?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
memory: fix usage of find_next_bit and find_next_zero_bit
[qemu/rayw.git] / block / iscsi.c
blob9fe76f48ec5e3a6cc24b1c21e1505f4c22f11915
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
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 bool request_timed_out;
74 } IscsiLun;
75
76 typedef struct IscsiTask {
77 int status;
78 int complete;
79 int retries;
80 int do_retry;
81 struct scsi_task *task;
82 Coroutine *co;
83 QEMUBH *bh;
84 IscsiLun *iscsilun;
85 QEMUTimer retry_timer;
86 bool force_next_flush;
87 int err_code;
88 } IscsiTask;
89
90 typedef struct IscsiAIOCB {
91 BlockAIOCB common;
92 QEMUIOVector *qiov;
93 QEMUBH *bh;
94 IscsiLun *iscsilun;
95 struct scsi_task *task;
96 uint8_t *buf;
97 int status;
98 int64_t sector_num;
99 int nb_sectors;
100 int ret;
101 #ifdef __linux__
102 sg_io_hdr_t *ioh;
103 #endif
104 } IscsiAIOCB;
105
106 /* libiscsi uses time_t so its enough to process events every second */
107 #define EVENT_INTERVAL 1000
108 #define NOP_INTERVAL 5000
109 #define MAX_NOP_FAILURES 3
110 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times)
111 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768};
112
113 /* this threshold is a trade-off knob to choose between
114 * the potential additional overhead of an extra GET_LBA_STATUS request
115 * vs. unnecessarily reading a lot of zero sectors over the wire.
116 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
117 * sectors we check the allocation status of the area covered by the
118 * request first if the allocationmap indicates that the area might be
119 * unallocated. */
120 #define ISCSI_CHECKALLOC_THRES 64
121
122 static void
123 iscsi_bh_cb(void *p)
124 {
125 IscsiAIOCB *acb = p;
126
127 qemu_bh_delete(acb->bh);
128
129 g_free(acb->buf);
130 acb->buf = NULL;
131
132 acb->common.cb(acb->common.opaque, acb->status);
133
134 if (acb->task != NULL) {
135 scsi_free_scsi_task(acb->task);
136 acb->task = NULL;
137 }
138
139 qemu_aio_unref(acb);
140 }
141
142 static void
143 iscsi_schedule_bh(IscsiAIOCB *acb)
144 {
145 if (acb->bh) {
146 return;
147 }
148 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb);
149 qemu_bh_schedule(acb->bh);
150 }
151
152 static void iscsi_co_generic_bh_cb(void *opaque)
153 {
154 struct IscsiTask *iTask = opaque;
155 iTask->complete = 1;
156 qemu_bh_delete(iTask->bh);
157 qemu_coroutine_enter(iTask->co, NULL);
158 }
159
160 static void iscsi_retry_timer_expired(void *opaque)
161 {
162 struct IscsiTask *iTask = opaque;
163 iTask->complete = 1;
164 if (iTask->co) {
165 qemu_coroutine_enter(iTask->co, NULL);
166 }
167 }
168
169 static inline unsigned exp_random(double mean)
170 {
171 return -mean * log((double)rand() / RAND_MAX);
172 }
173
174 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in
175 * libiscsi 1.10.0, together with other constants we need. Use it as
176 * a hint that we have to define them ourselves if needed, to keep the
177 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for
178 * the test because SCSI_STATUS_* is an enum.
179 *
180 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes
181 * an enum, check against the LIBISCSI_API_VERSION macro, which was
182 * introduced in 1.11.0. If it is present, there is no need to define
183 * anything.
184 */
185 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \
186 !defined(LIBISCSI_API_VERSION)
187 #define SCSI_STATUS_TASK_SET_FULL 0x28
188 #define SCSI_STATUS_TIMEOUT 0x0f000002
189 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600
190 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00
191 #endif
192
193 static int iscsi_translate_sense(struct scsi_sense *sense)
194 {
195 int ret;
196
197 switch (sense->key) {
198 case SCSI_SENSE_NOT_READY:
199 return -EBUSY;
200 case SCSI_SENSE_DATA_PROTECTION:
201 return -EACCES;
202 case SCSI_SENSE_COMMAND_ABORTED:
203 return -ECANCELED;
204 case SCSI_SENSE_ILLEGAL_REQUEST:
205 /* Parse ASCQ */
206 break;
207 default:
208 return -EIO;
209 }
210 switch (sense->ascq) {
211 case SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR:
212 case SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE:
213 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB:
214 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST:
215 ret = -EINVAL;
216 break;
217 case SCSI_SENSE_ASCQ_LBA_OUT_OF_RANGE:
218 ret = -ENOSPC;
219 break;
220 case SCSI_SENSE_ASCQ_LOGICAL_UNIT_NOT_SUPPORTED:
221 ret = -ENOTSUP;
222 break;
223 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT:
224 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_CLOSED:
225 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_OPEN:
226 ret = -ENOMEDIUM;
227 break;
228 case SCSI_SENSE_ASCQ_WRITE_PROTECTED:
229 ret = -EACCES;
230 break;
231 default:
232 ret = -EIO;
233 break;
234 }
235 return ret;
236 }
237
238 static void
239 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
240 void *command_data, void *opaque)
241 {
242 struct IscsiTask *iTask = opaque;
243 struct scsi_task *task = command_data;
244
245 iTask->status = status;
246 iTask->do_retry = 0;
247 iTask->task = task;
248
249 if (status != SCSI_STATUS_GOOD) {
250 if (iTask->retries++ < ISCSI_CMD_RETRIES) {
251 if (status == SCSI_STATUS_CHECK_CONDITION
252 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
253 error_report("iSCSI CheckCondition: %s",
254 iscsi_get_error(iscsi));
255 iTask->do_retry = 1;
256 goto out;
257 }
258 if (status == SCSI_STATUS_BUSY ||
259 status == SCSI_STATUS_TIMEOUT ||
260 status == SCSI_STATUS_TASK_SET_FULL) {
261 unsigned retry_time =
262 exp_random(iscsi_retry_times[iTask->retries - 1]);
263 if (status == SCSI_STATUS_TIMEOUT) {
264 /* make sure the request is rescheduled AFTER the
265 * reconnect is initiated */
266 retry_time = EVENT_INTERVAL * 2;
267 iTask->iscsilun->request_timed_out = true;
268 }
269 error_report("iSCSI Busy/TaskSetFull/TimeOut"
270 " (retry #%u in %u ms): %s",
271 iTask->retries, retry_time,
272 iscsi_get_error(iscsi));
273 aio_timer_init(iTask->iscsilun->aio_context,
274 &iTask->retry_timer, QEMU_CLOCK_REALTIME,
275 SCALE_MS, iscsi_retry_timer_expired, iTask);
276 timer_mod(&iTask->retry_timer,
277 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
278 iTask->do_retry = 1;
279 return;
280 }
281 }
282 iTask->err_code = iscsi_translate_sense(&task->sense);
283 error_report("iSCSI Failure: %s", iscsi_get_error(iscsi));
284 } else {
285 iTask->iscsilun->force_next_flush |= iTask->force_next_flush;
286 }
287
288 out:
289 if (iTask->co) {
290 iTask->bh = aio_bh_new(iTask->iscsilun->aio_context,
291 iscsi_co_generic_bh_cb, iTask);
292 qemu_bh_schedule(iTask->bh);
293 } else {
294 iTask->complete = 1;
295 }
296 }
297
298 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
299 {
300 *iTask = (struct IscsiTask) {
301 .co = qemu_coroutine_self(),
302 .iscsilun = iscsilun,
303 };
304 }
305
306 static void
307 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
308 void *private_data)
309 {
310 IscsiAIOCB *acb = private_data;
311
312 acb->status = -ECANCELED;
313 iscsi_schedule_bh(acb);
314 }
315
316 static void
317 iscsi_aio_cancel(BlockAIOCB *blockacb)
318 {
319 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
320 IscsiLun *iscsilun = acb->iscsilun;
321
322 if (acb->status != -EINPROGRESS) {
323 return;
324 }
325
326 /* send a task mgmt call to the target to cancel the task on the target */
327 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
328 iscsi_abort_task_cb, acb);
329
330 }
331
332 static const AIOCBInfo iscsi_aiocb_info = {
333 .aiocb_size = sizeof(IscsiAIOCB),
334 .cancel_async = iscsi_aio_cancel,
335 };
336
337
338 static void iscsi_process_read(void *arg);
339 static void iscsi_process_write(void *arg);
340
341 static void
342 iscsi_set_events(IscsiLun *iscsilun)
343 {
344 struct iscsi_context *iscsi = iscsilun->iscsi;
345 int ev = iscsi_which_events(iscsi);
346
347 if (ev != iscsilun->events) {
348 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi),
349 false,
350 (ev & POLLIN) ? iscsi_process_read : NULL,
351 (ev & POLLOUT) ? iscsi_process_write : NULL,
352 iscsilun);
353 iscsilun->events = ev;
354 }
355 }
356
357 static void iscsi_timed_check_events(void *opaque)
358 {
359 IscsiLun *iscsilun = opaque;
360
361 /* check for timed out requests */
362 iscsi_service(iscsilun->iscsi, 0);
363
364 if (iscsilun->request_timed_out) {
365 iscsilun->request_timed_out = false;
366 iscsi_reconnect(iscsilun->iscsi);
367 }
368
369 /* newer versions of libiscsi may return zero events. Ensure we are able
370 * to return to service once this situation changes. */
371 iscsi_set_events(iscsilun);
372
373 timer_mod(iscsilun->event_timer,
374 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
375 }
376
377 static void
378 iscsi_process_read(void *arg)
379 {
380 IscsiLun *iscsilun = arg;
381 struct iscsi_context *iscsi = iscsilun->iscsi;
382
383 iscsi_service(iscsi, POLLIN);
384 iscsi_set_events(iscsilun);
385 }
386
387 static void
388 iscsi_process_write(void *arg)
389 {
390 IscsiLun *iscsilun = arg;
391 struct iscsi_context *iscsi = iscsilun->iscsi;
392
393 iscsi_service(iscsi, POLLOUT);
394 iscsi_set_events(iscsilun);
395 }
396
397 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
398 {
399 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
400 }
401
402 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
403 {
404 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
405 }
406
407 static bool is_request_lun_aligned(int64_t sector_num, int nb_sectors,
408 IscsiLun *iscsilun)
409 {
410 if ((sector_num * BDRV_SECTOR_SIZE) % iscsilun->block_size ||
411 (nb_sectors * BDRV_SECTOR_SIZE) % iscsilun->block_size) {
412 error_report("iSCSI misaligned request: "
413 "iscsilun->block_size %u, sector_num %" PRIi64
414 ", nb_sectors %d",
415 iscsilun->block_size, sector_num, nb_sectors);
416 return 0;
417 }
418 return 1;
419 }
420
421 static unsigned long *iscsi_allocationmap_init(IscsiLun *iscsilun)
422 {
423 return bitmap_try_new(DIV_ROUND_UP(sector_lun2qemu(iscsilun->num_blocks,
424 iscsilun),
425 iscsilun->cluster_sectors));
426 }
427
428 static void iscsi_allocationmap_set(IscsiLun *iscsilun, int64_t sector_num,
429 int nb_sectors)
430 {
431 if (iscsilun->allocationmap == NULL) {
432 return;
433 }
434 bitmap_set(iscsilun->allocationmap,
435 sector_num / iscsilun->cluster_sectors,
436 DIV_ROUND_UP(nb_sectors, iscsilun->cluster_sectors));
437 }
438
439 static void iscsi_allocationmap_clear(IscsiLun *iscsilun, int64_t sector_num,
440 int nb_sectors)
441 {
442 int64_t cluster_num, nb_clusters;
443 if (iscsilun->allocationmap == NULL) {
444 return;
445 }
446 cluster_num = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors);
447 nb_clusters = (sector_num + nb_sectors) / iscsilun->cluster_sectors
448 - cluster_num;
449 if (nb_clusters > 0) {
450 bitmap_clear(iscsilun->allocationmap, cluster_num, nb_clusters);
451 }
452 }
453
454 static int coroutine_fn iscsi_co_writev(BlockDriverState *bs,
455 int64_t sector_num, int nb_sectors,
456 QEMUIOVector *iov)
457 {
458 IscsiLun *iscsilun = bs->opaque;
459 struct IscsiTask iTask;
460 uint64_t lba;
461 uint32_t num_sectors;
462 int fua;
463
464 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
465 return -EINVAL;
466 }
467
468 if (bs->bl.max_transfer_length && nb_sectors > bs->bl.max_transfer_length) {
469 error_report("iSCSI Error: Write of %d sectors exceeds max_xfer_len "
470 "of %d sectors", nb_sectors, bs->bl.max_transfer_length);
471 return -EINVAL;
472 }
473
474 lba = sector_qemu2lun(sector_num, iscsilun);
475 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
476 iscsi_co_init_iscsitask(iscsilun, &iTask);
477 retry:
478 fua = iscsilun->dpofua && !bs->enable_write_cache;
479 iTask.force_next_flush = !fua;
480 if (iscsilun->use_16_for_rw) {
481 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
482 NULL, num_sectors * iscsilun->block_size,
483 iscsilun->block_size, 0, 0, fua, 0, 0,
484 iscsi_co_generic_cb, &iTask);
485 } else {
486 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
487 NULL, num_sectors * iscsilun->block_size,
488 iscsilun->block_size, 0, 0, fua, 0, 0,
489 iscsi_co_generic_cb, &iTask);
490 }
491 if (iTask.task == NULL) {
492 return -ENOMEM;
493 }
494 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
495 iov->niov);
496 while (!iTask.complete) {
497 iscsi_set_events(iscsilun);
498 qemu_coroutine_yield();
499 }
500
501 if (iTask.task != NULL) {
502 scsi_free_scsi_task(iTask.task);
503 iTask.task = NULL;
504 }
505
506 if (iTask.do_retry) {
507 iTask.complete = 0;
508 goto retry;
509 }
510
511 if (iTask.status != SCSI_STATUS_GOOD) {
512 return iTask.err_code;
513 }
514
515 iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors);
516
517 return 0;
518 }
519
520
521 static bool iscsi_allocationmap_is_allocated(IscsiLun *iscsilun,
522 int64_t sector_num, int nb_sectors)
523 {
524 unsigned long size;
525 if (iscsilun->allocationmap == NULL) {
526 return true;
527 }
528 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
529 return !(find_next_bit(iscsilun->allocationmap, size,
530 sector_num / iscsilun->cluster_sectors) == size);
531 }
532
533 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
534 int64_t sector_num,
535 int nb_sectors, int *pnum,
536 BlockDriverState **file)
537 {
538 IscsiLun *iscsilun = bs->opaque;
539 struct scsi_get_lba_status *lbas = NULL;
540 struct scsi_lba_status_descriptor *lbasd = NULL;
541 struct IscsiTask iTask;
542 int64_t ret;
543
544 iscsi_co_init_iscsitask(iscsilun, &iTask);
545
546 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
547 ret = -EINVAL;
548 goto out;
549 }
550
551 /* default to all sectors allocated */
552 ret = BDRV_BLOCK_DATA;
553 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
554 *pnum = nb_sectors;
555
556 /* LUN does not support logical block provisioning */
557 if (!iscsilun->lbpme) {
558 goto out;
559 }
560
561 retry:
562 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
563 sector_qemu2lun(sector_num, iscsilun),
564 8 + 16, iscsi_co_generic_cb,
565 &iTask) == NULL) {
566 ret = -ENOMEM;
567 goto out;
568 }
569
570 while (!iTask.complete) {
571 iscsi_set_events(iscsilun);
572 qemu_coroutine_yield();
573 }
574
575 if (iTask.do_retry) {
576 if (iTask.task != NULL) {
577 scsi_free_scsi_task(iTask.task);
578 iTask.task = NULL;
579 }
580 iTask.complete = 0;
581 goto retry;
582 }
583
584 if (iTask.status != SCSI_STATUS_GOOD) {
585 /* in case the get_lba_status_callout fails (i.e.
586 * because the device is busy or the cmd is not
587 * supported) we pretend all blocks are allocated
588 * for backwards compatibility */
589 goto out;
590 }
591
592 lbas = scsi_datain_unmarshall(iTask.task);
593 if (lbas == NULL) {
594 ret = -EIO;
595 goto out;
596 }
597
598 lbasd = &lbas->descriptors[0];
599
600 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
601 ret = -EIO;
602 goto out;
603 }
604
605 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
606
607 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
608 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
609 ret &= ~BDRV_BLOCK_DATA;
610 if (iscsilun->lbprz) {
611 ret |= BDRV_BLOCK_ZERO;
612 }
613 }
614
615 if (ret & BDRV_BLOCK_ZERO) {
616 iscsi_allocationmap_clear(iscsilun, sector_num, *pnum);
617 } else {
618 iscsi_allocationmap_set(iscsilun, sector_num, *pnum);
619 }
620
621 if (*pnum > nb_sectors) {
622 *pnum = nb_sectors;
623 }
624 out:
625 if (iTask.task != NULL) {
626 scsi_free_scsi_task(iTask.task);
627 }
628 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) {
629 *file = bs;
630 }
631 return ret;
632 }
633
634 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
635 int64_t sector_num, int nb_sectors,
636 QEMUIOVector *iov)
637 {
638 IscsiLun *iscsilun = bs->opaque;
639 struct IscsiTask iTask;
640 uint64_t lba;
641 uint32_t num_sectors;
642
643 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
644 return -EINVAL;
645 }
646
647 if (bs->bl.max_transfer_length && nb_sectors > bs->bl.max_transfer_length) {
648 error_report("iSCSI Error: Read of %d sectors exceeds max_xfer_len "
649 "of %d sectors", nb_sectors, bs->bl.max_transfer_length);
650 return -EINVAL;
651 }
652
653 if (iscsilun->lbprz && nb_sectors >= ISCSI_CHECKALLOC_THRES &&
654 !iscsi_allocationmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
655 int64_t ret;
656 int pnum;
657 BlockDriverState *file;
658 ret = iscsi_co_get_block_status(bs, sector_num, INT_MAX, &pnum, &file);
659 if (ret < 0) {
660 return ret;
661 }
662 if (ret & BDRV_BLOCK_ZERO && pnum >= nb_sectors) {
663 qemu_iovec_memset(iov, 0, 0x00, iov->size);
664 return 0;
665 }
666 }
667
668 lba = sector_qemu2lun(sector_num, iscsilun);
669 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
670
671 iscsi_co_init_iscsitask(iscsilun, &iTask);
672 retry:
673 if (iscsilun->use_16_for_rw) {
674 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
675 num_sectors * iscsilun->block_size,
676 iscsilun->block_size, 0, 0, 0, 0, 0,
677 iscsi_co_generic_cb, &iTask);
678 } else {
679 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
680 num_sectors * iscsilun->block_size,
681 iscsilun->block_size,
682 0, 0, 0, 0, 0,
683 iscsi_co_generic_cb, &iTask);
684 }
685 if (iTask.task == NULL) {
686 return -ENOMEM;
687 }
688 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
689
690 while (!iTask.complete) {
691 iscsi_set_events(iscsilun);
692 qemu_coroutine_yield();
693 }
694
695 if (iTask.task != NULL) {
696 scsi_free_scsi_task(iTask.task);
697 iTask.task = NULL;
698 }
699
700 if (iTask.do_retry) {
701 iTask.complete = 0;
702 goto retry;
703 }
704
705 if (iTask.status != SCSI_STATUS_GOOD) {
706 return iTask.err_code;
707 }
708
709 return 0;
710 }
711
712 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
713 {
714 IscsiLun *iscsilun = bs->opaque;
715 struct IscsiTask iTask;
716
717 if (!iscsilun->force_next_flush) {
718 return 0;
719 }
720 iscsilun->force_next_flush = false;
721
722 iscsi_co_init_iscsitask(iscsilun, &iTask);
723 retry:
724 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
725 0, iscsi_co_generic_cb, &iTask) == NULL) {
726 return -ENOMEM;
727 }
728
729 while (!iTask.complete) {
730 iscsi_set_events(iscsilun);
731 qemu_coroutine_yield();
732 }
733
734 if (iTask.task != NULL) {
735 scsi_free_scsi_task(iTask.task);
736 iTask.task = NULL;
737 }
738
739 if (iTask.do_retry) {
740 iTask.complete = 0;
741 goto retry;
742 }
743
744 if (iTask.status != SCSI_STATUS_GOOD) {
745 return iTask.err_code;
746 }
747
748 return 0;
749 }
750
751 #ifdef __linux__
752 static void
753 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
754 void *command_data, void *opaque)
755 {
756 IscsiAIOCB *acb = opaque;
757
758 g_free(acb->buf);
759 acb->buf = NULL;
760
761 acb->status = 0;
762 if (status < 0) {
763 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
764 iscsi_get_error(iscsi));
765 acb->status = iscsi_translate_sense(&acb->task->sense);
766 }
767
768 acb->ioh->driver_status = 0;
769 acb->ioh->host_status = 0;
770 acb->ioh->resid = 0;
771
772 #define SG_ERR_DRIVER_SENSE 0x08
773
774 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
775 int ss;
776
777 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
778
779 acb->ioh->sb_len_wr = acb->task->datain.size - 2;
780 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
781 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
782 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
783 }
784
785 iscsi_schedule_bh(acb);
786 }
787
788 static void iscsi_ioctl_bh_completion(void *opaque)
789 {
790 IscsiAIOCB *acb = opaque;
791
792 qemu_bh_delete(acb->bh);
793 acb->common.cb(acb->common.opaque, acb->ret);
794 qemu_aio_unref(acb);
795 }
796
797 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
798 {
799 BlockDriverState *bs = acb->common.bs;
800 IscsiLun *iscsilun = bs->opaque;
801 int ret = 0;
802
803 switch (req) {
804 case SG_GET_VERSION_NUM:
805 *(int *)buf = 30000;
806 break;
807 case SG_GET_SCSI_ID:
808 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
809 break;
810 default:
811 ret = -EINVAL;
812 }
813 assert(!acb->bh);
814 acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
815 iscsi_ioctl_bh_completion, acb);
816 acb->ret = ret;
817 qemu_bh_schedule(acb->bh);
818 }
819
820 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
821 unsigned long int req, void *buf,
822 BlockCompletionFunc *cb, void *opaque)
823 {
824 IscsiLun *iscsilun = bs->opaque;
825 struct iscsi_context *iscsi = iscsilun->iscsi;
826 struct iscsi_data data;
827 IscsiAIOCB *acb;
828
829 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
830
831 acb->iscsilun = iscsilun;
832 acb->bh = NULL;
833 acb->status = -EINPROGRESS;
834 acb->buf = NULL;
835 acb->ioh = buf;
836
837 if (req != SG_IO) {
838 iscsi_ioctl_handle_emulated(acb, req, buf);
839 return &acb->common;
840 }
841
842 acb->task = malloc(sizeof(struct scsi_task));
843 if (acb->task == NULL) {
844 error_report("iSCSI: Failed to allocate task for scsi command. %s",
845 iscsi_get_error(iscsi));
846 qemu_aio_unref(acb);
847 return NULL;
848 }
849 memset(acb->task, 0, sizeof(struct scsi_task));
850
851 switch (acb->ioh->dxfer_direction) {
852 case SG_DXFER_TO_DEV:
853 acb->task->xfer_dir = SCSI_XFER_WRITE;
854 break;
855 case SG_DXFER_FROM_DEV:
856 acb->task->xfer_dir = SCSI_XFER_READ;
857 break;
858 default:
859 acb->task->xfer_dir = SCSI_XFER_NONE;
860 break;
861 }
862
863 acb->task->cdb_size = acb->ioh->cmd_len;
864 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
865 acb->task->expxferlen = acb->ioh->dxfer_len;
866
867 data.size = 0;
868 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
869 if (acb->ioh->iovec_count == 0) {
870 data.data = acb->ioh->dxferp;
871 data.size = acb->ioh->dxfer_len;
872 } else {
873 scsi_task_set_iov_out(acb->task,
874 (struct scsi_iovec *) acb->ioh->dxferp,
875 acb->ioh->iovec_count);
876 }
877 }
878
879 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
880 iscsi_aio_ioctl_cb,
881 (data.size > 0) ? &data : NULL,
882 acb) != 0) {
883 scsi_free_scsi_task(acb->task);
884 qemu_aio_unref(acb);
885 return NULL;
886 }
887
888 /* tell libiscsi to read straight into the buffer we got from ioctl */
889 if (acb->task->xfer_dir == SCSI_XFER_READ) {
890 if (acb->ioh->iovec_count == 0) {
891 scsi_task_add_data_in_buffer(acb->task,
892 acb->ioh->dxfer_len,
893 acb->ioh->dxferp);
894 } else {
895 scsi_task_set_iov_in(acb->task,
896 (struct scsi_iovec *) acb->ioh->dxferp,
897 acb->ioh->iovec_count);
898 }
899 }
900
901 iscsi_set_events(iscsilun);
902
903 return &acb->common;
904 }
905
906 #endif
907
908 static int64_t
909 iscsi_getlength(BlockDriverState *bs)
910 {
911 IscsiLun *iscsilun = bs->opaque;
912 int64_t len;
913
914 len = iscsilun->num_blocks;
915 len *= iscsilun->block_size;
916
917 return len;
918 }
919
920 static int
921 coroutine_fn iscsi_co_discard(BlockDriverState *bs, int64_t sector_num,
922 int nb_sectors)
923 {
924 IscsiLun *iscsilun = bs->opaque;
925 struct IscsiTask iTask;
926 struct unmap_list list;
927
928 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
929 return -EINVAL;
930 }
931
932 if (!iscsilun->lbp.lbpu) {
933 /* UNMAP is not supported by the target */
934 return 0;
935 }
936
937 list.lba = sector_qemu2lun(sector_num, iscsilun);
938 list.num = sector_qemu2lun(nb_sectors, iscsilun);
939
940 iscsi_co_init_iscsitask(iscsilun, &iTask);
941 retry:
942 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
943 iscsi_co_generic_cb, &iTask) == NULL) {
944 return -ENOMEM;
945 }
946
947 while (!iTask.complete) {
948 iscsi_set_events(iscsilun);
949 qemu_coroutine_yield();
950 }
951
952 if (iTask.task != NULL) {
953 scsi_free_scsi_task(iTask.task);
954 iTask.task = NULL;
955 }
956
957 if (iTask.do_retry) {
958 iTask.complete = 0;
959 goto retry;
960 }
961
962 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
963 /* the target might fail with a check condition if it
964 is not happy with the alignment of the UNMAP request
965 we silently fail in this case */
966 return 0;
967 }
968
969 if (iTask.status != SCSI_STATUS_GOOD) {
970 return iTask.err_code;
971 }
972
973 iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors);
974
975 return 0;
976 }
977
978 static int
979 coroutine_fn iscsi_co_write_zeroes(BlockDriverState *bs, int64_t sector_num,
980 int nb_sectors, BdrvRequestFlags flags)
981 {
982 IscsiLun *iscsilun = bs->opaque;
983 struct IscsiTask iTask;
984 uint64_t lba;
985 uint32_t nb_blocks;
986 bool use_16_for_ws = iscsilun->use_16_for_rw;
987
988 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
989 return -EINVAL;
990 }
991
992 if (flags & BDRV_REQ_MAY_UNMAP) {
993 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
994 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
995 use_16_for_ws = true;
996 }
997 if (use_16_for_ws && !iscsilun->lbp.lbpws) {
998 /* WRITESAME16 with UNMAP is not supported by the target,
999 * fall back and try WRITESAME10/16 without UNMAP */
1000 flags &= ~BDRV_REQ_MAY_UNMAP;
1001 use_16_for_ws = iscsilun->use_16_for_rw;
1002 }
1003 }
1004
1005 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1006 /* WRITESAME without UNMAP is not supported by the target */
1007 return -ENOTSUP;
1008 }
1009
1010 lba = sector_qemu2lun(sector_num, iscsilun);
1011 nb_blocks = sector_qemu2lun(nb_sectors, iscsilun);
1012
1013 if (iscsilun->zeroblock == NULL) {
1014 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1015 if (iscsilun->zeroblock == NULL) {
1016 return -ENOMEM;
1017 }
1018 }
1019
1020 iscsi_co_init_iscsitask(iscsilun, &iTask);
1021 iTask.force_next_flush = true;
1022 retry:
1023 if (use_16_for_ws) {
1024 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1025 iscsilun->zeroblock, iscsilun->block_size,
1026 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1027 0, 0, iscsi_co_generic_cb, &iTask);
1028 } else {
1029 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1030 iscsilun->zeroblock, iscsilun->block_size,
1031 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1032 0, 0, iscsi_co_generic_cb, &iTask);
1033 }
1034 if (iTask.task == NULL) {
1035 return -ENOMEM;
1036 }
1037
1038 while (!iTask.complete) {
1039 iscsi_set_events(iscsilun);
1040 qemu_coroutine_yield();
1041 }
1042
1043 if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1044 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1045 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1046 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1047 /* WRITE SAME is not supported by the target */
1048 iscsilun->has_write_same = false;
1049 scsi_free_scsi_task(iTask.task);
1050 return -ENOTSUP;
1051 }
1052
1053 if (iTask.task != NULL) {
1054 scsi_free_scsi_task(iTask.task);
1055 iTask.task = NULL;
1056 }
1057
1058 if (iTask.do_retry) {
1059 iTask.complete = 0;
1060 goto retry;
1061 }
1062
1063 if (iTask.status != SCSI_STATUS_GOOD) {
1064 return iTask.err_code;
1065 }
1066
1067 if (flags & BDRV_REQ_MAY_UNMAP) {
1068 iscsi_allocationmap_clear(iscsilun, sector_num, nb_sectors);
1069 } else {
1070 iscsi_allocationmap_set(iscsilun, sector_num, nb_sectors);
1071 }
1072
1073 return 0;
1074 }
1075
1076 static void parse_chap(struct iscsi_context *iscsi, const char *target,
1077 Error **errp)
1078 {
1079 QemuOptsList *list;
1080 QemuOpts *opts;
1081 const char *user = NULL;
1082 const char *password = NULL;
1083
1084 list = qemu_find_opts("iscsi");
1085 if (!list) {
1086 return;
1087 }
1088
1089 opts = qemu_opts_find(list, target);
1090 if (opts == NULL) {
1091 opts = QTAILQ_FIRST(&list->head);
1092 if (!opts) {
1093 return;
1094 }
1095 }
1096
1097 user = qemu_opt_get(opts, "user");
1098 if (!user) {
1099 return;
1100 }
1101
1102 password = qemu_opt_get(opts, "password");
1103 if (!password) {
1104 error_setg(errp, "CHAP username specified but no password was given");
1105 return;
1106 }
1107
1108 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1109 error_setg(errp, "Failed to set initiator username and password");
1110 }
1111 }
1112
1113 static void parse_header_digest(struct iscsi_context *iscsi, const char *target,
1114 Error **errp)
1115 {
1116 QemuOptsList *list;
1117 QemuOpts *opts;
1118 const char *digest = NULL;
1119
1120 list = qemu_find_opts("iscsi");
1121 if (!list) {
1122 return;
1123 }
1124
1125 opts = qemu_opts_find(list, target);
1126 if (opts == NULL) {
1127 opts = QTAILQ_FIRST(&list->head);
1128 if (!opts) {
1129 return;
1130 }
1131 }
1132
1133 digest = qemu_opt_get(opts, "header-digest");
1134 if (!digest) {
1135 return;
1136 }
1137
1138 if (!strcmp(digest, "CRC32C")) {
1139 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1140 } else if (!strcmp(digest, "NONE")) {
1141 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1142 } else if (!strcmp(digest, "CRC32C-NONE")) {
1143 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1144 } else if (!strcmp(digest, "NONE-CRC32C")) {
1145 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1146 } else {
1147 error_setg(errp, "Invalid header-digest setting : %s", digest);
1148 }
1149 }
1150
1151 static char *parse_initiator_name(const char *target)
1152 {
1153 QemuOptsList *list;
1154 QemuOpts *opts;
1155 const char *name;
1156 char *iscsi_name;
1157 UuidInfo *uuid_info;
1158
1159 list = qemu_find_opts("iscsi");
1160 if (list) {
1161 opts = qemu_opts_find(list, target);
1162 if (!opts) {
1163 opts = QTAILQ_FIRST(&list->head);
1164 }
1165 if (opts) {
1166 name = qemu_opt_get(opts, "initiator-name");
1167 if (name) {
1168 return g_strdup(name);
1169 }
1170 }
1171 }
1172
1173 uuid_info = qmp_query_uuid(NULL);
1174 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1175 name = qemu_get_vm_name();
1176 } else {
1177 name = uuid_info->UUID;
1178 }
1179 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1180 name ? ":" : "", name ? name : "");
1181 qapi_free_UuidInfo(uuid_info);
1182 return iscsi_name;
1183 }
1184
1185 static int parse_timeout(const char *target)
1186 {
1187 QemuOptsList *list;
1188 QemuOpts *opts;
1189 const char *timeout;
1190
1191 list = qemu_find_opts("iscsi");
1192 if (list) {
1193 opts = qemu_opts_find(list, target);
1194 if (!opts) {
1195 opts = QTAILQ_FIRST(&list->head);
1196 }
1197 if (opts) {
1198 timeout = qemu_opt_get(opts, "timeout");
1199 if (timeout) {
1200 return atoi(timeout);
1201 }
1202 }
1203 }
1204
1205 return 0;
1206 }
1207
1208 static void iscsi_nop_timed_event(void *opaque)
1209 {
1210 IscsiLun *iscsilun = opaque;
1211
1212 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1213 error_report("iSCSI: NOP timeout. Reconnecting...");
1214 iscsilun->request_timed_out = true;
1215 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1216 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1217 return;
1218 }
1219
1220 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1221 iscsi_set_events(iscsilun);
1222 }
1223
1224 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1225 {
1226 struct scsi_task *task = NULL;
1227 struct scsi_readcapacity10 *rc10 = NULL;
1228 struct scsi_readcapacity16 *rc16 = NULL;
1229 int retries = ISCSI_CMD_RETRIES;
1230
1231 do {
1232 if (task != NULL) {
1233 scsi_free_scsi_task(task);
1234 task = NULL;
1235 }
1236
1237 switch (iscsilun->type) {
1238 case TYPE_DISK:
1239 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1240 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1241 rc16 = scsi_datain_unmarshall(task);
1242 if (rc16 == NULL) {
1243 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1244 } else {
1245 iscsilun->block_size = rc16->block_length;
1246 iscsilun->num_blocks = rc16->returned_lba + 1;
1247 iscsilun->lbpme = !!rc16->lbpme;
1248 iscsilun->lbprz = !!rc16->lbprz;
1249 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1250 }
1251 break;
1252 }
1253 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1254 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1255 break;
1256 }
1257 /* Fall through and try READ CAPACITY(10) instead. */
1258 case TYPE_ROM:
1259 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1260 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1261 rc10 = scsi_datain_unmarshall(task);
1262 if (rc10 == NULL) {
1263 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1264 } else {
1265 iscsilun->block_size = rc10->block_size;
1266 if (rc10->lba == 0) {
1267 /* blank disk loaded */
1268 iscsilun->num_blocks = 0;
1269 } else {
1270 iscsilun->num_blocks = rc10->lba + 1;
1271 }
1272 }
1273 }
1274 break;
1275 default:
1276 return;
1277 }
1278 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1279 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1280 && retries-- > 0);
1281
1282 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1283 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1284 } else if (!iscsilun->block_size ||
1285 iscsilun->block_size % BDRV_SECTOR_SIZE) {
1286 error_setg(errp, "iSCSI: the target returned an invalid "
1287 "block size of %d.", iscsilun->block_size);
1288 }
1289 if (task) {
1290 scsi_free_scsi_task(task);
1291 }
1292 }
1293
1294 /* TODO Convert to fine grained options */
1295 static QemuOptsList runtime_opts = {
1296 .name = "iscsi",
1297 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1298 .desc = {
1299 {
1300 .name = "filename",
1301 .type = QEMU_OPT_STRING,
1302 .help = "URL to the iscsi image",
1303 },
1304 { /* end of list */ }
1305 },
1306 };
1307
1308 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1309 int evpd, int pc, void **inq, Error **errp)
1310 {
1311 int full_size;
1312 struct scsi_task *task = NULL;
1313 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1314 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1315 goto fail;
1316 }
1317 full_size = scsi_datain_getfullsize(task);
1318 if (full_size > task->datain.size) {
1319 scsi_free_scsi_task(task);
1320
1321 /* we need more data for the full list */
1322 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1323 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1324 goto fail;
1325 }
1326 }
1327
1328 *inq = scsi_datain_unmarshall(task);
1329 if (*inq == NULL) {
1330 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1331 goto fail_with_err;
1332 }
1333
1334 return task;
1335
1336 fail:
1337 error_setg(errp, "iSCSI: Inquiry command failed : %s",
1338 iscsi_get_error(iscsi));
1339 fail_with_err:
1340 if (task != NULL) {
1341 scsi_free_scsi_task(task);
1342 }
1343 return NULL;
1344 }
1345
1346 static void iscsi_detach_aio_context(BlockDriverState *bs)
1347 {
1348 IscsiLun *iscsilun = bs->opaque;
1349
1350 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1351 false, NULL, NULL, NULL);
1352 iscsilun->events = 0;
1353
1354 if (iscsilun->nop_timer) {
1355 timer_del(iscsilun->nop_timer);
1356 timer_free(iscsilun->nop_timer);
1357 iscsilun->nop_timer = NULL;
1358 }
1359 if (iscsilun->event_timer) {
1360 timer_del(iscsilun->event_timer);
1361 timer_free(iscsilun->event_timer);
1362 iscsilun->event_timer = NULL;
1363 }
1364 }
1365
1366 static void iscsi_attach_aio_context(BlockDriverState *bs,
1367 AioContext *new_context)
1368 {
1369 IscsiLun *iscsilun = bs->opaque;
1370
1371 iscsilun->aio_context = new_context;
1372 iscsi_set_events(iscsilun);
1373
1374 /* Set up a timer for sending out iSCSI NOPs */
1375 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1376 QEMU_CLOCK_REALTIME, SCALE_MS,
1377 iscsi_nop_timed_event, iscsilun);
1378 timer_mod(iscsilun->nop_timer,
1379 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1380
1381 /* Set up a timer for periodic calls to iscsi_set_events and to
1382 * scan for command timeout */
1383 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1384 QEMU_CLOCK_REALTIME, SCALE_MS,
1385 iscsi_timed_check_events, iscsilun);
1386 timer_mod(iscsilun->event_timer,
1387 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1388 }
1389
1390 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1391 {
1392 struct scsi_task *task;
1393 struct scsi_mode_sense *ms = NULL;
1394 iscsilun->write_protected = false;
1395 iscsilun->dpofua = false;
1396
1397 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1398 1, SCSI_MODESENSE_PC_CURRENT,
1399 0x3F, 0, 255);
1400 if (task == NULL) {
1401 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1402 iscsi_get_error(iscsilun->iscsi));
1403 goto out;
1404 }
1405
1406 if (task->status != SCSI_STATUS_GOOD) {
1407 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1408 goto out;
1409 }
1410 ms = scsi_datain_unmarshall(task);
1411 if (!ms) {
1412 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1413 iscsi_get_error(iscsilun->iscsi));
1414 goto out;
1415 }
1416 iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1417 iscsilun->dpofua = ms->device_specific_parameter & 0x10;
1418
1419 out:
1420 if (task) {
1421 scsi_free_scsi_task(task);
1422 }
1423 }
1424
1425 /*
1426 * We support iscsi url's on the form
1427 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1428 */
1429 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1430 Error **errp)
1431 {
1432 IscsiLun *iscsilun = bs->opaque;
1433 struct iscsi_context *iscsi = NULL;
1434 struct iscsi_url *iscsi_url = NULL;
1435 struct scsi_task *task = NULL;
1436 struct scsi_inquiry_standard *inq = NULL;
1437 struct scsi_inquiry_supported_pages *inq_vpd;
1438 char *initiator_name = NULL;
1439 QemuOpts *opts;
1440 Error *local_err = NULL;
1441 const char *filename;
1442 int i, ret = 0, timeout = 0;
1443
1444 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1445 qemu_opts_absorb_qdict(opts, options, &local_err);
1446 if (local_err) {
1447 error_propagate(errp, local_err);
1448 ret = -EINVAL;
1449 goto out;
1450 }
1451
1452 filename = qemu_opt_get(opts, "filename");
1453
1454 iscsi_url = iscsi_parse_full_url(iscsi, filename);
1455 if (iscsi_url == NULL) {
1456 error_setg(errp, "Failed to parse URL : %s", filename);
1457 ret = -EINVAL;
1458 goto out;
1459 }
1460
1461 memset(iscsilun, 0, sizeof(IscsiLun));
1462
1463 initiator_name = parse_initiator_name(iscsi_url->target);
1464
1465 iscsi = iscsi_create_context(initiator_name);
1466 if (iscsi == NULL) {
1467 error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1468 ret = -ENOMEM;
1469 goto out;
1470 }
1471
1472 if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1473 error_setg(errp, "iSCSI: Failed to set target name.");
1474 ret = -EINVAL;
1475 goto out;
1476 }
1477
1478 if (iscsi_url->user[0] != '\0') {
1479 ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1480 iscsi_url->passwd);
1481 if (ret != 0) {
1482 error_setg(errp, "Failed to set initiator username and password");
1483 ret = -EINVAL;
1484 goto out;
1485 }
1486 }
1487
1488 /* check if we got CHAP username/password via the options */
1489 parse_chap(iscsi, iscsi_url->target, &local_err);
1490 if (local_err != NULL) {
1491 error_propagate(errp, local_err);
1492 ret = -EINVAL;
1493 goto out;
1494 }
1495
1496 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1497 error_setg(errp, "iSCSI: Failed to set session type to normal.");
1498 ret = -EINVAL;
1499 goto out;
1500 }
1501
1502 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1503
1504 /* check if we got HEADER_DIGEST via the options */
1505 parse_header_digest(iscsi, iscsi_url->target, &local_err);
1506 if (local_err != NULL) {
1507 error_propagate(errp, local_err);
1508 ret = -EINVAL;
1509 goto out;
1510 }
1511
1512 /* timeout handling is broken in libiscsi before 1.15.0 */
1513 timeout = parse_timeout(iscsi_url->target);
1514 #if defined(LIBISCSI_API_VERSION) && LIBISCSI_API_VERSION >= 20150621
1515 iscsi_set_timeout(iscsi, timeout);
1516 #else
1517 if (timeout) {
1518 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1519 }
1520 #endif
1521
1522 if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1523 error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1524 iscsi_get_error(iscsi));
1525 ret = -EINVAL;
1526 goto out;
1527 }
1528
1529 iscsilun->iscsi = iscsi;
1530 iscsilun->aio_context = bdrv_get_aio_context(bs);
1531 iscsilun->lun = iscsi_url->lun;
1532 iscsilun->has_write_same = true;
1533
1534 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1535 (void **) &inq, errp);
1536 if (task == NULL) {
1537 ret = -EINVAL;
1538 goto out;
1539 }
1540 iscsilun->type = inq->periperal_device_type;
1541 scsi_free_scsi_task(task);
1542 task = NULL;
1543
1544 iscsi_modesense_sync(iscsilun);
1545
1546 /* Check the write protect flag of the LUN if we want to write */
1547 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1548 iscsilun->write_protected) {
1549 error_setg(errp, "Cannot open a write protected LUN as read-write");
1550 ret = -EACCES;
1551 goto out;
1552 }
1553
1554 iscsi_readcapacity_sync(iscsilun, &local_err);
1555 if (local_err != NULL) {
1556 error_propagate(errp, local_err);
1557 ret = -EINVAL;
1558 goto out;
1559 }
1560 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1561 bs->request_alignment = iscsilun->block_size;
1562
1563 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1564 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1565 * will try to read from the device to guess the image format.
1566 */
1567 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1568 bs->sg = 1;
1569 }
1570
1571 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1572 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1573 (void **) &inq_vpd, errp);
1574 if (task == NULL) {
1575 ret = -EINVAL;
1576 goto out;
1577 }
1578 for (i = 0; i < inq_vpd->num_pages; i++) {
1579 struct scsi_task *inq_task;
1580 struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1581 struct scsi_inquiry_block_limits *inq_bl;
1582 switch (inq_vpd->pages[i]) {
1583 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1584 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1585 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1586 (void **) &inq_lbp, errp);
1587 if (inq_task == NULL) {
1588 ret = -EINVAL;
1589 goto out;
1590 }
1591 memcpy(&iscsilun->lbp, inq_lbp,
1592 sizeof(struct scsi_inquiry_logical_block_provisioning));
1593 scsi_free_scsi_task(inq_task);
1594 break;
1595 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1596 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1597 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1598 (void **) &inq_bl, errp);
1599 if (inq_task == NULL) {
1600 ret = -EINVAL;
1601 goto out;
1602 }
1603 memcpy(&iscsilun->bl, inq_bl,
1604 sizeof(struct scsi_inquiry_block_limits));
1605 scsi_free_scsi_task(inq_task);
1606 break;
1607 default:
1608 break;
1609 }
1610 }
1611 scsi_free_scsi_task(task);
1612 task = NULL;
1613
1614 iscsi_attach_aio_context(bs, iscsilun->aio_context);
1615
1616 /* Guess the internal cluster (page) size of the iscsi target by the means
1617 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1618 * reasonable size */
1619 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
1620 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1621 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran *
1622 iscsilun->block_size) >> BDRV_SECTOR_BITS;
1623 if (iscsilun->lbprz) {
1624 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun);
1625 if (iscsilun->allocationmap == NULL) {
1626 ret = -ENOMEM;
1627 }
1628 }
1629 }
1630
1631 out:
1632 qemu_opts_del(opts);
1633 g_free(initiator_name);
1634 if (iscsi_url != NULL) {
1635 iscsi_destroy_url(iscsi_url);
1636 }
1637 if (task != NULL) {
1638 scsi_free_scsi_task(task);
1639 }
1640
1641 if (ret) {
1642 if (iscsi != NULL) {
1643 if (iscsi_is_logged_in(iscsi)) {
1644 iscsi_logout_sync(iscsi);
1645 }
1646 iscsi_destroy_context(iscsi);
1647 }
1648 memset(iscsilun, 0, sizeof(IscsiLun));
1649 }
1650 return ret;
1651 }
1652
1653 static void iscsi_close(BlockDriverState *bs)
1654 {
1655 IscsiLun *iscsilun = bs->opaque;
1656 struct iscsi_context *iscsi = iscsilun->iscsi;
1657
1658 iscsi_detach_aio_context(bs);
1659 if (iscsi_is_logged_in(iscsi)) {
1660 iscsi_logout_sync(iscsi);
1661 }
1662 iscsi_destroy_context(iscsi);
1663 g_free(iscsilun->zeroblock);
1664 g_free(iscsilun->allocationmap);
1665 memset(iscsilun, 0, sizeof(IscsiLun));
1666 }
1667
1668 static int sector_limits_lun2qemu(int64_t sector, IscsiLun *iscsilun)
1669 {
1670 return MIN(sector_lun2qemu(sector, iscsilun), INT_MAX / 2 + 1);
1671 }
1672
1673 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
1674 {
1675 /* We don't actually refresh here, but just return data queried in
1676 * iscsi_open(): iscsi targets don't change their limits. */
1677
1678 IscsiLun *iscsilun = bs->opaque;
1679 uint32_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
1680
1681 if (iscsilun->bl.max_xfer_len) {
1682 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
1683 }
1684
1685 bs->bl.max_transfer_length = sector_limits_lun2qemu(max_xfer_len, iscsilun);
1686
1687 if (iscsilun->lbp.lbpu) {
1688 if (iscsilun->bl.max_unmap < 0xffffffff) {
1689 bs->bl.max_discard =
1690 sector_limits_lun2qemu(iscsilun->bl.max_unmap, iscsilun);
1691 }
1692 bs->bl.discard_alignment =
1693 sector_limits_lun2qemu(iscsilun->bl.opt_unmap_gran, iscsilun);
1694 }
1695
1696 if (iscsilun->bl.max_ws_len < 0xffffffff) {
1697 bs->bl.max_write_zeroes =
1698 sector_limits_lun2qemu(iscsilun->bl.max_ws_len, iscsilun);
1699 }
1700 if (iscsilun->lbp.lbpws) {
1701 bs->bl.write_zeroes_alignment =
1702 sector_limits_lun2qemu(iscsilun->bl.opt_unmap_gran, iscsilun);
1703 }
1704 bs->bl.opt_transfer_length =
1705 sector_limits_lun2qemu(iscsilun->bl.opt_xfer_len, iscsilun);
1706 }
1707
1708 /* Note that this will not re-establish a connection with an iSCSI target - it
1709 * is effectively a NOP. */
1710 static int iscsi_reopen_prepare(BDRVReopenState *state,
1711 BlockReopenQueue *queue, Error **errp)
1712 {
1713 IscsiLun *iscsilun = state->bs->opaque;
1714
1715 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
1716 error_setg(errp, "Cannot open a write protected LUN as read-write");
1717 return -EACCES;
1718 }
1719 return 0;
1720 }
1721
1722 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1723 {
1724 IscsiLun *iscsilun = bs->opaque;
1725 Error *local_err = NULL;
1726
1727 if (iscsilun->type != TYPE_DISK) {
1728 return -ENOTSUP;
1729 }
1730
1731 iscsi_readcapacity_sync(iscsilun, &local_err);
1732 if (local_err != NULL) {
1733 error_free(local_err);
1734 return -EIO;
1735 }
1736
1737 if (offset > iscsi_getlength(bs)) {
1738 return -EINVAL;
1739 }
1740
1741 if (iscsilun->allocationmap != NULL) {
1742 g_free(iscsilun->allocationmap);
1743 iscsilun->allocationmap = iscsi_allocationmap_init(iscsilun);
1744 }
1745
1746 return 0;
1747 }
1748
1749 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp)
1750 {
1751 int ret = 0;
1752 int64_t total_size = 0;
1753 BlockDriverState *bs;
1754 IscsiLun *iscsilun = NULL;
1755 QDict *bs_options;
1756
1757 bs = bdrv_new();
1758
1759 /* Read out options */
1760 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1761 BDRV_SECTOR_SIZE);
1762 bs->opaque = g_new0(struct IscsiLun, 1);
1763 iscsilun = bs->opaque;
1764
1765 bs_options = qdict_new();
1766 qdict_put(bs_options, "filename", qstring_from_str(filename));
1767 ret = iscsi_open(bs, bs_options, 0, NULL);
1768 QDECREF(bs_options);
1769
1770 if (ret != 0) {
1771 goto out;
1772 }
1773 iscsi_detach_aio_context(bs);
1774 if (iscsilun->type != TYPE_DISK) {
1775 ret = -ENODEV;
1776 goto out;
1777 }
1778 if (bs->total_sectors < total_size) {
1779 ret = -ENOSPC;
1780 goto out;
1781 }
1782
1783 ret = 0;
1784 out:
1785 if (iscsilun->iscsi != NULL) {
1786 iscsi_destroy_context(iscsilun->iscsi);
1787 }
1788 g_free(bs->opaque);
1789 bs->opaque = NULL;
1790 bdrv_unref(bs);
1791 return ret;
1792 }
1793
1794 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1795 {
1796 IscsiLun *iscsilun = bs->opaque;
1797 bdi->unallocated_blocks_are_zero = iscsilun->lbprz;
1798 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
1799 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE;
1800 return 0;
1801 }
1802
1803 static QemuOptsList iscsi_create_opts = {
1804 .name = "iscsi-create-opts",
1805 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
1806 .desc = {
1807 {
1808 .name = BLOCK_OPT_SIZE,
1809 .type = QEMU_OPT_SIZE,
1810 .help = "Virtual disk size"
1811 },
1812 { /* end of list */ }
1813 }
1814 };
1815
1816 static BlockDriver bdrv_iscsi = {
1817 .format_name = "iscsi",
1818 .protocol_name = "iscsi",
1819
1820 .instance_size = sizeof(IscsiLun),
1821 .bdrv_needs_filename = true,
1822 .bdrv_file_open = iscsi_open,
1823 .bdrv_close = iscsi_close,
1824 .bdrv_create = iscsi_create,
1825 .create_opts = &iscsi_create_opts,
1826 .bdrv_reopen_prepare = iscsi_reopen_prepare,
1827
1828 .bdrv_getlength = iscsi_getlength,
1829 .bdrv_get_info = iscsi_get_info,
1830 .bdrv_truncate = iscsi_truncate,
1831 .bdrv_refresh_limits = iscsi_refresh_limits,
1832
1833 .bdrv_co_get_block_status = iscsi_co_get_block_status,
1834 .bdrv_co_discard = iscsi_co_discard,
1835 .bdrv_co_write_zeroes = iscsi_co_write_zeroes,
1836 .bdrv_co_readv = iscsi_co_readv,
1837 .bdrv_co_writev = iscsi_co_writev,
1838 .bdrv_co_flush_to_disk = iscsi_co_flush,
1839
1840 #ifdef __linux__
1841 .bdrv_aio_ioctl = iscsi_aio_ioctl,
1842 #endif
1843
1844 .bdrv_detach_aio_context = iscsi_detach_aio_context,
1845 .bdrv_attach_aio_context = iscsi_attach_aio_context,
1846 };
1847
1848 static QemuOptsList qemu_iscsi_opts = {
1849 .name = "iscsi",
1850 .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1851 .desc = {
1852 {
1853 .name = "user",
1854 .type = QEMU_OPT_STRING,
1855 .help = "username for CHAP authentication to target",
1856 },{
1857 .name = "password",
1858 .type = QEMU_OPT_STRING,
1859 .help = "password for CHAP authentication to target",
1860 },{
1861 .name = "header-digest",
1862 .type = QEMU_OPT_STRING,
1863 .help = "HeaderDigest setting. "
1864 "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1865 },{
1866 .name = "initiator-name",
1867 .type = QEMU_OPT_STRING,
1868 .help = "Initiator iqn name to use when connecting",
1869 },{
1870 .name = "timeout",
1871 .type = QEMU_OPT_NUMBER,
1872 .help = "Request timeout in seconds (default 0 = no timeout)",
1873 },
1874 { /* end of list */ }
1875 },
1876 };
1877
1878 static void iscsi_block_init(void)
1879 {
1880 bdrv_register(&bdrv_iscsi);
1881 qemu_add_opts(&qemu_iscsi_opts);
1882 }
1883
1884 block_init(iscsi_block_init);
Ray Wang's QEMU Repository