?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
migration: print features as on off
[qemu/kevin.git] / block / iscsi.c
blob5c0a9e55b6d99ed7227341291932617eb7dc2ec2
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 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
662 ret = -EINVAL;
663 goto out;
664 }
665
666 /* default to all sectors allocated */
667 ret = BDRV_BLOCK_DATA;
668 ret |= (sector_num << BDRV_SECTOR_BITS) | BDRV_BLOCK_OFFSET_VALID;
669 *pnum = nb_sectors;
670
671 /* LUN does not support logical block provisioning */
672 if (!iscsilun->lbpme) {
673 goto out;
674 }
675
676 lba = sector_qemu2lun(sector_num, iscsilun);
677
678 iscsi_co_init_iscsitask(iscsilun, &iTask);
679 qemu_mutex_lock(&iscsilun->mutex);
680 retry:
681 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
682 lba, 8 + 16, iscsi_co_generic_cb,
683 &iTask) == NULL) {
684 ret = -ENOMEM;
685 goto out_unlock;
686 }
687
688 while (!iTask.complete) {
689 iscsi_set_events(iscsilun);
690 qemu_mutex_unlock(&iscsilun->mutex);
691 qemu_coroutine_yield();
692 qemu_mutex_lock(&iscsilun->mutex);
693 }
694
695 if (iTask.do_retry) {
696 if (iTask.task != NULL) {
697 scsi_free_scsi_task(iTask.task);
698 iTask.task = NULL;
699 }
700 iTask.complete = 0;
701 goto retry;
702 }
703
704 if (iTask.status != SCSI_STATUS_GOOD) {
705 /* in case the get_lba_status_callout fails (i.e.
706 * because the device is busy or the cmd is not
707 * supported) we pretend all blocks are allocated
708 * for backwards compatibility */
709 error_report("iSCSI GET_LBA_STATUS failed at lba %" PRIu64 ": %s",
710 lba, iTask.err_str);
711 goto out_unlock;
712 }
713
714 lbas = scsi_datain_unmarshall(iTask.task);
715 if (lbas == NULL) {
716 ret = -EIO;
717 goto out_unlock;
718 }
719
720 lbasd = &lbas->descriptors[0];
721
722 if (sector_qemu2lun(sector_num, iscsilun) != lbasd->lba) {
723 ret = -EIO;
724 goto out_unlock;
725 }
726
727 *pnum = sector_lun2qemu(lbasd->num_blocks, iscsilun);
728
729 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
730 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
731 ret &= ~BDRV_BLOCK_DATA;
732 if (iscsilun->lbprz) {
733 ret |= BDRV_BLOCK_ZERO;
734 }
735 }
736
737 if (ret & BDRV_BLOCK_ZERO) {
738 iscsi_allocmap_set_unallocated(iscsilun, sector_num, *pnum);
739 } else {
740 iscsi_allocmap_set_allocated(iscsilun, sector_num, *pnum);
741 }
742
743 if (*pnum > nb_sectors) {
744 *pnum = nb_sectors;
745 }
746 out_unlock:
747 qemu_mutex_unlock(&iscsilun->mutex);
748 g_free(iTask.err_str);
749 out:
750 if (iTask.task != NULL) {
751 scsi_free_scsi_task(iTask.task);
752 }
753 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID) {
754 *file = bs;
755 }
756 return ret;
757 }
758
759 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
760 int64_t sector_num, int nb_sectors,
761 QEMUIOVector *iov)
762 {
763 IscsiLun *iscsilun = bs->opaque;
764 struct IscsiTask iTask;
765 uint64_t lba;
766 uint32_t num_sectors;
767 int r = 0;
768
769 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
770 return -EINVAL;
771 }
772
773 if (bs->bl.max_transfer) {
774 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
775 }
776
777 /* if cache.direct is off and we have a valid entry in our allocation map
778 * we can skip checking the block status and directly return zeroes if
779 * the request falls within an unallocated area */
780 if (iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) &&
781 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
782 qemu_iovec_memset(iov, 0, 0x00, iov->size);
783 return 0;
784 }
785
786 if (nb_sectors >= ISCSI_CHECKALLOC_THRES &&
787 !iscsi_allocmap_is_valid(iscsilun, sector_num, nb_sectors) &&
788 !iscsi_allocmap_is_allocated(iscsilun, sector_num, nb_sectors)) {
789 int pnum;
790 BlockDriverState *file;
791 /* check the block status from the beginning of the cluster
792 * containing the start sector */
793 int64_t ret = iscsi_co_get_block_status(bs,
794 sector_num - sector_num % iscsilun->cluster_sectors,
795 BDRV_REQUEST_MAX_SECTORS, &pnum, &file);
796 if (ret < 0) {
797 return ret;
798 }
799 /* if the whole request falls into an unallocated area we can avoid
800 * to read and directly return zeroes instead */
801 if (ret & BDRV_BLOCK_ZERO &&
802 pnum >= nb_sectors + sector_num % iscsilun->cluster_sectors) {
803 qemu_iovec_memset(iov, 0, 0x00, iov->size);
804 return 0;
805 }
806 }
807
808 lba = sector_qemu2lun(sector_num, iscsilun);
809 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
810
811 iscsi_co_init_iscsitask(iscsilun, &iTask);
812 qemu_mutex_lock(&iscsilun->mutex);
813 retry:
814 if (iscsilun->use_16_for_rw) {
815 #if LIBISCSI_API_VERSION >= (20160603)
816 iTask.task = iscsi_read16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
817 num_sectors * iscsilun->block_size,
818 iscsilun->block_size, 0, 0, 0, 0, 0,
819 iscsi_co_generic_cb, &iTask,
820 (struct scsi_iovec *)iov->iov, iov->niov);
821 } else {
822 iTask.task = iscsi_read10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
823 num_sectors * iscsilun->block_size,
824 iscsilun->block_size,
825 0, 0, 0, 0, 0,
826 iscsi_co_generic_cb, &iTask,
827 (struct scsi_iovec *)iov->iov, iov->niov);
828 }
829 #else
830 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
831 num_sectors * iscsilun->block_size,
832 iscsilun->block_size, 0, 0, 0, 0, 0,
833 iscsi_co_generic_cb, &iTask);
834 } else {
835 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
836 num_sectors * iscsilun->block_size,
837 iscsilun->block_size,
838 0, 0, 0, 0, 0,
839 iscsi_co_generic_cb, &iTask);
840 }
841 #endif
842 if (iTask.task == NULL) {
843 qemu_mutex_unlock(&iscsilun->mutex);
844 return -ENOMEM;
845 }
846 #if LIBISCSI_API_VERSION < (20160603)
847 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
848 #endif
849 while (!iTask.complete) {
850 iscsi_set_events(iscsilun);
851 qemu_mutex_unlock(&iscsilun->mutex);
852 qemu_coroutine_yield();
853 qemu_mutex_lock(&iscsilun->mutex);
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 error_report("iSCSI READ10/16 failed at lba %" PRIu64 ": %s",
868 lba, iTask.err_str);
869 r = iTask.err_code;
870 }
871
872 qemu_mutex_unlock(&iscsilun->mutex);
873 g_free(iTask.err_str);
874 return r;
875 }
876
877 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
878 {
879 IscsiLun *iscsilun = bs->opaque;
880 struct IscsiTask iTask;
881 int r = 0;
882
883 iscsi_co_init_iscsitask(iscsilun, &iTask);
884 qemu_mutex_lock(&iscsilun->mutex);
885 retry:
886 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
887 0, iscsi_co_generic_cb, &iTask) == NULL) {
888 qemu_mutex_unlock(&iscsilun->mutex);
889 return -ENOMEM;
890 }
891
892 while (!iTask.complete) {
893 iscsi_set_events(iscsilun);
894 qemu_mutex_unlock(&iscsilun->mutex);
895 qemu_coroutine_yield();
896 qemu_mutex_lock(&iscsilun->mutex);
897 }
898
899 if (iTask.task != NULL) {
900 scsi_free_scsi_task(iTask.task);
901 iTask.task = NULL;
902 }
903
904 if (iTask.do_retry) {
905 iTask.complete = 0;
906 goto retry;
907 }
908
909 if (iTask.status != SCSI_STATUS_GOOD) {
910 error_report("iSCSI SYNCHRONIZECACHE10 failed: %s", iTask.err_str);
911 r = iTask.err_code;
912 }
913
914 qemu_mutex_unlock(&iscsilun->mutex);
915 g_free(iTask.err_str);
916 return r;
917 }
918
919 #ifdef __linux__
920 /* Called (via iscsi_service) with QemuMutex held. */
921 static void
922 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
923 void *command_data, void *opaque)
924 {
925 IscsiAIOCB *acb = opaque;
926
927 g_free(acb->buf);
928 acb->buf = NULL;
929
930 acb->status = 0;
931 if (status < 0) {
932 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
933 iscsi_get_error(iscsi));
934 acb->status = iscsi_translate_sense(&acb->task->sense);
935 }
936
937 acb->ioh->driver_status = 0;
938 acb->ioh->host_status = 0;
939 acb->ioh->resid = 0;
940 acb->ioh->status = status;
941
942 #define SG_ERR_DRIVER_SENSE 0x08
943
944 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
945 int ss;
946
947 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
948
949 acb->ioh->sb_len_wr = acb->task->datain.size - 2;
950 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
951 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
952 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
953 }
954
955 iscsi_schedule_bh(acb);
956 }
957
958 static void iscsi_ioctl_bh_completion(void *opaque)
959 {
960 IscsiAIOCB *acb = opaque;
961
962 qemu_bh_delete(acb->bh);
963 acb->common.cb(acb->common.opaque, acb->ret);
964 qemu_aio_unref(acb);
965 }
966
967 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
968 {
969 BlockDriverState *bs = acb->common.bs;
970 IscsiLun *iscsilun = bs->opaque;
971 int ret = 0;
972
973 switch (req) {
974 case SG_GET_VERSION_NUM:
975 *(int *)buf = 30000;
976 break;
977 case SG_GET_SCSI_ID:
978 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
979 break;
980 default:
981 ret = -EINVAL;
982 }
983 assert(!acb->bh);
984 acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
985 iscsi_ioctl_bh_completion, acb);
986 acb->ret = ret;
987 qemu_bh_schedule(acb->bh);
988 }
989
990 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
991 unsigned long int req, void *buf,
992 BlockCompletionFunc *cb, void *opaque)
993 {
994 IscsiLun *iscsilun = bs->opaque;
995 struct iscsi_context *iscsi = iscsilun->iscsi;
996 struct iscsi_data data;
997 IscsiAIOCB *acb;
998
999 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
1000
1001 acb->iscsilun = iscsilun;
1002 acb->bh = NULL;
1003 acb->status = -EINPROGRESS;
1004 acb->buf = NULL;
1005 acb->ioh = buf;
1006
1007 if (req != SG_IO) {
1008 iscsi_ioctl_handle_emulated(acb, req, buf);
1009 return &acb->common;
1010 }
1011
1012 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) {
1013 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)",
1014 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE);
1015 qemu_aio_unref(acb);
1016 return NULL;
1017 }
1018
1019 acb->task = malloc(sizeof(struct scsi_task));
1020 if (acb->task == NULL) {
1021 error_report("iSCSI: Failed to allocate task for scsi command. %s",
1022 iscsi_get_error(iscsi));
1023 qemu_aio_unref(acb);
1024 return NULL;
1025 }
1026 memset(acb->task, 0, sizeof(struct scsi_task));
1027
1028 switch (acb->ioh->dxfer_direction) {
1029 case SG_DXFER_TO_DEV:
1030 acb->task->xfer_dir = SCSI_XFER_WRITE;
1031 break;
1032 case SG_DXFER_FROM_DEV:
1033 acb->task->xfer_dir = SCSI_XFER_READ;
1034 break;
1035 default:
1036 acb->task->xfer_dir = SCSI_XFER_NONE;
1037 break;
1038 }
1039
1040 acb->task->cdb_size = acb->ioh->cmd_len;
1041 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
1042 acb->task->expxferlen = acb->ioh->dxfer_len;
1043
1044 data.size = 0;
1045 qemu_mutex_lock(&iscsilun->mutex);
1046 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
1047 if (acb->ioh->iovec_count == 0) {
1048 data.data = acb->ioh->dxferp;
1049 data.size = acb->ioh->dxfer_len;
1050 } else {
1051 scsi_task_set_iov_out(acb->task,
1052 (struct scsi_iovec *) acb->ioh->dxferp,
1053 acb->ioh->iovec_count);
1054 }
1055 }
1056
1057 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
1058 iscsi_aio_ioctl_cb,
1059 (data.size > 0) ? &data : NULL,
1060 acb) != 0) {
1061 qemu_mutex_unlock(&iscsilun->mutex);
1062 scsi_free_scsi_task(acb->task);
1063 qemu_aio_unref(acb);
1064 return NULL;
1065 }
1066
1067 /* tell libiscsi to read straight into the buffer we got from ioctl */
1068 if (acb->task->xfer_dir == SCSI_XFER_READ) {
1069 if (acb->ioh->iovec_count == 0) {
1070 scsi_task_add_data_in_buffer(acb->task,
1071 acb->ioh->dxfer_len,
1072 acb->ioh->dxferp);
1073 } else {
1074 scsi_task_set_iov_in(acb->task,
1075 (struct scsi_iovec *) acb->ioh->dxferp,
1076 acb->ioh->iovec_count);
1077 }
1078 }
1079
1080 iscsi_set_events(iscsilun);
1081 qemu_mutex_unlock(&iscsilun->mutex);
1082
1083 return &acb->common;
1084 }
1085
1086 #endif
1087
1088 static int64_t
1089 iscsi_getlength(BlockDriverState *bs)
1090 {
1091 IscsiLun *iscsilun = bs->opaque;
1092 int64_t len;
1093
1094 len = iscsilun->num_blocks;
1095 len *= iscsilun->block_size;
1096
1097 return len;
1098 }
1099
1100 static int
1101 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
1102 {
1103 IscsiLun *iscsilun = bs->opaque;
1104 struct IscsiTask iTask;
1105 struct unmap_list list;
1106 int r = 0;
1107
1108 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) {
1109 return -ENOTSUP;
1110 }
1111
1112 if (!iscsilun->lbp.lbpu) {
1113 /* UNMAP is not supported by the target */
1114 return 0;
1115 }
1116
1117 list.lba = offset / iscsilun->block_size;
1118 list.num = bytes / iscsilun->block_size;
1119
1120 iscsi_co_init_iscsitask(iscsilun, &iTask);
1121 qemu_mutex_lock(&iscsilun->mutex);
1122 retry:
1123 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
1124 iscsi_co_generic_cb, &iTask) == NULL) {
1125 r = -ENOMEM;
1126 goto out_unlock;
1127 }
1128
1129 while (!iTask.complete) {
1130 iscsi_set_events(iscsilun);
1131 qemu_mutex_unlock(&iscsilun->mutex);
1132 qemu_coroutine_yield();
1133 qemu_mutex_lock(&iscsilun->mutex);
1134 }
1135
1136 if (iTask.task != NULL) {
1137 scsi_free_scsi_task(iTask.task);
1138 iTask.task = NULL;
1139 }
1140
1141 if (iTask.do_retry) {
1142 iTask.complete = 0;
1143 goto retry;
1144 }
1145
1146 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1147 bytes >> BDRV_SECTOR_BITS);
1148
1149 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
1150 /* the target might fail with a check condition if it
1151 is not happy with the alignment of the UNMAP request
1152 we silently fail in this case */
1153 goto out_unlock;
1154 }
1155
1156 if (iTask.status != SCSI_STATUS_GOOD) {
1157 error_report("iSCSI UNMAP failed at lba %" PRIu64 ": %s",
1158 list.lba, iTask.err_str);
1159 r = iTask.err_code;
1160 goto out_unlock;
1161 }
1162
1163 out_unlock:
1164 qemu_mutex_unlock(&iscsilun->mutex);
1165 g_free(iTask.err_str);
1166 return r;
1167 }
1168
1169 static int
1170 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1171 int bytes, BdrvRequestFlags flags)
1172 {
1173 IscsiLun *iscsilun = bs->opaque;
1174 struct IscsiTask iTask;
1175 uint64_t lba;
1176 uint32_t nb_blocks;
1177 bool use_16_for_ws = iscsilun->use_16_for_rw;
1178 int r = 0;
1179
1180 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) {
1181 return -ENOTSUP;
1182 }
1183
1184 if (flags & BDRV_REQ_MAY_UNMAP) {
1185 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
1186 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
1187 use_16_for_ws = true;
1188 }
1189 if (use_16_for_ws && !iscsilun->lbp.lbpws) {
1190 /* WRITESAME16 with UNMAP is not supported by the target,
1191 * fall back and try WRITESAME10/16 without UNMAP */
1192 flags &= ~BDRV_REQ_MAY_UNMAP;
1193 use_16_for_ws = iscsilun->use_16_for_rw;
1194 }
1195 }
1196
1197 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1198 /* WRITESAME without UNMAP is not supported by the target */
1199 return -ENOTSUP;
1200 }
1201
1202 lba = offset / iscsilun->block_size;
1203 nb_blocks = bytes / iscsilun->block_size;
1204
1205 if (iscsilun->zeroblock == NULL) {
1206 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1207 if (iscsilun->zeroblock == NULL) {
1208 return -ENOMEM;
1209 }
1210 }
1211
1212 qemu_mutex_lock(&iscsilun->mutex);
1213 iscsi_co_init_iscsitask(iscsilun, &iTask);
1214 retry:
1215 if (use_16_for_ws) {
1216 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1217 iscsilun->zeroblock, iscsilun->block_size,
1218 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1219 0, 0, iscsi_co_generic_cb, &iTask);
1220 } else {
1221 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1222 iscsilun->zeroblock, iscsilun->block_size,
1223 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1224 0, 0, iscsi_co_generic_cb, &iTask);
1225 }
1226 if (iTask.task == NULL) {
1227 qemu_mutex_unlock(&iscsilun->mutex);
1228 return -ENOMEM;
1229 }
1230
1231 while (!iTask.complete) {
1232 iscsi_set_events(iscsilun);
1233 qemu_mutex_unlock(&iscsilun->mutex);
1234 qemu_coroutine_yield();
1235 qemu_mutex_lock(&iscsilun->mutex);
1236 }
1237
1238 if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1239 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1240 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1241 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1242 /* WRITE SAME is not supported by the target */
1243 iscsilun->has_write_same = false;
1244 scsi_free_scsi_task(iTask.task);
1245 r = -ENOTSUP;
1246 goto out_unlock;
1247 }
1248
1249 if (iTask.task != NULL) {
1250 scsi_free_scsi_task(iTask.task);
1251 iTask.task = NULL;
1252 }
1253
1254 if (iTask.do_retry) {
1255 iTask.complete = 0;
1256 goto retry;
1257 }
1258
1259 if (iTask.status != SCSI_STATUS_GOOD) {
1260 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1261 bytes >> BDRV_SECTOR_BITS);
1262 error_report("iSCSI WRITESAME10/16 failed at lba %" PRIu64 ": %s",
1263 lba, iTask.err_str);
1264 r = iTask.err_code;
1265 goto out_unlock;
1266 }
1267
1268 if (flags & BDRV_REQ_MAY_UNMAP) {
1269 iscsi_allocmap_set_invalid(iscsilun, offset >> BDRV_SECTOR_BITS,
1270 bytes >> BDRV_SECTOR_BITS);
1271 } else {
1272 iscsi_allocmap_set_allocated(iscsilun, offset >> BDRV_SECTOR_BITS,
1273 bytes >> BDRV_SECTOR_BITS);
1274 }
1275
1276 out_unlock:
1277 qemu_mutex_unlock(&iscsilun->mutex);
1278 g_free(iTask.err_str);
1279 return r;
1280 }
1281
1282 static void apply_chap(struct iscsi_context *iscsi, QemuOpts *opts,
1283 Error **errp)
1284 {
1285 const char *user = NULL;
1286 const char *password = NULL;
1287 const char *secretid;
1288 char *secret = NULL;
1289
1290 user = qemu_opt_get(opts, "user");
1291 if (!user) {
1292 return;
1293 }
1294
1295 secretid = qemu_opt_get(opts, "password-secret");
1296 password = qemu_opt_get(opts, "password");
1297 if (secretid && password) {
1298 error_setg(errp, "'password' and 'password-secret' properties are "
1299 "mutually exclusive");
1300 return;
1301 }
1302 if (secretid) {
1303 secret = qcrypto_secret_lookup_as_utf8(secretid, errp);
1304 if (!secret) {
1305 return;
1306 }
1307 password = secret;
1308 } else if (!password) {
1309 error_setg(errp, "CHAP username specified but no password was given");
1310 return;
1311 }
1312
1313 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1314 error_setg(errp, "Failed to set initiator username and password");
1315 }
1316
1317 g_free(secret);
1318 }
1319
1320 static void apply_header_digest(struct iscsi_context *iscsi, QemuOpts *opts,
1321 Error **errp)
1322 {
1323 const char *digest = NULL;
1324
1325 digest = qemu_opt_get(opts, "header-digest");
1326 if (!digest) {
1327 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1328 } else if (!strcmp(digest, "crc32c")) {
1329 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1330 } else if (!strcmp(digest, "none")) {
1331 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1332 } else if (!strcmp(digest, "crc32c-none")) {
1333 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1334 } else if (!strcmp(digest, "none-crc32c")) {
1335 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1336 } else {
1337 error_setg(errp, "Invalid header-digest setting : %s", digest);
1338 }
1339 }
1340
1341 static char *get_initiator_name(QemuOpts *opts)
1342 {
1343 const char *name;
1344 char *iscsi_name;
1345 UuidInfo *uuid_info;
1346
1347 name = qemu_opt_get(opts, "initiator-name");
1348 if (name) {
1349 return g_strdup(name);
1350 }
1351
1352 uuid_info = qmp_query_uuid(NULL);
1353 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1354 name = qemu_get_vm_name();
1355 } else {
1356 name = uuid_info->UUID;
1357 }
1358 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1359 name ? ":" : "", name ? name : "");
1360 qapi_free_UuidInfo(uuid_info);
1361 return iscsi_name;
1362 }
1363
1364 static void iscsi_nop_timed_event(void *opaque)
1365 {
1366 IscsiLun *iscsilun = opaque;
1367
1368 qemu_mutex_lock(&iscsilun->mutex);
1369 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1370 error_report("iSCSI: NOP timeout. Reconnecting...");
1371 iscsilun->request_timed_out = true;
1372 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1373 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1374 goto out;
1375 }
1376
1377 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1378 iscsi_set_events(iscsilun);
1379
1380 out:
1381 qemu_mutex_unlock(&iscsilun->mutex);
1382 }
1383
1384 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1385 {
1386 struct scsi_task *task = NULL;
1387 struct scsi_readcapacity10 *rc10 = NULL;
1388 struct scsi_readcapacity16 *rc16 = NULL;
1389 int retries = ISCSI_CMD_RETRIES;
1390
1391 do {
1392 if (task != NULL) {
1393 scsi_free_scsi_task(task);
1394 task = NULL;
1395 }
1396
1397 switch (iscsilun->type) {
1398 case TYPE_DISK:
1399 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1400 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1401 rc16 = scsi_datain_unmarshall(task);
1402 if (rc16 == NULL) {
1403 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1404 } else {
1405 iscsilun->block_size = rc16->block_length;
1406 iscsilun->num_blocks = rc16->returned_lba + 1;
1407 iscsilun->lbpme = !!rc16->lbpme;
1408 iscsilun->lbprz = !!rc16->lbprz;
1409 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1410 }
1411 break;
1412 }
1413 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1414 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1415 break;
1416 }
1417 /* Fall through and try READ CAPACITY(10) instead. */
1418 case TYPE_ROM:
1419 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1420 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1421 rc10 = scsi_datain_unmarshall(task);
1422 if (rc10 == NULL) {
1423 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1424 } else {
1425 iscsilun->block_size = rc10->block_size;
1426 if (rc10->lba == 0) {
1427 /* blank disk loaded */
1428 iscsilun->num_blocks = 0;
1429 } else {
1430 iscsilun->num_blocks = rc10->lba + 1;
1431 }
1432 }
1433 }
1434 break;
1435 default:
1436 return;
1437 }
1438 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1439 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1440 && retries-- > 0);
1441
1442 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1443 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1444 } else if (!iscsilun->block_size ||
1445 iscsilun->block_size % BDRV_SECTOR_SIZE) {
1446 error_setg(errp, "iSCSI: the target returned an invalid "
1447 "block size of %d.", iscsilun->block_size);
1448 }
1449 if (task) {
1450 scsi_free_scsi_task(task);
1451 }
1452 }
1453
1454 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1455 int evpd, int pc, void **inq, Error **errp)
1456 {
1457 int full_size;
1458 struct scsi_task *task = NULL;
1459 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1460 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1461 goto fail;
1462 }
1463 full_size = scsi_datain_getfullsize(task);
1464 if (full_size > task->datain.size) {
1465 scsi_free_scsi_task(task);
1466
1467 /* we need more data for the full list */
1468 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1469 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1470 goto fail;
1471 }
1472 }
1473
1474 *inq = scsi_datain_unmarshall(task);
1475 if (*inq == NULL) {
1476 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1477 goto fail_with_err;
1478 }
1479
1480 return task;
1481
1482 fail:
1483 error_setg(errp, "iSCSI: Inquiry command failed : %s",
1484 iscsi_get_error(iscsi));
1485 fail_with_err:
1486 if (task != NULL) {
1487 scsi_free_scsi_task(task);
1488 }
1489 return NULL;
1490 }
1491
1492 static void iscsi_detach_aio_context(BlockDriverState *bs)
1493 {
1494 IscsiLun *iscsilun = bs->opaque;
1495
1496 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1497 false, NULL, NULL, NULL, NULL);
1498 iscsilun->events = 0;
1499
1500 if (iscsilun->nop_timer) {
1501 timer_del(iscsilun->nop_timer);
1502 timer_free(iscsilun->nop_timer);
1503 iscsilun->nop_timer = NULL;
1504 }
1505 if (iscsilun->event_timer) {
1506 timer_del(iscsilun->event_timer);
1507 timer_free(iscsilun->event_timer);
1508 iscsilun->event_timer = NULL;
1509 }
1510 }
1511
1512 static void iscsi_attach_aio_context(BlockDriverState *bs,
1513 AioContext *new_context)
1514 {
1515 IscsiLun *iscsilun = bs->opaque;
1516
1517 iscsilun->aio_context = new_context;
1518 iscsi_set_events(iscsilun);
1519
1520 /* Set up a timer for sending out iSCSI NOPs */
1521 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1522 QEMU_CLOCK_REALTIME, SCALE_MS,
1523 iscsi_nop_timed_event, iscsilun);
1524 timer_mod(iscsilun->nop_timer,
1525 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1526
1527 /* Set up a timer for periodic calls to iscsi_set_events and to
1528 * scan for command timeout */
1529 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1530 QEMU_CLOCK_REALTIME, SCALE_MS,
1531 iscsi_timed_check_events, iscsilun);
1532 timer_mod(iscsilun->event_timer,
1533 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1534 }
1535
1536 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1537 {
1538 struct scsi_task *task;
1539 struct scsi_mode_sense *ms = NULL;
1540 iscsilun->write_protected = false;
1541 iscsilun->dpofua = false;
1542
1543 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1544 1, SCSI_MODESENSE_PC_CURRENT,
1545 0x3F, 0, 255);
1546 if (task == NULL) {
1547 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1548 iscsi_get_error(iscsilun->iscsi));
1549 goto out;
1550 }
1551
1552 if (task->status != SCSI_STATUS_GOOD) {
1553 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1554 goto out;
1555 }
1556 ms = scsi_datain_unmarshall(task);
1557 if (!ms) {
1558 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1559 iscsi_get_error(iscsilun->iscsi));
1560 goto out;
1561 }
1562 iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1563 iscsilun->dpofua = ms->device_specific_parameter & 0x10;
1564
1565 out:
1566 if (task) {
1567 scsi_free_scsi_task(task);
1568 }
1569 }
1570
1571 static void iscsi_parse_iscsi_option(const char *target, QDict *options)
1572 {
1573 QemuOptsList *list;
1574 QemuOpts *opts;
1575 const char *user, *password, *password_secret, *initiator_name,
1576 *header_digest, *timeout;
1577
1578 list = qemu_find_opts("iscsi");
1579 if (!list) {
1580 return;
1581 }
1582
1583 opts = qemu_opts_find(list, target);
1584 if (opts == NULL) {
1585 opts = QTAILQ_FIRST(&list->head);
1586 if (!opts) {
1587 return;
1588 }
1589 }
1590
1591 user = qemu_opt_get(opts, "user");
1592 if (user) {
1593 qdict_set_default_str(options, "user", user);
1594 }
1595
1596 password = qemu_opt_get(opts, "password");
1597 if (password) {
1598 qdict_set_default_str(options, "password", password);
1599 }
1600
1601 password_secret = qemu_opt_get(opts, "password-secret");
1602 if (password_secret) {
1603 qdict_set_default_str(options, "password-secret", password_secret);
1604 }
1605
1606 initiator_name = qemu_opt_get(opts, "initiator-name");
1607 if (initiator_name) {
1608 qdict_set_default_str(options, "initiator-name", initiator_name);
1609 }
1610
1611 header_digest = qemu_opt_get(opts, "header-digest");
1612 if (header_digest) {
1613 /* -iscsi takes upper case values, but QAPI only supports lower case
1614 * enum constant names, so we have to convert here. */
1615 char *qapi_value = g_ascii_strdown(header_digest, -1);
1616 qdict_set_default_str(options, "header-digest", qapi_value);
1617 g_free(qapi_value);
1618 }
1619
1620 timeout = qemu_opt_get(opts, "timeout");
1621 if (timeout) {
1622 qdict_set_default_str(options, "timeout", timeout);
1623 }
1624 }
1625
1626 /*
1627 * We support iscsi url's on the form
1628 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1629 */
1630 static void iscsi_parse_filename(const char *filename, QDict *options,
1631 Error **errp)
1632 {
1633 struct iscsi_url *iscsi_url;
1634 const char *transport_name;
1635 char *lun_str;
1636
1637 iscsi_url = iscsi_parse_full_url(NULL, filename);
1638 if (iscsi_url == NULL) {
1639 error_setg(errp, "Failed to parse URL : %s", filename);
1640 return;
1641 }
1642
1643 #if LIBISCSI_API_VERSION >= (20160603)
1644 switch (iscsi_url->transport) {
1645 case TCP_TRANSPORT:
1646 transport_name = "tcp";
1647 break;
1648 case ISER_TRANSPORT:
1649 transport_name = "iser";
1650 break;
1651 default:
1652 error_setg(errp, "Unknown transport type (%d)",
1653 iscsi_url->transport);
1654 return;
1655 }
1656 #else
1657 transport_name = "tcp";
1658 #endif
1659
1660 qdict_set_default_str(options, "transport", transport_name);
1661 qdict_set_default_str(options, "portal", iscsi_url->portal);
1662 qdict_set_default_str(options, "target", iscsi_url->target);
1663
1664 lun_str = g_strdup_printf("%d", iscsi_url->lun);
1665 qdict_set_default_str(options, "lun", lun_str);
1666 g_free(lun_str);
1667
1668 /* User/password from -iscsi take precedence over those from the URL */
1669 iscsi_parse_iscsi_option(iscsi_url->target, options);
1670
1671 if (iscsi_url->user[0] != '\0') {
1672 qdict_set_default_str(options, "user", iscsi_url->user);
1673 qdict_set_default_str(options, "password", iscsi_url->passwd);
1674 }
1675
1676 iscsi_destroy_url(iscsi_url);
1677 }
1678
1679 static QemuOptsList runtime_opts = {
1680 .name = "iscsi",
1681 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1682 .desc = {
1683 {
1684 .name = "transport",
1685 .type = QEMU_OPT_STRING,
1686 },
1687 {
1688 .name = "portal",
1689 .type = QEMU_OPT_STRING,
1690 },
1691 {
1692 .name = "target",
1693 .type = QEMU_OPT_STRING,
1694 },
1695 {
1696 .name = "user",
1697 .type = QEMU_OPT_STRING,
1698 },
1699 {
1700 .name = "password",
1701 .type = QEMU_OPT_STRING,
1702 },
1703 {
1704 .name = "password-secret",
1705 .type = QEMU_OPT_STRING,
1706 },
1707 {
1708 .name = "lun",
1709 .type = QEMU_OPT_NUMBER,
1710 },
1711 {
1712 .name = "initiator-name",
1713 .type = QEMU_OPT_STRING,
1714 },
1715 {
1716 .name = "header-digest",
1717 .type = QEMU_OPT_STRING,
1718 },
1719 {
1720 .name = "timeout",
1721 .type = QEMU_OPT_NUMBER,
1722 },
1723 {
1724 .name = "filename",
1725 .type = QEMU_OPT_STRING,
1726 },
1727 { /* end of list */ }
1728 },
1729 };
1730
1731 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1732 Error **errp)
1733 {
1734 IscsiLun *iscsilun = bs->opaque;
1735 struct iscsi_context *iscsi = NULL;
1736 struct scsi_task *task = NULL;
1737 struct scsi_inquiry_standard *inq = NULL;
1738 struct scsi_inquiry_supported_pages *inq_vpd;
1739 char *initiator_name = NULL;
1740 QemuOpts *opts;
1741 Error *local_err = NULL;
1742 const char *transport_name, *portal, *target, *filename;
1743 #if LIBISCSI_API_VERSION >= (20160603)
1744 enum iscsi_transport_type transport;
1745 #endif
1746 int i, ret = 0, timeout = 0, lun;
1747
1748 /* If we are given a filename, parse the filename, with precedence given to
1749 * filename encoded options */
1750 filename = qdict_get_try_str(options, "filename");
1751 if (filename) {
1752 warn_report("'filename' option specified. "
1753 "This is an unsupported option, and may be deprecated "
1754 "in the future");
1755 iscsi_parse_filename(filename, options, &local_err);
1756 if (local_err) {
1757 ret = -EINVAL;
1758 error_propagate(errp, local_err);
1759 goto exit;
1760 }
1761 }
1762
1763 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1764 qemu_opts_absorb_qdict(opts, options, &local_err);
1765 if (local_err) {
1766 error_propagate(errp, local_err);
1767 ret = -EINVAL;
1768 goto out;
1769 }
1770
1771 transport_name = qemu_opt_get(opts, "transport");
1772 portal = qemu_opt_get(opts, "portal");
1773 target = qemu_opt_get(opts, "target");
1774 lun = qemu_opt_get_number(opts, "lun", 0);
1775
1776 if (!transport_name || !portal || !target) {
1777 error_setg(errp, "Need all of transport, portal and target options");
1778 ret = -EINVAL;
1779 goto out;
1780 }
1781
1782 if (!strcmp(transport_name, "tcp")) {
1783 #if LIBISCSI_API_VERSION >= (20160603)
1784 transport = TCP_TRANSPORT;
1785 } else if (!strcmp(transport_name, "iser")) {
1786 transport = ISER_TRANSPORT;
1787 #else
1788 /* TCP is what older libiscsi versions always use */
1789 #endif
1790 } else {
1791 error_setg(errp, "Unknown transport: %s", transport_name);
1792 ret = -EINVAL;
1793 goto out;
1794 }
1795
1796 memset(iscsilun, 0, sizeof(IscsiLun));
1797
1798 initiator_name = get_initiator_name(opts);
1799
1800 iscsi = iscsi_create_context(initiator_name);
1801 if (iscsi == NULL) {
1802 error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1803 ret = -ENOMEM;
1804 goto out;
1805 }
1806 #if LIBISCSI_API_VERSION >= (20160603)
1807 if (iscsi_init_transport(iscsi, transport)) {
1808 error_setg(errp, ("Error initializing transport."));
1809 ret = -EINVAL;
1810 goto out;
1811 }
1812 #endif
1813 if (iscsi_set_targetname(iscsi, target)) {
1814 error_setg(errp, "iSCSI: Failed to set target name.");
1815 ret = -EINVAL;
1816 goto out;
1817 }
1818
1819 /* check if we got CHAP username/password via the options */
1820 apply_chap(iscsi, opts, &local_err);
1821 if (local_err != NULL) {
1822 error_propagate(errp, local_err);
1823 ret = -EINVAL;
1824 goto out;
1825 }
1826
1827 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1828 error_setg(errp, "iSCSI: Failed to set session type to normal.");
1829 ret = -EINVAL;
1830 goto out;
1831 }
1832
1833 /* check if we got HEADER_DIGEST via the options */
1834 apply_header_digest(iscsi, opts, &local_err);
1835 if (local_err != NULL) {
1836 error_propagate(errp, local_err);
1837 ret = -EINVAL;
1838 goto out;
1839 }
1840
1841 /* timeout handling is broken in libiscsi before 1.15.0 */
1842 timeout = qemu_opt_get_number(opts, "timeout", 0);
1843 #if LIBISCSI_API_VERSION >= 20150621
1844 iscsi_set_timeout(iscsi, timeout);
1845 #else
1846 if (timeout) {
1847 error_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1848 }
1849 #endif
1850
1851 if (iscsi_full_connect_sync(iscsi, portal, lun) != 0) {
1852 error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1853 iscsi_get_error(iscsi));
1854 ret = -EINVAL;
1855 goto out;
1856 }
1857
1858 iscsilun->iscsi = iscsi;
1859 iscsilun->aio_context = bdrv_get_aio_context(bs);
1860 iscsilun->lun = lun;
1861 iscsilun->has_write_same = true;
1862
1863 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1864 (void **) &inq, errp);
1865 if (task == NULL) {
1866 ret = -EINVAL;
1867 goto out;
1868 }
1869 iscsilun->type = inq->periperal_device_type;
1870 scsi_free_scsi_task(task);
1871 task = NULL;
1872
1873 iscsi_modesense_sync(iscsilun);
1874 if (iscsilun->dpofua) {
1875 bs->supported_write_flags = BDRV_REQ_FUA;
1876 }
1877 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
1878
1879 /* Check the write protect flag of the LUN if we want to write */
1880 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1881 iscsilun->write_protected) {
1882 error_setg(errp, "Cannot open a write protected LUN as read-write");
1883 ret = -EACCES;
1884 goto out;
1885 }
1886
1887 iscsi_readcapacity_sync(iscsilun, &local_err);
1888 if (local_err != NULL) {
1889 error_propagate(errp, local_err);
1890 ret = -EINVAL;
1891 goto out;
1892 }
1893 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1894
1895 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1896 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1897 * will try to read from the device to guess the image format.
1898 */
1899 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1900 bs->sg = true;
1901 }
1902
1903 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1904 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1905 (void **) &inq_vpd, errp);
1906 if (task == NULL) {
1907 ret = -EINVAL;
1908 goto out;
1909 }
1910 for (i = 0; i < inq_vpd->num_pages; i++) {
1911 struct scsi_task *inq_task;
1912 struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1913 struct scsi_inquiry_block_limits *inq_bl;
1914 switch (inq_vpd->pages[i]) {
1915 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1916 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1917 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1918 (void **) &inq_lbp, errp);
1919 if (inq_task == NULL) {
1920 ret = -EINVAL;
1921 goto out;
1922 }
1923 memcpy(&iscsilun->lbp, inq_lbp,
1924 sizeof(struct scsi_inquiry_logical_block_provisioning));
1925 scsi_free_scsi_task(inq_task);
1926 break;
1927 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1928 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1929 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1930 (void **) &inq_bl, errp);
1931 if (inq_task == NULL) {
1932 ret = -EINVAL;
1933 goto out;
1934 }
1935 memcpy(&iscsilun->bl, inq_bl,
1936 sizeof(struct scsi_inquiry_block_limits));
1937 scsi_free_scsi_task(inq_task);
1938 break;
1939 default:
1940 break;
1941 }
1942 }
1943 scsi_free_scsi_task(task);
1944 task = NULL;
1945
1946 qemu_mutex_init(&iscsilun->mutex);
1947 iscsi_attach_aio_context(bs, iscsilun->aio_context);
1948
1949 /* Guess the internal cluster (page) size of the iscsi target by the means
1950 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
1951 * reasonable size */
1952 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
1953 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
1954 iscsilun->cluster_sectors = (iscsilun->bl.opt_unmap_gran *
1955 iscsilun->block_size) >> BDRV_SECTOR_BITS;
1956 if (iscsilun->lbprz) {
1957 ret = iscsi_allocmap_init(iscsilun, bs->open_flags);
1958 }
1959 }
1960
1961 out:
1962 qemu_opts_del(opts);
1963 g_free(initiator_name);
1964 if (task != NULL) {
1965 scsi_free_scsi_task(task);
1966 }
1967
1968 if (ret) {
1969 if (iscsi != NULL) {
1970 if (iscsi_is_logged_in(iscsi)) {
1971 iscsi_logout_sync(iscsi);
1972 }
1973 iscsi_destroy_context(iscsi);
1974 }
1975 memset(iscsilun, 0, sizeof(IscsiLun));
1976 }
1977 exit:
1978 return ret;
1979 }
1980
1981 static void iscsi_close(BlockDriverState *bs)
1982 {
1983 IscsiLun *iscsilun = bs->opaque;
1984 struct iscsi_context *iscsi = iscsilun->iscsi;
1985
1986 iscsi_detach_aio_context(bs);
1987 if (iscsi_is_logged_in(iscsi)) {
1988 iscsi_logout_sync(iscsi);
1989 }
1990 iscsi_destroy_context(iscsi);
1991 g_free(iscsilun->zeroblock);
1992 iscsi_allocmap_free(iscsilun);
1993 qemu_mutex_destroy(&iscsilun->mutex);
1994 memset(iscsilun, 0, sizeof(IscsiLun));
1995 }
1996
1997 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
1998 {
1999 /* We don't actually refresh here, but just return data queried in
2000 * iscsi_open(): iscsi targets don't change their limits. */
2001
2002 IscsiLun *iscsilun = bs->opaque;
2003 uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
2004 unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size);
2005
2006 assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bs->sg);
2007
2008 bs->bl.request_alignment = block_size;
2009
2010 if (iscsilun->bl.max_xfer_len) {
2011 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
2012 }
2013
2014 if (max_xfer_len * block_size < INT_MAX) {
2015 bs->bl.max_transfer = max_xfer_len * iscsilun->block_size;
2016 }
2017
2018 if (iscsilun->lbp.lbpu) {
2019 if (iscsilun->bl.max_unmap < 0xffffffff / block_size) {
2020 bs->bl.max_pdiscard =
2021 iscsilun->bl.max_unmap * iscsilun->block_size;
2022 }
2023 bs->bl.pdiscard_alignment =
2024 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2025 } else {
2026 bs->bl.pdiscard_alignment = iscsilun->block_size;
2027 }
2028
2029 if (iscsilun->bl.max_ws_len < 0xffffffff / block_size) {
2030 bs->bl.max_pwrite_zeroes =
2031 iscsilun->bl.max_ws_len * iscsilun->block_size;
2032 }
2033 if (iscsilun->lbp.lbpws) {
2034 bs->bl.pwrite_zeroes_alignment =
2035 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2036 } else {
2037 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size;
2038 }
2039 if (iscsilun->bl.opt_xfer_len &&
2040 iscsilun->bl.opt_xfer_len < INT_MAX / block_size) {
2041 bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len *
2042 iscsilun->block_size);
2043 }
2044 }
2045
2046 /* Note that this will not re-establish a connection with an iSCSI target - it
2047 * is effectively a NOP. */
2048 static int iscsi_reopen_prepare(BDRVReopenState *state,
2049 BlockReopenQueue *queue, Error **errp)
2050 {
2051 IscsiLun *iscsilun = state->bs->opaque;
2052
2053 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
2054 error_setg(errp, "Cannot open a write protected LUN as read-write");
2055 return -EACCES;
2056 }
2057 return 0;
2058 }
2059
2060 static void iscsi_reopen_commit(BDRVReopenState *reopen_state)
2061 {
2062 IscsiLun *iscsilun = reopen_state->bs->opaque;
2063
2064 /* the cache.direct status might have changed */
2065 if (iscsilun->allocmap != NULL) {
2066 iscsi_allocmap_init(iscsilun, reopen_state->flags);
2067 }
2068 }
2069
2070 static int iscsi_truncate(BlockDriverState *bs, int64_t offset,
2071 PreallocMode prealloc, Error **errp)
2072 {
2073 IscsiLun *iscsilun = bs->opaque;
2074 Error *local_err = NULL;
2075
2076 if (prealloc != PREALLOC_MODE_OFF) {
2077 error_setg(errp, "Unsupported preallocation mode '%s'",
2078 PreallocMode_str(prealloc));
2079 return -ENOTSUP;
2080 }
2081
2082 if (iscsilun->type != TYPE_DISK) {
2083 error_setg(errp, "Cannot resize non-disk iSCSI devices");
2084 return -ENOTSUP;
2085 }
2086
2087 iscsi_readcapacity_sync(iscsilun, &local_err);
2088 if (local_err != NULL) {
2089 error_propagate(errp, local_err);
2090 return -EIO;
2091 }
2092
2093 if (offset > iscsi_getlength(bs)) {
2094 error_setg(errp, "Cannot grow iSCSI devices");
2095 return -EINVAL;
2096 }
2097
2098 if (iscsilun->allocmap != NULL) {
2099 iscsi_allocmap_init(iscsilun, bs->open_flags);
2100 }
2101
2102 return 0;
2103 }
2104
2105 static int iscsi_create(const char *filename, QemuOpts *opts, Error **errp)
2106 {
2107 int ret = 0;
2108 int64_t total_size = 0;
2109 BlockDriverState *bs;
2110 IscsiLun *iscsilun = NULL;
2111 QDict *bs_options;
2112 Error *local_err = NULL;
2113
2114 bs = bdrv_new();
2115
2116 /* Read out options */
2117 total_size = DIV_ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2118 BDRV_SECTOR_SIZE);
2119 bs->opaque = g_new0(struct IscsiLun, 1);
2120 iscsilun = bs->opaque;
2121
2122 bs_options = qdict_new();
2123 iscsi_parse_filename(filename, bs_options, &local_err);
2124 if (local_err) {
2125 error_propagate(errp, local_err);
2126 ret = -EINVAL;
2127 } else {
2128 ret = iscsi_open(bs, bs_options, 0, NULL);
2129 }
2130 QDECREF(bs_options);
2131
2132 if (ret != 0) {
2133 goto out;
2134 }
2135 iscsi_detach_aio_context(bs);
2136 if (iscsilun->type != TYPE_DISK) {
2137 ret = -ENODEV;
2138 goto out;
2139 }
2140 if (bs->total_sectors < total_size) {
2141 ret = -ENOSPC;
2142 goto out;
2143 }
2144
2145 ret = 0;
2146 out:
2147 if (iscsilun->iscsi != NULL) {
2148 iscsi_destroy_context(iscsilun->iscsi);
2149 }
2150 g_free(bs->opaque);
2151 bs->opaque = NULL;
2152 bdrv_unref(bs);
2153 return ret;
2154 }
2155
2156 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2157 {
2158 IscsiLun *iscsilun = bs->opaque;
2159 bdi->unallocated_blocks_are_zero = iscsilun->lbprz;
2160 bdi->can_write_zeroes_with_unmap = iscsilun->lbprz && iscsilun->lbp.lbpws;
2161 bdi->cluster_size = iscsilun->cluster_sectors * BDRV_SECTOR_SIZE;
2162 return 0;
2163 }
2164
2165 static void iscsi_invalidate_cache(BlockDriverState *bs,
2166 Error **errp)
2167 {
2168 IscsiLun *iscsilun = bs->opaque;
2169 iscsi_allocmap_invalidate(iscsilun);
2170 }
2171
2172 static QemuOptsList iscsi_create_opts = {
2173 .name = "iscsi-create-opts",
2174 .head = QTAILQ_HEAD_INITIALIZER(iscsi_create_opts.head),
2175 .desc = {
2176 {
2177 .name = BLOCK_OPT_SIZE,
2178 .type = QEMU_OPT_SIZE,
2179 .help = "Virtual disk size"
2180 },
2181 { /* end of list */ }
2182 }
2183 };
2184
2185 static BlockDriver bdrv_iscsi = {
2186 .format_name = "iscsi",
2187 .protocol_name = "iscsi",
2188
2189 .instance_size = sizeof(IscsiLun),
2190 .bdrv_parse_filename = iscsi_parse_filename,
2191 .bdrv_file_open = iscsi_open,
2192 .bdrv_close = iscsi_close,
2193 .bdrv_create = iscsi_create,
2194 .create_opts = &iscsi_create_opts,
2195 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2196 .bdrv_reopen_commit = iscsi_reopen_commit,
2197 .bdrv_invalidate_cache = iscsi_invalidate_cache,
2198
2199 .bdrv_getlength = iscsi_getlength,
2200 .bdrv_get_info = iscsi_get_info,
2201 .bdrv_truncate = iscsi_truncate,
2202 .bdrv_refresh_limits = iscsi_refresh_limits,
2203
2204 .bdrv_co_get_block_status = iscsi_co_get_block_status,
2205 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2206 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2207 .bdrv_co_readv = iscsi_co_readv,
2208 .bdrv_co_writev_flags = iscsi_co_writev_flags,
2209 .bdrv_co_flush_to_disk = iscsi_co_flush,
2210
2211 #ifdef __linux__
2212 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2213 #endif
2214
2215 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2216 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2217 };
2218
2219 #if LIBISCSI_API_VERSION >= (20160603)
2220 static BlockDriver bdrv_iser = {
2221 .format_name = "iser",
2222 .protocol_name = "iser",
2223
2224 .instance_size = sizeof(IscsiLun),
2225 .bdrv_parse_filename = iscsi_parse_filename,
2226 .bdrv_file_open = iscsi_open,
2227 .bdrv_close = iscsi_close,
2228 .bdrv_create = iscsi_create,
2229 .create_opts = &iscsi_create_opts,
2230 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2231 .bdrv_reopen_commit = iscsi_reopen_commit,
2232 .bdrv_invalidate_cache = iscsi_invalidate_cache,
2233
2234 .bdrv_getlength = iscsi_getlength,
2235 .bdrv_get_info = iscsi_get_info,
2236 .bdrv_truncate = iscsi_truncate,
2237 .bdrv_refresh_limits = iscsi_refresh_limits,
2238
2239 .bdrv_co_get_block_status = iscsi_co_get_block_status,
2240 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2241 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2242 .bdrv_co_readv = iscsi_co_readv,
2243 .bdrv_co_writev_flags = iscsi_co_writev_flags,
2244 .bdrv_co_flush_to_disk = iscsi_co_flush,
2245
2246 #ifdef __linux__
2247 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2248 #endif
2249
2250 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2251 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2252 };
2253 #endif
2254
2255 static void iscsi_block_init(void)
2256 {
2257 bdrv_register(&bdrv_iscsi);
2258 #if LIBISCSI_API_VERSION >= (20160603)
2259 bdrv_register(&bdrv_iser);
2260 #endif
2261 }
2262
2263 block_init(iscsi_block_init);
Random patches