xref: /openbmc/qemu/hw/scsi/scsi-disk.c (revision efba1595)
1 /*
2  * SCSI Device emulation
3  *
4  * Copyright (c) 2006 CodeSourcery.
5  * Based on code by Fabrice Bellard
6  *
7  * Written by Paul Brook
8  * Modifications:
9  *  2009-Dec-12 Artyom Tarasenko : implemented stamdard inquiry for the case
10  *                                 when the allocation length of CDB is smaller
11  *                                 than 36.
12  *  2009-Oct-13 Artyom Tarasenko : implemented the block descriptor in the
13  *                                 MODE SENSE response.
14  *
15  * This code is licensed under the LGPL.
16  *
17  * Note that this file only handles the SCSI architecture model and device
18  * commands.  Emulation of interface/link layer protocols is handled by
19  * the host adapter emulator.
20  */
21 
22 #include "qemu/osdep.h"
23 #include "qemu/units.h"
24 #include "qapi/error.h"
25 #include "qemu/error-report.h"
26 #include "qemu/main-loop.h"
27 #include "qemu/module.h"
28 #include "hw/scsi/scsi.h"
29 #include "migration/qemu-file-types.h"
30 #include "migration/vmstate.h"
31 #include "hw/scsi/emulation.h"
32 #include "scsi/constants.h"
33 #include "sysemu/block-backend.h"
34 #include "sysemu/blockdev.h"
35 #include "hw/block/block.h"
36 #include "hw/qdev-properties.h"
37 #include "sysemu/dma.h"
38 #include "sysemu/sysemu.h"
39 #include "qemu/cutils.h"
40 #include "trace.h"
41 #include "qom/object.h"
42 
43 #ifdef __linux
44 #include <scsi/sg.h>
45 #endif
46 
47 #define SCSI_WRITE_SAME_MAX         (512 * KiB)
48 #define SCSI_DMA_BUF_SIZE           (128 * KiB)
49 #define SCSI_MAX_INQUIRY_LEN        256
50 #define SCSI_MAX_MODE_LEN           256
51 
52 #define DEFAULT_DISCARD_GRANULARITY (4 * KiB)
53 #define DEFAULT_MAX_UNMAP_SIZE      (1 * GiB)
54 #define DEFAULT_MAX_IO_SIZE         INT_MAX     /* 2 GB - 1 block */
55 
56 #define TYPE_SCSI_DISK_BASE         "scsi-disk-base"
57 
58 OBJECT_DECLARE_TYPE(SCSIDiskState, SCSIDiskClass, SCSI_DISK_BASE)
59 
60 struct SCSIDiskClass {
61     SCSIDeviceClass parent_class;
62     DMAIOFunc       *dma_readv;
63     DMAIOFunc       *dma_writev;
64     bool            (*need_fua_emulation)(SCSICommand *cmd);
65     void            (*update_sense)(SCSIRequest *r);
66 };
67 
68 typedef struct SCSIDiskReq {
69     SCSIRequest req;
70     /* Both sector and sector_count are in terms of BDRV_SECTOR_SIZE bytes.  */
71     uint64_t sector;
72     uint32_t sector_count;
73     uint32_t buflen;
74     bool started;
75     bool need_fua_emulation;
76     struct iovec iov;
77     QEMUIOVector qiov;
78     BlockAcctCookie acct;
79     unsigned char *status;
80 } SCSIDiskReq;
81 
82 #define SCSI_DISK_F_REMOVABLE             0
83 #define SCSI_DISK_F_DPOFUA                1
84 #define SCSI_DISK_F_NO_REMOVABLE_DEVOPS   2
85 
86 struct SCSIDiskState {
87     SCSIDevice qdev;
88     uint32_t features;
89     bool media_changed;
90     bool media_event;
91     bool eject_request;
92     uint16_t port_index;
93     uint64_t max_unmap_size;
94     uint64_t max_io_size;
95     QEMUBH *bh;
96     char *version;
97     char *serial;
98     char *vendor;
99     char *product;
100     char *device_id;
101     bool tray_open;
102     bool tray_locked;
103     /*
104      * 0x0000        - rotation rate not reported
105      * 0x0001        - non-rotating medium (SSD)
106      * 0x0002-0x0400 - reserved
107      * 0x0401-0xffe  - rotations per minute
108      * 0xffff        - reserved
109      */
110     uint16_t rotation_rate;
111 };
112 
113 static bool scsi_handle_rw_error(SCSIDiskReq *r, int error, bool acct_failed);
114 
115 static void scsi_free_request(SCSIRequest *req)
116 {
117     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
118 
119     qemu_vfree(r->iov.iov_base);
120 }
121 
122 /* Helper function for command completion with sense.  */
123 static void scsi_check_condition(SCSIDiskReq *r, SCSISense sense)
124 {
125     trace_scsi_disk_check_condition(r->req.tag, sense.key, sense.asc,
126                                     sense.ascq);
127     scsi_req_build_sense(&r->req, sense);
128     scsi_req_complete(&r->req, CHECK_CONDITION);
129 }
130 
131 static void scsi_init_iovec(SCSIDiskReq *r, size_t size)
132 {
133     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
134 
135     if (!r->iov.iov_base) {
136         r->buflen = size;
137         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
138     }
139     r->iov.iov_len = MIN(r->sector_count * BDRV_SECTOR_SIZE, r->buflen);
140     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
141 }
142 
143 static void scsi_disk_save_request(QEMUFile *f, SCSIRequest *req)
144 {
145     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
146 
147     qemu_put_be64s(f, &r->sector);
148     qemu_put_be32s(f, &r->sector_count);
149     qemu_put_be32s(f, &r->buflen);
150     if (r->buflen) {
151         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
152             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
153         } else if (!req->retry) {
154             uint32_t len = r->iov.iov_len;
155             qemu_put_be32s(f, &len);
156             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
157         }
158     }
159 }
160 
161 static void scsi_disk_load_request(QEMUFile *f, SCSIRequest *req)
162 {
163     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
164 
165     qemu_get_be64s(f, &r->sector);
166     qemu_get_be32s(f, &r->sector_count);
167     qemu_get_be32s(f, &r->buflen);
168     if (r->buflen) {
169         scsi_init_iovec(r, r->buflen);
170         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
171             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
172         } else if (!r->req.retry) {
173             uint32_t len;
174             qemu_get_be32s(f, &len);
175             r->iov.iov_len = len;
176             assert(r->iov.iov_len <= r->buflen);
177             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
178         }
179     }
180 
181     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
182 }
183 
184 static bool scsi_disk_req_check_error(SCSIDiskReq *r, int ret, bool acct_failed)
185 {
186     if (r->req.io_canceled) {
187         scsi_req_cancel_complete(&r->req);
188         return true;
189     }
190 
191     if (ret < 0 || (r->status && *r->status)) {
192         return scsi_handle_rw_error(r, -ret, acct_failed);
193     }
194 
195     return false;
196 }
197 
198 static void scsi_aio_complete(void *opaque, int ret)
199 {
200     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
201     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
202 
203     assert(r->req.aiocb != NULL);
204     r->req.aiocb = NULL;
205     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
206     if (scsi_disk_req_check_error(r, ret, true)) {
207         goto done;
208     }
209 
210     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
211     scsi_req_complete(&r->req, GOOD);
212 
213 done:
214     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
215     scsi_req_unref(&r->req);
216 }
217 
218 static bool scsi_is_cmd_fua(SCSICommand *cmd)
219 {
220     switch (cmd->buf[0]) {
221     case READ_10:
222     case READ_12:
223     case READ_16:
224     case WRITE_10:
225     case WRITE_12:
226     case WRITE_16:
227         return (cmd->buf[1] & 8) != 0;
228 
229     case VERIFY_10:
230     case VERIFY_12:
231     case VERIFY_16:
232     case WRITE_VERIFY_10:
233     case WRITE_VERIFY_12:
234     case WRITE_VERIFY_16:
235         return true;
236 
237     case READ_6:
238     case WRITE_6:
239     default:
240         return false;
241     }
242 }
243 
244 static void scsi_write_do_fua(SCSIDiskReq *r)
245 {
246     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
247 
248     assert(r->req.aiocb == NULL);
249     assert(!r->req.io_canceled);
250 
251     if (r->need_fua_emulation) {
252         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
253                          BLOCK_ACCT_FLUSH);
254         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
255         return;
256     }
257 
258     scsi_req_complete(&r->req, GOOD);
259     scsi_req_unref(&r->req);
260 }
261 
262 static void scsi_dma_complete_noio(SCSIDiskReq *r, int ret)
263 {
264     assert(r->req.aiocb == NULL);
265     if (scsi_disk_req_check_error(r, ret, false)) {
266         goto done;
267     }
268 
269     r->sector += r->sector_count;
270     r->sector_count = 0;
271     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
272         scsi_write_do_fua(r);
273         return;
274     } else {
275         scsi_req_complete(&r->req, GOOD);
276     }
277 
278 done:
279     scsi_req_unref(&r->req);
280 }
281 
282 static void scsi_dma_complete(void *opaque, int ret)
283 {
284     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
285     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
286 
287     assert(r->req.aiocb != NULL);
288     r->req.aiocb = NULL;
289 
290     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
291     if (ret < 0) {
292         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
293     } else {
294         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
295     }
296     scsi_dma_complete_noio(r, ret);
297     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
298 }
299 
300 static void scsi_read_complete_noio(SCSIDiskReq *r, int ret)
301 {
302     uint32_t n;
303 
304     assert(r->req.aiocb == NULL);
305     if (scsi_disk_req_check_error(r, ret, false)) {
306         goto done;
307     }
308 
309     n = r->qiov.size / BDRV_SECTOR_SIZE;
310     r->sector += n;
311     r->sector_count -= n;
312     scsi_req_data(&r->req, r->qiov.size);
313 
314 done:
315     scsi_req_unref(&r->req);
316 }
317 
318 static void scsi_read_complete(void *opaque, int ret)
319 {
320     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
321     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
322 
323     assert(r->req.aiocb != NULL);
324     r->req.aiocb = NULL;
325 
326     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
327     if (ret < 0) {
328         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
329     } else {
330         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
331         trace_scsi_disk_read_complete(r->req.tag, r->qiov.size);
332     }
333     scsi_read_complete_noio(r, ret);
334     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
335 }
336 
337 /* Actually issue a read to the block device.  */
338 static void scsi_do_read(SCSIDiskReq *r, int ret)
339 {
340     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
341     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
342 
343     assert (r->req.aiocb == NULL);
344     if (scsi_disk_req_check_error(r, ret, false)) {
345         goto done;
346     }
347 
348     /* The request is used as the AIO opaque value, so add a ref.  */
349     scsi_req_ref(&r->req);
350 
351     if (r->req.sg) {
352         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_READ);
353         r->req.resid -= r->req.sg->size;
354         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
355                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
356                                   BDRV_SECTOR_SIZE,
357                                   sdc->dma_readv, r, scsi_dma_complete, r,
358                                   DMA_DIRECTION_FROM_DEVICE);
359     } else {
360         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
361         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
362                          r->qiov.size, BLOCK_ACCT_READ);
363         r->req.aiocb = sdc->dma_readv(r->sector << BDRV_SECTOR_BITS, &r->qiov,
364                                       scsi_read_complete, r, r);
365     }
366 
367 done:
368     scsi_req_unref(&r->req);
369 }
370 
371 static void scsi_do_read_cb(void *opaque, int ret)
372 {
373     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
374     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
375 
376     assert (r->req.aiocb != NULL);
377     r->req.aiocb = NULL;
378 
379     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
380     if (ret < 0) {
381         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
382     } else {
383         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
384     }
385     scsi_do_read(opaque, ret);
386     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
387 }
388 
389 /* Read more data from scsi device into buffer.  */
390 static void scsi_read_data(SCSIRequest *req)
391 {
392     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
393     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
394     bool first;
395 
396     trace_scsi_disk_read_data_count(r->sector_count);
397     if (r->sector_count == 0) {
398         /* This also clears the sense buffer for REQUEST SENSE.  */
399         scsi_req_complete(&r->req, GOOD);
400         return;
401     }
402 
403     /* No data transfer may already be in progress */
404     assert(r->req.aiocb == NULL);
405 
406     /* The request is used as the AIO opaque value, so add a ref.  */
407     scsi_req_ref(&r->req);
408     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
409         trace_scsi_disk_read_data_invalid();
410         scsi_read_complete_noio(r, -EINVAL);
411         return;
412     }
413 
414     if (!blk_is_available(req->dev->conf.blk)) {
415         scsi_read_complete_noio(r, -ENOMEDIUM);
416         return;
417     }
418 
419     first = !r->started;
420     r->started = true;
421     if (first && r->need_fua_emulation) {
422         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
423                          BLOCK_ACCT_FLUSH);
424         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_do_read_cb, r);
425     } else {
426         scsi_do_read(r, 0);
427     }
428 }
429 
430 /*
431  * scsi_handle_rw_error has two return values.  False means that the error
432  * must be ignored, true means that the error has been processed and the
433  * caller should not do anything else for this request.  Note that
434  * scsi_handle_rw_error always manages its reference counts, independent
435  * of the return value.
436  */
437 static bool scsi_handle_rw_error(SCSIDiskReq *r, int error, bool acct_failed)
438 {
439     bool is_read = (r->req.cmd.mode == SCSI_XFER_FROM_DEV);
440     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
441     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
442     BlockErrorAction action = blk_get_error_action(s->qdev.conf.blk,
443                                                    is_read, error);
444 
445     if (action == BLOCK_ERROR_ACTION_REPORT) {
446         if (acct_failed) {
447             block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
448         }
449         switch (error) {
450         case 0:
451             /* A passthrough command has run and has produced sense data; check
452              * whether the error has to be handled by the guest or should rather
453              * pause the host.
454              */
455             assert(r->status && *r->status);
456             if (scsi_sense_buf_is_guest_recoverable(r->req.sense, sizeof(r->req.sense))) {
457                 /* These errors are handled by guest. */
458                 sdc->update_sense(&r->req);
459                 scsi_req_complete(&r->req, *r->status);
460                 return true;
461             }
462             error = scsi_sense_buf_to_errno(r->req.sense, sizeof(r->req.sense));
463             break;
464         case ENOMEDIUM:
465             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
466             break;
467         case ENOMEM:
468             scsi_check_condition(r, SENSE_CODE(TARGET_FAILURE));
469             break;
470         case EINVAL:
471             scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
472             break;
473         case ENOSPC:
474             scsi_check_condition(r, SENSE_CODE(SPACE_ALLOC_FAILED));
475             break;
476         default:
477             scsi_check_condition(r, SENSE_CODE(IO_ERROR));
478             break;
479         }
480     }
481 
482     blk_error_action(s->qdev.conf.blk, action, is_read, error);
483     if (action == BLOCK_ERROR_ACTION_IGNORE) {
484         scsi_req_complete(&r->req, 0);
485         return true;
486     }
487 
488     if (action == BLOCK_ERROR_ACTION_STOP) {
489         scsi_req_retry(&r->req);
490     }
491     return true;
492 }
493 
494 static void scsi_write_complete_noio(SCSIDiskReq *r, int ret)
495 {
496     uint32_t n;
497 
498     assert (r->req.aiocb == NULL);
499     if (scsi_disk_req_check_error(r, ret, false)) {
500         goto done;
501     }
502 
503     n = r->qiov.size / BDRV_SECTOR_SIZE;
504     r->sector += n;
505     r->sector_count -= n;
506     if (r->sector_count == 0) {
507         scsi_write_do_fua(r);
508         return;
509     } else {
510         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
511         trace_scsi_disk_write_complete_noio(r->req.tag, r->qiov.size);
512         scsi_req_data(&r->req, r->qiov.size);
513     }
514 
515 done:
516     scsi_req_unref(&r->req);
517 }
518 
519 static void scsi_write_complete(void * opaque, int ret)
520 {
521     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
522     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
523 
524     assert (r->req.aiocb != NULL);
525     r->req.aiocb = NULL;
526 
527     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
528     if (ret < 0) {
529         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
530     } else {
531         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
532     }
533     scsi_write_complete_noio(r, ret);
534     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
535 }
536 
537 static void scsi_write_data(SCSIRequest *req)
538 {
539     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
540     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
541     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
542 
543     /* No data transfer may already be in progress */
544     assert(r->req.aiocb == NULL);
545 
546     /* The request is used as the AIO opaque value, so add a ref.  */
547     scsi_req_ref(&r->req);
548     if (r->req.cmd.mode != SCSI_XFER_TO_DEV) {
549         trace_scsi_disk_write_data_invalid();
550         scsi_write_complete_noio(r, -EINVAL);
551         return;
552     }
553 
554     if (!r->req.sg && !r->qiov.size) {
555         /* Called for the first time.  Ask the driver to send us more data.  */
556         r->started = true;
557         scsi_write_complete_noio(r, 0);
558         return;
559     }
560     if (!blk_is_available(req->dev->conf.blk)) {
561         scsi_write_complete_noio(r, -ENOMEDIUM);
562         return;
563     }
564 
565     if (r->req.cmd.buf[0] == VERIFY_10 || r->req.cmd.buf[0] == VERIFY_12 ||
566         r->req.cmd.buf[0] == VERIFY_16) {
567         if (r->req.sg) {
568             scsi_dma_complete_noio(r, 0);
569         } else {
570             scsi_write_complete_noio(r, 0);
571         }
572         return;
573     }
574 
575     if (r->req.sg) {
576         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_WRITE);
577         r->req.resid -= r->req.sg->size;
578         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
579                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
580                                   BDRV_SECTOR_SIZE,
581                                   sdc->dma_writev, r, scsi_dma_complete, r,
582                                   DMA_DIRECTION_TO_DEVICE);
583     } else {
584         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
585                          r->qiov.size, BLOCK_ACCT_WRITE);
586         r->req.aiocb = sdc->dma_writev(r->sector << BDRV_SECTOR_BITS, &r->qiov,
587                                        scsi_write_complete, r, r);
588     }
589 }
590 
591 /* Return a pointer to the data buffer.  */
592 static uint8_t *scsi_get_buf(SCSIRequest *req)
593 {
594     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
595 
596     return (uint8_t *)r->iov.iov_base;
597 }
598 
599 static int scsi_disk_emulate_vpd_page(SCSIRequest *req, uint8_t *outbuf)
600 {
601     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
602     uint8_t page_code = req->cmd.buf[2];
603     int start, buflen = 0;
604 
605     outbuf[buflen++] = s->qdev.type & 0x1f;
606     outbuf[buflen++] = page_code;
607     outbuf[buflen++] = 0x00;
608     outbuf[buflen++] = 0x00;
609     start = buflen;
610 
611     switch (page_code) {
612     case 0x00: /* Supported page codes, mandatory */
613     {
614         trace_scsi_disk_emulate_vpd_page_00(req->cmd.xfer);
615         outbuf[buflen++] = 0x00; /* list of supported pages (this page) */
616         if (s->serial) {
617             outbuf[buflen++] = 0x80; /* unit serial number */
618         }
619         outbuf[buflen++] = 0x83; /* device identification */
620         if (s->qdev.type == TYPE_DISK) {
621             outbuf[buflen++] = 0xb0; /* block limits */
622             outbuf[buflen++] = 0xb1; /* block device characteristics */
623             outbuf[buflen++] = 0xb2; /* thin provisioning */
624         }
625         break;
626     }
627     case 0x80: /* Device serial number, optional */
628     {
629         int l;
630 
631         if (!s->serial) {
632             trace_scsi_disk_emulate_vpd_page_80_not_supported();
633             return -1;
634         }
635 
636         l = strlen(s->serial);
637         if (l > 36) {
638             l = 36;
639         }
640 
641         trace_scsi_disk_emulate_vpd_page_80(req->cmd.xfer);
642         memcpy(outbuf + buflen, s->serial, l);
643         buflen += l;
644         break;
645     }
646 
647     case 0x83: /* Device identification page, mandatory */
648     {
649         int id_len = s->device_id ? MIN(strlen(s->device_id), 255 - 8) : 0;
650 
651         trace_scsi_disk_emulate_vpd_page_83(req->cmd.xfer);
652 
653         if (id_len) {
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, s->device_id, id_len);
659             buflen += id_len;
660         }
661 
662         if (s->qdev.wwn) {
663             outbuf[buflen++] = 0x1; /* Binary */
664             outbuf[buflen++] = 0x3; /* NAA */
665             outbuf[buflen++] = 0;   /* reserved */
666             outbuf[buflen++] = 8;
667             stq_be_p(&outbuf[buflen], s->qdev.wwn);
668             buflen += 8;
669         }
670 
671         if (s->qdev.port_wwn) {
672             outbuf[buflen++] = 0x61; /* SAS / Binary */
673             outbuf[buflen++] = 0x93; /* PIV / Target port / NAA */
674             outbuf[buflen++] = 0;    /* reserved */
675             outbuf[buflen++] = 8;
676             stq_be_p(&outbuf[buflen], s->qdev.port_wwn);
677             buflen += 8;
678         }
679 
680         if (s->port_index) {
681             outbuf[buflen++] = 0x61; /* SAS / Binary */
682 
683             /* PIV/Target port/relative target port */
684             outbuf[buflen++] = 0x94;
685 
686             outbuf[buflen++] = 0;    /* reserved */
687             outbuf[buflen++] = 4;
688             stw_be_p(&outbuf[buflen + 2], s->port_index);
689             buflen += 4;
690         }
691         break;
692     }
693     case 0xb0: /* block limits */
694     {
695         SCSIBlockLimits bl = {};
696 
697         if (s->qdev.type == TYPE_ROM) {
698             trace_scsi_disk_emulate_vpd_page_b0_not_supported();
699             return -1;
700         }
701         bl.wsnz = 1;
702         bl.unmap_sectors =
703             s->qdev.conf.discard_granularity / s->qdev.blocksize;
704         bl.min_io_size =
705             s->qdev.conf.min_io_size / s->qdev.blocksize;
706         bl.opt_io_size =
707             s->qdev.conf.opt_io_size / s->qdev.blocksize;
708         bl.max_unmap_sectors =
709             s->max_unmap_size / s->qdev.blocksize;
710         bl.max_io_sectors =
711             s->max_io_size / s->qdev.blocksize;
712         /* 255 descriptors fit in 4 KiB with an 8-byte header */
713         bl.max_unmap_descr = 255;
714 
715         if (s->qdev.type == TYPE_DISK) {
716             int max_transfer_blk = blk_get_max_transfer(s->qdev.conf.blk);
717             int max_io_sectors_blk =
718                 max_transfer_blk / s->qdev.blocksize;
719 
720             bl.max_io_sectors =
721                 MIN_NON_ZERO(max_io_sectors_blk, bl.max_io_sectors);
722         }
723         buflen += scsi_emulate_block_limits(outbuf + buflen, &bl);
724         break;
725     }
726     case 0xb1: /* block device characteristics */
727     {
728         buflen = 0x40;
729         outbuf[4] = (s->rotation_rate >> 8) & 0xff;
730         outbuf[5] = s->rotation_rate & 0xff;
731         outbuf[6] = 0; /* PRODUCT TYPE */
732         outbuf[7] = 0; /* WABEREQ | WACEREQ | NOMINAL FORM FACTOR */
733         outbuf[8] = 0; /* VBULS */
734         break;
735     }
736     case 0xb2: /* thin provisioning */
737     {
738         buflen = 8;
739         outbuf[4] = 0;
740         outbuf[5] = 0xe0; /* unmap & write_same 10/16 all supported */
741         outbuf[6] = s->qdev.conf.discard_granularity ? 2 : 1;
742         outbuf[7] = 0;
743         break;
744     }
745     default:
746         return -1;
747     }
748     /* done with EVPD */
749     assert(buflen - start <= 255);
750     outbuf[start - 1] = buflen - start;
751     return buflen;
752 }
753 
754 static int scsi_disk_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf)
755 {
756     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
757     int buflen = 0;
758 
759     if (req->cmd.buf[1] & 0x1) {
760         /* Vital product data */
761         return scsi_disk_emulate_vpd_page(req, outbuf);
762     }
763 
764     /* Standard INQUIRY data */
765     if (req->cmd.buf[2] != 0) {
766         return -1;
767     }
768 
769     /* PAGE CODE == 0 */
770     buflen = req->cmd.xfer;
771     if (buflen > SCSI_MAX_INQUIRY_LEN) {
772         buflen = SCSI_MAX_INQUIRY_LEN;
773     }
774 
775     outbuf[0] = s->qdev.type & 0x1f;
776     outbuf[1] = (s->features & (1 << SCSI_DISK_F_REMOVABLE)) ? 0x80 : 0;
777 
778     strpadcpy((char *) &outbuf[16], 16, s->product, ' ');
779     strpadcpy((char *) &outbuf[8], 8, s->vendor, ' ');
780 
781     memset(&outbuf[32], 0, 4);
782     memcpy(&outbuf[32], s->version, MIN(4, strlen(s->version)));
783     /*
784      * We claim conformance to SPC-3, which is required for guests
785      * to ask for modern features like READ CAPACITY(16) or the
786      * block characteristics VPD page by default.  Not all of SPC-3
787      * is actually implemented, but we're good enough.
788      */
789     outbuf[2] = s->qdev.default_scsi_version;
790     outbuf[3] = 2 | 0x10; /* Format 2, HiSup */
791 
792     if (buflen > 36) {
793         outbuf[4] = buflen - 5; /* Additional Length = (Len - 1) - 4 */
794     } else {
795         /* If the allocation length of CDB is too small,
796                the additional length is not adjusted */
797         outbuf[4] = 36 - 5;
798     }
799 
800     /* Sync data transfer and TCQ.  */
801     outbuf[7] = 0x10 | (req->bus->info->tcq ? 0x02 : 0);
802     return buflen;
803 }
804 
805 static inline bool media_is_dvd(SCSIDiskState *s)
806 {
807     uint64_t nb_sectors;
808     if (s->qdev.type != TYPE_ROM) {
809         return false;
810     }
811     if (!blk_is_available(s->qdev.conf.blk)) {
812         return false;
813     }
814     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
815     return nb_sectors > CD_MAX_SECTORS;
816 }
817 
818 static inline bool media_is_cd(SCSIDiskState *s)
819 {
820     uint64_t nb_sectors;
821     if (s->qdev.type != TYPE_ROM) {
822         return false;
823     }
824     if (!blk_is_available(s->qdev.conf.blk)) {
825         return false;
826     }
827     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
828     return nb_sectors <= CD_MAX_SECTORS;
829 }
830 
831 static int scsi_read_disc_information(SCSIDiskState *s, SCSIDiskReq *r,
832                                       uint8_t *outbuf)
833 {
834     uint8_t type = r->req.cmd.buf[1] & 7;
835 
836     if (s->qdev.type != TYPE_ROM) {
837         return -1;
838     }
839 
840     /* Types 1/2 are only defined for Blu-Ray.  */
841     if (type != 0) {
842         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
843         return -1;
844     }
845 
846     memset(outbuf, 0, 34);
847     outbuf[1] = 32;
848     outbuf[2] = 0xe; /* last session complete, disc finalized */
849     outbuf[3] = 1;   /* first track on disc */
850     outbuf[4] = 1;   /* # of sessions */
851     outbuf[5] = 1;   /* first track of last session */
852     outbuf[6] = 1;   /* last track of last session */
853     outbuf[7] = 0x20; /* unrestricted use */
854     outbuf[8] = 0x00; /* CD-ROM or DVD-ROM */
855     /* 9-10-11: most significant byte corresponding bytes 4-5-6 */
856     /* 12-23: not meaningful for CD-ROM or DVD-ROM */
857     /* 24-31: disc bar code */
858     /* 32: disc application code */
859     /* 33: number of OPC tables */
860 
861     return 34;
862 }
863 
864 static int scsi_read_dvd_structure(SCSIDiskState *s, SCSIDiskReq *r,
865                                    uint8_t *outbuf)
866 {
867     static const int rds_caps_size[5] = {
868         [0] = 2048 + 4,
869         [1] = 4 + 4,
870         [3] = 188 + 4,
871         [4] = 2048 + 4,
872     };
873 
874     uint8_t media = r->req.cmd.buf[1];
875     uint8_t layer = r->req.cmd.buf[6];
876     uint8_t format = r->req.cmd.buf[7];
877     int size = -1;
878 
879     if (s->qdev.type != TYPE_ROM) {
880         return -1;
881     }
882     if (media != 0) {
883         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
884         return -1;
885     }
886 
887     if (format != 0xff) {
888         if (!blk_is_available(s->qdev.conf.blk)) {
889             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
890             return -1;
891         }
892         if (media_is_cd(s)) {
893             scsi_check_condition(r, SENSE_CODE(INCOMPATIBLE_FORMAT));
894             return -1;
895         }
896         if (format >= ARRAY_SIZE(rds_caps_size)) {
897             return -1;
898         }
899         size = rds_caps_size[format];
900         memset(outbuf, 0, size);
901     }
902 
903     switch (format) {
904     case 0x00: {
905         /* Physical format information */
906         uint64_t nb_sectors;
907         if (layer != 0) {
908             goto fail;
909         }
910         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
911 
912         outbuf[4] = 1;   /* DVD-ROM, part version 1 */
913         outbuf[5] = 0xf; /* 120mm disc, minimum rate unspecified */
914         outbuf[6] = 1;   /* one layer, read-only (per MMC-2 spec) */
915         outbuf[7] = 0;   /* default densities */
916 
917         stl_be_p(&outbuf[12], (nb_sectors >> 2) - 1); /* end sector */
918         stl_be_p(&outbuf[16], (nb_sectors >> 2) - 1); /* l0 end sector */
919         break;
920     }
921 
922     case 0x01: /* DVD copyright information, all zeros */
923         break;
924 
925     case 0x03: /* BCA information - invalid field for no BCA info */
926         return -1;
927 
928     case 0x04: /* DVD disc manufacturing information, all zeros */
929         break;
930 
931     case 0xff: { /* List capabilities */
932         int i;
933         size = 4;
934         for (i = 0; i < ARRAY_SIZE(rds_caps_size); i++) {
935             if (!rds_caps_size[i]) {
936                 continue;
937             }
938             outbuf[size] = i;
939             outbuf[size + 1] = 0x40; /* Not writable, readable */
940             stw_be_p(&outbuf[size + 2], rds_caps_size[i]);
941             size += 4;
942         }
943         break;
944      }
945 
946     default:
947         return -1;
948     }
949 
950     /* Size of buffer, not including 2 byte size field */
951     stw_be_p(outbuf, size - 2);
952     return size;
953 
954 fail:
955     return -1;
956 }
957 
958 static int scsi_event_status_media(SCSIDiskState *s, uint8_t *outbuf)
959 {
960     uint8_t event_code, media_status;
961 
962     media_status = 0;
963     if (s->tray_open) {
964         media_status = MS_TRAY_OPEN;
965     } else if (blk_is_inserted(s->qdev.conf.blk)) {
966         media_status = MS_MEDIA_PRESENT;
967     }
968 
969     /* Event notification descriptor */
970     event_code = MEC_NO_CHANGE;
971     if (media_status != MS_TRAY_OPEN) {
972         if (s->media_event) {
973             event_code = MEC_NEW_MEDIA;
974             s->media_event = false;
975         } else if (s->eject_request) {
976             event_code = MEC_EJECT_REQUESTED;
977             s->eject_request = false;
978         }
979     }
980 
981     outbuf[0] = event_code;
982     outbuf[1] = media_status;
983 
984     /* These fields are reserved, just clear them. */
985     outbuf[2] = 0;
986     outbuf[3] = 0;
987     return 4;
988 }
989 
990 static int scsi_get_event_status_notification(SCSIDiskState *s, SCSIDiskReq *r,
991                                               uint8_t *outbuf)
992 {
993     int size;
994     uint8_t *buf = r->req.cmd.buf;
995     uint8_t notification_class_request = buf[4];
996     if (s->qdev.type != TYPE_ROM) {
997         return -1;
998     }
999     if ((buf[1] & 1) == 0) {
1000         /* asynchronous */
1001         return -1;
1002     }
1003 
1004     size = 4;
1005     outbuf[0] = outbuf[1] = 0;
1006     outbuf[3] = 1 << GESN_MEDIA; /* supported events */
1007     if (notification_class_request & (1 << GESN_MEDIA)) {
1008         outbuf[2] = GESN_MEDIA;
1009         size += scsi_event_status_media(s, &outbuf[size]);
1010     } else {
1011         outbuf[2] = 0x80;
1012     }
1013     stw_be_p(outbuf, size - 4);
1014     return size;
1015 }
1016 
1017 static int scsi_get_configuration(SCSIDiskState *s, uint8_t *outbuf)
1018 {
1019     int current;
1020 
1021     if (s->qdev.type != TYPE_ROM) {
1022         return -1;
1023     }
1024 
1025     if (media_is_dvd(s)) {
1026         current = MMC_PROFILE_DVD_ROM;
1027     } else if (media_is_cd(s)) {
1028         current = MMC_PROFILE_CD_ROM;
1029     } else {
1030         current = MMC_PROFILE_NONE;
1031     }
1032 
1033     memset(outbuf, 0, 40);
1034     stl_be_p(&outbuf[0], 36); /* Bytes after the data length field */
1035     stw_be_p(&outbuf[6], current);
1036     /* outbuf[8] - outbuf[19]: Feature 0 - Profile list */
1037     outbuf[10] = 0x03; /* persistent, current */
1038     outbuf[11] = 8; /* two profiles */
1039     stw_be_p(&outbuf[12], MMC_PROFILE_DVD_ROM);
1040     outbuf[14] = (current == MMC_PROFILE_DVD_ROM);
1041     stw_be_p(&outbuf[16], MMC_PROFILE_CD_ROM);
1042     outbuf[18] = (current == MMC_PROFILE_CD_ROM);
1043     /* outbuf[20] - outbuf[31]: Feature 1 - Core feature */
1044     stw_be_p(&outbuf[20], 1);
1045     outbuf[22] = 0x08 | 0x03; /* version 2, persistent, current */
1046     outbuf[23] = 8;
1047     stl_be_p(&outbuf[24], 1); /* SCSI */
1048     outbuf[28] = 1; /* DBE = 1, mandatory */
1049     /* outbuf[32] - outbuf[39]: Feature 3 - Removable media feature */
1050     stw_be_p(&outbuf[32], 3);
1051     outbuf[34] = 0x08 | 0x03; /* version 2, persistent, current */
1052     outbuf[35] = 4;
1053     outbuf[36] = 0x39; /* tray, load=1, eject=1, unlocked at powerup, lock=1 */
1054     /* TODO: Random readable, CD read, DVD read, drive serial number,
1055        power management */
1056     return 40;
1057 }
1058 
1059 static int scsi_emulate_mechanism_status(SCSIDiskState *s, uint8_t *outbuf)
1060 {
1061     if (s->qdev.type != TYPE_ROM) {
1062         return -1;
1063     }
1064     memset(outbuf, 0, 8);
1065     outbuf[5] = 1; /* CD-ROM */
1066     return 8;
1067 }
1068 
1069 static int mode_sense_page(SCSIDiskState *s, int page, uint8_t **p_outbuf,
1070                            int page_control)
1071 {
1072     static const int mode_sense_valid[0x3f] = {
1073         [MODE_PAGE_HD_GEOMETRY]            = (1 << TYPE_DISK),
1074         [MODE_PAGE_FLEXIBLE_DISK_GEOMETRY] = (1 << TYPE_DISK),
1075         [MODE_PAGE_CACHING]                = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1076         [MODE_PAGE_R_W_ERROR]              = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1077         [MODE_PAGE_AUDIO_CTL]              = (1 << TYPE_ROM),
1078         [MODE_PAGE_CAPABILITIES]           = (1 << TYPE_ROM),
1079     };
1080 
1081     uint8_t *p = *p_outbuf + 2;
1082     int length;
1083 
1084     if ((mode_sense_valid[page] & (1 << s->qdev.type)) == 0) {
1085         return -1;
1086     }
1087 
1088     /*
1089      * If Changeable Values are requested, a mask denoting those mode parameters
1090      * that are changeable shall be returned. As we currently don't support
1091      * parameter changes via MODE_SELECT all bits are returned set to zero.
1092      * The buffer was already menset to zero by the caller of this function.
1093      *
1094      * The offsets here are off by two compared to the descriptions in the
1095      * SCSI specs, because those include a 2-byte header.  This is unfortunate,
1096      * but it is done so that offsets are consistent within our implementation
1097      * of MODE SENSE and MODE SELECT.  MODE SELECT has to deal with both
1098      * 2-byte and 4-byte headers.
1099      */
1100     switch (page) {
1101     case MODE_PAGE_HD_GEOMETRY:
1102         length = 0x16;
1103         if (page_control == 1) { /* Changeable Values */
1104             break;
1105         }
1106         /* if a geometry hint is available, use it */
1107         p[0] = (s->qdev.conf.cyls >> 16) & 0xff;
1108         p[1] = (s->qdev.conf.cyls >> 8) & 0xff;
1109         p[2] = s->qdev.conf.cyls & 0xff;
1110         p[3] = s->qdev.conf.heads & 0xff;
1111         /* Write precomp start cylinder, disabled */
1112         p[4] = (s->qdev.conf.cyls >> 16) & 0xff;
1113         p[5] = (s->qdev.conf.cyls >> 8) & 0xff;
1114         p[6] = s->qdev.conf.cyls & 0xff;
1115         /* Reduced current start cylinder, disabled */
1116         p[7] = (s->qdev.conf.cyls >> 16) & 0xff;
1117         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1118         p[9] = s->qdev.conf.cyls & 0xff;
1119         /* Device step rate [ns], 200ns */
1120         p[10] = 0;
1121         p[11] = 200;
1122         /* Landing zone cylinder */
1123         p[12] = 0xff;
1124         p[13] =  0xff;
1125         p[14] = 0xff;
1126         /* Medium rotation rate [rpm], 5400 rpm */
1127         p[18] = (5400 >> 8) & 0xff;
1128         p[19] = 5400 & 0xff;
1129         break;
1130 
1131     case MODE_PAGE_FLEXIBLE_DISK_GEOMETRY:
1132         length = 0x1e;
1133         if (page_control == 1) { /* Changeable Values */
1134             break;
1135         }
1136         /* Transfer rate [kbit/s], 5Mbit/s */
1137         p[0] = 5000 >> 8;
1138         p[1] = 5000 & 0xff;
1139         /* if a geometry hint is available, use it */
1140         p[2] = s->qdev.conf.heads & 0xff;
1141         p[3] = s->qdev.conf.secs & 0xff;
1142         p[4] = s->qdev.blocksize >> 8;
1143         p[6] = (s->qdev.conf.cyls >> 8) & 0xff;
1144         p[7] = s->qdev.conf.cyls & 0xff;
1145         /* Write precomp start cylinder, disabled */
1146         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1147         p[9] = s->qdev.conf.cyls & 0xff;
1148         /* Reduced current start cylinder, disabled */
1149         p[10] = (s->qdev.conf.cyls >> 8) & 0xff;
1150         p[11] = s->qdev.conf.cyls & 0xff;
1151         /* Device step rate [100us], 100us */
1152         p[12] = 0;
1153         p[13] = 1;
1154         /* Device step pulse width [us], 1us */
1155         p[14] = 1;
1156         /* Device head settle delay [100us], 100us */
1157         p[15] = 0;
1158         p[16] = 1;
1159         /* Motor on delay [0.1s], 0.1s */
1160         p[17] = 1;
1161         /* Motor off delay [0.1s], 0.1s */
1162         p[18] = 1;
1163         /* Medium rotation rate [rpm], 5400 rpm */
1164         p[26] = (5400 >> 8) & 0xff;
1165         p[27] = 5400 & 0xff;
1166         break;
1167 
1168     case MODE_PAGE_CACHING:
1169         length = 0x12;
1170         if (page_control == 1 || /* Changeable Values */
1171             blk_enable_write_cache(s->qdev.conf.blk)) {
1172             p[0] = 4; /* WCE */
1173         }
1174         break;
1175 
1176     case MODE_PAGE_R_W_ERROR:
1177         length = 10;
1178         if (page_control == 1) { /* Changeable Values */
1179             break;
1180         }
1181         p[0] = 0x80; /* Automatic Write Reallocation Enabled */
1182         if (s->qdev.type == TYPE_ROM) {
1183             p[1] = 0x20; /* Read Retry Count */
1184         }
1185         break;
1186 
1187     case MODE_PAGE_AUDIO_CTL:
1188         length = 14;
1189         break;
1190 
1191     case MODE_PAGE_CAPABILITIES:
1192         length = 0x14;
1193         if (page_control == 1) { /* Changeable Values */
1194             break;
1195         }
1196 
1197         p[0] = 0x3b; /* CD-R & CD-RW read */
1198         p[1] = 0; /* Writing not supported */
1199         p[2] = 0x7f; /* Audio, composite, digital out,
1200                         mode 2 form 1&2, multi session */
1201         p[3] = 0xff; /* CD DA, DA accurate, RW supported,
1202                         RW corrected, C2 errors, ISRC,
1203                         UPC, Bar code */
1204         p[4] = 0x2d | (s->tray_locked ? 2 : 0);
1205         /* Locking supported, jumper present, eject, tray */
1206         p[5] = 0; /* no volume & mute control, no
1207                      changer */
1208         p[6] = (50 * 176) >> 8; /* 50x read speed */
1209         p[7] = (50 * 176) & 0xff;
1210         p[8] = 2 >> 8; /* Two volume levels */
1211         p[9] = 2 & 0xff;
1212         p[10] = 2048 >> 8; /* 2M buffer */
1213         p[11] = 2048 & 0xff;
1214         p[12] = (16 * 176) >> 8; /* 16x read speed current */
1215         p[13] = (16 * 176) & 0xff;
1216         p[16] = (16 * 176) >> 8; /* 16x write speed */
1217         p[17] = (16 * 176) & 0xff;
1218         p[18] = (16 * 176) >> 8; /* 16x write speed current */
1219         p[19] = (16 * 176) & 0xff;
1220         break;
1221 
1222     default:
1223         return -1;
1224     }
1225 
1226     assert(length < 256);
1227     (*p_outbuf)[0] = page;
1228     (*p_outbuf)[1] = length;
1229     *p_outbuf += length + 2;
1230     return length + 2;
1231 }
1232 
1233 static int scsi_disk_emulate_mode_sense(SCSIDiskReq *r, uint8_t *outbuf)
1234 {
1235     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1236     uint64_t nb_sectors;
1237     bool dbd;
1238     int page, buflen, ret, page_control;
1239     uint8_t *p;
1240     uint8_t dev_specific_param;
1241 
1242     dbd = (r->req.cmd.buf[1] & 0x8) != 0;
1243     page = r->req.cmd.buf[2] & 0x3f;
1244     page_control = (r->req.cmd.buf[2] & 0xc0) >> 6;
1245 
1246     trace_scsi_disk_emulate_mode_sense((r->req.cmd.buf[0] == MODE_SENSE) ? 6 :
1247                                        10, page, r->req.cmd.xfer, page_control);
1248     memset(outbuf, 0, r->req.cmd.xfer);
1249     p = outbuf;
1250 
1251     if (s->qdev.type == TYPE_DISK) {
1252         dev_specific_param = s->features & (1 << SCSI_DISK_F_DPOFUA) ? 0x10 : 0;
1253         if (blk_is_read_only(s->qdev.conf.blk)) {
1254             dev_specific_param |= 0x80; /* Readonly.  */
1255         }
1256     } else {
1257         /* MMC prescribes that CD/DVD drives have no block descriptors,
1258          * and defines no device-specific parameter.  */
1259         dev_specific_param = 0x00;
1260         dbd = true;
1261     }
1262 
1263     if (r->req.cmd.buf[0] == MODE_SENSE) {
1264         p[1] = 0; /* Default media type.  */
1265         p[2] = dev_specific_param;
1266         p[3] = 0; /* Block descriptor length.  */
1267         p += 4;
1268     } else { /* MODE_SENSE_10 */
1269         p[2] = 0; /* Default media type.  */
1270         p[3] = dev_specific_param;
1271         p[6] = p[7] = 0; /* Block descriptor length.  */
1272         p += 8;
1273     }
1274 
1275     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1276     if (!dbd && nb_sectors) {
1277         if (r->req.cmd.buf[0] == MODE_SENSE) {
1278             outbuf[3] = 8; /* Block descriptor length  */
1279         } else { /* MODE_SENSE_10 */
1280             outbuf[7] = 8; /* Block descriptor length  */
1281         }
1282         nb_sectors /= (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1283         if (nb_sectors > 0xffffff) {
1284             nb_sectors = 0;
1285         }
1286         p[0] = 0; /* media density code */
1287         p[1] = (nb_sectors >> 16) & 0xff;
1288         p[2] = (nb_sectors >> 8) & 0xff;
1289         p[3] = nb_sectors & 0xff;
1290         p[4] = 0; /* reserved */
1291         p[5] = 0; /* bytes 5-7 are the sector size in bytes */
1292         p[6] = s->qdev.blocksize >> 8;
1293         p[7] = 0;
1294         p += 8;
1295     }
1296 
1297     if (page_control == 3) {
1298         /* Saved Values */
1299         scsi_check_condition(r, SENSE_CODE(SAVING_PARAMS_NOT_SUPPORTED));
1300         return -1;
1301     }
1302 
1303     if (page == 0x3f) {
1304         for (page = 0; page <= 0x3e; page++) {
1305             mode_sense_page(s, page, &p, page_control);
1306         }
1307     } else {
1308         ret = mode_sense_page(s, page, &p, page_control);
1309         if (ret == -1) {
1310             return -1;
1311         }
1312     }
1313 
1314     buflen = p - outbuf;
1315     /*
1316      * The mode data length field specifies the length in bytes of the
1317      * following data that is available to be transferred. The mode data
1318      * length does not include itself.
1319      */
1320     if (r->req.cmd.buf[0] == MODE_SENSE) {
1321         outbuf[0] = buflen - 1;
1322     } else { /* MODE_SENSE_10 */
1323         outbuf[0] = ((buflen - 2) >> 8) & 0xff;
1324         outbuf[1] = (buflen - 2) & 0xff;
1325     }
1326     return buflen;
1327 }
1328 
1329 static int scsi_disk_emulate_read_toc(SCSIRequest *req, uint8_t *outbuf)
1330 {
1331     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1332     int start_track, format, msf, toclen;
1333     uint64_t nb_sectors;
1334 
1335     msf = req->cmd.buf[1] & 2;
1336     format = req->cmd.buf[2] & 0xf;
1337     start_track = req->cmd.buf[6];
1338     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1339     trace_scsi_disk_emulate_read_toc(start_track, format, msf >> 1);
1340     nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
1341     switch (format) {
1342     case 0:
1343         toclen = cdrom_read_toc(nb_sectors, outbuf, msf, start_track);
1344         break;
1345     case 1:
1346         /* multi session : only a single session defined */
1347         toclen = 12;
1348         memset(outbuf, 0, 12);
1349         outbuf[1] = 0x0a;
1350         outbuf[2] = 0x01;
1351         outbuf[3] = 0x01;
1352         break;
1353     case 2:
1354         toclen = cdrom_read_toc_raw(nb_sectors, outbuf, msf, start_track);
1355         break;
1356     default:
1357         return -1;
1358     }
1359     return toclen;
1360 }
1361 
1362 static int scsi_disk_emulate_start_stop(SCSIDiskReq *r)
1363 {
1364     SCSIRequest *req = &r->req;
1365     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1366     bool start = req->cmd.buf[4] & 1;
1367     bool loej = req->cmd.buf[4] & 2; /* load on start, eject on !start */
1368     int pwrcnd = req->cmd.buf[4] & 0xf0;
1369 
1370     if (pwrcnd) {
1371         /* eject/load only happens for power condition == 0 */
1372         return 0;
1373     }
1374 
1375     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) && loej) {
1376         if (!start && !s->tray_open && s->tray_locked) {
1377             scsi_check_condition(r,
1378                                  blk_is_inserted(s->qdev.conf.blk)
1379                                  ? SENSE_CODE(ILLEGAL_REQ_REMOVAL_PREVENTED)
1380                                  : SENSE_CODE(NOT_READY_REMOVAL_PREVENTED));
1381             return -1;
1382         }
1383 
1384         if (s->tray_open != !start) {
1385             blk_eject(s->qdev.conf.blk, !start);
1386             s->tray_open = !start;
1387         }
1388     }
1389     return 0;
1390 }
1391 
1392 static void scsi_disk_emulate_read_data(SCSIRequest *req)
1393 {
1394     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1395     int buflen = r->iov.iov_len;
1396 
1397     if (buflen) {
1398         trace_scsi_disk_emulate_read_data(buflen);
1399         r->iov.iov_len = 0;
1400         r->started = true;
1401         scsi_req_data(&r->req, buflen);
1402         return;
1403     }
1404 
1405     /* This also clears the sense buffer for REQUEST SENSE.  */
1406     scsi_req_complete(&r->req, GOOD);
1407 }
1408 
1409 static int scsi_disk_check_mode_select(SCSIDiskState *s, int page,
1410                                        uint8_t *inbuf, int inlen)
1411 {
1412     uint8_t mode_current[SCSI_MAX_MODE_LEN];
1413     uint8_t mode_changeable[SCSI_MAX_MODE_LEN];
1414     uint8_t *p;
1415     int len, expected_len, changeable_len, i;
1416 
1417     /* The input buffer does not include the page header, so it is
1418      * off by 2 bytes.
1419      */
1420     expected_len = inlen + 2;
1421     if (expected_len > SCSI_MAX_MODE_LEN) {
1422         return -1;
1423     }
1424 
1425     p = mode_current;
1426     memset(mode_current, 0, inlen + 2);
1427     len = mode_sense_page(s, page, &p, 0);
1428     if (len < 0 || len != expected_len) {
1429         return -1;
1430     }
1431 
1432     p = mode_changeable;
1433     memset(mode_changeable, 0, inlen + 2);
1434     changeable_len = mode_sense_page(s, page, &p, 1);
1435     assert(changeable_len == len);
1436 
1437     /* Check that unchangeable bits are the same as what MODE SENSE
1438      * would return.
1439      */
1440     for (i = 2; i < len; i++) {
1441         if (((mode_current[i] ^ inbuf[i - 2]) & ~mode_changeable[i]) != 0) {
1442             return -1;
1443         }
1444     }
1445     return 0;
1446 }
1447 
1448 static void scsi_disk_apply_mode_select(SCSIDiskState *s, int page, uint8_t *p)
1449 {
1450     switch (page) {
1451     case MODE_PAGE_CACHING:
1452         blk_set_enable_write_cache(s->qdev.conf.blk, (p[0] & 4) != 0);
1453         break;
1454 
1455     default:
1456         break;
1457     }
1458 }
1459 
1460 static int mode_select_pages(SCSIDiskReq *r, uint8_t *p, int len, bool change)
1461 {
1462     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1463 
1464     while (len > 0) {
1465         int page, subpage, page_len;
1466 
1467         /* Parse both possible formats for the mode page headers.  */
1468         page = p[0] & 0x3f;
1469         if (p[0] & 0x40) {
1470             if (len < 4) {
1471                 goto invalid_param_len;
1472             }
1473             subpage = p[1];
1474             page_len = lduw_be_p(&p[2]);
1475             p += 4;
1476             len -= 4;
1477         } else {
1478             if (len < 2) {
1479                 goto invalid_param_len;
1480             }
1481             subpage = 0;
1482             page_len = p[1];
1483             p += 2;
1484             len -= 2;
1485         }
1486 
1487         if (subpage) {
1488             goto invalid_param;
1489         }
1490         if (page_len > len) {
1491             goto invalid_param_len;
1492         }
1493 
1494         if (!change) {
1495             if (scsi_disk_check_mode_select(s, page, p, page_len) < 0) {
1496                 goto invalid_param;
1497             }
1498         } else {
1499             scsi_disk_apply_mode_select(s, page, p);
1500         }
1501 
1502         p += page_len;
1503         len -= page_len;
1504     }
1505     return 0;
1506 
1507 invalid_param:
1508     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1509     return -1;
1510 
1511 invalid_param_len:
1512     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1513     return -1;
1514 }
1515 
1516 static void scsi_disk_emulate_mode_select(SCSIDiskReq *r, uint8_t *inbuf)
1517 {
1518     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1519     uint8_t *p = inbuf;
1520     int cmd = r->req.cmd.buf[0];
1521     int len = r->req.cmd.xfer;
1522     int hdr_len = (cmd == MODE_SELECT ? 4 : 8);
1523     int bd_len;
1524     int pass;
1525 
1526     /* We only support PF=1, SP=0.  */
1527     if ((r->req.cmd.buf[1] & 0x11) != 0x10) {
1528         goto invalid_field;
1529     }
1530 
1531     if (len < hdr_len) {
1532         goto invalid_param_len;
1533     }
1534 
1535     bd_len = (cmd == MODE_SELECT ? p[3] : lduw_be_p(&p[6]));
1536     len -= hdr_len;
1537     p += hdr_len;
1538     if (len < bd_len) {
1539         goto invalid_param_len;
1540     }
1541     if (bd_len != 0 && bd_len != 8) {
1542         goto invalid_param;
1543     }
1544 
1545     len -= bd_len;
1546     p += bd_len;
1547 
1548     /* Ensure no change is made if there is an error!  */
1549     for (pass = 0; pass < 2; pass++) {
1550         if (mode_select_pages(r, p, len, pass == 1) < 0) {
1551             assert(pass == 0);
1552             return;
1553         }
1554     }
1555     if (!blk_enable_write_cache(s->qdev.conf.blk)) {
1556         /* The request is used as the AIO opaque value, so add a ref.  */
1557         scsi_req_ref(&r->req);
1558         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
1559                          BLOCK_ACCT_FLUSH);
1560         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
1561         return;
1562     }
1563 
1564     scsi_req_complete(&r->req, GOOD);
1565     return;
1566 
1567 invalid_param:
1568     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1569     return;
1570 
1571 invalid_param_len:
1572     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1573     return;
1574 
1575 invalid_field:
1576     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1577 }
1578 
1579 static inline bool check_lba_range(SCSIDiskState *s,
1580                                    uint64_t sector_num, uint32_t nb_sectors)
1581 {
1582     /*
1583      * The first line tests that no overflow happens when computing the last
1584      * sector.  The second line tests that the last accessed sector is in
1585      * range.
1586      *
1587      * Careful, the computations should not underflow for nb_sectors == 0,
1588      * and a 0-block read to the first LBA beyond the end of device is
1589      * valid.
1590      */
1591     return (sector_num <= sector_num + nb_sectors &&
1592             sector_num + nb_sectors <= s->qdev.max_lba + 1);
1593 }
1594 
1595 typedef struct UnmapCBData {
1596     SCSIDiskReq *r;
1597     uint8_t *inbuf;
1598     int count;
1599 } UnmapCBData;
1600 
1601 static void scsi_unmap_complete(void *opaque, int ret);
1602 
1603 static void scsi_unmap_complete_noio(UnmapCBData *data, int ret)
1604 {
1605     SCSIDiskReq *r = data->r;
1606     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1607 
1608     assert(r->req.aiocb == NULL);
1609 
1610     if (data->count > 0) {
1611         r->sector = ldq_be_p(&data->inbuf[0])
1612             * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1613         r->sector_count = (ldl_be_p(&data->inbuf[8]) & 0xffffffffULL)
1614             * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1615         if (!check_lba_range(s, r->sector, r->sector_count)) {
1616             block_acct_invalid(blk_get_stats(s->qdev.conf.blk),
1617                                BLOCK_ACCT_UNMAP);
1618             scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1619             goto done;
1620         }
1621 
1622         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1623                          r->sector_count * BDRV_SECTOR_SIZE,
1624                          BLOCK_ACCT_UNMAP);
1625 
1626         r->req.aiocb = blk_aio_pdiscard(s->qdev.conf.blk,
1627                                         r->sector * BDRV_SECTOR_SIZE,
1628                                         r->sector_count * BDRV_SECTOR_SIZE,
1629                                         scsi_unmap_complete, data);
1630         data->count--;
1631         data->inbuf += 16;
1632         return;
1633     }
1634 
1635     scsi_req_complete(&r->req, GOOD);
1636 
1637 done:
1638     scsi_req_unref(&r->req);
1639     g_free(data);
1640 }
1641 
1642 static void scsi_unmap_complete(void *opaque, int ret)
1643 {
1644     UnmapCBData *data = opaque;
1645     SCSIDiskReq *r = data->r;
1646     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1647 
1648     assert(r->req.aiocb != NULL);
1649     r->req.aiocb = NULL;
1650 
1651     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
1652     if (scsi_disk_req_check_error(r, ret, true)) {
1653         scsi_req_unref(&r->req);
1654         g_free(data);
1655     } else {
1656         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1657         scsi_unmap_complete_noio(data, ret);
1658     }
1659     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1660 }
1661 
1662 static void scsi_disk_emulate_unmap(SCSIDiskReq *r, uint8_t *inbuf)
1663 {
1664     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1665     uint8_t *p = inbuf;
1666     int len = r->req.cmd.xfer;
1667     UnmapCBData *data;
1668 
1669     /* Reject ANCHOR=1.  */
1670     if (r->req.cmd.buf[1] & 0x1) {
1671         goto invalid_field;
1672     }
1673 
1674     if (len < 8) {
1675         goto invalid_param_len;
1676     }
1677     if (len < lduw_be_p(&p[0]) + 2) {
1678         goto invalid_param_len;
1679     }
1680     if (len < lduw_be_p(&p[2]) + 8) {
1681         goto invalid_param_len;
1682     }
1683     if (lduw_be_p(&p[2]) & 15) {
1684         goto invalid_param_len;
1685     }
1686 
1687     if (blk_is_read_only(s->qdev.conf.blk)) {
1688         block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1689         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1690         return;
1691     }
1692 
1693     data = g_new0(UnmapCBData, 1);
1694     data->r = r;
1695     data->inbuf = &p[8];
1696     data->count = lduw_be_p(&p[2]) >> 4;
1697 
1698     /* The matching unref is in scsi_unmap_complete, before data is freed.  */
1699     scsi_req_ref(&r->req);
1700     scsi_unmap_complete_noio(data, 0);
1701     return;
1702 
1703 invalid_param_len:
1704     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1705     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1706     return;
1707 
1708 invalid_field:
1709     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1710     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1711 }
1712 
1713 typedef struct WriteSameCBData {
1714     SCSIDiskReq *r;
1715     int64_t sector;
1716     int nb_sectors;
1717     QEMUIOVector qiov;
1718     struct iovec iov;
1719 } WriteSameCBData;
1720 
1721 static void scsi_write_same_complete(void *opaque, int ret)
1722 {
1723     WriteSameCBData *data = opaque;
1724     SCSIDiskReq *r = data->r;
1725     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1726 
1727     assert(r->req.aiocb != NULL);
1728     r->req.aiocb = NULL;
1729     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
1730     if (scsi_disk_req_check_error(r, ret, true)) {
1731         goto done;
1732     }
1733 
1734     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1735 
1736     data->nb_sectors -= data->iov.iov_len / BDRV_SECTOR_SIZE;
1737     data->sector += data->iov.iov_len / BDRV_SECTOR_SIZE;
1738     data->iov.iov_len = MIN(data->nb_sectors * BDRV_SECTOR_SIZE,
1739                             data->iov.iov_len);
1740     if (data->iov.iov_len) {
1741         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1742                          data->iov.iov_len, BLOCK_ACCT_WRITE);
1743         /* Reinitialize qiov, to handle unaligned WRITE SAME request
1744          * where final qiov may need smaller size */
1745         qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1746         r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1747                                        data->sector << BDRV_SECTOR_BITS,
1748                                        &data->qiov, 0,
1749                                        scsi_write_same_complete, data);
1750         aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1751         return;
1752     }
1753 
1754     scsi_req_complete(&r->req, GOOD);
1755 
1756 done:
1757     scsi_req_unref(&r->req);
1758     qemu_vfree(data->iov.iov_base);
1759     g_free(data);
1760     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1761 }
1762 
1763 static void scsi_disk_emulate_write_same(SCSIDiskReq *r, uint8_t *inbuf)
1764 {
1765     SCSIRequest *req = &r->req;
1766     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1767     uint32_t nb_sectors = scsi_data_cdb_xfer(r->req.cmd.buf);
1768     WriteSameCBData *data;
1769     uint8_t *buf;
1770     int i;
1771 
1772     /* Fail if PBDATA=1 or LBDATA=1 or ANCHOR=1.  */
1773     if (nb_sectors == 0 || (req->cmd.buf[1] & 0x16)) {
1774         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1775         return;
1776     }
1777 
1778     if (blk_is_read_only(s->qdev.conf.blk)) {
1779         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1780         return;
1781     }
1782     if (!check_lba_range(s, r->req.cmd.lba, nb_sectors)) {
1783         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1784         return;
1785     }
1786 
1787     if ((req->cmd.buf[1] & 0x1) || buffer_is_zero(inbuf, s->qdev.blocksize)) {
1788         int flags = (req->cmd.buf[1] & 0x8) ? BDRV_REQ_MAY_UNMAP : 0;
1789 
1790         /* The request is used as the AIO opaque value, so add a ref.  */
1791         scsi_req_ref(&r->req);
1792         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1793                          nb_sectors * s->qdev.blocksize,
1794                         BLOCK_ACCT_WRITE);
1795         r->req.aiocb = blk_aio_pwrite_zeroes(s->qdev.conf.blk,
1796                                 r->req.cmd.lba * s->qdev.blocksize,
1797                                 nb_sectors * s->qdev.blocksize,
1798                                 flags, scsi_aio_complete, r);
1799         return;
1800     }
1801 
1802     data = g_new0(WriteSameCBData, 1);
1803     data->r = r;
1804     data->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1805     data->nb_sectors = nb_sectors * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1806     data->iov.iov_len = MIN(data->nb_sectors * BDRV_SECTOR_SIZE,
1807                             SCSI_WRITE_SAME_MAX);
1808     data->iov.iov_base = buf = blk_blockalign(s->qdev.conf.blk,
1809                                               data->iov.iov_len);
1810     qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1811 
1812     for (i = 0; i < data->iov.iov_len; i += s->qdev.blocksize) {
1813         memcpy(&buf[i], inbuf, s->qdev.blocksize);
1814     }
1815 
1816     scsi_req_ref(&r->req);
1817     block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1818                      data->iov.iov_len, BLOCK_ACCT_WRITE);
1819     r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1820                                    data->sector << BDRV_SECTOR_BITS,
1821                                    &data->qiov, 0,
1822                                    scsi_write_same_complete, data);
1823 }
1824 
1825 static void scsi_disk_emulate_write_data(SCSIRequest *req)
1826 {
1827     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1828 
1829     if (r->iov.iov_len) {
1830         int buflen = r->iov.iov_len;
1831         trace_scsi_disk_emulate_write_data(buflen);
1832         r->iov.iov_len = 0;
1833         scsi_req_data(&r->req, buflen);
1834         return;
1835     }
1836 
1837     switch (req->cmd.buf[0]) {
1838     case MODE_SELECT:
1839     case MODE_SELECT_10:
1840         /* This also clears the sense buffer for REQUEST SENSE.  */
1841         scsi_disk_emulate_mode_select(r, r->iov.iov_base);
1842         break;
1843 
1844     case UNMAP:
1845         scsi_disk_emulate_unmap(r, r->iov.iov_base);
1846         break;
1847 
1848     case VERIFY_10:
1849     case VERIFY_12:
1850     case VERIFY_16:
1851         if (r->req.status == -1) {
1852             scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1853         }
1854         break;
1855 
1856     case WRITE_SAME_10:
1857     case WRITE_SAME_16:
1858         scsi_disk_emulate_write_same(r, r->iov.iov_base);
1859         break;
1860 
1861     default:
1862         abort();
1863     }
1864 }
1865 
1866 static int32_t scsi_disk_emulate_command(SCSIRequest *req, uint8_t *buf)
1867 {
1868     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1869     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1870     uint64_t nb_sectors;
1871     uint8_t *outbuf;
1872     int buflen;
1873 
1874     switch (req->cmd.buf[0]) {
1875     case INQUIRY:
1876     case MODE_SENSE:
1877     case MODE_SENSE_10:
1878     case RESERVE:
1879     case RESERVE_10:
1880     case RELEASE:
1881     case RELEASE_10:
1882     case START_STOP:
1883     case ALLOW_MEDIUM_REMOVAL:
1884     case GET_CONFIGURATION:
1885     case GET_EVENT_STATUS_NOTIFICATION:
1886     case MECHANISM_STATUS:
1887     case REQUEST_SENSE:
1888         break;
1889 
1890     default:
1891         if (!blk_is_available(s->qdev.conf.blk)) {
1892             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
1893             return 0;
1894         }
1895         break;
1896     }
1897 
1898     /*
1899      * FIXME: we shouldn't return anything bigger than 4k, but the code
1900      * requires the buffer to be as big as req->cmd.xfer in several
1901      * places.  So, do not allow CDBs with a very large ALLOCATION
1902      * LENGTH.  The real fix would be to modify scsi_read_data and
1903      * dma_buf_read, so that they return data beyond the buflen
1904      * as all zeros.
1905      */
1906     if (req->cmd.xfer > 65536) {
1907         goto illegal_request;
1908     }
1909     r->buflen = MAX(4096, req->cmd.xfer);
1910 
1911     if (!r->iov.iov_base) {
1912         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
1913     }
1914 
1915     outbuf = r->iov.iov_base;
1916     memset(outbuf, 0, r->buflen);
1917     switch (req->cmd.buf[0]) {
1918     case TEST_UNIT_READY:
1919         assert(blk_is_available(s->qdev.conf.blk));
1920         break;
1921     case INQUIRY:
1922         buflen = scsi_disk_emulate_inquiry(req, outbuf);
1923         if (buflen < 0) {
1924             goto illegal_request;
1925         }
1926         break;
1927     case MODE_SENSE:
1928     case MODE_SENSE_10:
1929         buflen = scsi_disk_emulate_mode_sense(r, outbuf);
1930         if (buflen < 0) {
1931             goto illegal_request;
1932         }
1933         break;
1934     case READ_TOC:
1935         buflen = scsi_disk_emulate_read_toc(req, outbuf);
1936         if (buflen < 0) {
1937             goto illegal_request;
1938         }
1939         break;
1940     case RESERVE:
1941         if (req->cmd.buf[1] & 1) {
1942             goto illegal_request;
1943         }
1944         break;
1945     case RESERVE_10:
1946         if (req->cmd.buf[1] & 3) {
1947             goto illegal_request;
1948         }
1949         break;
1950     case RELEASE:
1951         if (req->cmd.buf[1] & 1) {
1952             goto illegal_request;
1953         }
1954         break;
1955     case RELEASE_10:
1956         if (req->cmd.buf[1] & 3) {
1957             goto illegal_request;
1958         }
1959         break;
1960     case START_STOP:
1961         if (scsi_disk_emulate_start_stop(r) < 0) {
1962             return 0;
1963         }
1964         break;
1965     case ALLOW_MEDIUM_REMOVAL:
1966         s->tray_locked = req->cmd.buf[4] & 1;
1967         blk_lock_medium(s->qdev.conf.blk, req->cmd.buf[4] & 1);
1968         break;
1969     case READ_CAPACITY_10:
1970         /* The normal LEN field for this command is zero.  */
1971         memset(outbuf, 0, 8);
1972         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1973         if (!nb_sectors) {
1974             scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
1975             return 0;
1976         }
1977         if ((req->cmd.buf[8] & 1) == 0 && req->cmd.lba) {
1978             goto illegal_request;
1979         }
1980         nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
1981         /* Returned value is the address of the last sector.  */
1982         nb_sectors--;
1983         /* Remember the new size for read/write sanity checking. */
1984         s->qdev.max_lba = nb_sectors;
1985         /* Clip to 2TB, instead of returning capacity modulo 2TB. */
1986         if (nb_sectors > UINT32_MAX) {
1987             nb_sectors = UINT32_MAX;
1988         }
1989         outbuf[0] = (nb_sectors >> 24) & 0xff;
1990         outbuf[1] = (nb_sectors >> 16) & 0xff;
1991         outbuf[2] = (nb_sectors >> 8) & 0xff;
1992         outbuf[3] = nb_sectors & 0xff;
1993         outbuf[4] = 0;
1994         outbuf[5] = 0;
1995         outbuf[6] = s->qdev.blocksize >> 8;
1996         outbuf[7] = 0;
1997         break;
1998     case REQUEST_SENSE:
1999         /* Just return "NO SENSE".  */
2000         buflen = scsi_convert_sense(NULL, 0, outbuf, r->buflen,
2001                                     (req->cmd.buf[1] & 1) == 0);
2002         if (buflen < 0) {
2003             goto illegal_request;
2004         }
2005         break;
2006     case MECHANISM_STATUS:
2007         buflen = scsi_emulate_mechanism_status(s, outbuf);
2008         if (buflen < 0) {
2009             goto illegal_request;
2010         }
2011         break;
2012     case GET_CONFIGURATION:
2013         buflen = scsi_get_configuration(s, outbuf);
2014         if (buflen < 0) {
2015             goto illegal_request;
2016         }
2017         break;
2018     case GET_EVENT_STATUS_NOTIFICATION:
2019         buflen = scsi_get_event_status_notification(s, r, outbuf);
2020         if (buflen < 0) {
2021             goto illegal_request;
2022         }
2023         break;
2024     case READ_DISC_INFORMATION:
2025         buflen = scsi_read_disc_information(s, r, outbuf);
2026         if (buflen < 0) {
2027             goto illegal_request;
2028         }
2029         break;
2030     case READ_DVD_STRUCTURE:
2031         buflen = scsi_read_dvd_structure(s, r, outbuf);
2032         if (buflen < 0) {
2033             goto illegal_request;
2034         }
2035         break;
2036     case SERVICE_ACTION_IN_16:
2037         /* Service Action In subcommands. */
2038         if ((req->cmd.buf[1] & 31) == SAI_READ_CAPACITY_16) {
2039             trace_scsi_disk_emulate_command_SAI_16();
2040             memset(outbuf, 0, req->cmd.xfer);
2041             blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2042             if (!nb_sectors) {
2043                 scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
2044                 return 0;
2045             }
2046             if ((req->cmd.buf[14] & 1) == 0 && req->cmd.lba) {
2047                 goto illegal_request;
2048             }
2049             nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2050             /* Returned value is the address of the last sector.  */
2051             nb_sectors--;
2052             /* Remember the new size for read/write sanity checking. */
2053             s->qdev.max_lba = nb_sectors;
2054             outbuf[0] = (nb_sectors >> 56) & 0xff;
2055             outbuf[1] = (nb_sectors >> 48) & 0xff;
2056             outbuf[2] = (nb_sectors >> 40) & 0xff;
2057             outbuf[3] = (nb_sectors >> 32) & 0xff;
2058             outbuf[4] = (nb_sectors >> 24) & 0xff;
2059             outbuf[5] = (nb_sectors >> 16) & 0xff;
2060             outbuf[6] = (nb_sectors >> 8) & 0xff;
2061             outbuf[7] = nb_sectors & 0xff;
2062             outbuf[8] = 0;
2063             outbuf[9] = 0;
2064             outbuf[10] = s->qdev.blocksize >> 8;
2065             outbuf[11] = 0;
2066             outbuf[12] = 0;
2067             outbuf[13] = get_physical_block_exp(&s->qdev.conf);
2068 
2069             /* set TPE bit if the format supports discard */
2070             if (s->qdev.conf.discard_granularity) {
2071                 outbuf[14] = 0x80;
2072             }
2073 
2074             /* Protection, exponent and lowest lba field left blank. */
2075             break;
2076         }
2077         trace_scsi_disk_emulate_command_SAI_unsupported();
2078         goto illegal_request;
2079     case SYNCHRONIZE_CACHE:
2080         /* The request is used as the AIO opaque value, so add a ref.  */
2081         scsi_req_ref(&r->req);
2082         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
2083                          BLOCK_ACCT_FLUSH);
2084         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
2085         return 0;
2086     case SEEK_10:
2087         trace_scsi_disk_emulate_command_SEEK_10(r->req.cmd.lba);
2088         if (r->req.cmd.lba > s->qdev.max_lba) {
2089             goto illegal_lba;
2090         }
2091         break;
2092     case MODE_SELECT:
2093         trace_scsi_disk_emulate_command_MODE_SELECT(r->req.cmd.xfer);
2094         break;
2095     case MODE_SELECT_10:
2096         trace_scsi_disk_emulate_command_MODE_SELECT_10(r->req.cmd.xfer);
2097         break;
2098     case UNMAP:
2099         trace_scsi_disk_emulate_command_UNMAP(r->req.cmd.xfer);
2100         break;
2101     case VERIFY_10:
2102     case VERIFY_12:
2103     case VERIFY_16:
2104         trace_scsi_disk_emulate_command_VERIFY((req->cmd.buf[1] >> 1) & 3);
2105         if (req->cmd.buf[1] & 6) {
2106             goto illegal_request;
2107         }
2108         break;
2109     case WRITE_SAME_10:
2110     case WRITE_SAME_16:
2111         trace_scsi_disk_emulate_command_WRITE_SAME(
2112                 req->cmd.buf[0] == WRITE_SAME_10 ? 10 : 16, r->req.cmd.xfer);
2113         break;
2114     default:
2115         trace_scsi_disk_emulate_command_UNKNOWN(buf[0],
2116                                                 scsi_command_name(buf[0]));
2117         scsi_check_condition(r, SENSE_CODE(INVALID_OPCODE));
2118         return 0;
2119     }
2120     assert(!r->req.aiocb);
2121     r->iov.iov_len = MIN(r->buflen, req->cmd.xfer);
2122     if (r->iov.iov_len == 0) {
2123         scsi_req_complete(&r->req, GOOD);
2124     }
2125     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2126         assert(r->iov.iov_len == req->cmd.xfer);
2127         return -r->iov.iov_len;
2128     } else {
2129         return r->iov.iov_len;
2130     }
2131 
2132 illegal_request:
2133     if (r->req.status == -1) {
2134         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2135     }
2136     return 0;
2137 
2138 illegal_lba:
2139     scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2140     return 0;
2141 }
2142 
2143 /* Execute a scsi command.  Returns the length of the data expected by the
2144    command.  This will be Positive for data transfers from the device
2145    (eg. disk reads), negative for transfers to the device (eg. disk writes),
2146    and zero if the command does not transfer any data.  */
2147 
2148 static int32_t scsi_disk_dma_command(SCSIRequest *req, uint8_t *buf)
2149 {
2150     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
2151     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2152     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
2153     uint32_t len;
2154     uint8_t command;
2155 
2156     command = buf[0];
2157 
2158     if (!blk_is_available(s->qdev.conf.blk)) {
2159         scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
2160         return 0;
2161     }
2162 
2163     len = scsi_data_cdb_xfer(r->req.cmd.buf);
2164     switch (command) {
2165     case READ_6:
2166     case READ_10:
2167     case READ_12:
2168     case READ_16:
2169         trace_scsi_disk_dma_command_READ(r->req.cmd.lba, len);
2170         /* Protection information is not supported.  For SCSI versions 2 and
2171          * older (as determined by snooping the guest's INQUIRY commands),
2172          * there is no RD/WR/VRPROTECT, so skip this check in these versions.
2173          */
2174         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2175             goto illegal_request;
2176         }
2177         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2178             goto illegal_lba;
2179         }
2180         r->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2181         r->sector_count = len * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2182         break;
2183     case WRITE_6:
2184     case WRITE_10:
2185     case WRITE_12:
2186     case WRITE_16:
2187     case WRITE_VERIFY_10:
2188     case WRITE_VERIFY_12:
2189     case WRITE_VERIFY_16:
2190         if (blk_is_read_only(s->qdev.conf.blk)) {
2191             scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
2192             return 0;
2193         }
2194         trace_scsi_disk_dma_command_WRITE(
2195                 (command & 0xe) == 0xe ? "And Verify " : "",
2196                 r->req.cmd.lba, len);
2197         /* fall through */
2198     case VERIFY_10:
2199     case VERIFY_12:
2200     case VERIFY_16:
2201         /* We get here only for BYTCHK == 0x01 and only for scsi-block.
2202          * As far as DMA is concerned, we can treat it the same as a write;
2203          * scsi_block_do_sgio will send VERIFY commands.
2204          */
2205         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2206             goto illegal_request;
2207         }
2208         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2209             goto illegal_lba;
2210         }
2211         r->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2212         r->sector_count = len * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2213         break;
2214     default:
2215         abort();
2216     illegal_request:
2217         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2218         return 0;
2219     illegal_lba:
2220         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2221         return 0;
2222     }
2223     r->need_fua_emulation = sdc->need_fua_emulation(&r->req.cmd);
2224     if (r->sector_count == 0) {
2225         scsi_req_complete(&r->req, GOOD);
2226     }
2227     assert(r->iov.iov_len == 0);
2228     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2229         return -r->sector_count * BDRV_SECTOR_SIZE;
2230     } else {
2231         return r->sector_count * BDRV_SECTOR_SIZE;
2232     }
2233 }
2234 
2235 static void scsi_disk_reset(DeviceState *dev)
2236 {
2237     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev.qdev, dev);
2238     uint64_t nb_sectors;
2239 
2240     scsi_device_purge_requests(&s->qdev, SENSE_CODE(RESET));
2241 
2242     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2243     nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2244     if (nb_sectors) {
2245         nb_sectors--;
2246     }
2247     s->qdev.max_lba = nb_sectors;
2248     /* reset tray statuses */
2249     s->tray_locked = 0;
2250     s->tray_open = 0;
2251 
2252     s->qdev.scsi_version = s->qdev.default_scsi_version;
2253 }
2254 
2255 static void scsi_disk_resize_cb(void *opaque)
2256 {
2257     SCSIDiskState *s = opaque;
2258 
2259     /* SPC lists this sense code as available only for
2260      * direct-access devices.
2261      */
2262     if (s->qdev.type == TYPE_DISK) {
2263         scsi_device_report_change(&s->qdev, SENSE_CODE(CAPACITY_CHANGED));
2264     }
2265 }
2266 
2267 static void scsi_cd_change_media_cb(void *opaque, bool load, Error **errp)
2268 {
2269     SCSIDiskState *s = opaque;
2270 
2271     /*
2272      * When a CD gets changed, we have to report an ejected state and
2273      * then a loaded state to guests so that they detect tray
2274      * open/close and media change events.  Guests that do not use
2275      * GET_EVENT_STATUS_NOTIFICATION to detect such tray open/close
2276      * states rely on this behavior.
2277      *
2278      * media_changed governs the state machine used for unit attention
2279      * report.  media_event is used by GET EVENT STATUS NOTIFICATION.
2280      */
2281     s->media_changed = load;
2282     s->tray_open = !load;
2283     scsi_device_set_ua(&s->qdev, SENSE_CODE(UNIT_ATTENTION_NO_MEDIUM));
2284     s->media_event = true;
2285     s->eject_request = false;
2286 }
2287 
2288 static void scsi_cd_eject_request_cb(void *opaque, bool force)
2289 {
2290     SCSIDiskState *s = opaque;
2291 
2292     s->eject_request = true;
2293     if (force) {
2294         s->tray_locked = false;
2295     }
2296 }
2297 
2298 static bool scsi_cd_is_tray_open(void *opaque)
2299 {
2300     return ((SCSIDiskState *)opaque)->tray_open;
2301 }
2302 
2303 static bool scsi_cd_is_medium_locked(void *opaque)
2304 {
2305     return ((SCSIDiskState *)opaque)->tray_locked;
2306 }
2307 
2308 static const BlockDevOps scsi_disk_removable_block_ops = {
2309     .change_media_cb = scsi_cd_change_media_cb,
2310     .eject_request_cb = scsi_cd_eject_request_cb,
2311     .is_tray_open = scsi_cd_is_tray_open,
2312     .is_medium_locked = scsi_cd_is_medium_locked,
2313 
2314     .resize_cb = scsi_disk_resize_cb,
2315 };
2316 
2317 static const BlockDevOps scsi_disk_block_ops = {
2318     .resize_cb = scsi_disk_resize_cb,
2319 };
2320 
2321 static void scsi_disk_unit_attention_reported(SCSIDevice *dev)
2322 {
2323     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2324     if (s->media_changed) {
2325         s->media_changed = false;
2326         scsi_device_set_ua(&s->qdev, SENSE_CODE(MEDIUM_CHANGED));
2327     }
2328 }
2329 
2330 static void scsi_realize(SCSIDevice *dev, Error **errp)
2331 {
2332     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2333     bool read_only;
2334 
2335     if (!s->qdev.conf.blk) {
2336         error_setg(errp, "drive property not set");
2337         return;
2338     }
2339 
2340     if (!(s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2341         !blk_is_inserted(s->qdev.conf.blk)) {
2342         error_setg(errp, "Device needs media, but drive is empty");
2343         return;
2344     }
2345 
2346     if (!blkconf_blocksizes(&s->qdev.conf, errp)) {
2347         return;
2348     }
2349 
2350     if (blk_get_aio_context(s->qdev.conf.blk) != qemu_get_aio_context() &&
2351         !s->qdev.hba_supports_iothread)
2352     {
2353         error_setg(errp, "HBA does not support iothreads");
2354         return;
2355     }
2356 
2357     if (dev->type == TYPE_DISK) {
2358         if (!blkconf_geometry(&dev->conf, NULL, 65535, 255, 255, errp)) {
2359             return;
2360         }
2361     }
2362 
2363     read_only = blk_is_read_only(s->qdev.conf.blk);
2364     if (dev->type == TYPE_ROM) {
2365         read_only = true;
2366     }
2367 
2368     if (!blkconf_apply_backend_options(&dev->conf, read_only,
2369                                        dev->type == TYPE_DISK, errp)) {
2370         return;
2371     }
2372 
2373     if (s->qdev.conf.discard_granularity == -1) {
2374         s->qdev.conf.discard_granularity =
2375             MAX(s->qdev.conf.logical_block_size, DEFAULT_DISCARD_GRANULARITY);
2376     }
2377 
2378     if (!s->version) {
2379         s->version = g_strdup(qemu_hw_version());
2380     }
2381     if (!s->vendor) {
2382         s->vendor = g_strdup("QEMU");
2383     }
2384     if (!s->device_id) {
2385         if (s->serial) {
2386             s->device_id = g_strdup_printf("%.20s", s->serial);
2387         } else {
2388             const char *str = blk_name(s->qdev.conf.blk);
2389             if (str && *str) {
2390                 s->device_id = g_strdup(str);
2391             }
2392         }
2393     }
2394 
2395     if (blk_is_sg(s->qdev.conf.blk)) {
2396         error_setg(errp, "unwanted /dev/sg*");
2397         return;
2398     }
2399 
2400     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2401             !(s->features & (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS))) {
2402         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_removable_block_ops, s);
2403     } else {
2404         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_block_ops, s);
2405     }
2406     blk_set_guest_block_size(s->qdev.conf.blk, s->qdev.blocksize);
2407 
2408     blk_iostatus_enable(s->qdev.conf.blk);
2409 
2410     add_boot_device_lchs(&dev->qdev, NULL,
2411                          dev->conf.lcyls,
2412                          dev->conf.lheads,
2413                          dev->conf.lsecs);
2414 }
2415 
2416 static void scsi_unrealize(SCSIDevice *dev)
2417 {
2418     del_boot_device_lchs(&dev->qdev, NULL);
2419 }
2420 
2421 static void scsi_hd_realize(SCSIDevice *dev, Error **errp)
2422 {
2423     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2424     AioContext *ctx = NULL;
2425     /* can happen for devices without drive. The error message for missing
2426      * backend will be issued in scsi_realize
2427      */
2428     if (s->qdev.conf.blk) {
2429         ctx = blk_get_aio_context(s->qdev.conf.blk);
2430         aio_context_acquire(ctx);
2431         if (!blkconf_blocksizes(&s->qdev.conf, errp)) {
2432             goto out;
2433         }
2434     }
2435     s->qdev.blocksize = s->qdev.conf.logical_block_size;
2436     s->qdev.type = TYPE_DISK;
2437     if (!s->product) {
2438         s->product = g_strdup("QEMU HARDDISK");
2439     }
2440     scsi_realize(&s->qdev, errp);
2441 out:
2442     if (ctx) {
2443         aio_context_release(ctx);
2444     }
2445 }
2446 
2447 static void scsi_cd_realize(SCSIDevice *dev, Error **errp)
2448 {
2449     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2450     AioContext *ctx;
2451     int ret;
2452 
2453     if (!dev->conf.blk) {
2454         /* Anonymous BlockBackend for an empty drive. As we put it into
2455          * dev->conf, qdev takes care of detaching on unplug. */
2456         dev->conf.blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
2457         ret = blk_attach_dev(dev->conf.blk, &dev->qdev);
2458         assert(ret == 0);
2459     }
2460 
2461     ctx = blk_get_aio_context(dev->conf.blk);
2462     aio_context_acquire(ctx);
2463     s->qdev.blocksize = 2048;
2464     s->qdev.type = TYPE_ROM;
2465     s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2466     if (!s->product) {
2467         s->product = g_strdup("QEMU CD-ROM");
2468     }
2469     scsi_realize(&s->qdev, errp);
2470     aio_context_release(ctx);
2471 }
2472 
2473 static void scsi_disk_realize(SCSIDevice *dev, Error **errp)
2474 {
2475     DriveInfo *dinfo;
2476     Error *local_err = NULL;
2477 
2478     warn_report("'scsi-disk' is deprecated, "
2479                 "please use 'scsi-hd' or 'scsi-cd' instead");
2480 
2481     if (!dev->conf.blk) {
2482         scsi_realize(dev, &local_err);
2483         assert(local_err);
2484         error_propagate(errp, local_err);
2485         return;
2486     }
2487 
2488     dinfo = blk_legacy_dinfo(dev->conf.blk);
2489     if (dinfo && dinfo->media_cd) {
2490         scsi_cd_realize(dev, errp);
2491     } else {
2492         scsi_hd_realize(dev, errp);
2493     }
2494 }
2495 
2496 static const SCSIReqOps scsi_disk_emulate_reqops = {
2497     .size         = sizeof(SCSIDiskReq),
2498     .free_req     = scsi_free_request,
2499     .send_command = scsi_disk_emulate_command,
2500     .read_data    = scsi_disk_emulate_read_data,
2501     .write_data   = scsi_disk_emulate_write_data,
2502     .get_buf      = scsi_get_buf,
2503 };
2504 
2505 static const SCSIReqOps scsi_disk_dma_reqops = {
2506     .size         = sizeof(SCSIDiskReq),
2507     .free_req     = scsi_free_request,
2508     .send_command = scsi_disk_dma_command,
2509     .read_data    = scsi_read_data,
2510     .write_data   = scsi_write_data,
2511     .get_buf      = scsi_get_buf,
2512     .load_request = scsi_disk_load_request,
2513     .save_request = scsi_disk_save_request,
2514 };
2515 
2516 static const SCSIReqOps *const scsi_disk_reqops_dispatch[256] = {
2517     [TEST_UNIT_READY]                 = &scsi_disk_emulate_reqops,
2518     [INQUIRY]                         = &scsi_disk_emulate_reqops,
2519     [MODE_SENSE]                      = &scsi_disk_emulate_reqops,
2520     [MODE_SENSE_10]                   = &scsi_disk_emulate_reqops,
2521     [START_STOP]                      = &scsi_disk_emulate_reqops,
2522     [ALLOW_MEDIUM_REMOVAL]            = &scsi_disk_emulate_reqops,
2523     [READ_CAPACITY_10]                = &scsi_disk_emulate_reqops,
2524     [READ_TOC]                        = &scsi_disk_emulate_reqops,
2525     [READ_DVD_STRUCTURE]              = &scsi_disk_emulate_reqops,
2526     [READ_DISC_INFORMATION]           = &scsi_disk_emulate_reqops,
2527     [GET_CONFIGURATION]               = &scsi_disk_emulate_reqops,
2528     [GET_EVENT_STATUS_NOTIFICATION]   = &scsi_disk_emulate_reqops,
2529     [MECHANISM_STATUS]                = &scsi_disk_emulate_reqops,
2530     [SERVICE_ACTION_IN_16]            = &scsi_disk_emulate_reqops,
2531     [REQUEST_SENSE]                   = &scsi_disk_emulate_reqops,
2532     [SYNCHRONIZE_CACHE]               = &scsi_disk_emulate_reqops,
2533     [SEEK_10]                         = &scsi_disk_emulate_reqops,
2534     [MODE_SELECT]                     = &scsi_disk_emulate_reqops,
2535     [MODE_SELECT_10]                  = &scsi_disk_emulate_reqops,
2536     [UNMAP]                           = &scsi_disk_emulate_reqops,
2537     [WRITE_SAME_10]                   = &scsi_disk_emulate_reqops,
2538     [WRITE_SAME_16]                   = &scsi_disk_emulate_reqops,
2539     [VERIFY_10]                       = &scsi_disk_emulate_reqops,
2540     [VERIFY_12]                       = &scsi_disk_emulate_reqops,
2541     [VERIFY_16]                       = &scsi_disk_emulate_reqops,
2542 
2543     [READ_6]                          = &scsi_disk_dma_reqops,
2544     [READ_10]                         = &scsi_disk_dma_reqops,
2545     [READ_12]                         = &scsi_disk_dma_reqops,
2546     [READ_16]                         = &scsi_disk_dma_reqops,
2547     [WRITE_6]                         = &scsi_disk_dma_reqops,
2548     [WRITE_10]                        = &scsi_disk_dma_reqops,
2549     [WRITE_12]                        = &scsi_disk_dma_reqops,
2550     [WRITE_16]                        = &scsi_disk_dma_reqops,
2551     [WRITE_VERIFY_10]                 = &scsi_disk_dma_reqops,
2552     [WRITE_VERIFY_12]                 = &scsi_disk_dma_reqops,
2553     [WRITE_VERIFY_16]                 = &scsi_disk_dma_reqops,
2554 };
2555 
2556 static void scsi_disk_new_request_dump(uint32_t lun, uint32_t tag, uint8_t *buf)
2557 {
2558     int i;
2559     int len = scsi_cdb_length(buf);
2560     char *line_buffer, *p;
2561 
2562     line_buffer = g_malloc(len * 5 + 1);
2563 
2564     for (i = 0, p = line_buffer; i < len; i++) {
2565         p += sprintf(p, " 0x%02x", buf[i]);
2566     }
2567     trace_scsi_disk_new_request(lun, tag, line_buffer);
2568 
2569     g_free(line_buffer);
2570 }
2571 
2572 static SCSIRequest *scsi_new_request(SCSIDevice *d, uint32_t tag, uint32_t lun,
2573                                      uint8_t *buf, void *hba_private)
2574 {
2575     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2576     SCSIRequest *req;
2577     const SCSIReqOps *ops;
2578     uint8_t command;
2579 
2580     command = buf[0];
2581     ops = scsi_disk_reqops_dispatch[command];
2582     if (!ops) {
2583         ops = &scsi_disk_emulate_reqops;
2584     }
2585     req = scsi_req_alloc(ops, &s->qdev, tag, lun, hba_private);
2586 
2587     if (trace_event_get_state_backends(TRACE_SCSI_DISK_NEW_REQUEST)) {
2588         scsi_disk_new_request_dump(lun, tag, buf);
2589     }
2590 
2591     return req;
2592 }
2593 
2594 #ifdef __linux__
2595 static int get_device_type(SCSIDiskState *s)
2596 {
2597     uint8_t cmd[16];
2598     uint8_t buf[36];
2599     int ret;
2600 
2601     memset(cmd, 0, sizeof(cmd));
2602     memset(buf, 0, sizeof(buf));
2603     cmd[0] = INQUIRY;
2604     cmd[4] = sizeof(buf);
2605 
2606     ret = scsi_SG_IO_FROM_DEV(s->qdev.conf.blk, cmd, sizeof(cmd),
2607                               buf, sizeof(buf));
2608     if (ret < 0) {
2609         return -1;
2610     }
2611     s->qdev.type = buf[0];
2612     if (buf[1] & 0x80) {
2613         s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2614     }
2615     return 0;
2616 }
2617 
2618 static void scsi_block_realize(SCSIDevice *dev, Error **errp)
2619 {
2620     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2621     AioContext *ctx;
2622     int sg_version;
2623     int rc;
2624 
2625     if (!s->qdev.conf.blk) {
2626         error_setg(errp, "drive property not set");
2627         return;
2628     }
2629 
2630     if (s->rotation_rate) {
2631         error_report_once("rotation_rate is specified for scsi-block but is "
2632                           "not implemented. This option is deprecated and will "
2633                           "be removed in a future version");
2634     }
2635 
2636     ctx = blk_get_aio_context(s->qdev.conf.blk);
2637     aio_context_acquire(ctx);
2638 
2639     /* check we are using a driver managing SG_IO (version 3 and after) */
2640     rc = blk_ioctl(s->qdev.conf.blk, SG_GET_VERSION_NUM, &sg_version);
2641     if (rc < 0) {
2642         error_setg_errno(errp, -rc, "cannot get SG_IO version number");
2643         if (rc != -EPERM) {
2644             error_append_hint(errp, "Is this a SCSI device?\n");
2645         }
2646         goto out;
2647     }
2648     if (sg_version < 30000) {
2649         error_setg(errp, "scsi generic interface too old");
2650         goto out;
2651     }
2652 
2653     /* get device type from INQUIRY data */
2654     rc = get_device_type(s);
2655     if (rc < 0) {
2656         error_setg(errp, "INQUIRY failed");
2657         goto out;
2658     }
2659 
2660     /* Make a guess for the block size, we'll fix it when the guest sends.
2661      * READ CAPACITY.  If they don't, they likely would assume these sizes
2662      * anyway. (TODO: check in /sys).
2663      */
2664     if (s->qdev.type == TYPE_ROM || s->qdev.type == TYPE_WORM) {
2665         s->qdev.blocksize = 2048;
2666     } else {
2667         s->qdev.blocksize = 512;
2668     }
2669 
2670     /* Makes the scsi-block device not removable by using HMP and QMP eject
2671      * command.
2672      */
2673     s->features |= (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS);
2674 
2675     scsi_realize(&s->qdev, errp);
2676     scsi_generic_read_device_inquiry(&s->qdev);
2677 
2678 out:
2679     aio_context_release(ctx);
2680 }
2681 
2682 typedef struct SCSIBlockReq {
2683     SCSIDiskReq req;
2684     sg_io_hdr_t io_header;
2685 
2686     /* Selected bytes of the original CDB, copied into our own CDB.  */
2687     uint8_t cmd, cdb1, group_number;
2688 
2689     /* CDB passed to SG_IO.  */
2690     uint8_t cdb[16];
2691 } SCSIBlockReq;
2692 
2693 static BlockAIOCB *scsi_block_do_sgio(SCSIBlockReq *req,
2694                                       int64_t offset, QEMUIOVector *iov,
2695                                       int direction,
2696                                       BlockCompletionFunc *cb, void *opaque)
2697 {
2698     sg_io_hdr_t *io_header = &req->io_header;
2699     SCSIDiskReq *r = &req->req;
2700     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2701     int nb_logical_blocks;
2702     uint64_t lba;
2703     BlockAIOCB *aiocb;
2704 
2705     /* This is not supported yet.  It can only happen if the guest does
2706      * reads and writes that are not aligned to one logical sectors
2707      * _and_ cover multiple MemoryRegions.
2708      */
2709     assert(offset % s->qdev.blocksize == 0);
2710     assert(iov->size % s->qdev.blocksize == 0);
2711 
2712     io_header->interface_id = 'S';
2713 
2714     /* The data transfer comes from the QEMUIOVector.  */
2715     io_header->dxfer_direction = direction;
2716     io_header->dxfer_len = iov->size;
2717     io_header->dxferp = (void *)iov->iov;
2718     io_header->iovec_count = iov->niov;
2719     assert(io_header->iovec_count == iov->niov); /* no overflow! */
2720 
2721     /* Build a new CDB with the LBA and length patched in, in case
2722      * DMA helpers split the transfer in multiple segments.  Do not
2723      * build a CDB smaller than what the guest wanted, and only build
2724      * a larger one if strictly necessary.
2725      */
2726     io_header->cmdp = req->cdb;
2727     lba = offset / s->qdev.blocksize;
2728     nb_logical_blocks = io_header->dxfer_len / s->qdev.blocksize;
2729 
2730     if ((req->cmd >> 5) == 0 && lba <= 0x1ffff) {
2731         /* 6-byte CDB */
2732         stl_be_p(&req->cdb[0], lba | (req->cmd << 24));
2733         req->cdb[4] = nb_logical_blocks;
2734         req->cdb[5] = 0;
2735         io_header->cmd_len = 6;
2736     } else if ((req->cmd >> 5) <= 1 && lba <= 0xffffffffULL) {
2737         /* 10-byte CDB */
2738         req->cdb[0] = (req->cmd & 0x1f) | 0x20;
2739         req->cdb[1] = req->cdb1;
2740         stl_be_p(&req->cdb[2], lba);
2741         req->cdb[6] = req->group_number;
2742         stw_be_p(&req->cdb[7], nb_logical_blocks);
2743         req->cdb[9] = 0;
2744         io_header->cmd_len = 10;
2745     } else if ((req->cmd >> 5) != 4 && lba <= 0xffffffffULL) {
2746         /* 12-byte CDB */
2747         req->cdb[0] = (req->cmd & 0x1f) | 0xA0;
2748         req->cdb[1] = req->cdb1;
2749         stl_be_p(&req->cdb[2], lba);
2750         stl_be_p(&req->cdb[6], nb_logical_blocks);
2751         req->cdb[10] = req->group_number;
2752         req->cdb[11] = 0;
2753         io_header->cmd_len = 12;
2754     } else {
2755         /* 16-byte CDB */
2756         req->cdb[0] = (req->cmd & 0x1f) | 0x80;
2757         req->cdb[1] = req->cdb1;
2758         stq_be_p(&req->cdb[2], lba);
2759         stl_be_p(&req->cdb[10], nb_logical_blocks);
2760         req->cdb[14] = req->group_number;
2761         req->cdb[15] = 0;
2762         io_header->cmd_len = 16;
2763     }
2764 
2765     /* The rest is as in scsi-generic.c.  */
2766     io_header->mx_sb_len = sizeof(r->req.sense);
2767     io_header->sbp = r->req.sense;
2768     io_header->timeout = UINT_MAX;
2769     io_header->usr_ptr = r;
2770     io_header->flags |= SG_FLAG_DIRECT_IO;
2771 
2772     aiocb = blk_aio_ioctl(s->qdev.conf.blk, SG_IO, io_header, cb, opaque);
2773     assert(aiocb != NULL);
2774     return aiocb;
2775 }
2776 
2777 static bool scsi_block_no_fua(SCSICommand *cmd)
2778 {
2779     return false;
2780 }
2781 
2782 static BlockAIOCB *scsi_block_dma_readv(int64_t offset,
2783                                         QEMUIOVector *iov,
2784                                         BlockCompletionFunc *cb, void *cb_opaque,
2785                                         void *opaque)
2786 {
2787     SCSIBlockReq *r = opaque;
2788     return scsi_block_do_sgio(r, offset, iov,
2789                               SG_DXFER_FROM_DEV, cb, cb_opaque);
2790 }
2791 
2792 static BlockAIOCB *scsi_block_dma_writev(int64_t offset,
2793                                          QEMUIOVector *iov,
2794                                          BlockCompletionFunc *cb, void *cb_opaque,
2795                                          void *opaque)
2796 {
2797     SCSIBlockReq *r = opaque;
2798     return scsi_block_do_sgio(r, offset, iov,
2799                               SG_DXFER_TO_DEV, cb, cb_opaque);
2800 }
2801 
2802 static bool scsi_block_is_passthrough(SCSIDiskState *s, uint8_t *buf)
2803 {
2804     switch (buf[0]) {
2805     case VERIFY_10:
2806     case VERIFY_12:
2807     case VERIFY_16:
2808         /* Check if BYTCHK == 0x01 (data-out buffer contains data
2809          * for the number of logical blocks specified in the length
2810          * field).  For other modes, do not use scatter/gather operation.
2811          */
2812         if ((buf[1] & 6) == 2) {
2813             return false;
2814         }
2815         break;
2816 
2817     case READ_6:
2818     case READ_10:
2819     case READ_12:
2820     case READ_16:
2821     case WRITE_6:
2822     case WRITE_10:
2823     case WRITE_12:
2824     case WRITE_16:
2825     case WRITE_VERIFY_10:
2826     case WRITE_VERIFY_12:
2827     case WRITE_VERIFY_16:
2828         /* MMC writing cannot be done via DMA helpers, because it sometimes
2829          * involves writing beyond the maximum LBA or to negative LBA (lead-in).
2830          * We might use scsi_block_dma_reqops as long as no writing commands are
2831          * seen, but performance usually isn't paramount on optical media.  So,
2832          * just make scsi-block operate the same as scsi-generic for them.
2833          */
2834         if (s->qdev.type != TYPE_ROM) {
2835             return false;
2836         }
2837         break;
2838 
2839     default:
2840         break;
2841     }
2842 
2843     return true;
2844 }
2845 
2846 
2847 static int32_t scsi_block_dma_command(SCSIRequest *req, uint8_t *buf)
2848 {
2849     SCSIBlockReq *r = (SCSIBlockReq *)req;
2850     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2851 
2852     r->cmd = req->cmd.buf[0];
2853     switch (r->cmd >> 5) {
2854     case 0:
2855         /* 6-byte CDB.  */
2856         r->cdb1 = r->group_number = 0;
2857         break;
2858     case 1:
2859         /* 10-byte CDB.  */
2860         r->cdb1 = req->cmd.buf[1];
2861         r->group_number = req->cmd.buf[6];
2862         break;
2863     case 4:
2864         /* 12-byte CDB.  */
2865         r->cdb1 = req->cmd.buf[1];
2866         r->group_number = req->cmd.buf[10];
2867         break;
2868     case 5:
2869         /* 16-byte CDB.  */
2870         r->cdb1 = req->cmd.buf[1];
2871         r->group_number = req->cmd.buf[14];
2872         break;
2873     default:
2874         abort();
2875     }
2876 
2877     /* Protection information is not supported.  For SCSI versions 2 and
2878      * older (as determined by snooping the guest's INQUIRY commands),
2879      * there is no RD/WR/VRPROTECT, so skip this check in these versions.
2880      */
2881     if (s->qdev.scsi_version > 2 && (req->cmd.buf[1] & 0xe0)) {
2882         scsi_check_condition(&r->req, SENSE_CODE(INVALID_FIELD));
2883         return 0;
2884     }
2885 
2886     r->req.status = &r->io_header.status;
2887     return scsi_disk_dma_command(req, buf);
2888 }
2889 
2890 static const SCSIReqOps scsi_block_dma_reqops = {
2891     .size         = sizeof(SCSIBlockReq),
2892     .free_req     = scsi_free_request,
2893     .send_command = scsi_block_dma_command,
2894     .read_data    = scsi_read_data,
2895     .write_data   = scsi_write_data,
2896     .get_buf      = scsi_get_buf,
2897     .load_request = scsi_disk_load_request,
2898     .save_request = scsi_disk_save_request,
2899 };
2900 
2901 static SCSIRequest *scsi_block_new_request(SCSIDevice *d, uint32_t tag,
2902                                            uint32_t lun, uint8_t *buf,
2903                                            void *hba_private)
2904 {
2905     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2906 
2907     if (scsi_block_is_passthrough(s, buf)) {
2908         return scsi_req_alloc(&scsi_generic_req_ops, &s->qdev, tag, lun,
2909                               hba_private);
2910     } else {
2911         return scsi_req_alloc(&scsi_block_dma_reqops, &s->qdev, tag, lun,
2912                               hba_private);
2913     }
2914 }
2915 
2916 static int scsi_block_parse_cdb(SCSIDevice *d, SCSICommand *cmd,
2917                                   uint8_t *buf, void *hba_private)
2918 {
2919     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2920 
2921     if (scsi_block_is_passthrough(s, buf)) {
2922         return scsi_bus_parse_cdb(&s->qdev, cmd, buf, hba_private);
2923     } else {
2924         return scsi_req_parse_cdb(&s->qdev, cmd, buf);
2925     }
2926 }
2927 
2928 static void scsi_block_update_sense(SCSIRequest *req)
2929 {
2930     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
2931     SCSIBlockReq *br = DO_UPCAST(SCSIBlockReq, req, r);
2932     r->req.sense_len = MIN(br->io_header.sb_len_wr, sizeof(r->req.sense));
2933 }
2934 #endif
2935 
2936 static
2937 BlockAIOCB *scsi_dma_readv(int64_t offset, QEMUIOVector *iov,
2938                            BlockCompletionFunc *cb, void *cb_opaque,
2939                            void *opaque)
2940 {
2941     SCSIDiskReq *r = opaque;
2942     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2943     return blk_aio_preadv(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
2944 }
2945 
2946 static
2947 BlockAIOCB *scsi_dma_writev(int64_t offset, QEMUIOVector *iov,
2948                             BlockCompletionFunc *cb, void *cb_opaque,
2949                             void *opaque)
2950 {
2951     SCSIDiskReq *r = opaque;
2952     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2953     return blk_aio_pwritev(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
2954 }
2955 
2956 static void scsi_disk_base_class_initfn(ObjectClass *klass, void *data)
2957 {
2958     DeviceClass *dc = DEVICE_CLASS(klass);
2959     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
2960 
2961     dc->fw_name = "disk";
2962     dc->reset = scsi_disk_reset;
2963     sdc->dma_readv = scsi_dma_readv;
2964     sdc->dma_writev = scsi_dma_writev;
2965     sdc->need_fua_emulation = scsi_is_cmd_fua;
2966 }
2967 
2968 static const TypeInfo scsi_disk_base_info = {
2969     .name          = TYPE_SCSI_DISK_BASE,
2970     .parent        = TYPE_SCSI_DEVICE,
2971     .class_init    = scsi_disk_base_class_initfn,
2972     .instance_size = sizeof(SCSIDiskState),
2973     .class_size    = sizeof(SCSIDiskClass),
2974     .abstract      = true,
2975 };
2976 
2977 #define DEFINE_SCSI_DISK_PROPERTIES()                                   \
2978     DEFINE_PROP_DRIVE_IOTHREAD("drive", SCSIDiskState, qdev.conf.blk),  \
2979     DEFINE_BLOCK_PROPERTIES_BASE(SCSIDiskState, qdev.conf),             \
2980     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),            \
2981     DEFINE_PROP_STRING("ver", SCSIDiskState, version),                  \
2982     DEFINE_PROP_STRING("serial", SCSIDiskState, serial),                \
2983     DEFINE_PROP_STRING("vendor", SCSIDiskState, vendor),                \
2984     DEFINE_PROP_STRING("product", SCSIDiskState, product),              \
2985     DEFINE_PROP_STRING("device_id", SCSIDiskState, device_id)
2986 
2987 
2988 static Property scsi_hd_properties[] = {
2989     DEFINE_SCSI_DISK_PROPERTIES(),
2990     DEFINE_PROP_BIT("removable", SCSIDiskState, features,
2991                     SCSI_DISK_F_REMOVABLE, false),
2992     DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
2993                     SCSI_DISK_F_DPOFUA, false),
2994     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
2995     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
2996     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
2997     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
2998                        DEFAULT_MAX_UNMAP_SIZE),
2999     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3000                        DEFAULT_MAX_IO_SIZE),
3001     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3002     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3003                       5),
3004     DEFINE_BLOCK_CHS_PROPERTIES(SCSIDiskState, qdev.conf),
3005     DEFINE_PROP_END_OF_LIST(),
3006 };
3007 
3008 static const VMStateDescription vmstate_scsi_disk_state = {
3009     .name = "scsi-disk",
3010     .version_id = 1,
3011     .minimum_version_id = 1,
3012     .fields = (VMStateField[]) {
3013         VMSTATE_SCSI_DEVICE(qdev, SCSIDiskState),
3014         VMSTATE_BOOL(media_changed, SCSIDiskState),
3015         VMSTATE_BOOL(media_event, SCSIDiskState),
3016         VMSTATE_BOOL(eject_request, SCSIDiskState),
3017         VMSTATE_BOOL(tray_open, SCSIDiskState),
3018         VMSTATE_BOOL(tray_locked, SCSIDiskState),
3019         VMSTATE_END_OF_LIST()
3020     }
3021 };
3022 
3023 static void scsi_hd_class_initfn(ObjectClass *klass, void *data)
3024 {
3025     DeviceClass *dc = DEVICE_CLASS(klass);
3026     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3027 
3028     sc->realize      = scsi_hd_realize;
3029     sc->unrealize    = scsi_unrealize;
3030     sc->alloc_req    = scsi_new_request;
3031     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3032     dc->desc = "virtual SCSI disk";
3033     device_class_set_props(dc, scsi_hd_properties);
3034     dc->vmsd  = &vmstate_scsi_disk_state;
3035 }
3036 
3037 static const TypeInfo scsi_hd_info = {
3038     .name          = "scsi-hd",
3039     .parent        = TYPE_SCSI_DISK_BASE,
3040     .class_init    = scsi_hd_class_initfn,
3041 };
3042 
3043 static Property scsi_cd_properties[] = {
3044     DEFINE_SCSI_DISK_PROPERTIES(),
3045     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3046     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3047     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3048     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3049                        DEFAULT_MAX_IO_SIZE),
3050     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3051                       5),
3052     DEFINE_PROP_END_OF_LIST(),
3053 };
3054 
3055 static void scsi_cd_class_initfn(ObjectClass *klass, void *data)
3056 {
3057     DeviceClass *dc = DEVICE_CLASS(klass);
3058     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3059 
3060     sc->realize      = scsi_cd_realize;
3061     sc->alloc_req    = scsi_new_request;
3062     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3063     dc->desc = "virtual SCSI CD-ROM";
3064     device_class_set_props(dc, scsi_cd_properties);
3065     dc->vmsd  = &vmstate_scsi_disk_state;
3066 }
3067 
3068 static const TypeInfo scsi_cd_info = {
3069     .name          = "scsi-cd",
3070     .parent        = TYPE_SCSI_DISK_BASE,
3071     .class_init    = scsi_cd_class_initfn,
3072 };
3073 
3074 #ifdef __linux__
3075 static Property scsi_block_properties[] = {
3076     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),
3077     DEFINE_PROP_DRIVE("drive", SCSIDiskState, qdev.conf.blk),
3078     DEFINE_PROP_BOOL("share-rw", SCSIDiskState, qdev.conf.share_rw, false),
3079     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3080     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3081                        DEFAULT_MAX_UNMAP_SIZE),
3082     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3083                        DEFAULT_MAX_IO_SIZE),
3084     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3085                       -1),
3086     DEFINE_PROP_END_OF_LIST(),
3087 };
3088 
3089 static void scsi_block_class_initfn(ObjectClass *klass, void *data)
3090 {
3091     DeviceClass *dc = DEVICE_CLASS(klass);
3092     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3093     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
3094 
3095     sc->realize      = scsi_block_realize;
3096     sc->alloc_req    = scsi_block_new_request;
3097     sc->parse_cdb    = scsi_block_parse_cdb;
3098     sdc->dma_readv   = scsi_block_dma_readv;
3099     sdc->dma_writev  = scsi_block_dma_writev;
3100     sdc->update_sense = scsi_block_update_sense;
3101     sdc->need_fua_emulation = scsi_block_no_fua;
3102     dc->desc = "SCSI block device passthrough";
3103     device_class_set_props(dc, scsi_block_properties);
3104     dc->vmsd  = &vmstate_scsi_disk_state;
3105 }
3106 
3107 static const TypeInfo scsi_block_info = {
3108     .name          = "scsi-block",
3109     .parent        = TYPE_SCSI_DISK_BASE,
3110     .class_init    = scsi_block_class_initfn,
3111 };
3112 #endif
3113 
3114 static Property scsi_disk_properties[] = {
3115     DEFINE_SCSI_DISK_PROPERTIES(),
3116     DEFINE_PROP_BIT("removable", SCSIDiskState, features,
3117                     SCSI_DISK_F_REMOVABLE, false),
3118     DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
3119                     SCSI_DISK_F_DPOFUA, false),
3120     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3121     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3122     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3123     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3124                        DEFAULT_MAX_UNMAP_SIZE),
3125     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3126                        DEFAULT_MAX_IO_SIZE),
3127     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3128                       5),
3129     DEFINE_PROP_END_OF_LIST(),
3130 };
3131 
3132 static void scsi_disk_class_initfn(ObjectClass *klass, void *data)
3133 {
3134     DeviceClass *dc = DEVICE_CLASS(klass);
3135     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3136 
3137     sc->realize      = scsi_disk_realize;
3138     sc->alloc_req    = scsi_new_request;
3139     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3140     dc->fw_name = "disk";
3141     dc->desc = "virtual SCSI disk or CD-ROM (legacy)";
3142     dc->reset = scsi_disk_reset;
3143     device_class_set_props(dc, scsi_disk_properties);
3144     dc->vmsd  = &vmstate_scsi_disk_state;
3145 }
3146 
3147 static const TypeInfo scsi_disk_info = {
3148     .name          = "scsi-disk",
3149     .parent        = TYPE_SCSI_DISK_BASE,
3150     .class_init    = scsi_disk_class_initfn,
3151 };
3152 
3153 static void scsi_disk_register_types(void)
3154 {
3155     type_register_static(&scsi_disk_base_info);
3156     type_register_static(&scsi_hd_info);
3157     type_register_static(&scsi_cd_info);
3158 #ifdef __linux__
3159     type_register_static(&scsi_block_info);
3160 #endif
3161     type_register_static(&scsi_disk_info);
3162 }
3163 
3164 type_init(scsi_disk_register_types)
3165