target/arm: Move be test for regime into S1TranslateResult
[qemu/ar7.git] / block / iscsi.c
bloba316d46d9625865b0dff796d2266f3c4bfa3666c
1 /*
2 * QEMU Block driver for iSCSI images
4 * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
5 * Copyright (c) 2012-2017 Peter Lieven <pl@kamp.de>
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:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
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.
26 #include "qemu/osdep.h"
28 #include <poll.h>
29 #include <math.h>
30 #include <arpa/inet.h>
31 #include "sysemu/sysemu.h"
32 #include "qemu/config-file.h"
33 #include "qemu/error-report.h"
34 #include "qemu/bitops.h"
35 #include "qemu/bitmap.h"
36 #include "block/block_int.h"
37 #include "block/qdict.h"
38 #include "scsi/constants.h"
39 #include "qemu/iov.h"
40 #include "qemu/module.h"
41 #include "qemu/option.h"
42 #include "qemu/uuid.h"
43 #include "sysemu/replay.h"
44 #include "qapi/error.h"
45 #include "qapi/qapi-commands-machine.h"
46 #include "qapi/qmp/qdict.h"
47 #include "qapi/qmp/qstring.h"
48 #include "crypto/secret.h"
49 #include "scsi/utils.h"
50 #include "trace.h"
52 /* Conflict between scsi/utils.h and libiscsi! :( */
53 #define SCSI_XFER_NONE ISCSI_XFER_NONE
54 #include <iscsi/iscsi.h>
55 #define inline __attribute__((gnu_inline)) /* required for libiscsi v1.9.0 */
56 #include <iscsi/scsi-lowlevel.h>
57 #undef inline
58 #undef SCSI_XFER_NONE
59 QEMU_BUILD_BUG_ON((int)SCSI_XFER_NONE != (int)ISCSI_XFER_NONE);
61 #ifdef __linux__
62 #include <scsi/sg.h>
63 #endif
65 typedef struct IscsiLun {
66 struct iscsi_context *iscsi;
67 AioContext *aio_context;
68 int lun;
69 enum scsi_inquiry_peripheral_device_type type;
70 int block_size;
71 uint64_t num_blocks;
72 int events;
73 QEMUTimer *nop_timer;
74 QEMUTimer *event_timer;
75 QemuMutex mutex;
76 struct scsi_inquiry_logical_block_provisioning lbp;
77 struct scsi_inquiry_block_limits bl;
78 struct scsi_inquiry_device_designator *dd;
79 unsigned char *zeroblock;
80 /* The allocmap tracks which clusters (pages) on the iSCSI target are
81 * allocated and which are not. In case a target returns zeros for
82 * unallocated pages (iscsilun->lprz) we can directly return zeros instead
83 * of reading zeros over the wire if a read request falls within an
84 * unallocated block. As there are 3 possible states we need 2 bitmaps to
85 * track. allocmap_valid keeps track if QEMU's information about a page is
86 * valid. allocmap tracks if a page is allocated or not. In case QEMU has no
87 * valid information about a page the corresponding allocmap entry should be
88 * switched to unallocated as well to force a new lookup of the allocation
89 * status as lookups are generally skipped if a page is suspect to be
90 * allocated. If a iSCSI target is opened with cache.direct = on the
91 * allocmap_valid does not exist turning all cached information invalid so
92 * that a fresh lookup is made for any page even if allocmap entry returns
93 * it's unallocated. */
94 unsigned long *allocmap;
95 unsigned long *allocmap_valid;
96 long allocmap_size;
97 int cluster_size;
98 bool use_16_for_rw;
99 bool write_protected;
100 bool lbpme;
101 bool lbprz;
102 bool dpofua;
103 bool has_write_same;
104 bool request_timed_out;
105 } IscsiLun;
107 typedef struct IscsiTask {
108 int status;
109 int complete;
110 int retries;
111 int do_retry;
112 struct scsi_task *task;
113 Coroutine *co;
114 IscsiLun *iscsilun;
115 QEMUTimer retry_timer;
116 int err_code;
117 char *err_str;
118 } IscsiTask;
120 typedef struct IscsiAIOCB {
121 BlockAIOCB common;
122 QEMUBH *bh;
123 IscsiLun *iscsilun;
124 struct scsi_task *task;
125 int status;
126 int64_t sector_num;
127 int nb_sectors;
128 int ret;
129 #ifdef __linux__
130 sg_io_hdr_t *ioh;
131 #endif
132 bool cancelled;
133 } IscsiAIOCB;
135 /* libiscsi uses time_t so its enough to process events every second */
136 #define EVENT_INTERVAL 1000
137 #define NOP_INTERVAL 5000
138 #define MAX_NOP_FAILURES 3
139 #define ISCSI_CMD_RETRIES ARRAY_SIZE(iscsi_retry_times)
140 static const unsigned iscsi_retry_times[] = {8, 32, 128, 512, 2048, 8192, 32768};
142 /* this threshold is a trade-off knob to choose between
143 * the potential additional overhead of an extra GET_LBA_STATUS request
144 * vs. unnecessarily reading a lot of zero sectors over the wire.
145 * If a read request is greater or equal than ISCSI_CHECKALLOC_THRES
146 * sectors we check the allocation status of the area covered by the
147 * request first if the allocationmap indicates that the area might be
148 * unallocated. */
149 #define ISCSI_CHECKALLOC_THRES 64
151 #ifdef __linux__
153 static void
154 iscsi_bh_cb(void *p)
156 IscsiAIOCB *acb = p;
158 qemu_bh_delete(acb->bh);
160 acb->common.cb(acb->common.opaque, acb->status);
162 if (acb->task != NULL) {
163 scsi_free_scsi_task(acb->task);
164 acb->task = NULL;
167 qemu_aio_unref(acb);
170 static void
171 iscsi_schedule_bh(IscsiAIOCB *acb)
173 if (acb->bh) {
174 return;
176 acb->bh = aio_bh_new(acb->iscsilun->aio_context, iscsi_bh_cb, acb);
177 qemu_bh_schedule(acb->bh);
180 #endif
182 static void iscsi_co_generic_bh_cb(void *opaque)
184 struct IscsiTask *iTask = opaque;
186 iTask->complete = 1;
187 aio_co_wake(iTask->co);
190 static void iscsi_retry_timer_expired(void *opaque)
192 struct IscsiTask *iTask = opaque;
193 iTask->complete = 1;
194 if (iTask->co) {
195 aio_co_wake(iTask->co);
199 static inline unsigned exp_random(double mean)
201 return -mean * log((double)rand() / RAND_MAX);
204 /* SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST was introduced in
205 * libiscsi 1.10.0, together with other constants we need. Use it as
206 * a hint that we have to define them ourselves if needed, to keep the
207 * minimum required libiscsi version at 1.9.0. We use an ASCQ macro for
208 * the test because SCSI_STATUS_* is an enum.
210 * To guard against future changes where SCSI_SENSE_ASCQ_* also becomes
211 * an enum, check against the LIBISCSI_API_VERSION macro, which was
212 * introduced in 1.11.0. If it is present, there is no need to define
213 * anything.
215 #if !defined(SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST) && \
216 !defined(LIBISCSI_API_VERSION)
217 #define SCSI_STATUS_TASK_SET_FULL 0x28
218 #define SCSI_STATUS_TIMEOUT 0x0f000002
219 #define SCSI_SENSE_ASCQ_INVALID_FIELD_IN_PARAMETER_LIST 0x2600
220 #define SCSI_SENSE_ASCQ_PARAMETER_LIST_LENGTH_ERROR 0x1a00
221 #endif
223 #ifndef LIBISCSI_API_VERSION
224 #define LIBISCSI_API_VERSION 20130701
225 #endif
227 static int iscsi_translate_sense(struct scsi_sense *sense)
229 return scsi_sense_to_errno(sense->key,
230 (sense->ascq & 0xFF00) >> 8,
231 sense->ascq & 0xFF);
234 /* Called (via iscsi_service) with QemuMutex held. */
235 static void
236 iscsi_co_generic_cb(struct iscsi_context *iscsi, int status,
237 void *command_data, void *opaque)
239 struct IscsiTask *iTask = opaque;
240 struct scsi_task *task = command_data;
242 iTask->status = status;
243 iTask->do_retry = 0;
244 iTask->err_code = 0;
245 iTask->task = task;
247 if (status != SCSI_STATUS_GOOD) {
248 iTask->err_code = -EIO;
249 if (iTask->retries++ < ISCSI_CMD_RETRIES) {
250 if (status == SCSI_STATUS_BUSY ||
251 status == SCSI_STATUS_TIMEOUT ||
252 status == SCSI_STATUS_TASK_SET_FULL) {
253 unsigned retry_time =
254 exp_random(iscsi_retry_times[iTask->retries - 1]);
255 if (status == SCSI_STATUS_TIMEOUT) {
256 /* make sure the request is rescheduled AFTER the
257 * reconnect is initiated */
258 retry_time = EVENT_INTERVAL * 2;
259 iTask->iscsilun->request_timed_out = true;
261 error_report("iSCSI Busy/TaskSetFull/TimeOut"
262 " (retry #%u in %u ms): %s",
263 iTask->retries, retry_time,
264 iscsi_get_error(iscsi));
265 aio_timer_init(iTask->iscsilun->aio_context,
266 &iTask->retry_timer, QEMU_CLOCK_REALTIME,
267 SCALE_MS, iscsi_retry_timer_expired, iTask);
268 timer_mod(&iTask->retry_timer,
269 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + retry_time);
270 iTask->do_retry = 1;
271 } else if (status == SCSI_STATUS_CHECK_CONDITION) {
272 int error = iscsi_translate_sense(&task->sense);
273 if (error == EAGAIN) {
274 error_report("iSCSI CheckCondition: %s",
275 iscsi_get_error(iscsi));
276 iTask->do_retry = 1;
277 } else {
278 iTask->err_code = -error;
279 iTask->err_str = g_strdup(iscsi_get_error(iscsi));
285 if (iTask->co) {
286 replay_bh_schedule_oneshot_event(iTask->iscsilun->aio_context,
287 iscsi_co_generic_bh_cb, iTask);
288 } else {
289 iTask->complete = 1;
293 static void coroutine_fn
294 iscsi_co_init_iscsitask(IscsiLun *iscsilun, struct IscsiTask *iTask)
296 *iTask = (struct IscsiTask) {
297 .co = qemu_coroutine_self(),
298 .iscsilun = iscsilun,
302 #ifdef __linux__
304 /* Called (via iscsi_service) with QemuMutex held. */
305 static void
306 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
307 void *private_data)
309 IscsiAIOCB *acb = private_data;
311 /* If the command callback hasn't been called yet, drop the task */
312 if (!acb->bh) {
313 /* Call iscsi_aio_ioctl_cb() with SCSI_STATUS_CANCELLED */
314 iscsi_scsi_cancel_task(iscsi, acb->task);
317 qemu_aio_unref(acb); /* acquired in iscsi_aio_cancel() */
320 static void
321 iscsi_aio_cancel(BlockAIOCB *blockacb)
323 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
324 IscsiLun *iscsilun = acb->iscsilun;
326 WITH_QEMU_LOCK_GUARD(&iscsilun->mutex) {
328 /* If it was cancelled or completed already, our work is done here */
329 if (acb->cancelled || acb->status != -EINPROGRESS) {
330 return;
333 acb->cancelled = true;
335 qemu_aio_ref(acb); /* released in iscsi_abort_task_cb() */
337 /* send a task mgmt call to the target to cancel the task on the target */
338 if (iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
339 iscsi_abort_task_cb, acb) < 0) {
340 qemu_aio_unref(acb); /* since iscsi_abort_task_cb() won't be called */
345 static const AIOCBInfo iscsi_aiocb_info = {
346 .aiocb_size = sizeof(IscsiAIOCB),
347 .cancel_async = iscsi_aio_cancel,
350 #endif
352 static void iscsi_process_read(void *arg);
353 static void iscsi_process_write(void *arg);
355 /* Called with QemuMutex held. */
356 static void
357 iscsi_set_events(IscsiLun *iscsilun)
359 struct iscsi_context *iscsi = iscsilun->iscsi;
360 int ev = iscsi_which_events(iscsi);
362 if (ev != iscsilun->events) {
363 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsi),
364 false,
365 (ev & POLLIN) ? iscsi_process_read : NULL,
366 (ev & POLLOUT) ? iscsi_process_write : NULL,
367 NULL, NULL,
368 iscsilun);
369 iscsilun->events = ev;
373 static void iscsi_timed_check_events(void *opaque)
375 IscsiLun *iscsilun = opaque;
377 WITH_QEMU_LOCK_GUARD(&iscsilun->mutex) {
378 /* check for timed out requests */
379 iscsi_service(iscsilun->iscsi, 0);
381 if (iscsilun->request_timed_out) {
382 iscsilun->request_timed_out = false;
383 iscsi_reconnect(iscsilun->iscsi);
387 * newer versions of libiscsi may return zero events. Ensure we are
388 * able to return to service once this situation changes.
390 iscsi_set_events(iscsilun);
393 timer_mod(iscsilun->event_timer,
394 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
397 static void
398 iscsi_process_read(void *arg)
400 IscsiLun *iscsilun = arg;
401 struct iscsi_context *iscsi = iscsilun->iscsi;
403 qemu_mutex_lock(&iscsilun->mutex);
404 iscsi_service(iscsi, POLLIN);
405 iscsi_set_events(iscsilun);
406 qemu_mutex_unlock(&iscsilun->mutex);
409 static void
410 iscsi_process_write(void *arg)
412 IscsiLun *iscsilun = arg;
413 struct iscsi_context *iscsi = iscsilun->iscsi;
415 qemu_mutex_lock(&iscsilun->mutex);
416 iscsi_service(iscsi, POLLOUT);
417 iscsi_set_events(iscsilun);
418 qemu_mutex_unlock(&iscsilun->mutex);
421 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
423 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
426 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
428 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
431 static bool is_byte_request_lun_aligned(int64_t offset, int64_t bytes,
432 IscsiLun *iscsilun)
434 if (offset % iscsilun->block_size || bytes % iscsilun->block_size) {
435 error_report("iSCSI misaligned request: "
436 "iscsilun->block_size %u, offset %" PRIi64
437 ", bytes %" PRIi64,
438 iscsilun->block_size, offset, bytes);
439 return false;
441 return true;
444 static bool is_sector_request_lun_aligned(int64_t sector_num, int nb_sectors,
445 IscsiLun *iscsilun)
447 assert(nb_sectors <= BDRV_REQUEST_MAX_SECTORS);
448 return is_byte_request_lun_aligned(sector_num << BDRV_SECTOR_BITS,
449 nb_sectors << BDRV_SECTOR_BITS,
450 iscsilun);
453 static void iscsi_allocmap_free(IscsiLun *iscsilun)
455 g_free(iscsilun->allocmap);
456 g_free(iscsilun->allocmap_valid);
457 iscsilun->allocmap = NULL;
458 iscsilun->allocmap_valid = NULL;
462 static int iscsi_allocmap_init(IscsiLun *iscsilun, int open_flags)
464 iscsi_allocmap_free(iscsilun);
466 assert(iscsilun->cluster_size);
467 iscsilun->allocmap_size =
468 DIV_ROUND_UP(iscsilun->num_blocks * iscsilun->block_size,
469 iscsilun->cluster_size);
471 iscsilun->allocmap = bitmap_try_new(iscsilun->allocmap_size);
472 if (!iscsilun->allocmap) {
473 return -ENOMEM;
476 if (open_flags & BDRV_O_NOCACHE) {
477 /* when cache.direct = on all allocmap entries are
478 * treated as invalid to force a relookup of the block
479 * status on every read request */
480 return 0;
483 iscsilun->allocmap_valid = bitmap_try_new(iscsilun->allocmap_size);
484 if (!iscsilun->allocmap_valid) {
485 /* if we are under memory pressure free the allocmap as well */
486 iscsi_allocmap_free(iscsilun);
487 return -ENOMEM;
490 return 0;
493 static void
494 iscsi_allocmap_update(IscsiLun *iscsilun, int64_t offset,
495 int64_t bytes, bool allocated, bool valid)
497 int64_t cl_num_expanded, nb_cls_expanded, cl_num_shrunk, nb_cls_shrunk;
499 if (iscsilun->allocmap == NULL) {
500 return;
502 /* expand to entirely contain all affected clusters */
503 assert(iscsilun->cluster_size);
504 cl_num_expanded = offset / iscsilun->cluster_size;
505 nb_cls_expanded = DIV_ROUND_UP(offset + bytes,
506 iscsilun->cluster_size) - cl_num_expanded;
507 /* shrink to touch only completely contained clusters */
508 cl_num_shrunk = DIV_ROUND_UP(offset, iscsilun->cluster_size);
509 nb_cls_shrunk = (offset + bytes) / iscsilun->cluster_size - cl_num_shrunk;
510 if (allocated) {
511 bitmap_set(iscsilun->allocmap, cl_num_expanded, nb_cls_expanded);
512 } else {
513 if (nb_cls_shrunk > 0) {
514 bitmap_clear(iscsilun->allocmap, cl_num_shrunk, nb_cls_shrunk);
518 if (iscsilun->allocmap_valid == NULL) {
519 return;
521 if (valid) {
522 if (nb_cls_shrunk > 0) {
523 bitmap_set(iscsilun->allocmap_valid, cl_num_shrunk, nb_cls_shrunk);
525 } else {
526 bitmap_clear(iscsilun->allocmap_valid, cl_num_expanded,
527 nb_cls_expanded);
531 static void
532 iscsi_allocmap_set_allocated(IscsiLun *iscsilun, int64_t offset,
533 int64_t bytes)
535 iscsi_allocmap_update(iscsilun, offset, bytes, true, true);
538 static void
539 iscsi_allocmap_set_unallocated(IscsiLun *iscsilun, int64_t offset,
540 int64_t bytes)
542 /* Note: if cache.direct=on the fifth argument to iscsi_allocmap_update
543 * is ignored, so this will in effect be an iscsi_allocmap_set_invalid.
545 iscsi_allocmap_update(iscsilun, offset, bytes, false, true);
548 static void iscsi_allocmap_set_invalid(IscsiLun *iscsilun, int64_t offset,
549 int64_t bytes)
551 iscsi_allocmap_update(iscsilun, offset, bytes, false, false);
554 static void iscsi_allocmap_invalidate(IscsiLun *iscsilun)
556 if (iscsilun->allocmap) {
557 bitmap_zero(iscsilun->allocmap, iscsilun->allocmap_size);
559 if (iscsilun->allocmap_valid) {
560 bitmap_zero(iscsilun->allocmap_valid, iscsilun->allocmap_size);
564 static inline bool
565 iscsi_allocmap_is_allocated(IscsiLun *iscsilun, int64_t offset,
566 int64_t bytes)
568 unsigned long size;
569 if (iscsilun->allocmap == NULL) {
570 return true;
572 assert(iscsilun->cluster_size);
573 size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size);
574 return !(find_next_bit(iscsilun->allocmap, size,
575 offset / iscsilun->cluster_size) == size);
578 static inline bool iscsi_allocmap_is_valid(IscsiLun *iscsilun,
579 int64_t offset, int64_t bytes)
581 unsigned long size;
582 if (iscsilun->allocmap_valid == NULL) {
583 return false;
585 assert(iscsilun->cluster_size);
586 size = DIV_ROUND_UP(offset + bytes, iscsilun->cluster_size);
587 return (find_next_zero_bit(iscsilun->allocmap_valid, size,
588 offset / iscsilun->cluster_size) == size);
591 static void coroutine_fn iscsi_co_wait_for_task(IscsiTask *iTask,
592 IscsiLun *iscsilun)
594 while (!iTask->complete) {
595 iscsi_set_events(iscsilun);
596 qemu_mutex_unlock(&iscsilun->mutex);
597 qemu_coroutine_yield();
598 qemu_mutex_lock(&iscsilun->mutex);
602 static int coroutine_fn
603 iscsi_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
604 QEMUIOVector *iov, int flags)
606 IscsiLun *iscsilun = bs->opaque;
607 struct IscsiTask iTask;
608 uint64_t lba;
609 uint32_t num_sectors;
610 bool fua = flags & BDRV_REQ_FUA;
611 int r = 0;
613 if (fua) {
614 assert(iscsilun->dpofua);
616 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
617 return -EINVAL;
620 if (bs->bl.max_transfer) {
621 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
624 lba = sector_qemu2lun(sector_num, iscsilun);
625 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
626 iscsi_co_init_iscsitask(iscsilun, &iTask);
627 qemu_mutex_lock(&iscsilun->mutex);
628 retry:
629 if (iscsilun->use_16_for_rw) {
630 #if LIBISCSI_API_VERSION >= (20160603)
631 iTask.task = iscsi_write16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
632 NULL, num_sectors * iscsilun->block_size,
633 iscsilun->block_size, 0, 0, fua, 0, 0,
634 iscsi_co_generic_cb, &iTask,
635 (struct scsi_iovec *)iov->iov, iov->niov);
636 } else {
637 iTask.task = iscsi_write10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
638 NULL, num_sectors * iscsilun->block_size,
639 iscsilun->block_size, 0, 0, fua, 0, 0,
640 iscsi_co_generic_cb, &iTask,
641 (struct scsi_iovec *)iov->iov, iov->niov);
643 #else
644 iTask.task = iscsi_write16_task(iscsilun->iscsi, iscsilun->lun, lba,
645 NULL, num_sectors * iscsilun->block_size,
646 iscsilun->block_size, 0, 0, fua, 0, 0,
647 iscsi_co_generic_cb, &iTask);
648 } else {
649 iTask.task = iscsi_write10_task(iscsilun->iscsi, iscsilun->lun, lba,
650 NULL, num_sectors * iscsilun->block_size,
651 iscsilun->block_size, 0, 0, fua, 0, 0,
652 iscsi_co_generic_cb, &iTask);
654 #endif
655 if (iTask.task == NULL) {
656 qemu_mutex_unlock(&iscsilun->mutex);
657 return -ENOMEM;
659 #if LIBISCSI_API_VERSION < (20160603)
660 scsi_task_set_iov_out(iTask.task, (struct scsi_iovec *) iov->iov,
661 iov->niov);
662 #endif
663 iscsi_co_wait_for_task(&iTask, iscsilun);
665 if (iTask.task != NULL) {
666 scsi_free_scsi_task(iTask.task);
667 iTask.task = NULL;
670 if (iTask.do_retry) {
671 iTask.complete = 0;
672 goto retry;
675 if (iTask.status != SCSI_STATUS_GOOD) {
676 iscsi_allocmap_set_invalid(iscsilun, sector_num * BDRV_SECTOR_SIZE,
677 nb_sectors * BDRV_SECTOR_SIZE);
678 error_report("iSCSI WRITE10/16 failed at lba %" PRIu64 ": %s", lba,
679 iTask.err_str);
680 r = iTask.err_code;
681 goto out_unlock;
684 iscsi_allocmap_set_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE,
685 nb_sectors * BDRV_SECTOR_SIZE);
687 out_unlock:
688 qemu_mutex_unlock(&iscsilun->mutex);
689 g_free(iTask.err_str);
690 return r;
695 static int coroutine_fn iscsi_co_block_status(BlockDriverState *bs,
696 bool want_zero, int64_t offset,
697 int64_t bytes, int64_t *pnum,
698 int64_t *map,
699 BlockDriverState **file)
701 IscsiLun *iscsilun = bs->opaque;
702 struct scsi_get_lba_status *lbas = NULL;
703 struct scsi_lba_status_descriptor *lbasd = NULL;
704 struct IscsiTask iTask;
705 uint64_t lba, max_bytes;
706 int ret;
708 iscsi_co_init_iscsitask(iscsilun, &iTask);
710 assert(QEMU_IS_ALIGNED(offset | bytes, iscsilun->block_size));
712 /* default to all sectors allocated */
713 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
714 if (map) {
715 *map = offset;
717 *pnum = bytes;
719 /* LUN does not support logical block provisioning */
720 if (!iscsilun->lbpme) {
721 goto out;
724 lba = offset / iscsilun->block_size;
725 max_bytes = (iscsilun->num_blocks - lba) * iscsilun->block_size;
727 qemu_mutex_lock(&iscsilun->mutex);
728 retry:
729 if (iscsi_get_lba_status_task(iscsilun->iscsi, iscsilun->lun,
730 lba, 8 + 16, iscsi_co_generic_cb,
731 &iTask) == NULL) {
732 ret = -ENOMEM;
733 goto out_unlock;
735 iscsi_co_wait_for_task(&iTask, iscsilun);
737 if (iTask.do_retry) {
738 if (iTask.task != NULL) {
739 scsi_free_scsi_task(iTask.task);
740 iTask.task = NULL;
742 iTask.complete = 0;
743 goto retry;
746 if (iTask.status != SCSI_STATUS_GOOD) {
747 /* in case the get_lba_status_callout fails (i.e.
748 * because the device is busy or the cmd is not
749 * supported) we pretend all blocks are allocated
750 * for backwards compatibility */
751 error_report("iSCSI GET_LBA_STATUS failed at lba %" PRIu64 ": %s",
752 lba, iTask.err_str);
753 goto out_unlock;
756 lbas = scsi_datain_unmarshall(iTask.task);
757 if (lbas == NULL || lbas->num_descriptors == 0) {
758 ret = -EIO;
759 goto out_unlock;
762 lbasd = &lbas->descriptors[0];
764 if (lba != lbasd->lba) {
765 ret = -EIO;
766 goto out_unlock;
769 *pnum = MIN((int64_t) lbasd->num_blocks * iscsilun->block_size, max_bytes);
771 if (lbasd->provisioning == SCSI_PROVISIONING_TYPE_DEALLOCATED ||
772 lbasd->provisioning == SCSI_PROVISIONING_TYPE_ANCHORED) {
773 ret &= ~BDRV_BLOCK_DATA;
774 if (iscsilun->lbprz) {
775 ret |= BDRV_BLOCK_ZERO;
779 if (ret & BDRV_BLOCK_ZERO) {
780 iscsi_allocmap_set_unallocated(iscsilun, offset, *pnum);
781 } else {
782 iscsi_allocmap_set_allocated(iscsilun, offset, *pnum);
785 out_unlock:
786 qemu_mutex_unlock(&iscsilun->mutex);
787 g_free(iTask.err_str);
788 out:
789 if (iTask.task != NULL) {
790 scsi_free_scsi_task(iTask.task);
792 if (ret > 0 && ret & BDRV_BLOCK_OFFSET_VALID && file) {
793 *file = bs;
795 return ret;
798 static int coroutine_fn iscsi_co_readv(BlockDriverState *bs,
799 int64_t sector_num, int nb_sectors,
800 QEMUIOVector *iov)
802 IscsiLun *iscsilun = bs->opaque;
803 struct IscsiTask iTask;
804 uint64_t lba;
805 uint32_t num_sectors;
806 int r = 0;
808 if (!is_sector_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
809 return -EINVAL;
812 if (bs->bl.max_transfer) {
813 assert(nb_sectors << BDRV_SECTOR_BITS <= bs->bl.max_transfer);
816 /* if cache.direct is off and we have a valid entry in our allocation map
817 * we can skip checking the block status and directly return zeroes if
818 * the request falls within an unallocated area */
819 if (iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE,
820 nb_sectors * BDRV_SECTOR_SIZE) &&
821 !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE,
822 nb_sectors * BDRV_SECTOR_SIZE)) {
823 qemu_iovec_memset(iov, 0, 0x00, iov->size);
824 return 0;
827 if (nb_sectors >= ISCSI_CHECKALLOC_THRES &&
828 !iscsi_allocmap_is_valid(iscsilun, sector_num * BDRV_SECTOR_SIZE,
829 nb_sectors * BDRV_SECTOR_SIZE) &&
830 !iscsi_allocmap_is_allocated(iscsilun, sector_num * BDRV_SECTOR_SIZE,
831 nb_sectors * BDRV_SECTOR_SIZE)) {
832 int64_t pnum;
833 /* check the block status from the beginning of the cluster
834 * containing the start sector */
835 int64_t head;
836 int ret;
838 assert(iscsilun->cluster_size);
839 head = (sector_num * BDRV_SECTOR_SIZE) % iscsilun->cluster_size;
840 ret = iscsi_co_block_status(bs, true,
841 sector_num * BDRV_SECTOR_SIZE - head,
842 BDRV_REQUEST_MAX_BYTES, &pnum, NULL, NULL);
843 if (ret < 0) {
844 return ret;
846 /* if the whole request falls into an unallocated area we can avoid
847 * reading and directly return zeroes instead */
848 if (ret & BDRV_BLOCK_ZERO &&
849 pnum >= nb_sectors * BDRV_SECTOR_SIZE + head) {
850 qemu_iovec_memset(iov, 0, 0x00, iov->size);
851 return 0;
855 lba = sector_qemu2lun(sector_num, iscsilun);
856 num_sectors = sector_qemu2lun(nb_sectors, iscsilun);
858 iscsi_co_init_iscsitask(iscsilun, &iTask);
859 qemu_mutex_lock(&iscsilun->mutex);
860 retry:
861 if (iscsilun->use_16_for_rw) {
862 #if LIBISCSI_API_VERSION >= (20160603)
863 iTask.task = iscsi_read16_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
864 num_sectors * iscsilun->block_size,
865 iscsilun->block_size, 0, 0, 0, 0, 0,
866 iscsi_co_generic_cb, &iTask,
867 (struct scsi_iovec *)iov->iov, iov->niov);
868 } else {
869 iTask.task = iscsi_read10_iov_task(iscsilun->iscsi, iscsilun->lun, lba,
870 num_sectors * iscsilun->block_size,
871 iscsilun->block_size,
872 0, 0, 0, 0, 0,
873 iscsi_co_generic_cb, &iTask,
874 (struct scsi_iovec *)iov->iov, iov->niov);
876 #else
877 iTask.task = iscsi_read16_task(iscsilun->iscsi, iscsilun->lun, lba,
878 num_sectors * iscsilun->block_size,
879 iscsilun->block_size, 0, 0, 0, 0, 0,
880 iscsi_co_generic_cb, &iTask);
881 } else {
882 iTask.task = iscsi_read10_task(iscsilun->iscsi, iscsilun->lun, lba,
883 num_sectors * iscsilun->block_size,
884 iscsilun->block_size,
885 0, 0, 0, 0, 0,
886 iscsi_co_generic_cb, &iTask);
888 #endif
889 if (iTask.task == NULL) {
890 qemu_mutex_unlock(&iscsilun->mutex);
891 return -ENOMEM;
893 #if LIBISCSI_API_VERSION < (20160603)
894 scsi_task_set_iov_in(iTask.task, (struct scsi_iovec *) iov->iov, iov->niov);
895 #endif
897 iscsi_co_wait_for_task(&iTask, iscsilun);
898 if (iTask.task != NULL) {
899 scsi_free_scsi_task(iTask.task);
900 iTask.task = NULL;
903 if (iTask.do_retry) {
904 iTask.complete = 0;
905 goto retry;
908 if (iTask.status != SCSI_STATUS_GOOD) {
909 error_report("iSCSI READ10/16 failed at lba %" PRIu64 ": %s",
910 lba, iTask.err_str);
911 r = iTask.err_code;
914 qemu_mutex_unlock(&iscsilun->mutex);
915 g_free(iTask.err_str);
916 return r;
919 static int coroutine_fn iscsi_co_flush(BlockDriverState *bs)
921 IscsiLun *iscsilun = bs->opaque;
922 struct IscsiTask iTask;
923 int r = 0;
925 iscsi_co_init_iscsitask(iscsilun, &iTask);
926 qemu_mutex_lock(&iscsilun->mutex);
927 retry:
928 if (iscsi_synchronizecache10_task(iscsilun->iscsi, iscsilun->lun, 0, 0, 0,
929 0, iscsi_co_generic_cb, &iTask) == NULL) {
930 qemu_mutex_unlock(&iscsilun->mutex);
931 return -ENOMEM;
934 iscsi_co_wait_for_task(&iTask, iscsilun);
936 if (iTask.task != NULL) {
937 scsi_free_scsi_task(iTask.task);
938 iTask.task = NULL;
941 if (iTask.do_retry) {
942 iTask.complete = 0;
943 goto retry;
946 if (iTask.status != SCSI_STATUS_GOOD) {
947 error_report("iSCSI SYNCHRONIZECACHE10 failed: %s", iTask.err_str);
948 r = iTask.err_code;
951 qemu_mutex_unlock(&iscsilun->mutex);
952 g_free(iTask.err_str);
953 return r;
956 #ifdef __linux__
957 /* Called (via iscsi_service) with QemuMutex held. */
958 static void
959 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
960 void *command_data, void *opaque)
962 IscsiAIOCB *acb = opaque;
964 if (status == SCSI_STATUS_CANCELLED) {
965 if (!acb->bh) {
966 acb->status = -ECANCELED;
967 iscsi_schedule_bh(acb);
969 return;
972 acb->status = 0;
973 if (status < 0) {
974 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
975 iscsi_get_error(iscsi));
976 acb->status = -iscsi_translate_sense(&acb->task->sense);
979 acb->ioh->driver_status = 0;
980 acb->ioh->host_status = 0;
981 acb->ioh->resid = 0;
982 acb->ioh->status = status;
984 #define SG_ERR_DRIVER_SENSE 0x08
986 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
987 int ss;
989 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
991 acb->ioh->sb_len_wr = acb->task->datain.size - 2;
992 ss = MIN(acb->ioh->mx_sb_len, acb->ioh->sb_len_wr);
993 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
996 iscsi_schedule_bh(acb);
999 static void iscsi_ioctl_bh_completion(void *opaque)
1001 IscsiAIOCB *acb = opaque;
1003 qemu_bh_delete(acb->bh);
1004 acb->common.cb(acb->common.opaque, acb->ret);
1005 qemu_aio_unref(acb);
1008 static void iscsi_ioctl_handle_emulated(IscsiAIOCB *acb, int req, void *buf)
1010 BlockDriverState *bs = acb->common.bs;
1011 IscsiLun *iscsilun = bs->opaque;
1012 int ret = 0;
1014 switch (req) {
1015 case SG_GET_VERSION_NUM:
1016 *(int *)buf = 30000;
1017 break;
1018 case SG_GET_SCSI_ID:
1019 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
1020 break;
1021 default:
1022 ret = -EINVAL;
1024 assert(!acb->bh);
1025 acb->bh = aio_bh_new(bdrv_get_aio_context(bs),
1026 iscsi_ioctl_bh_completion, acb);
1027 acb->ret = ret;
1028 qemu_bh_schedule(acb->bh);
1031 static BlockAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
1032 unsigned long int req, void *buf,
1033 BlockCompletionFunc *cb, void *opaque)
1035 IscsiLun *iscsilun = bs->opaque;
1036 struct iscsi_context *iscsi = iscsilun->iscsi;
1037 struct iscsi_data data;
1038 IscsiAIOCB *acb;
1040 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
1042 acb->iscsilun = iscsilun;
1043 acb->bh = NULL;
1044 acb->status = -EINPROGRESS;
1045 acb->ioh = buf;
1046 acb->cancelled = false;
1048 if (req != SG_IO) {
1049 iscsi_ioctl_handle_emulated(acb, req, buf);
1050 return &acb->common;
1053 if (acb->ioh->cmd_len > SCSI_CDB_MAX_SIZE) {
1054 error_report("iSCSI: ioctl error CDB exceeds max size (%d > %d)",
1055 acb->ioh->cmd_len, SCSI_CDB_MAX_SIZE);
1056 qemu_aio_unref(acb);
1057 return NULL;
1060 acb->task = malloc(sizeof(struct scsi_task));
1061 if (acb->task == NULL) {
1062 error_report("iSCSI: Failed to allocate task for scsi command. %s",
1063 iscsi_get_error(iscsi));
1064 qemu_aio_unref(acb);
1065 return NULL;
1067 memset(acb->task, 0, sizeof(struct scsi_task));
1069 switch (acb->ioh->dxfer_direction) {
1070 case SG_DXFER_TO_DEV:
1071 acb->task->xfer_dir = SCSI_XFER_WRITE;
1072 break;
1073 case SG_DXFER_FROM_DEV:
1074 acb->task->xfer_dir = SCSI_XFER_READ;
1075 break;
1076 default:
1077 acb->task->xfer_dir = SCSI_XFER_NONE;
1078 break;
1081 acb->task->cdb_size = acb->ioh->cmd_len;
1082 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
1083 acb->task->expxferlen = acb->ioh->dxfer_len;
1085 data.size = 0;
1086 qemu_mutex_lock(&iscsilun->mutex);
1087 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
1088 if (acb->ioh->iovec_count == 0) {
1089 data.data = acb->ioh->dxferp;
1090 data.size = acb->ioh->dxfer_len;
1091 } else {
1092 scsi_task_set_iov_out(acb->task,
1093 (struct scsi_iovec *) acb->ioh->dxferp,
1094 acb->ioh->iovec_count);
1098 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
1099 iscsi_aio_ioctl_cb,
1100 (data.size > 0) ? &data : NULL,
1101 acb) != 0) {
1102 qemu_mutex_unlock(&iscsilun->mutex);
1103 scsi_free_scsi_task(acb->task);
1104 qemu_aio_unref(acb);
1105 return NULL;
1108 /* tell libiscsi to read straight into the buffer we got from ioctl */
1109 if (acb->task->xfer_dir == SCSI_XFER_READ) {
1110 if (acb->ioh->iovec_count == 0) {
1111 scsi_task_add_data_in_buffer(acb->task,
1112 acb->ioh->dxfer_len,
1113 acb->ioh->dxferp);
1114 } else {
1115 scsi_task_set_iov_in(acb->task,
1116 (struct scsi_iovec *) acb->ioh->dxferp,
1117 acb->ioh->iovec_count);
1121 iscsi_set_events(iscsilun);
1122 qemu_mutex_unlock(&iscsilun->mutex);
1124 return &acb->common;
1127 #endif
1129 static int64_t
1130 iscsi_getlength(BlockDriverState *bs)
1132 IscsiLun *iscsilun = bs->opaque;
1133 int64_t len;
1135 len = iscsilun->num_blocks;
1136 len *= iscsilun->block_size;
1138 return len;
1141 static int
1142 coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset,
1143 int64_t bytes)
1145 IscsiLun *iscsilun = bs->opaque;
1146 struct IscsiTask iTask;
1147 struct unmap_list list;
1148 int r = 0;
1150 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) {
1151 return -ENOTSUP;
1154 if (!iscsilun->lbp.lbpu) {
1155 /* UNMAP is not supported by the target */
1156 return 0;
1160 * We don't want to overflow list.num which is uint32_t.
1161 * We rely on our max_pdiscard.
1163 assert(bytes / iscsilun->block_size <= UINT32_MAX);
1165 list.lba = offset / iscsilun->block_size;
1166 list.num = bytes / iscsilun->block_size;
1168 iscsi_co_init_iscsitask(iscsilun, &iTask);
1169 qemu_mutex_lock(&iscsilun->mutex);
1170 retry:
1171 if (iscsi_unmap_task(iscsilun->iscsi, iscsilun->lun, 0, 0, &list, 1,
1172 iscsi_co_generic_cb, &iTask) == NULL) {
1173 r = -ENOMEM;
1174 goto out_unlock;
1177 iscsi_co_wait_for_task(&iTask, iscsilun);
1179 if (iTask.task != NULL) {
1180 scsi_free_scsi_task(iTask.task);
1181 iTask.task = NULL;
1184 if (iTask.do_retry) {
1185 iTask.complete = 0;
1186 goto retry;
1189 iscsi_allocmap_set_invalid(iscsilun, offset, bytes);
1191 if (iTask.status == SCSI_STATUS_CHECK_CONDITION) {
1192 /* the target might fail with a check condition if it
1193 is not happy with the alignment of the UNMAP request
1194 we silently fail in this case */
1195 goto out_unlock;
1198 if (iTask.status != SCSI_STATUS_GOOD) {
1199 error_report("iSCSI UNMAP failed at lba %" PRIu64 ": %s",
1200 list.lba, iTask.err_str);
1201 r = iTask.err_code;
1202 goto out_unlock;
1205 out_unlock:
1206 qemu_mutex_unlock(&iscsilun->mutex);
1207 g_free(iTask.err_str);
1208 return r;
1211 static int
1212 coroutine_fn iscsi_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
1213 int64_t bytes, BdrvRequestFlags flags)
1215 IscsiLun *iscsilun = bs->opaque;
1216 struct IscsiTask iTask;
1217 uint64_t lba;
1218 uint64_t nb_blocks;
1219 bool use_16_for_ws = iscsilun->use_16_for_rw;
1220 int r = 0;
1222 if (!is_byte_request_lun_aligned(offset, bytes, iscsilun)) {
1223 return -ENOTSUP;
1226 if (flags & BDRV_REQ_MAY_UNMAP) {
1227 if (!use_16_for_ws && !iscsilun->lbp.lbpws10) {
1228 /* WRITESAME10 with UNMAP is unsupported try WRITESAME16 */
1229 use_16_for_ws = true;
1231 if (use_16_for_ws && !iscsilun->lbp.lbpws) {
1232 /* WRITESAME16 with UNMAP is not supported by the target,
1233 * fall back and try WRITESAME10/16 without UNMAP */
1234 flags &= ~BDRV_REQ_MAY_UNMAP;
1235 use_16_for_ws = iscsilun->use_16_for_rw;
1239 if (!(flags & BDRV_REQ_MAY_UNMAP) && !iscsilun->has_write_same) {
1240 /* WRITESAME without UNMAP is not supported by the target */
1241 return -ENOTSUP;
1244 lba = offset / iscsilun->block_size;
1245 nb_blocks = bytes / iscsilun->block_size;
1247 if (iscsilun->zeroblock == NULL) {
1248 iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
1249 if (iscsilun->zeroblock == NULL) {
1250 return -ENOMEM;
1254 qemu_mutex_lock(&iscsilun->mutex);
1255 iscsi_co_init_iscsitask(iscsilun, &iTask);
1256 retry:
1257 if (use_16_for_ws) {
1259 * iscsi_writesame16_task num_blocks argument is uint32_t. We rely here
1260 * on our max_pwrite_zeroes limit.
1262 assert(nb_blocks <= UINT32_MAX);
1263 iTask.task = iscsi_writesame16_task(iscsilun->iscsi, iscsilun->lun, lba,
1264 iscsilun->zeroblock, iscsilun->block_size,
1265 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1266 0, 0, iscsi_co_generic_cb, &iTask);
1267 } else {
1269 * iscsi_writesame10_task num_blocks argument is uint16_t. We rely here
1270 * on our max_pwrite_zeroes limit.
1272 assert(nb_blocks <= UINT16_MAX);
1273 iTask.task = iscsi_writesame10_task(iscsilun->iscsi, iscsilun->lun, lba,
1274 iscsilun->zeroblock, iscsilun->block_size,
1275 nb_blocks, 0, !!(flags & BDRV_REQ_MAY_UNMAP),
1276 0, 0, iscsi_co_generic_cb, &iTask);
1278 if (iTask.task == NULL) {
1279 qemu_mutex_unlock(&iscsilun->mutex);
1280 return -ENOMEM;
1283 iscsi_co_wait_for_task(&iTask, iscsilun);
1285 if (iTask.status == SCSI_STATUS_CHECK_CONDITION &&
1286 iTask.task->sense.key == SCSI_SENSE_ILLEGAL_REQUEST &&
1287 (iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_OPERATION_CODE ||
1288 iTask.task->sense.ascq == SCSI_SENSE_ASCQ_INVALID_FIELD_IN_CDB)) {
1289 /* WRITE SAME is not supported by the target */
1290 iscsilun->has_write_same = false;
1291 scsi_free_scsi_task(iTask.task);
1292 r = -ENOTSUP;
1293 goto out_unlock;
1296 if (iTask.task != NULL) {
1297 scsi_free_scsi_task(iTask.task);
1298 iTask.task = NULL;
1301 if (iTask.do_retry) {
1302 iTask.complete = 0;
1303 goto retry;
1306 if (iTask.status != SCSI_STATUS_GOOD) {
1307 iscsi_allocmap_set_invalid(iscsilun, offset, bytes);
1308 error_report("iSCSI WRITESAME10/16 failed at lba %" PRIu64 ": %s",
1309 lba, iTask.err_str);
1310 r = iTask.err_code;
1311 goto out_unlock;
1314 if (flags & BDRV_REQ_MAY_UNMAP) {
1315 iscsi_allocmap_set_invalid(iscsilun, offset, bytes);
1316 } else {
1317 iscsi_allocmap_set_allocated(iscsilun, offset, bytes);
1320 out_unlock:
1321 qemu_mutex_unlock(&iscsilun->mutex);
1322 g_free(iTask.err_str);
1323 return r;
1326 static void apply_chap(struct iscsi_context *iscsi, QemuOpts *opts,
1327 Error **errp)
1329 const char *user = NULL;
1330 const char *password = NULL;
1331 const char *secretid;
1332 char *secret = NULL;
1334 user = qemu_opt_get(opts, "user");
1335 if (!user) {
1336 return;
1339 secretid = qemu_opt_get(opts, "password-secret");
1340 password = qemu_opt_get(opts, "password");
1341 if (secretid && password) {
1342 error_setg(errp, "'password' and 'password-secret' properties are "
1343 "mutually exclusive");
1344 return;
1346 if (secretid) {
1347 secret = qcrypto_secret_lookup_as_utf8(secretid, errp);
1348 if (!secret) {
1349 return;
1351 password = secret;
1352 } else if (!password) {
1353 error_setg(errp, "CHAP username specified but no password was given");
1354 return;
1357 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
1358 error_setg(errp, "Failed to set initiator username and password");
1361 g_free(secret);
1364 static void apply_header_digest(struct iscsi_context *iscsi, QemuOpts *opts,
1365 Error **errp)
1367 const char *digest = NULL;
1369 digest = qemu_opt_get(opts, "header-digest");
1370 if (!digest) {
1371 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1372 } else if (!strcmp(digest, "crc32c")) {
1373 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
1374 } else if (!strcmp(digest, "none")) {
1375 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
1376 } else if (!strcmp(digest, "crc32c-none")) {
1377 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
1378 } else if (!strcmp(digest, "none-crc32c")) {
1379 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1380 } else {
1381 error_setg(errp, "Invalid header-digest setting : %s", digest);
1385 static char *get_initiator_name(QemuOpts *opts)
1387 const char *name;
1388 char *iscsi_name;
1389 UuidInfo *uuid_info;
1391 name = qemu_opt_get(opts, "initiator-name");
1392 if (name) {
1393 return g_strdup(name);
1396 uuid_info = qmp_query_uuid(NULL);
1397 if (strcmp(uuid_info->UUID, UUID_NONE) == 0) {
1398 name = qemu_get_vm_name();
1399 } else {
1400 name = uuid_info->UUID;
1402 iscsi_name = g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
1403 name ? ":" : "", name ? name : "");
1404 qapi_free_UuidInfo(uuid_info);
1405 return iscsi_name;
1408 static void iscsi_nop_timed_event(void *opaque)
1410 IscsiLun *iscsilun = opaque;
1412 QEMU_LOCK_GUARD(&iscsilun->mutex);
1413 if (iscsi_get_nops_in_flight(iscsilun->iscsi) >= MAX_NOP_FAILURES) {
1414 error_report("iSCSI: NOP timeout. Reconnecting...");
1415 iscsilun->request_timed_out = true;
1416 } else if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
1417 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
1418 return;
1421 timer_mod(iscsilun->nop_timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1422 iscsi_set_events(iscsilun);
1425 static void iscsi_readcapacity_sync(IscsiLun *iscsilun, Error **errp)
1427 struct scsi_task *task = NULL;
1428 struct scsi_readcapacity10 *rc10 = NULL;
1429 struct scsi_readcapacity16 *rc16 = NULL;
1430 int retries = ISCSI_CMD_RETRIES;
1432 do {
1433 if (task != NULL) {
1434 scsi_free_scsi_task(task);
1435 task = NULL;
1438 switch (iscsilun->type) {
1439 case TYPE_DISK:
1440 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
1441 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1442 rc16 = scsi_datain_unmarshall(task);
1443 if (rc16 == NULL) {
1444 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity16 data.");
1445 } else {
1446 iscsilun->block_size = rc16->block_length;
1447 iscsilun->num_blocks = rc16->returned_lba + 1;
1448 iscsilun->lbpme = !!rc16->lbpme;
1449 iscsilun->lbprz = !!rc16->lbprz;
1450 iscsilun->use_16_for_rw = (rc16->returned_lba > 0xffffffff);
1452 break;
1454 if (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1455 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION) {
1456 break;
1458 /* Fall through and try READ CAPACITY(10) instead. */
1459 case TYPE_ROM:
1460 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1461 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1462 rc10 = scsi_datain_unmarshall(task);
1463 if (rc10 == NULL) {
1464 error_setg(errp, "iSCSI: Failed to unmarshall readcapacity10 data.");
1465 } else {
1466 iscsilun->block_size = rc10->block_size;
1467 if (rc10->lba == 0) {
1468 /* blank disk loaded */
1469 iscsilun->num_blocks = 0;
1470 } else {
1471 iscsilun->num_blocks = rc10->lba + 1;
1475 break;
1476 default:
1477 return;
1479 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1480 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1481 && retries-- > 0);
1483 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1484 error_setg(errp, "iSCSI: failed to send readcapacity10/16 command");
1485 } else if (!iscsilun->block_size ||
1486 iscsilun->block_size % BDRV_SECTOR_SIZE) {
1487 error_setg(errp, "iSCSI: the target returned an invalid "
1488 "block size of %d.", iscsilun->block_size);
1490 if (task) {
1491 scsi_free_scsi_task(task);
1495 static struct scsi_task *iscsi_do_inquiry(struct iscsi_context *iscsi, int lun,
1496 int evpd, int pc, void **inq, Error **errp)
1498 int full_size;
1499 struct scsi_task *task = NULL;
1500 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, 64);
1501 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1502 goto fail;
1504 full_size = scsi_datain_getfullsize(task);
1505 if (full_size > task->datain.size) {
1506 scsi_free_scsi_task(task);
1508 /* we need more data for the full list */
1509 task = iscsi_inquiry_sync(iscsi, lun, evpd, pc, full_size);
1510 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1511 goto fail;
1515 *inq = scsi_datain_unmarshall(task);
1516 if (*inq == NULL) {
1517 error_setg(errp, "iSCSI: failed to unmarshall inquiry datain blob");
1518 goto fail_with_err;
1521 return task;
1523 fail:
1524 error_setg(errp, "iSCSI: Inquiry command failed : %s",
1525 iscsi_get_error(iscsi));
1526 fail_with_err:
1527 if (task != NULL) {
1528 scsi_free_scsi_task(task);
1530 return NULL;
1533 static void iscsi_detach_aio_context(BlockDriverState *bs)
1535 IscsiLun *iscsilun = bs->opaque;
1537 aio_set_fd_handler(iscsilun->aio_context, iscsi_get_fd(iscsilun->iscsi),
1538 false, NULL, NULL, NULL, NULL, NULL);
1539 iscsilun->events = 0;
1541 if (iscsilun->nop_timer) {
1542 timer_free(iscsilun->nop_timer);
1543 iscsilun->nop_timer = NULL;
1545 if (iscsilun->event_timer) {
1546 timer_free(iscsilun->event_timer);
1547 iscsilun->event_timer = NULL;
1551 static void iscsi_attach_aio_context(BlockDriverState *bs,
1552 AioContext *new_context)
1554 IscsiLun *iscsilun = bs->opaque;
1556 iscsilun->aio_context = new_context;
1557 iscsi_set_events(iscsilun);
1559 /* Set up a timer for sending out iSCSI NOPs */
1560 iscsilun->nop_timer = aio_timer_new(iscsilun->aio_context,
1561 QEMU_CLOCK_REALTIME, SCALE_MS,
1562 iscsi_nop_timed_event, iscsilun);
1563 timer_mod(iscsilun->nop_timer,
1564 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + NOP_INTERVAL);
1566 /* Set up a timer for periodic calls to iscsi_set_events and to
1567 * scan for command timeout */
1568 iscsilun->event_timer = aio_timer_new(iscsilun->aio_context,
1569 QEMU_CLOCK_REALTIME, SCALE_MS,
1570 iscsi_timed_check_events, iscsilun);
1571 timer_mod(iscsilun->event_timer,
1572 qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + EVENT_INTERVAL);
1575 static void iscsi_modesense_sync(IscsiLun *iscsilun)
1577 struct scsi_task *task;
1578 struct scsi_mode_sense *ms = NULL;
1579 iscsilun->write_protected = false;
1580 iscsilun->dpofua = false;
1582 task = iscsi_modesense6_sync(iscsilun->iscsi, iscsilun->lun,
1583 1, SCSI_MODESENSE_PC_CURRENT,
1584 0x3F, 0, 255);
1585 if (task == NULL) {
1586 error_report("iSCSI: Failed to send MODE_SENSE(6) command: %s",
1587 iscsi_get_error(iscsilun->iscsi));
1588 goto out;
1591 if (task->status != SCSI_STATUS_GOOD) {
1592 error_report("iSCSI: Failed MODE_SENSE(6), LUN assumed writable");
1593 goto out;
1595 ms = scsi_datain_unmarshall(task);
1596 if (!ms) {
1597 error_report("iSCSI: Failed to unmarshall MODE_SENSE(6) data: %s",
1598 iscsi_get_error(iscsilun->iscsi));
1599 goto out;
1601 iscsilun->write_protected = ms->device_specific_parameter & 0x80;
1602 iscsilun->dpofua = ms->device_specific_parameter & 0x10;
1604 out:
1605 if (task) {
1606 scsi_free_scsi_task(task);
1610 static void iscsi_parse_iscsi_option(const char *target, QDict *options)
1612 QemuOptsList *list;
1613 QemuOpts *opts;
1614 const char *user, *password, *password_secret, *initiator_name,
1615 *header_digest, *timeout;
1617 list = qemu_find_opts("iscsi");
1618 if (!list) {
1619 return;
1622 opts = qemu_opts_find(list, target);
1623 if (opts == NULL) {
1624 opts = QTAILQ_FIRST(&list->head);
1625 if (!opts) {
1626 return;
1630 user = qemu_opt_get(opts, "user");
1631 if (user) {
1632 qdict_set_default_str(options, "user", user);
1635 password = qemu_opt_get(opts, "password");
1636 if (password) {
1637 qdict_set_default_str(options, "password", password);
1640 password_secret = qemu_opt_get(opts, "password-secret");
1641 if (password_secret) {
1642 qdict_set_default_str(options, "password-secret", password_secret);
1645 initiator_name = qemu_opt_get(opts, "initiator-name");
1646 if (initiator_name) {
1647 qdict_set_default_str(options, "initiator-name", initiator_name);
1650 header_digest = qemu_opt_get(opts, "header-digest");
1651 if (header_digest) {
1652 /* -iscsi takes upper case values, but QAPI only supports lower case
1653 * enum constant names, so we have to convert here. */
1654 char *qapi_value = g_ascii_strdown(header_digest, -1);
1655 qdict_set_default_str(options, "header-digest", qapi_value);
1656 g_free(qapi_value);
1659 timeout = qemu_opt_get(opts, "timeout");
1660 if (timeout) {
1661 qdict_set_default_str(options, "timeout", timeout);
1666 * We support iscsi url's on the form
1667 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1669 static void iscsi_parse_filename(const char *filename, QDict *options,
1670 Error **errp)
1672 struct iscsi_url *iscsi_url;
1673 const char *transport_name;
1674 char *lun_str;
1676 iscsi_url = iscsi_parse_full_url(NULL, filename);
1677 if (iscsi_url == NULL) {
1678 error_setg(errp, "Failed to parse URL : %s", filename);
1679 return;
1682 #if LIBISCSI_API_VERSION >= (20160603)
1683 switch (iscsi_url->transport) {
1684 case TCP_TRANSPORT:
1685 transport_name = "tcp";
1686 break;
1687 case ISER_TRANSPORT:
1688 transport_name = "iser";
1689 break;
1690 default:
1691 error_setg(errp, "Unknown transport type (%d)",
1692 iscsi_url->transport);
1693 return;
1695 #else
1696 transport_name = "tcp";
1697 #endif
1699 qdict_set_default_str(options, "transport", transport_name);
1700 qdict_set_default_str(options, "portal", iscsi_url->portal);
1701 qdict_set_default_str(options, "target", iscsi_url->target);
1703 lun_str = g_strdup_printf("%d", iscsi_url->lun);
1704 qdict_set_default_str(options, "lun", lun_str);
1705 g_free(lun_str);
1707 /* User/password from -iscsi take precedence over those from the URL */
1708 iscsi_parse_iscsi_option(iscsi_url->target, options);
1710 if (iscsi_url->user[0] != '\0') {
1711 qdict_set_default_str(options, "user", iscsi_url->user);
1712 qdict_set_default_str(options, "password", iscsi_url->passwd);
1715 iscsi_destroy_url(iscsi_url);
1718 static QemuOptsList runtime_opts = {
1719 .name = "iscsi",
1720 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1721 .desc = {
1723 .name = "transport",
1724 .type = QEMU_OPT_STRING,
1727 .name = "portal",
1728 .type = QEMU_OPT_STRING,
1731 .name = "target",
1732 .type = QEMU_OPT_STRING,
1735 .name = "user",
1736 .type = QEMU_OPT_STRING,
1739 .name = "password",
1740 .type = QEMU_OPT_STRING,
1743 .name = "password-secret",
1744 .type = QEMU_OPT_STRING,
1747 .name = "lun",
1748 .type = QEMU_OPT_NUMBER,
1751 .name = "initiator-name",
1752 .type = QEMU_OPT_STRING,
1755 .name = "header-digest",
1756 .type = QEMU_OPT_STRING,
1759 .name = "timeout",
1760 .type = QEMU_OPT_NUMBER,
1762 { /* end of list */ }
1766 static void iscsi_save_designator(IscsiLun *lun,
1767 struct scsi_inquiry_device_identification *inq_di)
1769 struct scsi_inquiry_device_designator *desig, *copy = NULL;
1771 for (desig = inq_di->designators; desig; desig = desig->next) {
1772 if (desig->association ||
1773 desig->designator_type > SCSI_DESIGNATOR_TYPE_NAA) {
1774 continue;
1776 /* NAA works better than T10 vendor ID based designator. */
1777 if (!copy || copy->designator_type < desig->designator_type) {
1778 copy = desig;
1781 if (copy) {
1782 lun->dd = g_new(struct scsi_inquiry_device_designator, 1);
1783 *lun->dd = *copy;
1784 lun->dd->next = NULL;
1785 lun->dd->designator = g_malloc(copy->designator_length);
1786 memcpy(lun->dd->designator, copy->designator, copy->designator_length);
1790 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags,
1791 Error **errp)
1793 IscsiLun *iscsilun = bs->opaque;
1794 struct iscsi_context *iscsi = NULL;
1795 struct scsi_task *task = NULL;
1796 struct scsi_inquiry_standard *inq = NULL;
1797 struct scsi_inquiry_supported_pages *inq_vpd;
1798 char *initiator_name = NULL;
1799 QemuOpts *opts;
1800 Error *local_err = NULL;
1801 const char *transport_name, *portal, *target;
1802 #if LIBISCSI_API_VERSION >= (20160603)
1803 enum iscsi_transport_type transport;
1804 #endif
1805 int i, ret = 0, timeout = 0, lun;
1807 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
1808 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1809 ret = -EINVAL;
1810 goto out;
1813 transport_name = qemu_opt_get(opts, "transport");
1814 portal = qemu_opt_get(opts, "portal");
1815 target = qemu_opt_get(opts, "target");
1816 lun = qemu_opt_get_number(opts, "lun", 0);
1818 if (!transport_name || !portal || !target) {
1819 error_setg(errp, "Need all of transport, portal and target options");
1820 ret = -EINVAL;
1821 goto out;
1824 if (!strcmp(transport_name, "tcp")) {
1825 #if LIBISCSI_API_VERSION >= (20160603)
1826 transport = TCP_TRANSPORT;
1827 } else if (!strcmp(transport_name, "iser")) {
1828 transport = ISER_TRANSPORT;
1829 #else
1830 /* TCP is what older libiscsi versions always use */
1831 #endif
1832 } else {
1833 error_setg(errp, "Unknown transport: %s", transport_name);
1834 ret = -EINVAL;
1835 goto out;
1838 memset(iscsilun, 0, sizeof(IscsiLun));
1840 initiator_name = get_initiator_name(opts);
1842 iscsi = iscsi_create_context(initiator_name);
1843 if (iscsi == NULL) {
1844 error_setg(errp, "iSCSI: Failed to create iSCSI context.");
1845 ret = -ENOMEM;
1846 goto out;
1848 #if LIBISCSI_API_VERSION >= (20160603)
1849 if (iscsi_init_transport(iscsi, transport)) {
1850 error_setg(errp, ("Error initializing transport."));
1851 ret = -EINVAL;
1852 goto out;
1854 #endif
1855 if (iscsi_set_targetname(iscsi, target)) {
1856 error_setg(errp, "iSCSI: Failed to set target name.");
1857 ret = -EINVAL;
1858 goto out;
1861 /* check if we got CHAP username/password via the options */
1862 apply_chap(iscsi, opts, &local_err);
1863 if (local_err != NULL) {
1864 error_propagate(errp, local_err);
1865 ret = -EINVAL;
1866 goto out;
1869 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1870 error_setg(errp, "iSCSI: Failed to set session type to normal.");
1871 ret = -EINVAL;
1872 goto out;
1875 /* check if we got HEADER_DIGEST via the options */
1876 apply_header_digest(iscsi, opts, &local_err);
1877 if (local_err != NULL) {
1878 error_propagate(errp, local_err);
1879 ret = -EINVAL;
1880 goto out;
1883 /* timeout handling is broken in libiscsi before 1.15.0 */
1884 timeout = qemu_opt_get_number(opts, "timeout", 0);
1885 #if LIBISCSI_API_VERSION >= 20150621
1886 iscsi_set_timeout(iscsi, timeout);
1887 #else
1888 if (timeout) {
1889 warn_report("iSCSI: ignoring timeout value for libiscsi <1.15.0");
1891 #endif
1893 if (iscsi_full_connect_sync(iscsi, portal, lun) != 0) {
1894 error_setg(errp, "iSCSI: Failed to connect to LUN : %s",
1895 iscsi_get_error(iscsi));
1896 ret = -EINVAL;
1897 goto out;
1900 iscsilun->iscsi = iscsi;
1901 iscsilun->aio_context = bdrv_get_aio_context(bs);
1902 iscsilun->lun = lun;
1903 iscsilun->has_write_same = true;
1905 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 0, 0,
1906 (void **) &inq, errp);
1907 if (task == NULL) {
1908 ret = -EINVAL;
1909 goto out;
1911 iscsilun->type = inq->periperal_device_type;
1912 scsi_free_scsi_task(task);
1913 task = NULL;
1915 iscsi_modesense_sync(iscsilun);
1916 if (iscsilun->dpofua) {
1917 bs->supported_write_flags = BDRV_REQ_FUA;
1920 /* Check the write protect flag of the LUN if we want to write */
1921 if (iscsilun->type == TYPE_DISK && (flags & BDRV_O_RDWR) &&
1922 iscsilun->write_protected) {
1923 ret = bdrv_apply_auto_read_only(bs, "LUN is write protected", errp);
1924 if (ret < 0) {
1925 goto out;
1927 flags &= ~BDRV_O_RDWR;
1930 iscsi_readcapacity_sync(iscsilun, &local_err);
1931 if (local_err != NULL) {
1932 error_propagate(errp, local_err);
1933 ret = -EINVAL;
1934 goto out;
1936 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1938 /* We don't have any emulation for devices other than disks and CD-ROMs, so
1939 * this must be sg ioctl compatible. We force it to be sg, otherwise qemu
1940 * will try to read from the device to guess the image format.
1942 if (iscsilun->type != TYPE_DISK && iscsilun->type != TYPE_ROM) {
1943 bs->sg = true;
1946 task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1947 SCSI_INQUIRY_PAGECODE_SUPPORTED_VPD_PAGES,
1948 (void **) &inq_vpd, errp);
1949 if (task == NULL) {
1950 ret = -EINVAL;
1951 goto out;
1953 for (i = 0; i < inq_vpd->num_pages; i++) {
1954 struct scsi_task *inq_task;
1955 struct scsi_inquiry_logical_block_provisioning *inq_lbp;
1956 struct scsi_inquiry_block_limits *inq_bl;
1957 struct scsi_inquiry_device_identification *inq_di;
1958 switch (inq_vpd->pages[i]) {
1959 case SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING:
1960 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1961 SCSI_INQUIRY_PAGECODE_LOGICAL_BLOCK_PROVISIONING,
1962 (void **) &inq_lbp, errp);
1963 if (inq_task == NULL) {
1964 ret = -EINVAL;
1965 goto out;
1967 memcpy(&iscsilun->lbp, inq_lbp,
1968 sizeof(struct scsi_inquiry_logical_block_provisioning));
1969 scsi_free_scsi_task(inq_task);
1970 break;
1971 case SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS:
1972 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1973 SCSI_INQUIRY_PAGECODE_BLOCK_LIMITS,
1974 (void **) &inq_bl, errp);
1975 if (inq_task == NULL) {
1976 ret = -EINVAL;
1977 goto out;
1979 memcpy(&iscsilun->bl, inq_bl,
1980 sizeof(struct scsi_inquiry_block_limits));
1981 scsi_free_scsi_task(inq_task);
1982 break;
1983 case SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION:
1984 inq_task = iscsi_do_inquiry(iscsilun->iscsi, iscsilun->lun, 1,
1985 SCSI_INQUIRY_PAGECODE_DEVICE_IDENTIFICATION,
1986 (void **) &inq_di, errp);
1987 if (inq_task == NULL) {
1988 ret = -EINVAL;
1989 goto out;
1991 iscsi_save_designator(iscsilun, inq_di);
1992 scsi_free_scsi_task(inq_task);
1993 break;
1994 default:
1995 break;
1998 scsi_free_scsi_task(task);
1999 task = NULL;
2001 qemu_mutex_init(&iscsilun->mutex);
2002 iscsi_attach_aio_context(bs, iscsilun->aio_context);
2004 /* Guess the internal cluster (page) size of the iscsi target by the means
2005 * of opt_unmap_gran. Transfer the unmap granularity only if it has a
2006 * reasonable size */
2007 if (iscsilun->bl.opt_unmap_gran * iscsilun->block_size >= 4 * 1024 &&
2008 iscsilun->bl.opt_unmap_gran * iscsilun->block_size <= 16 * 1024 * 1024) {
2009 iscsilun->cluster_size = iscsilun->bl.opt_unmap_gran *
2010 iscsilun->block_size;
2011 if (iscsilun->lbprz) {
2012 ret = iscsi_allocmap_init(iscsilun, flags);
2016 if (iscsilun->lbprz && iscsilun->lbp.lbpws) {
2017 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
2020 out:
2021 qemu_opts_del(opts);
2022 g_free(initiator_name);
2023 if (task != NULL) {
2024 scsi_free_scsi_task(task);
2027 if (ret) {
2028 if (iscsi != NULL) {
2029 if (iscsi_is_logged_in(iscsi)) {
2030 iscsi_logout_sync(iscsi);
2032 iscsi_destroy_context(iscsi);
2034 memset(iscsilun, 0, sizeof(IscsiLun));
2037 return ret;
2040 static void iscsi_close(BlockDriverState *bs)
2042 IscsiLun *iscsilun = bs->opaque;
2043 struct iscsi_context *iscsi = iscsilun->iscsi;
2045 iscsi_detach_aio_context(bs);
2046 if (iscsi_is_logged_in(iscsi)) {
2047 iscsi_logout_sync(iscsi);
2049 iscsi_destroy_context(iscsi);
2050 if (iscsilun->dd) {
2051 g_free(iscsilun->dd->designator);
2052 g_free(iscsilun->dd);
2054 g_free(iscsilun->zeroblock);
2055 iscsi_allocmap_free(iscsilun);
2056 qemu_mutex_destroy(&iscsilun->mutex);
2057 memset(iscsilun, 0, sizeof(IscsiLun));
2060 static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
2062 /* We don't actually refresh here, but just return data queried in
2063 * iscsi_open(): iscsi targets don't change their limits. */
2065 IscsiLun *iscsilun = bs->opaque;
2066 uint64_t max_xfer_len = iscsilun->use_16_for_rw ? 0xffffffff : 0xffff;
2067 unsigned int block_size = MAX(BDRV_SECTOR_SIZE, iscsilun->block_size);
2069 assert(iscsilun->block_size >= BDRV_SECTOR_SIZE || bdrv_is_sg(bs));
2071 bs->bl.request_alignment = block_size;
2073 if (iscsilun->bl.max_xfer_len) {
2074 max_xfer_len = MIN(max_xfer_len, iscsilun->bl.max_xfer_len);
2077 if (max_xfer_len * block_size < INT_MAX) {
2078 bs->bl.max_transfer = max_xfer_len * iscsilun->block_size;
2081 if (iscsilun->lbp.lbpu) {
2082 bs->bl.max_pdiscard =
2083 MIN_NON_ZERO(iscsilun->bl.max_unmap * iscsilun->block_size,
2084 (uint64_t)UINT32_MAX * iscsilun->block_size);
2085 bs->bl.pdiscard_alignment =
2086 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2087 } else {
2088 bs->bl.pdiscard_alignment = iscsilun->block_size;
2091 bs->bl.max_pwrite_zeroes =
2092 MIN_NON_ZERO(iscsilun->bl.max_ws_len * iscsilun->block_size,
2093 max_xfer_len * iscsilun->block_size);
2095 if (iscsilun->lbp.lbpws) {
2096 bs->bl.pwrite_zeroes_alignment =
2097 iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
2098 } else {
2099 bs->bl.pwrite_zeroes_alignment = iscsilun->block_size;
2101 if (iscsilun->bl.opt_xfer_len &&
2102 iscsilun->bl.opt_xfer_len < INT_MAX / block_size) {
2103 bs->bl.opt_transfer = pow2floor(iscsilun->bl.opt_xfer_len *
2104 iscsilun->block_size);
2108 /* Note that this will not re-establish a connection with an iSCSI target - it
2109 * is effectively a NOP. */
2110 static int iscsi_reopen_prepare(BDRVReopenState *state,
2111 BlockReopenQueue *queue, Error **errp)
2113 IscsiLun *iscsilun = state->bs->opaque;
2115 if (state->flags & BDRV_O_RDWR && iscsilun->write_protected) {
2116 error_setg(errp, "Cannot open a write protected LUN as read-write");
2117 return -EACCES;
2119 return 0;
2122 static void iscsi_reopen_commit(BDRVReopenState *reopen_state)
2124 IscsiLun *iscsilun = reopen_state->bs->opaque;
2126 /* the cache.direct status might have changed */
2127 if (iscsilun->allocmap != NULL) {
2128 iscsi_allocmap_init(iscsilun, reopen_state->flags);
2132 static int coroutine_fn iscsi_co_truncate(BlockDriverState *bs, int64_t offset,
2133 bool exact, PreallocMode prealloc,
2134 BdrvRequestFlags flags, Error **errp)
2136 IscsiLun *iscsilun = bs->opaque;
2137 int64_t cur_length;
2138 Error *local_err = NULL;
2140 if (prealloc != PREALLOC_MODE_OFF) {
2141 error_setg(errp, "Unsupported preallocation mode '%s'",
2142 PreallocMode_str(prealloc));
2143 return -ENOTSUP;
2146 if (iscsilun->type != TYPE_DISK) {
2147 error_setg(errp, "Cannot resize non-disk iSCSI devices");
2148 return -ENOTSUP;
2151 iscsi_readcapacity_sync(iscsilun, &local_err);
2152 if (local_err != NULL) {
2153 error_propagate(errp, local_err);
2154 return -EIO;
2157 cur_length = iscsi_getlength(bs);
2158 if (offset != cur_length && exact) {
2159 error_setg(errp, "Cannot resize iSCSI devices");
2160 return -ENOTSUP;
2161 } else if (offset > cur_length) {
2162 error_setg(errp, "Cannot grow iSCSI devices");
2163 return -EINVAL;
2166 if (iscsilun->allocmap != NULL) {
2167 iscsi_allocmap_init(iscsilun, bs->open_flags);
2170 return 0;
2173 static int iscsi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2175 IscsiLun *iscsilun = bs->opaque;
2176 bdi->cluster_size = iscsilun->cluster_size;
2177 return 0;
2180 static void coroutine_fn iscsi_co_invalidate_cache(BlockDriverState *bs,
2181 Error **errp)
2183 IscsiLun *iscsilun = bs->opaque;
2184 iscsi_allocmap_invalidate(iscsilun);
2187 static int coroutine_fn iscsi_co_copy_range_from(BlockDriverState *bs,
2188 BdrvChild *src,
2189 int64_t src_offset,
2190 BdrvChild *dst,
2191 int64_t dst_offset,
2192 int64_t bytes,
2193 BdrvRequestFlags read_flags,
2194 BdrvRequestFlags write_flags)
2196 return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes,
2197 read_flags, write_flags);
2200 static struct scsi_task *iscsi_xcopy_task(int param_len)
2202 struct scsi_task *task;
2204 task = g_new0(struct scsi_task, 1);
2206 task->cdb[0] = EXTENDED_COPY;
2207 task->cdb[10] = (param_len >> 24) & 0xFF;
2208 task->cdb[11] = (param_len >> 16) & 0xFF;
2209 task->cdb[12] = (param_len >> 8) & 0xFF;
2210 task->cdb[13] = param_len & 0xFF;
2211 task->cdb_size = 16;
2212 task->xfer_dir = SCSI_XFER_WRITE;
2213 task->expxferlen = param_len;
2215 return task;
2218 static void iscsi_populate_target_desc(unsigned char *desc, IscsiLun *lun)
2220 struct scsi_inquiry_device_designator *dd = lun->dd;
2222 memset(desc, 0, 32);
2223 desc[0] = 0xE4; /* IDENT_DESCR_TGT_DESCR */
2224 desc[4] = dd->code_set;
2225 desc[5] = (dd->designator_type & 0xF)
2226 | ((dd->association & 3) << 4);
2227 desc[7] = dd->designator_length;
2228 memcpy(desc + 8, dd->designator, MIN(dd->designator_length, 20));
2230 desc[28] = 0;
2231 desc[29] = (lun->block_size >> 16) & 0xFF;
2232 desc[30] = (lun->block_size >> 8) & 0xFF;
2233 desc[31] = lun->block_size & 0xFF;
2236 static void iscsi_xcopy_desc_hdr(uint8_t *hdr, int dc, int cat, int src_index,
2237 int dst_index)
2239 hdr[0] = 0x02; /* BLK_TO_BLK_SEG_DESCR */
2240 hdr[1] = ((dc << 1) | cat) & 0xFF;
2241 hdr[2] = (XCOPY_BLK2BLK_SEG_DESC_SIZE >> 8) & 0xFF;
2242 /* don't account for the first 4 bytes in descriptor header*/
2243 hdr[3] = (XCOPY_BLK2BLK_SEG_DESC_SIZE - 4 /* SEG_DESC_SRC_INDEX_OFFSET */) & 0xFF;
2244 hdr[4] = (src_index >> 8) & 0xFF;
2245 hdr[5] = src_index & 0xFF;
2246 hdr[6] = (dst_index >> 8) & 0xFF;
2247 hdr[7] = dst_index & 0xFF;
2250 static void iscsi_xcopy_populate_desc(uint8_t *desc, int dc, int cat,
2251 int src_index, int dst_index, int num_blks,
2252 uint64_t src_lba, uint64_t dst_lba)
2254 iscsi_xcopy_desc_hdr(desc, dc, cat, src_index, dst_index);
2256 /* The caller should verify the request size */
2257 assert(num_blks < 65536);
2258 desc[10] = (num_blks >> 8) & 0xFF;
2259 desc[11] = num_blks & 0xFF;
2260 desc[12] = (src_lba >> 56) & 0xFF;
2261 desc[13] = (src_lba >> 48) & 0xFF;
2262 desc[14] = (src_lba >> 40) & 0xFF;
2263 desc[15] = (src_lba >> 32) & 0xFF;
2264 desc[16] = (src_lba >> 24) & 0xFF;
2265 desc[17] = (src_lba >> 16) & 0xFF;
2266 desc[18] = (src_lba >> 8) & 0xFF;
2267 desc[19] = src_lba & 0xFF;
2268 desc[20] = (dst_lba >> 56) & 0xFF;
2269 desc[21] = (dst_lba >> 48) & 0xFF;
2270 desc[22] = (dst_lba >> 40) & 0xFF;
2271 desc[23] = (dst_lba >> 32) & 0xFF;
2272 desc[24] = (dst_lba >> 24) & 0xFF;
2273 desc[25] = (dst_lba >> 16) & 0xFF;
2274 desc[26] = (dst_lba >> 8) & 0xFF;
2275 desc[27] = dst_lba & 0xFF;
2278 static void iscsi_xcopy_populate_header(unsigned char *buf, int list_id, int str,
2279 int list_id_usage, int prio,
2280 int tgt_desc_len,
2281 int seg_desc_len, int inline_data_len)
2283 buf[0] = list_id;
2284 buf[1] = ((str & 1) << 5) | ((list_id_usage & 3) << 3) | (prio & 7);
2285 buf[2] = (tgt_desc_len >> 8) & 0xFF;
2286 buf[3] = tgt_desc_len & 0xFF;
2287 buf[8] = (seg_desc_len >> 24) & 0xFF;
2288 buf[9] = (seg_desc_len >> 16) & 0xFF;
2289 buf[10] = (seg_desc_len >> 8) & 0xFF;
2290 buf[11] = seg_desc_len & 0xFF;
2291 buf[12] = (inline_data_len >> 24) & 0xFF;
2292 buf[13] = (inline_data_len >> 16) & 0xFF;
2293 buf[14] = (inline_data_len >> 8) & 0xFF;
2294 buf[15] = inline_data_len & 0xFF;
2297 static void iscsi_xcopy_data(struct iscsi_data *data,
2298 IscsiLun *src, int64_t src_lba,
2299 IscsiLun *dst, int64_t dst_lba,
2300 uint16_t num_blocks)
2302 uint8_t *buf;
2303 const int src_offset = XCOPY_DESC_OFFSET;
2304 const int dst_offset = XCOPY_DESC_OFFSET + IDENT_DESCR_TGT_DESCR_SIZE;
2305 const int seg_offset = dst_offset + IDENT_DESCR_TGT_DESCR_SIZE;
2307 data->size = XCOPY_DESC_OFFSET +
2308 IDENT_DESCR_TGT_DESCR_SIZE * 2 +
2309 XCOPY_BLK2BLK_SEG_DESC_SIZE;
2310 data->data = g_malloc0(data->size);
2311 buf = data->data;
2313 /* Initialise the parameter list header */
2314 iscsi_xcopy_populate_header(buf, 1, 0, 2 /* LIST_ID_USAGE_DISCARD */,
2315 0, 2 * IDENT_DESCR_TGT_DESCR_SIZE,
2316 XCOPY_BLK2BLK_SEG_DESC_SIZE,
2319 /* Initialise CSCD list with one src + one dst descriptor */
2320 iscsi_populate_target_desc(&buf[src_offset], src);
2321 iscsi_populate_target_desc(&buf[dst_offset], dst);
2323 /* Initialise one segment descriptor */
2324 iscsi_xcopy_populate_desc(&buf[seg_offset], 0, 0, 0, 1, num_blocks,
2325 src_lba, dst_lba);
2328 static int coroutine_fn iscsi_co_copy_range_to(BlockDriverState *bs,
2329 BdrvChild *src,
2330 int64_t src_offset,
2331 BdrvChild *dst,
2332 int64_t dst_offset,
2333 int64_t bytes,
2334 BdrvRequestFlags read_flags,
2335 BdrvRequestFlags write_flags)
2337 IscsiLun *dst_lun = dst->bs->opaque;
2338 IscsiLun *src_lun;
2339 struct IscsiTask iscsi_task;
2340 struct iscsi_data data;
2341 int r = 0;
2342 int block_size;
2344 if (src->bs->drv->bdrv_co_copy_range_to != iscsi_co_copy_range_to) {
2345 return -ENOTSUP;
2347 src_lun = src->bs->opaque;
2349 if (!src_lun->dd || !dst_lun->dd) {
2350 return -ENOTSUP;
2352 if (!is_byte_request_lun_aligned(dst_offset, bytes, dst_lun)) {
2353 return -ENOTSUP;
2355 if (!is_byte_request_lun_aligned(src_offset, bytes, src_lun)) {
2356 return -ENOTSUP;
2358 if (dst_lun->block_size != src_lun->block_size ||
2359 !dst_lun->block_size) {
2360 return -ENOTSUP;
2363 block_size = dst_lun->block_size;
2364 if (bytes / block_size > 65535) {
2365 return -ENOTSUP;
2368 iscsi_xcopy_data(&data,
2369 src_lun, src_offset / block_size,
2370 dst_lun, dst_offset / block_size,
2371 bytes / block_size);
2373 iscsi_co_init_iscsitask(dst_lun, &iscsi_task);
2375 qemu_mutex_lock(&dst_lun->mutex);
2376 iscsi_task.task = iscsi_xcopy_task(data.size);
2377 retry:
2378 if (iscsi_scsi_command_async(dst_lun->iscsi, dst_lun->lun,
2379 iscsi_task.task, iscsi_co_generic_cb,
2380 &data,
2381 &iscsi_task) != 0) {
2382 r = -EIO;
2383 goto out_unlock;
2386 iscsi_co_wait_for_task(&iscsi_task, dst_lun);
2388 if (iscsi_task.do_retry) {
2389 iscsi_task.complete = 0;
2390 goto retry;
2393 if (iscsi_task.status != SCSI_STATUS_GOOD) {
2394 r = iscsi_task.err_code;
2395 goto out_unlock;
2398 out_unlock:
2400 trace_iscsi_xcopy(src_lun, src_offset, dst_lun, dst_offset, bytes, r);
2401 g_free(iscsi_task.task);
2402 qemu_mutex_unlock(&dst_lun->mutex);
2403 g_free(iscsi_task.err_str);
2404 return r;
2408 static const char *const iscsi_strong_runtime_opts[] = {
2409 "transport",
2410 "portal",
2411 "target",
2412 "user",
2413 "password",
2414 "password-secret",
2415 "lun",
2416 "initiator-name",
2417 "header-digest",
2419 NULL
2422 static BlockDriver bdrv_iscsi = {
2423 .format_name = "iscsi",
2424 .protocol_name = "iscsi",
2426 .instance_size = sizeof(IscsiLun),
2427 .bdrv_parse_filename = iscsi_parse_filename,
2428 .bdrv_file_open = iscsi_open,
2429 .bdrv_close = iscsi_close,
2430 .bdrv_co_create_opts = bdrv_co_create_opts_simple,
2431 .create_opts = &bdrv_create_opts_simple,
2432 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2433 .bdrv_reopen_commit = iscsi_reopen_commit,
2434 .bdrv_co_invalidate_cache = iscsi_co_invalidate_cache,
2436 .bdrv_getlength = iscsi_getlength,
2437 .bdrv_get_info = iscsi_get_info,
2438 .bdrv_co_truncate = iscsi_co_truncate,
2439 .bdrv_refresh_limits = iscsi_refresh_limits,
2441 .bdrv_co_block_status = iscsi_co_block_status,
2442 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2443 .bdrv_co_copy_range_from = iscsi_co_copy_range_from,
2444 .bdrv_co_copy_range_to = iscsi_co_copy_range_to,
2445 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2446 .bdrv_co_readv = iscsi_co_readv,
2447 .bdrv_co_writev = iscsi_co_writev,
2448 .bdrv_co_flush_to_disk = iscsi_co_flush,
2450 #ifdef __linux__
2451 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2452 #endif
2454 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2455 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2457 .strong_runtime_opts = iscsi_strong_runtime_opts,
2460 #if LIBISCSI_API_VERSION >= (20160603)
2461 static BlockDriver bdrv_iser = {
2462 .format_name = "iser",
2463 .protocol_name = "iser",
2465 .instance_size = sizeof(IscsiLun),
2466 .bdrv_parse_filename = iscsi_parse_filename,
2467 .bdrv_file_open = iscsi_open,
2468 .bdrv_close = iscsi_close,
2469 .bdrv_co_create_opts = bdrv_co_create_opts_simple,
2470 .create_opts = &bdrv_create_opts_simple,
2471 .bdrv_reopen_prepare = iscsi_reopen_prepare,
2472 .bdrv_reopen_commit = iscsi_reopen_commit,
2473 .bdrv_co_invalidate_cache = iscsi_co_invalidate_cache,
2475 .bdrv_getlength = iscsi_getlength,
2476 .bdrv_get_info = iscsi_get_info,
2477 .bdrv_co_truncate = iscsi_co_truncate,
2478 .bdrv_refresh_limits = iscsi_refresh_limits,
2480 .bdrv_co_block_status = iscsi_co_block_status,
2481 .bdrv_co_pdiscard = iscsi_co_pdiscard,
2482 .bdrv_co_copy_range_from = iscsi_co_copy_range_from,
2483 .bdrv_co_copy_range_to = iscsi_co_copy_range_to,
2484 .bdrv_co_pwrite_zeroes = iscsi_co_pwrite_zeroes,
2485 .bdrv_co_readv = iscsi_co_readv,
2486 .bdrv_co_writev = iscsi_co_writev,
2487 .bdrv_co_flush_to_disk = iscsi_co_flush,
2489 #ifdef __linux__
2490 .bdrv_aio_ioctl = iscsi_aio_ioctl,
2491 #endif
2493 .bdrv_detach_aio_context = iscsi_detach_aio_context,
2494 .bdrv_attach_aio_context = iscsi_attach_aio_context,
2496 .strong_runtime_opts = iscsi_strong_runtime_opts,
2498 #endif
2500 static void iscsi_block_init(void)
2502 bdrv_register(&bdrv_iscsi);
2503 #if LIBISCSI_API_VERSION >= (20160603)
2504 bdrv_register(&bdrv_iser);
2505 #endif
2508 block_init(iscsi_block_init);