xref: /openbmc/qemu/hw/scsi/scsi-disk.c (revision b21e2380)
1 /*
2  * SCSI Device emulation
3  *
4  * Copyright (c) 2006 CodeSourcery.
5  * Based on code by Fabrice Bellard
6  *
7  * Written by Paul Brook
8  * Modifications:
9  *  2009-Dec-12 Artyom Tarasenko : implemented stamdard inquiry for the case
10  *                                 when the allocation length of CDB is smaller
11  *                                 than 36.
12  *  2009-Oct-13 Artyom Tarasenko : implemented the block descriptor in the
13  *                                 MODE SENSE response.
14  *
15  * This code is licensed under the LGPL.
16  *
17  * Note that this file only handles the SCSI architecture model and device
18  * commands.  Emulation of interface/link layer protocols is handled by
19  * the host adapter emulator.
20  */
21 
22 #include "qemu/osdep.h"
23 #include "qemu/units.h"
24 #include "qapi/error.h"
25 #include "qemu/error-report.h"
26 #include "qemu/main-loop.h"
27 #include "qemu/module.h"
28 #include "qemu/hw-version.h"
29 #include "qemu/memalign.h"
30 #include "hw/scsi/scsi.h"
31 #include "migration/qemu-file-types.h"
32 #include "migration/vmstate.h"
33 #include "hw/scsi/emulation.h"
34 #include "scsi/constants.h"
35 #include "sysemu/block-backend.h"
36 #include "sysemu/blockdev.h"
37 #include "hw/block/block.h"
38 #include "hw/qdev-properties.h"
39 #include "hw/qdev-properties-system.h"
40 #include "sysemu/dma.h"
41 #include "sysemu/sysemu.h"
42 #include "qemu/cutils.h"
43 #include "trace.h"
44 #include "qom/object.h"
45 
46 #ifdef __linux
47 #include <scsi/sg.h>
48 #endif
49 
50 #define SCSI_WRITE_SAME_MAX         (512 * KiB)
51 #define SCSI_DMA_BUF_SIZE           (128 * KiB)
52 #define SCSI_MAX_INQUIRY_LEN        256
53 #define SCSI_MAX_MODE_LEN           256
54 
55 #define DEFAULT_DISCARD_GRANULARITY (4 * KiB)
56 #define DEFAULT_MAX_UNMAP_SIZE      (1 * GiB)
57 #define DEFAULT_MAX_IO_SIZE         INT_MAX     /* 2 GB - 1 block */
58 
59 #define TYPE_SCSI_DISK_BASE         "scsi-disk-base"
60 
61 OBJECT_DECLARE_TYPE(SCSIDiskState, SCSIDiskClass, SCSI_DISK_BASE)
62 
63 struct SCSIDiskClass {
64     SCSIDeviceClass parent_class;
65     DMAIOFunc       *dma_readv;
66     DMAIOFunc       *dma_writev;
67     bool            (*need_fua_emulation)(SCSICommand *cmd);
68     void            (*update_sense)(SCSIRequest *r);
69 };
70 
71 typedef struct SCSIDiskReq {
72     SCSIRequest req;
73     /* Both sector and sector_count are in terms of BDRV_SECTOR_SIZE bytes.  */
74     uint64_t sector;
75     uint32_t sector_count;
76     uint32_t buflen;
77     bool started;
78     bool need_fua_emulation;
79     struct iovec iov;
80     QEMUIOVector qiov;
81     BlockAcctCookie acct;
82 } SCSIDiskReq;
83 
84 #define SCSI_DISK_F_REMOVABLE             0
85 #define SCSI_DISK_F_DPOFUA                1
86 #define SCSI_DISK_F_NO_REMOVABLE_DEVOPS   2
87 
88 struct SCSIDiskState {
89     SCSIDevice qdev;
90     uint32_t features;
91     bool media_changed;
92     bool media_event;
93     bool eject_request;
94     uint16_t port_index;
95     uint64_t max_unmap_size;
96     uint64_t max_io_size;
97     QEMUBH *bh;
98     char *version;
99     char *serial;
100     char *vendor;
101     char *product;
102     char *device_id;
103     bool tray_open;
104     bool tray_locked;
105     /*
106      * 0x0000        - rotation rate not reported
107      * 0x0001        - non-rotating medium (SSD)
108      * 0x0002-0x0400 - reserved
109      * 0x0401-0xffe  - rotations per minute
110      * 0xffff        - reserved
111      */
112     uint16_t rotation_rate;
113 };
114 
115 static void scsi_free_request(SCSIRequest *req)
116 {
117     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
118 
119     qemu_vfree(r->iov.iov_base);
120 }
121 
122 /* Helper function for command completion with sense.  */
123 static void scsi_check_condition(SCSIDiskReq *r, SCSISense sense)
124 {
125     trace_scsi_disk_check_condition(r->req.tag, sense.key, sense.asc,
126                                     sense.ascq);
127     scsi_req_build_sense(&r->req, sense);
128     scsi_req_complete(&r->req, CHECK_CONDITION);
129 }
130 
131 static void scsi_init_iovec(SCSIDiskReq *r, size_t size)
132 {
133     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
134 
135     if (!r->iov.iov_base) {
136         r->buflen = size;
137         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
138     }
139     r->iov.iov_len = MIN(r->sector_count * BDRV_SECTOR_SIZE, r->buflen);
140     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
141 }
142 
143 static void scsi_disk_save_request(QEMUFile *f, SCSIRequest *req)
144 {
145     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
146 
147     qemu_put_be64s(f, &r->sector);
148     qemu_put_be32s(f, &r->sector_count);
149     qemu_put_be32s(f, &r->buflen);
150     if (r->buflen) {
151         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
152             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
153         } else if (!req->retry) {
154             uint32_t len = r->iov.iov_len;
155             qemu_put_be32s(f, &len);
156             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
157         }
158     }
159 }
160 
161 static void scsi_disk_load_request(QEMUFile *f, SCSIRequest *req)
162 {
163     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
164 
165     qemu_get_be64s(f, &r->sector);
166     qemu_get_be32s(f, &r->sector_count);
167     qemu_get_be32s(f, &r->buflen);
168     if (r->buflen) {
169         scsi_init_iovec(r, r->buflen);
170         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
171             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
172         } else if (!r->req.retry) {
173             uint32_t len;
174             qemu_get_be32s(f, &len);
175             r->iov.iov_len = len;
176             assert(r->iov.iov_len <= r->buflen);
177             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
178         }
179     }
180 
181     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
182 }
183 
184 /*
185  * scsi_handle_rw_error has two return values.  False means that the error
186  * must be ignored, true means that the error has been processed and the
187  * caller should not do anything else for this request.  Note that
188  * scsi_handle_rw_error always manages its reference counts, independent
189  * of the return value.
190  */
191 static bool scsi_handle_rw_error(SCSIDiskReq *r, int ret, bool acct_failed)
192 {
193     bool is_read = (r->req.cmd.mode == SCSI_XFER_FROM_DEV);
194     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
195     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
196     SCSISense sense = SENSE_CODE(NO_SENSE);
197     int error = 0;
198     bool req_has_sense = false;
199     BlockErrorAction action;
200     int status;
201 
202     if (ret < 0) {
203         status = scsi_sense_from_errno(-ret, &sense);
204         error = -ret;
205     } else {
206         /* A passthrough command has completed with nonzero status.  */
207         status = ret;
208         if (status == CHECK_CONDITION) {
209             req_has_sense = true;
210             error = scsi_sense_buf_to_errno(r->req.sense, sizeof(r->req.sense));
211         } else {
212             error = EINVAL;
213         }
214     }
215 
216     /*
217      * Check whether the error has to be handled by the guest or should
218      * rather follow the rerror=/werror= settings.  Guest-handled errors
219      * are usually retried immediately, so do not post them to QMP and
220      * do not account them as failed I/O.
221      */
222     if (req_has_sense &&
223         scsi_sense_buf_is_guest_recoverable(r->req.sense, sizeof(r->req.sense))) {
224         action = BLOCK_ERROR_ACTION_REPORT;
225         acct_failed = false;
226     } else {
227         action = blk_get_error_action(s->qdev.conf.blk, is_read, error);
228         blk_error_action(s->qdev.conf.blk, action, is_read, error);
229     }
230 
231     switch (action) {
232     case BLOCK_ERROR_ACTION_REPORT:
233         if (acct_failed) {
234             block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
235         }
236         if (req_has_sense) {
237             sdc->update_sense(&r->req);
238         } else if (status == CHECK_CONDITION) {
239             scsi_req_build_sense(&r->req, sense);
240         }
241         scsi_req_complete(&r->req, status);
242         return true;
243 
244     case BLOCK_ERROR_ACTION_IGNORE:
245         return false;
246 
247     case BLOCK_ERROR_ACTION_STOP:
248         scsi_req_retry(&r->req);
249         return true;
250 
251     default:
252         g_assert_not_reached();
253     }
254 }
255 
256 static bool scsi_disk_req_check_error(SCSIDiskReq *r, int ret, bool acct_failed)
257 {
258     if (r->req.io_canceled) {
259         scsi_req_cancel_complete(&r->req);
260         return true;
261     }
262 
263     if (ret < 0) {
264         return scsi_handle_rw_error(r, ret, acct_failed);
265     }
266 
267     return false;
268 }
269 
270 static void scsi_aio_complete(void *opaque, int ret)
271 {
272     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
273     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
274 
275     assert(r->req.aiocb != NULL);
276     r->req.aiocb = NULL;
277     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
278     if (scsi_disk_req_check_error(r, ret, true)) {
279         goto done;
280     }
281 
282     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
283     scsi_req_complete(&r->req, GOOD);
284 
285 done:
286     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
287     scsi_req_unref(&r->req);
288 }
289 
290 static bool scsi_is_cmd_fua(SCSICommand *cmd)
291 {
292     switch (cmd->buf[0]) {
293     case READ_10:
294     case READ_12:
295     case READ_16:
296     case WRITE_10:
297     case WRITE_12:
298     case WRITE_16:
299         return (cmd->buf[1] & 8) != 0;
300 
301     case VERIFY_10:
302     case VERIFY_12:
303     case VERIFY_16:
304     case WRITE_VERIFY_10:
305     case WRITE_VERIFY_12:
306     case WRITE_VERIFY_16:
307         return true;
308 
309     case READ_6:
310     case WRITE_6:
311     default:
312         return false;
313     }
314 }
315 
316 static void scsi_write_do_fua(SCSIDiskReq *r)
317 {
318     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
319 
320     assert(r->req.aiocb == NULL);
321     assert(!r->req.io_canceled);
322 
323     if (r->need_fua_emulation) {
324         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
325                          BLOCK_ACCT_FLUSH);
326         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
327         return;
328     }
329 
330     scsi_req_complete(&r->req, GOOD);
331     scsi_req_unref(&r->req);
332 }
333 
334 static void scsi_dma_complete_noio(SCSIDiskReq *r, int ret)
335 {
336     assert(r->req.aiocb == NULL);
337     if (scsi_disk_req_check_error(r, ret, false)) {
338         goto done;
339     }
340 
341     r->sector += r->sector_count;
342     r->sector_count = 0;
343     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
344         scsi_write_do_fua(r);
345         return;
346     } else {
347         scsi_req_complete(&r->req, GOOD);
348     }
349 
350 done:
351     scsi_req_unref(&r->req);
352 }
353 
354 static void scsi_dma_complete(void *opaque, int ret)
355 {
356     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
357     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
358 
359     assert(r->req.aiocb != NULL);
360     r->req.aiocb = NULL;
361 
362     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
363     if (ret < 0) {
364         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
365     } else {
366         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
367     }
368     scsi_dma_complete_noio(r, ret);
369     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
370 }
371 
372 static void scsi_read_complete_noio(SCSIDiskReq *r, int ret)
373 {
374     uint32_t n;
375 
376     assert(r->req.aiocb == NULL);
377     if (scsi_disk_req_check_error(r, ret, false)) {
378         goto done;
379     }
380 
381     n = r->qiov.size / BDRV_SECTOR_SIZE;
382     r->sector += n;
383     r->sector_count -= n;
384     scsi_req_data(&r->req, r->qiov.size);
385 
386 done:
387     scsi_req_unref(&r->req);
388 }
389 
390 static void scsi_read_complete(void *opaque, int ret)
391 {
392     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
393     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
394 
395     assert(r->req.aiocb != NULL);
396     r->req.aiocb = NULL;
397 
398     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
399     if (ret < 0) {
400         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
401     } else {
402         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
403         trace_scsi_disk_read_complete(r->req.tag, r->qiov.size);
404     }
405     scsi_read_complete_noio(r, ret);
406     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
407 }
408 
409 /* Actually issue a read to the block device.  */
410 static void scsi_do_read(SCSIDiskReq *r, int ret)
411 {
412     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
413     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
414 
415     assert (r->req.aiocb == NULL);
416     if (scsi_disk_req_check_error(r, ret, false)) {
417         goto done;
418     }
419 
420     /* The request is used as the AIO opaque value, so add a ref.  */
421     scsi_req_ref(&r->req);
422 
423     if (r->req.sg) {
424         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_READ);
425         r->req.residual -= r->req.sg->size;
426         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
427                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
428                                   BDRV_SECTOR_SIZE,
429                                   sdc->dma_readv, r, scsi_dma_complete, r,
430                                   DMA_DIRECTION_FROM_DEVICE);
431     } else {
432         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
433         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
434                          r->qiov.size, BLOCK_ACCT_READ);
435         r->req.aiocb = sdc->dma_readv(r->sector << BDRV_SECTOR_BITS, &r->qiov,
436                                       scsi_read_complete, r, r);
437     }
438 
439 done:
440     scsi_req_unref(&r->req);
441 }
442 
443 static void scsi_do_read_cb(void *opaque, int ret)
444 {
445     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
446     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
447 
448     assert (r->req.aiocb != NULL);
449     r->req.aiocb = NULL;
450 
451     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
452     if (ret < 0) {
453         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
454     } else {
455         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
456     }
457     scsi_do_read(opaque, ret);
458     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
459 }
460 
461 /* Read more data from scsi device into buffer.  */
462 static void scsi_read_data(SCSIRequest *req)
463 {
464     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
465     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
466     bool first;
467 
468     trace_scsi_disk_read_data_count(r->sector_count);
469     if (r->sector_count == 0) {
470         /* This also clears the sense buffer for REQUEST SENSE.  */
471         scsi_req_complete(&r->req, GOOD);
472         return;
473     }
474 
475     /* No data transfer may already be in progress */
476     assert(r->req.aiocb == NULL);
477 
478     /* The request is used as the AIO opaque value, so add a ref.  */
479     scsi_req_ref(&r->req);
480     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
481         trace_scsi_disk_read_data_invalid();
482         scsi_read_complete_noio(r, -EINVAL);
483         return;
484     }
485 
486     if (!blk_is_available(req->dev->conf.blk)) {
487         scsi_read_complete_noio(r, -ENOMEDIUM);
488         return;
489     }
490 
491     first = !r->started;
492     r->started = true;
493     if (first && r->need_fua_emulation) {
494         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
495                          BLOCK_ACCT_FLUSH);
496         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_do_read_cb, r);
497     } else {
498         scsi_do_read(r, 0);
499     }
500 }
501 
502 static void scsi_write_complete_noio(SCSIDiskReq *r, int ret)
503 {
504     uint32_t n;
505 
506     assert (r->req.aiocb == NULL);
507     if (scsi_disk_req_check_error(r, ret, false)) {
508         goto done;
509     }
510 
511     n = r->qiov.size / BDRV_SECTOR_SIZE;
512     r->sector += n;
513     r->sector_count -= n;
514     if (r->sector_count == 0) {
515         scsi_write_do_fua(r);
516         return;
517     } else {
518         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
519         trace_scsi_disk_write_complete_noio(r->req.tag, r->qiov.size);
520         scsi_req_data(&r->req, r->qiov.size);
521     }
522 
523 done:
524     scsi_req_unref(&r->req);
525 }
526 
527 static void scsi_write_complete(void * opaque, int ret)
528 {
529     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
530     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
531 
532     assert (r->req.aiocb != NULL);
533     r->req.aiocb = NULL;
534 
535     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
536     if (ret < 0) {
537         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
538     } else {
539         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
540     }
541     scsi_write_complete_noio(r, ret);
542     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
543 }
544 
545 static void scsi_write_data(SCSIRequest *req)
546 {
547     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
548     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
549     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
550 
551     /* No data transfer may already be in progress */
552     assert(r->req.aiocb == NULL);
553 
554     /* The request is used as the AIO opaque value, so add a ref.  */
555     scsi_req_ref(&r->req);
556     if (r->req.cmd.mode != SCSI_XFER_TO_DEV) {
557         trace_scsi_disk_write_data_invalid();
558         scsi_write_complete_noio(r, -EINVAL);
559         return;
560     }
561 
562     if (!r->req.sg && !r->qiov.size) {
563         /* Called for the first time.  Ask the driver to send us more data.  */
564         r->started = true;
565         scsi_write_complete_noio(r, 0);
566         return;
567     }
568     if (!blk_is_available(req->dev->conf.blk)) {
569         scsi_write_complete_noio(r, -ENOMEDIUM);
570         return;
571     }
572 
573     if (r->req.cmd.buf[0] == VERIFY_10 || r->req.cmd.buf[0] == VERIFY_12 ||
574         r->req.cmd.buf[0] == VERIFY_16) {
575         if (r->req.sg) {
576             scsi_dma_complete_noio(r, 0);
577         } else {
578             scsi_write_complete_noio(r, 0);
579         }
580         return;
581     }
582 
583     if (r->req.sg) {
584         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_WRITE);
585         r->req.residual -= r->req.sg->size;
586         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
587                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
588                                   BDRV_SECTOR_SIZE,
589                                   sdc->dma_writev, r, scsi_dma_complete, r,
590                                   DMA_DIRECTION_TO_DEVICE);
591     } else {
592         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
593                          r->qiov.size, BLOCK_ACCT_WRITE);
594         r->req.aiocb = sdc->dma_writev(r->sector << BDRV_SECTOR_BITS, &r->qiov,
595                                        scsi_write_complete, r, r);
596     }
597 }
598 
599 /* Return a pointer to the data buffer.  */
600 static uint8_t *scsi_get_buf(SCSIRequest *req)
601 {
602     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
603 
604     return (uint8_t *)r->iov.iov_base;
605 }
606 
607 static int scsi_disk_emulate_vpd_page(SCSIRequest *req, uint8_t *outbuf)
608 {
609     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
610     uint8_t page_code = req->cmd.buf[2];
611     int start, buflen = 0;
612 
613     outbuf[buflen++] = s->qdev.type & 0x1f;
614     outbuf[buflen++] = page_code;
615     outbuf[buflen++] = 0x00;
616     outbuf[buflen++] = 0x00;
617     start = buflen;
618 
619     switch (page_code) {
620     case 0x00: /* Supported page codes, mandatory */
621     {
622         trace_scsi_disk_emulate_vpd_page_00(req->cmd.xfer);
623         outbuf[buflen++] = 0x00; /* list of supported pages (this page) */
624         if (s->serial) {
625             outbuf[buflen++] = 0x80; /* unit serial number */
626         }
627         outbuf[buflen++] = 0x83; /* device identification */
628         if (s->qdev.type == TYPE_DISK) {
629             outbuf[buflen++] = 0xb0; /* block limits */
630             outbuf[buflen++] = 0xb1; /* block device characteristics */
631             outbuf[buflen++] = 0xb2; /* thin provisioning */
632         }
633         break;
634     }
635     case 0x80: /* Device serial number, optional */
636     {
637         int l;
638 
639         if (!s->serial) {
640             trace_scsi_disk_emulate_vpd_page_80_not_supported();
641             return -1;
642         }
643 
644         l = strlen(s->serial);
645         if (l > 36) {
646             l = 36;
647         }
648 
649         trace_scsi_disk_emulate_vpd_page_80(req->cmd.xfer);
650         memcpy(outbuf + buflen, s->serial, l);
651         buflen += l;
652         break;
653     }
654 
655     case 0x83: /* Device identification page, mandatory */
656     {
657         int id_len = s->device_id ? MIN(strlen(s->device_id), 255 - 8) : 0;
658 
659         trace_scsi_disk_emulate_vpd_page_83(req->cmd.xfer);
660 
661         if (id_len) {
662             outbuf[buflen++] = 0x2; /* ASCII */
663             outbuf[buflen++] = 0;   /* not officially assigned */
664             outbuf[buflen++] = 0;   /* reserved */
665             outbuf[buflen++] = id_len; /* length of data following */
666             memcpy(outbuf + buflen, s->device_id, id_len);
667             buflen += id_len;
668         }
669 
670         if (s->qdev.wwn) {
671             outbuf[buflen++] = 0x1; /* Binary */
672             outbuf[buflen++] = 0x3; /* NAA */
673             outbuf[buflen++] = 0;   /* reserved */
674             outbuf[buflen++] = 8;
675             stq_be_p(&outbuf[buflen], s->qdev.wwn);
676             buflen += 8;
677         }
678 
679         if (s->qdev.port_wwn) {
680             outbuf[buflen++] = 0x61; /* SAS / Binary */
681             outbuf[buflen++] = 0x93; /* PIV / Target port / NAA */
682             outbuf[buflen++] = 0;    /* reserved */
683             outbuf[buflen++] = 8;
684             stq_be_p(&outbuf[buflen], s->qdev.port_wwn);
685             buflen += 8;
686         }
687 
688         if (s->port_index) {
689             outbuf[buflen++] = 0x61; /* SAS / Binary */
690 
691             /* PIV/Target port/relative target port */
692             outbuf[buflen++] = 0x94;
693 
694             outbuf[buflen++] = 0;    /* reserved */
695             outbuf[buflen++] = 4;
696             stw_be_p(&outbuf[buflen + 2], s->port_index);
697             buflen += 4;
698         }
699         break;
700     }
701     case 0xb0: /* block limits */
702     {
703         SCSIBlockLimits bl = {};
704 
705         if (s->qdev.type == TYPE_ROM) {
706             trace_scsi_disk_emulate_vpd_page_b0_not_supported();
707             return -1;
708         }
709         bl.wsnz = 1;
710         bl.unmap_sectors =
711             s->qdev.conf.discard_granularity / s->qdev.blocksize;
712         bl.min_io_size =
713             s->qdev.conf.min_io_size / s->qdev.blocksize;
714         bl.opt_io_size =
715             s->qdev.conf.opt_io_size / s->qdev.blocksize;
716         bl.max_unmap_sectors =
717             s->max_unmap_size / s->qdev.blocksize;
718         bl.max_io_sectors =
719             s->max_io_size / s->qdev.blocksize;
720         /* 255 descriptors fit in 4 KiB with an 8-byte header */
721         bl.max_unmap_descr = 255;
722 
723         if (s->qdev.type == TYPE_DISK) {
724             int max_transfer_blk = blk_get_max_transfer(s->qdev.conf.blk);
725             int max_io_sectors_blk =
726                 max_transfer_blk / s->qdev.blocksize;
727 
728             bl.max_io_sectors =
729                 MIN_NON_ZERO(max_io_sectors_blk, bl.max_io_sectors);
730         }
731         buflen += scsi_emulate_block_limits(outbuf + buflen, &bl);
732         break;
733     }
734     case 0xb1: /* block device characteristics */
735     {
736         buflen = 0x40;
737         outbuf[4] = (s->rotation_rate >> 8) & 0xff;
738         outbuf[5] = s->rotation_rate & 0xff;
739         outbuf[6] = 0; /* PRODUCT TYPE */
740         outbuf[7] = 0; /* WABEREQ | WACEREQ | NOMINAL FORM FACTOR */
741         outbuf[8] = 0; /* VBULS */
742         break;
743     }
744     case 0xb2: /* thin provisioning */
745     {
746         buflen = 8;
747         outbuf[4] = 0;
748         outbuf[5] = 0xe0; /* unmap & write_same 10/16 all supported */
749         outbuf[6] = s->qdev.conf.discard_granularity ? 2 : 1;
750         outbuf[7] = 0;
751         break;
752     }
753     default:
754         return -1;
755     }
756     /* done with EVPD */
757     assert(buflen - start <= 255);
758     outbuf[start - 1] = buflen - start;
759     return buflen;
760 }
761 
762 static int scsi_disk_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf)
763 {
764     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
765     int buflen = 0;
766 
767     if (req->cmd.buf[1] & 0x1) {
768         /* Vital product data */
769         return scsi_disk_emulate_vpd_page(req, outbuf);
770     }
771 
772     /* Standard INQUIRY data */
773     if (req->cmd.buf[2] != 0) {
774         return -1;
775     }
776 
777     /* PAGE CODE == 0 */
778     buflen = req->cmd.xfer;
779     if (buflen > SCSI_MAX_INQUIRY_LEN) {
780         buflen = SCSI_MAX_INQUIRY_LEN;
781     }
782 
783     outbuf[0] = s->qdev.type & 0x1f;
784     outbuf[1] = (s->features & (1 << SCSI_DISK_F_REMOVABLE)) ? 0x80 : 0;
785 
786     strpadcpy((char *) &outbuf[16], 16, s->product, ' ');
787     strpadcpy((char *) &outbuf[8], 8, s->vendor, ' ');
788 
789     memset(&outbuf[32], 0, 4);
790     memcpy(&outbuf[32], s->version, MIN(4, strlen(s->version)));
791     /*
792      * We claim conformance to SPC-3, which is required for guests
793      * to ask for modern features like READ CAPACITY(16) or the
794      * block characteristics VPD page by default.  Not all of SPC-3
795      * is actually implemented, but we're good enough.
796      */
797     outbuf[2] = s->qdev.default_scsi_version;
798     outbuf[3] = 2 | 0x10; /* Format 2, HiSup */
799 
800     if (buflen > 36) {
801         outbuf[4] = buflen - 5; /* Additional Length = (Len - 1) - 4 */
802     } else {
803         /* If the allocation length of CDB is too small,
804                the additional length is not adjusted */
805         outbuf[4] = 36 - 5;
806     }
807 
808     /* Sync data transfer and TCQ.  */
809     outbuf[7] = 0x10 | (req->bus->info->tcq ? 0x02 : 0);
810     return buflen;
811 }
812 
813 static inline bool media_is_dvd(SCSIDiskState *s)
814 {
815     uint64_t nb_sectors;
816     if (s->qdev.type != TYPE_ROM) {
817         return false;
818     }
819     if (!blk_is_available(s->qdev.conf.blk)) {
820         return false;
821     }
822     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
823     return nb_sectors > CD_MAX_SECTORS;
824 }
825 
826 static inline bool media_is_cd(SCSIDiskState *s)
827 {
828     uint64_t nb_sectors;
829     if (s->qdev.type != TYPE_ROM) {
830         return false;
831     }
832     if (!blk_is_available(s->qdev.conf.blk)) {
833         return false;
834     }
835     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
836     return nb_sectors <= CD_MAX_SECTORS;
837 }
838 
839 static int scsi_read_disc_information(SCSIDiskState *s, SCSIDiskReq *r,
840                                       uint8_t *outbuf)
841 {
842     uint8_t type = r->req.cmd.buf[1] & 7;
843 
844     if (s->qdev.type != TYPE_ROM) {
845         return -1;
846     }
847 
848     /* Types 1/2 are only defined for Blu-Ray.  */
849     if (type != 0) {
850         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
851         return -1;
852     }
853 
854     memset(outbuf, 0, 34);
855     outbuf[1] = 32;
856     outbuf[2] = 0xe; /* last session complete, disc finalized */
857     outbuf[3] = 1;   /* first track on disc */
858     outbuf[4] = 1;   /* # of sessions */
859     outbuf[5] = 1;   /* first track of last session */
860     outbuf[6] = 1;   /* last track of last session */
861     outbuf[7] = 0x20; /* unrestricted use */
862     outbuf[8] = 0x00; /* CD-ROM or DVD-ROM */
863     /* 9-10-11: most significant byte corresponding bytes 4-5-6 */
864     /* 12-23: not meaningful for CD-ROM or DVD-ROM */
865     /* 24-31: disc bar code */
866     /* 32: disc application code */
867     /* 33: number of OPC tables */
868 
869     return 34;
870 }
871 
872 static int scsi_read_dvd_structure(SCSIDiskState *s, SCSIDiskReq *r,
873                                    uint8_t *outbuf)
874 {
875     static const int rds_caps_size[5] = {
876         [0] = 2048 + 4,
877         [1] = 4 + 4,
878         [3] = 188 + 4,
879         [4] = 2048 + 4,
880     };
881 
882     uint8_t media = r->req.cmd.buf[1];
883     uint8_t layer = r->req.cmd.buf[6];
884     uint8_t format = r->req.cmd.buf[7];
885     int size = -1;
886 
887     if (s->qdev.type != TYPE_ROM) {
888         return -1;
889     }
890     if (media != 0) {
891         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
892         return -1;
893     }
894 
895     if (format != 0xff) {
896         if (!blk_is_available(s->qdev.conf.blk)) {
897             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
898             return -1;
899         }
900         if (media_is_cd(s)) {
901             scsi_check_condition(r, SENSE_CODE(INCOMPATIBLE_FORMAT));
902             return -1;
903         }
904         if (format >= ARRAY_SIZE(rds_caps_size)) {
905             return -1;
906         }
907         size = rds_caps_size[format];
908         memset(outbuf, 0, size);
909     }
910 
911     switch (format) {
912     case 0x00: {
913         /* Physical format information */
914         uint64_t nb_sectors;
915         if (layer != 0) {
916             goto fail;
917         }
918         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
919 
920         outbuf[4] = 1;   /* DVD-ROM, part version 1 */
921         outbuf[5] = 0xf; /* 120mm disc, minimum rate unspecified */
922         outbuf[6] = 1;   /* one layer, read-only (per MMC-2 spec) */
923         outbuf[7] = 0;   /* default densities */
924 
925         stl_be_p(&outbuf[12], (nb_sectors >> 2) - 1); /* end sector */
926         stl_be_p(&outbuf[16], (nb_sectors >> 2) - 1); /* l0 end sector */
927         break;
928     }
929 
930     case 0x01: /* DVD copyright information, all zeros */
931         break;
932 
933     case 0x03: /* BCA information - invalid field for no BCA info */
934         return -1;
935 
936     case 0x04: /* DVD disc manufacturing information, all zeros */
937         break;
938 
939     case 0xff: { /* List capabilities */
940         int i;
941         size = 4;
942         for (i = 0; i < ARRAY_SIZE(rds_caps_size); i++) {
943             if (!rds_caps_size[i]) {
944                 continue;
945             }
946             outbuf[size] = i;
947             outbuf[size + 1] = 0x40; /* Not writable, readable */
948             stw_be_p(&outbuf[size + 2], rds_caps_size[i]);
949             size += 4;
950         }
951         break;
952      }
953 
954     default:
955         return -1;
956     }
957 
958     /* Size of buffer, not including 2 byte size field */
959     stw_be_p(outbuf, size - 2);
960     return size;
961 
962 fail:
963     return -1;
964 }
965 
966 static int scsi_event_status_media(SCSIDiskState *s, uint8_t *outbuf)
967 {
968     uint8_t event_code, media_status;
969 
970     media_status = 0;
971     if (s->tray_open) {
972         media_status = MS_TRAY_OPEN;
973     } else if (blk_is_inserted(s->qdev.conf.blk)) {
974         media_status = MS_MEDIA_PRESENT;
975     }
976 
977     /* Event notification descriptor */
978     event_code = MEC_NO_CHANGE;
979     if (media_status != MS_TRAY_OPEN) {
980         if (s->media_event) {
981             event_code = MEC_NEW_MEDIA;
982             s->media_event = false;
983         } else if (s->eject_request) {
984             event_code = MEC_EJECT_REQUESTED;
985             s->eject_request = false;
986         }
987     }
988 
989     outbuf[0] = event_code;
990     outbuf[1] = media_status;
991 
992     /* These fields are reserved, just clear them. */
993     outbuf[2] = 0;
994     outbuf[3] = 0;
995     return 4;
996 }
997 
998 static int scsi_get_event_status_notification(SCSIDiskState *s, SCSIDiskReq *r,
999                                               uint8_t *outbuf)
1000 {
1001     int size;
1002     uint8_t *buf = r->req.cmd.buf;
1003     uint8_t notification_class_request = buf[4];
1004     if (s->qdev.type != TYPE_ROM) {
1005         return -1;
1006     }
1007     if ((buf[1] & 1) == 0) {
1008         /* asynchronous */
1009         return -1;
1010     }
1011 
1012     size = 4;
1013     outbuf[0] = outbuf[1] = 0;
1014     outbuf[3] = 1 << GESN_MEDIA; /* supported events */
1015     if (notification_class_request & (1 << GESN_MEDIA)) {
1016         outbuf[2] = GESN_MEDIA;
1017         size += scsi_event_status_media(s, &outbuf[size]);
1018     } else {
1019         outbuf[2] = 0x80;
1020     }
1021     stw_be_p(outbuf, size - 4);
1022     return size;
1023 }
1024 
1025 static int scsi_get_configuration(SCSIDiskState *s, uint8_t *outbuf)
1026 {
1027     int current;
1028 
1029     if (s->qdev.type != TYPE_ROM) {
1030         return -1;
1031     }
1032 
1033     if (media_is_dvd(s)) {
1034         current = MMC_PROFILE_DVD_ROM;
1035     } else if (media_is_cd(s)) {
1036         current = MMC_PROFILE_CD_ROM;
1037     } else {
1038         current = MMC_PROFILE_NONE;
1039     }
1040 
1041     memset(outbuf, 0, 40);
1042     stl_be_p(&outbuf[0], 36); /* Bytes after the data length field */
1043     stw_be_p(&outbuf[6], current);
1044     /* outbuf[8] - outbuf[19]: Feature 0 - Profile list */
1045     outbuf[10] = 0x03; /* persistent, current */
1046     outbuf[11] = 8; /* two profiles */
1047     stw_be_p(&outbuf[12], MMC_PROFILE_DVD_ROM);
1048     outbuf[14] = (current == MMC_PROFILE_DVD_ROM);
1049     stw_be_p(&outbuf[16], MMC_PROFILE_CD_ROM);
1050     outbuf[18] = (current == MMC_PROFILE_CD_ROM);
1051     /* outbuf[20] - outbuf[31]: Feature 1 - Core feature */
1052     stw_be_p(&outbuf[20], 1);
1053     outbuf[22] = 0x08 | 0x03; /* version 2, persistent, current */
1054     outbuf[23] = 8;
1055     stl_be_p(&outbuf[24], 1); /* SCSI */
1056     outbuf[28] = 1; /* DBE = 1, mandatory */
1057     /* outbuf[32] - outbuf[39]: Feature 3 - Removable media feature */
1058     stw_be_p(&outbuf[32], 3);
1059     outbuf[34] = 0x08 | 0x03; /* version 2, persistent, current */
1060     outbuf[35] = 4;
1061     outbuf[36] = 0x39; /* tray, load=1, eject=1, unlocked at powerup, lock=1 */
1062     /* TODO: Random readable, CD read, DVD read, drive serial number,
1063        power management */
1064     return 40;
1065 }
1066 
1067 static int scsi_emulate_mechanism_status(SCSIDiskState *s, uint8_t *outbuf)
1068 {
1069     if (s->qdev.type != TYPE_ROM) {
1070         return -1;
1071     }
1072     memset(outbuf, 0, 8);
1073     outbuf[5] = 1; /* CD-ROM */
1074     return 8;
1075 }
1076 
1077 static int mode_sense_page(SCSIDiskState *s, int page, uint8_t **p_outbuf,
1078                            int page_control)
1079 {
1080     static const int mode_sense_valid[0x3f] = {
1081         [MODE_PAGE_HD_GEOMETRY]            = (1 << TYPE_DISK),
1082         [MODE_PAGE_FLEXIBLE_DISK_GEOMETRY] = (1 << TYPE_DISK),
1083         [MODE_PAGE_CACHING]                = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1084         [MODE_PAGE_R_W_ERROR]              = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1085         [MODE_PAGE_AUDIO_CTL]              = (1 << TYPE_ROM),
1086         [MODE_PAGE_CAPABILITIES]           = (1 << TYPE_ROM),
1087     };
1088 
1089     uint8_t *p = *p_outbuf + 2;
1090     int length;
1091 
1092     assert(page < ARRAY_SIZE(mode_sense_valid));
1093     if ((mode_sense_valid[page] & (1 << s->qdev.type)) == 0) {
1094         return -1;
1095     }
1096 
1097     /*
1098      * If Changeable Values are requested, a mask denoting those mode parameters
1099      * that are changeable shall be returned. As we currently don't support
1100      * parameter changes via MODE_SELECT all bits are returned set to zero.
1101      * The buffer was already menset to zero by the caller of this function.
1102      *
1103      * The offsets here are off by two compared to the descriptions in the
1104      * SCSI specs, because those include a 2-byte header.  This is unfortunate,
1105      * but it is done so that offsets are consistent within our implementation
1106      * of MODE SENSE and MODE SELECT.  MODE SELECT has to deal with both
1107      * 2-byte and 4-byte headers.
1108      */
1109     switch (page) {
1110     case MODE_PAGE_HD_GEOMETRY:
1111         length = 0x16;
1112         if (page_control == 1) { /* Changeable Values */
1113             break;
1114         }
1115         /* if a geometry hint is available, use it */
1116         p[0] = (s->qdev.conf.cyls >> 16) & 0xff;
1117         p[1] = (s->qdev.conf.cyls >> 8) & 0xff;
1118         p[2] = s->qdev.conf.cyls & 0xff;
1119         p[3] = s->qdev.conf.heads & 0xff;
1120         /* Write precomp start cylinder, disabled */
1121         p[4] = (s->qdev.conf.cyls >> 16) & 0xff;
1122         p[5] = (s->qdev.conf.cyls >> 8) & 0xff;
1123         p[6] = s->qdev.conf.cyls & 0xff;
1124         /* Reduced current start cylinder, disabled */
1125         p[7] = (s->qdev.conf.cyls >> 16) & 0xff;
1126         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1127         p[9] = s->qdev.conf.cyls & 0xff;
1128         /* Device step rate [ns], 200ns */
1129         p[10] = 0;
1130         p[11] = 200;
1131         /* Landing zone cylinder */
1132         p[12] = 0xff;
1133         p[13] =  0xff;
1134         p[14] = 0xff;
1135         /* Medium rotation rate [rpm], 5400 rpm */
1136         p[18] = (5400 >> 8) & 0xff;
1137         p[19] = 5400 & 0xff;
1138         break;
1139 
1140     case MODE_PAGE_FLEXIBLE_DISK_GEOMETRY:
1141         length = 0x1e;
1142         if (page_control == 1) { /* Changeable Values */
1143             break;
1144         }
1145         /* Transfer rate [kbit/s], 5Mbit/s */
1146         p[0] = 5000 >> 8;
1147         p[1] = 5000 & 0xff;
1148         /* if a geometry hint is available, use it */
1149         p[2] = s->qdev.conf.heads & 0xff;
1150         p[3] = s->qdev.conf.secs & 0xff;
1151         p[4] = s->qdev.blocksize >> 8;
1152         p[6] = (s->qdev.conf.cyls >> 8) & 0xff;
1153         p[7] = s->qdev.conf.cyls & 0xff;
1154         /* Write precomp start cylinder, disabled */
1155         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1156         p[9] = s->qdev.conf.cyls & 0xff;
1157         /* Reduced current start cylinder, disabled */
1158         p[10] = (s->qdev.conf.cyls >> 8) & 0xff;
1159         p[11] = s->qdev.conf.cyls & 0xff;
1160         /* Device step rate [100us], 100us */
1161         p[12] = 0;
1162         p[13] = 1;
1163         /* Device step pulse width [us], 1us */
1164         p[14] = 1;
1165         /* Device head settle delay [100us], 100us */
1166         p[15] = 0;
1167         p[16] = 1;
1168         /* Motor on delay [0.1s], 0.1s */
1169         p[17] = 1;
1170         /* Motor off delay [0.1s], 0.1s */
1171         p[18] = 1;
1172         /* Medium rotation rate [rpm], 5400 rpm */
1173         p[26] = (5400 >> 8) & 0xff;
1174         p[27] = 5400 & 0xff;
1175         break;
1176 
1177     case MODE_PAGE_CACHING:
1178         length = 0x12;
1179         if (page_control == 1 || /* Changeable Values */
1180             blk_enable_write_cache(s->qdev.conf.blk)) {
1181             p[0] = 4; /* WCE */
1182         }
1183         break;
1184 
1185     case MODE_PAGE_R_W_ERROR:
1186         length = 10;
1187         if (page_control == 1) { /* Changeable Values */
1188             break;
1189         }
1190         p[0] = 0x80; /* Automatic Write Reallocation Enabled */
1191         if (s->qdev.type == TYPE_ROM) {
1192             p[1] = 0x20; /* Read Retry Count */
1193         }
1194         break;
1195 
1196     case MODE_PAGE_AUDIO_CTL:
1197         length = 14;
1198         break;
1199 
1200     case MODE_PAGE_CAPABILITIES:
1201         length = 0x14;
1202         if (page_control == 1) { /* Changeable Values */
1203             break;
1204         }
1205 
1206         p[0] = 0x3b; /* CD-R & CD-RW read */
1207         p[1] = 0; /* Writing not supported */
1208         p[2] = 0x7f; /* Audio, composite, digital out,
1209                         mode 2 form 1&2, multi session */
1210         p[3] = 0xff; /* CD DA, DA accurate, RW supported,
1211                         RW corrected, C2 errors, ISRC,
1212                         UPC, Bar code */
1213         p[4] = 0x2d | (s->tray_locked ? 2 : 0);
1214         /* Locking supported, jumper present, eject, tray */
1215         p[5] = 0; /* no volume & mute control, no
1216                      changer */
1217         p[6] = (50 * 176) >> 8; /* 50x read speed */
1218         p[7] = (50 * 176) & 0xff;
1219         p[8] = 2 >> 8; /* Two volume levels */
1220         p[9] = 2 & 0xff;
1221         p[10] = 2048 >> 8; /* 2M buffer */
1222         p[11] = 2048 & 0xff;
1223         p[12] = (16 * 176) >> 8; /* 16x read speed current */
1224         p[13] = (16 * 176) & 0xff;
1225         p[16] = (16 * 176) >> 8; /* 16x write speed */
1226         p[17] = (16 * 176) & 0xff;
1227         p[18] = (16 * 176) >> 8; /* 16x write speed current */
1228         p[19] = (16 * 176) & 0xff;
1229         break;
1230 
1231     default:
1232         return -1;
1233     }
1234 
1235     assert(length < 256);
1236     (*p_outbuf)[0] = page;
1237     (*p_outbuf)[1] = length;
1238     *p_outbuf += length + 2;
1239     return length + 2;
1240 }
1241 
1242 static int scsi_disk_emulate_mode_sense(SCSIDiskReq *r, uint8_t *outbuf)
1243 {
1244     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1245     uint64_t nb_sectors;
1246     bool dbd;
1247     int page, buflen, ret, page_control;
1248     uint8_t *p;
1249     uint8_t dev_specific_param;
1250 
1251     dbd = (r->req.cmd.buf[1] & 0x8) != 0;
1252     page = r->req.cmd.buf[2] & 0x3f;
1253     page_control = (r->req.cmd.buf[2] & 0xc0) >> 6;
1254 
1255     trace_scsi_disk_emulate_mode_sense((r->req.cmd.buf[0] == MODE_SENSE) ? 6 :
1256                                        10, page, r->req.cmd.xfer, page_control);
1257     memset(outbuf, 0, r->req.cmd.xfer);
1258     p = outbuf;
1259 
1260     if (s->qdev.type == TYPE_DISK) {
1261         dev_specific_param = s->features & (1 << SCSI_DISK_F_DPOFUA) ? 0x10 : 0;
1262         if (!blk_is_writable(s->qdev.conf.blk)) {
1263             dev_specific_param |= 0x80; /* Readonly.  */
1264         }
1265     } else {
1266         /* MMC prescribes that CD/DVD drives have no block descriptors,
1267          * and defines no device-specific parameter.  */
1268         dev_specific_param = 0x00;
1269         dbd = true;
1270     }
1271 
1272     if (r->req.cmd.buf[0] == MODE_SENSE) {
1273         p[1] = 0; /* Default media type.  */
1274         p[2] = dev_specific_param;
1275         p[3] = 0; /* Block descriptor length.  */
1276         p += 4;
1277     } else { /* MODE_SENSE_10 */
1278         p[2] = 0; /* Default media type.  */
1279         p[3] = dev_specific_param;
1280         p[6] = p[7] = 0; /* Block descriptor length.  */
1281         p += 8;
1282     }
1283 
1284     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1285     if (!dbd && nb_sectors) {
1286         if (r->req.cmd.buf[0] == MODE_SENSE) {
1287             outbuf[3] = 8; /* Block descriptor length  */
1288         } else { /* MODE_SENSE_10 */
1289             outbuf[7] = 8; /* Block descriptor length  */
1290         }
1291         nb_sectors /= (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1292         if (nb_sectors > 0xffffff) {
1293             nb_sectors = 0;
1294         }
1295         p[0] = 0; /* media density code */
1296         p[1] = (nb_sectors >> 16) & 0xff;
1297         p[2] = (nb_sectors >> 8) & 0xff;
1298         p[3] = nb_sectors & 0xff;
1299         p[4] = 0; /* reserved */
1300         p[5] = 0; /* bytes 5-7 are the sector size in bytes */
1301         p[6] = s->qdev.blocksize >> 8;
1302         p[7] = 0;
1303         p += 8;
1304     }
1305 
1306     if (page_control == 3) {
1307         /* Saved Values */
1308         scsi_check_condition(r, SENSE_CODE(SAVING_PARAMS_NOT_SUPPORTED));
1309         return -1;
1310     }
1311 
1312     if (page == 0x3f) {
1313         for (page = 0; page <= 0x3e; page++) {
1314             mode_sense_page(s, page, &p, page_control);
1315         }
1316     } else {
1317         ret = mode_sense_page(s, page, &p, page_control);
1318         if (ret == -1) {
1319             return -1;
1320         }
1321     }
1322 
1323     buflen = p - outbuf;
1324     /*
1325      * The mode data length field specifies the length in bytes of the
1326      * following data that is available to be transferred. The mode data
1327      * length does not include itself.
1328      */
1329     if (r->req.cmd.buf[0] == MODE_SENSE) {
1330         outbuf[0] = buflen - 1;
1331     } else { /* MODE_SENSE_10 */
1332         outbuf[0] = ((buflen - 2) >> 8) & 0xff;
1333         outbuf[1] = (buflen - 2) & 0xff;
1334     }
1335     return buflen;
1336 }
1337 
1338 static int scsi_disk_emulate_read_toc(SCSIRequest *req, uint8_t *outbuf)
1339 {
1340     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1341     int start_track, format, msf, toclen;
1342     uint64_t nb_sectors;
1343 
1344     msf = req->cmd.buf[1] & 2;
1345     format = req->cmd.buf[2] & 0xf;
1346     start_track = req->cmd.buf[6];
1347     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1348     trace_scsi_disk_emulate_read_toc(start_track, format, msf >> 1);
1349     nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
1350     switch (format) {
1351     case 0:
1352         toclen = cdrom_read_toc(nb_sectors, outbuf, msf, start_track);
1353         break;
1354     case 1:
1355         /* multi session : only a single session defined */
1356         toclen = 12;
1357         memset(outbuf, 0, 12);
1358         outbuf[1] = 0x0a;
1359         outbuf[2] = 0x01;
1360         outbuf[3] = 0x01;
1361         break;
1362     case 2:
1363         toclen = cdrom_read_toc_raw(nb_sectors, outbuf, msf, start_track);
1364         break;
1365     default:
1366         return -1;
1367     }
1368     return toclen;
1369 }
1370 
1371 static int scsi_disk_emulate_start_stop(SCSIDiskReq *r)
1372 {
1373     SCSIRequest *req = &r->req;
1374     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1375     bool start = req->cmd.buf[4] & 1;
1376     bool loej = req->cmd.buf[4] & 2; /* load on start, eject on !start */
1377     int pwrcnd = req->cmd.buf[4] & 0xf0;
1378 
1379     if (pwrcnd) {
1380         /* eject/load only happens for power condition == 0 */
1381         return 0;
1382     }
1383 
1384     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) && loej) {
1385         if (!start && !s->tray_open && s->tray_locked) {
1386             scsi_check_condition(r,
1387                                  blk_is_inserted(s->qdev.conf.blk)
1388                                  ? SENSE_CODE(ILLEGAL_REQ_REMOVAL_PREVENTED)
1389                                  : SENSE_CODE(NOT_READY_REMOVAL_PREVENTED));
1390             return -1;
1391         }
1392 
1393         if (s->tray_open != !start) {
1394             blk_eject(s->qdev.conf.blk, !start);
1395             s->tray_open = !start;
1396         }
1397     }
1398     return 0;
1399 }
1400 
1401 static void scsi_disk_emulate_read_data(SCSIRequest *req)
1402 {
1403     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1404     int buflen = r->iov.iov_len;
1405 
1406     if (buflen) {
1407         trace_scsi_disk_emulate_read_data(buflen);
1408         r->iov.iov_len = 0;
1409         r->started = true;
1410         scsi_req_data(&r->req, buflen);
1411         return;
1412     }
1413 
1414     /* This also clears the sense buffer for REQUEST SENSE.  */
1415     scsi_req_complete(&r->req, GOOD);
1416 }
1417 
1418 static int scsi_disk_check_mode_select(SCSIDiskState *s, int page,
1419                                        uint8_t *inbuf, int inlen)
1420 {
1421     uint8_t mode_current[SCSI_MAX_MODE_LEN];
1422     uint8_t mode_changeable[SCSI_MAX_MODE_LEN];
1423     uint8_t *p;
1424     int len, expected_len, changeable_len, i;
1425 
1426     /* The input buffer does not include the page header, so it is
1427      * off by 2 bytes.
1428      */
1429     expected_len = inlen + 2;
1430     if (expected_len > SCSI_MAX_MODE_LEN) {
1431         return -1;
1432     }
1433 
1434     /* MODE_PAGE_ALLS is only valid for MODE SENSE commands */
1435     if (page == MODE_PAGE_ALLS) {
1436         return -1;
1437     }
1438 
1439     p = mode_current;
1440     memset(mode_current, 0, inlen + 2);
1441     len = mode_sense_page(s, page, &p, 0);
1442     if (len < 0 || len != expected_len) {
1443         return -1;
1444     }
1445 
1446     p = mode_changeable;
1447     memset(mode_changeable, 0, inlen + 2);
1448     changeable_len = mode_sense_page(s, page, &p, 1);
1449     assert(changeable_len == len);
1450 
1451     /* Check that unchangeable bits are the same as what MODE SENSE
1452      * would return.
1453      */
1454     for (i = 2; i < len; i++) {
1455         if (((mode_current[i] ^ inbuf[i - 2]) & ~mode_changeable[i]) != 0) {
1456             return -1;
1457         }
1458     }
1459     return 0;
1460 }
1461 
1462 static void scsi_disk_apply_mode_select(SCSIDiskState *s, int page, uint8_t *p)
1463 {
1464     switch (page) {
1465     case MODE_PAGE_CACHING:
1466         blk_set_enable_write_cache(s->qdev.conf.blk, (p[0] & 4) != 0);
1467         break;
1468 
1469     default:
1470         break;
1471     }
1472 }
1473 
1474 static int mode_select_pages(SCSIDiskReq *r, uint8_t *p, int len, bool change)
1475 {
1476     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1477 
1478     while (len > 0) {
1479         int page, subpage, page_len;
1480 
1481         /* Parse both possible formats for the mode page headers.  */
1482         page = p[0] & 0x3f;
1483         if (p[0] & 0x40) {
1484             if (len < 4) {
1485                 goto invalid_param_len;
1486             }
1487             subpage = p[1];
1488             page_len = lduw_be_p(&p[2]);
1489             p += 4;
1490             len -= 4;
1491         } else {
1492             if (len < 2) {
1493                 goto invalid_param_len;
1494             }
1495             subpage = 0;
1496             page_len = p[1];
1497             p += 2;
1498             len -= 2;
1499         }
1500 
1501         if (subpage) {
1502             goto invalid_param;
1503         }
1504         if (page_len > len) {
1505             goto invalid_param_len;
1506         }
1507 
1508         if (!change) {
1509             if (scsi_disk_check_mode_select(s, page, p, page_len) < 0) {
1510                 goto invalid_param;
1511             }
1512         } else {
1513             scsi_disk_apply_mode_select(s, page, p);
1514         }
1515 
1516         p += page_len;
1517         len -= page_len;
1518     }
1519     return 0;
1520 
1521 invalid_param:
1522     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1523     return -1;
1524 
1525 invalid_param_len:
1526     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1527     return -1;
1528 }
1529 
1530 static void scsi_disk_emulate_mode_select(SCSIDiskReq *r, uint8_t *inbuf)
1531 {
1532     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1533     uint8_t *p = inbuf;
1534     int cmd = r->req.cmd.buf[0];
1535     int len = r->req.cmd.xfer;
1536     int hdr_len = (cmd == MODE_SELECT ? 4 : 8);
1537     int bd_len;
1538     int pass;
1539 
1540     /* We only support PF=1, SP=0.  */
1541     if ((r->req.cmd.buf[1] & 0x11) != 0x10) {
1542         goto invalid_field;
1543     }
1544 
1545     if (len < hdr_len) {
1546         goto invalid_param_len;
1547     }
1548 
1549     bd_len = (cmd == MODE_SELECT ? p[3] : lduw_be_p(&p[6]));
1550     len -= hdr_len;
1551     p += hdr_len;
1552     if (len < bd_len) {
1553         goto invalid_param_len;
1554     }
1555     if (bd_len != 0 && bd_len != 8) {
1556         goto invalid_param;
1557     }
1558 
1559     len -= bd_len;
1560     p += bd_len;
1561 
1562     /* Ensure no change is made if there is an error!  */
1563     for (pass = 0; pass < 2; pass++) {
1564         if (mode_select_pages(r, p, len, pass == 1) < 0) {
1565             assert(pass == 0);
1566             return;
1567         }
1568     }
1569     if (!blk_enable_write_cache(s->qdev.conf.blk)) {
1570         /* The request is used as the AIO opaque value, so add a ref.  */
1571         scsi_req_ref(&r->req);
1572         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
1573                          BLOCK_ACCT_FLUSH);
1574         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
1575         return;
1576     }
1577 
1578     scsi_req_complete(&r->req, GOOD);
1579     return;
1580 
1581 invalid_param:
1582     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1583     return;
1584 
1585 invalid_param_len:
1586     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1587     return;
1588 
1589 invalid_field:
1590     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1591 }
1592 
1593 /* sector_num and nb_sectors expected to be in qdev blocksize */
1594 static inline bool check_lba_range(SCSIDiskState *s,
1595                                    uint64_t sector_num, uint32_t nb_sectors)
1596 {
1597     /*
1598      * The first line tests that no overflow happens when computing the last
1599      * sector.  The second line tests that the last accessed sector is in
1600      * range.
1601      *
1602      * Careful, the computations should not underflow for nb_sectors == 0,
1603      * and a 0-block read to the first LBA beyond the end of device is
1604      * valid.
1605      */
1606     return (sector_num <= sector_num + nb_sectors &&
1607             sector_num + nb_sectors <= s->qdev.max_lba + 1);
1608 }
1609 
1610 typedef struct UnmapCBData {
1611     SCSIDiskReq *r;
1612     uint8_t *inbuf;
1613     int count;
1614 } UnmapCBData;
1615 
1616 static void scsi_unmap_complete(void *opaque, int ret);
1617 
1618 static void scsi_unmap_complete_noio(UnmapCBData *data, int ret)
1619 {
1620     SCSIDiskReq *r = data->r;
1621     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1622 
1623     assert(r->req.aiocb == NULL);
1624 
1625     if (data->count > 0) {
1626         uint64_t sector_num = ldq_be_p(&data->inbuf[0]);
1627         uint32_t nb_sectors = ldl_be_p(&data->inbuf[8]) & 0xffffffffULL;
1628         r->sector = sector_num * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1629         r->sector_count = nb_sectors * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1630 
1631         if (!check_lba_range(s, sector_num, nb_sectors)) {
1632             block_acct_invalid(blk_get_stats(s->qdev.conf.blk),
1633                                BLOCK_ACCT_UNMAP);
1634             scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1635             goto done;
1636         }
1637 
1638         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1639                          r->sector_count * BDRV_SECTOR_SIZE,
1640                          BLOCK_ACCT_UNMAP);
1641 
1642         r->req.aiocb = blk_aio_pdiscard(s->qdev.conf.blk,
1643                                         r->sector * BDRV_SECTOR_SIZE,
1644                                         r->sector_count * BDRV_SECTOR_SIZE,
1645                                         scsi_unmap_complete, data);
1646         data->count--;
1647         data->inbuf += 16;
1648         return;
1649     }
1650 
1651     scsi_req_complete(&r->req, GOOD);
1652 
1653 done:
1654     scsi_req_unref(&r->req);
1655     g_free(data);
1656 }
1657 
1658 static void scsi_unmap_complete(void *opaque, int ret)
1659 {
1660     UnmapCBData *data = opaque;
1661     SCSIDiskReq *r = data->r;
1662     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1663 
1664     assert(r->req.aiocb != NULL);
1665     r->req.aiocb = NULL;
1666 
1667     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
1668     if (scsi_disk_req_check_error(r, ret, true)) {
1669         scsi_req_unref(&r->req);
1670         g_free(data);
1671     } else {
1672         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1673         scsi_unmap_complete_noio(data, ret);
1674     }
1675     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1676 }
1677 
1678 static void scsi_disk_emulate_unmap(SCSIDiskReq *r, uint8_t *inbuf)
1679 {
1680     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1681     uint8_t *p = inbuf;
1682     int len = r->req.cmd.xfer;
1683     UnmapCBData *data;
1684 
1685     /* Reject ANCHOR=1.  */
1686     if (r->req.cmd.buf[1] & 0x1) {
1687         goto invalid_field;
1688     }
1689 
1690     if (len < 8) {
1691         goto invalid_param_len;
1692     }
1693     if (len < lduw_be_p(&p[0]) + 2) {
1694         goto invalid_param_len;
1695     }
1696     if (len < lduw_be_p(&p[2]) + 8) {
1697         goto invalid_param_len;
1698     }
1699     if (lduw_be_p(&p[2]) & 15) {
1700         goto invalid_param_len;
1701     }
1702 
1703     if (!blk_is_writable(s->qdev.conf.blk)) {
1704         block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1705         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1706         return;
1707     }
1708 
1709     data = g_new0(UnmapCBData, 1);
1710     data->r = r;
1711     data->inbuf = &p[8];
1712     data->count = lduw_be_p(&p[2]) >> 4;
1713 
1714     /* The matching unref is in scsi_unmap_complete, before data is freed.  */
1715     scsi_req_ref(&r->req);
1716     scsi_unmap_complete_noio(data, 0);
1717     return;
1718 
1719 invalid_param_len:
1720     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1721     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1722     return;
1723 
1724 invalid_field:
1725     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1726     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1727 }
1728 
1729 typedef struct WriteSameCBData {
1730     SCSIDiskReq *r;
1731     int64_t sector;
1732     int nb_sectors;
1733     QEMUIOVector qiov;
1734     struct iovec iov;
1735 } WriteSameCBData;
1736 
1737 static void scsi_write_same_complete(void *opaque, int ret)
1738 {
1739     WriteSameCBData *data = opaque;
1740     SCSIDiskReq *r = data->r;
1741     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1742 
1743     assert(r->req.aiocb != NULL);
1744     r->req.aiocb = NULL;
1745     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
1746     if (scsi_disk_req_check_error(r, ret, true)) {
1747         goto done;
1748     }
1749 
1750     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1751 
1752     data->nb_sectors -= data->iov.iov_len / BDRV_SECTOR_SIZE;
1753     data->sector += data->iov.iov_len / BDRV_SECTOR_SIZE;
1754     data->iov.iov_len = MIN(data->nb_sectors * BDRV_SECTOR_SIZE,
1755                             data->iov.iov_len);
1756     if (data->iov.iov_len) {
1757         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1758                          data->iov.iov_len, BLOCK_ACCT_WRITE);
1759         /* Reinitialize qiov, to handle unaligned WRITE SAME request
1760          * where final qiov may need smaller size */
1761         qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1762         r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1763                                        data->sector << BDRV_SECTOR_BITS,
1764                                        &data->qiov, 0,
1765                                        scsi_write_same_complete, data);
1766         aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1767         return;
1768     }
1769 
1770     scsi_req_complete(&r->req, GOOD);
1771 
1772 done:
1773     scsi_req_unref(&r->req);
1774     qemu_vfree(data->iov.iov_base);
1775     g_free(data);
1776     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1777 }
1778 
1779 static void scsi_disk_emulate_write_same(SCSIDiskReq *r, uint8_t *inbuf)
1780 {
1781     SCSIRequest *req = &r->req;
1782     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1783     uint32_t nb_sectors = scsi_data_cdb_xfer(r->req.cmd.buf);
1784     WriteSameCBData *data;
1785     uint8_t *buf;
1786     int i;
1787 
1788     /* Fail if PBDATA=1 or LBDATA=1 or ANCHOR=1.  */
1789     if (nb_sectors == 0 || (req->cmd.buf[1] & 0x16)) {
1790         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1791         return;
1792     }
1793 
1794     if (!blk_is_writable(s->qdev.conf.blk)) {
1795         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1796         return;
1797     }
1798     if (!check_lba_range(s, r->req.cmd.lba, nb_sectors)) {
1799         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1800         return;
1801     }
1802 
1803     if ((req->cmd.buf[1] & 0x1) || buffer_is_zero(inbuf, s->qdev.blocksize)) {
1804         int flags = (req->cmd.buf[1] & 0x8) ? BDRV_REQ_MAY_UNMAP : 0;
1805 
1806         /* The request is used as the AIO opaque value, so add a ref.  */
1807         scsi_req_ref(&r->req);
1808         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1809                          nb_sectors * s->qdev.blocksize,
1810                         BLOCK_ACCT_WRITE);
1811         r->req.aiocb = blk_aio_pwrite_zeroes(s->qdev.conf.blk,
1812                                 r->req.cmd.lba * s->qdev.blocksize,
1813                                 nb_sectors * s->qdev.blocksize,
1814                                 flags, scsi_aio_complete, r);
1815         return;
1816     }
1817 
1818     data = g_new0(WriteSameCBData, 1);
1819     data->r = r;
1820     data->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1821     data->nb_sectors = nb_sectors * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1822     data->iov.iov_len = MIN(data->nb_sectors * BDRV_SECTOR_SIZE,
1823                             SCSI_WRITE_SAME_MAX);
1824     data->iov.iov_base = buf = blk_blockalign(s->qdev.conf.blk,
1825                                               data->iov.iov_len);
1826     qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1827 
1828     for (i = 0; i < data->iov.iov_len; i += s->qdev.blocksize) {
1829         memcpy(&buf[i], inbuf, s->qdev.blocksize);
1830     }
1831 
1832     scsi_req_ref(&r->req);
1833     block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1834                      data->iov.iov_len, BLOCK_ACCT_WRITE);
1835     r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1836                                    data->sector << BDRV_SECTOR_BITS,
1837                                    &data->qiov, 0,
1838                                    scsi_write_same_complete, data);
1839 }
1840 
1841 static void scsi_disk_emulate_write_data(SCSIRequest *req)
1842 {
1843     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1844 
1845     if (r->iov.iov_len) {
1846         int buflen = r->iov.iov_len;
1847         trace_scsi_disk_emulate_write_data(buflen);
1848         r->iov.iov_len = 0;
1849         scsi_req_data(&r->req, buflen);
1850         return;
1851     }
1852 
1853     switch (req->cmd.buf[0]) {
1854     case MODE_SELECT:
1855     case MODE_SELECT_10:
1856         /* This also clears the sense buffer for REQUEST SENSE.  */
1857         scsi_disk_emulate_mode_select(r, r->iov.iov_base);
1858         break;
1859 
1860     case UNMAP:
1861         scsi_disk_emulate_unmap(r, r->iov.iov_base);
1862         break;
1863 
1864     case VERIFY_10:
1865     case VERIFY_12:
1866     case VERIFY_16:
1867         if (r->req.status == -1) {
1868             scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1869         }
1870         break;
1871 
1872     case WRITE_SAME_10:
1873     case WRITE_SAME_16:
1874         scsi_disk_emulate_write_same(r, r->iov.iov_base);
1875         break;
1876 
1877     default:
1878         abort();
1879     }
1880 }
1881 
1882 static int32_t scsi_disk_emulate_command(SCSIRequest *req, uint8_t *buf)
1883 {
1884     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1885     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1886     uint64_t nb_sectors;
1887     uint8_t *outbuf;
1888     int buflen;
1889 
1890     switch (req->cmd.buf[0]) {
1891     case INQUIRY:
1892     case MODE_SENSE:
1893     case MODE_SENSE_10:
1894     case RESERVE:
1895     case RESERVE_10:
1896     case RELEASE:
1897     case RELEASE_10:
1898     case START_STOP:
1899     case ALLOW_MEDIUM_REMOVAL:
1900     case GET_CONFIGURATION:
1901     case GET_EVENT_STATUS_NOTIFICATION:
1902     case MECHANISM_STATUS:
1903     case REQUEST_SENSE:
1904         break;
1905 
1906     default:
1907         if (!blk_is_available(s->qdev.conf.blk)) {
1908             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
1909             return 0;
1910         }
1911         break;
1912     }
1913 
1914     /*
1915      * FIXME: we shouldn't return anything bigger than 4k, but the code
1916      * requires the buffer to be as big as req->cmd.xfer in several
1917      * places.  So, do not allow CDBs with a very large ALLOCATION
1918      * LENGTH.  The real fix would be to modify scsi_read_data and
1919      * dma_buf_read, so that they return data beyond the buflen
1920      * as all zeros.
1921      */
1922     if (req->cmd.xfer > 65536) {
1923         goto illegal_request;
1924     }
1925     r->buflen = MAX(4096, req->cmd.xfer);
1926 
1927     if (!r->iov.iov_base) {
1928         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
1929     }
1930 
1931     outbuf = r->iov.iov_base;
1932     memset(outbuf, 0, r->buflen);
1933     switch (req->cmd.buf[0]) {
1934     case TEST_UNIT_READY:
1935         assert(blk_is_available(s->qdev.conf.blk));
1936         break;
1937     case INQUIRY:
1938         buflen = scsi_disk_emulate_inquiry(req, outbuf);
1939         if (buflen < 0) {
1940             goto illegal_request;
1941         }
1942         break;
1943     case MODE_SENSE:
1944     case MODE_SENSE_10:
1945         buflen = scsi_disk_emulate_mode_sense(r, outbuf);
1946         if (buflen < 0) {
1947             goto illegal_request;
1948         }
1949         break;
1950     case READ_TOC:
1951         buflen = scsi_disk_emulate_read_toc(req, outbuf);
1952         if (buflen < 0) {
1953             goto illegal_request;
1954         }
1955         break;
1956     case RESERVE:
1957         if (req->cmd.buf[1] & 1) {
1958             goto illegal_request;
1959         }
1960         break;
1961     case RESERVE_10:
1962         if (req->cmd.buf[1] & 3) {
1963             goto illegal_request;
1964         }
1965         break;
1966     case RELEASE:
1967         if (req->cmd.buf[1] & 1) {
1968             goto illegal_request;
1969         }
1970         break;
1971     case RELEASE_10:
1972         if (req->cmd.buf[1] & 3) {
1973             goto illegal_request;
1974         }
1975         break;
1976     case START_STOP:
1977         if (scsi_disk_emulate_start_stop(r) < 0) {
1978             return 0;
1979         }
1980         break;
1981     case ALLOW_MEDIUM_REMOVAL:
1982         s->tray_locked = req->cmd.buf[4] & 1;
1983         blk_lock_medium(s->qdev.conf.blk, req->cmd.buf[4] & 1);
1984         break;
1985     case READ_CAPACITY_10:
1986         /* The normal LEN field for this command is zero.  */
1987         memset(outbuf, 0, 8);
1988         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1989         if (!nb_sectors) {
1990             scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
1991             return 0;
1992         }
1993         if ((req->cmd.buf[8] & 1) == 0 && req->cmd.lba) {
1994             goto illegal_request;
1995         }
1996         nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
1997         /* Returned value is the address of the last sector.  */
1998         nb_sectors--;
1999         /* Remember the new size for read/write sanity checking. */
2000         s->qdev.max_lba = nb_sectors;
2001         /* Clip to 2TB, instead of returning capacity modulo 2TB. */
2002         if (nb_sectors > UINT32_MAX) {
2003             nb_sectors = UINT32_MAX;
2004         }
2005         outbuf[0] = (nb_sectors >> 24) & 0xff;
2006         outbuf[1] = (nb_sectors >> 16) & 0xff;
2007         outbuf[2] = (nb_sectors >> 8) & 0xff;
2008         outbuf[3] = nb_sectors & 0xff;
2009         outbuf[4] = 0;
2010         outbuf[5] = 0;
2011         outbuf[6] = s->qdev.blocksize >> 8;
2012         outbuf[7] = 0;
2013         break;
2014     case REQUEST_SENSE:
2015         /* Just return "NO SENSE".  */
2016         buflen = scsi_convert_sense(NULL, 0, outbuf, r->buflen,
2017                                     (req->cmd.buf[1] & 1) == 0);
2018         if (buflen < 0) {
2019             goto illegal_request;
2020         }
2021         break;
2022     case MECHANISM_STATUS:
2023         buflen = scsi_emulate_mechanism_status(s, outbuf);
2024         if (buflen < 0) {
2025             goto illegal_request;
2026         }
2027         break;
2028     case GET_CONFIGURATION:
2029         buflen = scsi_get_configuration(s, outbuf);
2030         if (buflen < 0) {
2031             goto illegal_request;
2032         }
2033         break;
2034     case GET_EVENT_STATUS_NOTIFICATION:
2035         buflen = scsi_get_event_status_notification(s, r, outbuf);
2036         if (buflen < 0) {
2037             goto illegal_request;
2038         }
2039         break;
2040     case READ_DISC_INFORMATION:
2041         buflen = scsi_read_disc_information(s, r, outbuf);
2042         if (buflen < 0) {
2043             goto illegal_request;
2044         }
2045         break;
2046     case READ_DVD_STRUCTURE:
2047         buflen = scsi_read_dvd_structure(s, r, outbuf);
2048         if (buflen < 0) {
2049             goto illegal_request;
2050         }
2051         break;
2052     case SERVICE_ACTION_IN_16:
2053         /* Service Action In subcommands. */
2054         if ((req->cmd.buf[1] & 31) == SAI_READ_CAPACITY_16) {
2055             trace_scsi_disk_emulate_command_SAI_16();
2056             memset(outbuf, 0, req->cmd.xfer);
2057             blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2058             if (!nb_sectors) {
2059                 scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
2060                 return 0;
2061             }
2062             if ((req->cmd.buf[14] & 1) == 0 && req->cmd.lba) {
2063                 goto illegal_request;
2064             }
2065             nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2066             /* Returned value is the address of the last sector.  */
2067             nb_sectors--;
2068             /* Remember the new size for read/write sanity checking. */
2069             s->qdev.max_lba = nb_sectors;
2070             outbuf[0] = (nb_sectors >> 56) & 0xff;
2071             outbuf[1] = (nb_sectors >> 48) & 0xff;
2072             outbuf[2] = (nb_sectors >> 40) & 0xff;
2073             outbuf[3] = (nb_sectors >> 32) & 0xff;
2074             outbuf[4] = (nb_sectors >> 24) & 0xff;
2075             outbuf[5] = (nb_sectors >> 16) & 0xff;
2076             outbuf[6] = (nb_sectors >> 8) & 0xff;
2077             outbuf[7] = nb_sectors & 0xff;
2078             outbuf[8] = 0;
2079             outbuf[9] = 0;
2080             outbuf[10] = s->qdev.blocksize >> 8;
2081             outbuf[11] = 0;
2082             outbuf[12] = 0;
2083             outbuf[13] = get_physical_block_exp(&s->qdev.conf);
2084 
2085             /* set TPE bit if the format supports discard */
2086             if (s->qdev.conf.discard_granularity) {
2087                 outbuf[14] = 0x80;
2088             }
2089 
2090             /* Protection, exponent and lowest lba field left blank. */
2091             break;
2092         }
2093         trace_scsi_disk_emulate_command_SAI_unsupported();
2094         goto illegal_request;
2095     case SYNCHRONIZE_CACHE:
2096         /* The request is used as the AIO opaque value, so add a ref.  */
2097         scsi_req_ref(&r->req);
2098         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
2099                          BLOCK_ACCT_FLUSH);
2100         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
2101         return 0;
2102     case SEEK_10:
2103         trace_scsi_disk_emulate_command_SEEK_10(r->req.cmd.lba);
2104         if (r->req.cmd.lba > s->qdev.max_lba) {
2105             goto illegal_lba;
2106         }
2107         break;
2108     case MODE_SELECT:
2109         trace_scsi_disk_emulate_command_MODE_SELECT(r->req.cmd.xfer);
2110         break;
2111     case MODE_SELECT_10:
2112         trace_scsi_disk_emulate_command_MODE_SELECT_10(r->req.cmd.xfer);
2113         break;
2114     case UNMAP:
2115         trace_scsi_disk_emulate_command_UNMAP(r->req.cmd.xfer);
2116         break;
2117     case VERIFY_10:
2118     case VERIFY_12:
2119     case VERIFY_16:
2120         trace_scsi_disk_emulate_command_VERIFY((req->cmd.buf[1] >> 1) & 3);
2121         if (req->cmd.buf[1] & 6) {
2122             goto illegal_request;
2123         }
2124         break;
2125     case WRITE_SAME_10:
2126     case WRITE_SAME_16:
2127         trace_scsi_disk_emulate_command_WRITE_SAME(
2128                 req->cmd.buf[0] == WRITE_SAME_10 ? 10 : 16, r->req.cmd.xfer);
2129         break;
2130     default:
2131         trace_scsi_disk_emulate_command_UNKNOWN(buf[0],
2132                                                 scsi_command_name(buf[0]));
2133         scsi_check_condition(r, SENSE_CODE(INVALID_OPCODE));
2134         return 0;
2135     }
2136     assert(!r->req.aiocb);
2137     r->iov.iov_len = MIN(r->buflen, req->cmd.xfer);
2138     if (r->iov.iov_len == 0) {
2139         scsi_req_complete(&r->req, GOOD);
2140     }
2141     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2142         assert(r->iov.iov_len == req->cmd.xfer);
2143         return -r->iov.iov_len;
2144     } else {
2145         return r->iov.iov_len;
2146     }
2147 
2148 illegal_request:
2149     if (r->req.status == -1) {
2150         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2151     }
2152     return 0;
2153 
2154 illegal_lba:
2155     scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2156     return 0;
2157 }
2158 
2159 /* Execute a scsi command.  Returns the length of the data expected by the
2160    command.  This will be Positive for data transfers from the device
2161    (eg. disk reads), negative for transfers to the device (eg. disk writes),
2162    and zero if the command does not transfer any data.  */
2163 
2164 static int32_t scsi_disk_dma_command(SCSIRequest *req, uint8_t *buf)
2165 {
2166     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
2167     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2168     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
2169     uint32_t len;
2170     uint8_t command;
2171 
2172     command = buf[0];
2173 
2174     if (!blk_is_available(s->qdev.conf.blk)) {
2175         scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
2176         return 0;
2177     }
2178 
2179     len = scsi_data_cdb_xfer(r->req.cmd.buf);
2180     switch (command) {
2181     case READ_6:
2182     case READ_10:
2183     case READ_12:
2184     case READ_16:
2185         trace_scsi_disk_dma_command_READ(r->req.cmd.lba, len);
2186         /* Protection information is not supported.  For SCSI versions 2 and
2187          * older (as determined by snooping the guest's INQUIRY commands),
2188          * there is no RD/WR/VRPROTECT, so skip this check in these versions.
2189          */
2190         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2191             goto illegal_request;
2192         }
2193         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2194             goto illegal_lba;
2195         }
2196         r->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2197         r->sector_count = len * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2198         break;
2199     case WRITE_6:
2200     case WRITE_10:
2201     case WRITE_12:
2202     case WRITE_16:
2203     case WRITE_VERIFY_10:
2204     case WRITE_VERIFY_12:
2205     case WRITE_VERIFY_16:
2206         if (!blk_is_writable(s->qdev.conf.blk)) {
2207             scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
2208             return 0;
2209         }
2210         trace_scsi_disk_dma_command_WRITE(
2211                 (command & 0xe) == 0xe ? "And Verify " : "",
2212                 r->req.cmd.lba, len);
2213         /* fall through */
2214     case VERIFY_10:
2215     case VERIFY_12:
2216     case VERIFY_16:
2217         /* We get here only for BYTCHK == 0x01 and only for scsi-block.
2218          * As far as DMA is concerned, we can treat it the same as a write;
2219          * scsi_block_do_sgio will send VERIFY commands.
2220          */
2221         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2222             goto illegal_request;
2223         }
2224         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2225             goto illegal_lba;
2226         }
2227         r->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2228         r->sector_count = len * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2229         break;
2230     default:
2231         abort();
2232     illegal_request:
2233         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2234         return 0;
2235     illegal_lba:
2236         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2237         return 0;
2238     }
2239     r->need_fua_emulation = sdc->need_fua_emulation(&r->req.cmd);
2240     if (r->sector_count == 0) {
2241         scsi_req_complete(&r->req, GOOD);
2242     }
2243     assert(r->iov.iov_len == 0);
2244     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2245         return -r->sector_count * BDRV_SECTOR_SIZE;
2246     } else {
2247         return r->sector_count * BDRV_SECTOR_SIZE;
2248     }
2249 }
2250 
2251 static void scsi_disk_reset(DeviceState *dev)
2252 {
2253     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev.qdev, dev);
2254     uint64_t nb_sectors;
2255 
2256     scsi_device_purge_requests(&s->qdev, SENSE_CODE(RESET));
2257 
2258     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2259     nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2260     if (nb_sectors) {
2261         nb_sectors--;
2262     }
2263     s->qdev.max_lba = nb_sectors;
2264     /* reset tray statuses */
2265     s->tray_locked = 0;
2266     s->tray_open = 0;
2267 
2268     s->qdev.scsi_version = s->qdev.default_scsi_version;
2269 }
2270 
2271 static void scsi_disk_resize_cb(void *opaque)
2272 {
2273     SCSIDiskState *s = opaque;
2274 
2275     /* SPC lists this sense code as available only for
2276      * direct-access devices.
2277      */
2278     if (s->qdev.type == TYPE_DISK) {
2279         scsi_device_report_change(&s->qdev, SENSE_CODE(CAPACITY_CHANGED));
2280     }
2281 }
2282 
2283 static void scsi_cd_change_media_cb(void *opaque, bool load, Error **errp)
2284 {
2285     SCSIDiskState *s = opaque;
2286 
2287     /*
2288      * When a CD gets changed, we have to report an ejected state and
2289      * then a loaded state to guests so that they detect tray
2290      * open/close and media change events.  Guests that do not use
2291      * GET_EVENT_STATUS_NOTIFICATION to detect such tray open/close
2292      * states rely on this behavior.
2293      *
2294      * media_changed governs the state machine used for unit attention
2295      * report.  media_event is used by GET EVENT STATUS NOTIFICATION.
2296      */
2297     s->media_changed = load;
2298     s->tray_open = !load;
2299     scsi_device_set_ua(&s->qdev, SENSE_CODE(UNIT_ATTENTION_NO_MEDIUM));
2300     s->media_event = true;
2301     s->eject_request = false;
2302 }
2303 
2304 static void scsi_cd_eject_request_cb(void *opaque, bool force)
2305 {
2306     SCSIDiskState *s = opaque;
2307 
2308     s->eject_request = true;
2309     if (force) {
2310         s->tray_locked = false;
2311     }
2312 }
2313 
2314 static bool scsi_cd_is_tray_open(void *opaque)
2315 {
2316     return ((SCSIDiskState *)opaque)->tray_open;
2317 }
2318 
2319 static bool scsi_cd_is_medium_locked(void *opaque)
2320 {
2321     return ((SCSIDiskState *)opaque)->tray_locked;
2322 }
2323 
2324 static const BlockDevOps scsi_disk_removable_block_ops = {
2325     .change_media_cb = scsi_cd_change_media_cb,
2326     .eject_request_cb = scsi_cd_eject_request_cb,
2327     .is_tray_open = scsi_cd_is_tray_open,
2328     .is_medium_locked = scsi_cd_is_medium_locked,
2329 
2330     .resize_cb = scsi_disk_resize_cb,
2331 };
2332 
2333 static const BlockDevOps scsi_disk_block_ops = {
2334     .resize_cb = scsi_disk_resize_cb,
2335 };
2336 
2337 static void scsi_disk_unit_attention_reported(SCSIDevice *dev)
2338 {
2339     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2340     if (s->media_changed) {
2341         s->media_changed = false;
2342         scsi_device_set_ua(&s->qdev, SENSE_CODE(MEDIUM_CHANGED));
2343     }
2344 }
2345 
2346 static void scsi_realize(SCSIDevice *dev, Error **errp)
2347 {
2348     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2349     bool read_only;
2350 
2351     if (!s->qdev.conf.blk) {
2352         error_setg(errp, "drive property not set");
2353         return;
2354     }
2355 
2356     if (!(s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2357         !blk_is_inserted(s->qdev.conf.blk)) {
2358         error_setg(errp, "Device needs media, but drive is empty");
2359         return;
2360     }
2361 
2362     if (!blkconf_blocksizes(&s->qdev.conf, errp)) {
2363         return;
2364     }
2365 
2366     if (blk_get_aio_context(s->qdev.conf.blk) != qemu_get_aio_context() &&
2367         !s->qdev.hba_supports_iothread)
2368     {
2369         error_setg(errp, "HBA does not support iothreads");
2370         return;
2371     }
2372 
2373     if (dev->type == TYPE_DISK) {
2374         if (!blkconf_geometry(&dev->conf, NULL, 65535, 255, 255, errp)) {
2375             return;
2376         }
2377     }
2378 
2379     read_only = !blk_supports_write_perm(s->qdev.conf.blk);
2380     if (dev->type == TYPE_ROM) {
2381         read_only = true;
2382     }
2383 
2384     if (!blkconf_apply_backend_options(&dev->conf, read_only,
2385                                        dev->type == TYPE_DISK, errp)) {
2386         return;
2387     }
2388 
2389     if (s->qdev.conf.discard_granularity == -1) {
2390         s->qdev.conf.discard_granularity =
2391             MAX(s->qdev.conf.logical_block_size, DEFAULT_DISCARD_GRANULARITY);
2392     }
2393 
2394     if (!s->version) {
2395         s->version = g_strdup(qemu_hw_version());
2396     }
2397     if (!s->vendor) {
2398         s->vendor = g_strdup("QEMU");
2399     }
2400     if (!s->device_id) {
2401         if (s->serial) {
2402             s->device_id = g_strdup_printf("%.20s", s->serial);
2403         } else {
2404             const char *str = blk_name(s->qdev.conf.blk);
2405             if (str && *str) {
2406                 s->device_id = g_strdup(str);
2407             }
2408         }
2409     }
2410 
2411     if (blk_is_sg(s->qdev.conf.blk)) {
2412         error_setg(errp, "unwanted /dev/sg*");
2413         return;
2414     }
2415 
2416     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2417             !(s->features & (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS))) {
2418         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_removable_block_ops, s);
2419     } else {
2420         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_block_ops, s);
2421     }
2422     blk_set_guest_block_size(s->qdev.conf.blk, s->qdev.blocksize);
2423 
2424     blk_iostatus_enable(s->qdev.conf.blk);
2425 
2426     add_boot_device_lchs(&dev->qdev, NULL,
2427                          dev->conf.lcyls,
2428                          dev->conf.lheads,
2429                          dev->conf.lsecs);
2430 }
2431 
2432 static void scsi_unrealize(SCSIDevice *dev)
2433 {
2434     del_boot_device_lchs(&dev->qdev, NULL);
2435 }
2436 
2437 static void scsi_hd_realize(SCSIDevice *dev, Error **errp)
2438 {
2439     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2440     AioContext *ctx = NULL;
2441     /* can happen for devices without drive. The error message for missing
2442      * backend will be issued in scsi_realize
2443      */
2444     if (s->qdev.conf.blk) {
2445         ctx = blk_get_aio_context(s->qdev.conf.blk);
2446         aio_context_acquire(ctx);
2447         if (!blkconf_blocksizes(&s->qdev.conf, errp)) {
2448             goto out;
2449         }
2450     }
2451     s->qdev.blocksize = s->qdev.conf.logical_block_size;
2452     s->qdev.type = TYPE_DISK;
2453     if (!s->product) {
2454         s->product = g_strdup("QEMU HARDDISK");
2455     }
2456     scsi_realize(&s->qdev, errp);
2457 out:
2458     if (ctx) {
2459         aio_context_release(ctx);
2460     }
2461 }
2462 
2463 static void scsi_cd_realize(SCSIDevice *dev, Error **errp)
2464 {
2465     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2466     AioContext *ctx;
2467     int ret;
2468 
2469     if (!dev->conf.blk) {
2470         /* Anonymous BlockBackend for an empty drive. As we put it into
2471          * dev->conf, qdev takes care of detaching on unplug. */
2472         dev->conf.blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
2473         ret = blk_attach_dev(dev->conf.blk, &dev->qdev);
2474         assert(ret == 0);
2475     }
2476 
2477     ctx = blk_get_aio_context(dev->conf.blk);
2478     aio_context_acquire(ctx);
2479     s->qdev.blocksize = 2048;
2480     s->qdev.type = TYPE_ROM;
2481     s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2482     if (!s->product) {
2483         s->product = g_strdup("QEMU CD-ROM");
2484     }
2485     scsi_realize(&s->qdev, errp);
2486     aio_context_release(ctx);
2487 }
2488 
2489 
2490 static const SCSIReqOps scsi_disk_emulate_reqops = {
2491     .size         = sizeof(SCSIDiskReq),
2492     .free_req     = scsi_free_request,
2493     .send_command = scsi_disk_emulate_command,
2494     .read_data    = scsi_disk_emulate_read_data,
2495     .write_data   = scsi_disk_emulate_write_data,
2496     .get_buf      = scsi_get_buf,
2497 };
2498 
2499 static const SCSIReqOps scsi_disk_dma_reqops = {
2500     .size         = sizeof(SCSIDiskReq),
2501     .free_req     = scsi_free_request,
2502     .send_command = scsi_disk_dma_command,
2503     .read_data    = scsi_read_data,
2504     .write_data   = scsi_write_data,
2505     .get_buf      = scsi_get_buf,
2506     .load_request = scsi_disk_load_request,
2507     .save_request = scsi_disk_save_request,
2508 };
2509 
2510 static const SCSIReqOps *const scsi_disk_reqops_dispatch[256] = {
2511     [TEST_UNIT_READY]                 = &scsi_disk_emulate_reqops,
2512     [INQUIRY]                         = &scsi_disk_emulate_reqops,
2513     [MODE_SENSE]                      = &scsi_disk_emulate_reqops,
2514     [MODE_SENSE_10]                   = &scsi_disk_emulate_reqops,
2515     [START_STOP]                      = &scsi_disk_emulate_reqops,
2516     [ALLOW_MEDIUM_REMOVAL]            = &scsi_disk_emulate_reqops,
2517     [READ_CAPACITY_10]                = &scsi_disk_emulate_reqops,
2518     [READ_TOC]                        = &scsi_disk_emulate_reqops,
2519     [READ_DVD_STRUCTURE]              = &scsi_disk_emulate_reqops,
2520     [READ_DISC_INFORMATION]           = &scsi_disk_emulate_reqops,
2521     [GET_CONFIGURATION]               = &scsi_disk_emulate_reqops,
2522     [GET_EVENT_STATUS_NOTIFICATION]   = &scsi_disk_emulate_reqops,
2523     [MECHANISM_STATUS]                = &scsi_disk_emulate_reqops,
2524     [SERVICE_ACTION_IN_16]            = &scsi_disk_emulate_reqops,
2525     [REQUEST_SENSE]                   = &scsi_disk_emulate_reqops,
2526     [SYNCHRONIZE_CACHE]               = &scsi_disk_emulate_reqops,
2527     [SEEK_10]                         = &scsi_disk_emulate_reqops,
2528     [MODE_SELECT]                     = &scsi_disk_emulate_reqops,
2529     [MODE_SELECT_10]                  = &scsi_disk_emulate_reqops,
2530     [UNMAP]                           = &scsi_disk_emulate_reqops,
2531     [WRITE_SAME_10]                   = &scsi_disk_emulate_reqops,
2532     [WRITE_SAME_16]                   = &scsi_disk_emulate_reqops,
2533     [VERIFY_10]                       = &scsi_disk_emulate_reqops,
2534     [VERIFY_12]                       = &scsi_disk_emulate_reqops,
2535     [VERIFY_16]                       = &scsi_disk_emulate_reqops,
2536 
2537     [READ_6]                          = &scsi_disk_dma_reqops,
2538     [READ_10]                         = &scsi_disk_dma_reqops,
2539     [READ_12]                         = &scsi_disk_dma_reqops,
2540     [READ_16]                         = &scsi_disk_dma_reqops,
2541     [WRITE_6]                         = &scsi_disk_dma_reqops,
2542     [WRITE_10]                        = &scsi_disk_dma_reqops,
2543     [WRITE_12]                        = &scsi_disk_dma_reqops,
2544     [WRITE_16]                        = &scsi_disk_dma_reqops,
2545     [WRITE_VERIFY_10]                 = &scsi_disk_dma_reqops,
2546     [WRITE_VERIFY_12]                 = &scsi_disk_dma_reqops,
2547     [WRITE_VERIFY_16]                 = &scsi_disk_dma_reqops,
2548 };
2549 
2550 static void scsi_disk_new_request_dump(uint32_t lun, uint32_t tag, uint8_t *buf)
2551 {
2552     int i;
2553     int len = scsi_cdb_length(buf);
2554     char *line_buffer, *p;
2555 
2556     assert(len > 0 && len <= 16);
2557     line_buffer = g_malloc(len * 5 + 1);
2558 
2559     for (i = 0, p = line_buffer; i < len; i++) {
2560         p += sprintf(p, " 0x%02x", buf[i]);
2561     }
2562     trace_scsi_disk_new_request(lun, tag, line_buffer);
2563 
2564     g_free(line_buffer);
2565 }
2566 
2567 static SCSIRequest *scsi_new_request(SCSIDevice *d, uint32_t tag, uint32_t lun,
2568                                      uint8_t *buf, void *hba_private)
2569 {
2570     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2571     SCSIRequest *req;
2572     const SCSIReqOps *ops;
2573     uint8_t command;
2574 
2575     command = buf[0];
2576     ops = scsi_disk_reqops_dispatch[command];
2577     if (!ops) {
2578         ops = &scsi_disk_emulate_reqops;
2579     }
2580     req = scsi_req_alloc(ops, &s->qdev, tag, lun, hba_private);
2581 
2582     if (trace_event_get_state_backends(TRACE_SCSI_DISK_NEW_REQUEST)) {
2583         scsi_disk_new_request_dump(lun, tag, buf);
2584     }
2585 
2586     return req;
2587 }
2588 
2589 #ifdef __linux__
2590 static int get_device_type(SCSIDiskState *s)
2591 {
2592     uint8_t cmd[16];
2593     uint8_t buf[36];
2594     int ret;
2595 
2596     memset(cmd, 0, sizeof(cmd));
2597     memset(buf, 0, sizeof(buf));
2598     cmd[0] = INQUIRY;
2599     cmd[4] = sizeof(buf);
2600 
2601     ret = scsi_SG_IO_FROM_DEV(s->qdev.conf.blk, cmd, sizeof(cmd),
2602                               buf, sizeof(buf), s->qdev.io_timeout);
2603     if (ret < 0) {
2604         return -1;
2605     }
2606     s->qdev.type = buf[0];
2607     if (buf[1] & 0x80) {
2608         s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2609     }
2610     return 0;
2611 }
2612 
2613 static void scsi_block_realize(SCSIDevice *dev, Error **errp)
2614 {
2615     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2616     AioContext *ctx;
2617     int sg_version;
2618     int rc;
2619 
2620     if (!s->qdev.conf.blk) {
2621         error_setg(errp, "drive property not set");
2622         return;
2623     }
2624 
2625     if (s->rotation_rate) {
2626         error_report_once("rotation_rate is specified for scsi-block but is "
2627                           "not implemented. This option is deprecated and will "
2628                           "be removed in a future version");
2629     }
2630 
2631     ctx = blk_get_aio_context(s->qdev.conf.blk);
2632     aio_context_acquire(ctx);
2633 
2634     /* check we are using a driver managing SG_IO (version 3 and after) */
2635     rc = blk_ioctl(s->qdev.conf.blk, SG_GET_VERSION_NUM, &sg_version);
2636     if (rc < 0) {
2637         error_setg_errno(errp, -rc, "cannot get SG_IO version number");
2638         if (rc != -EPERM) {
2639             error_append_hint(errp, "Is this a SCSI device?\n");
2640         }
2641         goto out;
2642     }
2643     if (sg_version < 30000) {
2644         error_setg(errp, "scsi generic interface too old");
2645         goto out;
2646     }
2647 
2648     /* get device type from INQUIRY data */
2649     rc = get_device_type(s);
2650     if (rc < 0) {
2651         error_setg(errp, "INQUIRY failed");
2652         goto out;
2653     }
2654 
2655     /* Make a guess for the block size, we'll fix it when the guest sends.
2656      * READ CAPACITY.  If they don't, they likely would assume these sizes
2657      * anyway. (TODO: check in /sys).
2658      */
2659     if (s->qdev.type == TYPE_ROM || s->qdev.type == TYPE_WORM) {
2660         s->qdev.blocksize = 2048;
2661     } else {
2662         s->qdev.blocksize = 512;
2663     }
2664 
2665     /* Makes the scsi-block device not removable by using HMP and QMP eject
2666      * command.
2667      */
2668     s->features |= (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS);
2669 
2670     scsi_realize(&s->qdev, errp);
2671     scsi_generic_read_device_inquiry(&s->qdev);
2672 
2673 out:
2674     aio_context_release(ctx);
2675 }
2676 
2677 typedef struct SCSIBlockReq {
2678     SCSIDiskReq req;
2679     sg_io_hdr_t io_header;
2680 
2681     /* Selected bytes of the original CDB, copied into our own CDB.  */
2682     uint8_t cmd, cdb1, group_number;
2683 
2684     /* CDB passed to SG_IO.  */
2685     uint8_t cdb[16];
2686     BlockCompletionFunc *cb;
2687     void *cb_opaque;
2688 } SCSIBlockReq;
2689 
2690 static void scsi_block_sgio_complete(void *opaque, int ret)
2691 {
2692     SCSIBlockReq *req = (SCSIBlockReq *)opaque;
2693     SCSIDiskReq *r = &req->req;
2694     SCSIDevice *s = r->req.dev;
2695     sg_io_hdr_t *io_hdr = &req->io_header;
2696 
2697     if (ret == 0) {
2698         if (io_hdr->host_status != SCSI_HOST_OK) {
2699             scsi_req_complete_failed(&r->req, io_hdr->host_status);
2700             scsi_req_unref(&r->req);
2701             return;
2702         }
2703 
2704         if (io_hdr->driver_status & SG_ERR_DRIVER_TIMEOUT) {
2705             ret = BUSY;
2706         } else {
2707             ret = io_hdr->status;
2708         }
2709 
2710         if (ret > 0) {
2711             aio_context_acquire(blk_get_aio_context(s->conf.blk));
2712             if (scsi_handle_rw_error(r, ret, true)) {
2713                 aio_context_release(blk_get_aio_context(s->conf.blk));
2714                 scsi_req_unref(&r->req);
2715                 return;
2716             }
2717             aio_context_release(blk_get_aio_context(s->conf.blk));
2718 
2719             /* Ignore error.  */
2720             ret = 0;
2721         }
2722     }
2723 
2724     req->cb(req->cb_opaque, ret);
2725 }
2726 
2727 static BlockAIOCB *scsi_block_do_sgio(SCSIBlockReq *req,
2728                                       int64_t offset, QEMUIOVector *iov,
2729                                       int direction,
2730                                       BlockCompletionFunc *cb, void *opaque)
2731 {
2732     sg_io_hdr_t *io_header = &req->io_header;
2733     SCSIDiskReq *r = &req->req;
2734     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2735     int nb_logical_blocks;
2736     uint64_t lba;
2737     BlockAIOCB *aiocb;
2738 
2739     /* This is not supported yet.  It can only happen if the guest does
2740      * reads and writes that are not aligned to one logical sectors
2741      * _and_ cover multiple MemoryRegions.
2742      */
2743     assert(offset % s->qdev.blocksize == 0);
2744     assert(iov->size % s->qdev.blocksize == 0);
2745 
2746     io_header->interface_id = 'S';
2747 
2748     /* The data transfer comes from the QEMUIOVector.  */
2749     io_header->dxfer_direction = direction;
2750     io_header->dxfer_len = iov->size;
2751     io_header->dxferp = (void *)iov->iov;
2752     io_header->iovec_count = iov->niov;
2753     assert(io_header->iovec_count == iov->niov); /* no overflow! */
2754 
2755     /* Build a new CDB with the LBA and length patched in, in case
2756      * DMA helpers split the transfer in multiple segments.  Do not
2757      * build a CDB smaller than what the guest wanted, and only build
2758      * a larger one if strictly necessary.
2759      */
2760     io_header->cmdp = req->cdb;
2761     lba = offset / s->qdev.blocksize;
2762     nb_logical_blocks = io_header->dxfer_len / s->qdev.blocksize;
2763 
2764     if ((req->cmd >> 5) == 0 && lba <= 0x1ffff) {
2765         /* 6-byte CDB */
2766         stl_be_p(&req->cdb[0], lba | (req->cmd << 24));
2767         req->cdb[4] = nb_logical_blocks;
2768         req->cdb[5] = 0;
2769         io_header->cmd_len = 6;
2770     } else if ((req->cmd >> 5) <= 1 && lba <= 0xffffffffULL) {
2771         /* 10-byte CDB */
2772         req->cdb[0] = (req->cmd & 0x1f) | 0x20;
2773         req->cdb[1] = req->cdb1;
2774         stl_be_p(&req->cdb[2], lba);
2775         req->cdb[6] = req->group_number;
2776         stw_be_p(&req->cdb[7], nb_logical_blocks);
2777         req->cdb[9] = 0;
2778         io_header->cmd_len = 10;
2779     } else if ((req->cmd >> 5) != 4 && lba <= 0xffffffffULL) {
2780         /* 12-byte CDB */
2781         req->cdb[0] = (req->cmd & 0x1f) | 0xA0;
2782         req->cdb[1] = req->cdb1;
2783         stl_be_p(&req->cdb[2], lba);
2784         stl_be_p(&req->cdb[6], nb_logical_blocks);
2785         req->cdb[10] = req->group_number;
2786         req->cdb[11] = 0;
2787         io_header->cmd_len = 12;
2788     } else {
2789         /* 16-byte CDB */
2790         req->cdb[0] = (req->cmd & 0x1f) | 0x80;
2791         req->cdb[1] = req->cdb1;
2792         stq_be_p(&req->cdb[2], lba);
2793         stl_be_p(&req->cdb[10], nb_logical_blocks);
2794         req->cdb[14] = req->group_number;
2795         req->cdb[15] = 0;
2796         io_header->cmd_len = 16;
2797     }
2798 
2799     /* The rest is as in scsi-generic.c.  */
2800     io_header->mx_sb_len = sizeof(r->req.sense);
2801     io_header->sbp = r->req.sense;
2802     io_header->timeout = s->qdev.io_timeout * 1000;
2803     io_header->usr_ptr = r;
2804     io_header->flags |= SG_FLAG_DIRECT_IO;
2805     req->cb = cb;
2806     req->cb_opaque = opaque;
2807     trace_scsi_disk_aio_sgio_command(r->req.tag, req->cdb[0], lba,
2808                                      nb_logical_blocks, io_header->timeout);
2809     aiocb = blk_aio_ioctl(s->qdev.conf.blk, SG_IO, io_header, scsi_block_sgio_complete, req);
2810     assert(aiocb != NULL);
2811     return aiocb;
2812 }
2813 
2814 static bool scsi_block_no_fua(SCSICommand *cmd)
2815 {
2816     return false;
2817 }
2818 
2819 static BlockAIOCB *scsi_block_dma_readv(int64_t offset,
2820                                         QEMUIOVector *iov,
2821                                         BlockCompletionFunc *cb, void *cb_opaque,
2822                                         void *opaque)
2823 {
2824     SCSIBlockReq *r = opaque;
2825     return scsi_block_do_sgio(r, offset, iov,
2826                               SG_DXFER_FROM_DEV, cb, cb_opaque);
2827 }
2828 
2829 static BlockAIOCB *scsi_block_dma_writev(int64_t offset,
2830                                          QEMUIOVector *iov,
2831                                          BlockCompletionFunc *cb, void *cb_opaque,
2832                                          void *opaque)
2833 {
2834     SCSIBlockReq *r = opaque;
2835     return scsi_block_do_sgio(r, offset, iov,
2836                               SG_DXFER_TO_DEV, cb, cb_opaque);
2837 }
2838 
2839 static bool scsi_block_is_passthrough(SCSIDiskState *s, uint8_t *buf)
2840 {
2841     switch (buf[0]) {
2842     case VERIFY_10:
2843     case VERIFY_12:
2844     case VERIFY_16:
2845         /* Check if BYTCHK == 0x01 (data-out buffer contains data
2846          * for the number of logical blocks specified in the length
2847          * field).  For other modes, do not use scatter/gather operation.
2848          */
2849         if ((buf[1] & 6) == 2) {
2850             return false;
2851         }
2852         break;
2853 
2854     case READ_6:
2855     case READ_10:
2856     case READ_12:
2857     case READ_16:
2858     case WRITE_6:
2859     case WRITE_10:
2860     case WRITE_12:
2861     case WRITE_16:
2862     case WRITE_VERIFY_10:
2863     case WRITE_VERIFY_12:
2864     case WRITE_VERIFY_16:
2865         /* MMC writing cannot be done via DMA helpers, because it sometimes
2866          * involves writing beyond the maximum LBA or to negative LBA (lead-in).
2867          * We might use scsi_block_dma_reqops as long as no writing commands are
2868          * seen, but performance usually isn't paramount on optical media.  So,
2869          * just make scsi-block operate the same as scsi-generic for them.
2870          */
2871         if (s->qdev.type != TYPE_ROM) {
2872             return false;
2873         }
2874         break;
2875 
2876     default:
2877         break;
2878     }
2879 
2880     return true;
2881 }
2882 
2883 
2884 static int32_t scsi_block_dma_command(SCSIRequest *req, uint8_t *buf)
2885 {
2886     SCSIBlockReq *r = (SCSIBlockReq *)req;
2887     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2888 
2889     r->cmd = req->cmd.buf[0];
2890     switch (r->cmd >> 5) {
2891     case 0:
2892         /* 6-byte CDB.  */
2893         r->cdb1 = r->group_number = 0;
2894         break;
2895     case 1:
2896         /* 10-byte CDB.  */
2897         r->cdb1 = req->cmd.buf[1];
2898         r->group_number = req->cmd.buf[6];
2899         break;
2900     case 4:
2901         /* 12-byte CDB.  */
2902         r->cdb1 = req->cmd.buf[1];
2903         r->group_number = req->cmd.buf[10];
2904         break;
2905     case 5:
2906         /* 16-byte CDB.  */
2907         r->cdb1 = req->cmd.buf[1];
2908         r->group_number = req->cmd.buf[14];
2909         break;
2910     default:
2911         abort();
2912     }
2913 
2914     /* Protection information is not supported.  For SCSI versions 2 and
2915      * older (as determined by snooping the guest's INQUIRY commands),
2916      * there is no RD/WR/VRPROTECT, so skip this check in these versions.
2917      */
2918     if (s->qdev.scsi_version > 2 && (req->cmd.buf[1] & 0xe0)) {
2919         scsi_check_condition(&r->req, SENSE_CODE(INVALID_FIELD));
2920         return 0;
2921     }
2922 
2923     return scsi_disk_dma_command(req, buf);
2924 }
2925 
2926 static const SCSIReqOps scsi_block_dma_reqops = {
2927     .size         = sizeof(SCSIBlockReq),
2928     .free_req     = scsi_free_request,
2929     .send_command = scsi_block_dma_command,
2930     .read_data    = scsi_read_data,
2931     .write_data   = scsi_write_data,
2932     .get_buf      = scsi_get_buf,
2933     .load_request = scsi_disk_load_request,
2934     .save_request = scsi_disk_save_request,
2935 };
2936 
2937 static SCSIRequest *scsi_block_new_request(SCSIDevice *d, uint32_t tag,
2938                                            uint32_t lun, uint8_t *buf,
2939                                            void *hba_private)
2940 {
2941     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2942 
2943     if (scsi_block_is_passthrough(s, buf)) {
2944         return scsi_req_alloc(&scsi_generic_req_ops, &s->qdev, tag, lun,
2945                               hba_private);
2946     } else {
2947         return scsi_req_alloc(&scsi_block_dma_reqops, &s->qdev, tag, lun,
2948                               hba_private);
2949     }
2950 }
2951 
2952 static int scsi_block_parse_cdb(SCSIDevice *d, SCSICommand *cmd,
2953                                   uint8_t *buf, void *hba_private)
2954 {
2955     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2956 
2957     if (scsi_block_is_passthrough(s, buf)) {
2958         return scsi_bus_parse_cdb(&s->qdev, cmd, buf, hba_private);
2959     } else {
2960         return scsi_req_parse_cdb(&s->qdev, cmd, buf);
2961     }
2962 }
2963 
2964 static void scsi_block_update_sense(SCSIRequest *req)
2965 {
2966     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
2967     SCSIBlockReq *br = DO_UPCAST(SCSIBlockReq, req, r);
2968     r->req.sense_len = MIN(br->io_header.sb_len_wr, sizeof(r->req.sense));
2969 }
2970 #endif
2971 
2972 static
2973 BlockAIOCB *scsi_dma_readv(int64_t offset, QEMUIOVector *iov,
2974                            BlockCompletionFunc *cb, void *cb_opaque,
2975                            void *opaque)
2976 {
2977     SCSIDiskReq *r = opaque;
2978     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2979     return blk_aio_preadv(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
2980 }
2981 
2982 static
2983 BlockAIOCB *scsi_dma_writev(int64_t offset, QEMUIOVector *iov,
2984                             BlockCompletionFunc *cb, void *cb_opaque,
2985                             void *opaque)
2986 {
2987     SCSIDiskReq *r = opaque;
2988     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2989     return blk_aio_pwritev(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
2990 }
2991 
2992 static void scsi_disk_base_class_initfn(ObjectClass *klass, void *data)
2993 {
2994     DeviceClass *dc = DEVICE_CLASS(klass);
2995     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
2996 
2997     dc->fw_name = "disk";
2998     dc->reset = scsi_disk_reset;
2999     sdc->dma_readv = scsi_dma_readv;
3000     sdc->dma_writev = scsi_dma_writev;
3001     sdc->need_fua_emulation = scsi_is_cmd_fua;
3002 }
3003 
3004 static const TypeInfo scsi_disk_base_info = {
3005     .name          = TYPE_SCSI_DISK_BASE,
3006     .parent        = TYPE_SCSI_DEVICE,
3007     .class_init    = scsi_disk_base_class_initfn,
3008     .instance_size = sizeof(SCSIDiskState),
3009     .class_size    = sizeof(SCSIDiskClass),
3010     .abstract      = true,
3011 };
3012 
3013 #define DEFINE_SCSI_DISK_PROPERTIES()                                   \
3014     DEFINE_PROP_DRIVE_IOTHREAD("drive", SCSIDiskState, qdev.conf.blk),  \
3015     DEFINE_BLOCK_PROPERTIES_BASE(SCSIDiskState, qdev.conf),             \
3016     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),            \
3017     DEFINE_PROP_STRING("ver", SCSIDiskState, version),                  \
3018     DEFINE_PROP_STRING("serial", SCSIDiskState, serial),                \
3019     DEFINE_PROP_STRING("vendor", SCSIDiskState, vendor),                \
3020     DEFINE_PROP_STRING("product", SCSIDiskState, product),              \
3021     DEFINE_PROP_STRING("device_id", SCSIDiskState, device_id)
3022 
3023 
3024 static Property scsi_hd_properties[] = {
3025     DEFINE_SCSI_DISK_PROPERTIES(),
3026     DEFINE_PROP_BIT("removable", SCSIDiskState, features,
3027                     SCSI_DISK_F_REMOVABLE, false),
3028     DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
3029                     SCSI_DISK_F_DPOFUA, false),
3030     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3031     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3032     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3033     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3034                        DEFAULT_MAX_UNMAP_SIZE),
3035     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3036                        DEFAULT_MAX_IO_SIZE),
3037     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3038     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3039                       5),
3040     DEFINE_BLOCK_CHS_PROPERTIES(SCSIDiskState, qdev.conf),
3041     DEFINE_PROP_END_OF_LIST(),
3042 };
3043 
3044 static const VMStateDescription vmstate_scsi_disk_state = {
3045     .name = "scsi-disk",
3046     .version_id = 1,
3047     .minimum_version_id = 1,
3048     .fields = (VMStateField[]) {
3049         VMSTATE_SCSI_DEVICE(qdev, SCSIDiskState),
3050         VMSTATE_BOOL(media_changed, SCSIDiskState),
3051         VMSTATE_BOOL(media_event, SCSIDiskState),
3052         VMSTATE_BOOL(eject_request, SCSIDiskState),
3053         VMSTATE_BOOL(tray_open, SCSIDiskState),
3054         VMSTATE_BOOL(tray_locked, SCSIDiskState),
3055         VMSTATE_END_OF_LIST()
3056     }
3057 };
3058 
3059 static void scsi_hd_class_initfn(ObjectClass *klass, void *data)
3060 {
3061     DeviceClass *dc = DEVICE_CLASS(klass);
3062     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3063 
3064     sc->realize      = scsi_hd_realize;
3065     sc->unrealize    = scsi_unrealize;
3066     sc->alloc_req    = scsi_new_request;
3067     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3068     dc->desc = "virtual SCSI disk";
3069     device_class_set_props(dc, scsi_hd_properties);
3070     dc->vmsd  = &vmstate_scsi_disk_state;
3071 }
3072 
3073 static const TypeInfo scsi_hd_info = {
3074     .name          = "scsi-hd",
3075     .parent        = TYPE_SCSI_DISK_BASE,
3076     .class_init    = scsi_hd_class_initfn,
3077 };
3078 
3079 static Property scsi_cd_properties[] = {
3080     DEFINE_SCSI_DISK_PROPERTIES(),
3081     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3082     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3083     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3084     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3085                        DEFAULT_MAX_IO_SIZE),
3086     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3087                       5),
3088     DEFINE_PROP_END_OF_LIST(),
3089 };
3090 
3091 static void scsi_cd_class_initfn(ObjectClass *klass, void *data)
3092 {
3093     DeviceClass *dc = DEVICE_CLASS(klass);
3094     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3095 
3096     sc->realize      = scsi_cd_realize;
3097     sc->alloc_req    = scsi_new_request;
3098     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3099     dc->desc = "virtual SCSI CD-ROM";
3100     device_class_set_props(dc, scsi_cd_properties);
3101     dc->vmsd  = &vmstate_scsi_disk_state;
3102 }
3103 
3104 static const TypeInfo scsi_cd_info = {
3105     .name          = "scsi-cd",
3106     .parent        = TYPE_SCSI_DISK_BASE,
3107     .class_init    = scsi_cd_class_initfn,
3108 };
3109 
3110 #ifdef __linux__
3111 static Property scsi_block_properties[] = {
3112     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),
3113     DEFINE_PROP_DRIVE("drive", SCSIDiskState, qdev.conf.blk),
3114     DEFINE_PROP_BOOL("share-rw", SCSIDiskState, qdev.conf.share_rw, false),
3115     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3116     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3117                        DEFAULT_MAX_UNMAP_SIZE),
3118     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3119                        DEFAULT_MAX_IO_SIZE),
3120     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3121                       -1),
3122     DEFINE_PROP_UINT32("io_timeout", SCSIDiskState, qdev.io_timeout,
3123                        DEFAULT_IO_TIMEOUT),
3124     DEFINE_PROP_END_OF_LIST(),
3125 };
3126 
3127 static void scsi_block_class_initfn(ObjectClass *klass, void *data)
3128 {
3129     DeviceClass *dc = DEVICE_CLASS(klass);
3130     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3131     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
3132 
3133     sc->realize      = scsi_block_realize;
3134     sc->alloc_req    = scsi_block_new_request;
3135     sc->parse_cdb    = scsi_block_parse_cdb;
3136     sdc->dma_readv   = scsi_block_dma_readv;
3137     sdc->dma_writev  = scsi_block_dma_writev;
3138     sdc->update_sense = scsi_block_update_sense;
3139     sdc->need_fua_emulation = scsi_block_no_fua;
3140     dc->desc = "SCSI block device passthrough";
3141     device_class_set_props(dc, scsi_block_properties);
3142     dc->vmsd  = &vmstate_scsi_disk_state;
3143 }
3144 
3145 static const TypeInfo scsi_block_info = {
3146     .name          = "scsi-block",
3147     .parent        = TYPE_SCSI_DISK_BASE,
3148     .class_init    = scsi_block_class_initfn,
3149 };
3150 #endif
3151 
3152 static void scsi_disk_register_types(void)
3153 {
3154     type_register_static(&scsi_disk_base_info);
3155     type_register_static(&scsi_hd_info);
3156     type_register_static(&scsi_cd_info);
3157 #ifdef __linux__
3158     type_register_static(&scsi_block_info);
3159 #endif
3160 }
3161 
3162 type_init(scsi_disk_register_types)
3163