]> git.proxmox.com Git - qemu.git/blob - block/iscsi.c
rbd: avoid qemu_rbd_snap_list() memory leaks
[qemu.git] / block / iscsi.c
1 /*
2 * QEMU Block driver for iSCSI images
3 *
4 * Copyright (c) 2010-2011 Ronnie Sahlberg <ronniesahlberg@gmail.com>
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "config-host.h"
26
27 #include <poll.h>
28 #include <arpa/inet.h>
29 #include "qemu-common.h"
30 #include "qemu/config-file.h"
31 #include "qemu/error-report.h"
32 #include "block/block_int.h"
33 #include "trace.h"
34 #include "block/scsi.h"
35 #include "qemu/iov.h"
36
37 #include <iscsi/iscsi.h>
38 #include <iscsi/scsi-lowlevel.h>
39
40 #ifdef __linux__
41 #include <scsi/sg.h>
42 #include <block/scsi.h>
43 #endif
44
45 typedef struct IscsiLun {
46 struct iscsi_context *iscsi;
47 int lun;
48 enum scsi_inquiry_peripheral_device_type type;
49 int block_size;
50 uint64_t num_blocks;
51 int events;
52 QEMUTimer *nop_timer;
53 } IscsiLun;
54
55 typedef struct IscsiAIOCB {
56 BlockDriverAIOCB common;
57 QEMUIOVector *qiov;
58 QEMUBH *bh;
59 IscsiLun *iscsilun;
60 struct scsi_task *task;
61 uint8_t *buf;
62 int status;
63 int canceled;
64 int retries;
65 int64_t sector_num;
66 int nb_sectors;
67 #ifdef __linux__
68 sg_io_hdr_t *ioh;
69 #endif
70 } IscsiAIOCB;
71
72 #define NOP_INTERVAL 5000
73 #define MAX_NOP_FAILURES 3
74 #define ISCSI_CMD_RETRIES 5
75
76 static void
77 iscsi_bh_cb(void *p)
78 {
79 IscsiAIOCB *acb = p;
80
81 qemu_bh_delete(acb->bh);
82
83 g_free(acb->buf);
84 acb->buf = NULL;
85
86 if (acb->canceled == 0) {
87 acb->common.cb(acb->common.opaque, acb->status);
88 }
89
90 if (acb->task != NULL) {
91 scsi_free_scsi_task(acb->task);
92 acb->task = NULL;
93 }
94
95 qemu_aio_release(acb);
96 }
97
98 static void
99 iscsi_schedule_bh(IscsiAIOCB *acb)
100 {
101 if (acb->bh) {
102 return;
103 }
104 acb->bh = qemu_bh_new(iscsi_bh_cb, acb);
105 qemu_bh_schedule(acb->bh);
106 }
107
108
109 static void
110 iscsi_abort_task_cb(struct iscsi_context *iscsi, int status, void *command_data,
111 void *private_data)
112 {
113 IscsiAIOCB *acb = private_data;
114
115 acb->status = -ECANCELED;
116 iscsi_schedule_bh(acb);
117 }
118
119 static void
120 iscsi_aio_cancel(BlockDriverAIOCB *blockacb)
121 {
122 IscsiAIOCB *acb = (IscsiAIOCB *)blockacb;
123 IscsiLun *iscsilun = acb->iscsilun;
124
125 if (acb->status != -EINPROGRESS) {
126 return;
127 }
128
129 acb->canceled = 1;
130
131 /* send a task mgmt call to the target to cancel the task on the target */
132 iscsi_task_mgmt_abort_task_async(iscsilun->iscsi, acb->task,
133 iscsi_abort_task_cb, acb);
134
135 while (acb->status == -EINPROGRESS) {
136 qemu_aio_wait();
137 }
138 }
139
140 static const AIOCBInfo iscsi_aiocb_info = {
141 .aiocb_size = sizeof(IscsiAIOCB),
142 .cancel = iscsi_aio_cancel,
143 };
144
145
146 static void iscsi_process_read(void *arg);
147 static void iscsi_process_write(void *arg);
148
149 static int iscsi_process_flush(void *arg)
150 {
151 IscsiLun *iscsilun = arg;
152
153 return iscsi_queue_length(iscsilun->iscsi) > 0;
154 }
155
156 static void
157 iscsi_set_events(IscsiLun *iscsilun)
158 {
159 struct iscsi_context *iscsi = iscsilun->iscsi;
160 int ev;
161
162 /* We always register a read handler. */
163 ev = POLLIN;
164 ev |= iscsi_which_events(iscsi);
165 if (ev != iscsilun->events) {
166 qemu_aio_set_fd_handler(iscsi_get_fd(iscsi),
167 iscsi_process_read,
168 (ev & POLLOUT) ? iscsi_process_write : NULL,
169 iscsi_process_flush,
170 iscsilun);
171
172 }
173
174 iscsilun->events = ev;
175 }
176
177 static void
178 iscsi_process_read(void *arg)
179 {
180 IscsiLun *iscsilun = arg;
181 struct iscsi_context *iscsi = iscsilun->iscsi;
182
183 iscsi_service(iscsi, POLLIN);
184 iscsi_set_events(iscsilun);
185 }
186
187 static void
188 iscsi_process_write(void *arg)
189 {
190 IscsiLun *iscsilun = arg;
191 struct iscsi_context *iscsi = iscsilun->iscsi;
192
193 iscsi_service(iscsi, POLLOUT);
194 iscsi_set_events(iscsilun);
195 }
196
197 static int
198 iscsi_aio_writev_acb(IscsiAIOCB *acb);
199
200 static void
201 iscsi_aio_write16_cb(struct iscsi_context *iscsi, int status,
202 void *command_data, void *opaque)
203 {
204 IscsiAIOCB *acb = opaque;
205
206 trace_iscsi_aio_write16_cb(iscsi, status, acb, acb->canceled);
207
208 g_free(acb->buf);
209 acb->buf = NULL;
210
211 if (acb->canceled != 0) {
212 return;
213 }
214
215 acb->status = 0;
216 if (status != 0) {
217 if (status == SCSI_STATUS_CHECK_CONDITION
218 && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
219 && acb->retries-- > 0) {
220 scsi_free_scsi_task(acb->task);
221 acb->task = NULL;
222 if (iscsi_aio_writev_acb(acb) == 0) {
223 iscsi_set_events(acb->iscsilun);
224 return;
225 }
226 }
227 error_report("Failed to write16 data to iSCSI lun. %s",
228 iscsi_get_error(iscsi));
229 acb->status = -EIO;
230 }
231
232 iscsi_schedule_bh(acb);
233 }
234
235 static int64_t sector_lun2qemu(int64_t sector, IscsiLun *iscsilun)
236 {
237 return sector * iscsilun->block_size / BDRV_SECTOR_SIZE;
238 }
239
240 static int64_t sector_qemu2lun(int64_t sector, IscsiLun *iscsilun)
241 {
242 return sector * BDRV_SECTOR_SIZE / iscsilun->block_size;
243 }
244
245 static bool is_request_lun_aligned(int64_t sector_num, int nb_sectors,
246 IscsiLun *iscsilun)
247 {
248 if ((sector_num * BDRV_SECTOR_SIZE) % iscsilun->block_size ||
249 (nb_sectors * BDRV_SECTOR_SIZE) % iscsilun->block_size) {
250 error_report("iSCSI misaligned request: "
251 "iscsilun->block_size %u, sector_num %" PRIi64
252 ", nb_sectors %d",
253 iscsilun->block_size, sector_num, nb_sectors);
254 return 0;
255 }
256 return 1;
257 }
258
259 static int
260 iscsi_aio_writev_acb(IscsiAIOCB *acb)
261 {
262 struct iscsi_context *iscsi = acb->iscsilun->iscsi;
263 size_t size;
264 uint32_t num_sectors;
265 uint64_t lba;
266 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
267 struct iscsi_data data;
268 #endif
269 int ret;
270
271 acb->canceled = 0;
272 acb->bh = NULL;
273 acb->status = -EINPROGRESS;
274 acb->buf = NULL;
275
276 /* this will allow us to get rid of 'buf' completely */
277 size = acb->nb_sectors * BDRV_SECTOR_SIZE;
278
279 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
280 data.size = MIN(size, acb->qiov->size);
281
282 /* if the iovec only contains one buffer we can pass it directly */
283 if (acb->qiov->niov == 1) {
284 data.data = acb->qiov->iov[0].iov_base;
285 } else {
286 acb->buf = g_malloc(data.size);
287 qemu_iovec_to_buf(acb->qiov, 0, acb->buf, data.size);
288 data.data = acb->buf;
289 }
290 #endif
291
292 acb->task = malloc(sizeof(struct scsi_task));
293 if (acb->task == NULL) {
294 error_report("iSCSI: Failed to allocate task for scsi WRITE16 "
295 "command. %s", iscsi_get_error(iscsi));
296 return -1;
297 }
298 memset(acb->task, 0, sizeof(struct scsi_task));
299
300 acb->task->xfer_dir = SCSI_XFER_WRITE;
301 acb->task->cdb_size = 16;
302 acb->task->cdb[0] = 0x8a;
303 lba = sector_qemu2lun(acb->sector_num, acb->iscsilun);
304 *(uint32_t *)&acb->task->cdb[2] = htonl(lba >> 32);
305 *(uint32_t *)&acb->task->cdb[6] = htonl(lba & 0xffffffff);
306 num_sectors = sector_qemu2lun(acb->nb_sectors, acb->iscsilun);
307 *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
308 acb->task->expxferlen = size;
309
310 #if defined(LIBISCSI_FEATURE_IOVECTOR)
311 ret = iscsi_scsi_command_async(iscsi, acb->iscsilun->lun, acb->task,
312 iscsi_aio_write16_cb,
313 NULL,
314 acb);
315 #else
316 ret = iscsi_scsi_command_async(iscsi, acb->iscsilun->lun, acb->task,
317 iscsi_aio_write16_cb,
318 &data,
319 acb);
320 #endif
321 if (ret != 0) {
322 scsi_free_scsi_task(acb->task);
323 g_free(acb->buf);
324 return -1;
325 }
326
327 #if defined(LIBISCSI_FEATURE_IOVECTOR)
328 scsi_task_set_iov_out(acb->task, (struct scsi_iovec*) acb->qiov->iov, acb->qiov->niov);
329 #endif
330
331 return 0;
332 }
333
334 static BlockDriverAIOCB *
335 iscsi_aio_writev(BlockDriverState *bs, int64_t sector_num,
336 QEMUIOVector *qiov, int nb_sectors,
337 BlockDriverCompletionFunc *cb,
338 void *opaque)
339 {
340 IscsiLun *iscsilun = bs->opaque;
341 IscsiAIOCB *acb;
342
343 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
344 return NULL;
345 }
346
347 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
348 trace_iscsi_aio_writev(iscsilun->iscsi, sector_num, nb_sectors, opaque, acb);
349
350 acb->iscsilun = iscsilun;
351 acb->qiov = qiov;
352 acb->nb_sectors = nb_sectors;
353 acb->sector_num = sector_num;
354 acb->retries = ISCSI_CMD_RETRIES;
355
356 if (iscsi_aio_writev_acb(acb) != 0) {
357 qemu_aio_release(acb);
358 return NULL;
359 }
360
361 iscsi_set_events(iscsilun);
362 return &acb->common;
363 }
364
365 static int
366 iscsi_aio_readv_acb(IscsiAIOCB *acb);
367
368 static void
369 iscsi_aio_read16_cb(struct iscsi_context *iscsi, int status,
370 void *command_data, void *opaque)
371 {
372 IscsiAIOCB *acb = opaque;
373
374 trace_iscsi_aio_read16_cb(iscsi, status, acb, acb->canceled);
375
376 if (acb->canceled != 0) {
377 return;
378 }
379
380 acb->status = 0;
381 if (status != 0) {
382 if (status == SCSI_STATUS_CHECK_CONDITION
383 && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
384 && acb->retries-- > 0) {
385 scsi_free_scsi_task(acb->task);
386 acb->task = NULL;
387 if (iscsi_aio_readv_acb(acb) == 0) {
388 iscsi_set_events(acb->iscsilun);
389 return;
390 }
391 }
392 error_report("Failed to read16 data from iSCSI lun. %s",
393 iscsi_get_error(iscsi));
394 acb->status = -EIO;
395 }
396
397 iscsi_schedule_bh(acb);
398 }
399
400 static int
401 iscsi_aio_readv_acb(IscsiAIOCB *acb)
402 {
403 struct iscsi_context *iscsi = acb->iscsilun->iscsi;
404 size_t size;
405 uint64_t lba;
406 uint32_t num_sectors;
407 int ret;
408 #if !defined(LIBISCSI_FEATURE_IOVECTOR)
409 int i;
410 #endif
411
412 acb->canceled = 0;
413 acb->bh = NULL;
414 acb->status = -EINPROGRESS;
415 acb->buf = NULL;
416
417 size = acb->nb_sectors * BDRV_SECTOR_SIZE;
418
419 acb->task = malloc(sizeof(struct scsi_task));
420 if (acb->task == NULL) {
421 error_report("iSCSI: Failed to allocate task for scsi READ16 "
422 "command. %s", iscsi_get_error(iscsi));
423 return -1;
424 }
425 memset(acb->task, 0, sizeof(struct scsi_task));
426
427 acb->task->xfer_dir = SCSI_XFER_READ;
428 acb->task->expxferlen = size;
429 lba = sector_qemu2lun(acb->sector_num, acb->iscsilun);
430 num_sectors = sector_qemu2lun(acb->nb_sectors, acb->iscsilun);
431
432 switch (acb->iscsilun->type) {
433 case TYPE_DISK:
434 acb->task->cdb_size = 16;
435 acb->task->cdb[0] = 0x88;
436 *(uint32_t *)&acb->task->cdb[2] = htonl(lba >> 32);
437 *(uint32_t *)&acb->task->cdb[6] = htonl(lba & 0xffffffff);
438 *(uint32_t *)&acb->task->cdb[10] = htonl(num_sectors);
439 break;
440 default:
441 acb->task->cdb_size = 10;
442 acb->task->cdb[0] = 0x28;
443 *(uint32_t *)&acb->task->cdb[2] = htonl(lba);
444 *(uint16_t *)&acb->task->cdb[7] = htons(num_sectors);
445 break;
446 }
447
448 ret = iscsi_scsi_command_async(iscsi, acb->iscsilun->lun, acb->task,
449 iscsi_aio_read16_cb,
450 NULL,
451 acb);
452 if (ret != 0) {
453 scsi_free_scsi_task(acb->task);
454 return -1;
455 }
456
457 #if defined(LIBISCSI_FEATURE_IOVECTOR)
458 scsi_task_set_iov_in(acb->task, (struct scsi_iovec*) acb->qiov->iov, acb->qiov->niov);
459 #else
460 for (i = 0; i < acb->qiov->niov; i++) {
461 scsi_task_add_data_in_buffer(acb->task,
462 acb->qiov->iov[i].iov_len,
463 acb->qiov->iov[i].iov_base);
464 }
465 #endif
466 return 0;
467 }
468
469 static BlockDriverAIOCB *
470 iscsi_aio_readv(BlockDriverState *bs, int64_t sector_num,
471 QEMUIOVector *qiov, int nb_sectors,
472 BlockDriverCompletionFunc *cb,
473 void *opaque)
474 {
475 IscsiLun *iscsilun = bs->opaque;
476 IscsiAIOCB *acb;
477
478 if (!is_request_lun_aligned(sector_num, nb_sectors, iscsilun)) {
479 return NULL;
480 }
481
482 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
483 trace_iscsi_aio_readv(iscsilun->iscsi, sector_num, nb_sectors, opaque, acb);
484
485 acb->nb_sectors = nb_sectors;
486 acb->sector_num = sector_num;
487 acb->iscsilun = iscsilun;
488 acb->qiov = qiov;
489 acb->retries = ISCSI_CMD_RETRIES;
490
491 if (iscsi_aio_readv_acb(acb) != 0) {
492 qemu_aio_release(acb);
493 return NULL;
494 }
495
496 iscsi_set_events(iscsilun);
497 return &acb->common;
498 }
499
500 static int
501 iscsi_aio_flush_acb(IscsiAIOCB *acb);
502
503 static void
504 iscsi_synccache10_cb(struct iscsi_context *iscsi, int status,
505 void *command_data, void *opaque)
506 {
507 IscsiAIOCB *acb = opaque;
508
509 if (acb->canceled != 0) {
510 return;
511 }
512
513 acb->status = 0;
514 if (status != 0) {
515 if (status == SCSI_STATUS_CHECK_CONDITION
516 && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
517 && acb->retries-- > 0) {
518 scsi_free_scsi_task(acb->task);
519 acb->task = NULL;
520 if (iscsi_aio_flush_acb(acb) == 0) {
521 iscsi_set_events(acb->iscsilun);
522 return;
523 }
524 }
525 error_report("Failed to sync10 data on iSCSI lun. %s",
526 iscsi_get_error(iscsi));
527 acb->status = -EIO;
528 }
529
530 iscsi_schedule_bh(acb);
531 }
532
533 static int
534 iscsi_aio_flush_acb(IscsiAIOCB *acb)
535 {
536 struct iscsi_context *iscsi = acb->iscsilun->iscsi;
537
538 acb->canceled = 0;
539 acb->bh = NULL;
540 acb->status = -EINPROGRESS;
541 acb->buf = NULL;
542
543 acb->task = iscsi_synchronizecache10_task(iscsi, acb->iscsilun->lun,
544 0, 0, 0, 0,
545 iscsi_synccache10_cb,
546 acb);
547 if (acb->task == NULL) {
548 error_report("iSCSI: Failed to send synchronizecache10 command. %s",
549 iscsi_get_error(iscsi));
550 return -1;
551 }
552
553 return 0;
554 }
555
556 static BlockDriverAIOCB *
557 iscsi_aio_flush(BlockDriverState *bs,
558 BlockDriverCompletionFunc *cb, void *opaque)
559 {
560 IscsiLun *iscsilun = bs->opaque;
561
562 IscsiAIOCB *acb;
563
564 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
565
566 acb->iscsilun = iscsilun;
567 acb->retries = ISCSI_CMD_RETRIES;
568
569 if (iscsi_aio_flush_acb(acb) != 0) {
570 qemu_aio_release(acb);
571 return NULL;
572 }
573
574 iscsi_set_events(iscsilun);
575
576 return &acb->common;
577 }
578
579 static int iscsi_aio_discard_acb(IscsiAIOCB *acb);
580
581 static void
582 iscsi_unmap_cb(struct iscsi_context *iscsi, int status,
583 void *command_data, void *opaque)
584 {
585 IscsiAIOCB *acb = opaque;
586
587 if (acb->canceled != 0) {
588 return;
589 }
590
591 acb->status = 0;
592 if (status != 0) {
593 if (status == SCSI_STATUS_CHECK_CONDITION
594 && acb->task->sense.key == SCSI_SENSE_UNIT_ATTENTION
595 && acb->retries-- > 0) {
596 scsi_free_scsi_task(acb->task);
597 acb->task = NULL;
598 if (iscsi_aio_discard_acb(acb) == 0) {
599 iscsi_set_events(acb->iscsilun);
600 return;
601 }
602 }
603 error_report("Failed to unmap data on iSCSI lun. %s",
604 iscsi_get_error(iscsi));
605 acb->status = -EIO;
606 }
607
608 iscsi_schedule_bh(acb);
609 }
610
611 static int iscsi_aio_discard_acb(IscsiAIOCB *acb) {
612 struct iscsi_context *iscsi = acb->iscsilun->iscsi;
613 struct unmap_list list[1];
614
615 acb->canceled = 0;
616 acb->bh = NULL;
617 acb->status = -EINPROGRESS;
618 acb->buf = NULL;
619
620 list[0].lba = sector_qemu2lun(acb->sector_num, acb->iscsilun);
621 list[0].num = acb->nb_sectors * BDRV_SECTOR_SIZE / acb->iscsilun->block_size;
622
623 acb->task = iscsi_unmap_task(iscsi, acb->iscsilun->lun,
624 0, 0, &list[0], 1,
625 iscsi_unmap_cb,
626 acb);
627 if (acb->task == NULL) {
628 error_report("iSCSI: Failed to send unmap command. %s",
629 iscsi_get_error(iscsi));
630 return -1;
631 }
632
633 return 0;
634 }
635
636 static BlockDriverAIOCB *
637 iscsi_aio_discard(BlockDriverState *bs,
638 int64_t sector_num, int nb_sectors,
639 BlockDriverCompletionFunc *cb, void *opaque)
640 {
641 IscsiLun *iscsilun = bs->opaque;
642 IscsiAIOCB *acb;
643
644 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
645
646 acb->iscsilun = iscsilun;
647 acb->nb_sectors = nb_sectors;
648 acb->sector_num = sector_num;
649 acb->retries = ISCSI_CMD_RETRIES;
650
651 if (iscsi_aio_discard_acb(acb) != 0) {
652 qemu_aio_release(acb);
653 return NULL;
654 }
655
656 iscsi_set_events(iscsilun);
657
658 return &acb->common;
659 }
660
661 #ifdef __linux__
662 static void
663 iscsi_aio_ioctl_cb(struct iscsi_context *iscsi, int status,
664 void *command_data, void *opaque)
665 {
666 IscsiAIOCB *acb = opaque;
667
668 g_free(acb->buf);
669 acb->buf = NULL;
670
671 if (acb->canceled != 0) {
672 return;
673 }
674
675 acb->status = 0;
676 if (status < 0) {
677 error_report("Failed to ioctl(SG_IO) to iSCSI lun. %s",
678 iscsi_get_error(iscsi));
679 acb->status = -EIO;
680 }
681
682 acb->ioh->driver_status = 0;
683 acb->ioh->host_status = 0;
684 acb->ioh->resid = 0;
685
686 #define SG_ERR_DRIVER_SENSE 0x08
687
688 if (status == SCSI_STATUS_CHECK_CONDITION && acb->task->datain.size >= 2) {
689 int ss;
690
691 acb->ioh->driver_status |= SG_ERR_DRIVER_SENSE;
692
693 acb->ioh->sb_len_wr = acb->task->datain.size - 2;
694 ss = (acb->ioh->mx_sb_len >= acb->ioh->sb_len_wr) ?
695 acb->ioh->mx_sb_len : acb->ioh->sb_len_wr;
696 memcpy(acb->ioh->sbp, &acb->task->datain.data[2], ss);
697 }
698
699 iscsi_schedule_bh(acb);
700 }
701
702 static BlockDriverAIOCB *iscsi_aio_ioctl(BlockDriverState *bs,
703 unsigned long int req, void *buf,
704 BlockDriverCompletionFunc *cb, void *opaque)
705 {
706 IscsiLun *iscsilun = bs->opaque;
707 struct iscsi_context *iscsi = iscsilun->iscsi;
708 struct iscsi_data data;
709 IscsiAIOCB *acb;
710
711 assert(req == SG_IO);
712
713 acb = qemu_aio_get(&iscsi_aiocb_info, bs, cb, opaque);
714
715 acb->iscsilun = iscsilun;
716 acb->canceled = 0;
717 acb->bh = NULL;
718 acb->status = -EINPROGRESS;
719 acb->buf = NULL;
720 acb->ioh = buf;
721
722 acb->task = malloc(sizeof(struct scsi_task));
723 if (acb->task == NULL) {
724 error_report("iSCSI: Failed to allocate task for scsi command. %s",
725 iscsi_get_error(iscsi));
726 qemu_aio_release(acb);
727 return NULL;
728 }
729 memset(acb->task, 0, sizeof(struct scsi_task));
730
731 switch (acb->ioh->dxfer_direction) {
732 case SG_DXFER_TO_DEV:
733 acb->task->xfer_dir = SCSI_XFER_WRITE;
734 break;
735 case SG_DXFER_FROM_DEV:
736 acb->task->xfer_dir = SCSI_XFER_READ;
737 break;
738 default:
739 acb->task->xfer_dir = SCSI_XFER_NONE;
740 break;
741 }
742
743 acb->task->cdb_size = acb->ioh->cmd_len;
744 memcpy(&acb->task->cdb[0], acb->ioh->cmdp, acb->ioh->cmd_len);
745 acb->task->expxferlen = acb->ioh->dxfer_len;
746
747 data.size = 0;
748 if (acb->task->xfer_dir == SCSI_XFER_WRITE) {
749 if (acb->ioh->iovec_count == 0) {
750 data.data = acb->ioh->dxferp;
751 data.size = acb->ioh->dxfer_len;
752 } else {
753 #if defined(LIBISCSI_FEATURE_IOVECTOR)
754 scsi_task_set_iov_out(acb->task,
755 (struct scsi_iovec *) acb->ioh->dxferp,
756 acb->ioh->iovec_count);
757 #else
758 struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
759
760 acb->buf = g_malloc(acb->ioh->dxfer_len);
761 data.data = acb->buf;
762 data.size = iov_to_buf(iov, acb->ioh->iovec_count, 0,
763 acb->buf, acb->ioh->dxfer_len);
764 #endif
765 }
766 }
767
768 if (iscsi_scsi_command_async(iscsi, iscsilun->lun, acb->task,
769 iscsi_aio_ioctl_cb,
770 (data.size > 0) ? &data : NULL,
771 acb) != 0) {
772 scsi_free_scsi_task(acb->task);
773 qemu_aio_release(acb);
774 return NULL;
775 }
776
777 /* tell libiscsi to read straight into the buffer we got from ioctl */
778 if (acb->task->xfer_dir == SCSI_XFER_READ) {
779 if (acb->ioh->iovec_count == 0) {
780 scsi_task_add_data_in_buffer(acb->task,
781 acb->ioh->dxfer_len,
782 acb->ioh->dxferp);
783 } else {
784 #if defined(LIBISCSI_FEATURE_IOVECTOR)
785 scsi_task_set_iov_in(acb->task,
786 (struct scsi_iovec *) acb->ioh->dxferp,
787 acb->ioh->iovec_count);
788 #else
789 int i;
790 for (i = 0; i < acb->ioh->iovec_count; i++) {
791 struct iovec *iov = (struct iovec *)acb->ioh->dxferp;
792
793 scsi_task_add_data_in_buffer(acb->task,
794 iov[i].iov_len,
795 iov[i].iov_base);
796 }
797 #endif
798 }
799 }
800
801 iscsi_set_events(iscsilun);
802
803 return &acb->common;
804 }
805
806
807 static void ioctl_cb(void *opaque, int status)
808 {
809 int *p_status = opaque;
810 *p_status = status;
811 }
812
813 static int iscsi_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
814 {
815 IscsiLun *iscsilun = bs->opaque;
816 int status;
817
818 switch (req) {
819 case SG_GET_VERSION_NUM:
820 *(int *)buf = 30000;
821 break;
822 case SG_GET_SCSI_ID:
823 ((struct sg_scsi_id *)buf)->scsi_type = iscsilun->type;
824 break;
825 case SG_IO:
826 status = -EINPROGRESS;
827 iscsi_aio_ioctl(bs, req, buf, ioctl_cb, &status);
828
829 while (status == -EINPROGRESS) {
830 qemu_aio_wait();
831 }
832
833 return 0;
834 default:
835 return -1;
836 }
837 return 0;
838 }
839 #endif
840
841 static int64_t
842 iscsi_getlength(BlockDriverState *bs)
843 {
844 IscsiLun *iscsilun = bs->opaque;
845 int64_t len;
846
847 len = iscsilun->num_blocks;
848 len *= iscsilun->block_size;
849
850 return len;
851 }
852
853 static int parse_chap(struct iscsi_context *iscsi, const char *target)
854 {
855 QemuOptsList *list;
856 QemuOpts *opts;
857 const char *user = NULL;
858 const char *password = NULL;
859
860 list = qemu_find_opts("iscsi");
861 if (!list) {
862 return 0;
863 }
864
865 opts = qemu_opts_find(list, target);
866 if (opts == NULL) {
867 opts = QTAILQ_FIRST(&list->head);
868 if (!opts) {
869 return 0;
870 }
871 }
872
873 user = qemu_opt_get(opts, "user");
874 if (!user) {
875 return 0;
876 }
877
878 password = qemu_opt_get(opts, "password");
879 if (!password) {
880 error_report("CHAP username specified but no password was given");
881 return -1;
882 }
883
884 if (iscsi_set_initiator_username_pwd(iscsi, user, password)) {
885 error_report("Failed to set initiator username and password");
886 return -1;
887 }
888
889 return 0;
890 }
891
892 static void parse_header_digest(struct iscsi_context *iscsi, const char *target)
893 {
894 QemuOptsList *list;
895 QemuOpts *opts;
896 const char *digest = NULL;
897
898 list = qemu_find_opts("iscsi");
899 if (!list) {
900 return;
901 }
902
903 opts = qemu_opts_find(list, target);
904 if (opts == NULL) {
905 opts = QTAILQ_FIRST(&list->head);
906 if (!opts) {
907 return;
908 }
909 }
910
911 digest = qemu_opt_get(opts, "header-digest");
912 if (!digest) {
913 return;
914 }
915
916 if (!strcmp(digest, "CRC32C")) {
917 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C);
918 } else if (!strcmp(digest, "NONE")) {
919 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE);
920 } else if (!strcmp(digest, "CRC32C-NONE")) {
921 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_CRC32C_NONE);
922 } else if (!strcmp(digest, "NONE-CRC32C")) {
923 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
924 } else {
925 error_report("Invalid header-digest setting : %s", digest);
926 }
927 }
928
929 static char *parse_initiator_name(const char *target)
930 {
931 QemuOptsList *list;
932 QemuOpts *opts;
933 const char *name = NULL;
934 const char *iscsi_name = qemu_get_vm_name();
935
936 list = qemu_find_opts("iscsi");
937 if (list) {
938 opts = qemu_opts_find(list, target);
939 if (!opts) {
940 opts = QTAILQ_FIRST(&list->head);
941 }
942 if (opts) {
943 name = qemu_opt_get(opts, "initiator-name");
944 }
945 }
946
947 if (name) {
948 return g_strdup(name);
949 } else {
950 return g_strdup_printf("iqn.2008-11.org.linux-kvm%s%s",
951 iscsi_name ? ":" : "",
952 iscsi_name ? iscsi_name : "");
953 }
954 }
955
956 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
957 static void iscsi_nop_timed_event(void *opaque)
958 {
959 IscsiLun *iscsilun = opaque;
960
961 if (iscsi_get_nops_in_flight(iscsilun->iscsi) > MAX_NOP_FAILURES) {
962 error_report("iSCSI: NOP timeout. Reconnecting...");
963 iscsi_reconnect(iscsilun->iscsi);
964 }
965
966 if (iscsi_nop_out_async(iscsilun->iscsi, NULL, NULL, 0, NULL) != 0) {
967 error_report("iSCSI: failed to sent NOP-Out. Disabling NOP messages.");
968 return;
969 }
970
971 qemu_mod_timer(iscsilun->nop_timer, qemu_get_clock_ms(rt_clock) + NOP_INTERVAL);
972 iscsi_set_events(iscsilun);
973 }
974 #endif
975
976 static int iscsi_readcapacity_sync(IscsiLun *iscsilun)
977 {
978 struct scsi_task *task = NULL;
979 struct scsi_readcapacity10 *rc10 = NULL;
980 struct scsi_readcapacity16 *rc16 = NULL;
981 int ret = 0;
982 int retries = ISCSI_CMD_RETRIES;
983
984 do {
985 if (task != NULL) {
986 scsi_free_scsi_task(task);
987 task = NULL;
988 }
989
990 switch (iscsilun->type) {
991 case TYPE_DISK:
992 task = iscsi_readcapacity16_sync(iscsilun->iscsi, iscsilun->lun);
993 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
994 rc16 = scsi_datain_unmarshall(task);
995 if (rc16 == NULL) {
996 error_report("iSCSI: Failed to unmarshall readcapacity16 data.");
997 ret = -EINVAL;
998 } else {
999 iscsilun->block_size = rc16->block_length;
1000 iscsilun->num_blocks = rc16->returned_lba + 1;
1001 }
1002 }
1003 break;
1004 case TYPE_ROM:
1005 task = iscsi_readcapacity10_sync(iscsilun->iscsi, iscsilun->lun, 0, 0);
1006 if (task != NULL && task->status == SCSI_STATUS_GOOD) {
1007 rc10 = scsi_datain_unmarshall(task);
1008 if (rc10 == NULL) {
1009 error_report("iSCSI: Failed to unmarshall readcapacity10 data.");
1010 ret = -EINVAL;
1011 } else {
1012 iscsilun->block_size = rc10->block_size;
1013 if (rc10->lba == 0) {
1014 /* blank disk loaded */
1015 iscsilun->num_blocks = 0;
1016 } else {
1017 iscsilun->num_blocks = rc10->lba + 1;
1018 }
1019 }
1020 }
1021 break;
1022 default:
1023 return 0;
1024 }
1025 } while (task != NULL && task->status == SCSI_STATUS_CHECK_CONDITION
1026 && task->sense.key == SCSI_SENSE_UNIT_ATTENTION
1027 && retries-- > 0);
1028
1029 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1030 error_report("iSCSI: failed to send readcapacity10 command.");
1031 ret = -EINVAL;
1032 }
1033 if (task) {
1034 scsi_free_scsi_task(task);
1035 }
1036 return ret;
1037 }
1038
1039 /* TODO Convert to fine grained options */
1040 static QemuOptsList runtime_opts = {
1041 .name = "iscsi",
1042 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
1043 .desc = {
1044 {
1045 .name = "filename",
1046 .type = QEMU_OPT_STRING,
1047 .help = "URL to the iscsi image",
1048 },
1049 { /* end of list */ }
1050 },
1051 };
1052
1053 /*
1054 * We support iscsi url's on the form
1055 * iscsi://[<username>%<password>@]<host>[:<port>]/<targetname>/<lun>
1056 */
1057 static int iscsi_open(BlockDriverState *bs, QDict *options, int flags)
1058 {
1059 IscsiLun *iscsilun = bs->opaque;
1060 struct iscsi_context *iscsi = NULL;
1061 struct iscsi_url *iscsi_url = NULL;
1062 struct scsi_task *task = NULL;
1063 struct scsi_inquiry_standard *inq = NULL;
1064 char *initiator_name = NULL;
1065 QemuOpts *opts;
1066 Error *local_err = NULL;
1067 const char *filename;
1068 int ret;
1069
1070 if ((BDRV_SECTOR_SIZE % 512) != 0) {
1071 error_report("iSCSI: Invalid BDRV_SECTOR_SIZE. "
1072 "BDRV_SECTOR_SIZE(%lld) is not a multiple "
1073 "of 512", BDRV_SECTOR_SIZE);
1074 return -EINVAL;
1075 }
1076
1077 opts = qemu_opts_create_nofail(&runtime_opts);
1078 qemu_opts_absorb_qdict(opts, options, &local_err);
1079 if (error_is_set(&local_err)) {
1080 qerror_report_err(local_err);
1081 error_free(local_err);
1082 ret = -EINVAL;
1083 goto out;
1084 }
1085
1086 filename = qemu_opt_get(opts, "filename");
1087
1088
1089 iscsi_url = iscsi_parse_full_url(iscsi, filename);
1090 if (iscsi_url == NULL) {
1091 error_report("Failed to parse URL : %s", filename);
1092 ret = -EINVAL;
1093 goto out;
1094 }
1095
1096 memset(iscsilun, 0, sizeof(IscsiLun));
1097
1098 initiator_name = parse_initiator_name(iscsi_url->target);
1099
1100 iscsi = iscsi_create_context(initiator_name);
1101 if (iscsi == NULL) {
1102 error_report("iSCSI: Failed to create iSCSI context.");
1103 ret = -ENOMEM;
1104 goto out;
1105 }
1106
1107 if (iscsi_set_targetname(iscsi, iscsi_url->target)) {
1108 error_report("iSCSI: Failed to set target name.");
1109 ret = -EINVAL;
1110 goto out;
1111 }
1112
1113 if (iscsi_url->user != NULL) {
1114 ret = iscsi_set_initiator_username_pwd(iscsi, iscsi_url->user,
1115 iscsi_url->passwd);
1116 if (ret != 0) {
1117 error_report("Failed to set initiator username and password");
1118 ret = -EINVAL;
1119 goto out;
1120 }
1121 }
1122
1123 /* check if we got CHAP username/password via the options */
1124 if (parse_chap(iscsi, iscsi_url->target) != 0) {
1125 error_report("iSCSI: Failed to set CHAP user/password");
1126 ret = -EINVAL;
1127 goto out;
1128 }
1129
1130 if (iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL) != 0) {
1131 error_report("iSCSI: Failed to set session type to normal.");
1132 ret = -EINVAL;
1133 goto out;
1134 }
1135
1136 iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);
1137
1138 /* check if we got HEADER_DIGEST via the options */
1139 parse_header_digest(iscsi, iscsi_url->target);
1140
1141 if (iscsi_full_connect_sync(iscsi, iscsi_url->portal, iscsi_url->lun) != 0) {
1142 error_report("iSCSI: Failed to connect to LUN : %s",
1143 iscsi_get_error(iscsi));
1144 ret = -EINVAL;
1145 goto out;
1146 }
1147
1148 iscsilun->iscsi = iscsi;
1149 iscsilun->lun = iscsi_url->lun;
1150
1151 task = iscsi_inquiry_sync(iscsi, iscsilun->lun, 0, 0, 36);
1152
1153 if (task == NULL || task->status != SCSI_STATUS_GOOD) {
1154 error_report("iSCSI: failed to send inquiry command.");
1155 ret = -EINVAL;
1156 goto out;
1157 }
1158
1159 inq = scsi_datain_unmarshall(task);
1160 if (inq == NULL) {
1161 error_report("iSCSI: Failed to unmarshall inquiry data.");
1162 ret = -EINVAL;
1163 goto out;
1164 }
1165
1166 iscsilun->type = inq->periperal_device_type;
1167
1168 if ((ret = iscsi_readcapacity_sync(iscsilun)) != 0) {
1169 goto out;
1170 }
1171 bs->total_sectors = sector_lun2qemu(iscsilun->num_blocks, iscsilun);
1172
1173 /* Medium changer or tape. We dont have any emulation for this so this must
1174 * be sg ioctl compatible. We force it to be sg, otherwise qemu will try
1175 * to read from the device to guess the image format.
1176 */
1177 if (iscsilun->type == TYPE_MEDIUM_CHANGER ||
1178 iscsilun->type == TYPE_TAPE) {
1179 bs->sg = 1;
1180 }
1181
1182 #if defined(LIBISCSI_FEATURE_NOP_COUNTER)
1183 /* Set up a timer for sending out iSCSI NOPs */
1184 iscsilun->nop_timer = qemu_new_timer_ms(rt_clock, iscsi_nop_timed_event, iscsilun);
1185 qemu_mod_timer(iscsilun->nop_timer, qemu_get_clock_ms(rt_clock) + NOP_INTERVAL);
1186 #endif
1187
1188 out:
1189 qemu_opts_del(opts);
1190 if (initiator_name != NULL) {
1191 g_free(initiator_name);
1192 }
1193 if (iscsi_url != NULL) {
1194 iscsi_destroy_url(iscsi_url);
1195 }
1196 if (task != NULL) {
1197 scsi_free_scsi_task(task);
1198 }
1199
1200 if (ret) {
1201 if (iscsi != NULL) {
1202 iscsi_destroy_context(iscsi);
1203 }
1204 memset(iscsilun, 0, sizeof(IscsiLun));
1205 }
1206 return ret;
1207 }
1208
1209 static void iscsi_close(BlockDriverState *bs)
1210 {
1211 IscsiLun *iscsilun = bs->opaque;
1212 struct iscsi_context *iscsi = iscsilun->iscsi;
1213
1214 if (iscsilun->nop_timer) {
1215 qemu_del_timer(iscsilun->nop_timer);
1216 qemu_free_timer(iscsilun->nop_timer);
1217 }
1218 qemu_aio_set_fd_handler(iscsi_get_fd(iscsi), NULL, NULL, NULL, NULL);
1219 iscsi_destroy_context(iscsi);
1220 memset(iscsilun, 0, sizeof(IscsiLun));
1221 }
1222
1223 static int iscsi_truncate(BlockDriverState *bs, int64_t offset)
1224 {
1225 IscsiLun *iscsilun = bs->opaque;
1226 int ret = 0;
1227
1228 if (iscsilun->type != TYPE_DISK) {
1229 return -ENOTSUP;
1230 }
1231
1232 if ((ret = iscsi_readcapacity_sync(iscsilun)) != 0) {
1233 return ret;
1234 }
1235
1236 if (offset > iscsi_getlength(bs)) {
1237 return -EINVAL;
1238 }
1239
1240 return 0;
1241 }
1242
1243 static int iscsi_has_zero_init(BlockDriverState *bs)
1244 {
1245 return 0;
1246 }
1247
1248 static int iscsi_create(const char *filename, QEMUOptionParameter *options)
1249 {
1250 int ret = 0;
1251 int64_t total_size = 0;
1252 BlockDriverState bs;
1253 IscsiLun *iscsilun = NULL;
1254 QDict *bs_options;
1255
1256 memset(&bs, 0, sizeof(BlockDriverState));
1257
1258 /* Read out options */
1259 while (options && options->name) {
1260 if (!strcmp(options->name, "size")) {
1261 total_size = options->value.n / BDRV_SECTOR_SIZE;
1262 }
1263 options++;
1264 }
1265
1266 bs.opaque = g_malloc0(sizeof(struct IscsiLun));
1267 iscsilun = bs.opaque;
1268
1269 bs_options = qdict_new();
1270 qdict_put(bs_options, "filename", qstring_from_str(filename));
1271 ret = iscsi_open(&bs, bs_options, 0);
1272 QDECREF(bs_options);
1273
1274 if (ret != 0) {
1275 goto out;
1276 }
1277 if (iscsilun->nop_timer) {
1278 qemu_del_timer(iscsilun->nop_timer);
1279 qemu_free_timer(iscsilun->nop_timer);
1280 }
1281 if (iscsilun->type != TYPE_DISK) {
1282 ret = -ENODEV;
1283 goto out;
1284 }
1285 if (bs.total_sectors < total_size) {
1286 ret = -ENOSPC;
1287 goto out;
1288 }
1289
1290 ret = 0;
1291 out:
1292 if (iscsilun->iscsi != NULL) {
1293 iscsi_destroy_context(iscsilun->iscsi);
1294 }
1295 g_free(bs.opaque);
1296 return ret;
1297 }
1298
1299 static QEMUOptionParameter iscsi_create_options[] = {
1300 {
1301 .name = BLOCK_OPT_SIZE,
1302 .type = OPT_SIZE,
1303 .help = "Virtual disk size"
1304 },
1305 { NULL }
1306 };
1307
1308 static BlockDriver bdrv_iscsi = {
1309 .format_name = "iscsi",
1310 .protocol_name = "iscsi",
1311
1312 .instance_size = sizeof(IscsiLun),
1313 .bdrv_file_open = iscsi_open,
1314 .bdrv_close = iscsi_close,
1315 .bdrv_create = iscsi_create,
1316 .create_options = iscsi_create_options,
1317
1318 .bdrv_getlength = iscsi_getlength,
1319 .bdrv_truncate = iscsi_truncate,
1320
1321 .bdrv_aio_readv = iscsi_aio_readv,
1322 .bdrv_aio_writev = iscsi_aio_writev,
1323 .bdrv_aio_flush = iscsi_aio_flush,
1324
1325 .bdrv_aio_discard = iscsi_aio_discard,
1326 .bdrv_has_zero_init = iscsi_has_zero_init,
1327
1328 #ifdef __linux__
1329 .bdrv_ioctl = iscsi_ioctl,
1330 .bdrv_aio_ioctl = iscsi_aio_ioctl,
1331 #endif
1332 };
1333
1334 static QemuOptsList qemu_iscsi_opts = {
1335 .name = "iscsi",
1336 .head = QTAILQ_HEAD_INITIALIZER(qemu_iscsi_opts.head),
1337 .desc = {
1338 {
1339 .name = "user",
1340 .type = QEMU_OPT_STRING,
1341 .help = "username for CHAP authentication to target",
1342 },{
1343 .name = "password",
1344 .type = QEMU_OPT_STRING,
1345 .help = "password for CHAP authentication to target",
1346 },{
1347 .name = "header-digest",
1348 .type = QEMU_OPT_STRING,
1349 .help = "HeaderDigest setting. "
1350 "{CRC32C|CRC32C-NONE|NONE-CRC32C|NONE}",
1351 },{
1352 .name = "initiator-name",
1353 .type = QEMU_OPT_STRING,
1354 .help = "Initiator iqn name to use when connecting",
1355 },
1356 { /* end of list */ }
1357 },
1358 };
1359
1360 static void iscsi_block_init(void)
1361 {
1362 bdrv_register(&bdrv_iscsi);
1363 qemu_add_opts(&qemu_iscsi_opts);
1364 }
1365
1366 block_init(iscsi_block_init);