?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
kvm: Switch to unlocked PIO
[qemu/armbru.git] / block / iscsi.c
blob49cee4dda99b04b66d0348193c5467d6f5676187
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 "config-host.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
43 #include <iscsi/iscsi.h>
44 #include <iscsi/scsi-lowlevel.h>
45
46 #ifdef __linux__
47 #include <scsi/sg.h>
48 #include <block/scsi.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 force_next_flush;
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 bool force_next_flush;
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 #ifdef __linux__
99 sg_io_hdr_t *ioh;
100 #endif
101 } IscsiAIOCB;
102
103 #define EVENT_INTERVAL 250
104 #define NOP_INTERVAL 5000
105 #define MAX_NOP_FAILURES 3
106 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times)
107 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768};
108
109 /* this threshold is a trade-off knob to choose between
110 * the potential additional overhead of an extra GET_LBA_STATUS request
111 * vs. unnecessarily reading a lot of zero sectors over the wire.
112 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
113 * sectors we check the allocation status of the area covered by the
114 * request first if the allocationmap indicates that the area might be
115 * unallocated. */
116 #define ISCSI_CHECKALLOC_THRES 64
117
118 static void
119 iscsi_bh_cb(void *p)
120 {
121 IscsiAIOCB *acb = p;
122
123 qemu_bh_delete(acb->bh);
124
125 g_free(acb->buf);
126 acb->buf = NULL;
127
128 acb->common.cb(acb->common.opaque, acb->status);
129
130 if (acb->task != NULL) {
131 scsi_free_scsi_task(acb->task);
132 acb->task = NULL;
133 }
134
135 qemu_aio_unref(acb);
136 }
137
138 static void
139 iscsi_schedule_bh(IscsiAIOCB *acb)
140 {
141 if (acb->bh) {
142 return;
143 }
144 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb);
145 qemu_bh_schedule(acb->bh);
146 }
147
148 static void iscsi_co_generic_bh_cb(void *opaque)
149 {
150 struct IscsiTask *iTask = opaque;
151 iTask->complete = 1;
152 qemu_bh_delete(iTask->bh);
153 qemu_coroutine_enter(iTask->co, NULL);
154 }
155
156 static void iscsi_retry_timer_expired(void *opaque)
157 {
158 struct IscsiTask *iTask = opaque;
159 iTask->complete = 1;
160 if (iTask->co) {
161 qemu_coroutine_enter(iTask->co, NULL);
162 }
163 }
164
165 static inline unsigned exp_random(double mean)
166 {
167 return -mean * log((double)rand() / RAND_MAX);
168 }
169
170 static void
171 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
172 void *command_data, void *opaque)
173 {
174 struct IscsiTask *iTask = opaque;
175 struct scsi_task *task = command_data;
176
177 iTask->status = status;
178 iTask->do_retry = 0;
179 iTask->task = task;
180
181 if (status != SCSI_STATUS_GOOD) {
182 if (iTask->retries++ < ISCSI_CMD_RETRIES) {
183 if (status == SCSI_STATUS_CHECK_CONDITION
184 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
185 error_report("iSCSI CheckCondition: %s",
186 iscsi_get_error(iscsi));
187 iTask->do_retry = 1;
188 goto out;
189 }
190 /* status 0x28 is SCSI_TASK_SET_FULL. It was first introduced
191 * in libiscsi 1.10.0. Hardcode this value here to avoid
192 * the need to bump the libiscsi requirement to 1.10.0 */
193 if (status == SCSI_STATUS_BUSY || status == 0x28) {
194 unsigned retry_time =
195 exp_random(iscsi_retry_times[iTask->retries - 1]);
196 error_report("iSCSI Busy/TaskSetFull (retry #%u in %u ms): %s",
197 iTask->retries, retry_time,
198 iscsi_get_error(iscsi));
199 aio_timer_init(iTask->iscsilun->aio_context,
200 &iTask->retry_timer, QEMU_CLOCK_REALTIME,
201 SCALE_MS, iscsi_retry_timer_expired, iTask);
202 timer_mod(&iTask->retry_timer,
203 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
204 iTask->do_retry = 1;
205 return;
206 }
207 }
208 error_report("iSCSI Failure: %s", iscsi_get_error(iscsi));
209 } else {
210 iTask->iscsilun->force_next_flush |= iTask->force_next_flush;
211 }
212
213 out:
214 if (iTask->co) {
215 iTask->bh = aio_bh_new(iTask->iscsilun->aio_context,
216 iscsi_co_generic_bh_cb, iTask);
217 qemu_bh_schedule(iTask->bh);
218 } else {
219 iTask->complete = 1;
220 }
221 }
222
223 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
224 {
225 *iTask = (struct IscsiTask) {
226 .co = qemu_coroutine_self(),
227 .iscsilun = iscsilun,
228 };
229 }
230
231 static void
232 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
233 void *private_data)
234 {
235 IscsiAIOCB *acb = private_data;
236
237 acb->status = -ECANCELED;
238 iscsi_schedule_bh(acb);
239 }
240
241 static void
242 iscsi_aio_cancel(BlockAIOCB *blockacb)
243 {
244 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
245 IscsiLun *iscsilun = acb->iscsilun;
246
247 if (acb->status != -EINPROGRESS) {
248 return;
249 }
250
251 /* send a task mgmt call to the target to cancel the task on the target */
252 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
253 iscsi_abort_task_cb, acb);
254
255 }
256
257 static const AIOCBInfo iscsi_aiocb_info = {
258 .aiocb_size = sizeof(IscsiAIOCB),
259 .cancel_async = iscsi_aio_cancel,
260 };
261
262
263 static void iscsi_process_read(void *arg);
264 static void iscsi_process_write(void *arg);
265
266 static void
267 iscsi_set_events(IscsiLun *iscsilun)
268 {
269 struct iscsi_context *iscsi = iscsilun->iscsi;
270 int ev = iscsi_which_events(iscsi);
271
272 if (ev != iscsilun->events) {
273 aio_set_fd_handler(iscsilun->aio_context,
274 iscsi_get_fd(iscsi),
275 (ev & POLLIN) ? iscsi_process_read : NULL,
276 (ev & POLLOUT) ? iscsi_process_write : NULL,
277 iscsilun);
278 iscsilun->events = ev;
279 }
280
281 /* newer versions of libiscsi may return zero events. In this
282 * case start a timer to ensure we are able to return to service
283 * once this situation changes. */
284 if (!ev) {
285 timer_mod(iscsilun->event_timer,
286 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
287 }
288 }
289
290 static void iscsi_timed_set_events(void *opaque)
291 {
292 IscsiLun *iscsilun = opaque;
293 iscsi_set_events(iscsilun);
294 }
295
296 static void
297 iscsi_process_read(void *arg)
298 {
299 IscsiLun *iscsilun = arg;
300 struct iscsi_context *iscsi = iscsilun->iscsi;
301
302 iscsi_service(iscsi, POLLIN);
303 iscsi_set_events(iscsilun);
304 }
305
306 static void
307 iscsi_process_write(void *arg)
308 {
309 IscsiLun *iscsilun = arg;
310 struct iscsi_context *iscsi = iscsilun->iscsi;
311
312 iscsi_service(iscsi, POLLOUT);
313 iscsi_set_events(iscsilun);
314 }
315
316 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
317 {
318 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
319 }
320
321 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
322 {
323 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
324 }
325
326 static bool is_request_lun_aligned(int64_t sector_num, int nb_sectors,
327 IscsiLun *iscsilun)
328 {
329 if ((sector_num * BDRV_SECTOR_SIZE) % iscsilun->block_size ||
330 (nb_sectors * BDRV_SECTOR_SIZE) % iscsilun->block_size) {
331 error_report("iSCSI misaligned request: "
332 "iscsilun->block_size %u, sector_num %" PRIi64
333 ", nb_sectors %d",
334 iscsilun->block_size, sector_num, nb_sectors);
335 return 0;
336 }
337 return 1;
338 }
339
340 static unsigned long *iscsi_allocationmap_init(IscsiLun *iscsilun)
341 {
342 return bitmap_try_new(DIV_ROUND_UP(sector_lun2qemu(iscsilun->num_blocks,
343 iscsilun),
344 iscsilun->cluster_sectors));
345 }
346
347 static void iscsi_allocationmap_set(IscsiLun *iscsilun, int64_t sector_num,
348 int nb_sectors)
349 {
350 if (iscsilun->allocationmap == NULL) {
351 return;
352 }
353 bitmap_set(iscsilun->allocationmap,
354 sector_num / iscsilun->cluster_sectors,
355 DIV_ROUND_UP(nb_sectors, iscsilun->cluster_sectors));
356 }
357
358 static void iscsi_allocationmap_clear(IscsiLun *iscsilun, int64_t sector_num,
359 int nb_sectors)
360 {
361 int64_t cluster_num, nb_clusters;
362 if (iscsilun->allocationmap == NULL) {
363 return;
364 }
365 cluster_num = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors);
366 nb_clusters = (sector_num + nb_sectors) / iscsilun->cluster_sectors
367 - cluster_num;
368 if (nb_clusters > 0) {
369 bitmap_clear(iscsilun->allocationmap, cluster_num, nb_clusters);
370 }
371 }
372
373 static int coroutine_fn iscsi_co_writev(BlockDriverState *bs,
374 int64_t sector_num, int nb_sectors,
375 QEMUIOVector *iov)
376 {
377 IscsiLun *iscsilun = bs->opaque;
378 struct IscsiTask iTask;
379 uint64_t lba;
380 uint32_t num_sectors;
381 int fua;
382
383 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
384 return -EINVAL;
385 }
386
387 if (bs->bl.max_transfer_length && nb_sectors > bs->bl.max_transfer_length) {
388 error_report("iSCSI Error: Write of %d sectors exceeds max_xfer_len "
389 "of %d sectors", nb_sectors, bs->bl.max_transfer_length);
390 return -EINVAL;
391 }
392
393 lba = sector_qemu2lun(sector_num, iscsilun);
394 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
395 iscsi_co_init_iscsitask(iscsilun, &iTask);
396 retry:
397 fua = iscsilun->dpofua && !bs->enable_write_cache;
398 iTask.force_next_flush = !fua;
399 if (iscsilun->use_16_for_rw) {
400 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
401 NULL, num_sectors * iscsilun->block_size,
402 iscsilun->block_size, 0, 0, fua, 0, 0,
403 iscsi_co_generic_cb, &iTask);
404 } else {
405 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
406 NULL, num_sectors * iscsilun->block_size,
407 iscsilun->block_size, 0, 0, fua, 0, 0,
408 iscsi_co_generic_cb, &iTask);
409 }
410 if (iTask.task == NULL) {
411 return -ENOMEM;
412 }
413 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
414 iov->niov);
415 while (!iTask.complete) {
416 iscsi_set_events(iscsilun);
417 qemu_coroutine_yield();
418 }
419
420 if (iTask.task != NULL) {
421 scsi_free_scsi_task(iTask.task);
422 iTask.task = NULL;
423 }
424
425 if (iTask.do_retry) {
426 iTask.complete = 0;
427 goto retry;
428 }
429
430 if (iTask.status != SCSI_STATUS_GOOD) {
431 return -EIO;
432 }
433
434 iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors);
435
436 return 0;
437 }
438
439
440 static bool iscsi_allocationmap_is_allocated(IscsiLun *iscsilun,
441 int64_t sector_num, int nb_sectors)
442 {
443 unsigned long size;
444 if (iscsilun->allocationmap == NULL) {
445 return true;
446 }
447 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
448 return !(find_next_bit(iscsilun->allocationmap, size,
449 sector_num / iscsilun->cluster_sectors) == size);
450 }
451
452 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
453 int64_t sector_num,
454 int nb_sectors, int *pnum)
455 {
456 IscsiLun *iscsilun = bs->opaque;
457 struct scsi_get_lba_status *lbas = NULL;
458 struct scsi_lba_status_descriptor *lbasd = NULL;
459 struct IscsiTask iTask;
460 int64_t ret;
461
462 iscsi_co_init_iscsitask(iscsilun, &iTask);
463
464 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
465 ret = -EINVAL;
466 goto out;
467 }
468
469 /* default to all sectors allocated */
470 ret = BDRV_BLOCK_DATA;
471 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
472 *pnum = nb_sectors;
473
474 /* LUN does not support logical block provisioning */
475 if (!iscsilun->lbpme) {
476 goto out;
477 }
478
479 retry:
480 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
481 sector_qemu2lun(sector_num, iscsilun),
482 8 + 16, iscsi_co_generic_cb,
483 &iTask) == NULL) {
484 ret = -ENOMEM;
485 goto out;
486 }
487
488 while (!iTask.complete) {
489 iscsi_set_events(iscsilun);
490 qemu_coroutine_yield();
491 }
492
493 if (iTask.do_retry) {
494 if (iTask.task != NULL) {
495 scsi_free_scsi_task(iTask.task);
496 iTask.task = NULL;
497 }
498 iTask.complete = 0;
499 goto retry;
500 }
501
502 if (iTask.status != SCSI_STATUS_GOOD) {
503 /* in case the get_lba_status_callout fails (i.e.
504 * because the device is busy or the cmd is not
505 * supported) we pretend all blocks are allocated
506 * for backwards compatibility */
507 goto out;
508 }
509
510 lbas = scsi_datain_unmarshall(iTask.task);
511 if (lbas == NULL) {
512 ret = -EIO;
513 goto out;
514 }
515
516 lbasd = &lbas->descriptors[0];
517
518 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
519 ret = -EIO;
520 goto out;
521 }
522
523 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
524
525 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
526 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
527 ret &= ~BDRV_BLOCK_DATA;
528 if (iscsilun->lbprz) {
529 ret |= BDRV_BLOCK_ZERO;
530 }
531 }
532
533 if (ret & BDRV_BLOCK_ZERO) {
534 iscsi_allocationmap_clear(iscsilun, sector_num, *pnum);
535 } else {
536 iscsi_allocationmap_set(iscsilun, sector_num, *pnum);
537 }
538
539 if (*pnum > nb_sectors) {
540 *pnum = nb_sectors;
541 }
542 out:
543 if (iTask.task != NULL) {
544 scsi_free_scsi_task(iTask.task);
545 }
546 return ret;
547 }
548
549 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
550 int64_t sector_num, int nb_sectors,
551 QEMUIOVector *iov)
552 {
553 IscsiLun *iscsilun = bs->opaque;
554 struct IscsiTask iTask;
555 uint64_t lba;
556 uint32_t num_sectors;
557
558 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
559 return -EINVAL;
560 }
561
562 if (bs->bl.max_transfer_length && nb_sectors > bs->bl.max_transfer_length) {
563 error_report("iSCSI Error: Read of %d sectors exceeds max_xfer_len "
564 "of %d sectors", nb_sectors, bs->bl.max_transfer_length);
565 return -EINVAL;
566 }
567
568 if (iscsilun->lbprz && nb_sectors >= ISCSI_CHECKALLOC_THRES &&
569 !iscsi_allocationmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
570 int64_t ret;
571 int pnum;
572 ret = iscsi_co_get_block_status(bs, sector_num, INT_MAX, &pnum);
573 if (ret < 0) {
574 return ret;
575 }
576 if (ret & BDRV_BLOCK_ZERO && pnum >= nb_sectors) {
577 qemu_iovec_memset(iov, 0, 0x00, iov->size);
578 return 0;
579 }
580 }
581
582 lba = sector_qemu2lun(sector_num, iscsilun);
583 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
584
585 iscsi_co_init_iscsitask(iscsilun, &iTask);
586 retry:
587 if (iscsilun->use_16_for_rw) {
588 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
589 num_sectors * iscsilun->block_size,
590 iscsilun->block_size, 0, 0, 0, 0, 0,
591 iscsi_co_generic_cb, &iTask);
592 } else {
593 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
594 num_sectors * iscsilun->block_size,
595 iscsilun->block_size,
596 0, 0, 0, 0, 0,
597 iscsi_co_generic_cb, &iTask);
598 }
599 if (iTask.task == NULL) {
600 return -ENOMEM;
601 }
602 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
603
604 while (!iTask.complete) {
605 iscsi_set_events(iscsilun);
606 qemu_coroutine_yield();
607 }
608
609 if (iTask.task != NULL) {
610 scsi_free_scsi_task(iTask.task);
611 iTask.task = NULL;
612 }
613
614 if (iTask.do_retry) {
615 iTask.complete = 0;
616 goto retry;
617 }
618
619 if (iTask.status != SCSI_STATUS_GOOD) {
620 return -EIO;
621 }
622
623 return 0;
624 }
625
626 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
627 {
628 IscsiLun *iscsilun = bs->opaque;
629 struct IscsiTask iTask;
630
631 if (!iscsilun->force_next_flush) {
632 return 0;
633 }
634 iscsilun->force_next_flush = false;
635
636 iscsi_co_init_iscsitask(iscsilun, &iTask);
637 retry:
638 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
639 0, iscsi_co_generic_cb, &iTask) == NULL) {
640 return -ENOMEM;
641 }
642
643 while (!iTask.complete) {
644 iscsi_set_events(iscsilun);
645 qemu_coroutine_yield();
646 }
647
648 if (iTask.task != NULL) {
649 scsi_free_scsi_task(iTask.task);
650 iTask.task = NULL;
651 }
652
653 if (iTask.do_retry) {
654 iTask.complete = 0;
655 goto retry;
656 }
657
658 if (iTask.status != SCSI_STATUS_GOOD) {
659 return -EIO;
660 }
661
662 return 0;
663 }
664
665 #ifdef __linux__
666 static void
667 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
668 void *command_data, void *opaque)
669 {
670 IscsiAIOCB *acb = opaque;
671
672 g_free(acb->buf);
673 acb->buf = NULL;
674
675 acb->status = 0;
676 if (status < 0) {
677 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
678 iscsi_get_error(iscsi));
679 acb->status = -EIO;
680 }
681
682 acb->ioh->driver_status = 0;
683 acb->ioh->host_status = 0;
684 acb->ioh->resid = 0;
685
686 #define SG_ERR_DRIVER_SENSE 0x08
687
688 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
689 int ss;
690
691 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
692
693 acb->ioh->sb_len_wr = acb->task->datain.size - 2;
694 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
695 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
696 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
697 }
698
699 iscsi_schedule_bh(acb);
700 }
701
702 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
703 unsigned long int req, void *buf,
704 BlockCompletionFunc *cb, void *opaque)
705 {
706 IscsiLun *iscsilun = bs->opaque;
707 struct iscsi_context *iscsi = iscsilun->iscsi;
708 struct iscsi_data data;
709 IscsiAIOCB *acb;
710
711 assert(req == SG_IO);
712
713 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
714
715 acb->iscsilun = iscsilun;
716 acb->bh = NULL;
717 acb->status = -EINPROGRESS;
718 acb->buf = NULL;
719 acb->ioh = buf;
720
721 acb->task = malloc(sizeof(struct scsi_task));
722 if (acb->task == NULL) {
723 error_report("iSCSI: Failed to allocate task for scsi command. %s",
724 iscsi_get_error(iscsi));
725 qemu_aio_unref(acb);
726 return NULL;
727 }
728 memset(acb->task, 0, sizeof(struct scsi_task));
729
730 switch (acb->ioh->dxfer_direction) {
731 case SG_DXFER_TO_DEV:
732 acb->task->xfer_dir = SCSI_XFER_WRITE;
733 break;
734 case SG_DXFER_FROM_DEV:
735 acb->task->xfer_dir = SCSI_XFER_READ;
736 break;
737 default:
738 acb->task->xfer_dir = SCSI_XFER_NONE;
739 break;
740 }
741
742 acb->task->cdb_size = acb->ioh->cmd_len;
743 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
744 acb->task->expxferlen = acb->ioh->dxfer_len;
745
746 data.size = 0;
747 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
748 if (acb->ioh->iovec_count == 0) {
749 data.data = acb->ioh->dxferp;
750 data.size = acb->ioh->dxfer_len;
751 } else {
752 scsi_task_set_iov_out(acb->task,
753 (struct scsi_iovec *) acb->ioh->dxferp,
754 acb->ioh->iovec_count);
755 }
756 }
757
758 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
759 iscsi_aio_ioctl_cb,
760 (data.size > 0) ? &data : NULL,
761 acb) != 0) {
762 scsi_free_scsi_task(acb->task);
763 qemu_aio_unref(acb);
764 return NULL;
765 }
766
767 /* tell libiscsi to read straight into the buffer we got from ioctl */
768 if (acb->task->xfer_dir == SCSI_XFER_READ) {
769 if (acb->ioh->iovec_count == 0) {
770 scsi_task_add_data_in_buffer(acb->task,
771 acb->ioh->dxfer_len,
772 acb->ioh->dxferp);
773 } else {
774 scsi_task_set_iov_in(acb->task,
775 (struct scsi_iovec *) acb->ioh->dxferp,
776 acb->ioh->iovec_count);
777 }
778 }
779
780 iscsi_set_events(iscsilun);
781
782 return &acb->common;
783 }
784
785 static void ioctl_cb(void *opaque, int status)
786 {
787 int *p_status = opaque;
788 *p_status = status;
789 }
790
791 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
792 {
793 IscsiLun *iscsilun = bs->opaque;
794 int status;
795
796 switch (req) {
797 case SG_GET_VERSION_NUM:
798 *(int *)buf = 30000;
799 break;
800 case SG_GET_SCSI_ID:
801 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
802 break;
803 case SG_IO:
804 status = -EINPROGRESS;
805 iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status);
806
807 while (status == -EINPROGRESS) {
808 aio_poll(iscsilun->aio_context, true);
809 }
810
811 return 0;
812 default:
813 return -1;
814 }
815 return 0;
816 }
817 #endif
818
819 static int64_t
820 iscsi_getlength(BlockDriverState *bs)
821 {
822 IscsiLun *iscsilun = bs->opaque;
823 int64_t len;
824
825 len = iscsilun->num_blocks;
826 len *= iscsilun->block_size;
827
828 return len;
829 }
830
831 static int
832 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num,
833 int nb_sectors)
834 {
835 IscsiLun *iscsilun = bs->opaque;
836 struct IscsiTask iTask;
837 struct unmap_list list;
838
839 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
840 return -EINVAL;
841 }
842
843 if (!iscsilun->lbp.lbpu) {
844 /* UNMAP is not supported by the target */
845 return 0;
846 }
847
848 list.lba = sector_qemu2lun(sector_num, iscsilun);
849 list.num = sector_qemu2lun(nb_sectors, iscsilun);
850
851 iscsi_co_init_iscsitask(iscsilun, &iTask);
852 retry:
853 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
854 iscsi_co_generic_cb, &iTask) == NULL) {
855 return -ENOMEM;
856 }
857
858 while (!iTask.complete) {
859 iscsi_set_events(iscsilun);
860 qemu_coroutine_yield();
861 }
862
863 if (iTask.task != NULL) {
864 scsi_free_scsi_task(iTask.task);
865 iTask.task = NULL;
866 }
867
868 if (iTask.do_retry) {
869 iTask.complete = 0;
870 goto retry;
871 }
872
873 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
874 /* the target might fail with a check condition if it
875 is not happy with the alignment of the UNMAP request
876 we silently fail in this case */
877 return 0;
878 }
879
880 if (iTask.status != SCSI_STATUS_GOOD) {
881 return -EIO;
882 }
883
884 iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors);
885
886 return 0;
887 }
888
889 static int
890 coroutine_fn iscsi_co_write_zeroes(BlockDriverState *bs, int64_t sector_num,
891 int nb_sectors, BdrvRequestFlags flags)
892 {
893 IscsiLun *iscsilun = bs->opaque;
894 struct IscsiTask iTask;
895 uint64_t lba;
896 uint32_t nb_blocks;
897 bool use_16_for_ws = iscsilun->use_16_for_rw;
898
899 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
900 return -EINVAL;
901 }
902
903 if (flags & BDRV_REQ_MAY_UNMAP) {
904 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
905 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
906 use_16_for_ws = true;
907 }
908 if (use_16_for_ws && !iscsilun->lbp.lbpws) {
909 /* WRITESAME16 with UNMAP is not supported by the target,
910 * fall back and try WRITESAME10/16 without UNMAP */
911 flags &= ~BDRV_REQ_MAY_UNMAP;
912 use_16_for_ws = iscsilun->use_16_for_rw;
913 }
914 }
915
916 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
917 /* WRITESAME without UNMAP is not supported by the target */
918 return -ENOTSUP;
919 }
920
921 lba = sector_qemu2lun(sector_num, iscsilun);
922 nb_blocks = sector_qemu2lun(nb_sectors, iscsilun);
923
924 if (iscsilun->zeroblock == NULL) {
925 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
926 if (iscsilun->zeroblock == NULL) {
927 return -ENOMEM;
928 }
929 }
930
931 iscsi_co_init_iscsitask(iscsilun, &iTask);
932 iTask.force_next_flush = true;
933 retry:
934 if (use_16_for_ws) {
935 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
936 iscsilun->zeroblock, iscsilun->block_size,
937 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
938 0, 0, iscsi_co_generic_cb, &iTask);
939 } else {
940 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
941 iscsilun->zeroblock, iscsilun->block_size,
942 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
943 0, 0, iscsi_co_generic_cb, &iTask);
944 }
945 if (iTask.task == NULL) {
946 return -ENOMEM;
947 }
948
949 while (!iTask.complete) {
950 iscsi_set_events(iscsilun);
951 qemu_coroutine_yield();
952 }
953
954 if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
955 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
956 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
957 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
958 /* WRITE SAME is not supported by the target */
959 iscsilun->has_write_same = false;
960 scsi_free_scsi_task(iTask.task);
961 return -ENOTSUP;
962 }
963
964 if (iTask.task != NULL) {
965 scsi_free_scsi_task(iTask.task);
966 iTask.task = NULL;
967 }
968
969 if (iTask.do_retry) {
970 iTask.complete = 0;
971 goto retry;
972 }
973
974 if (iTask.status != SCSI_STATUS_GOOD) {
975 return -EIO;
976 }
977
978 if (flags & BDRV_REQ_MAY_UNMAP) {
979 iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors);
980 } else {
981 iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors);
982 }
983
984 return 0;
985 }
986
987 static void parse_chap(struct iscsi_context *iscsi, const char *target,
988 Error **errp)
989 {
990 QemuOptsList *list;
991 QemuOpts *opts;
992 const char *user = NULL;
993 const char *password = NULL;
994
995 list = qemu_find_opts("iscsi");
996 if (!list) {
997 return;
998 }
999
1000 opts = qemu_opts_find(list, target);
1001 if (opts == NULL) {
1002 opts = QTAILQ_FIRST(&list->head);
1003 if (!opts) {
1004 return;
1005 }
1006 }
1007
1008 user = qemu_opt_get(opts, "user");
1009 if (!user) {
1010 return;
1011 }
1012
1013 password = qemu_opt_get(opts, "password");
1014 if (!password) {
1015 error_setg(errp, "CHAP username specified but no password was given");
1016 return;
1017 }
1018
1019 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1020 error_setg(errp, "Failed to set initiator username and password");
1021 }
1022 }
1023
1024 static void parse_header_digest(struct iscsi_context *iscsi, const char *target,
1025 Error **errp)
1026 {
1027 QemuOptsList *list;
1028 QemuOpts *opts;
1029 const char *digest = NULL;
1030
1031 list = qemu_find_opts("iscsi");
1032 if (!list) {
1033 return;
1034 }
1035
1036 opts = qemu_opts_find(list, target);
1037 if (opts == NULL) {
1038 opts = QTAILQ_FIRST(&list->head);
1039 if (!opts) {
1040 return;
1041 }
1042 }
1043
1044 digest = qemu_opt_get(opts, "header-digest");
1045 if (!digest) {
1046 return;
1047 }
1048
1049 if (!strcmp(digest, "CRC32C")) {
1050 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1051 } else if (!strcmp(digest, "NONE")) {
1052 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1053 } else if (!strcmp(digest, "CRC32C-NONE")) {
1054 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1055 } else if (!strcmp(digest, "NONE-CRC32C")) {
1056 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1057 } else {
1058 error_setg(errp, "Invalid header-digest setting : %s", digest);
1059 }
1060 }
1061
1062 static char *parse_initiator_name(const char *target)
1063 {
1064 QemuOptsList *list;
1065 QemuOpts *opts;
1066 const char *name;
1067 char *iscsi_name;
1068 UuidInfo *uuid_info;
1069
1070 list = qemu_find_opts("iscsi");
1071 if (list) {
1072 opts = qemu_opts_find(list, target);
1073 if (!opts) {
1074 opts = QTAILQ_FIRST(&list->head);
1075 }
1076 if (opts) {
1077 name = qemu_opt_get(opts, "initiator-name");
1078 if (name) {
1079 return g_strdup(name);
1080 }
1081 }
1082 }
1083
1084 uuid_info = qmp_query_uuid(NULL);
1085 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1086 name = qemu_get_vm_name();
1087 } else {
1088 name = uuid_info->UUID;
1089 }
1090 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1091 name ? ":" : "", name ? name : "");
1092 qapi_free_UuidInfo(uuid_info);
1093 return iscsi_name;
1094 }
1095
1096 static void iscsi_nop_timed_event(void *opaque)
1097 {
1098 IscsiLun *iscsilun = opaque;
1099
1100 if (iscsi_get_nops_in_flight(iscsilun->iscsi) > MAX_NOP_FAILURES) {
1101 error_report("iSCSI: NOP timeout. Reconnecting...");
1102 iscsi_reconnect(iscsilun->iscsi);
1103 }
1104
1105 if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1106 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1107 return;
1108 }
1109
1110 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1111 iscsi_set_events(iscsilun);
1112 }
1113
1114 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1115 {
1116 struct scsi_task *task = NULL;
1117 struct scsi_readcapacity10 *rc10 = NULL;
1118 struct scsi_readcapacity16 *rc16 = NULL;
1119 int retries = ISCSI_CMD_RETRIES;
1120
1121 do {
1122 if (task != NULL) {
1123 scsi_free_scsi_task(task);
1124 task = NULL;
1125 }
1126
1127 switch (iscsilun->type) {
1128 case TYPE_DISK:
1129 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1130 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1131 rc16 = scsi_datain_unmarshall(task);
1132 if (rc16 == NULL) {
1133 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1134 } else {
1135 iscsilun->block_size = rc16->block_length;
1136 iscsilun->num_blocks = rc16->returned_lba + 1;
1137 iscsilun->lbpme = !!rc16->lbpme;
1138 iscsilun->lbprz = !!rc16->lbprz;
1139 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1140 }
1141 }
1142 break;
1143 case TYPE_ROM:
1144 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1145 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1146 rc10 = scsi_datain_unmarshall(task);
1147 if (rc10 == NULL) {
1148 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1149 } else {
1150 iscsilun->block_size = rc10->block_size;
1151 if (rc10->lba == 0) {
1152 /* blank disk loaded */
1153 iscsilun->num_blocks = 0;
1154 } else {
1155 iscsilun->num_blocks = rc10->lba + 1;
1156 }
1157 }
1158 }
1159 break;
1160 default:
1161 return;
1162 }
1163 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1164 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1165 && retries-- > 0);
1166
1167 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1168 error_setg(errp, "iSCSI: failed to send readcapacity10 command.");
1169 }
1170 if (task) {
1171 scsi_free_scsi_task(task);
1172 }
1173 }
1174
1175 /* TODO Convert to fine grained options */
1176 static QemuOptsList runtime_opts = {
1177 .name = "iscsi",
1178 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1179 .desc = {
1180 {
1181 .name = "filename",
1182 .type = QEMU_OPT_STRING,
1183 .help = "URL to the iscsi image",
1184 },
1185 { /* end of list */ }
1186 },
1187 };
1188
1189 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1190 int evpd, int pc, void **inq, Error **errp)
1191 {
1192 int full_size;
1193 struct scsi_task *task = NULL;
1194 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1195 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1196 goto fail;
1197 }
1198 full_size = scsi_datain_getfullsize(task);
1199 if (full_size > task->datain.size) {
1200 scsi_free_scsi_task(task);
1201
1202 /* we need more data for the full list */
1203 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1204 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1205 goto fail;
1206 }
1207 }
1208
1209 *inq = scsi_datain_unmarshall(task);
1210 if (*inq == NULL) {
1211 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1212 goto fail_with_err;
1213 }
1214
1215 return task;
1216
1217 fail:
1218 error_setg(errp, "iSCSI: Inquiry command failed : %s",
1219 iscsi_get_error(iscsi));
1220 fail_with_err:
1221 if (task != NULL) {
1222 scsi_free_scsi_task(task);
1223 }
1224 return NULL;
1225 }
1226
1227 static void iscsi_detach_aio_context(BlockDriverState *bs)
1228 {
1229 IscsiLun *iscsilun = bs->opaque;
1230
1231 aio_set_fd_handler(iscsilun->aio_context,
1232 iscsi_get_fd(iscsilun->iscsi),
1233 NULL, NULL, NULL);
1234 iscsilun->events = 0;
1235
1236 if (iscsilun->nop_timer) {
1237 timer_del(iscsilun->nop_timer);
1238 timer_free(iscsilun->nop_timer);
1239 iscsilun->nop_timer = NULL;
1240 }
1241 if (iscsilun->event_timer) {
1242 timer_del(iscsilun->event_timer);
1243 timer_free(iscsilun->event_timer);
1244 iscsilun->event_timer = NULL;
1245 }
1246 }
1247
1248 static void iscsi_attach_aio_context(BlockDriverState *bs,
1249 AioContext *new_context)
1250 {
1251 IscsiLun *iscsilun = bs->opaque;
1252
1253 iscsilun->aio_context = new_context;
1254 iscsi_set_events(iscsilun);
1255
1256 /* Set up a timer for sending out iSCSI NOPs */
1257 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1258 QEMU_CLOCK_REALTIME, SCALE_MS,
1259 iscsi_nop_timed_event, iscsilun);
1260 timer_mod(iscsilun->nop_timer,
1261 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1262
1263 /* Prepare a timer for a delayed call to iscsi_set_events */
1264 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1265 QEMU_CLOCK_REALTIME, SCALE_MS,
1266 iscsi_timed_set_events, iscsilun);
1267 }
1268
1269 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1270 {
1271 struct scsi_task *task;
1272 struct scsi_mode_sense *ms = NULL;
1273 iscsilun->write_protected = false;
1274 iscsilun->dpofua = false;
1275
1276 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1277 1, SCSI_MODESENSE_PC_CURRENT,
1278 0x3F, 0, 255);
1279 if (task == NULL) {
1280 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1281 iscsi_get_error(iscsilun->iscsi));
1282 goto out;
1283 }
1284
1285 if (task->status != SCSI_STATUS_GOOD) {
1286 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1287 goto out;
1288 }
1289 ms = scsi_datain_unmarshall(task);
1290 if (!ms) {
1291 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1292 iscsi_get_error(iscsilun->iscsi));
1293 goto out;
1294 }
1295 iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1296 iscsilun->dpofua = ms->device_specific_parameter & 0x10;
1297
1298 out:
1299 if (task) {
1300 scsi_free_scsi_task(task);
1301 }
1302 }
1303
1304 /*
1305 * We support iscsi url's on the form
1306 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1307 */
1308 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1309 Error **errp)
1310 {
1311 IscsiLun *iscsilun = bs->opaque;
1312 struct iscsi_context *iscsi = NULL;
1313 struct iscsi_url *iscsi_url = NULL;
1314 struct scsi_task *task = NULL;
1315 struct scsi_inquiry_standard *inq = NULL;
1316 struct scsi_inquiry_supported_pages *inq_vpd;
1317 char *initiator_name = NULL;
1318 QemuOpts *opts;
1319 Error *local_err = NULL;
1320 const char *filename;
1321 int i, ret = 0;
1322
1323 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1324 qemu_opts_absorb_qdict(opts, options, &local_err);
1325 if (local_err) {
1326 error_propagate(errp, local_err);
1327 ret = -EINVAL;
1328 goto out;
1329 }
1330
1331 filename = qemu_opt_get(opts, "filename");
1332
1333 iscsi_url = iscsi_parse_full_url(iscsi, filename);
1334 if (iscsi_url == NULL) {
1335 error_setg(errp, "Failed to parse URL : %s", filename);
1336 ret = -EINVAL;
1337 goto out;
1338 }
1339
1340 memset(iscsilun, 0, sizeof(IscsiLun));
1341
1342 initiator_name = parse_initiator_name(iscsi_url->target);
1343
1344 iscsi = iscsi_create_context(initiator_name);
1345 if (iscsi == NULL) {
1346 error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1347 ret = -ENOMEM;
1348 goto out;
1349 }
1350
1351 if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1352 error_setg(errp, "iSCSI: Failed to set target name.");
1353 ret = -EINVAL;
1354 goto out;
1355 }
1356
1357 if (iscsi_url->user[0] != '\0') {
1358 ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1359 iscsi_url->passwd);
1360 if (ret != 0) {
1361 error_setg(errp, "Failed to set initiator username and password");
1362 ret = -EINVAL;
1363 goto out;
1364 }
1365 }
1366
1367 /* check if we got CHAP username/password via the options */
1368 parse_chap(iscsi, iscsi_url->target, &local_err);
1369 if (local_err != NULL) {
1370 error_propagate(errp, local_err);
1371 ret = -EINVAL;
1372 goto out;
1373 }
1374
1375 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1376 error_setg(errp, "iSCSI: Failed to set session type to normal.");
1377 ret = -EINVAL;
1378 goto out;
1379 }
1380
1381 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1382
1383 /* check if we got HEADER_DIGEST via the options */
1384 parse_header_digest(iscsi, iscsi_url->target, &local_err);
1385 if (local_err != NULL) {
1386 error_propagate(errp, local_err);
1387 ret = -EINVAL;
1388 goto out;
1389 }
1390
1391 if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1392 error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1393 iscsi_get_error(iscsi));
1394 ret = -EINVAL;
1395 goto out;
1396 }
1397
1398 iscsilun->iscsi = iscsi;
1399 iscsilun->aio_context = bdrv_get_aio_context(bs);
1400 iscsilun->lun = iscsi_url->lun;
1401 iscsilun->has_write_same = true;
1402
1403 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1404 (void **) &inq, errp);
1405 if (task == NULL) {
1406 ret = -EINVAL;
1407 goto out;
1408 }
1409 iscsilun->type = inq->periperal_device_type;
1410 scsi_free_scsi_task(task);
1411 task = NULL;
1412
1413 iscsi_modesense_sync(iscsilun);
1414
1415 /* Check the write protect flag of the LUN if we want to write */
1416 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1417 iscsilun->write_protected) {
1418 error_setg(errp, "Cannot open a write protected LUN as read-write");
1419 ret = -EACCES;
1420 goto out;
1421 }
1422
1423 iscsi_readcapacity_sync(iscsilun, &local_err);
1424 if (local_err != NULL) {
1425 error_propagate(errp, local_err);
1426 ret = -EINVAL;
1427 goto out;
1428 }
1429 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1430 bs->request_alignment = iscsilun->block_size;
1431
1432 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1433 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1434 * will try to read from the device to guess the image format.
1435 */
1436 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1437 bs->sg = 1;
1438 }
1439
1440 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1441 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1442 (void **) &inq_vpd, errp);
1443 if (task == NULL) {
1444 ret = -EINVAL;
1445 goto out;
1446 }
1447 for (i = 0; i < inq_vpd->num_pages; i++) {
1448 struct scsi_task *inq_task;
1449 struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1450 struct scsi_inquiry_block_limits *inq_bl;
1451 switch (inq_vpd->pages[i]) {
1452 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1453 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1454 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1455 (void **) &inq_lbp, errp);
1456 if (inq_task == NULL) {
1457 ret = -EINVAL;
1458 goto out;
1459 }
1460 memcpy(&iscsilun->lbp, inq_lbp,
1461 sizeof(struct scsi_inquiry_logical_block_provisioning));
1462 scsi_free_scsi_task(inq_task);
1463 break;
1464 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1465 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1466 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1467 (void **) &inq_bl, errp);
1468 if (inq_task == NULL) {
1469 ret = -EINVAL;
1470 goto out;
1471 }
1472 memcpy(&iscsilun->bl, inq_bl,
1473 sizeof(struct scsi_inquiry_block_limits));
1474 scsi_free_scsi_task(inq_task);
1475 break;
1476 default:
1477 break;
1478 }
1479 }
1480 scsi_free_scsi_task(task);
1481 task = NULL;
1482
1483 iscsi_attach_aio_context(bs, iscsilun->aio_context);
1484
1485 /* Guess the internal cluster (page) size of the iscsi target by the means
1486 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1487 * reasonable size */
1488 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
1489 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1490 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran *
1491 iscsilun->block_size) >> BDRV_SECTOR_BITS;
1492 if (iscsilun->lbprz) {
1493 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun);
1494 if (iscsilun->allocationmap == NULL) {
1495 ret = -ENOMEM;
1496 }
1497 }
1498 }
1499
1500 out:
1501 qemu_opts_del(opts);
1502 g_free(initiator_name);
1503 if (iscsi_url != NULL) {
1504 iscsi_destroy_url(iscsi_url);
1505 }
1506 if (task != NULL) {
1507 scsi_free_scsi_task(task);
1508 }
1509
1510 if (ret) {
1511 if (iscsi != NULL) {
1512 if (iscsi_is_logged_in(iscsi)) {
1513 iscsi_logout_sync(iscsi);
1514 }
1515 iscsi_destroy_context(iscsi);
1516 }
1517 memset(iscsilun, 0, sizeof(IscsiLun));
1518 }
1519 return ret;
1520 }
1521
1522 static void iscsi_close(BlockDriverState *bs)
1523 {
1524 IscsiLun *iscsilun = bs->opaque;
1525 struct iscsi_context *iscsi = iscsilun->iscsi;
1526
1527 iscsi_detach_aio_context(bs);
1528 if (iscsi_is_logged_in(iscsi)) {
1529 iscsi_logout_sync(iscsi);
1530 }
1531 iscsi_destroy_context(iscsi);
1532 g_free(iscsilun->zeroblock);
1533 g_free(iscsilun->allocationmap);
1534 memset(iscsilun, 0, sizeof(IscsiLun));
1535 }
1536
1537 static int sector_limits_lun2qemu(int64_t sector, IscsiLun *iscsilun)
1538 {
1539 return MIN(sector_lun2qemu(sector, iscsilun), INT_MAX / 2 + 1);
1540 }
1541
1542 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
1543 {
1544 /* We don't actually refresh here, but just return data queried in
1545 * iscsi_open(): iscsi targets don't change their limits. */
1546
1547 IscsiLun *iscsilun = bs->opaque;
1548 uint32_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
1549
1550 if (iscsilun->bl.max_xfer_len) {
1551 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
1552 }
1553
1554 bs->bl.max_transfer_length = sector_limits_lun2qemu(max_xfer_len, iscsilun);
1555
1556 if (iscsilun->lbp.lbpu) {
1557 if (iscsilun->bl.max_unmap < 0xffffffff) {
1558 bs->bl.max_discard =
1559 sector_limits_lun2qemu(iscsilun->bl.max_unmap, iscsilun);
1560 }
1561 bs->bl.discard_alignment =
1562 sector_limits_lun2qemu(iscsilun->bl.opt_unmap_gran, iscsilun);
1563 }
1564
1565 if (iscsilun->bl.max_ws_len < 0xffffffff) {
1566 bs->bl.max_write_zeroes =
1567 sector_limits_lun2qemu(iscsilun->bl.max_ws_len, iscsilun);
1568 }
1569 if (iscsilun->lbp.lbpws) {
1570 bs->bl.write_zeroes_alignment =
1571 sector_limits_lun2qemu(iscsilun->bl.opt_unmap_gran, iscsilun);
1572 }
1573 bs->bl.opt_transfer_length =
1574 sector_limits_lun2qemu(iscsilun->bl.opt_xfer_len, iscsilun);
1575 }
1576
1577 /* Note that this will not re-establish a connection with an iSCSI target - it
1578 * is effectively a NOP. */
1579 static int iscsi_reopen_prepare(BDRVReopenState *state,
1580 BlockReopenQueue *queue, Error **errp)
1581 {
1582 IscsiLun *iscsilun = state->bs->opaque;
1583
1584 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
1585 error_setg(errp, "Cannot open a write protected LUN as read-write");
1586 return -EACCES;
1587 }
1588 return 0;
1589 }
1590
1591 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1592 {
1593 IscsiLun *iscsilun = bs->opaque;
1594 Error *local_err = NULL;
1595
1596 if (iscsilun->type != TYPE_DISK) {
1597 return -ENOTSUP;
1598 }
1599
1600 iscsi_readcapacity_sync(iscsilun, &local_err);
1601 if (local_err != NULL) {
1602 error_free(local_err);
1603 return -EIO;
1604 }
1605
1606 if (offset > iscsi_getlength(bs)) {
1607 return -EINVAL;
1608 }
1609
1610 if (iscsilun->allocationmap != NULL) {
1611 g_free(iscsilun->allocationmap);
1612 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun);
1613 }
1614
1615 return 0;
1616 }
1617
1618 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp)
1619 {
1620 int ret = 0;
1621 int64_t total_size = 0;
1622 BlockDriverState *bs;
1623 IscsiLun *iscsilun = NULL;
1624 QDict *bs_options;
1625
1626 bs = bdrv_new();
1627
1628 /* Read out options */
1629 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1630 BDRV_SECTOR_SIZE);
1631 bs->opaque = g_new0(struct IscsiLun, 1);
1632 iscsilun = bs->opaque;
1633
1634 bs_options = qdict_new();
1635 qdict_put(bs_options, "filename", qstring_from_str(filename));
1636 ret = iscsi_open(bs, bs_options, 0, NULL);
1637 QDECREF(bs_options);
1638
1639 if (ret != 0) {
1640 goto out;
1641 }
1642 iscsi_detach_aio_context(bs);
1643 if (iscsilun->type != TYPE_DISK) {
1644 ret = -ENODEV;
1645 goto out;
1646 }
1647 if (bs->total_sectors < total_size) {
1648 ret = -ENOSPC;
1649 goto out;
1650 }
1651
1652 ret = 0;
1653 out:
1654 if (iscsilun->iscsi != NULL) {
1655 iscsi_destroy_context(iscsilun->iscsi);
1656 }
1657 g_free(bs->opaque);
1658 bs->opaque = NULL;
1659 bdrv_unref(bs);
1660 return ret;
1661 }
1662
1663 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1664 {
1665 IscsiLun *iscsilun = bs->opaque;
1666 bdi->unallocated_blocks_are_zero = iscsilun->lbprz;
1667 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
1668 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE;
1669 return 0;
1670 }
1671
1672 static QemuOptsList iscsi_create_opts = {
1673 .name = "iscsi-create-opts",
1674 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
1675 .desc = {
1676 {
1677 .name = BLOCK_OPT_SIZE,
1678 .type = QEMU_OPT_SIZE,
1679 .help = "Virtual disk size"
1680 },
1681 { /* end of list */ }
1682 }
1683 };
1684
1685 static BlockDriver bdrv_iscsi = {
1686 .format_name = "iscsi",
1687 .protocol_name = "iscsi",
1688
1689 .instance_size = sizeof(IscsiLun),
1690 .bdrv_needs_filename = true,
1691 .bdrv_file_open = iscsi_open,
1692 .bdrv_close = iscsi_close,
1693 .bdrv_create = iscsi_create,
1694 .create_opts = &iscsi_create_opts,
1695 .bdrv_reopen_prepare = iscsi_reopen_prepare,
1696
1697 .bdrv_getlength = iscsi_getlength,
1698 .bdrv_get_info = iscsi_get_info,
1699 .bdrv_truncate = iscsi_truncate,
1700 .bdrv_refresh_limits = iscsi_refresh_limits,
1701
1702 .bdrv_co_get_block_status = iscsi_co_get_block_status,
1703 .bdrv_co_discard = iscsi_co_discard,
1704 .bdrv_co_write_zeroes = iscsi_co_write_zeroes,
1705 .bdrv_co_readv = iscsi_co_readv,
1706 .bdrv_co_writev = iscsi_co_writev,
1707 .bdrv_co_flush_to_disk = iscsi_co_flush,
1708
1709 #ifdef __linux__
1710 .bdrv_ioctl = iscsi_ioctl,
1711 .bdrv_aio_ioctl = iscsi_aio_ioctl,
1712 #endif
1713
1714 .bdrv_detach_aio_context = iscsi_detach_aio_context,
1715 .bdrv_attach_aio_context = iscsi_attach_aio_context,
1716 };
1717
1718 static QemuOptsList qemu_iscsi_opts = {
1719 .name = "iscsi",
1720 .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1721 .desc = {
1722 {
1723 .name = "user",
1724 .type = QEMU_OPT_STRING,
1725 .help = "username for CHAP authentication to target",
1726 },{
1727 .name = "password",
1728 .type = QEMU_OPT_STRING,
1729 .help = "password for CHAP authentication to target",
1730 },{
1731 .name = "header-digest",
1732 .type = QEMU_OPT_STRING,
1733 .help = "HeaderDigest setting. "
1734 "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1735 },{
1736 .name = "initiator-name",
1737 .type = QEMU_OPT_STRING,
1738 .help = "Initiator iqn name to use when connecting",
1739 },
1740 { /* end of list */ }
1741 },
1742 };
1743
1744 static void iscsi_block_init(void)
1745 {
1746 bdrv_register(&bdrv_iscsi);
1747 qemu_add_opts(&qemu_iscsi_opts);
1748 }
1749
1750 block_init(iscsi_block_init);
armbru's QEMU hackery