xref: /openbmc/qemu/hw/scsi/scsi-disk.c (revision afb81fe8)
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     uint32_t quirks;
98     QEMUBH *bh;
99     char *version;
100     char *serial;
101     char *vendor;
102     char *product;
103     char *device_id;
104     bool tray_open;
105     bool tray_locked;
106     /*
107      * 0x0000        - rotation rate not reported
108      * 0x0001        - non-rotating medium (SSD)
109      * 0x0002-0x0400 - reserved
110      * 0x0401-0xffe  - rotations per minute
111      * 0xffff        - reserved
112      */
113     uint16_t rotation_rate;
114 };
115 
116 static void scsi_free_request(SCSIRequest *req)
117 {
118     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
119 
120     qemu_vfree(r->iov.iov_base);
121 }
122 
123 /* Helper function for command completion with sense.  */
124 static void scsi_check_condition(SCSIDiskReq *r, SCSISense sense)
125 {
126     trace_scsi_disk_check_condition(r->req.tag, sense.key, sense.asc,
127                                     sense.ascq);
128     scsi_req_build_sense(&r->req, sense);
129     scsi_req_complete(&r->req, CHECK_CONDITION);
130 }
131 
132 static void scsi_init_iovec(SCSIDiskReq *r, size_t size)
133 {
134     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
135 
136     if (!r->iov.iov_base) {
137         r->buflen = size;
138         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
139     }
140     r->iov.iov_len = MIN(r->sector_count * BDRV_SECTOR_SIZE, r->buflen);
141     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
142 }
143 
144 static void scsi_disk_save_request(QEMUFile *f, SCSIRequest *req)
145 {
146     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
147 
148     qemu_put_be64s(f, &r->sector);
149     qemu_put_be32s(f, &r->sector_count);
150     qemu_put_be32s(f, &r->buflen);
151     if (r->buflen) {
152         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
153             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
154         } else if (!req->retry) {
155             uint32_t len = r->iov.iov_len;
156             qemu_put_be32s(f, &len);
157             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
158         }
159     }
160 }
161 
162 static void scsi_disk_load_request(QEMUFile *f, SCSIRequest *req)
163 {
164     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
165 
166     qemu_get_be64s(f, &r->sector);
167     qemu_get_be32s(f, &r->sector_count);
168     qemu_get_be32s(f, &r->buflen);
169     if (r->buflen) {
170         scsi_init_iovec(r, r->buflen);
171         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
172             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
173         } else if (!r->req.retry) {
174             uint32_t len;
175             qemu_get_be32s(f, &len);
176             r->iov.iov_len = len;
177             assert(r->iov.iov_len <= r->buflen);
178             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
179         }
180     }
181 
182     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
183 }
184 
185 /*
186  * scsi_handle_rw_error has two return values.  False means that the error
187  * must be ignored, true means that the error has been processed and the
188  * caller should not do anything else for this request.  Note that
189  * scsi_handle_rw_error always manages its reference counts, independent
190  * of the return value.
191  */
192 static bool scsi_handle_rw_error(SCSIDiskReq *r, int ret, bool acct_failed)
193 {
194     bool is_read = (r->req.cmd.mode == SCSI_XFER_FROM_DEV);
195     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
196     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
197     SCSISense sense = SENSE_CODE(NO_SENSE);
198     int error = 0;
199     bool req_has_sense = false;
200     BlockErrorAction action;
201     int status;
202 
203     if (ret < 0) {
204         status = scsi_sense_from_errno(-ret, &sense);
205         error = -ret;
206     } else {
207         /* A passthrough command has completed with nonzero status.  */
208         status = ret;
209         if (status == CHECK_CONDITION) {
210             req_has_sense = true;
211             error = scsi_sense_buf_to_errno(r->req.sense, sizeof(r->req.sense));
212         } else {
213             error = EINVAL;
214         }
215     }
216 
217     /*
218      * Check whether the error has to be handled by the guest or should
219      * rather follow the rerror=/werror= settings.  Guest-handled errors
220      * are usually retried immediately, so do not post them to QMP and
221      * do not account them as failed I/O.
222      */
223     if (req_has_sense &&
224         scsi_sense_buf_is_guest_recoverable(r->req.sense, sizeof(r->req.sense))) {
225         action = BLOCK_ERROR_ACTION_REPORT;
226         acct_failed = false;
227     } else {
228         action = blk_get_error_action(s->qdev.conf.blk, is_read, error);
229         blk_error_action(s->qdev.conf.blk, action, is_read, error);
230     }
231 
232     switch (action) {
233     case BLOCK_ERROR_ACTION_REPORT:
234         if (acct_failed) {
235             block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
236         }
237         if (req_has_sense) {
238             sdc->update_sense(&r->req);
239         } else if (status == CHECK_CONDITION) {
240             scsi_req_build_sense(&r->req, sense);
241         }
242         scsi_req_complete(&r->req, status);
243         return true;
244 
245     case BLOCK_ERROR_ACTION_IGNORE:
246         return false;
247 
248     case BLOCK_ERROR_ACTION_STOP:
249         scsi_req_retry(&r->req);
250         return true;
251 
252     default:
253         g_assert_not_reached();
254     }
255 }
256 
257 static bool scsi_disk_req_check_error(SCSIDiskReq *r, int ret, bool acct_failed)
258 {
259     if (r->req.io_canceled) {
260         scsi_req_cancel_complete(&r->req);
261         return true;
262     }
263 
264     if (ret < 0) {
265         return scsi_handle_rw_error(r, ret, acct_failed);
266     }
267 
268     return false;
269 }
270 
271 static void scsi_aio_complete(void *opaque, int ret)
272 {
273     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
274     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
275 
276     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
277 
278     assert(r->req.aiocb != NULL);
279     r->req.aiocb = NULL;
280 
281     if (scsi_disk_req_check_error(r, ret, true)) {
282         goto done;
283     }
284 
285     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
286     scsi_req_complete(&r->req, GOOD);
287 
288 done:
289     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
290     scsi_req_unref(&r->req);
291 }
292 
293 static bool scsi_is_cmd_fua(SCSICommand *cmd)
294 {
295     switch (cmd->buf[0]) {
296     case READ_10:
297     case READ_12:
298     case READ_16:
299     case WRITE_10:
300     case WRITE_12:
301     case WRITE_16:
302         return (cmd->buf[1] & 8) != 0;
303 
304     case VERIFY_10:
305     case VERIFY_12:
306     case VERIFY_16:
307     case WRITE_VERIFY_10:
308     case WRITE_VERIFY_12:
309     case WRITE_VERIFY_16:
310         return true;
311 
312     case READ_6:
313     case WRITE_6:
314     default:
315         return false;
316     }
317 }
318 
319 static void scsi_write_do_fua(SCSIDiskReq *r)
320 {
321     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
322 
323     assert(r->req.aiocb == NULL);
324     assert(!r->req.io_canceled);
325 
326     if (r->need_fua_emulation) {
327         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
328                          BLOCK_ACCT_FLUSH);
329         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
330         return;
331     }
332 
333     scsi_req_complete(&r->req, GOOD);
334     scsi_req_unref(&r->req);
335 }
336 
337 static void scsi_dma_complete_noio(SCSIDiskReq *r, int ret)
338 {
339     assert(r->req.aiocb == NULL);
340     if (scsi_disk_req_check_error(r, ret, false)) {
341         goto done;
342     }
343 
344     r->sector += r->sector_count;
345     r->sector_count = 0;
346     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
347         scsi_write_do_fua(r);
348         return;
349     } else {
350         scsi_req_complete(&r->req, GOOD);
351     }
352 
353 done:
354     scsi_req_unref(&r->req);
355 }
356 
357 /* Called with AioContext lock held */
358 static void scsi_dma_complete(void *opaque, int ret)
359 {
360     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
361     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
362 
363     assert(r->req.aiocb != NULL);
364     r->req.aiocb = NULL;
365 
366     if (ret < 0) {
367         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
368     } else {
369         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
370     }
371     scsi_dma_complete_noio(r, ret);
372 }
373 
374 static void scsi_read_complete_noio(SCSIDiskReq *r, int ret)
375 {
376     uint32_t n;
377 
378     assert(r->req.aiocb == NULL);
379     if (scsi_disk_req_check_error(r, ret, false)) {
380         goto done;
381     }
382 
383     n = r->qiov.size / BDRV_SECTOR_SIZE;
384     r->sector += n;
385     r->sector_count -= n;
386     scsi_req_data(&r->req, r->qiov.size);
387 
388 done:
389     scsi_req_unref(&r->req);
390 }
391 
392 static void scsi_read_complete(void *opaque, int ret)
393 {
394     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
395     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
396 
397     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
398 
399     assert(r->req.aiocb != NULL);
400     r->req.aiocb = NULL;
401 
402     if (ret < 0) {
403         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
404     } else {
405         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
406         trace_scsi_disk_read_complete(r->req.tag, r->qiov.size);
407     }
408     scsi_read_complete_noio(r, ret);
409     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
410 }
411 
412 /* Actually issue a read to the block device.  */
413 static void scsi_do_read(SCSIDiskReq *r, int ret)
414 {
415     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
416     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
417 
418     assert (r->req.aiocb == NULL);
419     if (scsi_disk_req_check_error(r, ret, false)) {
420         goto done;
421     }
422 
423     /* The request is used as the AIO opaque value, so add a ref.  */
424     scsi_req_ref(&r->req);
425 
426     if (r->req.sg) {
427         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_READ);
428         r->req.residual -= r->req.sg->size;
429         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
430                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
431                                   BDRV_SECTOR_SIZE,
432                                   sdc->dma_readv, r, scsi_dma_complete, r,
433                                   DMA_DIRECTION_FROM_DEVICE);
434     } else {
435         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
436         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
437                          r->qiov.size, BLOCK_ACCT_READ);
438         r->req.aiocb = sdc->dma_readv(r->sector << BDRV_SECTOR_BITS, &r->qiov,
439                                       scsi_read_complete, r, r);
440     }
441 
442 done:
443     scsi_req_unref(&r->req);
444 }
445 
446 static void scsi_do_read_cb(void *opaque, int ret)
447 {
448     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
449     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
450 
451     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
452 
453     assert (r->req.aiocb != NULL);
454     r->req.aiocb = NULL;
455 
456     if (ret < 0) {
457         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
458     } else {
459         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
460     }
461     scsi_do_read(opaque, ret);
462     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
463 }
464 
465 /* Read more data from scsi device into buffer.  */
466 static void scsi_read_data(SCSIRequest *req)
467 {
468     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
469     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
470     bool first;
471 
472     trace_scsi_disk_read_data_count(r->sector_count);
473     if (r->sector_count == 0) {
474         /* This also clears the sense buffer for REQUEST SENSE.  */
475         scsi_req_complete(&r->req, GOOD);
476         return;
477     }
478 
479     /* No data transfer may already be in progress */
480     assert(r->req.aiocb == NULL);
481 
482     /* The request is used as the AIO opaque value, so add a ref.  */
483     scsi_req_ref(&r->req);
484     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
485         trace_scsi_disk_read_data_invalid();
486         scsi_read_complete_noio(r, -EINVAL);
487         return;
488     }
489 
490     if (!blk_is_available(req->dev->conf.blk)) {
491         scsi_read_complete_noio(r, -ENOMEDIUM);
492         return;
493     }
494 
495     first = !r->started;
496     r->started = true;
497     if (first && r->need_fua_emulation) {
498         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
499                          BLOCK_ACCT_FLUSH);
500         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_do_read_cb, r);
501     } else {
502         scsi_do_read(r, 0);
503     }
504 }
505 
506 static void scsi_write_complete_noio(SCSIDiskReq *r, int ret)
507 {
508     uint32_t n;
509 
510     assert (r->req.aiocb == NULL);
511     if (scsi_disk_req_check_error(r, ret, false)) {
512         goto done;
513     }
514 
515     n = r->qiov.size / BDRV_SECTOR_SIZE;
516     r->sector += n;
517     r->sector_count -= n;
518     if (r->sector_count == 0) {
519         scsi_write_do_fua(r);
520         return;
521     } else {
522         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
523         trace_scsi_disk_write_complete_noio(r->req.tag, r->qiov.size);
524         scsi_req_data(&r->req, r->qiov.size);
525     }
526 
527 done:
528     scsi_req_unref(&r->req);
529 }
530 
531 static void scsi_write_complete(void * opaque, int ret)
532 {
533     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
534     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
535 
536     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
537 
538     assert (r->req.aiocb != NULL);
539     r->req.aiocb = NULL;
540 
541     if (ret < 0) {
542         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
543     } else {
544         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
545     }
546     scsi_write_complete_noio(r, ret);
547     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
548 }
549 
550 static void scsi_write_data(SCSIRequest *req)
551 {
552     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
553     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
554     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
555 
556     /* No data transfer may already be in progress */
557     assert(r->req.aiocb == NULL);
558 
559     /* The request is used as the AIO opaque value, so add a ref.  */
560     scsi_req_ref(&r->req);
561     if (r->req.cmd.mode != SCSI_XFER_TO_DEV) {
562         trace_scsi_disk_write_data_invalid();
563         scsi_write_complete_noio(r, -EINVAL);
564         return;
565     }
566 
567     if (!r->req.sg && !r->qiov.size) {
568         /* Called for the first time.  Ask the driver to send us more data.  */
569         r->started = true;
570         scsi_write_complete_noio(r, 0);
571         return;
572     }
573     if (!blk_is_available(req->dev->conf.blk)) {
574         scsi_write_complete_noio(r, -ENOMEDIUM);
575         return;
576     }
577 
578     if (r->req.cmd.buf[0] == VERIFY_10 || r->req.cmd.buf[0] == VERIFY_12 ||
579         r->req.cmd.buf[0] == VERIFY_16) {
580         if (r->req.sg) {
581             scsi_dma_complete_noio(r, 0);
582         } else {
583             scsi_write_complete_noio(r, 0);
584         }
585         return;
586     }
587 
588     if (r->req.sg) {
589         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_WRITE);
590         r->req.residual -= r->req.sg->size;
591         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
592                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
593                                   BDRV_SECTOR_SIZE,
594                                   sdc->dma_writev, r, scsi_dma_complete, r,
595                                   DMA_DIRECTION_TO_DEVICE);
596     } else {
597         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
598                          r->qiov.size, BLOCK_ACCT_WRITE);
599         r->req.aiocb = sdc->dma_writev(r->sector << BDRV_SECTOR_BITS, &r->qiov,
600                                        scsi_write_complete, r, r);
601     }
602 }
603 
604 /* Return a pointer to the data buffer.  */
605 static uint8_t *scsi_get_buf(SCSIRequest *req)
606 {
607     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
608 
609     return (uint8_t *)r->iov.iov_base;
610 }
611 
612 static int scsi_disk_emulate_vpd_page(SCSIRequest *req, uint8_t *outbuf)
613 {
614     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
615     uint8_t page_code = req->cmd.buf[2];
616     int start, buflen = 0;
617 
618     outbuf[buflen++] = s->qdev.type & 0x1f;
619     outbuf[buflen++] = page_code;
620     outbuf[buflen++] = 0x00;
621     outbuf[buflen++] = 0x00;
622     start = buflen;
623 
624     switch (page_code) {
625     case 0x00: /* Supported page codes, mandatory */
626     {
627         trace_scsi_disk_emulate_vpd_page_00(req->cmd.xfer);
628         outbuf[buflen++] = 0x00; /* list of supported pages (this page) */
629         if (s->serial) {
630             outbuf[buflen++] = 0x80; /* unit serial number */
631         }
632         outbuf[buflen++] = 0x83; /* device identification */
633         if (s->qdev.type == TYPE_DISK) {
634             outbuf[buflen++] = 0xb0; /* block limits */
635             outbuf[buflen++] = 0xb1; /* block device characteristics */
636             outbuf[buflen++] = 0xb2; /* thin provisioning */
637         }
638         break;
639     }
640     case 0x80: /* Device serial number, optional */
641     {
642         int l;
643 
644         if (!s->serial) {
645             trace_scsi_disk_emulate_vpd_page_80_not_supported();
646             return -1;
647         }
648 
649         l = strlen(s->serial);
650         if (l > 36) {
651             l = 36;
652         }
653 
654         trace_scsi_disk_emulate_vpd_page_80(req->cmd.xfer);
655         memcpy(outbuf + buflen, s->serial, l);
656         buflen += l;
657         break;
658     }
659 
660     case 0x83: /* Device identification page, mandatory */
661     {
662         int id_len = s->device_id ? MIN(strlen(s->device_id), 255 - 8) : 0;
663 
664         trace_scsi_disk_emulate_vpd_page_83(req->cmd.xfer);
665 
666         if (id_len) {
667             outbuf[buflen++] = 0x2; /* ASCII */
668             outbuf[buflen++] = 0;   /* not officially assigned */
669             outbuf[buflen++] = 0;   /* reserved */
670             outbuf[buflen++] = id_len; /* length of data following */
671             memcpy(outbuf + buflen, s->device_id, id_len);
672             buflen += id_len;
673         }
674 
675         if (s->qdev.wwn) {
676             outbuf[buflen++] = 0x1; /* Binary */
677             outbuf[buflen++] = 0x3; /* NAA */
678             outbuf[buflen++] = 0;   /* reserved */
679             outbuf[buflen++] = 8;
680             stq_be_p(&outbuf[buflen], s->qdev.wwn);
681             buflen += 8;
682         }
683 
684         if (s->qdev.port_wwn) {
685             outbuf[buflen++] = 0x61; /* SAS / Binary */
686             outbuf[buflen++] = 0x93; /* PIV / Target port / NAA */
687             outbuf[buflen++] = 0;    /* reserved */
688             outbuf[buflen++] = 8;
689             stq_be_p(&outbuf[buflen], s->qdev.port_wwn);
690             buflen += 8;
691         }
692 
693         if (s->port_index) {
694             outbuf[buflen++] = 0x61; /* SAS / Binary */
695 
696             /* PIV/Target port/relative target port */
697             outbuf[buflen++] = 0x94;
698 
699             outbuf[buflen++] = 0;    /* reserved */
700             outbuf[buflen++] = 4;
701             stw_be_p(&outbuf[buflen + 2], s->port_index);
702             buflen += 4;
703         }
704         break;
705     }
706     case 0xb0: /* block limits */
707     {
708         SCSIBlockLimits bl = {};
709 
710         if (s->qdev.type == TYPE_ROM) {
711             trace_scsi_disk_emulate_vpd_page_b0_not_supported();
712             return -1;
713         }
714         bl.wsnz = 1;
715         bl.unmap_sectors =
716             s->qdev.conf.discard_granularity / s->qdev.blocksize;
717         bl.min_io_size =
718             s->qdev.conf.min_io_size / s->qdev.blocksize;
719         bl.opt_io_size =
720             s->qdev.conf.opt_io_size / s->qdev.blocksize;
721         bl.max_unmap_sectors =
722             s->max_unmap_size / s->qdev.blocksize;
723         bl.max_io_sectors =
724             s->max_io_size / s->qdev.blocksize;
725         /* 255 descriptors fit in 4 KiB with an 8-byte header */
726         bl.max_unmap_descr = 255;
727 
728         if (s->qdev.type == TYPE_DISK) {
729             int max_transfer_blk = blk_get_max_transfer(s->qdev.conf.blk);
730             int max_io_sectors_blk =
731                 max_transfer_blk / s->qdev.blocksize;
732 
733             bl.max_io_sectors =
734                 MIN_NON_ZERO(max_io_sectors_blk, bl.max_io_sectors);
735         }
736         buflen += scsi_emulate_block_limits(outbuf + buflen, &bl);
737         break;
738     }
739     case 0xb1: /* block device characteristics */
740     {
741         buflen = 0x40;
742         outbuf[4] = (s->rotation_rate >> 8) & 0xff;
743         outbuf[5] = s->rotation_rate & 0xff;
744         outbuf[6] = 0; /* PRODUCT TYPE */
745         outbuf[7] = 0; /* WABEREQ | WACEREQ | NOMINAL FORM FACTOR */
746         outbuf[8] = 0; /* VBULS */
747         break;
748     }
749     case 0xb2: /* thin provisioning */
750     {
751         buflen = 8;
752         outbuf[4] = 0;
753         outbuf[5] = 0xe0; /* unmap & write_same 10/16 all supported */
754         outbuf[6] = s->qdev.conf.discard_granularity ? 2 : 1;
755         outbuf[7] = 0;
756         break;
757     }
758     default:
759         return -1;
760     }
761     /* done with EVPD */
762     assert(buflen - start <= 255);
763     outbuf[start - 1] = buflen - start;
764     return buflen;
765 }
766 
767 static int scsi_disk_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf)
768 {
769     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
770     int buflen = 0;
771 
772     if (req->cmd.buf[1] & 0x1) {
773         /* Vital product data */
774         return scsi_disk_emulate_vpd_page(req, outbuf);
775     }
776 
777     /* Standard INQUIRY data */
778     if (req->cmd.buf[2] != 0) {
779         return -1;
780     }
781 
782     /* PAGE CODE == 0 */
783     buflen = req->cmd.xfer;
784     if (buflen > SCSI_MAX_INQUIRY_LEN) {
785         buflen = SCSI_MAX_INQUIRY_LEN;
786     }
787 
788     outbuf[0] = s->qdev.type & 0x1f;
789     outbuf[1] = (s->features & (1 << SCSI_DISK_F_REMOVABLE)) ? 0x80 : 0;
790 
791     strpadcpy((char *) &outbuf[16], 16, s->product, ' ');
792     strpadcpy((char *) &outbuf[8], 8, s->vendor, ' ');
793 
794     memset(&outbuf[32], 0, 4);
795     memcpy(&outbuf[32], s->version, MIN(4, strlen(s->version)));
796     /*
797      * We claim conformance to SPC-3, which is required for guests
798      * to ask for modern features like READ CAPACITY(16) or the
799      * block characteristics VPD page by default.  Not all of SPC-3
800      * is actually implemented, but we're good enough.
801      */
802     outbuf[2] = s->qdev.default_scsi_version;
803     outbuf[3] = 2 | 0x10; /* Format 2, HiSup */
804 
805     if (buflen > 36) {
806         outbuf[4] = buflen - 5; /* Additional Length = (Len - 1) - 4 */
807     } else {
808         /* If the allocation length of CDB is too small,
809                the additional length is not adjusted */
810         outbuf[4] = 36 - 5;
811     }
812 
813     /* Sync data transfer and TCQ.  */
814     outbuf[7] = 0x10 | (req->bus->info->tcq ? 0x02 : 0);
815     return buflen;
816 }
817 
818 static inline bool media_is_dvd(SCSIDiskState *s)
819 {
820     uint64_t nb_sectors;
821     if (s->qdev.type != TYPE_ROM) {
822         return false;
823     }
824     if (!blk_is_available(s->qdev.conf.blk)) {
825         return false;
826     }
827     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
828     return nb_sectors > CD_MAX_SECTORS;
829 }
830 
831 static inline bool media_is_cd(SCSIDiskState *s)
832 {
833     uint64_t nb_sectors;
834     if (s->qdev.type != TYPE_ROM) {
835         return false;
836     }
837     if (!blk_is_available(s->qdev.conf.blk)) {
838         return false;
839     }
840     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
841     return nb_sectors <= CD_MAX_SECTORS;
842 }
843 
844 static int scsi_read_disc_information(SCSIDiskState *s, SCSIDiskReq *r,
845                                       uint8_t *outbuf)
846 {
847     uint8_t type = r->req.cmd.buf[1] & 7;
848 
849     if (s->qdev.type != TYPE_ROM) {
850         return -1;
851     }
852 
853     /* Types 1/2 are only defined for Blu-Ray.  */
854     if (type != 0) {
855         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
856         return -1;
857     }
858 
859     memset(outbuf, 0, 34);
860     outbuf[1] = 32;
861     outbuf[2] = 0xe; /* last session complete, disc finalized */
862     outbuf[3] = 1;   /* first track on disc */
863     outbuf[4] = 1;   /* # of sessions */
864     outbuf[5] = 1;   /* first track of last session */
865     outbuf[6] = 1;   /* last track of last session */
866     outbuf[7] = 0x20; /* unrestricted use */
867     outbuf[8] = 0x00; /* CD-ROM or DVD-ROM */
868     /* 9-10-11: most significant byte corresponding bytes 4-5-6 */
869     /* 12-23: not meaningful for CD-ROM or DVD-ROM */
870     /* 24-31: disc bar code */
871     /* 32: disc application code */
872     /* 33: number of OPC tables */
873 
874     return 34;
875 }
876 
877 static int scsi_read_dvd_structure(SCSIDiskState *s, SCSIDiskReq *r,
878                                    uint8_t *outbuf)
879 {
880     static const int rds_caps_size[5] = {
881         [0] = 2048 + 4,
882         [1] = 4 + 4,
883         [3] = 188 + 4,
884         [4] = 2048 + 4,
885     };
886 
887     uint8_t media = r->req.cmd.buf[1];
888     uint8_t layer = r->req.cmd.buf[6];
889     uint8_t format = r->req.cmd.buf[7];
890     int size = -1;
891 
892     if (s->qdev.type != TYPE_ROM) {
893         return -1;
894     }
895     if (media != 0) {
896         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
897         return -1;
898     }
899 
900     if (format != 0xff) {
901         if (!blk_is_available(s->qdev.conf.blk)) {
902             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
903             return -1;
904         }
905         if (media_is_cd(s)) {
906             scsi_check_condition(r, SENSE_CODE(INCOMPATIBLE_FORMAT));
907             return -1;
908         }
909         if (format >= ARRAY_SIZE(rds_caps_size)) {
910             return -1;
911         }
912         size = rds_caps_size[format];
913         memset(outbuf, 0, size);
914     }
915 
916     switch (format) {
917     case 0x00: {
918         /* Physical format information */
919         uint64_t nb_sectors;
920         if (layer != 0) {
921             goto fail;
922         }
923         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
924 
925         outbuf[4] = 1;   /* DVD-ROM, part version 1 */
926         outbuf[5] = 0xf; /* 120mm disc, minimum rate unspecified */
927         outbuf[6] = 1;   /* one layer, read-only (per MMC-2 spec) */
928         outbuf[7] = 0;   /* default densities */
929 
930         stl_be_p(&outbuf[12], (nb_sectors >> 2) - 1); /* end sector */
931         stl_be_p(&outbuf[16], (nb_sectors >> 2) - 1); /* l0 end sector */
932         break;
933     }
934 
935     case 0x01: /* DVD copyright information, all zeros */
936         break;
937 
938     case 0x03: /* BCA information - invalid field for no BCA info */
939         return -1;
940 
941     case 0x04: /* DVD disc manufacturing information, all zeros */
942         break;
943 
944     case 0xff: { /* List capabilities */
945         int i;
946         size = 4;
947         for (i = 0; i < ARRAY_SIZE(rds_caps_size); i++) {
948             if (!rds_caps_size[i]) {
949                 continue;
950             }
951             outbuf[size] = i;
952             outbuf[size + 1] = 0x40; /* Not writable, readable */
953             stw_be_p(&outbuf[size + 2], rds_caps_size[i]);
954             size += 4;
955         }
956         break;
957      }
958 
959     default:
960         return -1;
961     }
962 
963     /* Size of buffer, not including 2 byte size field */
964     stw_be_p(outbuf, size - 2);
965     return size;
966 
967 fail:
968     return -1;
969 }
970 
971 static int scsi_event_status_media(SCSIDiskState *s, uint8_t *outbuf)
972 {
973     uint8_t event_code, media_status;
974 
975     media_status = 0;
976     if (s->tray_open) {
977         media_status = MS_TRAY_OPEN;
978     } else if (blk_is_inserted(s->qdev.conf.blk)) {
979         media_status = MS_MEDIA_PRESENT;
980     }
981 
982     /* Event notification descriptor */
983     event_code = MEC_NO_CHANGE;
984     if (media_status != MS_TRAY_OPEN) {
985         if (s->media_event) {
986             event_code = MEC_NEW_MEDIA;
987             s->media_event = false;
988         } else if (s->eject_request) {
989             event_code = MEC_EJECT_REQUESTED;
990             s->eject_request = false;
991         }
992     }
993 
994     outbuf[0] = event_code;
995     outbuf[1] = media_status;
996 
997     /* These fields are reserved, just clear them. */
998     outbuf[2] = 0;
999     outbuf[3] = 0;
1000     return 4;
1001 }
1002 
1003 static int scsi_get_event_status_notification(SCSIDiskState *s, SCSIDiskReq *r,
1004                                               uint8_t *outbuf)
1005 {
1006     int size;
1007     uint8_t *buf = r->req.cmd.buf;
1008     uint8_t notification_class_request = buf[4];
1009     if (s->qdev.type != TYPE_ROM) {
1010         return -1;
1011     }
1012     if ((buf[1] & 1) == 0) {
1013         /* asynchronous */
1014         return -1;
1015     }
1016 
1017     size = 4;
1018     outbuf[0] = outbuf[1] = 0;
1019     outbuf[3] = 1 << GESN_MEDIA; /* supported events */
1020     if (notification_class_request & (1 << GESN_MEDIA)) {
1021         outbuf[2] = GESN_MEDIA;
1022         size += scsi_event_status_media(s, &outbuf[size]);
1023     } else {
1024         outbuf[2] = 0x80;
1025     }
1026     stw_be_p(outbuf, size - 4);
1027     return size;
1028 }
1029 
1030 static int scsi_get_configuration(SCSIDiskState *s, uint8_t *outbuf)
1031 {
1032     int current;
1033 
1034     if (s->qdev.type != TYPE_ROM) {
1035         return -1;
1036     }
1037 
1038     if (media_is_dvd(s)) {
1039         current = MMC_PROFILE_DVD_ROM;
1040     } else if (media_is_cd(s)) {
1041         current = MMC_PROFILE_CD_ROM;
1042     } else {
1043         current = MMC_PROFILE_NONE;
1044     }
1045 
1046     memset(outbuf, 0, 40);
1047     stl_be_p(&outbuf[0], 36); /* Bytes after the data length field */
1048     stw_be_p(&outbuf[6], current);
1049     /* outbuf[8] - outbuf[19]: Feature 0 - Profile list */
1050     outbuf[10] = 0x03; /* persistent, current */
1051     outbuf[11] = 8; /* two profiles */
1052     stw_be_p(&outbuf[12], MMC_PROFILE_DVD_ROM);
1053     outbuf[14] = (current == MMC_PROFILE_DVD_ROM);
1054     stw_be_p(&outbuf[16], MMC_PROFILE_CD_ROM);
1055     outbuf[18] = (current == MMC_PROFILE_CD_ROM);
1056     /* outbuf[20] - outbuf[31]: Feature 1 - Core feature */
1057     stw_be_p(&outbuf[20], 1);
1058     outbuf[22] = 0x08 | 0x03; /* version 2, persistent, current */
1059     outbuf[23] = 8;
1060     stl_be_p(&outbuf[24], 1); /* SCSI */
1061     outbuf[28] = 1; /* DBE = 1, mandatory */
1062     /* outbuf[32] - outbuf[39]: Feature 3 - Removable media feature */
1063     stw_be_p(&outbuf[32], 3);
1064     outbuf[34] = 0x08 | 0x03; /* version 2, persistent, current */
1065     outbuf[35] = 4;
1066     outbuf[36] = 0x39; /* tray, load=1, eject=1, unlocked at powerup, lock=1 */
1067     /* TODO: Random readable, CD read, DVD read, drive serial number,
1068        power management */
1069     return 40;
1070 }
1071 
1072 static int scsi_emulate_mechanism_status(SCSIDiskState *s, uint8_t *outbuf)
1073 {
1074     if (s->qdev.type != TYPE_ROM) {
1075         return -1;
1076     }
1077     memset(outbuf, 0, 8);
1078     outbuf[5] = 1; /* CD-ROM */
1079     return 8;
1080 }
1081 
1082 static int mode_sense_page(SCSIDiskState *s, int page, uint8_t **p_outbuf,
1083                            int page_control)
1084 {
1085     static const int mode_sense_valid[0x3f] = {
1086         [MODE_PAGE_VENDOR_SPECIFIC]        = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1087         [MODE_PAGE_HD_GEOMETRY]            = (1 << TYPE_DISK),
1088         [MODE_PAGE_FLEXIBLE_DISK_GEOMETRY] = (1 << TYPE_DISK),
1089         [MODE_PAGE_CACHING]                = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1090         [MODE_PAGE_R_W_ERROR]              = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1091         [MODE_PAGE_AUDIO_CTL]              = (1 << TYPE_ROM),
1092         [MODE_PAGE_CAPABILITIES]           = (1 << TYPE_ROM),
1093         [MODE_PAGE_APPLE_VENDOR]           = (1 << TYPE_ROM),
1094     };
1095 
1096     uint8_t *p = *p_outbuf + 2;
1097     int length;
1098 
1099     assert(page < ARRAY_SIZE(mode_sense_valid));
1100     if ((mode_sense_valid[page] & (1 << s->qdev.type)) == 0) {
1101         return -1;
1102     }
1103 
1104     /*
1105      * If Changeable Values are requested, a mask denoting those mode parameters
1106      * that are changeable shall be returned. As we currently don't support
1107      * parameter changes via MODE_SELECT all bits are returned set to zero.
1108      * The buffer was already menset to zero by the caller of this function.
1109      *
1110      * The offsets here are off by two compared to the descriptions in the
1111      * SCSI specs, because those include a 2-byte header.  This is unfortunate,
1112      * but it is done so that offsets are consistent within our implementation
1113      * of MODE SENSE and MODE SELECT.  MODE SELECT has to deal with both
1114      * 2-byte and 4-byte headers.
1115      */
1116     switch (page) {
1117     case MODE_PAGE_HD_GEOMETRY:
1118         length = 0x16;
1119         if (page_control == 1) { /* Changeable Values */
1120             break;
1121         }
1122         /* if a geometry hint is available, use it */
1123         p[0] = (s->qdev.conf.cyls >> 16) & 0xff;
1124         p[1] = (s->qdev.conf.cyls >> 8) & 0xff;
1125         p[2] = s->qdev.conf.cyls & 0xff;
1126         p[3] = s->qdev.conf.heads & 0xff;
1127         /* Write precomp start cylinder, disabled */
1128         p[4] = (s->qdev.conf.cyls >> 16) & 0xff;
1129         p[5] = (s->qdev.conf.cyls >> 8) & 0xff;
1130         p[6] = s->qdev.conf.cyls & 0xff;
1131         /* Reduced current start cylinder, disabled */
1132         p[7] = (s->qdev.conf.cyls >> 16) & 0xff;
1133         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1134         p[9] = s->qdev.conf.cyls & 0xff;
1135         /* Device step rate [ns], 200ns */
1136         p[10] = 0;
1137         p[11] = 200;
1138         /* Landing zone cylinder */
1139         p[12] = 0xff;
1140         p[13] =  0xff;
1141         p[14] = 0xff;
1142         /* Medium rotation rate [rpm], 5400 rpm */
1143         p[18] = (5400 >> 8) & 0xff;
1144         p[19] = 5400 & 0xff;
1145         break;
1146 
1147     case MODE_PAGE_FLEXIBLE_DISK_GEOMETRY:
1148         length = 0x1e;
1149         if (page_control == 1) { /* Changeable Values */
1150             break;
1151         }
1152         /* Transfer rate [kbit/s], 5Mbit/s */
1153         p[0] = 5000 >> 8;
1154         p[1] = 5000 & 0xff;
1155         /* if a geometry hint is available, use it */
1156         p[2] = s->qdev.conf.heads & 0xff;
1157         p[3] = s->qdev.conf.secs & 0xff;
1158         p[4] = s->qdev.blocksize >> 8;
1159         p[6] = (s->qdev.conf.cyls >> 8) & 0xff;
1160         p[7] = s->qdev.conf.cyls & 0xff;
1161         /* Write precomp start cylinder, disabled */
1162         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1163         p[9] = s->qdev.conf.cyls & 0xff;
1164         /* Reduced current start cylinder, disabled */
1165         p[10] = (s->qdev.conf.cyls >> 8) & 0xff;
1166         p[11] = s->qdev.conf.cyls & 0xff;
1167         /* Device step rate [100us], 100us */
1168         p[12] = 0;
1169         p[13] = 1;
1170         /* Device step pulse width [us], 1us */
1171         p[14] = 1;
1172         /* Device head settle delay [100us], 100us */
1173         p[15] = 0;
1174         p[16] = 1;
1175         /* Motor on delay [0.1s], 0.1s */
1176         p[17] = 1;
1177         /* Motor off delay [0.1s], 0.1s */
1178         p[18] = 1;
1179         /* Medium rotation rate [rpm], 5400 rpm */
1180         p[26] = (5400 >> 8) & 0xff;
1181         p[27] = 5400 & 0xff;
1182         break;
1183 
1184     case MODE_PAGE_CACHING:
1185         length = 0x12;
1186         if (page_control == 1 || /* Changeable Values */
1187             blk_enable_write_cache(s->qdev.conf.blk)) {
1188             p[0] = 4; /* WCE */
1189         }
1190         break;
1191 
1192     case MODE_PAGE_R_W_ERROR:
1193         length = 10;
1194         if (page_control == 1) { /* Changeable Values */
1195             if (s->qdev.type == TYPE_ROM) {
1196                 /* Automatic Write Reallocation Enabled */
1197                 p[0] = 0x80;
1198             }
1199             break;
1200         }
1201         p[0] = 0x80; /* Automatic Write Reallocation Enabled */
1202         if (s->qdev.type == TYPE_ROM) {
1203             p[1] = 0x20; /* Read Retry Count */
1204         }
1205         break;
1206 
1207     case MODE_PAGE_AUDIO_CTL:
1208         length = 14;
1209         break;
1210 
1211     case MODE_PAGE_CAPABILITIES:
1212         length = 0x14;
1213         if (page_control == 1) { /* Changeable Values */
1214             break;
1215         }
1216 
1217         p[0] = 0x3b; /* CD-R & CD-RW read */
1218         p[1] = 0; /* Writing not supported */
1219         p[2] = 0x7f; /* Audio, composite, digital out,
1220                         mode 2 form 1&2, multi session */
1221         p[3] = 0xff; /* CD DA, DA accurate, RW supported,
1222                         RW corrected, C2 errors, ISRC,
1223                         UPC, Bar code */
1224         p[4] = 0x2d | (s->tray_locked ? 2 : 0);
1225         /* Locking supported, jumper present, eject, tray */
1226         p[5] = 0; /* no volume & mute control, no
1227                      changer */
1228         p[6] = (50 * 176) >> 8; /* 50x read speed */
1229         p[7] = (50 * 176) & 0xff;
1230         p[8] = 2 >> 8; /* Two volume levels */
1231         p[9] = 2 & 0xff;
1232         p[10] = 2048 >> 8; /* 2M buffer */
1233         p[11] = 2048 & 0xff;
1234         p[12] = (16 * 176) >> 8; /* 16x read speed current */
1235         p[13] = (16 * 176) & 0xff;
1236         p[16] = (16 * 176) >> 8; /* 16x write speed */
1237         p[17] = (16 * 176) & 0xff;
1238         p[18] = (16 * 176) >> 8; /* 16x write speed current */
1239         p[19] = (16 * 176) & 0xff;
1240         break;
1241 
1242      case MODE_PAGE_APPLE_VENDOR:
1243         if (s->quirks & (1 << SCSI_DISK_QUIRK_MODE_PAGE_APPLE_VENDOR)) {
1244             length = 0x1e;
1245             if (page_control == 1) { /* Changeable Values */
1246                 break;
1247             }
1248 
1249             memset(p, 0, length);
1250             strcpy((char *)p + 8, "APPLE COMPUTER, INC   ");
1251             break;
1252         } else {
1253             return -1;
1254         }
1255 
1256     case MODE_PAGE_VENDOR_SPECIFIC:
1257         if (s->qdev.type == TYPE_DISK && (s->quirks &
1258             (1 << SCSI_DISK_QUIRK_MODE_PAGE_VENDOR_SPECIFIC_APPLE))) {
1259             length = 0x2;
1260             if (page_control == 1) { /* Changeable Values */
1261                 p[0] = 0xff;
1262                 p[1] = 0xff;
1263                 break;
1264             }
1265             p[0] = 0;
1266             p[1] = 0;
1267             break;
1268         } else {
1269             return -1;
1270         }
1271 
1272     default:
1273         return -1;
1274     }
1275 
1276     assert(length < 256);
1277     (*p_outbuf)[0] = page;
1278     (*p_outbuf)[1] = length;
1279     *p_outbuf += length + 2;
1280     return length + 2;
1281 }
1282 
1283 static int scsi_disk_emulate_mode_sense(SCSIDiskReq *r, uint8_t *outbuf)
1284 {
1285     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1286     uint64_t nb_sectors;
1287     bool dbd;
1288     int page, buflen, ret, page_control;
1289     uint8_t *p;
1290     uint8_t dev_specific_param;
1291 
1292     dbd = (r->req.cmd.buf[1] & 0x8) != 0;
1293     page = r->req.cmd.buf[2] & 0x3f;
1294     page_control = (r->req.cmd.buf[2] & 0xc0) >> 6;
1295 
1296     trace_scsi_disk_emulate_mode_sense((r->req.cmd.buf[0] == MODE_SENSE) ? 6 :
1297                                        10, page, r->req.cmd.xfer, page_control);
1298     memset(outbuf, 0, r->req.cmd.xfer);
1299     p = outbuf;
1300 
1301     if (s->qdev.type == TYPE_DISK) {
1302         dev_specific_param = s->features & (1 << SCSI_DISK_F_DPOFUA) ? 0x10 : 0;
1303         if (!blk_is_writable(s->qdev.conf.blk)) {
1304             dev_specific_param |= 0x80; /* Readonly.  */
1305         }
1306     } else {
1307         if (s->quirks & (1 << SCSI_DISK_QUIRK_MODE_SENSE_ROM_USE_DBD)) {
1308             /* Use DBD from the request... */
1309             dev_specific_param = 0x00;
1310 
1311             /*
1312              * ... unless we receive a request for MODE_PAGE_APPLE_VENDOR
1313              * which should never return a block descriptor even though DBD is
1314              * not set, otherwise CDROM detection fails in MacOS
1315              */
1316             if (s->quirks & (1 << SCSI_DISK_QUIRK_MODE_PAGE_APPLE_VENDOR) &&
1317                 page == MODE_PAGE_APPLE_VENDOR) {
1318                 dbd = true;
1319             }
1320         } else {
1321             /*
1322              * MMC prescribes that CD/DVD drives have no block descriptors,
1323              * and defines no device-specific parameter.
1324              */
1325             dev_specific_param = 0x00;
1326             dbd = true;
1327         }
1328     }
1329 
1330     if (r->req.cmd.buf[0] == MODE_SENSE) {
1331         p[1] = 0; /* Default media type.  */
1332         p[2] = dev_specific_param;
1333         p[3] = 0; /* Block descriptor length.  */
1334         p += 4;
1335     } else { /* MODE_SENSE_10 */
1336         p[2] = 0; /* Default media type.  */
1337         p[3] = dev_specific_param;
1338         p[6] = p[7] = 0; /* Block descriptor length.  */
1339         p += 8;
1340     }
1341 
1342     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1343     if (!dbd && nb_sectors) {
1344         if (r->req.cmd.buf[0] == MODE_SENSE) {
1345             outbuf[3] = 8; /* Block descriptor length  */
1346         } else { /* MODE_SENSE_10 */
1347             outbuf[7] = 8; /* Block descriptor length  */
1348         }
1349         nb_sectors /= (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1350         if (nb_sectors > 0xffffff) {
1351             nb_sectors = 0;
1352         }
1353         p[0] = 0; /* media density code */
1354         p[1] = (nb_sectors >> 16) & 0xff;
1355         p[2] = (nb_sectors >> 8) & 0xff;
1356         p[3] = nb_sectors & 0xff;
1357         p[4] = 0; /* reserved */
1358         p[5] = 0; /* bytes 5-7 are the sector size in bytes */
1359         p[6] = s->qdev.blocksize >> 8;
1360         p[7] = 0;
1361         p += 8;
1362     }
1363 
1364     if (page_control == 3) {
1365         /* Saved Values */
1366         scsi_check_condition(r, SENSE_CODE(SAVING_PARAMS_NOT_SUPPORTED));
1367         return -1;
1368     }
1369 
1370     if (page == 0x3f) {
1371         for (page = 0; page <= 0x3e; page++) {
1372             mode_sense_page(s, page, &p, page_control);
1373         }
1374     } else {
1375         ret = mode_sense_page(s, page, &p, page_control);
1376         if (ret == -1) {
1377             return -1;
1378         }
1379     }
1380 
1381     buflen = p - outbuf;
1382     /*
1383      * The mode data length field specifies the length in bytes of the
1384      * following data that is available to be transferred. The mode data
1385      * length does not include itself.
1386      */
1387     if (r->req.cmd.buf[0] == MODE_SENSE) {
1388         outbuf[0] = buflen - 1;
1389     } else { /* MODE_SENSE_10 */
1390         outbuf[0] = ((buflen - 2) >> 8) & 0xff;
1391         outbuf[1] = (buflen - 2) & 0xff;
1392     }
1393     return buflen;
1394 }
1395 
1396 static int scsi_disk_emulate_read_toc(SCSIRequest *req, uint8_t *outbuf)
1397 {
1398     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1399     int start_track, format, msf, toclen;
1400     uint64_t nb_sectors;
1401 
1402     msf = req->cmd.buf[1] & 2;
1403     format = req->cmd.buf[2] & 0xf;
1404     start_track = req->cmd.buf[6];
1405     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1406     trace_scsi_disk_emulate_read_toc(start_track, format, msf >> 1);
1407     nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
1408     switch (format) {
1409     case 0:
1410         toclen = cdrom_read_toc(nb_sectors, outbuf, msf, start_track);
1411         break;
1412     case 1:
1413         /* multi session : only a single session defined */
1414         toclen = 12;
1415         memset(outbuf, 0, 12);
1416         outbuf[1] = 0x0a;
1417         outbuf[2] = 0x01;
1418         outbuf[3] = 0x01;
1419         break;
1420     case 2:
1421         toclen = cdrom_read_toc_raw(nb_sectors, outbuf, msf, start_track);
1422         break;
1423     default:
1424         return -1;
1425     }
1426     return toclen;
1427 }
1428 
1429 static int scsi_disk_emulate_start_stop(SCSIDiskReq *r)
1430 {
1431     SCSIRequest *req = &r->req;
1432     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1433     bool start = req->cmd.buf[4] & 1;
1434     bool loej = req->cmd.buf[4] & 2; /* load on start, eject on !start */
1435     int pwrcnd = req->cmd.buf[4] & 0xf0;
1436 
1437     if (pwrcnd) {
1438         /* eject/load only happens for power condition == 0 */
1439         return 0;
1440     }
1441 
1442     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) && loej) {
1443         if (!start && !s->tray_open && s->tray_locked) {
1444             scsi_check_condition(r,
1445                                  blk_is_inserted(s->qdev.conf.blk)
1446                                  ? SENSE_CODE(ILLEGAL_REQ_REMOVAL_PREVENTED)
1447                                  : SENSE_CODE(NOT_READY_REMOVAL_PREVENTED));
1448             return -1;
1449         }
1450 
1451         if (s->tray_open != !start) {
1452             blk_eject(s->qdev.conf.blk, !start);
1453             s->tray_open = !start;
1454         }
1455     }
1456     return 0;
1457 }
1458 
1459 static void scsi_disk_emulate_read_data(SCSIRequest *req)
1460 {
1461     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1462     int buflen = r->iov.iov_len;
1463 
1464     if (buflen) {
1465         trace_scsi_disk_emulate_read_data(buflen);
1466         r->iov.iov_len = 0;
1467         r->started = true;
1468         scsi_req_data(&r->req, buflen);
1469         return;
1470     }
1471 
1472     /* This also clears the sense buffer for REQUEST SENSE.  */
1473     scsi_req_complete(&r->req, GOOD);
1474 }
1475 
1476 static int scsi_disk_check_mode_select(SCSIDiskState *s, int page,
1477                                        uint8_t *inbuf, int inlen)
1478 {
1479     uint8_t mode_current[SCSI_MAX_MODE_LEN];
1480     uint8_t mode_changeable[SCSI_MAX_MODE_LEN];
1481     uint8_t *p;
1482     int len, expected_len, changeable_len, i;
1483 
1484     /* The input buffer does not include the page header, so it is
1485      * off by 2 bytes.
1486      */
1487     expected_len = inlen + 2;
1488     if (expected_len > SCSI_MAX_MODE_LEN) {
1489         return -1;
1490     }
1491 
1492     /* MODE_PAGE_ALLS is only valid for MODE SENSE commands */
1493     if (page == MODE_PAGE_ALLS) {
1494         return -1;
1495     }
1496 
1497     p = mode_current;
1498     memset(mode_current, 0, inlen + 2);
1499     len = mode_sense_page(s, page, &p, 0);
1500     if (len < 0 || len != expected_len) {
1501         return -1;
1502     }
1503 
1504     p = mode_changeable;
1505     memset(mode_changeable, 0, inlen + 2);
1506     changeable_len = mode_sense_page(s, page, &p, 1);
1507     assert(changeable_len == len);
1508 
1509     /* Check that unchangeable bits are the same as what MODE SENSE
1510      * would return.
1511      */
1512     for (i = 2; i < len; i++) {
1513         if (((mode_current[i] ^ inbuf[i - 2]) & ~mode_changeable[i]) != 0) {
1514             return -1;
1515         }
1516     }
1517     return 0;
1518 }
1519 
1520 static void scsi_disk_apply_mode_select(SCSIDiskState *s, int page, uint8_t *p)
1521 {
1522     switch (page) {
1523     case MODE_PAGE_CACHING:
1524         blk_set_enable_write_cache(s->qdev.conf.blk, (p[0] & 4) != 0);
1525         break;
1526 
1527     default:
1528         break;
1529     }
1530 }
1531 
1532 static int mode_select_pages(SCSIDiskReq *r, uint8_t *p, int len, bool change)
1533 {
1534     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1535 
1536     while (len > 0) {
1537         int page, subpage, page_len;
1538 
1539         /* Parse both possible formats for the mode page headers.  */
1540         page = p[0] & 0x3f;
1541         if (p[0] & 0x40) {
1542             if (len < 4) {
1543                 goto invalid_param_len;
1544             }
1545             subpage = p[1];
1546             page_len = lduw_be_p(&p[2]);
1547             p += 4;
1548             len -= 4;
1549         } else {
1550             if (len < 2) {
1551                 goto invalid_param_len;
1552             }
1553             subpage = 0;
1554             page_len = p[1];
1555             p += 2;
1556             len -= 2;
1557         }
1558 
1559         if (subpage) {
1560             goto invalid_param;
1561         }
1562         if (page_len > len) {
1563             if (!(s->quirks & SCSI_DISK_QUIRK_MODE_PAGE_TRUNCATED)) {
1564                 goto invalid_param_len;
1565             }
1566             trace_scsi_disk_mode_select_page_truncated(page, page_len, len);
1567         }
1568 
1569         if (!change) {
1570             if (scsi_disk_check_mode_select(s, page, p, page_len) < 0) {
1571                 goto invalid_param;
1572             }
1573         } else {
1574             scsi_disk_apply_mode_select(s, page, p);
1575         }
1576 
1577         p += page_len;
1578         len -= page_len;
1579     }
1580     return 0;
1581 
1582 invalid_param:
1583     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1584     return -1;
1585 
1586 invalid_param_len:
1587     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1588     return -1;
1589 }
1590 
1591 static void scsi_disk_emulate_mode_select(SCSIDiskReq *r, uint8_t *inbuf)
1592 {
1593     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1594     uint8_t *p = inbuf;
1595     int cmd = r->req.cmd.buf[0];
1596     int len = r->req.cmd.xfer;
1597     int hdr_len = (cmd == MODE_SELECT ? 4 : 8);
1598     int bd_len, bs;
1599     int pass;
1600 
1601     if ((r->req.cmd.buf[1] & 0x11) != 0x10) {
1602         if (!(s->quirks &
1603             (1 << SCSI_DISK_QUIRK_MODE_PAGE_VENDOR_SPECIFIC_APPLE))) {
1604             /* We only support PF=1, SP=0.  */
1605             goto invalid_field;
1606         }
1607     }
1608 
1609     if (len < hdr_len) {
1610         goto invalid_param_len;
1611     }
1612 
1613     bd_len = (cmd == MODE_SELECT ? p[3] : lduw_be_p(&p[6]));
1614     len -= hdr_len;
1615     p += hdr_len;
1616     if (len < bd_len) {
1617         goto invalid_param_len;
1618     }
1619     if (bd_len != 0 && bd_len != 8) {
1620         goto invalid_param;
1621     }
1622 
1623     /* Allow changing the block size */
1624     if (bd_len) {
1625         bs = p[5] << 16 | p[6] << 8 | p[7];
1626 
1627         /*
1628          * Since the existing code only checks/updates bits 8-15 of the block
1629          * size, restrict ourselves to the same requirement for now to ensure
1630          * that a block size set by a block descriptor and then read back by
1631          * a subsequent SCSI command will be the same. Also disallow a block
1632          * size of 256 since we cannot handle anything below BDRV_SECTOR_SIZE.
1633          */
1634         if (bs && !(bs & ~0xfe00) && bs != s->qdev.blocksize) {
1635             s->qdev.blocksize = bs;
1636             trace_scsi_disk_mode_select_set_blocksize(s->qdev.blocksize);
1637         }
1638     }
1639 
1640     len -= bd_len;
1641     p += bd_len;
1642 
1643     /* Ensure no change is made if there is an error!  */
1644     for (pass = 0; pass < 2; pass++) {
1645         if (mode_select_pages(r, p, len, pass == 1) < 0) {
1646             assert(pass == 0);
1647             return;
1648         }
1649     }
1650     if (!blk_enable_write_cache(s->qdev.conf.blk)) {
1651         /* The request is used as the AIO opaque value, so add a ref.  */
1652         scsi_req_ref(&r->req);
1653         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
1654                          BLOCK_ACCT_FLUSH);
1655         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
1656         return;
1657     }
1658 
1659     scsi_req_complete(&r->req, GOOD);
1660     return;
1661 
1662 invalid_param:
1663     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1664     return;
1665 
1666 invalid_param_len:
1667     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1668     return;
1669 
1670 invalid_field:
1671     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1672 }
1673 
1674 /* sector_num and nb_sectors expected to be in qdev blocksize */
1675 static inline bool check_lba_range(SCSIDiskState *s,
1676                                    uint64_t sector_num, uint32_t nb_sectors)
1677 {
1678     /*
1679      * The first line tests that no overflow happens when computing the last
1680      * sector.  The second line tests that the last accessed sector is in
1681      * range.
1682      *
1683      * Careful, the computations should not underflow for nb_sectors == 0,
1684      * and a 0-block read to the first LBA beyond the end of device is
1685      * valid.
1686      */
1687     return (sector_num <= sector_num + nb_sectors &&
1688             sector_num + nb_sectors <= s->qdev.max_lba + 1);
1689 }
1690 
1691 typedef struct UnmapCBData {
1692     SCSIDiskReq *r;
1693     uint8_t *inbuf;
1694     int count;
1695 } UnmapCBData;
1696 
1697 static void scsi_unmap_complete(void *opaque, int ret);
1698 
1699 static void scsi_unmap_complete_noio(UnmapCBData *data, int ret)
1700 {
1701     SCSIDiskReq *r = data->r;
1702     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1703 
1704     assert(r->req.aiocb == NULL);
1705 
1706     if (data->count > 0) {
1707         uint64_t sector_num = ldq_be_p(&data->inbuf[0]);
1708         uint32_t nb_sectors = ldl_be_p(&data->inbuf[8]) & 0xffffffffULL;
1709         r->sector = sector_num * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1710         r->sector_count = nb_sectors * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1711 
1712         if (!check_lba_range(s, sector_num, nb_sectors)) {
1713             block_acct_invalid(blk_get_stats(s->qdev.conf.blk),
1714                                BLOCK_ACCT_UNMAP);
1715             scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1716             goto done;
1717         }
1718 
1719         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1720                          r->sector_count * BDRV_SECTOR_SIZE,
1721                          BLOCK_ACCT_UNMAP);
1722 
1723         r->req.aiocb = blk_aio_pdiscard(s->qdev.conf.blk,
1724                                         r->sector * BDRV_SECTOR_SIZE,
1725                                         r->sector_count * BDRV_SECTOR_SIZE,
1726                                         scsi_unmap_complete, data);
1727         data->count--;
1728         data->inbuf += 16;
1729         return;
1730     }
1731 
1732     scsi_req_complete(&r->req, GOOD);
1733 
1734 done:
1735     scsi_req_unref(&r->req);
1736     g_free(data);
1737 }
1738 
1739 static void scsi_unmap_complete(void *opaque, int ret)
1740 {
1741     UnmapCBData *data = opaque;
1742     SCSIDiskReq *r = data->r;
1743     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1744 
1745     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
1746 
1747     assert(r->req.aiocb != NULL);
1748     r->req.aiocb = NULL;
1749 
1750     if (scsi_disk_req_check_error(r, ret, true)) {
1751         scsi_req_unref(&r->req);
1752         g_free(data);
1753     } else {
1754         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1755         scsi_unmap_complete_noio(data, ret);
1756     }
1757     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1758 }
1759 
1760 static void scsi_disk_emulate_unmap(SCSIDiskReq *r, uint8_t *inbuf)
1761 {
1762     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1763     uint8_t *p = inbuf;
1764     int len = r->req.cmd.xfer;
1765     UnmapCBData *data;
1766 
1767     /* Reject ANCHOR=1.  */
1768     if (r->req.cmd.buf[1] & 0x1) {
1769         goto invalid_field;
1770     }
1771 
1772     if (len < 8) {
1773         goto invalid_param_len;
1774     }
1775     if (len < lduw_be_p(&p[0]) + 2) {
1776         goto invalid_param_len;
1777     }
1778     if (len < lduw_be_p(&p[2]) + 8) {
1779         goto invalid_param_len;
1780     }
1781     if (lduw_be_p(&p[2]) & 15) {
1782         goto invalid_param_len;
1783     }
1784 
1785     if (!blk_is_writable(s->qdev.conf.blk)) {
1786         block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1787         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1788         return;
1789     }
1790 
1791     data = g_new0(UnmapCBData, 1);
1792     data->r = r;
1793     data->inbuf = &p[8];
1794     data->count = lduw_be_p(&p[2]) >> 4;
1795 
1796     /* The matching unref is in scsi_unmap_complete, before data is freed.  */
1797     scsi_req_ref(&r->req);
1798     scsi_unmap_complete_noio(data, 0);
1799     return;
1800 
1801 invalid_param_len:
1802     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1803     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1804     return;
1805 
1806 invalid_field:
1807     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1808     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1809 }
1810 
1811 typedef struct WriteSameCBData {
1812     SCSIDiskReq *r;
1813     int64_t sector;
1814     int nb_sectors;
1815     QEMUIOVector qiov;
1816     struct iovec iov;
1817 } WriteSameCBData;
1818 
1819 static void scsi_write_same_complete(void *opaque, int ret)
1820 {
1821     WriteSameCBData *data = opaque;
1822     SCSIDiskReq *r = data->r;
1823     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1824 
1825     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
1826 
1827     assert(r->req.aiocb != NULL);
1828     r->req.aiocb = NULL;
1829 
1830     if (scsi_disk_req_check_error(r, ret, true)) {
1831         goto done;
1832     }
1833 
1834     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1835 
1836     data->nb_sectors -= data->iov.iov_len / BDRV_SECTOR_SIZE;
1837     data->sector += data->iov.iov_len / BDRV_SECTOR_SIZE;
1838     data->iov.iov_len = MIN(data->nb_sectors * BDRV_SECTOR_SIZE,
1839                             data->iov.iov_len);
1840     if (data->iov.iov_len) {
1841         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1842                          data->iov.iov_len, BLOCK_ACCT_WRITE);
1843         /* Reinitialize qiov, to handle unaligned WRITE SAME request
1844          * where final qiov may need smaller size */
1845         qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1846         r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1847                                        data->sector << BDRV_SECTOR_BITS,
1848                                        &data->qiov, 0,
1849                                        scsi_write_same_complete, data);
1850         aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1851         return;
1852     }
1853 
1854     scsi_req_complete(&r->req, GOOD);
1855 
1856 done:
1857     scsi_req_unref(&r->req);
1858     qemu_vfree(data->iov.iov_base);
1859     g_free(data);
1860     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1861 }
1862 
1863 static void scsi_disk_emulate_write_same(SCSIDiskReq *r, uint8_t *inbuf)
1864 {
1865     SCSIRequest *req = &r->req;
1866     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1867     uint32_t nb_sectors = scsi_data_cdb_xfer(r->req.cmd.buf);
1868     WriteSameCBData *data;
1869     uint8_t *buf;
1870     int i, l;
1871 
1872     /* Fail if PBDATA=1 or LBDATA=1 or ANCHOR=1.  */
1873     if (nb_sectors == 0 || (req->cmd.buf[1] & 0x16)) {
1874         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1875         return;
1876     }
1877 
1878     if (!blk_is_writable(s->qdev.conf.blk)) {
1879         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1880         return;
1881     }
1882     if (!check_lba_range(s, r->req.cmd.lba, nb_sectors)) {
1883         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1884         return;
1885     }
1886 
1887     if ((req->cmd.buf[1] & 0x1) || buffer_is_zero(inbuf, s->qdev.blocksize)) {
1888         int flags = (req->cmd.buf[1] & 0x8) ? BDRV_REQ_MAY_UNMAP : 0;
1889 
1890         /* The request is used as the AIO opaque value, so add a ref.  */
1891         scsi_req_ref(&r->req);
1892         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1893                          nb_sectors * s->qdev.blocksize,
1894                         BLOCK_ACCT_WRITE);
1895         r->req.aiocb = blk_aio_pwrite_zeroes(s->qdev.conf.blk,
1896                                 r->req.cmd.lba * s->qdev.blocksize,
1897                                 nb_sectors * s->qdev.blocksize,
1898                                 flags, scsi_aio_complete, r);
1899         return;
1900     }
1901 
1902     data = g_new0(WriteSameCBData, 1);
1903     data->r = r;
1904     data->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1905     data->nb_sectors = nb_sectors * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1906     data->iov.iov_len = MIN(data->nb_sectors * BDRV_SECTOR_SIZE,
1907                             SCSI_WRITE_SAME_MAX);
1908     data->iov.iov_base = buf = blk_blockalign(s->qdev.conf.blk,
1909                                               data->iov.iov_len);
1910     qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1911 
1912     for (i = 0; i < data->iov.iov_len; i += l) {
1913         l = MIN(s->qdev.blocksize, data->iov.iov_len - i);
1914         memcpy(&buf[i], inbuf, l);
1915     }
1916 
1917     scsi_req_ref(&r->req);
1918     block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1919                      data->iov.iov_len, BLOCK_ACCT_WRITE);
1920     r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1921                                    data->sector << BDRV_SECTOR_BITS,
1922                                    &data->qiov, 0,
1923                                    scsi_write_same_complete, data);
1924 }
1925 
1926 static void scsi_disk_emulate_write_data(SCSIRequest *req)
1927 {
1928     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1929 
1930     if (r->iov.iov_len) {
1931         int buflen = r->iov.iov_len;
1932         trace_scsi_disk_emulate_write_data(buflen);
1933         r->iov.iov_len = 0;
1934         scsi_req_data(&r->req, buflen);
1935         return;
1936     }
1937 
1938     switch (req->cmd.buf[0]) {
1939     case MODE_SELECT:
1940     case MODE_SELECT_10:
1941         /* This also clears the sense buffer for REQUEST SENSE.  */
1942         scsi_disk_emulate_mode_select(r, r->iov.iov_base);
1943         break;
1944 
1945     case UNMAP:
1946         scsi_disk_emulate_unmap(r, r->iov.iov_base);
1947         break;
1948 
1949     case VERIFY_10:
1950     case VERIFY_12:
1951     case VERIFY_16:
1952         if (r->req.status == -1) {
1953             scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1954         }
1955         break;
1956 
1957     case WRITE_SAME_10:
1958     case WRITE_SAME_16:
1959         scsi_disk_emulate_write_same(r, r->iov.iov_base);
1960         break;
1961 
1962     case FORMAT_UNIT:
1963         scsi_req_complete(&r->req, GOOD);
1964         break;
1965 
1966     default:
1967         abort();
1968     }
1969 }
1970 
1971 static int32_t scsi_disk_emulate_command(SCSIRequest *req, uint8_t *buf)
1972 {
1973     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1974     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1975     uint64_t nb_sectors;
1976     uint8_t *outbuf;
1977     int buflen;
1978 
1979     switch (req->cmd.buf[0]) {
1980     case INQUIRY:
1981     case MODE_SENSE:
1982     case MODE_SENSE_10:
1983     case RESERVE:
1984     case RESERVE_10:
1985     case RELEASE:
1986     case RELEASE_10:
1987     case START_STOP:
1988     case ALLOW_MEDIUM_REMOVAL:
1989     case GET_CONFIGURATION:
1990     case GET_EVENT_STATUS_NOTIFICATION:
1991     case MECHANISM_STATUS:
1992     case REQUEST_SENSE:
1993         break;
1994 
1995     default:
1996         if (!blk_is_available(s->qdev.conf.blk)) {
1997             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
1998             return 0;
1999         }
2000         break;
2001     }
2002 
2003     /*
2004      * FIXME: we shouldn't return anything bigger than 4k, but the code
2005      * requires the buffer to be as big as req->cmd.xfer in several
2006      * places.  So, do not allow CDBs with a very large ALLOCATION
2007      * LENGTH.  The real fix would be to modify scsi_read_data and
2008      * dma_buf_read, so that they return data beyond the buflen
2009      * as all zeros.
2010      */
2011     if (req->cmd.xfer > 65536) {
2012         goto illegal_request;
2013     }
2014     r->buflen = MAX(4096, req->cmd.xfer);
2015 
2016     if (!r->iov.iov_base) {
2017         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
2018     }
2019 
2020     outbuf = r->iov.iov_base;
2021     memset(outbuf, 0, r->buflen);
2022     switch (req->cmd.buf[0]) {
2023     case TEST_UNIT_READY:
2024         assert(blk_is_available(s->qdev.conf.blk));
2025         break;
2026     case INQUIRY:
2027         buflen = scsi_disk_emulate_inquiry(req, outbuf);
2028         if (buflen < 0) {
2029             goto illegal_request;
2030         }
2031         break;
2032     case MODE_SENSE:
2033     case MODE_SENSE_10:
2034         buflen = scsi_disk_emulate_mode_sense(r, outbuf);
2035         if (buflen < 0) {
2036             goto illegal_request;
2037         }
2038         break;
2039     case READ_TOC:
2040         buflen = scsi_disk_emulate_read_toc(req, outbuf);
2041         if (buflen < 0) {
2042             goto illegal_request;
2043         }
2044         break;
2045     case RESERVE:
2046         if (req->cmd.buf[1] & 1) {
2047             goto illegal_request;
2048         }
2049         break;
2050     case RESERVE_10:
2051         if (req->cmd.buf[1] & 3) {
2052             goto illegal_request;
2053         }
2054         break;
2055     case RELEASE:
2056         if (req->cmd.buf[1] & 1) {
2057             goto illegal_request;
2058         }
2059         break;
2060     case RELEASE_10:
2061         if (req->cmd.buf[1] & 3) {
2062             goto illegal_request;
2063         }
2064         break;
2065     case START_STOP:
2066         if (scsi_disk_emulate_start_stop(r) < 0) {
2067             return 0;
2068         }
2069         break;
2070     case ALLOW_MEDIUM_REMOVAL:
2071         s->tray_locked = req->cmd.buf[4] & 1;
2072         blk_lock_medium(s->qdev.conf.blk, req->cmd.buf[4] & 1);
2073         break;
2074     case READ_CAPACITY_10:
2075         /* The normal LEN field for this command is zero.  */
2076         memset(outbuf, 0, 8);
2077         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2078         if (!nb_sectors) {
2079             scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
2080             return 0;
2081         }
2082         if ((req->cmd.buf[8] & 1) == 0 && req->cmd.lba) {
2083             goto illegal_request;
2084         }
2085         nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2086         /* Returned value is the address of the last sector.  */
2087         nb_sectors--;
2088         /* Remember the new size for read/write sanity checking. */
2089         s->qdev.max_lba = nb_sectors;
2090         /* Clip to 2TB, instead of returning capacity modulo 2TB. */
2091         if (nb_sectors > UINT32_MAX) {
2092             nb_sectors = UINT32_MAX;
2093         }
2094         outbuf[0] = (nb_sectors >> 24) & 0xff;
2095         outbuf[1] = (nb_sectors >> 16) & 0xff;
2096         outbuf[2] = (nb_sectors >> 8) & 0xff;
2097         outbuf[3] = nb_sectors & 0xff;
2098         outbuf[4] = 0;
2099         outbuf[5] = 0;
2100         outbuf[6] = s->qdev.blocksize >> 8;
2101         outbuf[7] = 0;
2102         break;
2103     case REQUEST_SENSE:
2104         /* Just return "NO SENSE".  */
2105         buflen = scsi_convert_sense(NULL, 0, outbuf, r->buflen,
2106                                     (req->cmd.buf[1] & 1) == 0);
2107         if (buflen < 0) {
2108             goto illegal_request;
2109         }
2110         break;
2111     case MECHANISM_STATUS:
2112         buflen = scsi_emulate_mechanism_status(s, outbuf);
2113         if (buflen < 0) {
2114             goto illegal_request;
2115         }
2116         break;
2117     case GET_CONFIGURATION:
2118         buflen = scsi_get_configuration(s, outbuf);
2119         if (buflen < 0) {
2120             goto illegal_request;
2121         }
2122         break;
2123     case GET_EVENT_STATUS_NOTIFICATION:
2124         buflen = scsi_get_event_status_notification(s, r, outbuf);
2125         if (buflen < 0) {
2126             goto illegal_request;
2127         }
2128         break;
2129     case READ_DISC_INFORMATION:
2130         buflen = scsi_read_disc_information(s, r, outbuf);
2131         if (buflen < 0) {
2132             goto illegal_request;
2133         }
2134         break;
2135     case READ_DVD_STRUCTURE:
2136         buflen = scsi_read_dvd_structure(s, r, outbuf);
2137         if (buflen < 0) {
2138             goto illegal_request;
2139         }
2140         break;
2141     case SERVICE_ACTION_IN_16:
2142         /* Service Action In subcommands. */
2143         if ((req->cmd.buf[1] & 31) == SAI_READ_CAPACITY_16) {
2144             trace_scsi_disk_emulate_command_SAI_16();
2145             memset(outbuf, 0, req->cmd.xfer);
2146             blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2147             if (!nb_sectors) {
2148                 scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
2149                 return 0;
2150             }
2151             if ((req->cmd.buf[14] & 1) == 0 && req->cmd.lba) {
2152                 goto illegal_request;
2153             }
2154             nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2155             /* Returned value is the address of the last sector.  */
2156             nb_sectors--;
2157             /* Remember the new size for read/write sanity checking. */
2158             s->qdev.max_lba = nb_sectors;
2159             outbuf[0] = (nb_sectors >> 56) & 0xff;
2160             outbuf[1] = (nb_sectors >> 48) & 0xff;
2161             outbuf[2] = (nb_sectors >> 40) & 0xff;
2162             outbuf[3] = (nb_sectors >> 32) & 0xff;
2163             outbuf[4] = (nb_sectors >> 24) & 0xff;
2164             outbuf[5] = (nb_sectors >> 16) & 0xff;
2165             outbuf[6] = (nb_sectors >> 8) & 0xff;
2166             outbuf[7] = nb_sectors & 0xff;
2167             outbuf[8] = 0;
2168             outbuf[9] = 0;
2169             outbuf[10] = s->qdev.blocksize >> 8;
2170             outbuf[11] = 0;
2171             outbuf[12] = 0;
2172             outbuf[13] = get_physical_block_exp(&s->qdev.conf);
2173 
2174             /* set TPE bit if the format supports discard */
2175             if (s->qdev.conf.discard_granularity) {
2176                 outbuf[14] = 0x80;
2177             }
2178 
2179             /* Protection, exponent and lowest lba field left blank. */
2180             break;
2181         }
2182         trace_scsi_disk_emulate_command_SAI_unsupported();
2183         goto illegal_request;
2184     case SYNCHRONIZE_CACHE:
2185         /* The request is used as the AIO opaque value, so add a ref.  */
2186         scsi_req_ref(&r->req);
2187         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
2188                          BLOCK_ACCT_FLUSH);
2189         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
2190         return 0;
2191     case SEEK_10:
2192         trace_scsi_disk_emulate_command_SEEK_10(r->req.cmd.lba);
2193         if (r->req.cmd.lba > s->qdev.max_lba) {
2194             goto illegal_lba;
2195         }
2196         break;
2197     case MODE_SELECT:
2198         trace_scsi_disk_emulate_command_MODE_SELECT(r->req.cmd.xfer);
2199         break;
2200     case MODE_SELECT_10:
2201         trace_scsi_disk_emulate_command_MODE_SELECT_10(r->req.cmd.xfer);
2202         break;
2203     case UNMAP:
2204         trace_scsi_disk_emulate_command_UNMAP(r->req.cmd.xfer);
2205         break;
2206     case VERIFY_10:
2207     case VERIFY_12:
2208     case VERIFY_16:
2209         trace_scsi_disk_emulate_command_VERIFY((req->cmd.buf[1] >> 1) & 3);
2210         if (req->cmd.buf[1] & 6) {
2211             goto illegal_request;
2212         }
2213         break;
2214     case WRITE_SAME_10:
2215     case WRITE_SAME_16:
2216         trace_scsi_disk_emulate_command_WRITE_SAME(
2217                 req->cmd.buf[0] == WRITE_SAME_10 ? 10 : 16, r->req.cmd.xfer);
2218         break;
2219     case FORMAT_UNIT:
2220         trace_scsi_disk_emulate_command_FORMAT_UNIT(r->req.cmd.xfer);
2221         break;
2222     default:
2223         trace_scsi_disk_emulate_command_UNKNOWN(buf[0],
2224                                                 scsi_command_name(buf[0]));
2225         scsi_check_condition(r, SENSE_CODE(INVALID_OPCODE));
2226         return 0;
2227     }
2228     assert(!r->req.aiocb);
2229     r->iov.iov_len = MIN(r->buflen, req->cmd.xfer);
2230     if (r->iov.iov_len == 0) {
2231         scsi_req_complete(&r->req, GOOD);
2232     }
2233     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2234         assert(r->iov.iov_len == req->cmd.xfer);
2235         return -r->iov.iov_len;
2236     } else {
2237         return r->iov.iov_len;
2238     }
2239 
2240 illegal_request:
2241     if (r->req.status == -1) {
2242         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2243     }
2244     return 0;
2245 
2246 illegal_lba:
2247     scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2248     return 0;
2249 }
2250 
2251 /* Execute a scsi command.  Returns the length of the data expected by the
2252    command.  This will be Positive for data transfers from the device
2253    (eg. disk reads), negative for transfers to the device (eg. disk writes),
2254    and zero if the command does not transfer any data.  */
2255 
2256 static int32_t scsi_disk_dma_command(SCSIRequest *req, uint8_t *buf)
2257 {
2258     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
2259     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2260     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
2261     uint32_t len;
2262     uint8_t command;
2263 
2264     command = buf[0];
2265 
2266     if (!blk_is_available(s->qdev.conf.blk)) {
2267         scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
2268         return 0;
2269     }
2270 
2271     len = scsi_data_cdb_xfer(r->req.cmd.buf);
2272     switch (command) {
2273     case READ_6:
2274     case READ_10:
2275     case READ_12:
2276     case READ_16:
2277         trace_scsi_disk_dma_command_READ(r->req.cmd.lba, len);
2278         /* Protection information is not supported.  For SCSI versions 2 and
2279          * older (as determined by snooping the guest's INQUIRY commands),
2280          * there is no RD/WR/VRPROTECT, so skip this check in these versions.
2281          */
2282         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2283             goto illegal_request;
2284         }
2285         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2286             goto illegal_lba;
2287         }
2288         r->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2289         r->sector_count = len * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2290         break;
2291     case WRITE_6:
2292     case WRITE_10:
2293     case WRITE_12:
2294     case WRITE_16:
2295     case WRITE_VERIFY_10:
2296     case WRITE_VERIFY_12:
2297     case WRITE_VERIFY_16:
2298         if (!blk_is_writable(s->qdev.conf.blk)) {
2299             scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
2300             return 0;
2301         }
2302         trace_scsi_disk_dma_command_WRITE(
2303                 (command & 0xe) == 0xe ? "And Verify " : "",
2304                 r->req.cmd.lba, len);
2305         /* fall through */
2306     case VERIFY_10:
2307     case VERIFY_12:
2308     case VERIFY_16:
2309         /* We get here only for BYTCHK == 0x01 and only for scsi-block.
2310          * As far as DMA is concerned, we can treat it the same as a write;
2311          * scsi_block_do_sgio will send VERIFY commands.
2312          */
2313         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2314             goto illegal_request;
2315         }
2316         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2317             goto illegal_lba;
2318         }
2319         r->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2320         r->sector_count = len * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2321         break;
2322     default:
2323         abort();
2324     illegal_request:
2325         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2326         return 0;
2327     illegal_lba:
2328         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2329         return 0;
2330     }
2331     r->need_fua_emulation = sdc->need_fua_emulation(&r->req.cmd);
2332     if (r->sector_count == 0) {
2333         scsi_req_complete(&r->req, GOOD);
2334     }
2335     assert(r->iov.iov_len == 0);
2336     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2337         return -r->sector_count * BDRV_SECTOR_SIZE;
2338     } else {
2339         return r->sector_count * BDRV_SECTOR_SIZE;
2340     }
2341 }
2342 
2343 static void scsi_disk_reset(DeviceState *dev)
2344 {
2345     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev.qdev, dev);
2346     uint64_t nb_sectors;
2347     AioContext *ctx;
2348 
2349     scsi_device_purge_requests(&s->qdev, SENSE_CODE(RESET));
2350 
2351     ctx = blk_get_aio_context(s->qdev.conf.blk);
2352     aio_context_acquire(ctx);
2353     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2354     aio_context_release(ctx);
2355 
2356     nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2357     if (nb_sectors) {
2358         nb_sectors--;
2359     }
2360     s->qdev.max_lba = nb_sectors;
2361     /* reset tray statuses */
2362     s->tray_locked = 0;
2363     s->tray_open = 0;
2364 
2365     s->qdev.scsi_version = s->qdev.default_scsi_version;
2366 }
2367 
2368 static void scsi_disk_drained_begin(void *opaque)
2369 {
2370     SCSIDiskState *s = opaque;
2371 
2372     scsi_device_drained_begin(&s->qdev);
2373 }
2374 
2375 static void scsi_disk_drained_end(void *opaque)
2376 {
2377     SCSIDiskState *s = opaque;
2378 
2379     scsi_device_drained_end(&s->qdev);
2380 }
2381 
2382 static void scsi_disk_resize_cb(void *opaque)
2383 {
2384     SCSIDiskState *s = opaque;
2385 
2386     /* SPC lists this sense code as available only for
2387      * direct-access devices.
2388      */
2389     if (s->qdev.type == TYPE_DISK) {
2390         scsi_device_report_change(&s->qdev, SENSE_CODE(CAPACITY_CHANGED));
2391     }
2392 }
2393 
2394 static void scsi_cd_change_media_cb(void *opaque, bool load, Error **errp)
2395 {
2396     SCSIDiskState *s = opaque;
2397 
2398     /*
2399      * When a CD gets changed, we have to report an ejected state and
2400      * then a loaded state to guests so that they detect tray
2401      * open/close and media change events.  Guests that do not use
2402      * GET_EVENT_STATUS_NOTIFICATION to detect such tray open/close
2403      * states rely on this behavior.
2404      *
2405      * media_changed governs the state machine used for unit attention
2406      * report.  media_event is used by GET EVENT STATUS NOTIFICATION.
2407      */
2408     s->media_changed = load;
2409     s->tray_open = !load;
2410     scsi_device_set_ua(&s->qdev, SENSE_CODE(UNIT_ATTENTION_NO_MEDIUM));
2411     s->media_event = true;
2412     s->eject_request = false;
2413 }
2414 
2415 static void scsi_cd_eject_request_cb(void *opaque, bool force)
2416 {
2417     SCSIDiskState *s = opaque;
2418 
2419     s->eject_request = true;
2420     if (force) {
2421         s->tray_locked = false;
2422     }
2423 }
2424 
2425 static bool scsi_cd_is_tray_open(void *opaque)
2426 {
2427     return ((SCSIDiskState *)opaque)->tray_open;
2428 }
2429 
2430 static bool scsi_cd_is_medium_locked(void *opaque)
2431 {
2432     return ((SCSIDiskState *)opaque)->tray_locked;
2433 }
2434 
2435 static const BlockDevOps scsi_disk_removable_block_ops = {
2436     .change_media_cb  = scsi_cd_change_media_cb,
2437     .drained_begin    = scsi_disk_drained_begin,
2438     .drained_end      = scsi_disk_drained_end,
2439     .eject_request_cb = scsi_cd_eject_request_cb,
2440     .is_medium_locked = scsi_cd_is_medium_locked,
2441     .is_tray_open     = scsi_cd_is_tray_open,
2442     .resize_cb        = scsi_disk_resize_cb,
2443 };
2444 
2445 static const BlockDevOps scsi_disk_block_ops = {
2446     .drained_begin = scsi_disk_drained_begin,
2447     .drained_end   = scsi_disk_drained_end,
2448     .resize_cb     = scsi_disk_resize_cb,
2449 };
2450 
2451 static void scsi_disk_unit_attention_reported(SCSIDevice *dev)
2452 {
2453     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2454     if (s->media_changed) {
2455         s->media_changed = false;
2456         scsi_device_set_ua(&s->qdev, SENSE_CODE(MEDIUM_CHANGED));
2457     }
2458 }
2459 
2460 static void scsi_realize(SCSIDevice *dev, Error **errp)
2461 {
2462     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2463     bool read_only;
2464 
2465     if (!s->qdev.conf.blk) {
2466         error_setg(errp, "drive property not set");
2467         return;
2468     }
2469 
2470     if (!(s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2471         !blk_is_inserted(s->qdev.conf.blk)) {
2472         error_setg(errp, "Device needs media, but drive is empty");
2473         return;
2474     }
2475 
2476     if (!blkconf_blocksizes(&s->qdev.conf, errp)) {
2477         return;
2478     }
2479 
2480     if (blk_get_aio_context(s->qdev.conf.blk) != qemu_get_aio_context() &&
2481         !s->qdev.hba_supports_iothread)
2482     {
2483         error_setg(errp, "HBA does not support iothreads");
2484         return;
2485     }
2486 
2487     if (dev->type == TYPE_DISK) {
2488         if (!blkconf_geometry(&dev->conf, NULL, 65535, 255, 255, errp)) {
2489             return;
2490         }
2491     }
2492 
2493     read_only = !blk_supports_write_perm(s->qdev.conf.blk);
2494     if (dev->type == TYPE_ROM) {
2495         read_only = true;
2496     }
2497 
2498     if (!blkconf_apply_backend_options(&dev->conf, read_only,
2499                                        dev->type == TYPE_DISK, errp)) {
2500         return;
2501     }
2502 
2503     if (s->qdev.conf.discard_granularity == -1) {
2504         s->qdev.conf.discard_granularity =
2505             MAX(s->qdev.conf.logical_block_size, DEFAULT_DISCARD_GRANULARITY);
2506     }
2507 
2508     if (!s->version) {
2509         s->version = g_strdup(qemu_hw_version());
2510     }
2511     if (!s->vendor) {
2512         s->vendor = g_strdup("QEMU");
2513     }
2514     if (!s->device_id) {
2515         if (s->serial) {
2516             s->device_id = g_strdup_printf("%.20s", s->serial);
2517         } else {
2518             const char *str = blk_name(s->qdev.conf.blk);
2519             if (str && *str) {
2520                 s->device_id = g_strdup(str);
2521             }
2522         }
2523     }
2524 
2525     if (blk_is_sg(s->qdev.conf.blk)) {
2526         error_setg(errp, "unwanted /dev/sg*");
2527         return;
2528     }
2529 
2530     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2531             !(s->features & (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS))) {
2532         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_removable_block_ops, s);
2533     } else {
2534         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_block_ops, s);
2535     }
2536 
2537     blk_iostatus_enable(s->qdev.conf.blk);
2538 
2539     add_boot_device_lchs(&dev->qdev, NULL,
2540                          dev->conf.lcyls,
2541                          dev->conf.lheads,
2542                          dev->conf.lsecs);
2543 }
2544 
2545 static void scsi_unrealize(SCSIDevice *dev)
2546 {
2547     del_boot_device_lchs(&dev->qdev, NULL);
2548 }
2549 
2550 static void scsi_hd_realize(SCSIDevice *dev, Error **errp)
2551 {
2552     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2553     AioContext *ctx = NULL;
2554     /* can happen for devices without drive. The error message for missing
2555      * backend will be issued in scsi_realize
2556      */
2557     if (s->qdev.conf.blk) {
2558         ctx = blk_get_aio_context(s->qdev.conf.blk);
2559         aio_context_acquire(ctx);
2560         if (!blkconf_blocksizes(&s->qdev.conf, errp)) {
2561             goto out;
2562         }
2563     }
2564     s->qdev.blocksize = s->qdev.conf.logical_block_size;
2565     s->qdev.type = TYPE_DISK;
2566     if (!s->product) {
2567         s->product = g_strdup("QEMU HARDDISK");
2568     }
2569     scsi_realize(&s->qdev, errp);
2570 out:
2571     if (ctx) {
2572         aio_context_release(ctx);
2573     }
2574 }
2575 
2576 static void scsi_cd_realize(SCSIDevice *dev, Error **errp)
2577 {
2578     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2579     AioContext *ctx;
2580     int ret;
2581     uint32_t blocksize = 2048;
2582 
2583     if (!dev->conf.blk) {
2584         /* Anonymous BlockBackend for an empty drive. As we put it into
2585          * dev->conf, qdev takes care of detaching on unplug. */
2586         dev->conf.blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
2587         ret = blk_attach_dev(dev->conf.blk, &dev->qdev);
2588         assert(ret == 0);
2589     }
2590 
2591     if (dev->conf.physical_block_size != 0) {
2592         blocksize = dev->conf.physical_block_size;
2593     }
2594 
2595     ctx = blk_get_aio_context(dev->conf.blk);
2596     aio_context_acquire(ctx);
2597     s->qdev.blocksize = blocksize;
2598     s->qdev.type = TYPE_ROM;
2599     s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2600     if (!s->product) {
2601         s->product = g_strdup("QEMU CD-ROM");
2602     }
2603     scsi_realize(&s->qdev, errp);
2604     aio_context_release(ctx);
2605 }
2606 
2607 
2608 static const SCSIReqOps scsi_disk_emulate_reqops = {
2609     .size         = sizeof(SCSIDiskReq),
2610     .free_req     = scsi_free_request,
2611     .send_command = scsi_disk_emulate_command,
2612     .read_data    = scsi_disk_emulate_read_data,
2613     .write_data   = scsi_disk_emulate_write_data,
2614     .get_buf      = scsi_get_buf,
2615 };
2616 
2617 static const SCSIReqOps scsi_disk_dma_reqops = {
2618     .size         = sizeof(SCSIDiskReq),
2619     .free_req     = scsi_free_request,
2620     .send_command = scsi_disk_dma_command,
2621     .read_data    = scsi_read_data,
2622     .write_data   = scsi_write_data,
2623     .get_buf      = scsi_get_buf,
2624     .load_request = scsi_disk_load_request,
2625     .save_request = scsi_disk_save_request,
2626 };
2627 
2628 static const SCSIReqOps *const scsi_disk_reqops_dispatch[256] = {
2629     [TEST_UNIT_READY]                 = &scsi_disk_emulate_reqops,
2630     [INQUIRY]                         = &scsi_disk_emulate_reqops,
2631     [MODE_SENSE]                      = &scsi_disk_emulate_reqops,
2632     [MODE_SENSE_10]                   = &scsi_disk_emulate_reqops,
2633     [START_STOP]                      = &scsi_disk_emulate_reqops,
2634     [ALLOW_MEDIUM_REMOVAL]            = &scsi_disk_emulate_reqops,
2635     [READ_CAPACITY_10]                = &scsi_disk_emulate_reqops,
2636     [READ_TOC]                        = &scsi_disk_emulate_reqops,
2637     [READ_DVD_STRUCTURE]              = &scsi_disk_emulate_reqops,
2638     [READ_DISC_INFORMATION]           = &scsi_disk_emulate_reqops,
2639     [GET_CONFIGURATION]               = &scsi_disk_emulate_reqops,
2640     [GET_EVENT_STATUS_NOTIFICATION]   = &scsi_disk_emulate_reqops,
2641     [MECHANISM_STATUS]                = &scsi_disk_emulate_reqops,
2642     [SERVICE_ACTION_IN_16]            = &scsi_disk_emulate_reqops,
2643     [REQUEST_SENSE]                   = &scsi_disk_emulate_reqops,
2644     [SYNCHRONIZE_CACHE]               = &scsi_disk_emulate_reqops,
2645     [SEEK_10]                         = &scsi_disk_emulate_reqops,
2646     [MODE_SELECT]                     = &scsi_disk_emulate_reqops,
2647     [MODE_SELECT_10]                  = &scsi_disk_emulate_reqops,
2648     [UNMAP]                           = &scsi_disk_emulate_reqops,
2649     [WRITE_SAME_10]                   = &scsi_disk_emulate_reqops,
2650     [WRITE_SAME_16]                   = &scsi_disk_emulate_reqops,
2651     [VERIFY_10]                       = &scsi_disk_emulate_reqops,
2652     [VERIFY_12]                       = &scsi_disk_emulate_reqops,
2653     [VERIFY_16]                       = &scsi_disk_emulate_reqops,
2654     [FORMAT_UNIT]                     = &scsi_disk_emulate_reqops,
2655 
2656     [READ_6]                          = &scsi_disk_dma_reqops,
2657     [READ_10]                         = &scsi_disk_dma_reqops,
2658     [READ_12]                         = &scsi_disk_dma_reqops,
2659     [READ_16]                         = &scsi_disk_dma_reqops,
2660     [WRITE_6]                         = &scsi_disk_dma_reqops,
2661     [WRITE_10]                        = &scsi_disk_dma_reqops,
2662     [WRITE_12]                        = &scsi_disk_dma_reqops,
2663     [WRITE_16]                        = &scsi_disk_dma_reqops,
2664     [WRITE_VERIFY_10]                 = &scsi_disk_dma_reqops,
2665     [WRITE_VERIFY_12]                 = &scsi_disk_dma_reqops,
2666     [WRITE_VERIFY_16]                 = &scsi_disk_dma_reqops,
2667 };
2668 
2669 static void scsi_disk_new_request_dump(uint32_t lun, uint32_t tag, uint8_t *buf)
2670 {
2671     int i;
2672     int len = scsi_cdb_length(buf);
2673     char *line_buffer, *p;
2674 
2675     assert(len > 0 && len <= 16);
2676     line_buffer = g_malloc(len * 5 + 1);
2677 
2678     for (i = 0, p = line_buffer; i < len; i++) {
2679         p += sprintf(p, " 0x%02x", buf[i]);
2680     }
2681     trace_scsi_disk_new_request(lun, tag, line_buffer);
2682 
2683     g_free(line_buffer);
2684 }
2685 
2686 static SCSIRequest *scsi_new_request(SCSIDevice *d, uint32_t tag, uint32_t lun,
2687                                      uint8_t *buf, void *hba_private)
2688 {
2689     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2690     SCSIRequest *req;
2691     const SCSIReqOps *ops;
2692     uint8_t command;
2693 
2694     command = buf[0];
2695     ops = scsi_disk_reqops_dispatch[command];
2696     if (!ops) {
2697         ops = &scsi_disk_emulate_reqops;
2698     }
2699     req = scsi_req_alloc(ops, &s->qdev, tag, lun, hba_private);
2700 
2701     if (trace_event_get_state_backends(TRACE_SCSI_DISK_NEW_REQUEST)) {
2702         scsi_disk_new_request_dump(lun, tag, buf);
2703     }
2704 
2705     return req;
2706 }
2707 
2708 #ifdef __linux__
2709 static int get_device_type(SCSIDiskState *s)
2710 {
2711     uint8_t cmd[16];
2712     uint8_t buf[36];
2713     int ret;
2714 
2715     memset(cmd, 0, sizeof(cmd));
2716     memset(buf, 0, sizeof(buf));
2717     cmd[0] = INQUIRY;
2718     cmd[4] = sizeof(buf);
2719 
2720     ret = scsi_SG_IO_FROM_DEV(s->qdev.conf.blk, cmd, sizeof(cmd),
2721                               buf, sizeof(buf), s->qdev.io_timeout);
2722     if (ret < 0) {
2723         return -1;
2724     }
2725     s->qdev.type = buf[0];
2726     if (buf[1] & 0x80) {
2727         s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2728     }
2729     return 0;
2730 }
2731 
2732 static void scsi_block_realize(SCSIDevice *dev, Error **errp)
2733 {
2734     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2735     AioContext *ctx;
2736     int sg_version;
2737     int rc;
2738 
2739     if (!s->qdev.conf.blk) {
2740         error_setg(errp, "drive property not set");
2741         return;
2742     }
2743 
2744     if (s->rotation_rate) {
2745         error_report_once("rotation_rate is specified for scsi-block but is "
2746                           "not implemented. This option is deprecated and will "
2747                           "be removed in a future version");
2748     }
2749 
2750     ctx = blk_get_aio_context(s->qdev.conf.blk);
2751     aio_context_acquire(ctx);
2752 
2753     /* check we are using a driver managing SG_IO (version 3 and after) */
2754     rc = blk_ioctl(s->qdev.conf.blk, SG_GET_VERSION_NUM, &sg_version);
2755     if (rc < 0) {
2756         error_setg_errno(errp, -rc, "cannot get SG_IO version number");
2757         if (rc != -EPERM) {
2758             error_append_hint(errp, "Is this a SCSI device?\n");
2759         }
2760         goto out;
2761     }
2762     if (sg_version < 30000) {
2763         error_setg(errp, "scsi generic interface too old");
2764         goto out;
2765     }
2766 
2767     /* get device type from INQUIRY data */
2768     rc = get_device_type(s);
2769     if (rc < 0) {
2770         error_setg(errp, "INQUIRY failed");
2771         goto out;
2772     }
2773 
2774     /* Make a guess for the block size, we'll fix it when the guest sends.
2775      * READ CAPACITY.  If they don't, they likely would assume these sizes
2776      * anyway. (TODO: check in /sys).
2777      */
2778     if (s->qdev.type == TYPE_ROM || s->qdev.type == TYPE_WORM) {
2779         s->qdev.blocksize = 2048;
2780     } else {
2781         s->qdev.blocksize = 512;
2782     }
2783 
2784     /* Makes the scsi-block device not removable by using HMP and QMP eject
2785      * command.
2786      */
2787     s->features |= (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS);
2788 
2789     scsi_realize(&s->qdev, errp);
2790     scsi_generic_read_device_inquiry(&s->qdev);
2791 
2792 out:
2793     aio_context_release(ctx);
2794 }
2795 
2796 typedef struct SCSIBlockReq {
2797     SCSIDiskReq req;
2798     sg_io_hdr_t io_header;
2799 
2800     /* Selected bytes of the original CDB, copied into our own CDB.  */
2801     uint8_t cmd, cdb1, group_number;
2802 
2803     /* CDB passed to SG_IO.  */
2804     uint8_t cdb[16];
2805     BlockCompletionFunc *cb;
2806     void *cb_opaque;
2807 } SCSIBlockReq;
2808 
2809 static void scsi_block_sgio_complete(void *opaque, int ret)
2810 {
2811     SCSIBlockReq *req = (SCSIBlockReq *)opaque;
2812     SCSIDiskReq *r = &req->req;
2813     SCSIDevice *s = r->req.dev;
2814     sg_io_hdr_t *io_hdr = &req->io_header;
2815 
2816     if (ret == 0) {
2817         if (io_hdr->host_status != SCSI_HOST_OK) {
2818             scsi_req_complete_failed(&r->req, io_hdr->host_status);
2819             scsi_req_unref(&r->req);
2820             return;
2821         }
2822 
2823         if (io_hdr->driver_status & SG_ERR_DRIVER_TIMEOUT) {
2824             ret = BUSY;
2825         } else {
2826             ret = io_hdr->status;
2827         }
2828 
2829         if (ret > 0) {
2830             aio_context_acquire(blk_get_aio_context(s->conf.blk));
2831             if (scsi_handle_rw_error(r, ret, true)) {
2832                 aio_context_release(blk_get_aio_context(s->conf.blk));
2833                 scsi_req_unref(&r->req);
2834                 return;
2835             }
2836             aio_context_release(blk_get_aio_context(s->conf.blk));
2837 
2838             /* Ignore error.  */
2839             ret = 0;
2840         }
2841     }
2842 
2843     req->cb(req->cb_opaque, ret);
2844 }
2845 
2846 static BlockAIOCB *scsi_block_do_sgio(SCSIBlockReq *req,
2847                                       int64_t offset, QEMUIOVector *iov,
2848                                       int direction,
2849                                       BlockCompletionFunc *cb, void *opaque)
2850 {
2851     sg_io_hdr_t *io_header = &req->io_header;
2852     SCSIDiskReq *r = &req->req;
2853     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2854     int nb_logical_blocks;
2855     uint64_t lba;
2856     BlockAIOCB *aiocb;
2857 
2858     /* This is not supported yet.  It can only happen if the guest does
2859      * reads and writes that are not aligned to one logical sectors
2860      * _and_ cover multiple MemoryRegions.
2861      */
2862     assert(offset % s->qdev.blocksize == 0);
2863     assert(iov->size % s->qdev.blocksize == 0);
2864 
2865     io_header->interface_id = 'S';
2866 
2867     /* The data transfer comes from the QEMUIOVector.  */
2868     io_header->dxfer_direction = direction;
2869     io_header->dxfer_len = iov->size;
2870     io_header->dxferp = (void *)iov->iov;
2871     io_header->iovec_count = iov->niov;
2872     assert(io_header->iovec_count == iov->niov); /* no overflow! */
2873 
2874     /* Build a new CDB with the LBA and length patched in, in case
2875      * DMA helpers split the transfer in multiple segments.  Do not
2876      * build a CDB smaller than what the guest wanted, and only build
2877      * a larger one if strictly necessary.
2878      */
2879     io_header->cmdp = req->cdb;
2880     lba = offset / s->qdev.blocksize;
2881     nb_logical_blocks = io_header->dxfer_len / s->qdev.blocksize;
2882 
2883     if ((req->cmd >> 5) == 0 && lba <= 0x1ffff) {
2884         /* 6-byte CDB */
2885         stl_be_p(&req->cdb[0], lba | (req->cmd << 24));
2886         req->cdb[4] = nb_logical_blocks;
2887         req->cdb[5] = 0;
2888         io_header->cmd_len = 6;
2889     } else if ((req->cmd >> 5) <= 1 && lba <= 0xffffffffULL) {
2890         /* 10-byte CDB */
2891         req->cdb[0] = (req->cmd & 0x1f) | 0x20;
2892         req->cdb[1] = req->cdb1;
2893         stl_be_p(&req->cdb[2], lba);
2894         req->cdb[6] = req->group_number;
2895         stw_be_p(&req->cdb[7], nb_logical_blocks);
2896         req->cdb[9] = 0;
2897         io_header->cmd_len = 10;
2898     } else if ((req->cmd >> 5) != 4 && lba <= 0xffffffffULL) {
2899         /* 12-byte CDB */
2900         req->cdb[0] = (req->cmd & 0x1f) | 0xA0;
2901         req->cdb[1] = req->cdb1;
2902         stl_be_p(&req->cdb[2], lba);
2903         stl_be_p(&req->cdb[6], nb_logical_blocks);
2904         req->cdb[10] = req->group_number;
2905         req->cdb[11] = 0;
2906         io_header->cmd_len = 12;
2907     } else {
2908         /* 16-byte CDB */
2909         req->cdb[0] = (req->cmd & 0x1f) | 0x80;
2910         req->cdb[1] = req->cdb1;
2911         stq_be_p(&req->cdb[2], lba);
2912         stl_be_p(&req->cdb[10], nb_logical_blocks);
2913         req->cdb[14] = req->group_number;
2914         req->cdb[15] = 0;
2915         io_header->cmd_len = 16;
2916     }
2917 
2918     /* The rest is as in scsi-generic.c.  */
2919     io_header->mx_sb_len = sizeof(r->req.sense);
2920     io_header->sbp = r->req.sense;
2921     io_header->timeout = s->qdev.io_timeout * 1000;
2922     io_header->usr_ptr = r;
2923     io_header->flags |= SG_FLAG_DIRECT_IO;
2924     req->cb = cb;
2925     req->cb_opaque = opaque;
2926     trace_scsi_disk_aio_sgio_command(r->req.tag, req->cdb[0], lba,
2927                                      nb_logical_blocks, io_header->timeout);
2928     aiocb = blk_aio_ioctl(s->qdev.conf.blk, SG_IO, io_header, scsi_block_sgio_complete, req);
2929     assert(aiocb != NULL);
2930     return aiocb;
2931 }
2932 
2933 static bool scsi_block_no_fua(SCSICommand *cmd)
2934 {
2935     return false;
2936 }
2937 
2938 static BlockAIOCB *scsi_block_dma_readv(int64_t offset,
2939                                         QEMUIOVector *iov,
2940                                         BlockCompletionFunc *cb, void *cb_opaque,
2941                                         void *opaque)
2942 {
2943     SCSIBlockReq *r = opaque;
2944     return scsi_block_do_sgio(r, offset, iov,
2945                               SG_DXFER_FROM_DEV, cb, cb_opaque);
2946 }
2947 
2948 static BlockAIOCB *scsi_block_dma_writev(int64_t offset,
2949                                          QEMUIOVector *iov,
2950                                          BlockCompletionFunc *cb, void *cb_opaque,
2951                                          void *opaque)
2952 {
2953     SCSIBlockReq *r = opaque;
2954     return scsi_block_do_sgio(r, offset, iov,
2955                               SG_DXFER_TO_DEV, cb, cb_opaque);
2956 }
2957 
2958 static bool scsi_block_is_passthrough(SCSIDiskState *s, uint8_t *buf)
2959 {
2960     switch (buf[0]) {
2961     case VERIFY_10:
2962     case VERIFY_12:
2963     case VERIFY_16:
2964         /* Check if BYTCHK == 0x01 (data-out buffer contains data
2965          * for the number of logical blocks specified in the length
2966          * field).  For other modes, do not use scatter/gather operation.
2967          */
2968         if ((buf[1] & 6) == 2) {
2969             return false;
2970         }
2971         break;
2972 
2973     case READ_6:
2974     case READ_10:
2975     case READ_12:
2976     case READ_16:
2977     case WRITE_6:
2978     case WRITE_10:
2979     case WRITE_12:
2980     case WRITE_16:
2981     case WRITE_VERIFY_10:
2982     case WRITE_VERIFY_12:
2983     case WRITE_VERIFY_16:
2984         /* MMC writing cannot be done via DMA helpers, because it sometimes
2985          * involves writing beyond the maximum LBA or to negative LBA (lead-in).
2986          * We might use scsi_block_dma_reqops as long as no writing commands are
2987          * seen, but performance usually isn't paramount on optical media.  So,
2988          * just make scsi-block operate the same as scsi-generic for them.
2989          */
2990         if (s->qdev.type != TYPE_ROM) {
2991             return false;
2992         }
2993         break;
2994 
2995     default:
2996         break;
2997     }
2998 
2999     return true;
3000 }
3001 
3002 
3003 static int32_t scsi_block_dma_command(SCSIRequest *req, uint8_t *buf)
3004 {
3005     SCSIBlockReq *r = (SCSIBlockReq *)req;
3006     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
3007 
3008     r->cmd = req->cmd.buf[0];
3009     switch (r->cmd >> 5) {
3010     case 0:
3011         /* 6-byte CDB.  */
3012         r->cdb1 = r->group_number = 0;
3013         break;
3014     case 1:
3015         /* 10-byte CDB.  */
3016         r->cdb1 = req->cmd.buf[1];
3017         r->group_number = req->cmd.buf[6];
3018         break;
3019     case 4:
3020         /* 12-byte CDB.  */
3021         r->cdb1 = req->cmd.buf[1];
3022         r->group_number = req->cmd.buf[10];
3023         break;
3024     case 5:
3025         /* 16-byte CDB.  */
3026         r->cdb1 = req->cmd.buf[1];
3027         r->group_number = req->cmd.buf[14];
3028         break;
3029     default:
3030         abort();
3031     }
3032 
3033     /* Protection information is not supported.  For SCSI versions 2 and
3034      * older (as determined by snooping the guest's INQUIRY commands),
3035      * there is no RD/WR/VRPROTECT, so skip this check in these versions.
3036      */
3037     if (s->qdev.scsi_version > 2 && (req->cmd.buf[1] & 0xe0)) {
3038         scsi_check_condition(&r->req, SENSE_CODE(INVALID_FIELD));
3039         return 0;
3040     }
3041 
3042     return scsi_disk_dma_command(req, buf);
3043 }
3044 
3045 static const SCSIReqOps scsi_block_dma_reqops = {
3046     .size         = sizeof(SCSIBlockReq),
3047     .free_req     = scsi_free_request,
3048     .send_command = scsi_block_dma_command,
3049     .read_data    = scsi_read_data,
3050     .write_data   = scsi_write_data,
3051     .get_buf      = scsi_get_buf,
3052     .load_request = scsi_disk_load_request,
3053     .save_request = scsi_disk_save_request,
3054 };
3055 
3056 static SCSIRequest *scsi_block_new_request(SCSIDevice *d, uint32_t tag,
3057                                            uint32_t lun, uint8_t *buf,
3058                                            void *hba_private)
3059 {
3060     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
3061 
3062     if (scsi_block_is_passthrough(s, buf)) {
3063         return scsi_req_alloc(&scsi_generic_req_ops, &s->qdev, tag, lun,
3064                               hba_private);
3065     } else {
3066         return scsi_req_alloc(&scsi_block_dma_reqops, &s->qdev, tag, lun,
3067                               hba_private);
3068     }
3069 }
3070 
3071 static int scsi_block_parse_cdb(SCSIDevice *d, SCSICommand *cmd,
3072                                   uint8_t *buf, size_t buf_len,
3073                                   void *hba_private)
3074 {
3075     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
3076 
3077     if (scsi_block_is_passthrough(s, buf)) {
3078         return scsi_bus_parse_cdb(&s->qdev, cmd, buf, buf_len, hba_private);
3079     } else {
3080         return scsi_req_parse_cdb(&s->qdev, cmd, buf, buf_len);
3081     }
3082 }
3083 
3084 static void scsi_block_update_sense(SCSIRequest *req)
3085 {
3086     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
3087     SCSIBlockReq *br = DO_UPCAST(SCSIBlockReq, req, r);
3088     r->req.sense_len = MIN(br->io_header.sb_len_wr, sizeof(r->req.sense));
3089 }
3090 #endif
3091 
3092 static
3093 BlockAIOCB *scsi_dma_readv(int64_t offset, QEMUIOVector *iov,
3094                            BlockCompletionFunc *cb, void *cb_opaque,
3095                            void *opaque)
3096 {
3097     SCSIDiskReq *r = opaque;
3098     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
3099     return blk_aio_preadv(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
3100 }
3101 
3102 static
3103 BlockAIOCB *scsi_dma_writev(int64_t offset, QEMUIOVector *iov,
3104                             BlockCompletionFunc *cb, void *cb_opaque,
3105                             void *opaque)
3106 {
3107     SCSIDiskReq *r = opaque;
3108     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
3109     return blk_aio_pwritev(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
3110 }
3111 
3112 static void scsi_disk_base_class_initfn(ObjectClass *klass, void *data)
3113 {
3114     DeviceClass *dc = DEVICE_CLASS(klass);
3115     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
3116 
3117     dc->fw_name = "disk";
3118     dc->reset = scsi_disk_reset;
3119     sdc->dma_readv = scsi_dma_readv;
3120     sdc->dma_writev = scsi_dma_writev;
3121     sdc->need_fua_emulation = scsi_is_cmd_fua;
3122 }
3123 
3124 static const TypeInfo scsi_disk_base_info = {
3125     .name          = TYPE_SCSI_DISK_BASE,
3126     .parent        = TYPE_SCSI_DEVICE,
3127     .class_init    = scsi_disk_base_class_initfn,
3128     .instance_size = sizeof(SCSIDiskState),
3129     .class_size    = sizeof(SCSIDiskClass),
3130     .abstract      = true,
3131 };
3132 
3133 #define DEFINE_SCSI_DISK_PROPERTIES()                                   \
3134     DEFINE_PROP_DRIVE_IOTHREAD("drive", SCSIDiskState, qdev.conf.blk),  \
3135     DEFINE_BLOCK_PROPERTIES_BASE(SCSIDiskState, qdev.conf),             \
3136     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),            \
3137     DEFINE_PROP_STRING("ver", SCSIDiskState, version),                  \
3138     DEFINE_PROP_STRING("serial", SCSIDiskState, serial),                \
3139     DEFINE_PROP_STRING("vendor", SCSIDiskState, vendor),                \
3140     DEFINE_PROP_STRING("product", SCSIDiskState, product),              \
3141     DEFINE_PROP_STRING("device_id", SCSIDiskState, device_id)
3142 
3143 
3144 static Property scsi_hd_properties[] = {
3145     DEFINE_SCSI_DISK_PROPERTIES(),
3146     DEFINE_PROP_BIT("removable", SCSIDiskState, features,
3147                     SCSI_DISK_F_REMOVABLE, false),
3148     DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
3149                     SCSI_DISK_F_DPOFUA, false),
3150     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3151     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3152     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3153     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3154                        DEFAULT_MAX_UNMAP_SIZE),
3155     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3156                        DEFAULT_MAX_IO_SIZE),
3157     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3158     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3159                       5),
3160     DEFINE_PROP_BIT("quirk_mode_page_vendor_specific_apple", SCSIDiskState,
3161                     quirks, SCSI_DISK_QUIRK_MODE_PAGE_VENDOR_SPECIFIC_APPLE,
3162                     0),
3163     DEFINE_BLOCK_CHS_PROPERTIES(SCSIDiskState, qdev.conf),
3164     DEFINE_PROP_END_OF_LIST(),
3165 };
3166 
3167 static const VMStateDescription vmstate_scsi_disk_state = {
3168     .name = "scsi-disk",
3169     .version_id = 1,
3170     .minimum_version_id = 1,
3171     .fields = (VMStateField[]) {
3172         VMSTATE_SCSI_DEVICE(qdev, SCSIDiskState),
3173         VMSTATE_BOOL(media_changed, SCSIDiskState),
3174         VMSTATE_BOOL(media_event, SCSIDiskState),
3175         VMSTATE_BOOL(eject_request, SCSIDiskState),
3176         VMSTATE_BOOL(tray_open, SCSIDiskState),
3177         VMSTATE_BOOL(tray_locked, SCSIDiskState),
3178         VMSTATE_END_OF_LIST()
3179     }
3180 };
3181 
3182 static void scsi_hd_class_initfn(ObjectClass *klass, void *data)
3183 {
3184     DeviceClass *dc = DEVICE_CLASS(klass);
3185     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3186 
3187     sc->realize      = scsi_hd_realize;
3188     sc->unrealize    = scsi_unrealize;
3189     sc->alloc_req    = scsi_new_request;
3190     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3191     dc->desc = "virtual SCSI disk";
3192     device_class_set_props(dc, scsi_hd_properties);
3193     dc->vmsd  = &vmstate_scsi_disk_state;
3194 }
3195 
3196 static const TypeInfo scsi_hd_info = {
3197     .name          = "scsi-hd",
3198     .parent        = TYPE_SCSI_DISK_BASE,
3199     .class_init    = scsi_hd_class_initfn,
3200 };
3201 
3202 static Property scsi_cd_properties[] = {
3203     DEFINE_SCSI_DISK_PROPERTIES(),
3204     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3205     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3206     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3207     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3208                        DEFAULT_MAX_IO_SIZE),
3209     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3210                       5),
3211     DEFINE_PROP_BIT("quirk_mode_page_apple_vendor", SCSIDiskState, quirks,
3212                     SCSI_DISK_QUIRK_MODE_PAGE_APPLE_VENDOR, 0),
3213     DEFINE_PROP_BIT("quirk_mode_sense_rom_use_dbd", SCSIDiskState, quirks,
3214                     SCSI_DISK_QUIRK_MODE_SENSE_ROM_USE_DBD, 0),
3215     DEFINE_PROP_BIT("quirk_mode_page_vendor_specific_apple", SCSIDiskState,
3216                     quirks, SCSI_DISK_QUIRK_MODE_PAGE_VENDOR_SPECIFIC_APPLE,
3217                     0),
3218     DEFINE_PROP_BIT("quirk_mode_page_truncated", SCSIDiskState, quirks,
3219                     SCSI_DISK_QUIRK_MODE_PAGE_TRUNCATED, 0),
3220     DEFINE_PROP_END_OF_LIST(),
3221 };
3222 
3223 static void scsi_cd_class_initfn(ObjectClass *klass, void *data)
3224 {
3225     DeviceClass *dc = DEVICE_CLASS(klass);
3226     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3227 
3228     sc->realize      = scsi_cd_realize;
3229     sc->alloc_req    = scsi_new_request;
3230     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3231     dc->desc = "virtual SCSI CD-ROM";
3232     device_class_set_props(dc, scsi_cd_properties);
3233     dc->vmsd  = &vmstate_scsi_disk_state;
3234 }
3235 
3236 static const TypeInfo scsi_cd_info = {
3237     .name          = "scsi-cd",
3238     .parent        = TYPE_SCSI_DISK_BASE,
3239     .class_init    = scsi_cd_class_initfn,
3240 };
3241 
3242 #ifdef __linux__
3243 static Property scsi_block_properties[] = {
3244     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),
3245     DEFINE_PROP_DRIVE("drive", SCSIDiskState, qdev.conf.blk),
3246     DEFINE_PROP_BOOL("share-rw", SCSIDiskState, qdev.conf.share_rw, false),
3247     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3248     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3249                        DEFAULT_MAX_UNMAP_SIZE),
3250     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3251                        DEFAULT_MAX_IO_SIZE),
3252     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3253                       -1),
3254     DEFINE_PROP_UINT32("io_timeout", SCSIDiskState, qdev.io_timeout,
3255                        DEFAULT_IO_TIMEOUT),
3256     DEFINE_PROP_END_OF_LIST(),
3257 };
3258 
3259 static void scsi_block_class_initfn(ObjectClass *klass, void *data)
3260 {
3261     DeviceClass *dc = DEVICE_CLASS(klass);
3262     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3263     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
3264 
3265     sc->realize      = scsi_block_realize;
3266     sc->alloc_req    = scsi_block_new_request;
3267     sc->parse_cdb    = scsi_block_parse_cdb;
3268     sdc->dma_readv   = scsi_block_dma_readv;
3269     sdc->dma_writev  = scsi_block_dma_writev;
3270     sdc->update_sense = scsi_block_update_sense;
3271     sdc->need_fua_emulation = scsi_block_no_fua;
3272     dc->desc = "SCSI block device passthrough";
3273     device_class_set_props(dc, scsi_block_properties);
3274     dc->vmsd  = &vmstate_scsi_disk_state;
3275 }
3276 
3277 static const TypeInfo scsi_block_info = {
3278     .name          = "scsi-block",
3279     .parent        = TYPE_SCSI_DISK_BASE,
3280     .class_init    = scsi_block_class_initfn,
3281 };
3282 #endif
3283 
3284 static void scsi_disk_register_types(void)
3285 {
3286     type_register_static(&scsi_disk_base_info);
3287     type_register_static(&scsi_hd_info);
3288     type_register_static(&scsi_cd_info);
3289 #ifdef __linux__
3290     type_register_static(&scsi_block_info);
3291 #endif
3292 }
3293 
3294 type_init(scsi_disk_register_types)
3295