xref: /openbmc/qemu/hw/scsi/scsi-disk.c (revision 89854803)
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 //#define DEBUG_SCSI
23 
24 #ifdef DEBUG_SCSI
25 #define DPRINTF(fmt, ...) \
26 do { printf("scsi-disk: " fmt , ## __VA_ARGS__); } while (0)
27 #else
28 #define DPRINTF(fmt, ...) do {} while(0)
29 #endif
30 
31 #include "qemu/osdep.h"
32 #include "qapi/error.h"
33 #include "qemu/error-report.h"
34 #include "hw/scsi/scsi.h"
35 #include "scsi/constants.h"
36 #include "sysemu/sysemu.h"
37 #include "sysemu/block-backend.h"
38 #include "sysemu/blockdev.h"
39 #include "hw/block/block.h"
40 #include "sysemu/dma.h"
41 #include "qemu/cutils.h"
42 
43 #ifdef __linux
44 #include <scsi/sg.h>
45 #endif
46 
47 #define SCSI_WRITE_SAME_MAX         524288
48 #define SCSI_DMA_BUF_SIZE           131072
49 #define SCSI_MAX_INQUIRY_LEN        256
50 #define SCSI_MAX_MODE_LEN           256
51 
52 #define DEFAULT_DISCARD_GRANULARITY 4096
53 #define DEFAULT_MAX_UNMAP_SIZE      (1 << 30)   /* 1 GB */
54 #define DEFAULT_MAX_IO_SIZE         INT_MAX     /* 2 GB - 1 block */
55 
56 #define TYPE_SCSI_DISK_BASE         "scsi-disk-base"
57 
58 #define SCSI_DISK_BASE(obj) \
59      OBJECT_CHECK(SCSIDiskState, (obj), TYPE_SCSI_DISK_BASE)
60 #define SCSI_DISK_BASE_CLASS(klass) \
61      OBJECT_CLASS_CHECK(SCSIDiskClass, (klass), TYPE_SCSI_DISK_BASE)
62 #define SCSI_DISK_BASE_GET_CLASS(obj) \
63      OBJECT_GET_CLASS(SCSIDiskClass, (obj), TYPE_SCSI_DISK_BASE)
64 
65 typedef struct SCSIDiskClass {
66     SCSIDeviceClass parent_class;
67     DMAIOFunc       *dma_readv;
68     DMAIOFunc       *dma_writev;
69     bool            (*need_fua_emulation)(SCSICommand *cmd);
70 } SCSIDiskClass;
71 
72 typedef struct SCSIDiskReq {
73     SCSIRequest req;
74     /* Both sector and sector_count are in terms of qemu 512 byte blocks.  */
75     uint64_t sector;
76     uint32_t sector_count;
77     uint32_t buflen;
78     bool started;
79     bool need_fua_emulation;
80     struct iovec iov;
81     QEMUIOVector qiov;
82     BlockAcctCookie acct;
83     unsigned char *status;
84 } SCSIDiskReq;
85 
86 #define SCSI_DISK_F_REMOVABLE             0
87 #define SCSI_DISK_F_DPOFUA                1
88 #define SCSI_DISK_F_NO_REMOVABLE_DEVOPS   2
89 
90 typedef struct SCSIDiskState
91 {
92     SCSIDevice qdev;
93     uint32_t features;
94     bool media_changed;
95     bool media_event;
96     bool eject_request;
97     uint16_t port_index;
98     uint64_t max_unmap_size;
99     uint64_t max_io_size;
100     QEMUBH *bh;
101     char *version;
102     char *serial;
103     char *vendor;
104     char *product;
105     bool tray_open;
106     bool tray_locked;
107     /*
108      * 0x0000        - rotation rate not reported
109      * 0x0001        - non-rotating medium (SSD)
110      * 0x0002-0x0400 - reserved
111      * 0x0401-0xffe  - rotations per minute
112      * 0xffff        - reserved
113      */
114     uint16_t rotation_rate;
115 } SCSIDiskState;
116 
117 static bool scsi_handle_rw_error(SCSIDiskReq *r, int error, bool acct_failed);
118 
119 static void scsi_free_request(SCSIRequest *req)
120 {
121     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
122 
123     qemu_vfree(r->iov.iov_base);
124 }
125 
126 /* Helper function for command completion with sense.  */
127 static void scsi_check_condition(SCSIDiskReq *r, SCSISense sense)
128 {
129     DPRINTF("Command complete tag=0x%x sense=%d/%d/%d\n",
130             r->req.tag, sense.key, sense.asc, sense.ascq);
131     scsi_req_build_sense(&r->req, sense);
132     scsi_req_complete(&r->req, CHECK_CONDITION);
133 }
134 
135 static void scsi_init_iovec(SCSIDiskReq *r, size_t size)
136 {
137     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
138 
139     if (!r->iov.iov_base) {
140         r->buflen = size;
141         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
142     }
143     r->iov.iov_len = MIN(r->sector_count * 512, r->buflen);
144     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
145 }
146 
147 static void scsi_disk_save_request(QEMUFile *f, SCSIRequest *req)
148 {
149     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
150 
151     qemu_put_be64s(f, &r->sector);
152     qemu_put_be32s(f, &r->sector_count);
153     qemu_put_be32s(f, &r->buflen);
154     if (r->buflen) {
155         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
156             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
157         } else if (!req->retry) {
158             uint32_t len = r->iov.iov_len;
159             qemu_put_be32s(f, &len);
160             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
161         }
162     }
163 }
164 
165 static void scsi_disk_load_request(QEMUFile *f, SCSIRequest *req)
166 {
167     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
168 
169     qemu_get_be64s(f, &r->sector);
170     qemu_get_be32s(f, &r->sector_count);
171     qemu_get_be32s(f, &r->buflen);
172     if (r->buflen) {
173         scsi_init_iovec(r, r->buflen);
174         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
175             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
176         } else if (!r->req.retry) {
177             uint32_t len;
178             qemu_get_be32s(f, &len);
179             r->iov.iov_len = len;
180             assert(r->iov.iov_len <= r->buflen);
181             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
182         }
183     }
184 
185     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
186 }
187 
188 static bool scsi_disk_req_check_error(SCSIDiskReq *r, int ret, bool acct_failed)
189 {
190     if (r->req.io_canceled) {
191         scsi_req_cancel_complete(&r->req);
192         return true;
193     }
194 
195     if (ret < 0 || (r->status && *r->status)) {
196         return scsi_handle_rw_error(r, -ret, acct_failed);
197     }
198 
199     return false;
200 }
201 
202 static void scsi_aio_complete(void *opaque, int ret)
203 {
204     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
205     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
206 
207     assert(r->req.aiocb != NULL);
208     r->req.aiocb = NULL;
209     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
210     if (scsi_disk_req_check_error(r, ret, true)) {
211         goto done;
212     }
213 
214     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
215     scsi_req_complete(&r->req, GOOD);
216 
217 done:
218     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
219     scsi_req_unref(&r->req);
220 }
221 
222 static bool scsi_is_cmd_fua(SCSICommand *cmd)
223 {
224     switch (cmd->buf[0]) {
225     case READ_10:
226     case READ_12:
227     case READ_16:
228     case WRITE_10:
229     case WRITE_12:
230     case WRITE_16:
231         return (cmd->buf[1] & 8) != 0;
232 
233     case VERIFY_10:
234     case VERIFY_12:
235     case VERIFY_16:
236     case WRITE_VERIFY_10:
237     case WRITE_VERIFY_12:
238     case WRITE_VERIFY_16:
239         return true;
240 
241     case READ_6:
242     case WRITE_6:
243     default:
244         return false;
245     }
246 }
247 
248 static void scsi_write_do_fua(SCSIDiskReq *r)
249 {
250     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
251 
252     assert(r->req.aiocb == NULL);
253     assert(!r->req.io_canceled);
254 
255     if (r->need_fua_emulation) {
256         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
257                          BLOCK_ACCT_FLUSH);
258         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
259         return;
260     }
261 
262     scsi_req_complete(&r->req, GOOD);
263     scsi_req_unref(&r->req);
264 }
265 
266 static void scsi_dma_complete_noio(SCSIDiskReq *r, int ret)
267 {
268     assert(r->req.aiocb == NULL);
269     if (scsi_disk_req_check_error(r, ret, false)) {
270         goto done;
271     }
272 
273     r->sector += r->sector_count;
274     r->sector_count = 0;
275     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
276         scsi_write_do_fua(r);
277         return;
278     } else {
279         scsi_req_complete(&r->req, GOOD);
280     }
281 
282 done:
283     scsi_req_unref(&r->req);
284 }
285 
286 static void scsi_dma_complete(void *opaque, int ret)
287 {
288     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
289     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
290 
291     assert(r->req.aiocb != NULL);
292     r->req.aiocb = NULL;
293 
294     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
295     if (ret < 0) {
296         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
297     } else {
298         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
299     }
300     scsi_dma_complete_noio(r, ret);
301     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
302 }
303 
304 static void scsi_read_complete(void * opaque, int ret)
305 {
306     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
307     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
308     int n;
309 
310     assert(r->req.aiocb != NULL);
311     r->req.aiocb = NULL;
312     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
313     if (scsi_disk_req_check_error(r, ret, true)) {
314         goto done;
315     }
316 
317     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
318     DPRINTF("Data ready tag=0x%x len=%zd\n", r->req.tag, r->qiov.size);
319 
320     n = r->qiov.size / 512;
321     r->sector += n;
322     r->sector_count -= n;
323     scsi_req_data(&r->req, r->qiov.size);
324 
325 done:
326     scsi_req_unref(&r->req);
327     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
328 }
329 
330 /* Actually issue a read to the block device.  */
331 static void scsi_do_read(SCSIDiskReq *r, int ret)
332 {
333     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
334     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
335 
336     assert (r->req.aiocb == NULL);
337     if (scsi_disk_req_check_error(r, ret, false)) {
338         goto done;
339     }
340 
341     /* The request is used as the AIO opaque value, so add a ref.  */
342     scsi_req_ref(&r->req);
343 
344     if (r->req.sg) {
345         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_READ);
346         r->req.resid -= r->req.sg->size;
347         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
348                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
349                                   BDRV_SECTOR_SIZE,
350                                   sdc->dma_readv, r, scsi_dma_complete, r,
351                                   DMA_DIRECTION_FROM_DEVICE);
352     } else {
353         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
354         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
355                          r->qiov.size, BLOCK_ACCT_READ);
356         r->req.aiocb = sdc->dma_readv(r->sector << BDRV_SECTOR_BITS, &r->qiov,
357                                       scsi_read_complete, r, r);
358     }
359 
360 done:
361     scsi_req_unref(&r->req);
362 }
363 
364 static void scsi_do_read_cb(void *opaque, int ret)
365 {
366     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
367     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
368 
369     assert (r->req.aiocb != NULL);
370     r->req.aiocb = NULL;
371 
372     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
373     if (ret < 0) {
374         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
375     } else {
376         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
377     }
378     scsi_do_read(opaque, ret);
379     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
380 }
381 
382 /* Read more data from scsi device into buffer.  */
383 static void scsi_read_data(SCSIRequest *req)
384 {
385     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
386     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
387     bool first;
388 
389     DPRINTF("Read sector_count=%d\n", r->sector_count);
390     if (r->sector_count == 0) {
391         /* This also clears the sense buffer for REQUEST SENSE.  */
392         scsi_req_complete(&r->req, GOOD);
393         return;
394     }
395 
396     /* No data transfer may already be in progress */
397     assert(r->req.aiocb == NULL);
398 
399     /* The request is used as the AIO opaque value, so add a ref.  */
400     scsi_req_ref(&r->req);
401     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
402         DPRINTF("Data transfer direction invalid\n");
403         scsi_read_complete(r, -EINVAL);
404         return;
405     }
406 
407     if (!blk_is_available(req->dev->conf.blk)) {
408         scsi_read_complete(r, -ENOMEDIUM);
409         return;
410     }
411 
412     first = !r->started;
413     r->started = true;
414     if (first && r->need_fua_emulation) {
415         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
416                          BLOCK_ACCT_FLUSH);
417         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_do_read_cb, r);
418     } else {
419         scsi_do_read(r, 0);
420     }
421 }
422 
423 /*
424  * scsi_handle_rw_error has two return values.  False means that the error
425  * must be ignored, true means that the error has been processed and the
426  * caller should not do anything else for this request.  Note that
427  * scsi_handle_rw_error always manages its reference counts, independent
428  * of the return value.
429  */
430 static bool scsi_handle_rw_error(SCSIDiskReq *r, int error, bool acct_failed)
431 {
432     bool is_read = (r->req.cmd.mode == SCSI_XFER_FROM_DEV);
433     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
434     BlockErrorAction action = blk_get_error_action(s->qdev.conf.blk,
435                                                    is_read, error);
436 
437     if (action == BLOCK_ERROR_ACTION_REPORT) {
438         if (acct_failed) {
439             block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
440         }
441         switch (error) {
442         case 0:
443             /* The command has run, no need to fake sense.  */
444             assert(r->status && *r->status);
445             scsi_req_complete(&r->req, *r->status);
446             break;
447         case ENOMEDIUM:
448             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
449             break;
450         case ENOMEM:
451             scsi_check_condition(r, SENSE_CODE(TARGET_FAILURE));
452             break;
453         case EINVAL:
454             scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
455             break;
456         case ENOSPC:
457             scsi_check_condition(r, SENSE_CODE(SPACE_ALLOC_FAILED));
458             break;
459         default:
460             scsi_check_condition(r, SENSE_CODE(IO_ERROR));
461             break;
462         }
463     }
464     if (!error) {
465         assert(r->status && *r->status);
466         error = scsi_sense_buf_to_errno(r->req.sense, sizeof(r->req.sense));
467 
468         if (error == ECANCELED || error == EAGAIN || error == ENOTCONN ||
469             error == 0)  {
470             /* These errors are handled by guest. */
471             scsi_req_complete(&r->req, *r->status);
472             return true;
473         }
474     }
475 
476     blk_error_action(s->qdev.conf.blk, action, is_read, error);
477     if (action == BLOCK_ERROR_ACTION_STOP) {
478         scsi_req_retry(&r->req);
479     }
480     return action != BLOCK_ERROR_ACTION_IGNORE;
481 }
482 
483 static void scsi_write_complete_noio(SCSIDiskReq *r, int ret)
484 {
485     uint32_t n;
486 
487     assert (r->req.aiocb == NULL);
488     if (scsi_disk_req_check_error(r, ret, false)) {
489         goto done;
490     }
491 
492     n = r->qiov.size / 512;
493     r->sector += n;
494     r->sector_count -= n;
495     if (r->sector_count == 0) {
496         scsi_write_do_fua(r);
497         return;
498     } else {
499         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
500         DPRINTF("Write complete tag=0x%x more=%zd\n", r->req.tag, r->qiov.size);
501         scsi_req_data(&r->req, r->qiov.size);
502     }
503 
504 done:
505     scsi_req_unref(&r->req);
506 }
507 
508 static void scsi_write_complete(void * opaque, int ret)
509 {
510     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
511     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
512 
513     assert (r->req.aiocb != NULL);
514     r->req.aiocb = NULL;
515 
516     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
517     if (ret < 0) {
518         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
519     } else {
520         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
521     }
522     scsi_write_complete_noio(r, ret);
523     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
524 }
525 
526 static void scsi_write_data(SCSIRequest *req)
527 {
528     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
529     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
530     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
531 
532     /* No data transfer may already be in progress */
533     assert(r->req.aiocb == NULL);
534 
535     /* The request is used as the AIO opaque value, so add a ref.  */
536     scsi_req_ref(&r->req);
537     if (r->req.cmd.mode != SCSI_XFER_TO_DEV) {
538         DPRINTF("Data transfer direction invalid\n");
539         scsi_write_complete_noio(r, -EINVAL);
540         return;
541     }
542 
543     if (!r->req.sg && !r->qiov.size) {
544         /* Called for the first time.  Ask the driver to send us more data.  */
545         r->started = true;
546         scsi_write_complete_noio(r, 0);
547         return;
548     }
549     if (!blk_is_available(req->dev->conf.blk)) {
550         scsi_write_complete_noio(r, -ENOMEDIUM);
551         return;
552     }
553 
554     if (r->req.cmd.buf[0] == VERIFY_10 || r->req.cmd.buf[0] == VERIFY_12 ||
555         r->req.cmd.buf[0] == VERIFY_16) {
556         if (r->req.sg) {
557             scsi_dma_complete_noio(r, 0);
558         } else {
559             scsi_write_complete_noio(r, 0);
560         }
561         return;
562     }
563 
564     if (r->req.sg) {
565         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_WRITE);
566         r->req.resid -= r->req.sg->size;
567         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
568                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
569                                   BDRV_SECTOR_SIZE,
570                                   sdc->dma_writev, r, scsi_dma_complete, r,
571                                   DMA_DIRECTION_TO_DEVICE);
572     } else {
573         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
574                          r->qiov.size, BLOCK_ACCT_WRITE);
575         r->req.aiocb = sdc->dma_writev(r->sector << BDRV_SECTOR_BITS, &r->qiov,
576                                        scsi_write_complete, r, r);
577     }
578 }
579 
580 /* Return a pointer to the data buffer.  */
581 static uint8_t *scsi_get_buf(SCSIRequest *req)
582 {
583     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
584 
585     return (uint8_t *)r->iov.iov_base;
586 }
587 
588 static int scsi_disk_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf)
589 {
590     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
591     int buflen = 0;
592     int start;
593 
594     if (req->cmd.buf[1] & 0x1) {
595         /* Vital product data */
596         uint8_t page_code = req->cmd.buf[2];
597 
598         outbuf[buflen++] = s->qdev.type & 0x1f;
599         outbuf[buflen++] = page_code ; // this page
600         outbuf[buflen++] = 0x00;
601         outbuf[buflen++] = 0x00;
602         start = buflen;
603 
604         switch (page_code) {
605         case 0x00: /* Supported page codes, mandatory */
606         {
607             DPRINTF("Inquiry EVPD[Supported pages] "
608                     "buffer size %zd\n", req->cmd.xfer);
609             outbuf[buflen++] = 0x00; // list of supported pages (this page)
610             if (s->serial) {
611                 outbuf[buflen++] = 0x80; // unit serial number
612             }
613             outbuf[buflen++] = 0x83; // device identification
614             if (s->qdev.type == TYPE_DISK) {
615                 outbuf[buflen++] = 0xb0; // block limits
616                 outbuf[buflen++] = 0xb1; /* block device characteristics */
617                 outbuf[buflen++] = 0xb2; // thin provisioning
618             }
619             break;
620         }
621         case 0x80: /* Device serial number, optional */
622         {
623             int l;
624 
625             if (!s->serial) {
626                 DPRINTF("Inquiry (EVPD[Serial number] not supported\n");
627                 return -1;
628             }
629 
630             l = strlen(s->serial);
631             if (l > 36) {
632                 l = 36;
633             }
634 
635             DPRINTF("Inquiry EVPD[Serial number] "
636                     "buffer size %zd\n", req->cmd.xfer);
637             memcpy(outbuf+buflen, s->serial, l);
638             buflen += l;
639             break;
640         }
641 
642         case 0x83: /* Device identification page, mandatory */
643         {
644             const char *str = s->serial ?: blk_name(s->qdev.conf.blk);
645             int max_len = s->serial ? 20 : 255 - 8;
646             int id_len = strlen(str);
647 
648             if (id_len > max_len) {
649                 id_len = max_len;
650             }
651             DPRINTF("Inquiry EVPD[Device identification] "
652                     "buffer size %zd\n", req->cmd.xfer);
653 
654             outbuf[buflen++] = 0x2; // ASCII
655             outbuf[buflen++] = 0;   // not officially assigned
656             outbuf[buflen++] = 0;   // reserved
657             outbuf[buflen++] = id_len; // length of data following
658             memcpy(outbuf+buflen, str, id_len);
659             buflen += id_len;
660 
661             if (s->qdev.wwn) {
662                 outbuf[buflen++] = 0x1; // Binary
663                 outbuf[buflen++] = 0x3; // NAA
664                 outbuf[buflen++] = 0;   // reserved
665                 outbuf[buflen++] = 8;
666                 stq_be_p(&outbuf[buflen], s->qdev.wwn);
667                 buflen += 8;
668             }
669 
670             if (s->qdev.port_wwn) {
671                 outbuf[buflen++] = 0x61; // SAS / Binary
672                 outbuf[buflen++] = 0x93; // PIV / Target port / NAA
673                 outbuf[buflen++] = 0;    // reserved
674                 outbuf[buflen++] = 8;
675                 stq_be_p(&outbuf[buflen], s->qdev.port_wwn);
676                 buflen += 8;
677             }
678 
679             if (s->port_index) {
680                 outbuf[buflen++] = 0x61; // SAS / Binary
681                 outbuf[buflen++] = 0x94; // PIV / Target port / relative target port
682                 outbuf[buflen++] = 0;    // reserved
683                 outbuf[buflen++] = 4;
684                 stw_be_p(&outbuf[buflen + 2], s->port_index);
685                 buflen += 4;
686             }
687             break;
688         }
689         case 0xb0: /* block limits */
690         {
691             unsigned int unmap_sectors =
692                     s->qdev.conf.discard_granularity / s->qdev.blocksize;
693             unsigned int min_io_size =
694                     s->qdev.conf.min_io_size / s->qdev.blocksize;
695             unsigned int opt_io_size =
696                     s->qdev.conf.opt_io_size / s->qdev.blocksize;
697             unsigned int max_unmap_sectors =
698                     s->max_unmap_size / s->qdev.blocksize;
699             unsigned int max_io_sectors =
700                     s->max_io_size / s->qdev.blocksize;
701 
702             if (s->qdev.type == TYPE_ROM) {
703                 DPRINTF("Inquiry (EVPD[%02X] not supported for CDROM\n",
704                         page_code);
705                 return -1;
706             }
707             if (s->qdev.type == TYPE_DISK) {
708                 int max_transfer_blk = blk_get_max_transfer(s->qdev.conf.blk);
709                 int max_io_sectors_blk =
710                     max_transfer_blk / s->qdev.blocksize;
711 
712                 max_io_sectors =
713                     MIN_NON_ZERO(max_io_sectors_blk, max_io_sectors);
714 
715                 /* min_io_size and opt_io_size can't be greater than
716                  * max_io_sectors */
717                 if (min_io_size) {
718                     min_io_size = MIN(min_io_size, max_io_sectors);
719                 }
720                 if (opt_io_size) {
721                     opt_io_size = MIN(opt_io_size, max_io_sectors);
722                 }
723             }
724             /* required VPD size with unmap support */
725             buflen = 0x40;
726             memset(outbuf + 4, 0, buflen - 4);
727 
728             outbuf[4] = 0x1; /* wsnz */
729 
730             /* optimal transfer length granularity */
731             outbuf[6] = (min_io_size >> 8) & 0xff;
732             outbuf[7] = min_io_size & 0xff;
733 
734             /* maximum transfer length */
735             outbuf[8] = (max_io_sectors >> 24) & 0xff;
736             outbuf[9] = (max_io_sectors >> 16) & 0xff;
737             outbuf[10] = (max_io_sectors >> 8) & 0xff;
738             outbuf[11] = max_io_sectors & 0xff;
739 
740             /* optimal transfer length */
741             outbuf[12] = (opt_io_size >> 24) & 0xff;
742             outbuf[13] = (opt_io_size >> 16) & 0xff;
743             outbuf[14] = (opt_io_size >> 8) & 0xff;
744             outbuf[15] = opt_io_size & 0xff;
745 
746             /* max unmap LBA count, default is 1GB */
747             outbuf[20] = (max_unmap_sectors >> 24) & 0xff;
748             outbuf[21] = (max_unmap_sectors >> 16) & 0xff;
749             outbuf[22] = (max_unmap_sectors >> 8) & 0xff;
750             outbuf[23] = max_unmap_sectors & 0xff;
751 
752             /* max unmap descriptors, 255 fit in 4 kb with an 8-byte header.  */
753             outbuf[24] = 0;
754             outbuf[25] = 0;
755             outbuf[26] = 0;
756             outbuf[27] = 255;
757 
758             /* optimal unmap granularity */
759             outbuf[28] = (unmap_sectors >> 24) & 0xff;
760             outbuf[29] = (unmap_sectors >> 16) & 0xff;
761             outbuf[30] = (unmap_sectors >> 8) & 0xff;
762             outbuf[31] = unmap_sectors & 0xff;
763 
764             /* max write same size */
765             outbuf[36] = 0;
766             outbuf[37] = 0;
767             outbuf[38] = 0;
768             outbuf[39] = 0;
769 
770             outbuf[40] = (max_io_sectors >> 24) & 0xff;
771             outbuf[41] = (max_io_sectors >> 16) & 0xff;
772             outbuf[42] = (max_io_sectors >> 8) & 0xff;
773             outbuf[43] = max_io_sectors & 0xff;
774             break;
775         }
776         case 0xb1: /* block device characteristics */
777         {
778             buflen = 8;
779             outbuf[4] = (s->rotation_rate >> 8) & 0xff;
780             outbuf[5] = s->rotation_rate & 0xff;
781             outbuf[6] = 0;
782             outbuf[7] = 0;
783             break;
784         }
785         case 0xb2: /* thin provisioning */
786         {
787             buflen = 8;
788             outbuf[4] = 0;
789             outbuf[5] = 0xe0; /* unmap & write_same 10/16 all supported */
790             outbuf[6] = s->qdev.conf.discard_granularity ? 2 : 1;
791             outbuf[7] = 0;
792             break;
793         }
794         default:
795             return -1;
796         }
797         /* done with EVPD */
798         assert(buflen - start <= 255);
799         outbuf[start - 1] = buflen - start;
800         return buflen;
801     }
802 
803     /* Standard INQUIRY data */
804     if (req->cmd.buf[2] != 0) {
805         return -1;
806     }
807 
808     /* PAGE CODE == 0 */
809     buflen = req->cmd.xfer;
810     if (buflen > SCSI_MAX_INQUIRY_LEN) {
811         buflen = SCSI_MAX_INQUIRY_LEN;
812     }
813 
814     outbuf[0] = s->qdev.type & 0x1f;
815     outbuf[1] = (s->features & (1 << SCSI_DISK_F_REMOVABLE)) ? 0x80 : 0;
816 
817     strpadcpy((char *) &outbuf[16], 16, s->product, ' ');
818     strpadcpy((char *) &outbuf[8], 8, s->vendor, ' ');
819 
820     memset(&outbuf[32], 0, 4);
821     memcpy(&outbuf[32], s->version, MIN(4, strlen(s->version)));
822     /*
823      * We claim conformance to SPC-3, which is required for guests
824      * to ask for modern features like READ CAPACITY(16) or the
825      * block characteristics VPD page by default.  Not all of SPC-3
826      * is actually implemented, but we're good enough.
827      */
828     outbuf[2] = s->qdev.default_scsi_version;
829     outbuf[3] = 2 | 0x10; /* Format 2, HiSup */
830 
831     if (buflen > 36) {
832         outbuf[4] = buflen - 5; /* Additional Length = (Len - 1) - 4 */
833     } else {
834         /* If the allocation length of CDB is too small,
835                the additional length is not adjusted */
836         outbuf[4] = 36 - 5;
837     }
838 
839     /* Sync data transfer and TCQ.  */
840     outbuf[7] = 0x10 | (req->bus->info->tcq ? 0x02 : 0);
841     return buflen;
842 }
843 
844 static inline bool media_is_dvd(SCSIDiskState *s)
845 {
846     uint64_t nb_sectors;
847     if (s->qdev.type != TYPE_ROM) {
848         return false;
849     }
850     if (!blk_is_available(s->qdev.conf.blk)) {
851         return false;
852     }
853     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
854     return nb_sectors > CD_MAX_SECTORS;
855 }
856 
857 static inline bool media_is_cd(SCSIDiskState *s)
858 {
859     uint64_t nb_sectors;
860     if (s->qdev.type != TYPE_ROM) {
861         return false;
862     }
863     if (!blk_is_available(s->qdev.conf.blk)) {
864         return false;
865     }
866     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
867     return nb_sectors <= CD_MAX_SECTORS;
868 }
869 
870 static int scsi_read_disc_information(SCSIDiskState *s, SCSIDiskReq *r,
871                                       uint8_t *outbuf)
872 {
873     uint8_t type = r->req.cmd.buf[1] & 7;
874 
875     if (s->qdev.type != TYPE_ROM) {
876         return -1;
877     }
878 
879     /* Types 1/2 are only defined for Blu-Ray.  */
880     if (type != 0) {
881         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
882         return -1;
883     }
884 
885     memset(outbuf, 0, 34);
886     outbuf[1] = 32;
887     outbuf[2] = 0xe; /* last session complete, disc finalized */
888     outbuf[3] = 1;   /* first track on disc */
889     outbuf[4] = 1;   /* # of sessions */
890     outbuf[5] = 1;   /* first track of last session */
891     outbuf[6] = 1;   /* last track of last session */
892     outbuf[7] = 0x20; /* unrestricted use */
893     outbuf[8] = 0x00; /* CD-ROM or DVD-ROM */
894     /* 9-10-11: most significant byte corresponding bytes 4-5-6 */
895     /* 12-23: not meaningful for CD-ROM or DVD-ROM */
896     /* 24-31: disc bar code */
897     /* 32: disc application code */
898     /* 33: number of OPC tables */
899 
900     return 34;
901 }
902 
903 static int scsi_read_dvd_structure(SCSIDiskState *s, SCSIDiskReq *r,
904                                    uint8_t *outbuf)
905 {
906     static const int rds_caps_size[5] = {
907         [0] = 2048 + 4,
908         [1] = 4 + 4,
909         [3] = 188 + 4,
910         [4] = 2048 + 4,
911     };
912 
913     uint8_t media = r->req.cmd.buf[1];
914     uint8_t layer = r->req.cmd.buf[6];
915     uint8_t format = r->req.cmd.buf[7];
916     int size = -1;
917 
918     if (s->qdev.type != TYPE_ROM) {
919         return -1;
920     }
921     if (media != 0) {
922         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
923         return -1;
924     }
925 
926     if (format != 0xff) {
927         if (!blk_is_available(s->qdev.conf.blk)) {
928             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
929             return -1;
930         }
931         if (media_is_cd(s)) {
932             scsi_check_condition(r, SENSE_CODE(INCOMPATIBLE_FORMAT));
933             return -1;
934         }
935         if (format >= ARRAY_SIZE(rds_caps_size)) {
936             return -1;
937         }
938         size = rds_caps_size[format];
939         memset(outbuf, 0, size);
940     }
941 
942     switch (format) {
943     case 0x00: {
944         /* Physical format information */
945         uint64_t nb_sectors;
946         if (layer != 0) {
947             goto fail;
948         }
949         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
950 
951         outbuf[4] = 1;   /* DVD-ROM, part version 1 */
952         outbuf[5] = 0xf; /* 120mm disc, minimum rate unspecified */
953         outbuf[6] = 1;   /* one layer, read-only (per MMC-2 spec) */
954         outbuf[7] = 0;   /* default densities */
955 
956         stl_be_p(&outbuf[12], (nb_sectors >> 2) - 1); /* end sector */
957         stl_be_p(&outbuf[16], (nb_sectors >> 2) - 1); /* l0 end sector */
958         break;
959     }
960 
961     case 0x01: /* DVD copyright information, all zeros */
962         break;
963 
964     case 0x03: /* BCA information - invalid field for no BCA info */
965         return -1;
966 
967     case 0x04: /* DVD disc manufacturing information, all zeros */
968         break;
969 
970     case 0xff: { /* List capabilities */
971         int i;
972         size = 4;
973         for (i = 0; i < ARRAY_SIZE(rds_caps_size); i++) {
974             if (!rds_caps_size[i]) {
975                 continue;
976             }
977             outbuf[size] = i;
978             outbuf[size + 1] = 0x40; /* Not writable, readable */
979             stw_be_p(&outbuf[size + 2], rds_caps_size[i]);
980             size += 4;
981         }
982         break;
983      }
984 
985     default:
986         return -1;
987     }
988 
989     /* Size of buffer, not including 2 byte size field */
990     stw_be_p(outbuf, size - 2);
991     return size;
992 
993 fail:
994     return -1;
995 }
996 
997 static int scsi_event_status_media(SCSIDiskState *s, uint8_t *outbuf)
998 {
999     uint8_t event_code, media_status;
1000 
1001     media_status = 0;
1002     if (s->tray_open) {
1003         media_status = MS_TRAY_OPEN;
1004     } else if (blk_is_inserted(s->qdev.conf.blk)) {
1005         media_status = MS_MEDIA_PRESENT;
1006     }
1007 
1008     /* Event notification descriptor */
1009     event_code = MEC_NO_CHANGE;
1010     if (media_status != MS_TRAY_OPEN) {
1011         if (s->media_event) {
1012             event_code = MEC_NEW_MEDIA;
1013             s->media_event = false;
1014         } else if (s->eject_request) {
1015             event_code = MEC_EJECT_REQUESTED;
1016             s->eject_request = false;
1017         }
1018     }
1019 
1020     outbuf[0] = event_code;
1021     outbuf[1] = media_status;
1022 
1023     /* These fields are reserved, just clear them. */
1024     outbuf[2] = 0;
1025     outbuf[3] = 0;
1026     return 4;
1027 }
1028 
1029 static int scsi_get_event_status_notification(SCSIDiskState *s, SCSIDiskReq *r,
1030                                               uint8_t *outbuf)
1031 {
1032     int size;
1033     uint8_t *buf = r->req.cmd.buf;
1034     uint8_t notification_class_request = buf[4];
1035     if (s->qdev.type != TYPE_ROM) {
1036         return -1;
1037     }
1038     if ((buf[1] & 1) == 0) {
1039         /* asynchronous */
1040         return -1;
1041     }
1042 
1043     size = 4;
1044     outbuf[0] = outbuf[1] = 0;
1045     outbuf[3] = 1 << GESN_MEDIA; /* supported events */
1046     if (notification_class_request & (1 << GESN_MEDIA)) {
1047         outbuf[2] = GESN_MEDIA;
1048         size += scsi_event_status_media(s, &outbuf[size]);
1049     } else {
1050         outbuf[2] = 0x80;
1051     }
1052     stw_be_p(outbuf, size - 4);
1053     return size;
1054 }
1055 
1056 static int scsi_get_configuration(SCSIDiskState *s, uint8_t *outbuf)
1057 {
1058     int current;
1059 
1060     if (s->qdev.type != TYPE_ROM) {
1061         return -1;
1062     }
1063 
1064     if (media_is_dvd(s)) {
1065         current = MMC_PROFILE_DVD_ROM;
1066     } else if (media_is_cd(s)) {
1067         current = MMC_PROFILE_CD_ROM;
1068     } else {
1069         current = MMC_PROFILE_NONE;
1070     }
1071 
1072     memset(outbuf, 0, 40);
1073     stl_be_p(&outbuf[0], 36); /* Bytes after the data length field */
1074     stw_be_p(&outbuf[6], current);
1075     /* outbuf[8] - outbuf[19]: Feature 0 - Profile list */
1076     outbuf[10] = 0x03; /* persistent, current */
1077     outbuf[11] = 8; /* two profiles */
1078     stw_be_p(&outbuf[12], MMC_PROFILE_DVD_ROM);
1079     outbuf[14] = (current == MMC_PROFILE_DVD_ROM);
1080     stw_be_p(&outbuf[16], MMC_PROFILE_CD_ROM);
1081     outbuf[18] = (current == MMC_PROFILE_CD_ROM);
1082     /* outbuf[20] - outbuf[31]: Feature 1 - Core feature */
1083     stw_be_p(&outbuf[20], 1);
1084     outbuf[22] = 0x08 | 0x03; /* version 2, persistent, current */
1085     outbuf[23] = 8;
1086     stl_be_p(&outbuf[24], 1); /* SCSI */
1087     outbuf[28] = 1; /* DBE = 1, mandatory */
1088     /* outbuf[32] - outbuf[39]: Feature 3 - Removable media feature */
1089     stw_be_p(&outbuf[32], 3);
1090     outbuf[34] = 0x08 | 0x03; /* version 2, persistent, current */
1091     outbuf[35] = 4;
1092     outbuf[36] = 0x39; /* tray, load=1, eject=1, unlocked at powerup, lock=1 */
1093     /* TODO: Random readable, CD read, DVD read, drive serial number,
1094        power management */
1095     return 40;
1096 }
1097 
1098 static int scsi_emulate_mechanism_status(SCSIDiskState *s, uint8_t *outbuf)
1099 {
1100     if (s->qdev.type != TYPE_ROM) {
1101         return -1;
1102     }
1103     memset(outbuf, 0, 8);
1104     outbuf[5] = 1; /* CD-ROM */
1105     return 8;
1106 }
1107 
1108 static int mode_sense_page(SCSIDiskState *s, int page, uint8_t **p_outbuf,
1109                            int page_control)
1110 {
1111     static const int mode_sense_valid[0x3f] = {
1112         [MODE_PAGE_HD_GEOMETRY]            = (1 << TYPE_DISK),
1113         [MODE_PAGE_FLEXIBLE_DISK_GEOMETRY] = (1 << TYPE_DISK),
1114         [MODE_PAGE_CACHING]                = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1115         [MODE_PAGE_R_W_ERROR]              = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1116         [MODE_PAGE_AUDIO_CTL]              = (1 << TYPE_ROM),
1117         [MODE_PAGE_CAPABILITIES]           = (1 << TYPE_ROM),
1118     };
1119 
1120     uint8_t *p = *p_outbuf + 2;
1121     int length;
1122 
1123     if ((mode_sense_valid[page] & (1 << s->qdev.type)) == 0) {
1124         return -1;
1125     }
1126 
1127     /*
1128      * If Changeable Values are requested, a mask denoting those mode parameters
1129      * that are changeable shall be returned. As we currently don't support
1130      * parameter changes via MODE_SELECT all bits are returned set to zero.
1131      * The buffer was already menset to zero by the caller of this function.
1132      *
1133      * The offsets here are off by two compared to the descriptions in the
1134      * SCSI specs, because those include a 2-byte header.  This is unfortunate,
1135      * but it is done so that offsets are consistent within our implementation
1136      * of MODE SENSE and MODE SELECT.  MODE SELECT has to deal with both
1137      * 2-byte and 4-byte headers.
1138      */
1139     switch (page) {
1140     case MODE_PAGE_HD_GEOMETRY:
1141         length = 0x16;
1142         if (page_control == 1) { /* Changeable Values */
1143             break;
1144         }
1145         /* if a geometry hint is available, use it */
1146         p[0] = (s->qdev.conf.cyls >> 16) & 0xff;
1147         p[1] = (s->qdev.conf.cyls >> 8) & 0xff;
1148         p[2] = s->qdev.conf.cyls & 0xff;
1149         p[3] = s->qdev.conf.heads & 0xff;
1150         /* Write precomp start cylinder, disabled */
1151         p[4] = (s->qdev.conf.cyls >> 16) & 0xff;
1152         p[5] = (s->qdev.conf.cyls >> 8) & 0xff;
1153         p[6] = s->qdev.conf.cyls & 0xff;
1154         /* Reduced current start cylinder, disabled */
1155         p[7] = (s->qdev.conf.cyls >> 16) & 0xff;
1156         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1157         p[9] = s->qdev.conf.cyls & 0xff;
1158         /* Device step rate [ns], 200ns */
1159         p[10] = 0;
1160         p[11] = 200;
1161         /* Landing zone cylinder */
1162         p[12] = 0xff;
1163         p[13] =  0xff;
1164         p[14] = 0xff;
1165         /* Medium rotation rate [rpm], 5400 rpm */
1166         p[18] = (5400 >> 8) & 0xff;
1167         p[19] = 5400 & 0xff;
1168         break;
1169 
1170     case MODE_PAGE_FLEXIBLE_DISK_GEOMETRY:
1171         length = 0x1e;
1172         if (page_control == 1) { /* Changeable Values */
1173             break;
1174         }
1175         /* Transfer rate [kbit/s], 5Mbit/s */
1176         p[0] = 5000 >> 8;
1177         p[1] = 5000 & 0xff;
1178         /* if a geometry hint is available, use it */
1179         p[2] = s->qdev.conf.heads & 0xff;
1180         p[3] = s->qdev.conf.secs & 0xff;
1181         p[4] = s->qdev.blocksize >> 8;
1182         p[6] = (s->qdev.conf.cyls >> 8) & 0xff;
1183         p[7] = s->qdev.conf.cyls & 0xff;
1184         /* Write precomp start cylinder, disabled */
1185         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1186         p[9] = s->qdev.conf.cyls & 0xff;
1187         /* Reduced current start cylinder, disabled */
1188         p[10] = (s->qdev.conf.cyls >> 8) & 0xff;
1189         p[11] = s->qdev.conf.cyls & 0xff;
1190         /* Device step rate [100us], 100us */
1191         p[12] = 0;
1192         p[13] = 1;
1193         /* Device step pulse width [us], 1us */
1194         p[14] = 1;
1195         /* Device head settle delay [100us], 100us */
1196         p[15] = 0;
1197         p[16] = 1;
1198         /* Motor on delay [0.1s], 0.1s */
1199         p[17] = 1;
1200         /* Motor off delay [0.1s], 0.1s */
1201         p[18] = 1;
1202         /* Medium rotation rate [rpm], 5400 rpm */
1203         p[26] = (5400 >> 8) & 0xff;
1204         p[27] = 5400 & 0xff;
1205         break;
1206 
1207     case MODE_PAGE_CACHING:
1208         length = 0x12;
1209         if (page_control == 1 || /* Changeable Values */
1210             blk_enable_write_cache(s->qdev.conf.blk)) {
1211             p[0] = 4; /* WCE */
1212         }
1213         break;
1214 
1215     case MODE_PAGE_R_W_ERROR:
1216         length = 10;
1217         if (page_control == 1) { /* Changeable Values */
1218             break;
1219         }
1220         p[0] = 0x80; /* Automatic Write Reallocation Enabled */
1221         if (s->qdev.type == TYPE_ROM) {
1222             p[1] = 0x20; /* Read Retry Count */
1223         }
1224         break;
1225 
1226     case MODE_PAGE_AUDIO_CTL:
1227         length = 14;
1228         break;
1229 
1230     case MODE_PAGE_CAPABILITIES:
1231         length = 0x14;
1232         if (page_control == 1) { /* Changeable Values */
1233             break;
1234         }
1235 
1236         p[0] = 0x3b; /* CD-R & CD-RW read */
1237         p[1] = 0; /* Writing not supported */
1238         p[2] = 0x7f; /* Audio, composite, digital out,
1239                         mode 2 form 1&2, multi session */
1240         p[3] = 0xff; /* CD DA, DA accurate, RW supported,
1241                         RW corrected, C2 errors, ISRC,
1242                         UPC, Bar code */
1243         p[4] = 0x2d | (s->tray_locked ? 2 : 0);
1244         /* Locking supported, jumper present, eject, tray */
1245         p[5] = 0; /* no volume & mute control, no
1246                      changer */
1247         p[6] = (50 * 176) >> 8; /* 50x read speed */
1248         p[7] = (50 * 176) & 0xff;
1249         p[8] = 2 >> 8; /* Two volume levels */
1250         p[9] = 2 & 0xff;
1251         p[10] = 2048 >> 8; /* 2M buffer */
1252         p[11] = 2048 & 0xff;
1253         p[12] = (16 * 176) >> 8; /* 16x read speed current */
1254         p[13] = (16 * 176) & 0xff;
1255         p[16] = (16 * 176) >> 8; /* 16x write speed */
1256         p[17] = (16 * 176) & 0xff;
1257         p[18] = (16 * 176) >> 8; /* 16x write speed current */
1258         p[19] = (16 * 176) & 0xff;
1259         break;
1260 
1261     default:
1262         return -1;
1263     }
1264 
1265     assert(length < 256);
1266     (*p_outbuf)[0] = page;
1267     (*p_outbuf)[1] = length;
1268     *p_outbuf += length + 2;
1269     return length + 2;
1270 }
1271 
1272 static int scsi_disk_emulate_mode_sense(SCSIDiskReq *r, uint8_t *outbuf)
1273 {
1274     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1275     uint64_t nb_sectors;
1276     bool dbd;
1277     int page, buflen, ret, page_control;
1278     uint8_t *p;
1279     uint8_t dev_specific_param;
1280 
1281     dbd = (r->req.cmd.buf[1] & 0x8) != 0;
1282     page = r->req.cmd.buf[2] & 0x3f;
1283     page_control = (r->req.cmd.buf[2] & 0xc0) >> 6;
1284     DPRINTF("Mode Sense(%d) (page %d, xfer %zd, page_control %d)\n",
1285         (r->req.cmd.buf[0] == MODE_SENSE) ? 6 : 10, page, r->req.cmd.xfer, page_control);
1286     memset(outbuf, 0, r->req.cmd.xfer);
1287     p = outbuf;
1288 
1289     if (s->qdev.type == TYPE_DISK) {
1290         dev_specific_param = s->features & (1 << SCSI_DISK_F_DPOFUA) ? 0x10 : 0;
1291         if (blk_is_read_only(s->qdev.conf.blk)) {
1292             dev_specific_param |= 0x80; /* Readonly.  */
1293         }
1294     } else {
1295         /* MMC prescribes that CD/DVD drives have no block descriptors,
1296          * and defines no device-specific parameter.  */
1297         dev_specific_param = 0x00;
1298         dbd = true;
1299     }
1300 
1301     if (r->req.cmd.buf[0] == MODE_SENSE) {
1302         p[1] = 0; /* Default media type.  */
1303         p[2] = dev_specific_param;
1304         p[3] = 0; /* Block descriptor length.  */
1305         p += 4;
1306     } else { /* MODE_SENSE_10 */
1307         p[2] = 0; /* Default media type.  */
1308         p[3] = dev_specific_param;
1309         p[6] = p[7] = 0; /* Block descriptor length.  */
1310         p += 8;
1311     }
1312 
1313     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1314     if (!dbd && nb_sectors) {
1315         if (r->req.cmd.buf[0] == MODE_SENSE) {
1316             outbuf[3] = 8; /* Block descriptor length  */
1317         } else { /* MODE_SENSE_10 */
1318             outbuf[7] = 8; /* Block descriptor length  */
1319         }
1320         nb_sectors /= (s->qdev.blocksize / 512);
1321         if (nb_sectors > 0xffffff) {
1322             nb_sectors = 0;
1323         }
1324         p[0] = 0; /* media density code */
1325         p[1] = (nb_sectors >> 16) & 0xff;
1326         p[2] = (nb_sectors >> 8) & 0xff;
1327         p[3] = nb_sectors & 0xff;
1328         p[4] = 0; /* reserved */
1329         p[5] = 0; /* bytes 5-7 are the sector size in bytes */
1330         p[6] = s->qdev.blocksize >> 8;
1331         p[7] = 0;
1332         p += 8;
1333     }
1334 
1335     if (page_control == 3) {
1336         /* Saved Values */
1337         scsi_check_condition(r, SENSE_CODE(SAVING_PARAMS_NOT_SUPPORTED));
1338         return -1;
1339     }
1340 
1341     if (page == 0x3f) {
1342         for (page = 0; page <= 0x3e; page++) {
1343             mode_sense_page(s, page, &p, page_control);
1344         }
1345     } else {
1346         ret = mode_sense_page(s, page, &p, page_control);
1347         if (ret == -1) {
1348             return -1;
1349         }
1350     }
1351 
1352     buflen = p - outbuf;
1353     /*
1354      * The mode data length field specifies the length in bytes of the
1355      * following data that is available to be transferred. The mode data
1356      * length does not include itself.
1357      */
1358     if (r->req.cmd.buf[0] == MODE_SENSE) {
1359         outbuf[0] = buflen - 1;
1360     } else { /* MODE_SENSE_10 */
1361         outbuf[0] = ((buflen - 2) >> 8) & 0xff;
1362         outbuf[1] = (buflen - 2) & 0xff;
1363     }
1364     return buflen;
1365 }
1366 
1367 static int scsi_disk_emulate_read_toc(SCSIRequest *req, uint8_t *outbuf)
1368 {
1369     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1370     int start_track, format, msf, toclen;
1371     uint64_t nb_sectors;
1372 
1373     msf = req->cmd.buf[1] & 2;
1374     format = req->cmd.buf[2] & 0xf;
1375     start_track = req->cmd.buf[6];
1376     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1377     DPRINTF("Read TOC (track %d format %d msf %d)\n", start_track, format, msf >> 1);
1378     nb_sectors /= s->qdev.blocksize / 512;
1379     switch (format) {
1380     case 0:
1381         toclen = cdrom_read_toc(nb_sectors, outbuf, msf, start_track);
1382         break;
1383     case 1:
1384         /* multi session : only a single session defined */
1385         toclen = 12;
1386         memset(outbuf, 0, 12);
1387         outbuf[1] = 0x0a;
1388         outbuf[2] = 0x01;
1389         outbuf[3] = 0x01;
1390         break;
1391     case 2:
1392         toclen = cdrom_read_toc_raw(nb_sectors, outbuf, msf, start_track);
1393         break;
1394     default:
1395         return -1;
1396     }
1397     return toclen;
1398 }
1399 
1400 static int scsi_disk_emulate_start_stop(SCSIDiskReq *r)
1401 {
1402     SCSIRequest *req = &r->req;
1403     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1404     bool start = req->cmd.buf[4] & 1;
1405     bool loej = req->cmd.buf[4] & 2; /* load on start, eject on !start */
1406     int pwrcnd = req->cmd.buf[4] & 0xf0;
1407 
1408     if (pwrcnd) {
1409         /* eject/load only happens for power condition == 0 */
1410         return 0;
1411     }
1412 
1413     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) && loej) {
1414         if (!start && !s->tray_open && s->tray_locked) {
1415             scsi_check_condition(r,
1416                                  blk_is_inserted(s->qdev.conf.blk)
1417                                  ? SENSE_CODE(ILLEGAL_REQ_REMOVAL_PREVENTED)
1418                                  : SENSE_CODE(NOT_READY_REMOVAL_PREVENTED));
1419             return -1;
1420         }
1421 
1422         if (s->tray_open != !start) {
1423             blk_eject(s->qdev.conf.blk, !start);
1424             s->tray_open = !start;
1425         }
1426     }
1427     return 0;
1428 }
1429 
1430 static void scsi_disk_emulate_read_data(SCSIRequest *req)
1431 {
1432     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1433     int buflen = r->iov.iov_len;
1434 
1435     if (buflen) {
1436         DPRINTF("Read buf_len=%d\n", buflen);
1437         r->iov.iov_len = 0;
1438         r->started = true;
1439         scsi_req_data(&r->req, buflen);
1440         return;
1441     }
1442 
1443     /* This also clears the sense buffer for REQUEST SENSE.  */
1444     scsi_req_complete(&r->req, GOOD);
1445 }
1446 
1447 static int scsi_disk_check_mode_select(SCSIDiskState *s, int page,
1448                                        uint8_t *inbuf, int inlen)
1449 {
1450     uint8_t mode_current[SCSI_MAX_MODE_LEN];
1451     uint8_t mode_changeable[SCSI_MAX_MODE_LEN];
1452     uint8_t *p;
1453     int len, expected_len, changeable_len, i;
1454 
1455     /* The input buffer does not include the page header, so it is
1456      * off by 2 bytes.
1457      */
1458     expected_len = inlen + 2;
1459     if (expected_len > SCSI_MAX_MODE_LEN) {
1460         return -1;
1461     }
1462 
1463     p = mode_current;
1464     memset(mode_current, 0, inlen + 2);
1465     len = mode_sense_page(s, page, &p, 0);
1466     if (len < 0 || len != expected_len) {
1467         return -1;
1468     }
1469 
1470     p = mode_changeable;
1471     memset(mode_changeable, 0, inlen + 2);
1472     changeable_len = mode_sense_page(s, page, &p, 1);
1473     assert(changeable_len == len);
1474 
1475     /* Check that unchangeable bits are the same as what MODE SENSE
1476      * would return.
1477      */
1478     for (i = 2; i < len; i++) {
1479         if (((mode_current[i] ^ inbuf[i - 2]) & ~mode_changeable[i]) != 0) {
1480             return -1;
1481         }
1482     }
1483     return 0;
1484 }
1485 
1486 static void scsi_disk_apply_mode_select(SCSIDiskState *s, int page, uint8_t *p)
1487 {
1488     switch (page) {
1489     case MODE_PAGE_CACHING:
1490         blk_set_enable_write_cache(s->qdev.conf.blk, (p[0] & 4) != 0);
1491         break;
1492 
1493     default:
1494         break;
1495     }
1496 }
1497 
1498 static int mode_select_pages(SCSIDiskReq *r, uint8_t *p, int len, bool change)
1499 {
1500     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1501 
1502     while (len > 0) {
1503         int page, subpage, page_len;
1504 
1505         /* Parse both possible formats for the mode page headers.  */
1506         page = p[0] & 0x3f;
1507         if (p[0] & 0x40) {
1508             if (len < 4) {
1509                 goto invalid_param_len;
1510             }
1511             subpage = p[1];
1512             page_len = lduw_be_p(&p[2]);
1513             p += 4;
1514             len -= 4;
1515         } else {
1516             if (len < 2) {
1517                 goto invalid_param_len;
1518             }
1519             subpage = 0;
1520             page_len = p[1];
1521             p += 2;
1522             len -= 2;
1523         }
1524 
1525         if (subpage) {
1526             goto invalid_param;
1527         }
1528         if (page_len > len) {
1529             goto invalid_param_len;
1530         }
1531 
1532         if (!change) {
1533             if (scsi_disk_check_mode_select(s, page, p, page_len) < 0) {
1534                 goto invalid_param;
1535             }
1536         } else {
1537             scsi_disk_apply_mode_select(s, page, p);
1538         }
1539 
1540         p += page_len;
1541         len -= page_len;
1542     }
1543     return 0;
1544 
1545 invalid_param:
1546     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1547     return -1;
1548 
1549 invalid_param_len:
1550     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1551     return -1;
1552 }
1553 
1554 static void scsi_disk_emulate_mode_select(SCSIDiskReq *r, uint8_t *inbuf)
1555 {
1556     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1557     uint8_t *p = inbuf;
1558     int cmd = r->req.cmd.buf[0];
1559     int len = r->req.cmd.xfer;
1560     int hdr_len = (cmd == MODE_SELECT ? 4 : 8);
1561     int bd_len;
1562     int pass;
1563 
1564     /* We only support PF=1, SP=0.  */
1565     if ((r->req.cmd.buf[1] & 0x11) != 0x10) {
1566         goto invalid_field;
1567     }
1568 
1569     if (len < hdr_len) {
1570         goto invalid_param_len;
1571     }
1572 
1573     bd_len = (cmd == MODE_SELECT ? p[3] : lduw_be_p(&p[6]));
1574     len -= hdr_len;
1575     p += hdr_len;
1576     if (len < bd_len) {
1577         goto invalid_param_len;
1578     }
1579     if (bd_len != 0 && bd_len != 8) {
1580         goto invalid_param;
1581     }
1582 
1583     len -= bd_len;
1584     p += bd_len;
1585 
1586     /* Ensure no change is made if there is an error!  */
1587     for (pass = 0; pass < 2; pass++) {
1588         if (mode_select_pages(r, p, len, pass == 1) < 0) {
1589             assert(pass == 0);
1590             return;
1591         }
1592     }
1593     if (!blk_enable_write_cache(s->qdev.conf.blk)) {
1594         /* The request is used as the AIO opaque value, so add a ref.  */
1595         scsi_req_ref(&r->req);
1596         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
1597                          BLOCK_ACCT_FLUSH);
1598         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
1599         return;
1600     }
1601 
1602     scsi_req_complete(&r->req, GOOD);
1603     return;
1604 
1605 invalid_param:
1606     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1607     return;
1608 
1609 invalid_param_len:
1610     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1611     return;
1612 
1613 invalid_field:
1614     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1615 }
1616 
1617 static inline bool check_lba_range(SCSIDiskState *s,
1618                                    uint64_t sector_num, uint32_t nb_sectors)
1619 {
1620     /*
1621      * The first line tests that no overflow happens when computing the last
1622      * sector.  The second line tests that the last accessed sector is in
1623      * range.
1624      *
1625      * Careful, the computations should not underflow for nb_sectors == 0,
1626      * and a 0-block read to the first LBA beyond the end of device is
1627      * valid.
1628      */
1629     return (sector_num <= sector_num + nb_sectors &&
1630             sector_num + nb_sectors <= s->qdev.max_lba + 1);
1631 }
1632 
1633 typedef struct UnmapCBData {
1634     SCSIDiskReq *r;
1635     uint8_t *inbuf;
1636     int count;
1637 } UnmapCBData;
1638 
1639 static void scsi_unmap_complete(void *opaque, int ret);
1640 
1641 static void scsi_unmap_complete_noio(UnmapCBData *data, int ret)
1642 {
1643     SCSIDiskReq *r = data->r;
1644     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1645     uint64_t sector_num;
1646     uint32_t nb_sectors;
1647 
1648     assert(r->req.aiocb == NULL);
1649     if (scsi_disk_req_check_error(r, ret, false)) {
1650         goto done;
1651     }
1652 
1653     if (data->count > 0) {
1654         sector_num = ldq_be_p(&data->inbuf[0]);
1655         nb_sectors = ldl_be_p(&data->inbuf[8]) & 0xffffffffULL;
1656         if (!check_lba_range(s, sector_num, nb_sectors)) {
1657             scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1658             goto done;
1659         }
1660 
1661         r->req.aiocb = blk_aio_pdiscard(s->qdev.conf.blk,
1662                                         sector_num * s->qdev.blocksize,
1663                                         nb_sectors * s->qdev.blocksize,
1664                                         scsi_unmap_complete, data);
1665         data->count--;
1666         data->inbuf += 16;
1667         return;
1668     }
1669 
1670     scsi_req_complete(&r->req, GOOD);
1671 
1672 done:
1673     scsi_req_unref(&r->req);
1674     g_free(data);
1675 }
1676 
1677 static void scsi_unmap_complete(void *opaque, int ret)
1678 {
1679     UnmapCBData *data = opaque;
1680     SCSIDiskReq *r = data->r;
1681     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1682 
1683     assert(r->req.aiocb != NULL);
1684     r->req.aiocb = NULL;
1685 
1686     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
1687     scsi_unmap_complete_noio(data, ret);
1688     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1689 }
1690 
1691 static void scsi_disk_emulate_unmap(SCSIDiskReq *r, uint8_t *inbuf)
1692 {
1693     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1694     uint8_t *p = inbuf;
1695     int len = r->req.cmd.xfer;
1696     UnmapCBData *data;
1697 
1698     /* Reject ANCHOR=1.  */
1699     if (r->req.cmd.buf[1] & 0x1) {
1700         goto invalid_field;
1701     }
1702 
1703     if (len < 8) {
1704         goto invalid_param_len;
1705     }
1706     if (len < lduw_be_p(&p[0]) + 2) {
1707         goto invalid_param_len;
1708     }
1709     if (len < lduw_be_p(&p[2]) + 8) {
1710         goto invalid_param_len;
1711     }
1712     if (lduw_be_p(&p[2]) & 15) {
1713         goto invalid_param_len;
1714     }
1715 
1716     if (blk_is_read_only(s->qdev.conf.blk)) {
1717         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1718         return;
1719     }
1720 
1721     data = g_new0(UnmapCBData, 1);
1722     data->r = r;
1723     data->inbuf = &p[8];
1724     data->count = lduw_be_p(&p[2]) >> 4;
1725 
1726     /* The matching unref is in scsi_unmap_complete, before data is freed.  */
1727     scsi_req_ref(&r->req);
1728     scsi_unmap_complete_noio(data, 0);
1729     return;
1730 
1731 invalid_param_len:
1732     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1733     return;
1734 
1735 invalid_field:
1736     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1737 }
1738 
1739 typedef struct WriteSameCBData {
1740     SCSIDiskReq *r;
1741     int64_t sector;
1742     int nb_sectors;
1743     QEMUIOVector qiov;
1744     struct iovec iov;
1745 } WriteSameCBData;
1746 
1747 static void scsi_write_same_complete(void *opaque, int ret)
1748 {
1749     WriteSameCBData *data = opaque;
1750     SCSIDiskReq *r = data->r;
1751     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1752 
1753     assert(r->req.aiocb != NULL);
1754     r->req.aiocb = NULL;
1755     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
1756     if (scsi_disk_req_check_error(r, ret, true)) {
1757         goto done;
1758     }
1759 
1760     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1761 
1762     data->nb_sectors -= data->iov.iov_len / 512;
1763     data->sector += data->iov.iov_len / 512;
1764     data->iov.iov_len = MIN(data->nb_sectors * 512, data->iov.iov_len);
1765     if (data->iov.iov_len) {
1766         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1767                          data->iov.iov_len, BLOCK_ACCT_WRITE);
1768         /* Reinitialize qiov, to handle unaligned WRITE SAME request
1769          * where final qiov may need smaller size */
1770         qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1771         r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1772                                        data->sector << BDRV_SECTOR_BITS,
1773                                        &data->qiov, 0,
1774                                        scsi_write_same_complete, data);
1775         aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1776         return;
1777     }
1778 
1779     scsi_req_complete(&r->req, GOOD);
1780 
1781 done:
1782     scsi_req_unref(&r->req);
1783     qemu_vfree(data->iov.iov_base);
1784     g_free(data);
1785     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1786 }
1787 
1788 static void scsi_disk_emulate_write_same(SCSIDiskReq *r, uint8_t *inbuf)
1789 {
1790     SCSIRequest *req = &r->req;
1791     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1792     uint32_t nb_sectors = scsi_data_cdb_xfer(r->req.cmd.buf);
1793     WriteSameCBData *data;
1794     uint8_t *buf;
1795     int i;
1796 
1797     /* Fail if PBDATA=1 or LBDATA=1 or ANCHOR=1.  */
1798     if (nb_sectors == 0 || (req->cmd.buf[1] & 0x16)) {
1799         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1800         return;
1801     }
1802 
1803     if (blk_is_read_only(s->qdev.conf.blk)) {
1804         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1805         return;
1806     }
1807     if (!check_lba_range(s, r->req.cmd.lba, nb_sectors)) {
1808         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1809         return;
1810     }
1811 
1812     if ((req->cmd.buf[1] & 0x1) || buffer_is_zero(inbuf, s->qdev.blocksize)) {
1813         int flags = (req->cmd.buf[1] & 0x8) ? BDRV_REQ_MAY_UNMAP : 0;
1814 
1815         /* The request is used as the AIO opaque value, so add a ref.  */
1816         scsi_req_ref(&r->req);
1817         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1818                          nb_sectors * s->qdev.blocksize,
1819                         BLOCK_ACCT_WRITE);
1820         r->req.aiocb = blk_aio_pwrite_zeroes(s->qdev.conf.blk,
1821                                 r->req.cmd.lba * s->qdev.blocksize,
1822                                 nb_sectors * s->qdev.blocksize,
1823                                 flags, scsi_aio_complete, r);
1824         return;
1825     }
1826 
1827     data = g_new0(WriteSameCBData, 1);
1828     data->r = r;
1829     data->sector = r->req.cmd.lba * (s->qdev.blocksize / 512);
1830     data->nb_sectors = nb_sectors * (s->qdev.blocksize / 512);
1831     data->iov.iov_len = MIN(data->nb_sectors * 512, SCSI_WRITE_SAME_MAX);
1832     data->iov.iov_base = buf = blk_blockalign(s->qdev.conf.blk,
1833                                               data->iov.iov_len);
1834     qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1835 
1836     for (i = 0; i < data->iov.iov_len; i += s->qdev.blocksize) {
1837         memcpy(&buf[i], inbuf, s->qdev.blocksize);
1838     }
1839 
1840     scsi_req_ref(&r->req);
1841     block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1842                      data->iov.iov_len, BLOCK_ACCT_WRITE);
1843     r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1844                                    data->sector << BDRV_SECTOR_BITS,
1845                                    &data->qiov, 0,
1846                                    scsi_write_same_complete, data);
1847 }
1848 
1849 static void scsi_disk_emulate_write_data(SCSIRequest *req)
1850 {
1851     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1852 
1853     if (r->iov.iov_len) {
1854         int buflen = r->iov.iov_len;
1855         DPRINTF("Write buf_len=%d\n", buflen);
1856         r->iov.iov_len = 0;
1857         scsi_req_data(&r->req, buflen);
1858         return;
1859     }
1860 
1861     switch (req->cmd.buf[0]) {
1862     case MODE_SELECT:
1863     case MODE_SELECT_10:
1864         /* This also clears the sense buffer for REQUEST SENSE.  */
1865         scsi_disk_emulate_mode_select(r, r->iov.iov_base);
1866         break;
1867 
1868     case UNMAP:
1869         scsi_disk_emulate_unmap(r, r->iov.iov_base);
1870         break;
1871 
1872     case VERIFY_10:
1873     case VERIFY_12:
1874     case VERIFY_16:
1875         if (r->req.status == -1) {
1876             scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1877         }
1878         break;
1879 
1880     case WRITE_SAME_10:
1881     case WRITE_SAME_16:
1882         scsi_disk_emulate_write_same(r, r->iov.iov_base);
1883         break;
1884 
1885     default:
1886         abort();
1887     }
1888 }
1889 
1890 static int32_t scsi_disk_emulate_command(SCSIRequest *req, uint8_t *buf)
1891 {
1892     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1893     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1894     uint64_t nb_sectors;
1895     uint8_t *outbuf;
1896     int buflen;
1897 
1898     switch (req->cmd.buf[0]) {
1899     case INQUIRY:
1900     case MODE_SENSE:
1901     case MODE_SENSE_10:
1902     case RESERVE:
1903     case RESERVE_10:
1904     case RELEASE:
1905     case RELEASE_10:
1906     case START_STOP:
1907     case ALLOW_MEDIUM_REMOVAL:
1908     case GET_CONFIGURATION:
1909     case GET_EVENT_STATUS_NOTIFICATION:
1910     case MECHANISM_STATUS:
1911     case REQUEST_SENSE:
1912         break;
1913 
1914     default:
1915         if (!blk_is_available(s->qdev.conf.blk)) {
1916             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
1917             return 0;
1918         }
1919         break;
1920     }
1921 
1922     /*
1923      * FIXME: we shouldn't return anything bigger than 4k, but the code
1924      * requires the buffer to be as big as req->cmd.xfer in several
1925      * places.  So, do not allow CDBs with a very large ALLOCATION
1926      * LENGTH.  The real fix would be to modify scsi_read_data and
1927      * dma_buf_read, so that they return data beyond the buflen
1928      * as all zeros.
1929      */
1930     if (req->cmd.xfer > 65536) {
1931         goto illegal_request;
1932     }
1933     r->buflen = MAX(4096, req->cmd.xfer);
1934 
1935     if (!r->iov.iov_base) {
1936         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
1937     }
1938 
1939     buflen = req->cmd.xfer;
1940     outbuf = r->iov.iov_base;
1941     memset(outbuf, 0, r->buflen);
1942     switch (req->cmd.buf[0]) {
1943     case TEST_UNIT_READY:
1944         assert(blk_is_available(s->qdev.conf.blk));
1945         break;
1946     case INQUIRY:
1947         buflen = scsi_disk_emulate_inquiry(req, outbuf);
1948         if (buflen < 0) {
1949             goto illegal_request;
1950         }
1951         break;
1952     case MODE_SENSE:
1953     case MODE_SENSE_10:
1954         buflen = scsi_disk_emulate_mode_sense(r, outbuf);
1955         if (buflen < 0) {
1956             goto illegal_request;
1957         }
1958         break;
1959     case READ_TOC:
1960         buflen = scsi_disk_emulate_read_toc(req, outbuf);
1961         if (buflen < 0) {
1962             goto illegal_request;
1963         }
1964         break;
1965     case RESERVE:
1966         if (req->cmd.buf[1] & 1) {
1967             goto illegal_request;
1968         }
1969         break;
1970     case RESERVE_10:
1971         if (req->cmd.buf[1] & 3) {
1972             goto illegal_request;
1973         }
1974         break;
1975     case RELEASE:
1976         if (req->cmd.buf[1] & 1) {
1977             goto illegal_request;
1978         }
1979         break;
1980     case RELEASE_10:
1981         if (req->cmd.buf[1] & 3) {
1982             goto illegal_request;
1983         }
1984         break;
1985     case START_STOP:
1986         if (scsi_disk_emulate_start_stop(r) < 0) {
1987             return 0;
1988         }
1989         break;
1990     case ALLOW_MEDIUM_REMOVAL:
1991         s->tray_locked = req->cmd.buf[4] & 1;
1992         blk_lock_medium(s->qdev.conf.blk, req->cmd.buf[4] & 1);
1993         break;
1994     case READ_CAPACITY_10:
1995         /* The normal LEN field for this command is zero.  */
1996         memset(outbuf, 0, 8);
1997         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1998         if (!nb_sectors) {
1999             scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
2000             return 0;
2001         }
2002         if ((req->cmd.buf[8] & 1) == 0 && req->cmd.lba) {
2003             goto illegal_request;
2004         }
2005         nb_sectors /= s->qdev.blocksize / 512;
2006         /* Returned value is the address of the last sector.  */
2007         nb_sectors--;
2008         /* Remember the new size for read/write sanity checking. */
2009         s->qdev.max_lba = nb_sectors;
2010         /* Clip to 2TB, instead of returning capacity modulo 2TB. */
2011         if (nb_sectors > UINT32_MAX) {
2012             nb_sectors = UINT32_MAX;
2013         }
2014         outbuf[0] = (nb_sectors >> 24) & 0xff;
2015         outbuf[1] = (nb_sectors >> 16) & 0xff;
2016         outbuf[2] = (nb_sectors >> 8) & 0xff;
2017         outbuf[3] = nb_sectors & 0xff;
2018         outbuf[4] = 0;
2019         outbuf[5] = 0;
2020         outbuf[6] = s->qdev.blocksize >> 8;
2021         outbuf[7] = 0;
2022         break;
2023     case REQUEST_SENSE:
2024         /* Just return "NO SENSE".  */
2025         buflen = scsi_convert_sense(NULL, 0, outbuf, r->buflen,
2026                                     (req->cmd.buf[1] & 1) == 0);
2027         if (buflen < 0) {
2028             goto illegal_request;
2029         }
2030         break;
2031     case MECHANISM_STATUS:
2032         buflen = scsi_emulate_mechanism_status(s, outbuf);
2033         if (buflen < 0) {
2034             goto illegal_request;
2035         }
2036         break;
2037     case GET_CONFIGURATION:
2038         buflen = scsi_get_configuration(s, outbuf);
2039         if (buflen < 0) {
2040             goto illegal_request;
2041         }
2042         break;
2043     case GET_EVENT_STATUS_NOTIFICATION:
2044         buflen = scsi_get_event_status_notification(s, r, outbuf);
2045         if (buflen < 0) {
2046             goto illegal_request;
2047         }
2048         break;
2049     case READ_DISC_INFORMATION:
2050         buflen = scsi_read_disc_information(s, r, outbuf);
2051         if (buflen < 0) {
2052             goto illegal_request;
2053         }
2054         break;
2055     case READ_DVD_STRUCTURE:
2056         buflen = scsi_read_dvd_structure(s, r, outbuf);
2057         if (buflen < 0) {
2058             goto illegal_request;
2059         }
2060         break;
2061     case SERVICE_ACTION_IN_16:
2062         /* Service Action In subcommands. */
2063         if ((req->cmd.buf[1] & 31) == SAI_READ_CAPACITY_16) {
2064             DPRINTF("SAI READ CAPACITY(16)\n");
2065             memset(outbuf, 0, req->cmd.xfer);
2066             blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2067             if (!nb_sectors) {
2068                 scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
2069                 return 0;
2070             }
2071             if ((req->cmd.buf[14] & 1) == 0 && req->cmd.lba) {
2072                 goto illegal_request;
2073             }
2074             nb_sectors /= s->qdev.blocksize / 512;
2075             /* Returned value is the address of the last sector.  */
2076             nb_sectors--;
2077             /* Remember the new size for read/write sanity checking. */
2078             s->qdev.max_lba = nb_sectors;
2079             outbuf[0] = (nb_sectors >> 56) & 0xff;
2080             outbuf[1] = (nb_sectors >> 48) & 0xff;
2081             outbuf[2] = (nb_sectors >> 40) & 0xff;
2082             outbuf[3] = (nb_sectors >> 32) & 0xff;
2083             outbuf[4] = (nb_sectors >> 24) & 0xff;
2084             outbuf[5] = (nb_sectors >> 16) & 0xff;
2085             outbuf[6] = (nb_sectors >> 8) & 0xff;
2086             outbuf[7] = nb_sectors & 0xff;
2087             outbuf[8] = 0;
2088             outbuf[9] = 0;
2089             outbuf[10] = s->qdev.blocksize >> 8;
2090             outbuf[11] = 0;
2091             outbuf[12] = 0;
2092             outbuf[13] = get_physical_block_exp(&s->qdev.conf);
2093 
2094             /* set TPE bit if the format supports discard */
2095             if (s->qdev.conf.discard_granularity) {
2096                 outbuf[14] = 0x80;
2097             }
2098 
2099             /* Protection, exponent and lowest lba field left blank. */
2100             break;
2101         }
2102         DPRINTF("Unsupported Service Action In\n");
2103         goto illegal_request;
2104     case SYNCHRONIZE_CACHE:
2105         /* The request is used as the AIO opaque value, so add a ref.  */
2106         scsi_req_ref(&r->req);
2107         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
2108                          BLOCK_ACCT_FLUSH);
2109         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
2110         return 0;
2111     case SEEK_10:
2112         DPRINTF("Seek(10) (sector %" PRId64 ")\n", r->req.cmd.lba);
2113         if (r->req.cmd.lba > s->qdev.max_lba) {
2114             goto illegal_lba;
2115         }
2116         break;
2117     case MODE_SELECT:
2118         DPRINTF("Mode Select(6) (len %lu)\n", (unsigned long)r->req.cmd.xfer);
2119         break;
2120     case MODE_SELECT_10:
2121         DPRINTF("Mode Select(10) (len %lu)\n", (unsigned long)r->req.cmd.xfer);
2122         break;
2123     case UNMAP:
2124         DPRINTF("Unmap (len %lu)\n", (unsigned long)r->req.cmd.xfer);
2125         break;
2126     case VERIFY_10:
2127     case VERIFY_12:
2128     case VERIFY_16:
2129         DPRINTF("Verify (bytchk %d)\n", (req->cmd.buf[1] >> 1) & 3);
2130         if (req->cmd.buf[1] & 6) {
2131             goto illegal_request;
2132         }
2133         break;
2134     case WRITE_SAME_10:
2135     case WRITE_SAME_16:
2136         DPRINTF("WRITE SAME %d (len %lu)\n",
2137                 req->cmd.buf[0] == WRITE_SAME_10 ? 10 : 16,
2138                 (unsigned long)r->req.cmd.xfer);
2139         break;
2140     default:
2141         DPRINTF("Unknown SCSI command (%2.2x=%s)\n", buf[0],
2142                 scsi_command_name(buf[0]));
2143         scsi_check_condition(r, SENSE_CODE(INVALID_OPCODE));
2144         return 0;
2145     }
2146     assert(!r->req.aiocb);
2147     r->iov.iov_len = MIN(r->buflen, req->cmd.xfer);
2148     if (r->iov.iov_len == 0) {
2149         scsi_req_complete(&r->req, GOOD);
2150     }
2151     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2152         assert(r->iov.iov_len == req->cmd.xfer);
2153         return -r->iov.iov_len;
2154     } else {
2155         return r->iov.iov_len;
2156     }
2157 
2158 illegal_request:
2159     if (r->req.status == -1) {
2160         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2161     }
2162     return 0;
2163 
2164 illegal_lba:
2165     scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2166     return 0;
2167 }
2168 
2169 /* Execute a scsi command.  Returns the length of the data expected by the
2170    command.  This will be Positive for data transfers from the device
2171    (eg. disk reads), negative for transfers to the device (eg. disk writes),
2172    and zero if the command does not transfer any data.  */
2173 
2174 static int32_t scsi_disk_dma_command(SCSIRequest *req, uint8_t *buf)
2175 {
2176     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
2177     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2178     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
2179     uint32_t len;
2180     uint8_t command;
2181 
2182     command = buf[0];
2183 
2184     if (!blk_is_available(s->qdev.conf.blk)) {
2185         scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
2186         return 0;
2187     }
2188 
2189     len = scsi_data_cdb_xfer(r->req.cmd.buf);
2190     switch (command) {
2191     case READ_6:
2192     case READ_10:
2193     case READ_12:
2194     case READ_16:
2195         DPRINTF("Read (sector %" PRId64 ", count %u)\n", r->req.cmd.lba, len);
2196         /* Protection information is not supported.  For SCSI versions 2 and
2197          * older (as determined by snooping the guest's INQUIRY commands),
2198          * there is no RD/WR/VRPROTECT, so skip this check in these versions.
2199          */
2200         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2201             goto illegal_request;
2202         }
2203         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2204             goto illegal_lba;
2205         }
2206         r->sector = r->req.cmd.lba * (s->qdev.blocksize / 512);
2207         r->sector_count = len * (s->qdev.blocksize / 512);
2208         break;
2209     case WRITE_6:
2210     case WRITE_10:
2211     case WRITE_12:
2212     case WRITE_16:
2213     case WRITE_VERIFY_10:
2214     case WRITE_VERIFY_12:
2215     case WRITE_VERIFY_16:
2216         if (blk_is_read_only(s->qdev.conf.blk)) {
2217             scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
2218             return 0;
2219         }
2220         DPRINTF("Write %s(sector %" PRId64 ", count %u)\n",
2221                 (command & 0xe) == 0xe ? "And Verify " : "",
2222                 r->req.cmd.lba, len);
2223         /* fall through */
2224     case VERIFY_10:
2225     case VERIFY_12:
2226     case VERIFY_16:
2227         /* We get here only for BYTCHK == 0x01 and only for scsi-block.
2228          * As far as DMA is concerned, we can treat it the same as a write;
2229          * scsi_block_do_sgio will send VERIFY commands.
2230          */
2231         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2232             goto illegal_request;
2233         }
2234         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2235             goto illegal_lba;
2236         }
2237         r->sector = r->req.cmd.lba * (s->qdev.blocksize / 512);
2238         r->sector_count = len * (s->qdev.blocksize / 512);
2239         break;
2240     default:
2241         abort();
2242     illegal_request:
2243         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2244         return 0;
2245     illegal_lba:
2246         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2247         return 0;
2248     }
2249     r->need_fua_emulation = sdc->need_fua_emulation(&r->req.cmd);
2250     if (r->sector_count == 0) {
2251         scsi_req_complete(&r->req, GOOD);
2252     }
2253     assert(r->iov.iov_len == 0);
2254     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2255         return -r->sector_count * 512;
2256     } else {
2257         return r->sector_count * 512;
2258     }
2259 }
2260 
2261 static void scsi_disk_reset(DeviceState *dev)
2262 {
2263     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev.qdev, dev);
2264     uint64_t nb_sectors;
2265 
2266     scsi_device_purge_requests(&s->qdev, SENSE_CODE(RESET));
2267 
2268     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2269     nb_sectors /= s->qdev.blocksize / 512;
2270     if (nb_sectors) {
2271         nb_sectors--;
2272     }
2273     s->qdev.max_lba = nb_sectors;
2274     /* reset tray statuses */
2275     s->tray_locked = 0;
2276     s->tray_open = 0;
2277 
2278     s->qdev.scsi_version = s->qdev.default_scsi_version;
2279 }
2280 
2281 static void scsi_disk_resize_cb(void *opaque)
2282 {
2283     SCSIDiskState *s = opaque;
2284 
2285     /* SPC lists this sense code as available only for
2286      * direct-access devices.
2287      */
2288     if (s->qdev.type == TYPE_DISK) {
2289         scsi_device_report_change(&s->qdev, SENSE_CODE(CAPACITY_CHANGED));
2290     }
2291 }
2292 
2293 static void scsi_cd_change_media_cb(void *opaque, bool load, Error **errp)
2294 {
2295     SCSIDiskState *s = opaque;
2296 
2297     /*
2298      * When a CD gets changed, we have to report an ejected state and
2299      * then a loaded state to guests so that they detect tray
2300      * open/close and media change events.  Guests that do not use
2301      * GET_EVENT_STATUS_NOTIFICATION to detect such tray open/close
2302      * states rely on this behavior.
2303      *
2304      * media_changed governs the state machine used for unit attention
2305      * report.  media_event is used by GET EVENT STATUS NOTIFICATION.
2306      */
2307     s->media_changed = load;
2308     s->tray_open = !load;
2309     scsi_device_set_ua(&s->qdev, SENSE_CODE(UNIT_ATTENTION_NO_MEDIUM));
2310     s->media_event = true;
2311     s->eject_request = false;
2312 }
2313 
2314 static void scsi_cd_eject_request_cb(void *opaque, bool force)
2315 {
2316     SCSIDiskState *s = opaque;
2317 
2318     s->eject_request = true;
2319     if (force) {
2320         s->tray_locked = false;
2321     }
2322 }
2323 
2324 static bool scsi_cd_is_tray_open(void *opaque)
2325 {
2326     return ((SCSIDiskState *)opaque)->tray_open;
2327 }
2328 
2329 static bool scsi_cd_is_medium_locked(void *opaque)
2330 {
2331     return ((SCSIDiskState *)opaque)->tray_locked;
2332 }
2333 
2334 static const BlockDevOps scsi_disk_removable_block_ops = {
2335     .change_media_cb = scsi_cd_change_media_cb,
2336     .eject_request_cb = scsi_cd_eject_request_cb,
2337     .is_tray_open = scsi_cd_is_tray_open,
2338     .is_medium_locked = scsi_cd_is_medium_locked,
2339 
2340     .resize_cb = scsi_disk_resize_cb,
2341 };
2342 
2343 static const BlockDevOps scsi_disk_block_ops = {
2344     .resize_cb = scsi_disk_resize_cb,
2345 };
2346 
2347 static void scsi_disk_unit_attention_reported(SCSIDevice *dev)
2348 {
2349     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2350     if (s->media_changed) {
2351         s->media_changed = false;
2352         scsi_device_set_ua(&s->qdev, SENSE_CODE(MEDIUM_CHANGED));
2353     }
2354 }
2355 
2356 static void scsi_realize(SCSIDevice *dev, Error **errp)
2357 {
2358     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2359 
2360     if (!s->qdev.conf.blk) {
2361         error_setg(errp, "drive property not set");
2362         return;
2363     }
2364 
2365     if (!(s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2366         !blk_is_inserted(s->qdev.conf.blk)) {
2367         error_setg(errp, "Device needs media, but drive is empty");
2368         return;
2369     }
2370 
2371     blkconf_serial(&s->qdev.conf, &s->serial);
2372     blkconf_blocksizes(&s->qdev.conf);
2373 
2374     if (s->qdev.conf.logical_block_size >
2375         s->qdev.conf.physical_block_size) {
2376         error_setg(errp,
2377                    "logical_block_size > physical_block_size not supported");
2378         return;
2379     }
2380 
2381     if (dev->type == TYPE_DISK) {
2382         if (!blkconf_geometry(&dev->conf, NULL, 65535, 255, 255, errp)) {
2383             return;
2384         }
2385     }
2386     if (!blkconf_apply_backend_options(&dev->conf,
2387                                        blk_is_read_only(s->qdev.conf.blk),
2388                                        dev->type == TYPE_DISK, errp)) {
2389         return;
2390     }
2391 
2392     if (s->qdev.conf.discard_granularity == -1) {
2393         s->qdev.conf.discard_granularity =
2394             MAX(s->qdev.conf.logical_block_size, DEFAULT_DISCARD_GRANULARITY);
2395     }
2396 
2397     if (!s->version) {
2398         s->version = g_strdup(qemu_hw_version());
2399     }
2400     if (!s->vendor) {
2401         s->vendor = g_strdup("QEMU");
2402     }
2403 
2404     if (blk_is_sg(s->qdev.conf.blk)) {
2405         error_setg(errp, "unwanted /dev/sg*");
2406         return;
2407     }
2408 
2409     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2410             !(s->features & (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS))) {
2411         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_removable_block_ops, s);
2412     } else {
2413         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_block_ops, s);
2414     }
2415     blk_set_guest_block_size(s->qdev.conf.blk, s->qdev.blocksize);
2416 
2417     blk_iostatus_enable(s->qdev.conf.blk);
2418 }
2419 
2420 static void scsi_hd_realize(SCSIDevice *dev, Error **errp)
2421 {
2422     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2423     /* can happen for devices without drive. The error message for missing
2424      * backend will be issued in scsi_realize
2425      */
2426     if (s->qdev.conf.blk) {
2427         blkconf_blocksizes(&s->qdev.conf);
2428     }
2429     s->qdev.blocksize = s->qdev.conf.logical_block_size;
2430     s->qdev.type = TYPE_DISK;
2431     if (!s->product) {
2432         s->product = g_strdup("QEMU HARDDISK");
2433     }
2434     scsi_realize(&s->qdev, errp);
2435 }
2436 
2437 static void scsi_cd_realize(SCSIDevice *dev, Error **errp)
2438 {
2439     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2440     int ret;
2441 
2442     if (!dev->conf.blk) {
2443         /* Anonymous BlockBackend for an empty drive. As we put it into
2444          * dev->conf, qdev takes care of detaching on unplug. */
2445         dev->conf.blk = blk_new(0, BLK_PERM_ALL);
2446         ret = blk_attach_dev(dev->conf.blk, &dev->qdev);
2447         assert(ret == 0);
2448     }
2449 
2450     s->qdev.blocksize = 2048;
2451     s->qdev.type = TYPE_ROM;
2452     s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2453     if (!s->product) {
2454         s->product = g_strdup("QEMU CD-ROM");
2455     }
2456     scsi_realize(&s->qdev, errp);
2457 }
2458 
2459 static void scsi_disk_realize(SCSIDevice *dev, Error **errp)
2460 {
2461     DriveInfo *dinfo;
2462     Error *local_err = NULL;
2463 
2464     if (!dev->conf.blk) {
2465         scsi_realize(dev, &local_err);
2466         assert(local_err);
2467         error_propagate(errp, local_err);
2468         return;
2469     }
2470 
2471     dinfo = blk_legacy_dinfo(dev->conf.blk);
2472     if (dinfo && dinfo->media_cd) {
2473         scsi_cd_realize(dev, errp);
2474     } else {
2475         scsi_hd_realize(dev, errp);
2476     }
2477 }
2478 
2479 static const SCSIReqOps scsi_disk_emulate_reqops = {
2480     .size         = sizeof(SCSIDiskReq),
2481     .free_req     = scsi_free_request,
2482     .send_command = scsi_disk_emulate_command,
2483     .read_data    = scsi_disk_emulate_read_data,
2484     .write_data   = scsi_disk_emulate_write_data,
2485     .get_buf      = scsi_get_buf,
2486 };
2487 
2488 static const SCSIReqOps scsi_disk_dma_reqops = {
2489     .size         = sizeof(SCSIDiskReq),
2490     .free_req     = scsi_free_request,
2491     .send_command = scsi_disk_dma_command,
2492     .read_data    = scsi_read_data,
2493     .write_data   = scsi_write_data,
2494     .get_buf      = scsi_get_buf,
2495     .load_request = scsi_disk_load_request,
2496     .save_request = scsi_disk_save_request,
2497 };
2498 
2499 static const SCSIReqOps *const scsi_disk_reqops_dispatch[256] = {
2500     [TEST_UNIT_READY]                 = &scsi_disk_emulate_reqops,
2501     [INQUIRY]                         = &scsi_disk_emulate_reqops,
2502     [MODE_SENSE]                      = &scsi_disk_emulate_reqops,
2503     [MODE_SENSE_10]                   = &scsi_disk_emulate_reqops,
2504     [START_STOP]                      = &scsi_disk_emulate_reqops,
2505     [ALLOW_MEDIUM_REMOVAL]            = &scsi_disk_emulate_reqops,
2506     [READ_CAPACITY_10]                = &scsi_disk_emulate_reqops,
2507     [READ_TOC]                        = &scsi_disk_emulate_reqops,
2508     [READ_DVD_STRUCTURE]              = &scsi_disk_emulate_reqops,
2509     [READ_DISC_INFORMATION]           = &scsi_disk_emulate_reqops,
2510     [GET_CONFIGURATION]               = &scsi_disk_emulate_reqops,
2511     [GET_EVENT_STATUS_NOTIFICATION]   = &scsi_disk_emulate_reqops,
2512     [MECHANISM_STATUS]                = &scsi_disk_emulate_reqops,
2513     [SERVICE_ACTION_IN_16]            = &scsi_disk_emulate_reqops,
2514     [REQUEST_SENSE]                   = &scsi_disk_emulate_reqops,
2515     [SYNCHRONIZE_CACHE]               = &scsi_disk_emulate_reqops,
2516     [SEEK_10]                         = &scsi_disk_emulate_reqops,
2517     [MODE_SELECT]                     = &scsi_disk_emulate_reqops,
2518     [MODE_SELECT_10]                  = &scsi_disk_emulate_reqops,
2519     [UNMAP]                           = &scsi_disk_emulate_reqops,
2520     [WRITE_SAME_10]                   = &scsi_disk_emulate_reqops,
2521     [WRITE_SAME_16]                   = &scsi_disk_emulate_reqops,
2522     [VERIFY_10]                       = &scsi_disk_emulate_reqops,
2523     [VERIFY_12]                       = &scsi_disk_emulate_reqops,
2524     [VERIFY_16]                       = &scsi_disk_emulate_reqops,
2525 
2526     [READ_6]                          = &scsi_disk_dma_reqops,
2527     [READ_10]                         = &scsi_disk_dma_reqops,
2528     [READ_12]                         = &scsi_disk_dma_reqops,
2529     [READ_16]                         = &scsi_disk_dma_reqops,
2530     [WRITE_6]                         = &scsi_disk_dma_reqops,
2531     [WRITE_10]                        = &scsi_disk_dma_reqops,
2532     [WRITE_12]                        = &scsi_disk_dma_reqops,
2533     [WRITE_16]                        = &scsi_disk_dma_reqops,
2534     [WRITE_VERIFY_10]                 = &scsi_disk_dma_reqops,
2535     [WRITE_VERIFY_12]                 = &scsi_disk_dma_reqops,
2536     [WRITE_VERIFY_16]                 = &scsi_disk_dma_reqops,
2537 };
2538 
2539 static SCSIRequest *scsi_new_request(SCSIDevice *d, uint32_t tag, uint32_t lun,
2540                                      uint8_t *buf, void *hba_private)
2541 {
2542     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2543     SCSIRequest *req;
2544     const SCSIReqOps *ops;
2545     uint8_t command;
2546 
2547     command = buf[0];
2548     ops = scsi_disk_reqops_dispatch[command];
2549     if (!ops) {
2550         ops = &scsi_disk_emulate_reqops;
2551     }
2552     req = scsi_req_alloc(ops, &s->qdev, tag, lun, hba_private);
2553 
2554 #ifdef DEBUG_SCSI
2555     DPRINTF("Command: lun=%d tag=0x%x data=0x%02x", lun, tag, buf[0]);
2556     {
2557         int i;
2558         for (i = 1; i < scsi_cdb_length(buf); i++) {
2559             printf(" 0x%02x", buf[i]);
2560         }
2561         printf("\n");
2562     }
2563 #endif
2564 
2565     return req;
2566 }
2567 
2568 #ifdef __linux__
2569 static int get_device_type(SCSIDiskState *s)
2570 {
2571     uint8_t cmd[16];
2572     uint8_t buf[36];
2573     uint8_t sensebuf[8];
2574     sg_io_hdr_t io_header;
2575     int ret;
2576 
2577     memset(cmd, 0, sizeof(cmd));
2578     memset(buf, 0, sizeof(buf));
2579     cmd[0] = INQUIRY;
2580     cmd[4] = sizeof(buf);
2581 
2582     memset(&io_header, 0, sizeof(io_header));
2583     io_header.interface_id = 'S';
2584     io_header.dxfer_direction = SG_DXFER_FROM_DEV;
2585     io_header.dxfer_len = sizeof(buf);
2586     io_header.dxferp = buf;
2587     io_header.cmdp = cmd;
2588     io_header.cmd_len = sizeof(cmd);
2589     io_header.mx_sb_len = sizeof(sensebuf);
2590     io_header.sbp = sensebuf;
2591     io_header.timeout = 6000; /* XXX */
2592 
2593     ret = blk_ioctl(s->qdev.conf.blk, SG_IO, &io_header);
2594     if (ret < 0 || io_header.driver_status || io_header.host_status) {
2595         return -1;
2596     }
2597     s->qdev.type = buf[0];
2598     if (buf[1] & 0x80) {
2599         s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2600     }
2601     return 0;
2602 }
2603 
2604 static void scsi_block_realize(SCSIDevice *dev, Error **errp)
2605 {
2606     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2607     int sg_version;
2608     int rc;
2609 
2610     if (!s->qdev.conf.blk) {
2611         error_setg(errp, "drive property not set");
2612         return;
2613     }
2614 
2615     /* check we are using a driver managing SG_IO (version 3 and after) */
2616     rc = blk_ioctl(s->qdev.conf.blk, SG_GET_VERSION_NUM, &sg_version);
2617     if (rc < 0) {
2618         error_setg_errno(errp, -rc, "cannot get SG_IO version number");
2619         if (rc != -EPERM) {
2620             error_append_hint(errp, "Is this a SCSI device?\n");
2621         }
2622         return;
2623     }
2624     if (sg_version < 30000) {
2625         error_setg(errp, "scsi generic interface too old");
2626         return;
2627     }
2628 
2629     /* get device type from INQUIRY data */
2630     rc = get_device_type(s);
2631     if (rc < 0) {
2632         error_setg(errp, "INQUIRY failed");
2633         return;
2634     }
2635 
2636     /* Make a guess for the block size, we'll fix it when the guest sends.
2637      * READ CAPACITY.  If they don't, they likely would assume these sizes
2638      * anyway. (TODO: check in /sys).
2639      */
2640     if (s->qdev.type == TYPE_ROM || s->qdev.type == TYPE_WORM) {
2641         s->qdev.blocksize = 2048;
2642     } else {
2643         s->qdev.blocksize = 512;
2644     }
2645 
2646     /* Makes the scsi-block device not removable by using HMP and QMP eject
2647      * command.
2648      */
2649     s->features |= (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS);
2650 
2651     scsi_realize(&s->qdev, errp);
2652     scsi_generic_read_device_identification(&s->qdev);
2653 }
2654 
2655 typedef struct SCSIBlockReq {
2656     SCSIDiskReq req;
2657     sg_io_hdr_t io_header;
2658 
2659     /* Selected bytes of the original CDB, copied into our own CDB.  */
2660     uint8_t cmd, cdb1, group_number;
2661 
2662     /* CDB passed to SG_IO.  */
2663     uint8_t cdb[16];
2664 } SCSIBlockReq;
2665 
2666 static BlockAIOCB *scsi_block_do_sgio(SCSIBlockReq *req,
2667                                       int64_t offset, QEMUIOVector *iov,
2668                                       int direction,
2669                                       BlockCompletionFunc *cb, void *opaque)
2670 {
2671     sg_io_hdr_t *io_header = &req->io_header;
2672     SCSIDiskReq *r = &req->req;
2673     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2674     int nb_logical_blocks;
2675     uint64_t lba;
2676     BlockAIOCB *aiocb;
2677 
2678     /* This is not supported yet.  It can only happen if the guest does
2679      * reads and writes that are not aligned to one logical sectors
2680      * _and_ cover multiple MemoryRegions.
2681      */
2682     assert(offset % s->qdev.blocksize == 0);
2683     assert(iov->size % s->qdev.blocksize == 0);
2684 
2685     io_header->interface_id = 'S';
2686 
2687     /* The data transfer comes from the QEMUIOVector.  */
2688     io_header->dxfer_direction = direction;
2689     io_header->dxfer_len = iov->size;
2690     io_header->dxferp = (void *)iov->iov;
2691     io_header->iovec_count = iov->niov;
2692     assert(io_header->iovec_count == iov->niov); /* no overflow! */
2693 
2694     /* Build a new CDB with the LBA and length patched in, in case
2695      * DMA helpers split the transfer in multiple segments.  Do not
2696      * build a CDB smaller than what the guest wanted, and only build
2697      * a larger one if strictly necessary.
2698      */
2699     io_header->cmdp = req->cdb;
2700     lba = offset / s->qdev.blocksize;
2701     nb_logical_blocks = io_header->dxfer_len / s->qdev.blocksize;
2702 
2703     if ((req->cmd >> 5) == 0 && lba <= 0x1ffff) {
2704         /* 6-byte CDB */
2705         stl_be_p(&req->cdb[0], lba | (req->cmd << 24));
2706         req->cdb[4] = nb_logical_blocks;
2707         req->cdb[5] = 0;
2708         io_header->cmd_len = 6;
2709     } else if ((req->cmd >> 5) <= 1 && lba <= 0xffffffffULL) {
2710         /* 10-byte CDB */
2711         req->cdb[0] = (req->cmd & 0x1f) | 0x20;
2712         req->cdb[1] = req->cdb1;
2713         stl_be_p(&req->cdb[2], lba);
2714         req->cdb[6] = req->group_number;
2715         stw_be_p(&req->cdb[7], nb_logical_blocks);
2716         req->cdb[9] = 0;
2717         io_header->cmd_len = 10;
2718     } else if ((req->cmd >> 5) != 4 && lba <= 0xffffffffULL) {
2719         /* 12-byte CDB */
2720         req->cdb[0] = (req->cmd & 0x1f) | 0xA0;
2721         req->cdb[1] = req->cdb1;
2722         stl_be_p(&req->cdb[2], lba);
2723         stl_be_p(&req->cdb[6], nb_logical_blocks);
2724         req->cdb[10] = req->group_number;
2725         req->cdb[11] = 0;
2726         io_header->cmd_len = 12;
2727     } else {
2728         /* 16-byte CDB */
2729         req->cdb[0] = (req->cmd & 0x1f) | 0x80;
2730         req->cdb[1] = req->cdb1;
2731         stq_be_p(&req->cdb[2], lba);
2732         stl_be_p(&req->cdb[10], nb_logical_blocks);
2733         req->cdb[14] = req->group_number;
2734         req->cdb[15] = 0;
2735         io_header->cmd_len = 16;
2736     }
2737 
2738     /* The rest is as in scsi-generic.c.  */
2739     io_header->mx_sb_len = sizeof(r->req.sense);
2740     io_header->sbp = r->req.sense;
2741     io_header->timeout = UINT_MAX;
2742     io_header->usr_ptr = r;
2743     io_header->flags |= SG_FLAG_DIRECT_IO;
2744 
2745     aiocb = blk_aio_ioctl(s->qdev.conf.blk, SG_IO, io_header, cb, opaque);
2746     assert(aiocb != NULL);
2747     return aiocb;
2748 }
2749 
2750 static bool scsi_block_no_fua(SCSICommand *cmd)
2751 {
2752     return false;
2753 }
2754 
2755 static BlockAIOCB *scsi_block_dma_readv(int64_t offset,
2756                                         QEMUIOVector *iov,
2757                                         BlockCompletionFunc *cb, void *cb_opaque,
2758                                         void *opaque)
2759 {
2760     SCSIBlockReq *r = opaque;
2761     return scsi_block_do_sgio(r, offset, iov,
2762                               SG_DXFER_FROM_DEV, cb, cb_opaque);
2763 }
2764 
2765 static BlockAIOCB *scsi_block_dma_writev(int64_t offset,
2766                                          QEMUIOVector *iov,
2767                                          BlockCompletionFunc *cb, void *cb_opaque,
2768                                          void *opaque)
2769 {
2770     SCSIBlockReq *r = opaque;
2771     return scsi_block_do_sgio(r, offset, iov,
2772                               SG_DXFER_TO_DEV, cb, cb_opaque);
2773 }
2774 
2775 static bool scsi_block_is_passthrough(SCSIDiskState *s, uint8_t *buf)
2776 {
2777     switch (buf[0]) {
2778     case VERIFY_10:
2779     case VERIFY_12:
2780     case VERIFY_16:
2781         /* Check if BYTCHK == 0x01 (data-out buffer contains data
2782          * for the number of logical blocks specified in the length
2783          * field).  For other modes, do not use scatter/gather operation.
2784          */
2785         if ((buf[1] & 6) == 2) {
2786             return false;
2787         }
2788         break;
2789 
2790     case READ_6:
2791     case READ_10:
2792     case READ_12:
2793     case READ_16:
2794     case WRITE_6:
2795     case WRITE_10:
2796     case WRITE_12:
2797     case WRITE_16:
2798     case WRITE_VERIFY_10:
2799     case WRITE_VERIFY_12:
2800     case WRITE_VERIFY_16:
2801         /* MMC writing cannot be done via DMA helpers, because it sometimes
2802          * involves writing beyond the maximum LBA or to negative LBA (lead-in).
2803          * We might use scsi_block_dma_reqops as long as no writing commands are
2804          * seen, but performance usually isn't paramount on optical media.  So,
2805          * just make scsi-block operate the same as scsi-generic for them.
2806          */
2807         if (s->qdev.type != TYPE_ROM) {
2808             return false;
2809         }
2810         break;
2811 
2812     default:
2813         break;
2814     }
2815 
2816     return true;
2817 }
2818 
2819 
2820 static int32_t scsi_block_dma_command(SCSIRequest *req, uint8_t *buf)
2821 {
2822     SCSIBlockReq *r = (SCSIBlockReq *)req;
2823     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2824 
2825     r->cmd = req->cmd.buf[0];
2826     switch (r->cmd >> 5) {
2827     case 0:
2828         /* 6-byte CDB.  */
2829         r->cdb1 = r->group_number = 0;
2830         break;
2831     case 1:
2832         /* 10-byte CDB.  */
2833         r->cdb1 = req->cmd.buf[1];
2834         r->group_number = req->cmd.buf[6];
2835         break;
2836     case 4:
2837         /* 12-byte CDB.  */
2838         r->cdb1 = req->cmd.buf[1];
2839         r->group_number = req->cmd.buf[10];
2840         break;
2841     case 5:
2842         /* 16-byte CDB.  */
2843         r->cdb1 = req->cmd.buf[1];
2844         r->group_number = req->cmd.buf[14];
2845         break;
2846     default:
2847         abort();
2848     }
2849 
2850     /* Protection information is not supported.  For SCSI versions 2 and
2851      * older (as determined by snooping the guest's INQUIRY commands),
2852      * there is no RD/WR/VRPROTECT, so skip this check in these versions.
2853      */
2854     if (s->qdev.scsi_version > 2 && (req->cmd.buf[1] & 0xe0)) {
2855         scsi_check_condition(&r->req, SENSE_CODE(INVALID_FIELD));
2856         return 0;
2857     }
2858 
2859     r->req.status = &r->io_header.status;
2860     return scsi_disk_dma_command(req, buf);
2861 }
2862 
2863 static const SCSIReqOps scsi_block_dma_reqops = {
2864     .size         = sizeof(SCSIBlockReq),
2865     .free_req     = scsi_free_request,
2866     .send_command = scsi_block_dma_command,
2867     .read_data    = scsi_read_data,
2868     .write_data   = scsi_write_data,
2869     .get_buf      = scsi_get_buf,
2870     .load_request = scsi_disk_load_request,
2871     .save_request = scsi_disk_save_request,
2872 };
2873 
2874 static SCSIRequest *scsi_block_new_request(SCSIDevice *d, uint32_t tag,
2875                                            uint32_t lun, uint8_t *buf,
2876                                            void *hba_private)
2877 {
2878     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2879 
2880     if (scsi_block_is_passthrough(s, buf)) {
2881         return scsi_req_alloc(&scsi_generic_req_ops, &s->qdev, tag, lun,
2882                               hba_private);
2883     } else {
2884         return scsi_req_alloc(&scsi_block_dma_reqops, &s->qdev, tag, lun,
2885                               hba_private);
2886     }
2887 }
2888 
2889 static int scsi_block_parse_cdb(SCSIDevice *d, SCSICommand *cmd,
2890                                   uint8_t *buf, void *hba_private)
2891 {
2892     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2893 
2894     if (scsi_block_is_passthrough(s, buf)) {
2895         return scsi_bus_parse_cdb(&s->qdev, cmd, buf, hba_private);
2896     } else {
2897         return scsi_req_parse_cdb(&s->qdev, cmd, buf);
2898     }
2899 }
2900 
2901 #endif
2902 
2903 static
2904 BlockAIOCB *scsi_dma_readv(int64_t offset, QEMUIOVector *iov,
2905                            BlockCompletionFunc *cb, void *cb_opaque,
2906                            void *opaque)
2907 {
2908     SCSIDiskReq *r = opaque;
2909     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2910     return blk_aio_preadv(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
2911 }
2912 
2913 static
2914 BlockAIOCB *scsi_dma_writev(int64_t offset, QEMUIOVector *iov,
2915                             BlockCompletionFunc *cb, void *cb_opaque,
2916                             void *opaque)
2917 {
2918     SCSIDiskReq *r = opaque;
2919     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2920     return blk_aio_pwritev(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
2921 }
2922 
2923 static void scsi_disk_base_class_initfn(ObjectClass *klass, void *data)
2924 {
2925     DeviceClass *dc = DEVICE_CLASS(klass);
2926     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
2927 
2928     dc->fw_name = "disk";
2929     dc->reset = scsi_disk_reset;
2930     sdc->dma_readv = scsi_dma_readv;
2931     sdc->dma_writev = scsi_dma_writev;
2932     sdc->need_fua_emulation = scsi_is_cmd_fua;
2933 }
2934 
2935 static const TypeInfo scsi_disk_base_info = {
2936     .name          = TYPE_SCSI_DISK_BASE,
2937     .parent        = TYPE_SCSI_DEVICE,
2938     .class_init    = scsi_disk_base_class_initfn,
2939     .instance_size = sizeof(SCSIDiskState),
2940     .class_size    = sizeof(SCSIDiskClass),
2941     .abstract      = true,
2942 };
2943 
2944 #define DEFINE_SCSI_DISK_PROPERTIES()                                \
2945     DEFINE_BLOCK_PROPERTIES(SCSIDiskState, qdev.conf),               \
2946     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),         \
2947     DEFINE_PROP_STRING("ver", SCSIDiskState, version),               \
2948     DEFINE_PROP_STRING("serial", SCSIDiskState, serial),             \
2949     DEFINE_PROP_STRING("vendor", SCSIDiskState, vendor),             \
2950     DEFINE_PROP_STRING("product", SCSIDiskState, product)
2951 
2952 static Property scsi_hd_properties[] = {
2953     DEFINE_SCSI_DISK_PROPERTIES(),
2954     DEFINE_PROP_BIT("removable", SCSIDiskState, features,
2955                     SCSI_DISK_F_REMOVABLE, false),
2956     DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
2957                     SCSI_DISK_F_DPOFUA, false),
2958     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
2959     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
2960     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
2961     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
2962                        DEFAULT_MAX_UNMAP_SIZE),
2963     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
2964                        DEFAULT_MAX_IO_SIZE),
2965     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
2966     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
2967                       5),
2968     DEFINE_BLOCK_CHS_PROPERTIES(SCSIDiskState, qdev.conf),
2969     DEFINE_PROP_END_OF_LIST(),
2970 };
2971 
2972 static const VMStateDescription vmstate_scsi_disk_state = {
2973     .name = "scsi-disk",
2974     .version_id = 1,
2975     .minimum_version_id = 1,
2976     .fields = (VMStateField[]) {
2977         VMSTATE_SCSI_DEVICE(qdev, SCSIDiskState),
2978         VMSTATE_BOOL(media_changed, SCSIDiskState),
2979         VMSTATE_BOOL(media_event, SCSIDiskState),
2980         VMSTATE_BOOL(eject_request, SCSIDiskState),
2981         VMSTATE_BOOL(tray_open, SCSIDiskState),
2982         VMSTATE_BOOL(tray_locked, SCSIDiskState),
2983         VMSTATE_END_OF_LIST()
2984     }
2985 };
2986 
2987 static void scsi_hd_class_initfn(ObjectClass *klass, void *data)
2988 {
2989     DeviceClass *dc = DEVICE_CLASS(klass);
2990     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
2991 
2992     sc->realize      = scsi_hd_realize;
2993     sc->alloc_req    = scsi_new_request;
2994     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
2995     dc->desc = "virtual SCSI disk";
2996     dc->props = scsi_hd_properties;
2997     dc->vmsd  = &vmstate_scsi_disk_state;
2998 }
2999 
3000 static const TypeInfo scsi_hd_info = {
3001     .name          = "scsi-hd",
3002     .parent        = TYPE_SCSI_DISK_BASE,
3003     .class_init    = scsi_hd_class_initfn,
3004 };
3005 
3006 static Property scsi_cd_properties[] = {
3007     DEFINE_SCSI_DISK_PROPERTIES(),
3008     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3009     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3010     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3011     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3012                        DEFAULT_MAX_IO_SIZE),
3013     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3014                       5),
3015     DEFINE_PROP_END_OF_LIST(),
3016 };
3017 
3018 static void scsi_cd_class_initfn(ObjectClass *klass, void *data)
3019 {
3020     DeviceClass *dc = DEVICE_CLASS(klass);
3021     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3022 
3023     sc->realize      = scsi_cd_realize;
3024     sc->alloc_req    = scsi_new_request;
3025     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3026     dc->desc = "virtual SCSI CD-ROM";
3027     dc->props = scsi_cd_properties;
3028     dc->vmsd  = &vmstate_scsi_disk_state;
3029 }
3030 
3031 static const TypeInfo scsi_cd_info = {
3032     .name          = "scsi-cd",
3033     .parent        = TYPE_SCSI_DISK_BASE,
3034     .class_init    = scsi_cd_class_initfn,
3035 };
3036 
3037 #ifdef __linux__
3038 static Property scsi_block_properties[] = {
3039     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),         \
3040     DEFINE_PROP_DRIVE("drive", SCSIDiskState, qdev.conf.blk),
3041     DEFINE_PROP_BOOL("share-rw", SCSIDiskState, qdev.conf.share_rw, false),
3042     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3043     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3044                       -1),
3045     DEFINE_PROP_END_OF_LIST(),
3046 };
3047 
3048 static void scsi_block_class_initfn(ObjectClass *klass, void *data)
3049 {
3050     DeviceClass *dc = DEVICE_CLASS(klass);
3051     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3052     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
3053 
3054     sc->realize      = scsi_block_realize;
3055     sc->alloc_req    = scsi_block_new_request;
3056     sc->parse_cdb    = scsi_block_parse_cdb;
3057     sdc->dma_readv   = scsi_block_dma_readv;
3058     sdc->dma_writev  = scsi_block_dma_writev;
3059     sdc->need_fua_emulation = scsi_block_no_fua;
3060     dc->desc = "SCSI block device passthrough";
3061     dc->props = scsi_block_properties;
3062     dc->vmsd  = &vmstate_scsi_disk_state;
3063 }
3064 
3065 static const TypeInfo scsi_block_info = {
3066     .name          = "scsi-block",
3067     .parent        = TYPE_SCSI_DISK_BASE,
3068     .class_init    = scsi_block_class_initfn,
3069 };
3070 #endif
3071 
3072 static Property scsi_disk_properties[] = {
3073     DEFINE_SCSI_DISK_PROPERTIES(),
3074     DEFINE_PROP_BIT("removable", SCSIDiskState, features,
3075                     SCSI_DISK_F_REMOVABLE, false),
3076     DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
3077                     SCSI_DISK_F_DPOFUA, false),
3078     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3079     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3080     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3081     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3082                        DEFAULT_MAX_UNMAP_SIZE),
3083     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3084                        DEFAULT_MAX_IO_SIZE),
3085     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3086                       5),
3087     DEFINE_PROP_END_OF_LIST(),
3088 };
3089 
3090 static void scsi_disk_class_initfn(ObjectClass *klass, void *data)
3091 {
3092     DeviceClass *dc = DEVICE_CLASS(klass);
3093     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3094 
3095     sc->realize      = scsi_disk_realize;
3096     sc->alloc_req    = scsi_new_request;
3097     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3098     dc->fw_name = "disk";
3099     dc->desc = "virtual SCSI disk or CD-ROM (legacy)";
3100     dc->reset = scsi_disk_reset;
3101     dc->props = scsi_disk_properties;
3102     dc->vmsd  = &vmstate_scsi_disk_state;
3103 }
3104 
3105 static const TypeInfo scsi_disk_info = {
3106     .name          = "scsi-disk",
3107     .parent        = TYPE_SCSI_DISK_BASE,
3108     .class_init    = scsi_disk_class_initfn,
3109 };
3110 
3111 static void scsi_disk_register_types(void)
3112 {
3113     type_register_static(&scsi_disk_base_info);
3114     type_register_static(&scsi_hd_info);
3115     type_register_static(&scsi_cd_info);
3116 #ifdef __linux__
3117     type_register_static(&scsi_block_info);
3118 #endif
3119     type_register_static(&scsi_disk_info);
3120 }
3121 
3122 type_init(scsi_disk_register_types)
3123