1875 .bdrv_co_flush_to_disk = iscsi_co_flush,
1876
1877 #ifdef __linux__
1878 .bdrv_aio_ioctl = iscsi_aio_ioctl,
1879 #endif
1880
1881 .bdrv_detach_aio_context = iscsi_detach_aio_context,
1882 .bdrv_attach_aio_context = iscsi_attach_aio_context,
1883 };
1884
1885 static QemuOptsList qemu_iscsi_opts = {
1886 .name = "iscsi",
1887 .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1888 .desc = {
1889 {
1890 .name = "user",
1891 .type = QEMU_OPT_STRING,
1892 .help = "username for CHAP authentication to target",
1893 },{
1894 .name = "password",
1895 .type = QEMU_OPT_STRING,
1896 .help = "password for CHAP authentication to target",
1897 },{
1898 .name = "password-secret",
1899 .type = QEMU_OPT_STRING,
1900 .help = "ID of the secret providing password for CHAP "
1901 "authentication to target",
1902 },{
1903 .name = "header-digest",
1904 .type = QEMU_OPT_STRING,
1905 .help = "HeaderDigest setting. "
1906 "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1907 },{
1908 .name = "initiator-name",
1909 .type = QEMU_OPT_STRING,
1910 .help = "Initiator iqn name to use when connecting",
1911 },{
1912 .name = "timeout",
1913 .type = QEMU_OPT_NUMBER,
1914 .help = "Request timeout in seconds (default 0 = no timeout)",
1915 },
1916 { /* end of list */ }
1917 },
1918 };
1919
1920 static void iscsi_block_init(void)
1921 {
1922 bdrv_register(&bdrv_iscsi);
1923 qemu_add_opts(&qemu_iscsi_opts);
1924 }
1925
1926 block_init(iscsi_block_init);
QEMU mirror