]> git.proxmox.com Git - mirror_qemu.git/blob - block/iscsi.c
projects / mirror_qemu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
qcow2: Optimize the refcount-block overlap check
[mirror_qemu.git] / block / iscsi.c
1 /*
2 * QEMU Block driver for iSCSI images
3 *
4 * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
5 * Copyright (c) 2012-2016 Peter Lieven <pl@kamp.de>
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23 * THE SOFTWARE.
24 */
25
26 #include "qemu/osdep.h"
27
28 #include <poll.h>
29 #include <math.h>
30 #include <arpa/inet.h>
31 #include "qemu-common.h"
32 #include "qemu/config-file.h"
33 #include "qemu/error-report.h"
34 #include "qemu/bitops.h"
35 #include "qemu/bitmap.h"
36 #include "block/block_int.h"
37 #include "block/scsi.h"
38 #include "qemu/iov.h"
39 #include "qemu/uuid.h"
40 #include "qmp-commands.h"
41 #include "qapi/qmp/qstring.h"
42 #include "crypto/secret.h"
43
44 #include <iscsi/iscsi.h>
45 #include <iscsi/scsi-lowlevel.h>
46
47 #ifdef __linux__
48 #include <scsi/sg.h>
49 #endif
50
51 typedef struct IscsiLun {
52 struct iscsi_context *iscsi;
53 AioContext *aio_context;
54 int lun;
55 enum scsi_inquiry_peripheral_device_type type;
56 int block_size;
57 uint64_t num_blocks;
58 int events;
59 QEMUTimer *nop_timer;
60 QEMUTimer *event_timer;
61 struct scsi_inquiry_logical_block_provisioning lbp;
62 struct scsi_inquiry_block_limits bl;
63 unsigned char *zeroblock;
64 /* The allocmap tracks which clusters (pages) on the iSCSI target are
65 * allocated and which are not. In case a target returns zeros for
66 * unallocated pages (iscsilun->lprz) we can directly return zeros instead
67 * of reading zeros over the wire if a read request falls within an
68 * unallocated block. As there are 3 possible states we need 2 bitmaps to
69 * track. allocmap_valid keeps track if QEMU's information about a page is
70 * valid. allocmap tracks if a page is allocated or not. In case QEMU has no
71 * valid information about a page the corresponding allocmap entry should be
72 * switched to unallocated as well to force a new lookup of the allocation
73 * status as lookups are generally skipped if a page is suspect to be
74 * allocated. If a iSCSI target is opened with cache.direct = on the
75 * allocmap_valid does not exist turning all cached information invalid so
76 * that a fresh lookup is made for any page even if allocmap entry returns
77 * it's unallocated. */
78 unsigned long *allocmap;
79 unsigned long *allocmap_valid;
80 long allocmap_size;
81 int cluster_sectors;
82 bool use_16_for_rw;
83 bool write_protected;
84 bool lbpme;
85 bool lbprz;
86 bool dpofua;
87 bool has_write_same;
88 bool request_timed_out;
89 } IscsiLun;
90
91 typedef struct IscsiTask {
92 int status;
93 int complete;
94 int retries;
95 int do_retry;
96 struct scsi_task *task;
97 Coroutine *co;
98 IscsiLun *iscsilun;
99 QEMUTimer retry_timer;
100 int err_code;
101 } IscsiTask;
102
103 typedef struct IscsiAIOCB {
104 BlockAIOCB common;
105 QEMUIOVector *qiov;
106 QEMUBH *bh;
107 IscsiLun *iscsilun;
108 struct scsi_task *task;
109 uint8_t *buf;
110 int status;
111 int64_t sector_num;
112 int nb_sectors;
113 int ret;
114 #ifdef __linux__
115 sg_io_hdr_t *ioh;
116 #endif
117 } IscsiAIOCB;
118
119 /* libiscsi uses time_t so its enough to process events every second */
120 #define EVENT_INTERVAL 1000
121 #define NOP_INTERVAL 5000
122 #define MAX_NOP_FAILURES 3
123 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times)
124 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768};
125
126 /* this threshold is a trade-off knob to choose between
127 * the potential additional overhead of an extra GET_LBA_STATUS request
128 * vs. unnecessarily reading a lot of zero sectors over the wire.
129 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
130 * sectors we check the allocation status of the area covered by the
131 * request first if the allocationmap indicates that the area might be
132 * unallocated. */
133 #define ISCSI_CHECKALLOC_THRES 64
134
135 static void
136 iscsi_bh_cb(void *p)
137 {
138 IscsiAIOCB *acb = p;
139
140 qemu_bh_delete(acb->bh);
141
142 g_free(acb->buf);
143 acb->buf = NULL;
144
145 acb->common.cb(acb->common.opaque, acb->status);
146
147 if (acb->task != NULL) {
148 scsi_free_scsi_task(acb->task);
149 acb->task = NULL;
150 }
151
152 qemu_aio_unref(acb);
153 }
154
155 static void
156 iscsi_schedule_bh(IscsiAIOCB *acb)
157 {
158 if (acb->bh) {
159 return;
160 }
161 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb);
162 qemu_bh_schedule(acb->bh);
163 }
164
165 static void iscsi_co_generic_bh_cb(void *opaque)
166 {
167 struct IscsiTask *iTask = opaque;
168 iTask->complete = 1;
169 qemu_coroutine_enter(iTask->co);
170 }
171
172 static void iscsi_retry_timer_expired(void *opaque)
173 {
174 struct IscsiTask *iTask = opaque;
175 iTask->complete = 1;
176 if (iTask->co) {
177 qemu_coroutine_enter(iTask->co);
178 }
179 }
180
181 static inline unsigned exp_random(double mean)
182 {
183 return -mean * log((double)rand() / RAND_MAX);
184 }
185
186 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in
187 * libiscsi 1.10.0, together with other constants we need. Use it as
188 * a hint that we have to define them ourselves if needed, to keep the
189 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for
190 * the test because SCSI_STATUS_* is an enum.
191 *
192 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes
193 * an enum, check against the LIBISCSI_API_VERSION macro, which was
194 * introduced in 1.11.0. If it is present, there is no need to define
195 * anything.
196 */
197 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \
198 !defined(LIBISCSI_API_VERSION)
199 #define SCSI_STATUS_TASK_SET_FULL 0x28
200 #define SCSI_STATUS_TIMEOUT 0x0f000002
201 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600
202 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00
203 #endif
204
205 #ifndef LIBISCSI_API_VERSION
206 #define LIBISCSI_API_VERSION 20130701
207 #endif
208
209 static int iscsi_translate_sense(struct scsi_sense *sense)
210 {
211 int ret;
212
213 switch (sense->key) {
214 case SCSI_SENSE_NOT_READY:
215 return -EBUSY;
216 case SCSI_SENSE_DATA_PROTECTION:
217 return -EACCES;
218 case SCSI_SENSE_COMMAND_ABORTED:
219 return -ECANCELED;
220 case SCSI_SENSE_ILLEGAL_REQUEST:
221 /* Parse ASCQ */
222 break;
223 default:
224 return -EIO;
225 }
226 switch (sense->ascq) {
227 case SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR:
228 case SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE:
229 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB:
230 case SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST:
231 ret = -EINVAL;
232 break;
233 case SCSI_SENSE_ASCQ_LBA_OUT_OF_RANGE:
234 ret = -ENOSPC;
235 break;
236 case SCSI_SENSE_ASCQ_LOGICAL_UNIT_NOT_SUPPORTED:
237 ret = -ENOTSUP;
238 break;
239 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT:
240 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_CLOSED:
241 case SCSI_SENSE_ASCQ_MEDIUM_NOT_PRESENT_TRAY_OPEN:
242 ret = -ENOMEDIUM;
243 break;
244 case SCSI_SENSE_ASCQ_WRITE_PROTECTED:
245 ret = -EACCES;
246 break;
247 default:
248 ret = -EIO;
249 break;
250 }
251 return ret;
252 }
253
254 static void
255 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
256 void *command_data, void *opaque)
257 {
258 struct IscsiTask *iTask = opaque;
259 struct scsi_task *task = command_data;
260
261 iTask->status = status;
262 iTask->do_retry = 0;
263 iTask->task = task;
264
265 if (status != SCSI_STATUS_GOOD) {
266 if (iTask->retries++ < ISCSI_CMD_RETRIES) {
267 if (status == SCSI_STATUS_CHECK_CONDITION
268 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
269 error_report("iSCSI CheckCondition: %s",
270 iscsi_get_error(iscsi));
271 iTask->do_retry = 1;
272 goto out;
273 }
274 if (status == SCSI_STATUS_BUSY ||
275 status == SCSI_STATUS_TIMEOUT ||
276 status == SCSI_STATUS_TASK_SET_FULL) {
277 unsigned retry_time =
278 exp_random(iscsi_retry_times[iTask->retries - 1]);
279 if (status == SCSI_STATUS_TIMEOUT) {
280 /* make sure the request is rescheduled AFTER the
281 * reconnect is initiated */
282 retry_time = EVENT_INTERVAL * 2;
283 iTask->iscsilun->request_timed_out = true;
284 }
285 error_report("iSCSI Busy/TaskSetFull/TimeOut"
286 " (retry #%u in %u ms): %s",
287 iTask->retries, retry_time,
288 iscsi_get_error(iscsi));
289 aio_timer_init(iTask->iscsilun->aio_context,
290 &iTask->retry_timer, QEMU_CLOCK_REALTIME,
291 SCALE_MS, iscsi_retry_timer_expired, iTask);
292 timer_mod(&iTask->retry_timer,
293 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
294 iTask->do_retry = 1;
295 return;
296 }
297 }
298 iTask->err_code = iscsi_translate_sense(&task->sense);
299 error_report("iSCSI Failure: %s", iscsi_get_error(iscsi));
300 }
301
302 out:
303 if (iTask->co) {
304 aio_bh_schedule_oneshot(iTask->iscsilun->aio_context,
305 iscsi_co_generic_bh_cb, iTask);
306 } else {
307 iTask->complete = 1;
308 }
309 }
310
311 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
312 {
313 *iTask = (struct IscsiTask) {
314 .co = qemu_coroutine_self(),
315 .iscsilun = iscsilun,
316 };
317 }
318
319 static void
320 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
321 void *private_data)
322 {
323 IscsiAIOCB *acb = private_data;
324
325 acb->status = -ECANCELED;
326 iscsi_schedule_bh(acb);
327 }
328
329 static void
330 iscsi_aio_cancel(BlockAIOCB *blockacb)
331 {
332 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
333 IscsiLun *iscsilun = acb->iscsilun;
334
335 if (acb->status != -EINPROGRESS) {
336 return;
337 }
338
339 /* send a task mgmt call to the target to cancel the task on the target */
340 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
341 iscsi_abort_task_cb, acb);
342
343 }
344
345 static const AIOCBInfo iscsi_aiocb_info = {
346 .aiocb_size = sizeof(IscsiAIOCB),
347 .cancel_async = iscsi_aio_cancel,
348 };
349
350
351 static void iscsi_process_read(void *arg);
352 static void iscsi_process_write(void *arg);
353
354 static void
355 iscsi_set_events(IscsiLun *iscsilun)
356 {
357 struct iscsi_context *iscsi = iscsilun->iscsi;
358 int ev = iscsi_which_events(iscsi);
359
360 if (ev != iscsilun->events) {
361 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi),
362 false,
363 (ev & POLLIN) ? iscsi_process_read : NULL,
364 (ev & POLLOUT) ? iscsi_process_write : NULL,
365 NULL,
366 iscsilun);
367 iscsilun->events = ev;
368 }
369 }
370
371 static void iscsi_timed_check_events(void *opaque)
372 {
373 IscsiLun *iscsilun = opaque;
374
375 /* check for timed out requests */
376 iscsi_service(iscsilun->iscsi, 0);
377
378 if (iscsilun->request_timed_out) {
379 iscsilun->request_timed_out = false;
380 iscsi_reconnect(iscsilun->iscsi);
381 }
382
383 /* newer versions of libiscsi may return zero events. Ensure we are able
384 * to return to service once this situation changes. */
385 iscsi_set_events(iscsilun);
386
387 timer_mod(iscsilun->event_timer,
388 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
389 }
390
391 static void
392 iscsi_process_read(void *arg)
393 {
394 IscsiLun *iscsilun = arg;
395 struct iscsi_context *iscsi = iscsilun->iscsi;
396
397 iscsi_service(iscsi, POLLIN);
398 iscsi_set_events(iscsilun);
399 }
400
401 static void
402 iscsi_process_write(void *arg)
403 {
404 IscsiLun *iscsilun = arg;
405 struct iscsi_context *iscsi = iscsilun->iscsi;
406
407 iscsi_service(iscsi, POLLOUT);
408 iscsi_set_events(iscsilun);
409 }
410
411 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
412 {
413 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
414 }
415
416 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
417 {
418 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
419 }
420
421 static bool is_byte_request_lun_aligned(int64_t offset, int count,
422 IscsiLun *iscsilun)
423 {
424 if (offset % iscsilun->block_size || count % iscsilun->block_size) {
425 error_report("iSCSI misaligned request: "
426 "iscsilun->block_size %u, offset %" PRIi64
427 ", count %d",
428 iscsilun->block_size, offset, count);
429 return false;
430 }
431 return true;
432 }
433
434 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors,
435 IscsiLun *iscsilun)
436 {
437 assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS);
438 return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS,
439 nb_sectors << BDRV_SECTOR_BITS,
440 iscsilun);
441 }
442
443 static void iscsi_allocmap_free(IscsiLun *iscsilun)
444 {
445 g_free(iscsilun->allocmap);
446 g_free(iscsilun->allocmap_valid);
447 iscsilun->allocmap = NULL;
448 iscsilun->allocmap_valid = NULL;
449 }
450
451
452 static int iscsi_allocmap_init(IscsiLun *iscsilun, int open_flags)
453 {
454 iscsi_allocmap_free(iscsilun);
455
456 iscsilun->allocmap_size =
457 DIV_ROUND_UP(sector_lun2qemu(iscsilun->num_blocks, iscsilun),
458 iscsilun->cluster_sectors);
459
460 iscsilun->allocmap = bitmap_try_new(iscsilun->allocmap_size);
461 if (!iscsilun->allocmap) {
462 return -ENOMEM;
463 }
464
465 if (open_flags & BDRV_O_NOCACHE) {
466 /* in case that cache.direct = on all allocmap entries are
467 * treated as invalid to force a relookup of the block
468 * status on every read request */
469 return 0;
470 }
471
472 iscsilun->allocmap_valid = bitmap_try_new(iscsilun->allocmap_size);
473 if (!iscsilun->allocmap_valid) {
474 /* if we are under memory pressure free the allocmap as well */
475 iscsi_allocmap_free(iscsilun);
476 return -ENOMEM;
477 }
478
479 return 0;
480 }
481
482 static void
483 iscsi_allocmap_update(IscsiLun *iscsilun, int64_t sector_num,
484 int nb_sectors, bool allocated, bool valid)
485 {
486 int64_t cl_num_expanded, nb_cls_expanded, cl_num_shrunk, nb_cls_shrunk;
487
488 if (iscsilun->allocmap == NULL) {
489 return;
490 }
491 /* expand to entirely contain all affected clusters */
492 cl_num_expanded = sector_num / iscsilun->cluster_sectors;
493 nb_cls_expanded = DIV_ROUND_UP(sector_num + nb_sectors,
494 iscsilun->cluster_sectors) - cl_num_expanded;
495 /* shrink to touch only completely contained clusters */
496 cl_num_shrunk = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors);
497 nb_cls_shrunk = (sector_num + nb_sectors) / iscsilun->cluster_sectors
498 - cl_num_shrunk;
499 if (allocated) {
500 bitmap_set(iscsilun->allocmap, cl_num_expanded, nb_cls_expanded);
501 } else {
502 if (nb_cls_shrunk > 0) {
503 bitmap_clear(iscsilun->allocmap, cl_num_shrunk, nb_cls_shrunk);
504 }
505 }
506
507 if (iscsilun->allocmap_valid == NULL) {
508 return;
509 }
510 if (valid) {
511 if (nb_cls_shrunk > 0) {
512 bitmap_set(iscsilun->allocmap_valid, cl_num_shrunk, nb_cls_shrunk);
513 }
514 } else {
515 bitmap_clear(iscsilun->allocmap_valid, cl_num_expanded,
516 nb_cls_expanded);
517 }
518 }
519
520 static void
521 iscsi_allocmap_set_allocated(IscsiLun *iscsilun, int64_t sector_num,
522 int nb_sectors)
523 {
524 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, true, true);
525 }
526
527 static void
528 iscsi_allocmap_set_unallocated(IscsiLun *iscsilun, int64_t sector_num,
529 int nb_sectors)
530 {
531 /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update
532 * is ignored, so this will in effect be an iscsi_allocmap_set_invalid.
533 */
534 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, false, true);
535 }
536
537 static void iscsi_allocmap_set_invalid(IscsiLun *iscsilun, int64_t sector_num,
538 int nb_sectors)
539 {
540 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, false, false);
541 }
542
543 static void iscsi_allocmap_invalidate(IscsiLun *iscsilun)
544 {
545 if (iscsilun->allocmap) {
546 bitmap_zero(iscsilun->allocmap, iscsilun->allocmap_size);
547 }
548 if (iscsilun->allocmap_valid) {
549 bitmap_zero(iscsilun->allocmap_valid, iscsilun->allocmap_size);
550 }
551 }
552
553 static inline bool
554 iscsi_allocmap_is_allocated(IscsiLun *iscsilun, int64_t sector_num,
555 int nb_sectors)
556 {
557 unsigned long size;
558 if (iscsilun->allocmap == NULL) {
559 return true;
560 }
561 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
562 return !(find_next_bit(iscsilun->allocmap, size,
563 sector_num / iscsilun->cluster_sectors) == size);
564 }
565
566 static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun,
567 int64_t sector_num, int nb_sectors)
568 {
569 unsigned long size;
570 if (iscsilun->allocmap_valid == NULL) {
571 return false;
572 }
573 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
574 return (find_next_zero_bit(iscsilun->allocmap_valid, size,
575 sector_num / iscsilun->cluster_sectors) == size);
576 }
577
578 static int coroutine_fn
579 iscsi_co_writev_flags(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
580 QEMUIOVector *iov, int flags)
581 {
582 IscsiLun *iscsilun = bs->opaque;
583 struct IscsiTask iTask;
584 uint64_t lba;
585 uint32_t num_sectors;
586 bool fua = flags & BDRV_REQ_FUA;
587
588 if (fua) {
589 assert(iscsilun->dpofua);
590 }
591 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
592 return -EINVAL;
593 }
594
595 if (bs->bl.max_transfer) {
596 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
597 }
598
599 lba = sector_qemu2lun(sector_num, iscsilun);
600 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
601 iscsi_co_init_iscsitask(iscsilun, &iTask);
602 retry:
603 if (iscsilun->use_16_for_rw) {
604 #if LIBISCSI_API_VERSION >= (20160603)
605 iTask.task = iscsi_write16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
606 NULL, num_sectors * iscsilun->block_size,
607 iscsilun->block_size, 0, 0, fua, 0, 0,
608 iscsi_co_generic_cb, &iTask,
609 (struct scsi_iovec *)iov->iov, iov->niov);
610 } else {
611 iTask.task = iscsi_write10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
612 NULL, num_sectors * iscsilun->block_size,
613 iscsilun->block_size, 0, 0, fua, 0, 0,
614 iscsi_co_generic_cb, &iTask,
615 (struct scsi_iovec *)iov->iov, iov->niov);
616 }
617 #else
618 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
619 NULL, num_sectors * iscsilun->block_size,
620 iscsilun->block_size, 0, 0, fua, 0, 0,
621 iscsi_co_generic_cb, &iTask);
622 } else {
623 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
624 NULL, num_sectors * iscsilun->block_size,
625 iscsilun->block_size, 0, 0, fua, 0, 0,
626 iscsi_co_generic_cb, &iTask);
627 }
628 #endif
629 if (iTask.task == NULL) {
630 return -ENOMEM;
631 }
632 #if LIBISCSI_API_VERSION < (20160603)
633 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
634 iov->niov);
635 #endif
636 while (!iTask.complete) {
637 iscsi_set_events(iscsilun);
638 qemu_coroutine_yield();
639 }
640
641 if (iTask.task != NULL) {
642 scsi_free_scsi_task(iTask.task);
643 iTask.task = NULL;
644 }
645
646 if (iTask.do_retry) {
647 iTask.complete = 0;
648 goto retry;
649 }
650
651 if (iTask.status != SCSI_STATUS_GOOD) {
652 iscsi_allocmap_set_invalid(iscsilun, sector_num, nb_sectors);
653 return iTask.err_code;
654 }
655
656 iscsi_allocmap_set_allocated(iscsilun, sector_num, nb_sectors);
657
658 return 0;
659 }
660
661
662
663 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
664 int64_t sector_num,
665 int nb_sectors, int *pnum,
666 BlockDriverState **file)
667 {
668 IscsiLun *iscsilun = bs->opaque;
669 struct scsi_get_lba_status *lbas = NULL;
670 struct scsi_lba_status_descriptor *lbasd = NULL;
671 struct IscsiTask iTask;
672 int64_t ret;
673
674 iscsi_co_init_iscsitask(iscsilun, &iTask);
675
676 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
677 ret = -EINVAL;
678 goto out;
679 }
680
681 /* default to all sectors allocated */
682 ret = BDRV_BLOCK_DATA;
683 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
684 *pnum = nb_sectors;
685
686 /* LUN does not support logical block provisioning */
687 if (!iscsilun->lbpme) {
688 goto out;
689 }
690
691 retry:
692 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
693 sector_qemu2lun(sector_num, iscsilun),
694 8 + 16, iscsi_co_generic_cb,
695 &iTask) == NULL) {
696 ret = -ENOMEM;
697 goto out;
698 }
699
700 while (!iTask.complete) {
701 iscsi_set_events(iscsilun);
702 qemu_coroutine_yield();
703 }
704
705 if (iTask.do_retry) {
706 if (iTask.task != NULL) {
707 scsi_free_scsi_task(iTask.task);
708 iTask.task = NULL;
709 }
710 iTask.complete = 0;
711 goto retry;
712 }
713
714 if (iTask.status != SCSI_STATUS_GOOD) {
715 /* in case the get_lba_status_callout fails (i.e.
716 * because the device is busy or the cmd is not
717 * supported) we pretend all blocks are allocated
718 * for backwards compatibility */
719 goto out;
720 }
721
722 lbas = scsi_datain_unmarshall(iTask.task);
723 if (lbas == NULL) {
724 ret = -EIO;
725 goto out;
726 }
727
728 lbasd = &lbas->descriptors[0];
729
730 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
731 ret = -EIO;
732 goto out;
733 }
734
735 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
736
737 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
738 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
739 ret &= ~BDRV_BLOCK_DATA;
740 if (iscsilun->lbprz) {
741 ret |= BDRV_BLOCK_ZERO;
742 }
743 }
744
745 if (ret & BDRV_BLOCK_ZERO) {
746 iscsi_allocmap_set_unallocated(iscsilun, sector_num, *pnum);
747 } else {
748 iscsi_allocmap_set_allocated(iscsilun, sector_num, *pnum);
749 }
750
751 if (*pnum > nb_sectors) {
752 *pnum = nb_sectors;
753 }
754 out:
755 if (iTask.task != NULL) {
756 scsi_free_scsi_task(iTask.task);
757 }
758 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) {
759 *file = bs;
760 }
761 return ret;
762 }
763
764 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
765 int64_t sector_num, int nb_sectors,
766 QEMUIOVector *iov)
767 {
768 IscsiLun *iscsilun = bs->opaque;
769 struct IscsiTask iTask;
770 uint64_t lba;
771 uint32_t num_sectors;
772
773 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
774 return -EINVAL;
775 }
776
777 if (bs->bl.max_transfer) {
778 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
779 }
780
781 /* if cache.direct is off and we have a valid entry in our allocation map
782 * we can skip checking the block status and directly return zeroes if
783 * the request falls within an unallocated area */
784 if (iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) &&
785 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
786 qemu_iovec_memset(iov, 0, 0x00, iov->size);
787 return 0;
788 }
789
790 if (nb_sectors >= ISCSI_CHECKALLOC_THRES &&
791 !iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) &&
792 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
793 int pnum;
794 BlockDriverState *file;
795 /* check the block status from the beginning of the cluster
796 * containing the start sector */
797 int64_t ret = iscsi_co_get_block_status(bs,
798 sector_num - sector_num % iscsilun->cluster_sectors,
799 BDRV_REQUEST_MAX_SECTORS, &pnum, &file);
800 if (ret < 0) {
801 return ret;
802 }
803 /* if the whole request falls into an unallocated area we can avoid
804 * to read and directly return zeroes instead */
805 if (ret & BDRV_BLOCK_ZERO &&
806 pnum >= nb_sectors + sector_num % iscsilun->cluster_sectors) {
807 qemu_iovec_memset(iov, 0, 0x00, iov->size);
808 return 0;
809 }
810 }
811
812 lba = sector_qemu2lun(sector_num, iscsilun);
813 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
814
815 iscsi_co_init_iscsitask(iscsilun, &iTask);
816 retry:
817 if (iscsilun->use_16_for_rw) {
818 #if LIBISCSI_API_VERSION >= (20160603)
819 iTask.task = iscsi_read16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
820 num_sectors * iscsilun->block_size,
821 iscsilun->block_size, 0, 0, 0, 0, 0,
822 iscsi_co_generic_cb, &iTask,
823 (struct scsi_iovec *)iov->iov, iov->niov);
824 } else {
825 iTask.task = iscsi_read10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
826 num_sectors * iscsilun->block_size,
827 iscsilun->block_size,
828 0, 0, 0, 0, 0,
829 iscsi_co_generic_cb, &iTask,
830 (struct scsi_iovec *)iov->iov, iov->niov);
831 }
832 #else
833 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
834 num_sectors * iscsilun->block_size,
835 iscsilun->block_size, 0, 0, 0, 0, 0,
836 iscsi_co_generic_cb, &iTask);
837 } else {
838 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
839 num_sectors * iscsilun->block_size,
840 iscsilun->block_size,
841 0, 0, 0, 0, 0,
842 iscsi_co_generic_cb, &iTask);
843 }
844 #endif
845 if (iTask.task == NULL) {
846 return -ENOMEM;
847 }
848 #if LIBISCSI_API_VERSION < (20160603)
849 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
850 #endif
851 while (!iTask.complete) {
852 iscsi_set_events(iscsilun);
853 qemu_coroutine_yield();
854 }
855
856 if (iTask.task != NULL) {
857 scsi_free_scsi_task(iTask.task);
858 iTask.task = NULL;
859 }
860
861 if (iTask.do_retry) {
862 iTask.complete = 0;
863 goto retry;
864 }
865
866 if (iTask.status != SCSI_STATUS_GOOD) {
867 return iTask.err_code;
868 }
869
870 return 0;
871 }
872
873 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
874 {
875 IscsiLun *iscsilun = bs->opaque;
876 struct IscsiTask iTask;
877
878 iscsi_co_init_iscsitask(iscsilun, &iTask);
879 retry:
880 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
881 0, iscsi_co_generic_cb, &iTask) == NULL) {
882 return -ENOMEM;
883 }
884
885 while (!iTask.complete) {
886 iscsi_set_events(iscsilun);
887 qemu_coroutine_yield();
888 }
889
890 if (iTask.task != NULL) {
891 scsi_free_scsi_task(iTask.task);
892 iTask.task = NULL;
893 }
894
895 if (iTask.do_retry) {
896 iTask.complete = 0;
897 goto retry;
898 }
899
900 if (iTask.status != SCSI_STATUS_GOOD) {
901 return iTask.err_code;
902 }
903
904 return 0;
905 }
906
907 #ifdef __linux__
908 static void
909 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
910 void *command_data, void *opaque)
911 {
912 IscsiAIOCB *acb = opaque;
913
914 g_free(acb->buf);
915 acb->buf = NULL;
916
917 acb->status = 0;
918 if (status < 0) {
919 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
920 iscsi_get_error(iscsi));
921 acb->status = iscsi_translate_sense(&acb->task->sense);
922 }
923
924 acb->ioh->driver_status = 0;
925 acb->ioh->host_status = 0;
926 acb->ioh->resid = 0;
927 acb->ioh->status = status;
928
929 #define SG_ERR_DRIVER_SENSE 0x08
930
931 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
932 int ss;
933
934 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
935
936 acb->ioh->sb_len_wr = acb->task->datain.size - 2;
937 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
938 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
939 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
940 }
941
942 iscsi_schedule_bh(acb);
943 }
944
945 static void iscsi_ioctl_bh_completion(void *opaque)
946 {
947 IscsiAIOCB *acb = opaque;
948
949 qemu_bh_delete(acb->bh);
950 acb->common.cb(acb->common.opaque, acb->ret);
951 qemu_aio_unref(acb);
952 }
953
954 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
955 {
956 BlockDriverState *bs = acb->common.bs;
957 IscsiLun *iscsilun = bs->opaque;
958 int ret = 0;
959
960 switch (req) {
961 case SG_GET_VERSION_NUM:
962 *(int *)buf = 30000;
963 break;
964 case SG_GET_SCSI_ID:
965 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
966 break;
967 default:
968 ret = -EINVAL;
969 }
970 assert(!acb->bh);
971 acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
972 iscsi_ioctl_bh_completion, acb);
973 acb->ret = ret;
974 qemu_bh_schedule(acb->bh);
975 }
976
977 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
978 unsigned long int req, void *buf,
979 BlockCompletionFunc *cb, void *opaque)
980 {
981 IscsiLun *iscsilun = bs->opaque;
982 struct iscsi_context *iscsi = iscsilun->iscsi;
983 struct iscsi_data data;
984 IscsiAIOCB *acb;
985
986 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
987
988 acb->iscsilun = iscsilun;
989 acb->bh = NULL;
990 acb->status = -EINPROGRESS;
991 acb->buf = NULL;
992 acb->ioh = buf;
993
994 if (req != SG_IO) {
995 iscsi_ioctl_handle_emulated(acb, req, buf);
996 return &acb->common;
997 }
998
999 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) {
1000 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)",
1001 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE);
1002 qemu_aio_unref(acb);
1003 return NULL;
1004 }
1005
1006 acb->task = malloc(sizeof(struct scsi_task));
1007 if (acb->task == NULL) {
1008 error_report("iSCSI: Failed to allocate task for scsi command. %s",
1009 iscsi_get_error(iscsi));
1010 qemu_aio_unref(acb);
1011 return NULL;
1012 }
1013 memset(acb->task, 0, sizeof(struct scsi_task));
1014
1015 switch (acb->ioh->dxfer_direction) {
1016 case SG_DXFER_TO_DEV:
1017 acb->task->xfer_dir = SCSI_XFER_WRITE;
1018 break;
1019 case SG_DXFER_FROM_DEV:
1020 acb->task->xfer_dir = SCSI_XFER_READ;
1021 break;
1022 default:
1023 acb->task->xfer_dir = SCSI_XFER_NONE;
1024 break;
1025 }
1026
1027 acb->task->cdb_size = acb->ioh->cmd_len;
1028 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
1029 acb->task->expxferlen = acb->ioh->dxfer_len;
1030
1031 data.size = 0;
1032 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
1033 if (acb->ioh->iovec_count == 0) {
1034 data.data = acb->ioh->dxferp;
1035 data.size = acb->ioh->dxfer_len;
1036 } else {
1037 scsi_task_set_iov_out(acb->task,
1038 (struct scsi_iovec *) acb->ioh->dxferp,
1039 acb->ioh->iovec_count);
1040 }
1041 }
1042
1043 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
1044 iscsi_aio_ioctl_cb,
1045 (data.size > 0) ? &data : NULL,
1046 acb) != 0) {
1047 scsi_free_scsi_task(acb->task);
1048 qemu_aio_unref(acb);
1049 return NULL;
1050 }
1051
1052 /* tell libiscsi to read straight into the buffer we got from ioctl */
1053 if (acb->task->xfer_dir == SCSI_XFER_READ) {
1054 if (acb->ioh->iovec_count == 0) {
1055 scsi_task_add_data_in_buffer(acb->task,
1056 acb->ioh->dxfer_len,
1057 acb->ioh->dxferp);
1058 } else {
1059 scsi_task_set_iov_in(acb->task,
1060 (struct scsi_iovec *) acb->ioh->dxferp,
1061 acb->ioh->iovec_count);
1062 }
1063 }
1064
1065 iscsi_set_events(iscsilun);
1066
1067 return &acb->common;
1068 }
1069
1070 #endif
1071
1072 static int64_t
1073 iscsi_getlength(BlockDriverState *bs)
1074 {
1075 IscsiLun *iscsilun = bs->opaque;
1076 int64_t len;
1077
1078 len = iscsilun->num_blocks;
1079 len *= iscsilun->block_size;
1080
1081 return len;
1082 }
1083
1084 static int
1085 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, int count)
1086 {
1087 IscsiLun *iscsilun = bs->opaque;
1088 struct IscsiTask iTask;
1089 struct unmap_list list;
1090
1091 if (!is_byte_request_lun_aligned(offset, count, iscsilun)) {
1092 return -ENOTSUP;
1093 }
1094
1095 if (!iscsilun->lbp.lbpu) {
1096 /* UNMAP is not supported by the target */
1097 return 0;
1098 }
1099
1100 list.lba = offset / iscsilun->block_size;
1101 list.num = count / iscsilun->block_size;
1102
1103 iscsi_co_init_iscsitask(iscsilun, &iTask);
1104 retry:
1105 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
1106 iscsi_co_generic_cb, &iTask) == NULL) {
1107 return -ENOMEM;
1108 }
1109
1110 while (!iTask.complete) {
1111 iscsi_set_events(iscsilun);
1112 qemu_coroutine_yield();
1113 }
1114
1115 if (iTask.task != NULL) {
1116 scsi_free_scsi_task(iTask.task);
1117 iTask.task = NULL;
1118 }
1119
1120 if (iTask.do_retry) {
1121 iTask.complete = 0;
1122 goto retry;
1123 }
1124
1125 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
1126 /* the target might fail with a check condition if it
1127 is not happy with the alignment of the UNMAP request
1128 we silently fail in this case */
1129 return 0;
1130 }
1131
1132 if (iTask.status != SCSI_STATUS_GOOD) {
1133 return iTask.err_code;
1134 }
1135
1136 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1137 count >> BDRV_SECTOR_BITS);
1138
1139 return 0;
1140 }
1141
1142 static int
1143 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1144 int count, BdrvRequestFlags flags)
1145 {
1146 IscsiLun *iscsilun = bs->opaque;
1147 struct IscsiTask iTask;
1148 uint64_t lba;
1149 uint32_t nb_blocks;
1150 bool use_16_for_ws = iscsilun->use_16_for_rw;
1151
1152 if (!is_byte_request_lun_aligned(offset, count, iscsilun)) {
1153 return -ENOTSUP;
1154 }
1155
1156 if (flags & BDRV_REQ_MAY_UNMAP) {
1157 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
1158 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
1159 use_16_for_ws = true;
1160 }
1161 if (use_16_for_ws && !iscsilun->lbp.lbpws) {
1162 /* WRITESAME16 with UNMAP is not supported by the target,
1163 * fall back and try WRITESAME10/16 without UNMAP */
1164 flags &= ~BDRV_REQ_MAY_UNMAP;
1165 use_16_for_ws = iscsilun->use_16_for_rw;
1166 }
1167 }
1168
1169 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1170 /* WRITESAME without UNMAP is not supported by the target */
1171 return -ENOTSUP;
1172 }
1173
1174 lba = offset / iscsilun->block_size;
1175 nb_blocks = count / iscsilun->block_size;
1176
1177 if (iscsilun->zeroblock == NULL) {
1178 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1179 if (iscsilun->zeroblock == NULL) {
1180 return -ENOMEM;
1181 }
1182 }
1183
1184 iscsi_co_init_iscsitask(iscsilun, &iTask);
1185 retry:
1186 if (use_16_for_ws) {
1187 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1188 iscsilun->zeroblock, iscsilun->block_size,
1189 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1190 0, 0, iscsi_co_generic_cb, &iTask);
1191 } else {
1192 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1193 iscsilun->zeroblock, iscsilun->block_size,
1194 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1195 0, 0, iscsi_co_generic_cb, &iTask);
1196 }
1197 if (iTask.task == NULL) {
1198 return -ENOMEM;
1199 }
1200
1201 while (!iTask.complete) {
1202 iscsi_set_events(iscsilun);
1203 qemu_coroutine_yield();
1204 }
1205
1206 if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1207 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1208 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1209 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1210 /* WRITE SAME is not supported by the target */
1211 iscsilun->has_write_same = false;
1212 scsi_free_scsi_task(iTask.task);
1213 return -ENOTSUP;
1214 }
1215
1216 if (iTask.task != NULL) {
1217 scsi_free_scsi_task(iTask.task);
1218 iTask.task = NULL;
1219 }
1220
1221 if (iTask.do_retry) {
1222 iTask.complete = 0;
1223 goto retry;
1224 }
1225
1226 if (iTask.status != SCSI_STATUS_GOOD) {
1227 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1228 count >> BDRV_SECTOR_BITS);
1229 return iTask.err_code;
1230 }
1231
1232 if (flags & BDRV_REQ_MAY_UNMAP) {
1233 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1234 count >> BDRV_SECTOR_BITS);
1235 } else {
1236 iscsi_allocmap_set_allocated(iscsilun, offset >> BDRV_SECTOR_BITS,
1237 count >> BDRV_SECTOR_BITS);
1238 }
1239
1240 return 0;
1241 }
1242
1243 static void parse_chap(struct iscsi_context *iscsi, const char *target,
1244 Error **errp)
1245 {
1246 QemuOptsList *list;
1247 QemuOpts *opts;
1248 const char *user = NULL;
1249 const char *password = NULL;
1250 const char *secretid;
1251 char *secret = NULL;
1252
1253 list = qemu_find_opts("iscsi");
1254 if (!list) {
1255 return;
1256 }
1257
1258 opts = qemu_opts_find(list, target);
1259 if (opts == NULL) {
1260 opts = QTAILQ_FIRST(&list->head);
1261 if (!opts) {
1262 return;
1263 }
1264 }
1265
1266 user = qemu_opt_get(opts, "user");
1267 if (!user) {
1268 return;
1269 }
1270
1271 secretid = qemu_opt_get(opts, "password-secret");
1272 password = qemu_opt_get(opts, "password");
1273 if (secretid && password) {
1274 error_setg(errp, "'password' and 'password-secret' properties are "
1275 "mutually exclusive");
1276 return;
1277 }
1278 if (secretid) {
1279 secret = qcrypto_secret_lookup_as_utf8(secretid, errp);
1280 if (!secret) {
1281 return;
1282 }
1283 password = secret;
1284 } else if (!password) {
1285 error_setg(errp, "CHAP username specified but no password was given");
1286 return;
1287 }
1288
1289 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1290 error_setg(errp, "Failed to set initiator username and password");
1291 }
1292
1293 g_free(secret);
1294 }
1295
1296 static void parse_header_digest(struct iscsi_context *iscsi, const char *target,
1297 Error **errp)
1298 {
1299 QemuOptsList *list;
1300 QemuOpts *opts;
1301 const char *digest = NULL;
1302
1303 list = qemu_find_opts("iscsi");
1304 if (!list) {
1305 return;
1306 }
1307
1308 opts = qemu_opts_find(list, target);
1309 if (opts == NULL) {
1310 opts = QTAILQ_FIRST(&list->head);
1311 if (!opts) {
1312 return;
1313 }
1314 }
1315
1316 digest = qemu_opt_get(opts, "header-digest");
1317 if (!digest) {
1318 return;
1319 }
1320
1321 if (!strcmp(digest, "CRC32C")) {
1322 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1323 } else if (!strcmp(digest, "NONE")) {
1324 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1325 } else if (!strcmp(digest, "CRC32C-NONE")) {
1326 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1327 } else if (!strcmp(digest, "NONE-CRC32C")) {
1328 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1329 } else {
1330 error_setg(errp, "Invalid header-digest setting : %s", digest);
1331 }
1332 }
1333
1334 static char *parse_initiator_name(const char *target)
1335 {
1336 QemuOptsList *list;
1337 QemuOpts *opts;
1338 const char *name;
1339 char *iscsi_name;
1340 UuidInfo *uuid_info;
1341
1342 list = qemu_find_opts("iscsi");
1343 if (list) {
1344 opts = qemu_opts_find(list, target);
1345 if (!opts) {
1346 opts = QTAILQ_FIRST(&list->head);
1347 }
1348 if (opts) {
1349 name = qemu_opt_get(opts, "initiator-name");
1350 if (name) {
1351 return g_strdup(name);
1352 }
1353 }
1354 }
1355
1356 uuid_info = qmp_query_uuid(NULL);
1357 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1358 name = qemu_get_vm_name();
1359 } else {
1360 name = uuid_info->UUID;
1361 }
1362 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1363 name ? ":" : "", name ? name : "");
1364 qapi_free_UuidInfo(uuid_info);
1365 return iscsi_name;
1366 }
1367
1368 static int parse_timeout(const char *target)
1369 {
1370 QemuOptsList *list;
1371 QemuOpts *opts;
1372 const char *timeout;
1373
1374 list = qemu_find_opts("iscsi");
1375 if (list) {
1376 opts = qemu_opts_find(list, target);
1377 if (!opts) {
1378 opts = QTAILQ_FIRST(&list->head);
1379 }
1380 if (opts) {
1381 timeout = qemu_opt_get(opts, "timeout");
1382 if (timeout) {
1383 return atoi(timeout);
1384 }
1385 }
1386 }
1387
1388 return 0;
1389 }
1390
1391 static void iscsi_nop_timed_event(void *opaque)
1392 {
1393 IscsiLun *iscsilun = opaque;
1394
1395 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1396 error_report("iSCSI: NOP timeout. Reconnecting...");
1397 iscsilun->request_timed_out = true;
1398 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1399 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1400 return;
1401 }
1402
1403 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1404 iscsi_set_events(iscsilun);
1405 }
1406
1407 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1408 {
1409 struct scsi_task *task = NULL;
1410 struct scsi_readcapacity10 *rc10 = NULL;
1411 struct scsi_readcapacity16 *rc16 = NULL;
1412 int retries = ISCSI_CMD_RETRIES;
1413
1414 do {
1415 if (task != NULL) {
1416 scsi_free_scsi_task(task);
1417 task = NULL;
1418 }
1419
1420 switch (iscsilun->type) {
1421 case TYPE_DISK:
1422 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1423 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1424 rc16 = scsi_datain_unmarshall(task);
1425 if (rc16 == NULL) {
1426 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1427 } else {
1428 iscsilun->block_size = rc16->block_length;
1429 iscsilun->num_blocks = rc16->returned_lba + 1;
1430 iscsilun->lbpme = !!rc16->lbpme;
1431 iscsilun->lbprz = !!rc16->lbprz;
1432 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1433 }
1434 break;
1435 }
1436 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1437 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1438 break;
1439 }
1440 /* Fall through and try READ CAPACITY(10) instead. */
1441 case TYPE_ROM:
1442 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1443 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1444 rc10 = scsi_datain_unmarshall(task);
1445 if (rc10 == NULL) {
1446 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1447 } else {
1448 iscsilun->block_size = rc10->block_size;
1449 if (rc10->lba == 0) {
1450 /* blank disk loaded */
1451 iscsilun->num_blocks = 0;
1452 } else {
1453 iscsilun->num_blocks = rc10->lba + 1;
1454 }
1455 }
1456 }
1457 break;
1458 default:
1459 return;
1460 }
1461 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1462 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1463 && retries-- > 0);
1464
1465 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1466 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1467 } else if (!iscsilun->block_size ||
1468 iscsilun->block_size % BDRV_SECTOR_SIZE) {
1469 error_setg(errp, "iSCSI: the target returned an invalid "
1470 "block size of %d.", iscsilun->block_size);
1471 }
1472 if (task) {
1473 scsi_free_scsi_task(task);
1474 }
1475 }
1476
1477 /* TODO Convert to fine grained options */
1478 static QemuOptsList runtime_opts = {
1479 .name = "iscsi",
1480 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1481 .desc = {
1482 {
1483 .name = "filename",
1484 .type = QEMU_OPT_STRING,
1485 .help = "URL to the iscsi image",
1486 },
1487 { /* end of list */ }
1488 },
1489 };
1490
1491 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1492 int evpd, int pc, void **inq, Error **errp)
1493 {
1494 int full_size;
1495 struct scsi_task *task = NULL;
1496 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1497 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1498 goto fail;
1499 }
1500 full_size = scsi_datain_getfullsize(task);
1501 if (full_size > task->datain.size) {
1502 scsi_free_scsi_task(task);
1503
1504 /* we need more data for the full list */
1505 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1506 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1507 goto fail;
1508 }
1509 }
1510
1511 *inq = scsi_datain_unmarshall(task);
1512 if (*inq == NULL) {
1513 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1514 goto fail_with_err;
1515 }
1516
1517 return task;
1518
1519 fail:
1520 error_setg(errp, "iSCSI: Inquiry command failed : %s",
1521 iscsi_get_error(iscsi));
1522 fail_with_err:
1523 if (task != NULL) {
1524 scsi_free_scsi_task(task);
1525 }
1526 return NULL;
1527 }
1528
1529 static void iscsi_detach_aio_context(BlockDriverState *bs)
1530 {
1531 IscsiLun *iscsilun = bs->opaque;
1532
1533 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1534 false, NULL, NULL, NULL, NULL);
1535 iscsilun->events = 0;
1536
1537 if (iscsilun->nop_timer) {
1538 timer_del(iscsilun->nop_timer);
1539 timer_free(iscsilun->nop_timer);
1540 iscsilun->nop_timer = NULL;
1541 }
1542 if (iscsilun->event_timer) {
1543 timer_del(iscsilun->event_timer);
1544 timer_free(iscsilun->event_timer);
1545 iscsilun->event_timer = NULL;
1546 }
1547 }
1548
1549 static void iscsi_attach_aio_context(BlockDriverState *bs,
1550 AioContext *new_context)
1551 {
1552 IscsiLun *iscsilun = bs->opaque;
1553
1554 iscsilun->aio_context = new_context;
1555 iscsi_set_events(iscsilun);
1556
1557 /* Set up a timer for sending out iSCSI NOPs */
1558 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1559 QEMU_CLOCK_REALTIME, SCALE_MS,
1560 iscsi_nop_timed_event, iscsilun);
1561 timer_mod(iscsilun->nop_timer,
1562 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1563
1564 /* Set up a timer for periodic calls to iscsi_set_events and to
1565 * scan for command timeout */
1566 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1567 QEMU_CLOCK_REALTIME, SCALE_MS,
1568 iscsi_timed_check_events, iscsilun);
1569 timer_mod(iscsilun->event_timer,
1570 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1571 }
1572
1573 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1574 {
1575 struct scsi_task *task;
1576 struct scsi_mode_sense *ms = NULL;
1577 iscsilun->write_protected = false;
1578 iscsilun->dpofua = false;
1579
1580 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1581 1, SCSI_MODESENSE_PC_CURRENT,
1582 0x3F, 0, 255);
1583 if (task == NULL) {
1584 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1585 iscsi_get_error(iscsilun->iscsi));
1586 goto out;
1587 }
1588
1589 if (task->status != SCSI_STATUS_GOOD) {
1590 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1591 goto out;
1592 }
1593 ms = scsi_datain_unmarshall(task);
1594 if (!ms) {
1595 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1596 iscsi_get_error(iscsilun->iscsi));
1597 goto out;
1598 }
1599 iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1600 iscsilun->dpofua = ms->device_specific_parameter & 0x10;
1601
1602 out:
1603 if (task) {
1604 scsi_free_scsi_task(task);
1605 }
1606 }
1607
1608 /*
1609 * We support iscsi url's on the form
1610 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1611 */
1612 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1613 Error **errp)
1614 {
1615 IscsiLun *iscsilun = bs->opaque;
1616 struct iscsi_context *iscsi = NULL;
1617 struct iscsi_url *iscsi_url = NULL;
1618 struct scsi_task *task = NULL;
1619 struct scsi_inquiry_standard *inq = NULL;
1620 struct scsi_inquiry_supported_pages *inq_vpd;
1621 char *initiator_name = NULL;
1622 QemuOpts *opts;
1623 Error *local_err = NULL;
1624 const char *filename;
1625 int i, ret = 0, timeout = 0;
1626
1627 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1628 qemu_opts_absorb_qdict(opts, options, &local_err);
1629 if (local_err) {
1630 error_propagate(errp, local_err);
1631 ret = -EINVAL;
1632 goto out;
1633 }
1634
1635 filename = qemu_opt_get(opts, "filename");
1636
1637 iscsi_url = iscsi_parse_full_url(iscsi, filename);
1638 if (iscsi_url == NULL) {
1639 error_setg(errp, "Failed to parse URL : %s", filename);
1640 ret = -EINVAL;
1641 goto out;
1642 }
1643
1644 memset(iscsilun, 0, sizeof(IscsiLun));
1645
1646 initiator_name = parse_initiator_name(iscsi_url->target);
1647
1648 iscsi = iscsi_create_context(initiator_name);
1649 if (iscsi == NULL) {
1650 error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1651 ret = -ENOMEM;
1652 goto out;
1653 }
1654 #if LIBISCSI_API_VERSION >= (20160603)
1655 if (iscsi_init_transport(iscsi, iscsi_url->transport)) {
1656 error_setg(errp, ("Error initializing transport."));
1657 ret = -EINVAL;
1658 goto out;
1659 }
1660 #endif
1661 if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1662 error_setg(errp, "iSCSI: Failed to set target name.");
1663 ret = -EINVAL;
1664 goto out;
1665 }
1666
1667 if (iscsi_url->user[0] != '\0') {
1668 ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1669 iscsi_url->passwd);
1670 if (ret != 0) {
1671 error_setg(errp, "Failed to set initiator username and password");
1672 ret = -EINVAL;
1673 goto out;
1674 }
1675 }
1676
1677 /* check if we got CHAP username/password via the options */
1678 parse_chap(iscsi, iscsi_url->target, &local_err);
1679 if (local_err != NULL) {
1680 error_propagate(errp, local_err);
1681 ret = -EINVAL;
1682 goto out;
1683 }
1684
1685 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1686 error_setg(errp, "iSCSI: Failed to set session type to normal.");
1687 ret = -EINVAL;
1688 goto out;
1689 }
1690
1691 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1692
1693 /* check if we got HEADER_DIGEST via the options */
1694 parse_header_digest(iscsi, iscsi_url->target, &local_err);
1695 if (local_err != NULL) {
1696 error_propagate(errp, local_err);
1697 ret = -EINVAL;
1698 goto out;
1699 }
1700
1701 /* timeout handling is broken in libiscsi before 1.15.0 */
1702 timeout = parse_timeout(iscsi_url->target);
1703 #if LIBISCSI_API_VERSION >= 20150621
1704 iscsi_set_timeout(iscsi, timeout);
1705 #else
1706 if (timeout) {
1707 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1708 }
1709 #endif
1710
1711 if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1712 error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1713 iscsi_get_error(iscsi));
1714 ret = -EINVAL;
1715 goto out;
1716 }
1717
1718 iscsilun->iscsi = iscsi;
1719 iscsilun->aio_context = bdrv_get_aio_context(bs);
1720 iscsilun->lun = iscsi_url->lun;
1721 iscsilun->has_write_same = true;
1722
1723 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1724 (void **) &inq, errp);
1725 if (task == NULL) {
1726 ret = -EINVAL;
1727 goto out;
1728 }
1729 iscsilun->type = inq->periperal_device_type;
1730 scsi_free_scsi_task(task);
1731 task = NULL;
1732
1733 iscsi_modesense_sync(iscsilun);
1734 if (iscsilun->dpofua) {
1735 bs->supported_write_flags = BDRV_REQ_FUA;
1736 }
1737 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
1738
1739 /* Check the write protect flag of the LUN if we want to write */
1740 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1741 iscsilun->write_protected) {
1742 error_setg(errp, "Cannot open a write protected LUN as read-write");
1743 ret = -EACCES;
1744 goto out;
1745 }
1746
1747 iscsi_readcapacity_sync(iscsilun, &local_err);
1748 if (local_err != NULL) {
1749 error_propagate(errp, local_err);
1750 ret = -EINVAL;
1751 goto out;
1752 }
1753 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1754
1755 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1756 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1757 * will try to read from the device to guess the image format.
1758 */
1759 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1760 bs->sg = true;
1761 }
1762
1763 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1764 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1765 (void **) &inq_vpd, errp);
1766 if (task == NULL) {
1767 ret = -EINVAL;
1768 goto out;
1769 }
1770 for (i = 0; i < inq_vpd->num_pages; i++) {
1771 struct scsi_task *inq_task;
1772 struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1773 struct scsi_inquiry_block_limits *inq_bl;
1774 switch (inq_vpd->pages[i]) {
1775 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1776 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1777 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1778 (void **) &inq_lbp, errp);
1779 if (inq_task == NULL) {
1780 ret = -EINVAL;
1781 goto out;
1782 }
1783 memcpy(&iscsilun->lbp, inq_lbp,
1784 sizeof(struct scsi_inquiry_logical_block_provisioning));
1785 scsi_free_scsi_task(inq_task);
1786 break;
1787 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1788 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1789 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1790 (void **) &inq_bl, errp);
1791 if (inq_task == NULL) {
1792 ret = -EINVAL;
1793 goto out;
1794 }
1795 memcpy(&iscsilun->bl, inq_bl,
1796 sizeof(struct scsi_inquiry_block_limits));
1797 scsi_free_scsi_task(inq_task);
1798 break;
1799 default:
1800 break;
1801 }
1802 }
1803 scsi_free_scsi_task(task);
1804 task = NULL;
1805
1806 iscsi_attach_aio_context(bs, iscsilun->aio_context);
1807
1808 /* Guess the internal cluster (page) size of the iscsi target by the means
1809 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1810 * reasonable size */
1811 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
1812 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1813 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran *
1814 iscsilun->block_size) >> BDRV_SECTOR_BITS;
1815 if (iscsilun->lbprz) {
1816 ret = iscsi_allocmap_init(iscsilun, bs->open_flags);
1817 }
1818 }
1819
1820 out:
1821 qemu_opts_del(opts);
1822 g_free(initiator_name);
1823 if (iscsi_url != NULL) {
1824 iscsi_destroy_url(iscsi_url);
1825 }
1826 if (task != NULL) {
1827 scsi_free_scsi_task(task);
1828 }
1829
1830 if (ret) {
1831 if (iscsi != NULL) {
1832 if (iscsi_is_logged_in(iscsi)) {
1833 iscsi_logout_sync(iscsi);
1834 }
1835 iscsi_destroy_context(iscsi);
1836 }
1837 memset(iscsilun, 0, sizeof(IscsiLun));
1838 }
1839 return ret;
1840 }
1841
1842 static void iscsi_close(BlockDriverState *bs)
1843 {
1844 IscsiLun *iscsilun = bs->opaque;
1845 struct iscsi_context *iscsi = iscsilun->iscsi;
1846
1847 iscsi_detach_aio_context(bs);
1848 if (iscsi_is_logged_in(iscsi)) {
1849 iscsi_logout_sync(iscsi);
1850 }
1851 iscsi_destroy_context(iscsi);
1852 g_free(iscsilun->zeroblock);
1853 iscsi_allocmap_free(iscsilun);
1854 memset(iscsilun, 0, sizeof(IscsiLun));
1855 }
1856
1857 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
1858 {
1859 /* We don't actually refresh here, but just return data queried in
1860 * iscsi_open(): iscsi targets don't change their limits. */
1861
1862 IscsiLun *iscsilun = bs->opaque;
1863 uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
1864 unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size);
1865
1866 assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bs->sg);
1867
1868 bs->bl.request_alignment = block_size;
1869
1870 if (iscsilun->bl.max_xfer_len) {
1871 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
1872 }
1873
1874 if (max_xfer_len * block_size < INT_MAX) {
1875 bs->bl.max_transfer = max_xfer_len * iscsilun->block_size;
1876 }
1877
1878 if (iscsilun->lbp.lbpu) {
1879 if (iscsilun->bl.max_unmap < 0xffffffff / block_size) {
1880 bs->bl.max_pdiscard =
1881 iscsilun->bl.max_unmap * iscsilun->block_size;
1882 }
1883 bs->bl.pdiscard_alignment =
1884 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
1885 } else {
1886 bs->bl.pdiscard_alignment = iscsilun->block_size;
1887 }
1888
1889 if (iscsilun->bl.max_ws_len < 0xffffffff / block_size) {
1890 bs->bl.max_pwrite_zeroes =
1891 iscsilun->bl.max_ws_len * iscsilun->block_size;
1892 }
1893 if (iscsilun->lbp.lbpws) {
1894 bs->bl.pwrite_zeroes_alignment =
1895 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
1896 } else {
1897 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size;
1898 }
1899 if (iscsilun->bl.opt_xfer_len &&
1900 iscsilun->bl.opt_xfer_len < INT_MAX / block_size) {
1901 bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len *
1902 iscsilun->block_size);
1903 }
1904 }
1905
1906 /* Note that this will not re-establish a connection with an iSCSI target - it
1907 * is effectively a NOP. */
1908 static int iscsi_reopen_prepare(BDRVReopenState *state,
1909 BlockReopenQueue *queue, Error **errp)
1910 {
1911 IscsiLun *iscsilun = state->bs->opaque;
1912
1913 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
1914 error_setg(errp, "Cannot open a write protected LUN as read-write");
1915 return -EACCES;
1916 }
1917 return 0;
1918 }
1919
1920 static void iscsi_reopen_commit(BDRVReopenState *reopen_state)
1921 {
1922 IscsiLun *iscsilun = reopen_state->bs->opaque;
1923
1924 /* the cache.direct status might have changed */
1925 if (iscsilun->allocmap != NULL) {
1926 iscsi_allocmap_init(iscsilun, reopen_state->flags);
1927 }
1928 }
1929
1930 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1931 {
1932 IscsiLun *iscsilun = bs->opaque;
1933 Error *local_err = NULL;
1934
1935 if (iscsilun->type != TYPE_DISK) {
1936 return -ENOTSUP;
1937 }
1938
1939 iscsi_readcapacity_sync(iscsilun, &local_err);
1940 if (local_err != NULL) {
1941 error_free(local_err);
1942 return -EIO;
1943 }
1944
1945 if (offset > iscsi_getlength(bs)) {
1946 return -EINVAL;
1947 }
1948
1949 if (iscsilun->allocmap != NULL) {
1950 iscsi_allocmap_init(iscsilun, bs->open_flags);
1951 }
1952
1953 return 0;
1954 }
1955
1956 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp)
1957 {
1958 int ret = 0;
1959 int64_t total_size = 0;
1960 BlockDriverState *bs;
1961 IscsiLun *iscsilun = NULL;
1962 QDict *bs_options;
1963
1964 bs = bdrv_new();
1965
1966 /* Read out options */
1967 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1968 BDRV_SECTOR_SIZE);
1969 bs->opaque = g_new0(struct IscsiLun, 1);
1970 iscsilun = bs->opaque;
1971
1972 bs_options = qdict_new();
1973 qdict_put(bs_options, "filename", qstring_from_str(filename));
1974 ret = iscsi_open(bs, bs_options, 0, NULL);
1975 QDECREF(bs_options);
1976
1977 if (ret != 0) {
1978 goto out;
1979 }
1980 iscsi_detach_aio_context(bs);
1981 if (iscsilun->type != TYPE_DISK) {
1982 ret = -ENODEV;
1983 goto out;
1984 }
1985 if (bs->total_sectors < total_size) {
1986 ret = -ENOSPC;
1987 goto out;
1988 }
1989
1990 ret = 0;
1991 out:
1992 if (iscsilun->iscsi != NULL) {
1993 iscsi_destroy_context(iscsilun->iscsi);
1994 }
1995 g_free(bs->opaque);
1996 bs->opaque = NULL;
1997 bdrv_unref(bs);
1998 return ret;
1999 }
2000
2001 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2002 {
2003 IscsiLun *iscsilun = bs->opaque;
2004 bdi->unallocated_blocks_are_zero = iscsilun->lbprz;
2005 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
2006 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE;
2007 return 0;
2008 }
2009
2010 static void iscsi_invalidate_cache(BlockDriverState *bs,
2011 Error **errp)
2012 {
2013 IscsiLun *iscsilun = bs->opaque;
2014 iscsi_allocmap_invalidate(iscsilun);
2015 }
2016
2017 static QemuOptsList iscsi_create_opts = {
2018 .name = "iscsi-create-opts",
2019 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
2020 .desc = {
2021 {
2022 .name = BLOCK_OPT_SIZE,
2023 .type = QEMU_OPT_SIZE,
2024 .help = "Virtual disk size"
2025 },
2026 { /* end of list */ }
2027 }
2028 };
2029
2030 static BlockDriver bdrv_iscsi = {
2031 .format_name = "iscsi",
2032 .protocol_name = "iscsi",
2033
2034 .instance_size = sizeof(IscsiLun),
2035 .bdrv_needs_filename = true,
2036 .bdrv_file_open = iscsi_open,
2037 .bdrv_close = iscsi_close,
2038 .bdrv_create = iscsi_create,
2039 .create_opts = &iscsi_create_opts,
2040 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2041 .bdrv_reopen_commit = iscsi_reopen_commit,
2042 .bdrv_invalidate_cache = iscsi_invalidate_cache,
2043
2044 .bdrv_getlength = iscsi_getlength,
2045 .bdrv_get_info = iscsi_get_info,
2046 .bdrv_truncate = iscsi_truncate,
2047 .bdrv_refresh_limits = iscsi_refresh_limits,
2048
2049 .bdrv_co_get_block_status = iscsi_co_get_block_status,
2050 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2051 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2052 .bdrv_co_readv = iscsi_co_readv,
2053 .bdrv_co_writev_flags = iscsi_co_writev_flags,
2054 .bdrv_co_flush_to_disk = iscsi_co_flush,
2055
2056 #ifdef __linux__
2057 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2058 #endif
2059
2060 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2061 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2062 };
2063
2064 #if LIBISCSI_API_VERSION >= (20160603)
2065 static BlockDriver bdrv_iser = {
2066 .format_name = "iser",
2067 .protocol_name = "iser",
2068
2069 .instance_size = sizeof(IscsiLun),
2070 .bdrv_needs_filename = true,
2071 .bdrv_file_open = iscsi_open,
2072 .bdrv_close = iscsi_close,
2073 .bdrv_create = iscsi_create,
2074 .create_opts = &iscsi_create_opts,
2075 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2076 .bdrv_reopen_commit = iscsi_reopen_commit,
2077 .bdrv_invalidate_cache = iscsi_invalidate_cache,
2078
2079 .bdrv_getlength = iscsi_getlength,
2080 .bdrv_get_info = iscsi_get_info,
2081 .bdrv_truncate = iscsi_truncate,
2082 .bdrv_refresh_limits = iscsi_refresh_limits,
2083
2084 .bdrv_co_get_block_status = iscsi_co_get_block_status,
2085 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2086 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2087 .bdrv_co_readv = iscsi_co_readv,
2088 .bdrv_co_writev_flags = iscsi_co_writev_flags,
2089 .bdrv_co_flush_to_disk = iscsi_co_flush,
2090
2091 #ifdef __linux__
2092 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2093 #endif
2094
2095 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2096 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2097 };
2098 #endif
2099
2100 static void iscsi_block_init(void)
2101 {
2102 bdrv_register(&bdrv_iscsi);
2103 #if LIBISCSI_API_VERSION >= (20160603)
2104 bdrv_register(&bdrv_iser);
2105 #endif
2106 }
2107
2108 block_init(iscsi_block_init);
QEMU mirror