?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
qapi: add nbd-server-remove
[qemu/ar7.git] / block / iscsi.c
blob6a1c53711a7ba2310d2c853035aadc3c289ceb58
1 /*
2 * QEMU Block driver for iSCSI images
3 *
4 * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
5 * Copyright (c) 2012-2017 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 "scsi/constants.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 #include "scsi/utils.h"
44
45 /* Conflict between scsi/utils.h and libiscsi! :( */
46 #define SCSI_XFER_NONE ISCSI_XFER_NONE
47 #include <iscsi/iscsi.h>
48 #include <iscsi/scsi-lowlevel.h>
49 #undef SCSI_XFER_NONE
50 QEMU_BUILD_BUG_ON((int)SCSI_XFER_NONE != (int)ISCSI_XFER_NONE);
51
52 #ifdef __linux__
53 #include <scsi/sg.h>
54 #endif
55
56 typedef struct IscsiLun {
57 struct iscsi_context *iscsi;
58 AioContext *aio_context;
59 int lun;
60 enum scsi_inquiry_peripheral_device_type type;
61 int block_size;
62 uint64_t num_blocks;
63 int events;
64 QEMUTimer *nop_timer;
65 QEMUTimer *event_timer;
66 QemuMutex mutex;
67 struct scsi_inquiry_logical_block_provisioning lbp;
68 struct scsi_inquiry_block_limits bl;
69 unsigned char *zeroblock;
70 /* The allocmap tracks which clusters (pages) on the iSCSI target are
71 * allocated and which are not. In case a target returns zeros for
72 * unallocated pages (iscsilun->lprz) we can directly return zeros instead
73 * of reading zeros over the wire if a read request falls within an
74 * unallocated block. As there are 3 possible states we need 2 bitmaps to
75 * track. allocmap_valid keeps track if QEMU's information about a page is
76 * valid. allocmap tracks if a page is allocated or not. In case QEMU has no
77 * valid information about a page the corresponding allocmap entry should be
78 * switched to unallocated as well to force a new lookup of the allocation
79 * status as lookups are generally skipped if a page is suspect to be
80 * allocated. If a iSCSI target is opened with cache.direct = on the
81 * allocmap_valid does not exist turning all cached information invalid so
82 * that a fresh lookup is made for any page even if allocmap entry returns
83 * it's unallocated. */
84 unsigned long *allocmap;
85 unsigned long *allocmap_valid;
86 long allocmap_size;
87 int cluster_sectors;
88 bool use_16_for_rw;
89 bool write_protected;
90 bool lbpme;
91 bool lbprz;
92 bool dpofua;
93 bool has_write_same;
94 bool request_timed_out;
95 } IscsiLun;
96
97 typedef struct IscsiTask {
98 int status;
99 int complete;
100 int retries;
101 int do_retry;
102 struct scsi_task *task;
103 Coroutine *co;
104 IscsiLun *iscsilun;
105 QEMUTimer retry_timer;
106 int err_code;
107 char *err_str;
108 } IscsiTask;
109
110 typedef struct IscsiAIOCB {
111 BlockAIOCB common;
112 QEMUBH *bh;
113 IscsiLun *iscsilun;
114 struct scsi_task *task;
115 uint8_t *buf;
116 int status;
117 int64_t sector_num;
118 int nb_sectors;
119 int ret;
120 #ifdef __linux__
121 sg_io_hdr_t *ioh;
122 #endif
123 } IscsiAIOCB;
124
125 /* libiscsi uses time_t so its enough to process events every second */
126 #define EVENT_INTERVAL 1000
127 #define NOP_INTERVAL 5000
128 #define MAX_NOP_FAILURES 3
129 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times)
130 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768};
131
132 /* this threshold is a trade-off knob to choose between
133 * the potential additional overhead of an extra GET_LBA_STATUS request
134 * vs. unnecessarily reading a lot of zero sectors over the wire.
135 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
136 * sectors we check the allocation status of the area covered by the
137 * request first if the allocationmap indicates that the area might be
138 * unallocated. */
139 #define ISCSI_CHECKALLOC_THRES 64
140
141 static void
142 iscsi_bh_cb(void *p)
143 {
144 IscsiAIOCB *acb = p;
145
146 qemu_bh_delete(acb->bh);
147
148 g_free(acb->buf);
149 acb->buf = NULL;
150
151 acb->common.cb(acb->common.opaque, acb->status);
152
153 if (acb->task != NULL) {
154 scsi_free_scsi_task(acb->task);
155 acb->task = NULL;
156 }
157
158 qemu_aio_unref(acb);
159 }
160
161 static void
162 iscsi_schedule_bh(IscsiAIOCB *acb)
163 {
164 if (acb->bh) {
165 return;
166 }
167 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb);
168 qemu_bh_schedule(acb->bh);
169 }
170
171 static void iscsi_co_generic_bh_cb(void *opaque)
172 {
173 struct IscsiTask *iTask = opaque;
174
175 iTask->complete = 1;
176 aio_co_wake(iTask->co);
177 }
178
179 static void iscsi_retry_timer_expired(void *opaque)
180 {
181 struct IscsiTask *iTask = opaque;
182 iTask->complete = 1;
183 if (iTask->co) {
184 aio_co_wake(iTask->co);
185 }
186 }
187
188 static inline unsigned exp_random(double mean)
189 {
190 return -mean * log((double)rand() / RAND_MAX);
191 }
192
193 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in
194 * libiscsi 1.10.0, together with other constants we need. Use it as
195 * a hint that we have to define them ourselves if needed, to keep the
196 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for
197 * the test because SCSI_STATUS_* is an enum.
198 *
199 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes
200 * an enum, check against the LIBISCSI_API_VERSION macro, which was
201 * introduced in 1.11.0. If it is present, there is no need to define
202 * anything.
203 */
204 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \
205 !defined(LIBISCSI_API_VERSION)
206 #define SCSI_STATUS_TASK_SET_FULL 0x28
207 #define SCSI_STATUS_TIMEOUT 0x0f000002
208 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600
209 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00
210 #endif
211
212 #ifndef LIBISCSI_API_VERSION
213 #define LIBISCSI_API_VERSION 20130701
214 #endif
215
216 static int iscsi_translate_sense(struct scsi_sense *sense)
217 {
218 return - scsi_sense_to_errno(sense->key,
219 (sense->ascq & 0xFF00) >> 8,
220 sense->ascq & 0xFF);
221 }
222
223 /* Called (via iscsi_service) with QemuMutex held. */
224 static void
225 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
226 void *command_data, void *opaque)
227 {
228 struct IscsiTask *iTask = opaque;
229 struct scsi_task *task = command_data;
230
231 iTask->status = status;
232 iTask->do_retry = 0;
233 iTask->task = task;
234
235 if (status != SCSI_STATUS_GOOD) {
236 if (iTask->retries++ < ISCSI_CMD_RETRIES) {
237 if (status == SCSI_STATUS_CHECK_CONDITION
238 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
239 error_report("iSCSI CheckCondition: %s",
240 iscsi_get_error(iscsi));
241 iTask->do_retry = 1;
242 goto out;
243 }
244 if (status == SCSI_STATUS_BUSY ||
245 status == SCSI_STATUS_TIMEOUT ||
246 status == SCSI_STATUS_TASK_SET_FULL) {
247 unsigned retry_time =
248 exp_random(iscsi_retry_times[iTask->retries - 1]);
249 if (status == SCSI_STATUS_TIMEOUT) {
250 /* make sure the request is rescheduled AFTER the
251 * reconnect is initiated */
252 retry_time = EVENT_INTERVAL * 2;
253 iTask->iscsilun->request_timed_out = true;
254 }
255 error_report("iSCSI Busy/TaskSetFull/TimeOut"
256 " (retry #%u in %u ms): %s",
257 iTask->retries, retry_time,
258 iscsi_get_error(iscsi));
259 aio_timer_init(iTask->iscsilun->aio_context,
260 &iTask->retry_timer, QEMU_CLOCK_REALTIME,
261 SCALE_MS, iscsi_retry_timer_expired, iTask);
262 timer_mod(&iTask->retry_timer,
263 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
264 iTask->do_retry = 1;
265 return;
266 }
267 }
268 iTask->err_code = iscsi_translate_sense(&task->sense);
269 iTask->err_str = g_strdup(iscsi_get_error(iscsi));
270 }
271
272 out:
273 if (iTask->co) {
274 aio_bh_schedule_oneshot(iTask->iscsilun->aio_context,
275 iscsi_co_generic_bh_cb, iTask);
276 } else {
277 iTask->complete = 1;
278 }
279 }
280
281 static void iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
282 {
283 *iTask = (struct IscsiTask) {
284 .co = qemu_coroutine_self(),
285 .iscsilun = iscsilun,
286 };
287 }
288
289 static void
290 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
291 void *private_data)
292 {
293 IscsiAIOCB *acb = private_data;
294
295 acb->status = -ECANCELED;
296 iscsi_schedule_bh(acb);
297 }
298
299 static void
300 iscsi_aio_cancel(BlockAIOCB *blockacb)
301 {
302 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
303 IscsiLun *iscsilun = acb->iscsilun;
304
305 if (acb->status != -EINPROGRESS) {
306 return;
307 }
308
309 /* send a task mgmt call to the target to cancel the task on the target */
310 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
311 iscsi_abort_task_cb, acb);
312
313 }
314
315 static const AIOCBInfo iscsi_aiocb_info = {
316 .aiocb_size = sizeof(IscsiAIOCB),
317 .cancel_async = iscsi_aio_cancel,
318 };
319
320
321 static void iscsi_process_read(void *arg);
322 static void iscsi_process_write(void *arg);
323
324 /* Called with QemuMutex held. */
325 static void
326 iscsi_set_events(IscsiLun *iscsilun)
327 {
328 struct iscsi_context *iscsi = iscsilun->iscsi;
329 int ev = iscsi_which_events(iscsi);
330
331 if (ev != iscsilun->events) {
332 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi),
333 false,
334 (ev & POLLIN) ? iscsi_process_read : NULL,
335 (ev & POLLOUT) ? iscsi_process_write : NULL,
336 NULL,
337 iscsilun);
338 iscsilun->events = ev;
339 }
340 }
341
342 static void iscsi_timed_check_events(void *opaque)
343 {
344 IscsiLun *iscsilun = opaque;
345
346 /* check for timed out requests */
347 iscsi_service(iscsilun->iscsi, 0);
348
349 if (iscsilun->request_timed_out) {
350 iscsilun->request_timed_out = false;
351 iscsi_reconnect(iscsilun->iscsi);
352 }
353
354 /* newer versions of libiscsi may return zero events. Ensure we are able
355 * to return to service once this situation changes. */
356 iscsi_set_events(iscsilun);
357
358 timer_mod(iscsilun->event_timer,
359 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
360 }
361
362 static void
363 iscsi_process_read(void *arg)
364 {
365 IscsiLun *iscsilun = arg;
366 struct iscsi_context *iscsi = iscsilun->iscsi;
367
368 qemu_mutex_lock(&iscsilun->mutex);
369 iscsi_service(iscsi, POLLIN);
370 iscsi_set_events(iscsilun);
371 qemu_mutex_unlock(&iscsilun->mutex);
372 }
373
374 static void
375 iscsi_process_write(void *arg)
376 {
377 IscsiLun *iscsilun = arg;
378 struct iscsi_context *iscsi = iscsilun->iscsi;
379
380 qemu_mutex_lock(&iscsilun->mutex);
381 iscsi_service(iscsi, POLLOUT);
382 iscsi_set_events(iscsilun);
383 qemu_mutex_unlock(&iscsilun->mutex);
384 }
385
386 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
387 {
388 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
389 }
390
391 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
392 {
393 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
394 }
395
396 static bool is_byte_request_lun_aligned(int64_t offset, int count,
397 IscsiLun *iscsilun)
398 {
399 if (offset % iscsilun->block_size || count % iscsilun->block_size) {
400 error_report("iSCSI misaligned request: "
401 "iscsilun->block_size %u, offset %" PRIi64
402 ", count %d",
403 iscsilun->block_size, offset, count);
404 return false;
405 }
406 return true;
407 }
408
409 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors,
410 IscsiLun *iscsilun)
411 {
412 assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS);
413 return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS,
414 nb_sectors << BDRV_SECTOR_BITS,
415 iscsilun);
416 }
417
418 static void iscsi_allocmap_free(IscsiLun *iscsilun)
419 {
420 g_free(iscsilun->allocmap);
421 g_free(iscsilun->allocmap_valid);
422 iscsilun->allocmap = NULL;
423 iscsilun->allocmap_valid = NULL;
424 }
425
426
427 static int iscsi_allocmap_init(IscsiLun *iscsilun, int open_flags)
428 {
429 iscsi_allocmap_free(iscsilun);
430
431 iscsilun->allocmap_size =
432 DIV_ROUND_UP(sector_lun2qemu(iscsilun->num_blocks, iscsilun),
433 iscsilun->cluster_sectors);
434
435 iscsilun->allocmap = bitmap_try_new(iscsilun->allocmap_size);
436 if (!iscsilun->allocmap) {
437 return -ENOMEM;
438 }
439
440 if (open_flags & BDRV_O_NOCACHE) {
441 /* in case that cache.direct = on all allocmap entries are
442 * treated as invalid to force a relookup of the block
443 * status on every read request */
444 return 0;
445 }
446
447 iscsilun->allocmap_valid = bitmap_try_new(iscsilun->allocmap_size);
448 if (!iscsilun->allocmap_valid) {
449 /* if we are under memory pressure free the allocmap as well */
450 iscsi_allocmap_free(iscsilun);
451 return -ENOMEM;
452 }
453
454 return 0;
455 }
456
457 static void
458 iscsi_allocmap_update(IscsiLun *iscsilun, int64_t sector_num,
459 int nb_sectors, bool allocated, bool valid)
460 {
461 int64_t cl_num_expanded, nb_cls_expanded, cl_num_shrunk, nb_cls_shrunk;
462
463 if (iscsilun->allocmap == NULL) {
464 return;
465 }
466 /* expand to entirely contain all affected clusters */
467 cl_num_expanded = sector_num / iscsilun->cluster_sectors;
468 nb_cls_expanded = DIV_ROUND_UP(sector_num + nb_sectors,
469 iscsilun->cluster_sectors) - cl_num_expanded;
470 /* shrink to touch only completely contained clusters */
471 cl_num_shrunk = DIV_ROUND_UP(sector_num, iscsilun->cluster_sectors);
472 nb_cls_shrunk = (sector_num + nb_sectors) / iscsilun->cluster_sectors
473 - cl_num_shrunk;
474 if (allocated) {
475 bitmap_set(iscsilun->allocmap, cl_num_expanded, nb_cls_expanded);
476 } else {
477 if (nb_cls_shrunk > 0) {
478 bitmap_clear(iscsilun->allocmap, cl_num_shrunk, nb_cls_shrunk);
479 }
480 }
481
482 if (iscsilun->allocmap_valid == NULL) {
483 return;
484 }
485 if (valid) {
486 if (nb_cls_shrunk > 0) {
487 bitmap_set(iscsilun->allocmap_valid, cl_num_shrunk, nb_cls_shrunk);
488 }
489 } else {
490 bitmap_clear(iscsilun->allocmap_valid, cl_num_expanded,
491 nb_cls_expanded);
492 }
493 }
494
495 static void
496 iscsi_allocmap_set_allocated(IscsiLun *iscsilun, int64_t sector_num,
497 int nb_sectors)
498 {
499 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, true, true);
500 }
501
502 static void
503 iscsi_allocmap_set_unallocated(IscsiLun *iscsilun, int64_t sector_num,
504 int nb_sectors)
505 {
506 /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update
507 * is ignored, so this will in effect be an iscsi_allocmap_set_invalid.
508 */
509 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, false, true);
510 }
511
512 static void iscsi_allocmap_set_invalid(IscsiLun *iscsilun, int64_t sector_num,
513 int nb_sectors)
514 {
515 iscsi_allocmap_update(iscsilun, sector_num, nb_sectors, false, false);
516 }
517
518 static void iscsi_allocmap_invalidate(IscsiLun *iscsilun)
519 {
520 if (iscsilun->allocmap) {
521 bitmap_zero(iscsilun->allocmap, iscsilun->allocmap_size);
522 }
523 if (iscsilun->allocmap_valid) {
524 bitmap_zero(iscsilun->allocmap_valid, iscsilun->allocmap_size);
525 }
526 }
527
528 static inline bool
529 iscsi_allocmap_is_allocated(IscsiLun *iscsilun, int64_t sector_num,
530 int nb_sectors)
531 {
532 unsigned long size;
533 if (iscsilun->allocmap == NULL) {
534 return true;
535 }
536 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
537 return !(find_next_bit(iscsilun->allocmap, size,
538 sector_num / iscsilun->cluster_sectors) == size);
539 }
540
541 static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun,
542 int64_t sector_num, int nb_sectors)
543 {
544 unsigned long size;
545 if (iscsilun->allocmap_valid == NULL) {
546 return false;
547 }
548 size = DIV_ROUND_UP(sector_num + nb_sectors, iscsilun->cluster_sectors);
549 return (find_next_zero_bit(iscsilun->allocmap_valid, size,
550 sector_num / iscsilun->cluster_sectors) == size);
551 }
552
553 static int coroutine_fn
554 iscsi_co_writev_flags(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
555 QEMUIOVector *iov, int flags)
556 {
557 IscsiLun *iscsilun = bs->opaque;
558 struct IscsiTask iTask;
559 uint64_t lba;
560 uint32_t num_sectors;
561 bool fua = flags & BDRV_REQ_FUA;
562 int r = 0;
563
564 if (fua) {
565 assert(iscsilun->dpofua);
566 }
567 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
568 return -EINVAL;
569 }
570
571 if (bs->bl.max_transfer) {
572 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
573 }
574
575 lba = sector_qemu2lun(sector_num, iscsilun);
576 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
577 iscsi_co_init_iscsitask(iscsilun, &iTask);
578 qemu_mutex_lock(&iscsilun->mutex);
579 retry:
580 if (iscsilun->use_16_for_rw) {
581 #if LIBISCSI_API_VERSION >= (20160603)
582 iTask.task = iscsi_write16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
583 NULL, num_sectors * iscsilun->block_size,
584 iscsilun->block_size, 0, 0, fua, 0, 0,
585 iscsi_co_generic_cb, &iTask,
586 (struct scsi_iovec *)iov->iov, iov->niov);
587 } else {
588 iTask.task = iscsi_write10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
589 NULL, num_sectors * iscsilun->block_size,
590 iscsilun->block_size, 0, 0, fua, 0, 0,
591 iscsi_co_generic_cb, &iTask,
592 (struct scsi_iovec *)iov->iov, iov->niov);
593 }
594 #else
595 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
596 NULL, num_sectors * iscsilun->block_size,
597 iscsilun->block_size, 0, 0, fua, 0, 0,
598 iscsi_co_generic_cb, &iTask);
599 } else {
600 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
601 NULL, num_sectors * iscsilun->block_size,
602 iscsilun->block_size, 0, 0, fua, 0, 0,
603 iscsi_co_generic_cb, &iTask);
604 }
605 #endif
606 if (iTask.task == NULL) {
607 qemu_mutex_unlock(&iscsilun->mutex);
608 return -ENOMEM;
609 }
610 #if LIBISCSI_API_VERSION < (20160603)
611 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
612 iov->niov);
613 #endif
614 while (!iTask.complete) {
615 iscsi_set_events(iscsilun);
616 qemu_mutex_unlock(&iscsilun->mutex);
617 qemu_coroutine_yield();
618 qemu_mutex_lock(&iscsilun->mutex);
619 }
620
621 if (iTask.task != NULL) {
622 scsi_free_scsi_task(iTask.task);
623 iTask.task = NULL;
624 }
625
626 if (iTask.do_retry) {
627 iTask.complete = 0;
628 goto retry;
629 }
630
631 if (iTask.status != SCSI_STATUS_GOOD) {
632 iscsi_allocmap_set_invalid(iscsilun, sector_num, nb_sectors);
633 error_report("iSCSI WRITE10/16 failed at lba %" PRIu64 ": %s", lba,
634 iTask.err_str);
635 r = iTask.err_code;
636 goto out_unlock;
637 }
638
639 iscsi_allocmap_set_allocated(iscsilun, sector_num, nb_sectors);
640
641 out_unlock:
642 qemu_mutex_unlock(&iscsilun->mutex);
643 g_free(iTask.err_str);
644 return r;
645 }
646
647
648
649 static int64_t coroutine_fn iscsi_co_get_block_status(BlockDriverState *bs,
650 int64_t sector_num,
651 int nb_sectors, int *pnum,
652 BlockDriverState **file)
653 {
654 IscsiLun *iscsilun = bs->opaque;
655 struct scsi_get_lba_status *lbas = NULL;
656 struct scsi_lba_status_descriptor *lbasd = NULL;
657 struct IscsiTask iTask;
658 uint64_t lba;
659 int64_t ret;
660
661 iscsi_co_init_iscsitask(iscsilun, &iTask);
662
663 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
664 ret = -EINVAL;
665 goto out;
666 }
667
668 /* default to all sectors allocated */
669 ret = BDRV_BLOCK_DATA;
670 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
671 *pnum = nb_sectors;
672
673 /* LUN does not support logical block provisioning */
674 if (!iscsilun->lbpme) {
675 goto out;
676 }
677
678 lba = sector_qemu2lun(sector_num, iscsilun);
679
680 qemu_mutex_lock(&iscsilun->mutex);
681 retry:
682 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
683 lba, 8 + 16, iscsi_co_generic_cb,
684 &iTask) == NULL) {
685 ret = -ENOMEM;
686 goto out_unlock;
687 }
688
689 while (!iTask.complete) {
690 iscsi_set_events(iscsilun);
691 qemu_mutex_unlock(&iscsilun->mutex);
692 qemu_coroutine_yield();
693 qemu_mutex_lock(&iscsilun->mutex);
694 }
695
696 if (iTask.do_retry) {
697 if (iTask.task != NULL) {
698 scsi_free_scsi_task(iTask.task);
699 iTask.task = NULL;
700 }
701 iTask.complete = 0;
702 goto retry;
703 }
704
705 if (iTask.status != SCSI_STATUS_GOOD) {
706 /* in case the get_lba_status_callout fails (i.e.
707 * because the device is busy or the cmd is not
708 * supported) we pretend all blocks are allocated
709 * for backwards compatibility */
710 error_report("iSCSI GET_LBA_STATUS failed at lba %" PRIu64 ": %s",
711 lba, iTask.err_str);
712 goto out_unlock;
713 }
714
715 lbas = scsi_datain_unmarshall(iTask.task);
716 if (lbas == NULL) {
717 ret = -EIO;
718 goto out_unlock;
719 }
720
721 lbasd = &lbas->descriptors[0];
722
723 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
724 ret = -EIO;
725 goto out_unlock;
726 }
727
728 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
729
730 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
731 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
732 ret &= ~BDRV_BLOCK_DATA;
733 if (iscsilun->lbprz) {
734 ret |= BDRV_BLOCK_ZERO;
735 }
736 }
737
738 if (ret & BDRV_BLOCK_ZERO) {
739 iscsi_allocmap_set_unallocated(iscsilun, sector_num, *pnum);
740 } else {
741 iscsi_allocmap_set_allocated(iscsilun, sector_num, *pnum);
742 }
743
744 if (*pnum > nb_sectors) {
745 *pnum = nb_sectors;
746 }
747 out_unlock:
748 qemu_mutex_unlock(&iscsilun->mutex);
749 g_free(iTask.err_str);
750 out:
751 if (iTask.task != NULL) {
752 scsi_free_scsi_task(iTask.task);
753 }
754 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) {
755 *file = bs;
756 }
757 return ret;
758 }
759
760 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
761 int64_t sector_num, int nb_sectors,
762 QEMUIOVector *iov)
763 {
764 IscsiLun *iscsilun = bs->opaque;
765 struct IscsiTask iTask;
766 uint64_t lba;
767 uint32_t num_sectors;
768 int r = 0;
769
770 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
771 return -EINVAL;
772 }
773
774 if (bs->bl.max_transfer) {
775 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
776 }
777
778 /* if cache.direct is off and we have a valid entry in our allocation map
779 * we can skip checking the block status and directly return zeroes if
780 * the request falls within an unallocated area */
781 if (iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) &&
782 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
783 qemu_iovec_memset(iov, 0, 0x00, iov->size);
784 return 0;
785 }
786
787 if (nb_sectors >= ISCSI_CHECKALLOC_THRES &&
788 !iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) &&
789 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
790 int pnum;
791 BlockDriverState *file;
792 /* check the block status from the beginning of the cluster
793 * containing the start sector */
794 int64_t ret = iscsi_co_get_block_status(bs,
795 sector_num - sector_num % iscsilun->cluster_sectors,
796 BDRV_REQUEST_MAX_SECTORS, &pnum, &file);
797 if (ret < 0) {
798 return ret;
799 }
800 /* if the whole request falls into an unallocated area we can avoid
801 * to read and directly return zeroes instead */
802 if (ret & BDRV_BLOCK_ZERO &&
803 pnum >= nb_sectors + sector_num % iscsilun->cluster_sectors) {
804 qemu_iovec_memset(iov, 0, 0x00, iov->size);
805 return 0;
806 }
807 }
808
809 lba = sector_qemu2lun(sector_num, iscsilun);
810 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
811
812 iscsi_co_init_iscsitask(iscsilun, &iTask);
813 qemu_mutex_lock(&iscsilun->mutex);
814 retry:
815 if (iscsilun->use_16_for_rw) {
816 #if LIBISCSI_API_VERSION >= (20160603)
817 iTask.task = iscsi_read16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
818 num_sectors * iscsilun->block_size,
819 iscsilun->block_size, 0, 0, 0, 0, 0,
820 iscsi_co_generic_cb, &iTask,
821 (struct scsi_iovec *)iov->iov, iov->niov);
822 } else {
823 iTask.task = iscsi_read10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
824 num_sectors * iscsilun->block_size,
825 iscsilun->block_size,
826 0, 0, 0, 0, 0,
827 iscsi_co_generic_cb, &iTask,
828 (struct scsi_iovec *)iov->iov, iov->niov);
829 }
830 #else
831 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
832 num_sectors * iscsilun->block_size,
833 iscsilun->block_size, 0, 0, 0, 0, 0,
834 iscsi_co_generic_cb, &iTask);
835 } else {
836 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
837 num_sectors * iscsilun->block_size,
838 iscsilun->block_size,
839 0, 0, 0, 0, 0,
840 iscsi_co_generic_cb, &iTask);
841 }
842 #endif
843 if (iTask.task == NULL) {
844 qemu_mutex_unlock(&iscsilun->mutex);
845 return -ENOMEM;
846 }
847 #if LIBISCSI_API_VERSION < (20160603)
848 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
849 #endif
850 while (!iTask.complete) {
851 iscsi_set_events(iscsilun);
852 qemu_mutex_unlock(&iscsilun->mutex);
853 qemu_coroutine_yield();
854 qemu_mutex_lock(&iscsilun->mutex);
855 }
856
857 if (iTask.task != NULL) {
858 scsi_free_scsi_task(iTask.task);
859 iTask.task = NULL;
860 }
861
862 if (iTask.do_retry) {
863 iTask.complete = 0;
864 goto retry;
865 }
866
867 if (iTask.status != SCSI_STATUS_GOOD) {
868 error_report("iSCSI READ10/16 failed at lba %" PRIu64 ": %s",
869 lba, iTask.err_str);
870 r = iTask.err_code;
871 }
872
873 qemu_mutex_unlock(&iscsilun->mutex);
874 g_free(iTask.err_str);
875 return r;
876 }
877
878 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
879 {
880 IscsiLun *iscsilun = bs->opaque;
881 struct IscsiTask iTask;
882 int r = 0;
883
884 iscsi_co_init_iscsitask(iscsilun, &iTask);
885 qemu_mutex_lock(&iscsilun->mutex);
886 retry:
887 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
888 0, iscsi_co_generic_cb, &iTask) == NULL) {
889 qemu_mutex_unlock(&iscsilun->mutex);
890 return -ENOMEM;
891 }
892
893 while (!iTask.complete) {
894 iscsi_set_events(iscsilun);
895 qemu_mutex_unlock(&iscsilun->mutex);
896 qemu_coroutine_yield();
897 qemu_mutex_lock(&iscsilun->mutex);
898 }
899
900 if (iTask.task != NULL) {
901 scsi_free_scsi_task(iTask.task);
902 iTask.task = NULL;
903 }
904
905 if (iTask.do_retry) {
906 iTask.complete = 0;
907 goto retry;
908 }
909
910 if (iTask.status != SCSI_STATUS_GOOD) {
911 error_report("iSCSI SYNCHRONIZECACHE10 failed: %s", iTask.err_str);
912 r = iTask.err_code;
913 }
914
915 qemu_mutex_unlock(&iscsilun->mutex);
916 g_free(iTask.err_str);
917 return r;
918 }
919
920 #ifdef __linux__
921 /* Called (via iscsi_service) with QemuMutex held. */
922 static void
923 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
924 void *command_data, void *opaque)
925 {
926 IscsiAIOCB *acb = opaque;
927
928 g_free(acb->buf);
929 acb->buf = NULL;
930
931 acb->status = 0;
932 if (status < 0) {
933 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
934 iscsi_get_error(iscsi));
935 acb->status = iscsi_translate_sense(&acb->task->sense);
936 }
937
938 acb->ioh->driver_status = 0;
939 acb->ioh->host_status = 0;
940 acb->ioh->resid = 0;
941 acb->ioh->status = status;
942
943 #define SG_ERR_DRIVER_SENSE 0x08
944
945 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
946 int ss;
947
948 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
949
950 acb->ioh->sb_len_wr = acb->task->datain.size - 2;
951 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
952 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
953 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
954 }
955
956 iscsi_schedule_bh(acb);
957 }
958
959 static void iscsi_ioctl_bh_completion(void *opaque)
960 {
961 IscsiAIOCB *acb = opaque;
962
963 qemu_bh_delete(acb->bh);
964 acb->common.cb(acb->common.opaque, acb->ret);
965 qemu_aio_unref(acb);
966 }
967
968 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
969 {
970 BlockDriverState *bs = acb->common.bs;
971 IscsiLun *iscsilun = bs->opaque;
972 int ret = 0;
973
974 switch (req) {
975 case SG_GET_VERSION_NUM:
976 *(int *)buf = 30000;
977 break;
978 case SG_GET_SCSI_ID:
979 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
980 break;
981 default:
982 ret = -EINVAL;
983 }
984 assert(!acb->bh);
985 acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
986 iscsi_ioctl_bh_completion, acb);
987 acb->ret = ret;
988 qemu_bh_schedule(acb->bh);
989 }
990
991 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
992 unsigned long int req, void *buf,
993 BlockCompletionFunc *cb, void *opaque)
994 {
995 IscsiLun *iscsilun = bs->opaque;
996 struct iscsi_context *iscsi = iscsilun->iscsi;
997 struct iscsi_data data;
998 IscsiAIOCB *acb;
999
1000 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
1001
1002 acb->iscsilun = iscsilun;
1003 acb->bh = NULL;
1004 acb->status = -EINPROGRESS;
1005 acb->buf = NULL;
1006 acb->ioh = buf;
1007
1008 if (req != SG_IO) {
1009 iscsi_ioctl_handle_emulated(acb, req, buf);
1010 return &acb->common;
1011 }
1012
1013 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) {
1014 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)",
1015 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE);
1016 qemu_aio_unref(acb);
1017 return NULL;
1018 }
1019
1020 acb->task = malloc(sizeof(struct scsi_task));
1021 if (acb->task == NULL) {
1022 error_report("iSCSI: Failed to allocate task for scsi command. %s",
1023 iscsi_get_error(iscsi));
1024 qemu_aio_unref(acb);
1025 return NULL;
1026 }
1027 memset(acb->task, 0, sizeof(struct scsi_task));
1028
1029 switch (acb->ioh->dxfer_direction) {
1030 case SG_DXFER_TO_DEV:
1031 acb->task->xfer_dir = SCSI_XFER_WRITE;
1032 break;
1033 case SG_DXFER_FROM_DEV:
1034 acb->task->xfer_dir = SCSI_XFER_READ;
1035 break;
1036 default:
1037 acb->task->xfer_dir = SCSI_XFER_NONE;
1038 break;
1039 }
1040
1041 acb->task->cdb_size = acb->ioh->cmd_len;
1042 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
1043 acb->task->expxferlen = acb->ioh->dxfer_len;
1044
1045 data.size = 0;
1046 qemu_mutex_lock(&iscsilun->mutex);
1047 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
1048 if (acb->ioh->iovec_count == 0) {
1049 data.data = acb->ioh->dxferp;
1050 data.size = acb->ioh->dxfer_len;
1051 } else {
1052 scsi_task_set_iov_out(acb->task,
1053 (struct scsi_iovec *) acb->ioh->dxferp,
1054 acb->ioh->iovec_count);
1055 }
1056 }
1057
1058 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
1059 iscsi_aio_ioctl_cb,
1060 (data.size > 0) ? &data : NULL,
1061 acb) != 0) {
1062 qemu_mutex_unlock(&iscsilun->mutex);
1063 scsi_free_scsi_task(acb->task);
1064 qemu_aio_unref(acb);
1065 return NULL;
1066 }
1067
1068 /* tell libiscsi to read straight into the buffer we got from ioctl */
1069 if (acb->task->xfer_dir == SCSI_XFER_READ) {
1070 if (acb->ioh->iovec_count == 0) {
1071 scsi_task_add_data_in_buffer(acb->task,
1072 acb->ioh->dxfer_len,
1073 acb->ioh->dxferp);
1074 } else {
1075 scsi_task_set_iov_in(acb->task,
1076 (struct scsi_iovec *) acb->ioh->dxferp,
1077 acb->ioh->iovec_count);
1078 }
1079 }
1080
1081 iscsi_set_events(iscsilun);
1082 qemu_mutex_unlock(&iscsilun->mutex);
1083
1084 return &acb->common;
1085 }
1086
1087 #endif
1088
1089 static int64_t
1090 iscsi_getlength(BlockDriverState *bs)
1091 {
1092 IscsiLun *iscsilun = bs->opaque;
1093 int64_t len;
1094
1095 len = iscsilun->num_blocks;
1096 len *= iscsilun->block_size;
1097
1098 return len;
1099 }
1100
1101 static int
1102 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
1103 {
1104 IscsiLun *iscsilun = bs->opaque;
1105 struct IscsiTask iTask;
1106 struct unmap_list list;
1107 int r = 0;
1108
1109 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) {
1110 return -ENOTSUP;
1111 }
1112
1113 if (!iscsilun->lbp.lbpu) {
1114 /* UNMAP is not supported by the target */
1115 return 0;
1116 }
1117
1118 list.lba = offset / iscsilun->block_size;
1119 list.num = bytes / iscsilun->block_size;
1120
1121 iscsi_co_init_iscsitask(iscsilun, &iTask);
1122 qemu_mutex_lock(&iscsilun->mutex);
1123 retry:
1124 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
1125 iscsi_co_generic_cb, &iTask) == NULL) {
1126 r = -ENOMEM;
1127 goto out_unlock;
1128 }
1129
1130 while (!iTask.complete) {
1131 iscsi_set_events(iscsilun);
1132 qemu_mutex_unlock(&iscsilun->mutex);
1133 qemu_coroutine_yield();
1134 qemu_mutex_lock(&iscsilun->mutex);
1135 }
1136
1137 if (iTask.task != NULL) {
1138 scsi_free_scsi_task(iTask.task);
1139 iTask.task = NULL;
1140 }
1141
1142 if (iTask.do_retry) {
1143 iTask.complete = 0;
1144 goto retry;
1145 }
1146
1147 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1148 bytes >> BDRV_SECTOR_BITS);
1149
1150 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
1151 /* the target might fail with a check condition if it
1152 is not happy with the alignment of the UNMAP request
1153 we silently fail in this case */
1154 goto out_unlock;
1155 }
1156
1157 if (iTask.status != SCSI_STATUS_GOOD) {
1158 error_report("iSCSI UNMAP failed at lba %" PRIu64 ": %s",
1159 list.lba, iTask.err_str);
1160 r = iTask.err_code;
1161 goto out_unlock;
1162 }
1163
1164 out_unlock:
1165 qemu_mutex_unlock(&iscsilun->mutex);
1166 g_free(iTask.err_str);
1167 return r;
1168 }
1169
1170 static int
1171 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1172 int bytes, BdrvRequestFlags flags)
1173 {
1174 IscsiLun *iscsilun = bs->opaque;
1175 struct IscsiTask iTask;
1176 uint64_t lba;
1177 uint32_t nb_blocks;
1178 bool use_16_for_ws = iscsilun->use_16_for_rw;
1179 int r = 0;
1180
1181 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) {
1182 return -ENOTSUP;
1183 }
1184
1185 if (flags & BDRV_REQ_MAY_UNMAP) {
1186 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
1187 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
1188 use_16_for_ws = true;
1189 }
1190 if (use_16_for_ws && !iscsilun->lbp.lbpws) {
1191 /* WRITESAME16 with UNMAP is not supported by the target,
1192 * fall back and try WRITESAME10/16 without UNMAP */
1193 flags &= ~BDRV_REQ_MAY_UNMAP;
1194 use_16_for_ws = iscsilun->use_16_for_rw;
1195 }
1196 }
1197
1198 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1199 /* WRITESAME without UNMAP is not supported by the target */
1200 return -ENOTSUP;
1201 }
1202
1203 lba = offset / iscsilun->block_size;
1204 nb_blocks = bytes / iscsilun->block_size;
1205
1206 if (iscsilun->zeroblock == NULL) {
1207 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1208 if (iscsilun->zeroblock == NULL) {
1209 return -ENOMEM;
1210 }
1211 }
1212
1213 qemu_mutex_lock(&iscsilun->mutex);
1214 iscsi_co_init_iscsitask(iscsilun, &iTask);
1215 retry:
1216 if (use_16_for_ws) {
1217 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1218 iscsilun->zeroblock, iscsilun->block_size,
1219 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1220 0, 0, iscsi_co_generic_cb, &iTask);
1221 } else {
1222 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1223 iscsilun->zeroblock, iscsilun->block_size,
1224 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1225 0, 0, iscsi_co_generic_cb, &iTask);
1226 }
1227 if (iTask.task == NULL) {
1228 qemu_mutex_unlock(&iscsilun->mutex);
1229 return -ENOMEM;
1230 }
1231
1232 while (!iTask.complete) {
1233 iscsi_set_events(iscsilun);
1234 qemu_mutex_unlock(&iscsilun->mutex);
1235 qemu_coroutine_yield();
1236 qemu_mutex_lock(&iscsilun->mutex);
1237 }
1238
1239 if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1240 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1241 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1242 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1243 /* WRITE SAME is not supported by the target */
1244 iscsilun->has_write_same = false;
1245 scsi_free_scsi_task(iTask.task);
1246 r = -ENOTSUP;
1247 goto out_unlock;
1248 }
1249
1250 if (iTask.task != NULL) {
1251 scsi_free_scsi_task(iTask.task);
1252 iTask.task = NULL;
1253 }
1254
1255 if (iTask.do_retry) {
1256 iTask.complete = 0;
1257 goto retry;
1258 }
1259
1260 if (iTask.status != SCSI_STATUS_GOOD) {
1261 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1262 bytes >> BDRV_SECTOR_BITS);
1263 error_report("iSCSI WRITESAME10/16 failed at lba %" PRIu64 ": %s",
1264 lba, iTask.err_str);
1265 r = iTask.err_code;
1266 goto out_unlock;
1267 }
1268
1269 if (flags & BDRV_REQ_MAY_UNMAP) {
1270 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1271 bytes >> BDRV_SECTOR_BITS);
1272 } else {
1273 iscsi_allocmap_set_allocated(iscsilun, offset >> BDRV_SECTOR_BITS,
1274 bytes >> BDRV_SECTOR_BITS);
1275 }
1276
1277 out_unlock:
1278 qemu_mutex_unlock(&iscsilun->mutex);
1279 g_free(iTask.err_str);
1280 return r;
1281 }
1282
1283 static void apply_chap(struct iscsi_context *iscsi, QemuOpts *opts,
1284 Error **errp)
1285 {
1286 const char *user = NULL;
1287 const char *password = NULL;
1288 const char *secretid;
1289 char *secret = NULL;
1290
1291 user = qemu_opt_get(opts, "user");
1292 if (!user) {
1293 return;
1294 }
1295
1296 secretid = qemu_opt_get(opts, "password-secret");
1297 password = qemu_opt_get(opts, "password");
1298 if (secretid && password) {
1299 error_setg(errp, "'password' and 'password-secret' properties are "
1300 "mutually exclusive");
1301 return;
1302 }
1303 if (secretid) {
1304 secret = qcrypto_secret_lookup_as_utf8(secretid, errp);
1305 if (!secret) {
1306 return;
1307 }
1308 password = secret;
1309 } else if (!password) {
1310 error_setg(errp, "CHAP username specified but no password was given");
1311 return;
1312 }
1313
1314 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1315 error_setg(errp, "Failed to set initiator username and password");
1316 }
1317
1318 g_free(secret);
1319 }
1320
1321 static void apply_header_digest(struct iscsi_context *iscsi, QemuOpts *opts,
1322 Error **errp)
1323 {
1324 const char *digest = NULL;
1325
1326 digest = qemu_opt_get(opts, "header-digest");
1327 if (!digest) {
1328 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1329 } else if (!strcmp(digest, "crc32c")) {
1330 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1331 } else if (!strcmp(digest, "none")) {
1332 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1333 } else if (!strcmp(digest, "crc32c-none")) {
1334 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1335 } else if (!strcmp(digest, "none-crc32c")) {
1336 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1337 } else {
1338 error_setg(errp, "Invalid header-digest setting : %s", digest);
1339 }
1340 }
1341
1342 static char *get_initiator_name(QemuOpts *opts)
1343 {
1344 const char *name;
1345 char *iscsi_name;
1346 UuidInfo *uuid_info;
1347
1348 name = qemu_opt_get(opts, "initiator-name");
1349 if (name) {
1350 return g_strdup(name);
1351 }
1352
1353 uuid_info = qmp_query_uuid(NULL);
1354 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1355 name = qemu_get_vm_name();
1356 } else {
1357 name = uuid_info->UUID;
1358 }
1359 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1360 name ? ":" : "", name ? name : "");
1361 qapi_free_UuidInfo(uuid_info);
1362 return iscsi_name;
1363 }
1364
1365 static void iscsi_nop_timed_event(void *opaque)
1366 {
1367 IscsiLun *iscsilun = opaque;
1368
1369 qemu_mutex_lock(&iscsilun->mutex);
1370 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1371 error_report("iSCSI: NOP timeout. Reconnecting...");
1372 iscsilun->request_timed_out = true;
1373 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1374 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1375 goto out;
1376 }
1377
1378 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1379 iscsi_set_events(iscsilun);
1380
1381 out:
1382 qemu_mutex_unlock(&iscsilun->mutex);
1383 }
1384
1385 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1386 {
1387 struct scsi_task *task = NULL;
1388 struct scsi_readcapacity10 *rc10 = NULL;
1389 struct scsi_readcapacity16 *rc16 = NULL;
1390 int retries = ISCSI_CMD_RETRIES;
1391
1392 do {
1393 if (task != NULL) {
1394 scsi_free_scsi_task(task);
1395 task = NULL;
1396 }
1397
1398 switch (iscsilun->type) {
1399 case TYPE_DISK:
1400 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1401 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1402 rc16 = scsi_datain_unmarshall(task);
1403 if (rc16 == NULL) {
1404 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1405 } else {
1406 iscsilun->block_size = rc16->block_length;
1407 iscsilun->num_blocks = rc16->returned_lba + 1;
1408 iscsilun->lbpme = !!rc16->lbpme;
1409 iscsilun->lbprz = !!rc16->lbprz;
1410 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1411 }
1412 break;
1413 }
1414 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1415 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1416 break;
1417 }
1418 /* Fall through and try READ CAPACITY(10) instead. */
1419 case TYPE_ROM:
1420 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1421 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1422 rc10 = scsi_datain_unmarshall(task);
1423 if (rc10 == NULL) {
1424 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1425 } else {
1426 iscsilun->block_size = rc10->block_size;
1427 if (rc10->lba == 0) {
1428 /* blank disk loaded */
1429 iscsilun->num_blocks = 0;
1430 } else {
1431 iscsilun->num_blocks = rc10->lba + 1;
1432 }
1433 }
1434 }
1435 break;
1436 default:
1437 return;
1438 }
1439 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1440 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1441 && retries-- > 0);
1442
1443 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1444 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1445 } else if (!iscsilun->block_size ||
1446 iscsilun->block_size % BDRV_SECTOR_SIZE) {
1447 error_setg(errp, "iSCSI: the target returned an invalid "
1448 "block size of %d.", iscsilun->block_size);
1449 }
1450 if (task) {
1451 scsi_free_scsi_task(task);
1452 }
1453 }
1454
1455 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1456 int evpd, int pc, void **inq, Error **errp)
1457 {
1458 int full_size;
1459 struct scsi_task *task = NULL;
1460 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1461 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1462 goto fail;
1463 }
1464 full_size = scsi_datain_getfullsize(task);
1465 if (full_size > task->datain.size) {
1466 scsi_free_scsi_task(task);
1467
1468 /* we need more data for the full list */
1469 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1470 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1471 goto fail;
1472 }
1473 }
1474
1475 *inq = scsi_datain_unmarshall(task);
1476 if (*inq == NULL) {
1477 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1478 goto fail_with_err;
1479 }
1480
1481 return task;
1482
1483 fail:
1484 error_setg(errp, "iSCSI: Inquiry command failed : %s",
1485 iscsi_get_error(iscsi));
1486 fail_with_err:
1487 if (task != NULL) {
1488 scsi_free_scsi_task(task);
1489 }
1490 return NULL;
1491 }
1492
1493 static void iscsi_detach_aio_context(BlockDriverState *bs)
1494 {
1495 IscsiLun *iscsilun = bs->opaque;
1496
1497 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1498 false, NULL, NULL, NULL, NULL);
1499 iscsilun->events = 0;
1500
1501 if (iscsilun->nop_timer) {
1502 timer_del(iscsilun->nop_timer);
1503 timer_free(iscsilun->nop_timer);
1504 iscsilun->nop_timer = NULL;
1505 }
1506 if (iscsilun->event_timer) {
1507 timer_del(iscsilun->event_timer);
1508 timer_free(iscsilun->event_timer);
1509 iscsilun->event_timer = NULL;
1510 }
1511 }
1512
1513 static void iscsi_attach_aio_context(BlockDriverState *bs,
1514 AioContext *new_context)
1515 {
1516 IscsiLun *iscsilun = bs->opaque;
1517
1518 iscsilun->aio_context = new_context;
1519 iscsi_set_events(iscsilun);
1520
1521 /* Set up a timer for sending out iSCSI NOPs */
1522 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1523 QEMU_CLOCK_REALTIME, SCALE_MS,
1524 iscsi_nop_timed_event, iscsilun);
1525 timer_mod(iscsilun->nop_timer,
1526 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1527
1528 /* Set up a timer for periodic calls to iscsi_set_events and to
1529 * scan for command timeout */
1530 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1531 QEMU_CLOCK_REALTIME, SCALE_MS,
1532 iscsi_timed_check_events, iscsilun);
1533 timer_mod(iscsilun->event_timer,
1534 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1535 }
1536
1537 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1538 {
1539 struct scsi_task *task;
1540 struct scsi_mode_sense *ms = NULL;
1541 iscsilun->write_protected = false;
1542 iscsilun->dpofua = false;
1543
1544 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1545 1, SCSI_MODESENSE_PC_CURRENT,
1546 0x3F, 0, 255);
1547 if (task == NULL) {
1548 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1549 iscsi_get_error(iscsilun->iscsi));
1550 goto out;
1551 }
1552
1553 if (task->status != SCSI_STATUS_GOOD) {
1554 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1555 goto out;
1556 }
1557 ms = scsi_datain_unmarshall(task);
1558 if (!ms) {
1559 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1560 iscsi_get_error(iscsilun->iscsi));
1561 goto out;
1562 }
1563 iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1564 iscsilun->dpofua = ms->device_specific_parameter & 0x10;
1565
1566 out:
1567 if (task) {
1568 scsi_free_scsi_task(task);
1569 }
1570 }
1571
1572 static void iscsi_parse_iscsi_option(const char *target, QDict *options)
1573 {
1574 QemuOptsList *list;
1575 QemuOpts *opts;
1576 const char *user, *password, *password_secret, *initiator_name,
1577 *header_digest, *timeout;
1578
1579 list = qemu_find_opts("iscsi");
1580 if (!list) {
1581 return;
1582 }
1583
1584 opts = qemu_opts_find(list, target);
1585 if (opts == NULL) {
1586 opts = QTAILQ_FIRST(&list->head);
1587 if (!opts) {
1588 return;
1589 }
1590 }
1591
1592 user = qemu_opt_get(opts, "user");
1593 if (user) {
1594 qdict_set_default_str(options, "user", user);
1595 }
1596
1597 password = qemu_opt_get(opts, "password");
1598 if (password) {
1599 qdict_set_default_str(options, "password", password);
1600 }
1601
1602 password_secret = qemu_opt_get(opts, "password-secret");
1603 if (password_secret) {
1604 qdict_set_default_str(options, "password-secret", password_secret);
1605 }
1606
1607 initiator_name = qemu_opt_get(opts, "initiator-name");
1608 if (initiator_name) {
1609 qdict_set_default_str(options, "initiator-name", initiator_name);
1610 }
1611
1612 header_digest = qemu_opt_get(opts, "header-digest");
1613 if (header_digest) {
1614 /* -iscsi takes upper case values, but QAPI only supports lower case
1615 * enum constant names, so we have to convert here. */
1616 char *qapi_value = g_ascii_strdown(header_digest, -1);
1617 qdict_set_default_str(options, "header-digest", qapi_value);
1618 g_free(qapi_value);
1619 }
1620
1621 timeout = qemu_opt_get(opts, "timeout");
1622 if (timeout) {
1623 qdict_set_default_str(options, "timeout", timeout);
1624 }
1625 }
1626
1627 /*
1628 * We support iscsi url's on the form
1629 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1630 */
1631 static void iscsi_parse_filename(const char *filename, QDict *options,
1632 Error **errp)
1633 {
1634 struct iscsi_url *iscsi_url;
1635 const char *transport_name;
1636 char *lun_str;
1637
1638 iscsi_url = iscsi_parse_full_url(NULL, filename);
1639 if (iscsi_url == NULL) {
1640 error_setg(errp, "Failed to parse URL : %s", filename);
1641 return;
1642 }
1643
1644 #if LIBISCSI_API_VERSION >= (20160603)
1645 switch (iscsi_url->transport) {
1646 case TCP_TRANSPORT:
1647 transport_name = "tcp";
1648 break;
1649 case ISER_TRANSPORT:
1650 transport_name = "iser";
1651 break;
1652 default:
1653 error_setg(errp, "Unknown transport type (%d)",
1654 iscsi_url->transport);
1655 return;
1656 }
1657 #else
1658 transport_name = "tcp";
1659 #endif
1660
1661 qdict_set_default_str(options, "transport", transport_name);
1662 qdict_set_default_str(options, "portal", iscsi_url->portal);
1663 qdict_set_default_str(options, "target", iscsi_url->target);
1664
1665 lun_str = g_strdup_printf("%d", iscsi_url->lun);
1666 qdict_set_default_str(options, "lun", lun_str);
1667 g_free(lun_str);
1668
1669 /* User/password from -iscsi take precedence over those from the URL */
1670 iscsi_parse_iscsi_option(iscsi_url->target, options);
1671
1672 if (iscsi_url->user[0] != '\0') {
1673 qdict_set_default_str(options, "user", iscsi_url->user);
1674 qdict_set_default_str(options, "password", iscsi_url->passwd);
1675 }
1676
1677 iscsi_destroy_url(iscsi_url);
1678 }
1679
1680 static QemuOptsList runtime_opts = {
1681 .name = "iscsi",
1682 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1683 .desc = {
1684 {
1685 .name = "transport",
1686 .type = QEMU_OPT_STRING,
1687 },
1688 {
1689 .name = "portal",
1690 .type = QEMU_OPT_STRING,
1691 },
1692 {
1693 .name = "target",
1694 .type = QEMU_OPT_STRING,
1695 },
1696 {
1697 .name = "user",
1698 .type = QEMU_OPT_STRING,
1699 },
1700 {
1701 .name = "password",
1702 .type = QEMU_OPT_STRING,
1703 },
1704 {
1705 .name = "password-secret",
1706 .type = QEMU_OPT_STRING,
1707 },
1708 {
1709 .name = "lun",
1710 .type = QEMU_OPT_NUMBER,
1711 },
1712 {
1713 .name = "initiator-name",
1714 .type = QEMU_OPT_STRING,
1715 },
1716 {
1717 .name = "header-digest",
1718 .type = QEMU_OPT_STRING,
1719 },
1720 {
1721 .name = "timeout",
1722 .type = QEMU_OPT_NUMBER,
1723 },
1724 {
1725 .name = "filename",
1726 .type = QEMU_OPT_STRING,
1727 },
1728 { /* end of list */ }
1729 },
1730 };
1731
1732 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1733 Error **errp)
1734 {
1735 IscsiLun *iscsilun = bs->opaque;
1736 struct iscsi_context *iscsi = NULL;
1737 struct scsi_task *task = NULL;
1738 struct scsi_inquiry_standard *inq = NULL;
1739 struct scsi_inquiry_supported_pages *inq_vpd;
1740 char *initiator_name = NULL;
1741 QemuOpts *opts;
1742 Error *local_err = NULL;
1743 const char *transport_name, *portal, *target, *filename;
1744 #if LIBISCSI_API_VERSION >= (20160603)
1745 enum iscsi_transport_type transport;
1746 #endif
1747 int i, ret = 0, timeout = 0, lun;
1748
1749 /* If we are given a filename, parse the filename, with precedence given to
1750 * filename encoded options */
1751 filename = qdict_get_try_str(options, "filename");
1752 if (filename) {
1753 warn_report("'filename' option specified. "
1754 "This is an unsupported option, and may be deprecated "
1755 "in the future");
1756 iscsi_parse_filename(filename, options, &local_err);
1757 if (local_err) {
1758 ret = -EINVAL;
1759 error_propagate(errp, local_err);
1760 goto exit;
1761 }
1762 }
1763
1764 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1765 qemu_opts_absorb_qdict(opts, options, &local_err);
1766 if (local_err) {
1767 error_propagate(errp, local_err);
1768 ret = -EINVAL;
1769 goto out;
1770 }
1771
1772 transport_name = qemu_opt_get(opts, "transport");
1773 portal = qemu_opt_get(opts, "portal");
1774 target = qemu_opt_get(opts, "target");
1775 lun = qemu_opt_get_number(opts, "lun", 0);
1776
1777 if (!transport_name || !portal || !target) {
1778 error_setg(errp, "Need all of transport, portal and target options");
1779 ret = -EINVAL;
1780 goto out;
1781 }
1782
1783 if (!strcmp(transport_name, "tcp")) {
1784 #if LIBISCSI_API_VERSION >= (20160603)
1785 transport = TCP_TRANSPORT;
1786 } else if (!strcmp(transport_name, "iser")) {
1787 transport = ISER_TRANSPORT;
1788 #else
1789 /* TCP is what older libiscsi versions always use */
1790 #endif
1791 } else {
1792 error_setg(errp, "Unknown transport: %s", transport_name);
1793 ret = -EINVAL;
1794 goto out;
1795 }
1796
1797 memset(iscsilun, 0, sizeof(IscsiLun));
1798
1799 initiator_name = get_initiator_name(opts);
1800
1801 iscsi = iscsi_create_context(initiator_name);
1802 if (iscsi == NULL) {
1803 error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1804 ret = -ENOMEM;
1805 goto out;
1806 }
1807 #if LIBISCSI_API_VERSION >= (20160603)
1808 if (iscsi_init_transport(iscsi, transport)) {
1809 error_setg(errp, ("Error initializing transport."));
1810 ret = -EINVAL;
1811 goto out;
1812 }
1813 #endif
1814 if (iscsi_set_targetname(iscsi, target)) {
1815 error_setg(errp, "iSCSI: Failed to set target name.");
1816 ret = -EINVAL;
1817 goto out;
1818 }
1819
1820 /* check if we got CHAP username/password via the options */
1821 apply_chap(iscsi, opts, &local_err);
1822 if (local_err != NULL) {
1823 error_propagate(errp, local_err);
1824 ret = -EINVAL;
1825 goto out;
1826 }
1827
1828 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1829 error_setg(errp, "iSCSI: Failed to set session type to normal.");
1830 ret = -EINVAL;
1831 goto out;
1832 }
1833
1834 /* check if we got HEADER_DIGEST via the options */
1835 apply_header_digest(iscsi, opts, &local_err);
1836 if (local_err != NULL) {
1837 error_propagate(errp, local_err);
1838 ret = -EINVAL;
1839 goto out;
1840 }
1841
1842 /* timeout handling is broken in libiscsi before 1.15.0 */
1843 timeout = qemu_opt_get_number(opts, "timeout", 0);
1844 #if LIBISCSI_API_VERSION >= 20150621
1845 iscsi_set_timeout(iscsi, timeout);
1846 #else
1847 if (timeout) {
1848 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1849 }
1850 #endif
1851
1852 if (iscsi_full_connect_sync(iscsi, portal, lun) != 0) {
1853 error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1854 iscsi_get_error(iscsi));
1855 ret = -EINVAL;
1856 goto out;
1857 }
1858
1859 iscsilun->iscsi = iscsi;
1860 iscsilun->aio_context = bdrv_get_aio_context(bs);
1861 iscsilun->lun = lun;
1862 iscsilun->has_write_same = true;
1863
1864 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1865 (void **) &inq, errp);
1866 if (task == NULL) {
1867 ret = -EINVAL;
1868 goto out;
1869 }
1870 iscsilun->type = inq->periperal_device_type;
1871 scsi_free_scsi_task(task);
1872 task = NULL;
1873
1874 iscsi_modesense_sync(iscsilun);
1875 if (iscsilun->dpofua) {
1876 bs->supported_write_flags = BDRV_REQ_FUA;
1877 }
1878 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
1879
1880 /* Check the write protect flag of the LUN if we want to write */
1881 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1882 iscsilun->write_protected) {
1883 error_setg(errp, "Cannot open a write protected LUN as read-write");
1884 ret = -EACCES;
1885 goto out;
1886 }
1887
1888 iscsi_readcapacity_sync(iscsilun, &local_err);
1889 if (local_err != NULL) {
1890 error_propagate(errp, local_err);
1891 ret = -EINVAL;
1892 goto out;
1893 }
1894 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1895
1896 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1897 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1898 * will try to read from the device to guess the image format.
1899 */
1900 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1901 bs->sg = true;
1902 }
1903
1904 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1905 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1906 (void **) &inq_vpd, errp);
1907 if (task == NULL) {
1908 ret = -EINVAL;
1909 goto out;
1910 }
1911 for (i = 0; i < inq_vpd->num_pages; i++) {
1912 struct scsi_task *inq_task;
1913 struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1914 struct scsi_inquiry_block_limits *inq_bl;
1915 switch (inq_vpd->pages[i]) {
1916 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1917 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1918 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1919 (void **) &inq_lbp, errp);
1920 if (inq_task == NULL) {
1921 ret = -EINVAL;
1922 goto out;
1923 }
1924 memcpy(&iscsilun->lbp, inq_lbp,
1925 sizeof(struct scsi_inquiry_logical_block_provisioning));
1926 scsi_free_scsi_task(inq_task);
1927 break;
1928 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1929 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1930 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1931 (void **) &inq_bl, errp);
1932 if (inq_task == NULL) {
1933 ret = -EINVAL;
1934 goto out;
1935 }
1936 memcpy(&iscsilun->bl, inq_bl,
1937 sizeof(struct scsi_inquiry_block_limits));
1938 scsi_free_scsi_task(inq_task);
1939 break;
1940 default:
1941 break;
1942 }
1943 }
1944 scsi_free_scsi_task(task);
1945 task = NULL;
1946
1947 qemu_mutex_init(&iscsilun->mutex);
1948 iscsi_attach_aio_context(bs, iscsilun->aio_context);
1949
1950 /* Guess the internal cluster (page) size of the iscsi target by the means
1951 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1952 * reasonable size */
1953 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
1954 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1955 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran *
1956 iscsilun->block_size) >> BDRV_SECTOR_BITS;
1957 if (iscsilun->lbprz) {
1958 ret = iscsi_allocmap_init(iscsilun, bs->open_flags);
1959 }
1960 }
1961
1962 out:
1963 qemu_opts_del(opts);
1964 g_free(initiator_name);
1965 if (task != NULL) {
1966 scsi_free_scsi_task(task);
1967 }
1968
1969 if (ret) {
1970 if (iscsi != NULL) {
1971 if (iscsi_is_logged_in(iscsi)) {
1972 iscsi_logout_sync(iscsi);
1973 }
1974 iscsi_destroy_context(iscsi);
1975 }
1976 memset(iscsilun, 0, sizeof(IscsiLun));
1977 }
1978 exit:
1979 return ret;
1980 }
1981
1982 static void iscsi_close(BlockDriverState *bs)
1983 {
1984 IscsiLun *iscsilun = bs->opaque;
1985 struct iscsi_context *iscsi = iscsilun->iscsi;
1986
1987 iscsi_detach_aio_context(bs);
1988 if (iscsi_is_logged_in(iscsi)) {
1989 iscsi_logout_sync(iscsi);
1990 }
1991 iscsi_destroy_context(iscsi);
1992 g_free(iscsilun->zeroblock);
1993 iscsi_allocmap_free(iscsilun);
1994 qemu_mutex_destroy(&iscsilun->mutex);
1995 memset(iscsilun, 0, sizeof(IscsiLun));
1996 }
1997
1998 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
1999 {
2000 /* We don't actually refresh here, but just return data queried in
2001 * iscsi_open(): iscsi targets don't change their limits. */
2002
2003 IscsiLun *iscsilun = bs->opaque;
2004 uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
2005 unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size);
2006
2007 assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bs->sg);
2008
2009 bs->bl.request_alignment = block_size;
2010
2011 if (iscsilun->bl.max_xfer_len) {
2012 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
2013 }
2014
2015 if (max_xfer_len * block_size < INT_MAX) {
2016 bs->bl.max_transfer = max_xfer_len * iscsilun->block_size;
2017 }
2018
2019 if (iscsilun->lbp.lbpu) {
2020 if (iscsilun->bl.max_unmap < 0xffffffff / block_size) {
2021 bs->bl.max_pdiscard =
2022 iscsilun->bl.max_unmap * iscsilun->block_size;
2023 }
2024 bs->bl.pdiscard_alignment =
2025 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2026 } else {
2027 bs->bl.pdiscard_alignment = iscsilun->block_size;
2028 }
2029
2030 if (iscsilun->bl.max_ws_len < 0xffffffff / block_size) {
2031 bs->bl.max_pwrite_zeroes =
2032 iscsilun->bl.max_ws_len * iscsilun->block_size;
2033 }
2034 if (iscsilun->lbp.lbpws) {
2035 bs->bl.pwrite_zeroes_alignment =
2036 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2037 } else {
2038 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size;
2039 }
2040 if (iscsilun->bl.opt_xfer_len &&
2041 iscsilun->bl.opt_xfer_len < INT_MAX / block_size) {
2042 bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len *
2043 iscsilun->block_size);
2044 }
2045 }
2046
2047 /* Note that this will not re-establish a connection with an iSCSI target - it
2048 * is effectively a NOP. */
2049 static int iscsi_reopen_prepare(BDRVReopenState *state,
2050 BlockReopenQueue *queue, Error **errp)
2051 {
2052 IscsiLun *iscsilun = state->bs->opaque;
2053
2054 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
2055 error_setg(errp, "Cannot open a write protected LUN as read-write");
2056 return -EACCES;
2057 }
2058 return 0;
2059 }
2060
2061 static void iscsi_reopen_commit(BDRVReopenState *reopen_state)
2062 {
2063 IscsiLun *iscsilun = reopen_state->bs->opaque;
2064
2065 /* the cache.direct status might have changed */
2066 if (iscsilun->allocmap != NULL) {
2067 iscsi_allocmap_init(iscsilun, reopen_state->flags);
2068 }
2069 }
2070
2071 static int iscsi_truncate(BlockDriverState *bs, int64_t offset,
2072 PreallocMode prealloc, Error **errp)
2073 {
2074 IscsiLun *iscsilun = bs->opaque;
2075 Error *local_err = NULL;
2076
2077 if (prealloc != PREALLOC_MODE_OFF) {
2078 error_setg(errp, "Unsupported preallocation mode '%s'",
2079 PreallocMode_str(prealloc));
2080 return -ENOTSUP;
2081 }
2082
2083 if (iscsilun->type != TYPE_DISK) {
2084 error_setg(errp, "Cannot resize non-disk iSCSI devices");
2085 return -ENOTSUP;
2086 }
2087
2088 iscsi_readcapacity_sync(iscsilun, &local_err);
2089 if (local_err != NULL) {
2090 error_propagate(errp, local_err);
2091 return -EIO;
2092 }
2093
2094 if (offset > iscsi_getlength(bs)) {
2095 error_setg(errp, "Cannot grow iSCSI devices");
2096 return -EINVAL;
2097 }
2098
2099 if (iscsilun->allocmap != NULL) {
2100 iscsi_allocmap_init(iscsilun, bs->open_flags);
2101 }
2102
2103 return 0;
2104 }
2105
2106 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp)
2107 {
2108 int ret = 0;
2109 int64_t total_size = 0;
2110 BlockDriverState *bs;
2111 IscsiLun *iscsilun = NULL;
2112 QDict *bs_options;
2113 Error *local_err = NULL;
2114
2115 bs = bdrv_new();
2116
2117 /* Read out options */
2118 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2119 BDRV_SECTOR_SIZE);
2120 bs->opaque = g_new0(struct IscsiLun, 1);
2121 iscsilun = bs->opaque;
2122
2123 bs_options = qdict_new();
2124 iscsi_parse_filename(filename, bs_options, &local_err);
2125 if (local_err) {
2126 error_propagate(errp, local_err);
2127 ret = -EINVAL;
2128 } else {
2129 ret = iscsi_open(bs, bs_options, 0, NULL);
2130 }
2131 QDECREF(bs_options);
2132
2133 if (ret != 0) {
2134 goto out;
2135 }
2136 iscsi_detach_aio_context(bs);
2137 if (iscsilun->type != TYPE_DISK) {
2138 ret = -ENODEV;
2139 goto out;
2140 }
2141 if (bs->total_sectors < total_size) {
2142 ret = -ENOSPC;
2143 goto out;
2144 }
2145
2146 ret = 0;
2147 out:
2148 if (iscsilun->iscsi != NULL) {
2149 iscsi_destroy_context(iscsilun->iscsi);
2150 }
2151 g_free(bs->opaque);
2152 bs->opaque = NULL;
2153 bdrv_unref(bs);
2154 return ret;
2155 }
2156
2157 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2158 {
2159 IscsiLun *iscsilun = bs->opaque;
2160 bdi->unallocated_blocks_are_zero = iscsilun->lbprz;
2161 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
2162 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE;
2163 return 0;
2164 }
2165
2166 static void iscsi_invalidate_cache(BlockDriverState *bs,
2167 Error **errp)
2168 {
2169 IscsiLun *iscsilun = bs->opaque;
2170 iscsi_allocmap_invalidate(iscsilun);
2171 }
2172
2173 static QemuOptsList iscsi_create_opts = {
2174 .name = "iscsi-create-opts",
2175 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
2176 .desc = {
2177 {
2178 .name = BLOCK_OPT_SIZE,
2179 .type = QEMU_OPT_SIZE,
2180 .help = "Virtual disk size"
2181 },
2182 { /* end of list */ }
2183 }
2184 };
2185
2186 static BlockDriver bdrv_iscsi = {
2187 .format_name = "iscsi",
2188 .protocol_name = "iscsi",
2189
2190 .instance_size = sizeof(IscsiLun),
2191 .bdrv_parse_filename = iscsi_parse_filename,
2192 .bdrv_file_open = iscsi_open,
2193 .bdrv_close = iscsi_close,
2194 .bdrv_create = iscsi_create,
2195 .create_opts = &iscsi_create_opts,
2196 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2197 .bdrv_reopen_commit = iscsi_reopen_commit,
2198 .bdrv_invalidate_cache = iscsi_invalidate_cache,
2199
2200 .bdrv_getlength = iscsi_getlength,
2201 .bdrv_get_info = iscsi_get_info,
2202 .bdrv_truncate = iscsi_truncate,
2203 .bdrv_refresh_limits = iscsi_refresh_limits,
2204
2205 .bdrv_co_get_block_status = iscsi_co_get_block_status,
2206 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2207 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2208 .bdrv_co_readv = iscsi_co_readv,
2209 .bdrv_co_writev_flags = iscsi_co_writev_flags,
2210 .bdrv_co_flush_to_disk = iscsi_co_flush,
2211
2212 #ifdef __linux__
2213 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2214 #endif
2215
2216 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2217 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2218 };
2219
2220 #if LIBISCSI_API_VERSION >= (20160603)
2221 static BlockDriver bdrv_iser = {
2222 .format_name = "iser",
2223 .protocol_name = "iser",
2224
2225 .instance_size = sizeof(IscsiLun),
2226 .bdrv_parse_filename = iscsi_parse_filename,
2227 .bdrv_file_open = iscsi_open,
2228 .bdrv_close = iscsi_close,
2229 .bdrv_create = iscsi_create,
2230 .create_opts = &iscsi_create_opts,
2231 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2232 .bdrv_reopen_commit = iscsi_reopen_commit,
2233 .bdrv_invalidate_cache = iscsi_invalidate_cache,
2234
2235 .bdrv_getlength = iscsi_getlength,
2236 .bdrv_get_info = iscsi_get_info,
2237 .bdrv_truncate = iscsi_truncate,
2238 .bdrv_refresh_limits = iscsi_refresh_limits,
2239
2240 .bdrv_co_get_block_status = iscsi_co_get_block_status,
2241 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2242 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2243 .bdrv_co_readv = iscsi_co_readv,
2244 .bdrv_co_writev_flags = iscsi_co_writev_flags,
2245 .bdrv_co_flush_to_disk = iscsi_co_flush,
2246
2247 #ifdef __linux__
2248 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2249 #endif
2250
2251 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2252 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2253 };
2254 #endif
2255
2256 static void iscsi_block_init(void)
2257 {
2258 bdrv_register(&bdrv_iscsi);
2259 #if LIBISCSI_API_VERSION >= (20160603)
2260 bdrv_register(&bdrv_iser);
2261 #endif
2262 }
2263
2264 block_init(iscsi_block_init);
AR7 emulation for QEMU