xref: /openbmc/qemu/hw/scsi/scsi-disk.c (revision e1ecf8c8)
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 "qemu/cutils.h"
39 #include "trace.h"
40 
41 #ifdef __linux
42 #include <scsi/sg.h>
43 #endif
44 
45 #define SCSI_WRITE_SAME_MAX         (512 * KiB)
46 #define SCSI_DMA_BUF_SIZE           (128 * KiB)
47 #define SCSI_MAX_INQUIRY_LEN        256
48 #define SCSI_MAX_MODE_LEN           256
49 
50 #define DEFAULT_DISCARD_GRANULARITY (4 * KiB)
51 #define DEFAULT_MAX_UNMAP_SIZE      (1 * GiB)
52 #define DEFAULT_MAX_IO_SIZE         INT_MAX     /* 2 GB - 1 block */
53 
54 #define TYPE_SCSI_DISK_BASE         "scsi-disk-base"
55 
56 #define SCSI_DISK_BASE(obj) \
57      OBJECT_CHECK(SCSIDiskState, (obj), TYPE_SCSI_DISK_BASE)
58 #define SCSI_DISK_BASE_CLASS(klass) \
59      OBJECT_CLASS_CHECK(SCSIDiskClass, (klass), TYPE_SCSI_DISK_BASE)
60 #define SCSI_DISK_BASE_GET_CLASS(obj) \
61      OBJECT_GET_CLASS(SCSIDiskClass, (obj), TYPE_SCSI_DISK_BASE)
62 
63 typedef struct SCSIDiskClass {
64     SCSIDeviceClass parent_class;
65     DMAIOFunc       *dma_readv;
66     DMAIOFunc       *dma_writev;
67     bool            (*need_fua_emulation)(SCSICommand *cmd);
68     void            (*update_sense)(SCSIRequest *r);
69 } SCSIDiskClass;
70 
71 typedef struct SCSIDiskReq {
72     SCSIRequest req;
73     /* Both sector and sector_count are in terms of qemu 512 byte blocks.  */
74     uint64_t sector;
75     uint32_t sector_count;
76     uint32_t buflen;
77     bool started;
78     bool need_fua_emulation;
79     struct iovec iov;
80     QEMUIOVector qiov;
81     BlockAcctCookie acct;
82     unsigned char *status;
83 } SCSIDiskReq;
84 
85 #define SCSI_DISK_F_REMOVABLE             0
86 #define SCSI_DISK_F_DPOFUA                1
87 #define SCSI_DISK_F_NO_REMOVABLE_DEVOPS   2
88 
89 typedef struct SCSIDiskState
90 {
91     SCSIDevice qdev;
92     uint32_t features;
93     bool media_changed;
94     bool media_event;
95     bool eject_request;
96     uint16_t port_index;
97     uint64_t max_unmap_size;
98     uint64_t max_io_size;
99     QEMUBH *bh;
100     char *version;
101     char *serial;
102     char *vendor;
103     char *product;
104     char *device_id;
105     bool tray_open;
106     bool tray_locked;
107     /*
108      * 0x0000        - rotation rate not reported
109      * 0x0001        - non-rotating medium (SSD)
110      * 0x0002-0x0400 - reserved
111      * 0x0401-0xffe  - rotations per minute
112      * 0xffff        - reserved
113      */
114     uint16_t rotation_rate;
115 } SCSIDiskState;
116 
117 static bool scsi_handle_rw_error(SCSIDiskReq *r, int error, bool acct_failed);
118 
119 static void scsi_free_request(SCSIRequest *req)
120 {
121     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
122 
123     qemu_vfree(r->iov.iov_base);
124 }
125 
126 /* Helper function for command completion with sense.  */
127 static void scsi_check_condition(SCSIDiskReq *r, SCSISense sense)
128 {
129     trace_scsi_disk_check_condition(r->req.tag, sense.key, sense.asc,
130                                     sense.ascq);
131     scsi_req_build_sense(&r->req, sense);
132     scsi_req_complete(&r->req, CHECK_CONDITION);
133 }
134 
135 static void scsi_init_iovec(SCSIDiskReq *r, size_t size)
136 {
137     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
138 
139     if (!r->iov.iov_base) {
140         r->buflen = size;
141         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
142     }
143     r->iov.iov_len = MIN(r->sector_count * 512, r->buflen);
144     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
145 }
146 
147 static void scsi_disk_save_request(QEMUFile *f, SCSIRequest *req)
148 {
149     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
150 
151     qemu_put_be64s(f, &r->sector);
152     qemu_put_be32s(f, &r->sector_count);
153     qemu_put_be32s(f, &r->buflen);
154     if (r->buflen) {
155         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
156             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
157         } else if (!req->retry) {
158             uint32_t len = r->iov.iov_len;
159             qemu_put_be32s(f, &len);
160             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
161         }
162     }
163 }
164 
165 static void scsi_disk_load_request(QEMUFile *f, SCSIRequest *req)
166 {
167     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
168 
169     qemu_get_be64s(f, &r->sector);
170     qemu_get_be32s(f, &r->sector_count);
171     qemu_get_be32s(f, &r->buflen);
172     if (r->buflen) {
173         scsi_init_iovec(r, r->buflen);
174         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
175             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
176         } else if (!r->req.retry) {
177             uint32_t len;
178             qemu_get_be32s(f, &len);
179             r->iov.iov_len = len;
180             assert(r->iov.iov_len <= r->buflen);
181             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
182         }
183     }
184 
185     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
186 }
187 
188 static bool scsi_disk_req_check_error(SCSIDiskReq *r, int ret, bool acct_failed)
189 {
190     if (r->req.io_canceled) {
191         scsi_req_cancel_complete(&r->req);
192         return true;
193     }
194 
195     if (ret < 0 || (r->status && *r->status)) {
196         return scsi_handle_rw_error(r, -ret, acct_failed);
197     }
198 
199     return false;
200 }
201 
202 static void scsi_aio_complete(void *opaque, int ret)
203 {
204     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
205     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
206 
207     assert(r->req.aiocb != NULL);
208     r->req.aiocb = NULL;
209     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
210     if (scsi_disk_req_check_error(r, ret, true)) {
211         goto done;
212     }
213 
214     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
215     scsi_req_complete(&r->req, GOOD);
216 
217 done:
218     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
219     scsi_req_unref(&r->req);
220 }
221 
222 static bool scsi_is_cmd_fua(SCSICommand *cmd)
223 {
224     switch (cmd->buf[0]) {
225     case READ_10:
226     case READ_12:
227     case READ_16:
228     case WRITE_10:
229     case WRITE_12:
230     case WRITE_16:
231         return (cmd->buf[1] & 8) != 0;
232 
233     case VERIFY_10:
234     case VERIFY_12:
235     case VERIFY_16:
236     case WRITE_VERIFY_10:
237     case WRITE_VERIFY_12:
238     case WRITE_VERIFY_16:
239         return true;
240 
241     case READ_6:
242     case WRITE_6:
243     default:
244         return false;
245     }
246 }
247 
248 static void scsi_write_do_fua(SCSIDiskReq *r)
249 {
250     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
251 
252     assert(r->req.aiocb == NULL);
253     assert(!r->req.io_canceled);
254 
255     if (r->need_fua_emulation) {
256         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
257                          BLOCK_ACCT_FLUSH);
258         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
259         return;
260     }
261 
262     scsi_req_complete(&r->req, GOOD);
263     scsi_req_unref(&r->req);
264 }
265 
266 static void scsi_dma_complete_noio(SCSIDiskReq *r, int ret)
267 {
268     assert(r->req.aiocb == NULL);
269     if (scsi_disk_req_check_error(r, ret, false)) {
270         goto done;
271     }
272 
273     r->sector += r->sector_count;
274     r->sector_count = 0;
275     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
276         scsi_write_do_fua(r);
277         return;
278     } else {
279         scsi_req_complete(&r->req, GOOD);
280     }
281 
282 done:
283     scsi_req_unref(&r->req);
284 }
285 
286 static void scsi_dma_complete(void *opaque, int ret)
287 {
288     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
289     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
290 
291     assert(r->req.aiocb != NULL);
292     r->req.aiocb = NULL;
293 
294     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
295     if (ret < 0) {
296         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
297     } else {
298         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
299     }
300     scsi_dma_complete_noio(r, ret);
301     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
302 }
303 
304 static void scsi_read_complete_noio(SCSIDiskReq *r, int ret)
305 {
306     uint32_t n;
307 
308     assert(r->req.aiocb == NULL);
309     if (scsi_disk_req_check_error(r, ret, false)) {
310         goto done;
311     }
312 
313     n = r->qiov.size / 512;
314     r->sector += n;
315     r->sector_count -= n;
316     scsi_req_data(&r->req, r->qiov.size);
317 
318 done:
319     scsi_req_unref(&r->req);
320 }
321 
322 static void scsi_read_complete(void *opaque, int ret)
323 {
324     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
325     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
326 
327     assert(r->req.aiocb != NULL);
328     r->req.aiocb = NULL;
329 
330     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
331     if (ret < 0) {
332         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
333     } else {
334         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
335         trace_scsi_disk_read_complete(r->req.tag, r->qiov.size);
336     }
337     scsi_read_complete_noio(r, ret);
338     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
339 }
340 
341 /* Actually issue a read to the block device.  */
342 static void scsi_do_read(SCSIDiskReq *r, int ret)
343 {
344     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
345     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
346 
347     assert (r->req.aiocb == NULL);
348     if (scsi_disk_req_check_error(r, ret, false)) {
349         goto done;
350     }
351 
352     /* The request is used as the AIO opaque value, so add a ref.  */
353     scsi_req_ref(&r->req);
354 
355     if (r->req.sg) {
356         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_READ);
357         r->req.resid -= r->req.sg->size;
358         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
359                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
360                                   BDRV_SECTOR_SIZE,
361                                   sdc->dma_readv, r, scsi_dma_complete, r,
362                                   DMA_DIRECTION_FROM_DEVICE);
363     } else {
364         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
365         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
366                          r->qiov.size, BLOCK_ACCT_READ);
367         r->req.aiocb = sdc->dma_readv(r->sector << BDRV_SECTOR_BITS, &r->qiov,
368                                       scsi_read_complete, r, r);
369     }
370 
371 done:
372     scsi_req_unref(&r->req);
373 }
374 
375 static void scsi_do_read_cb(void *opaque, int ret)
376 {
377     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
378     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
379 
380     assert (r->req.aiocb != NULL);
381     r->req.aiocb = NULL;
382 
383     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
384     if (ret < 0) {
385         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
386     } else {
387         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
388     }
389     scsi_do_read(opaque, ret);
390     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
391 }
392 
393 /* Read more data from scsi device into buffer.  */
394 static void scsi_read_data(SCSIRequest *req)
395 {
396     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
397     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
398     bool first;
399 
400     trace_scsi_disk_read_data_count(r->sector_count);
401     if (r->sector_count == 0) {
402         /* This also clears the sense buffer for REQUEST SENSE.  */
403         scsi_req_complete(&r->req, GOOD);
404         return;
405     }
406 
407     /* No data transfer may already be in progress */
408     assert(r->req.aiocb == NULL);
409 
410     /* The request is used as the AIO opaque value, so add a ref.  */
411     scsi_req_ref(&r->req);
412     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
413         trace_scsi_disk_read_data_invalid();
414         scsi_read_complete_noio(r, -EINVAL);
415         return;
416     }
417 
418     if (!blk_is_available(req->dev->conf.blk)) {
419         scsi_read_complete_noio(r, -ENOMEDIUM);
420         return;
421     }
422 
423     first = !r->started;
424     r->started = true;
425     if (first && r->need_fua_emulation) {
426         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
427                          BLOCK_ACCT_FLUSH);
428         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_do_read_cb, r);
429     } else {
430         scsi_do_read(r, 0);
431     }
432 }
433 
434 /*
435  * scsi_handle_rw_error has two return values.  False means that the error
436  * must be ignored, true means that the error has been processed and the
437  * caller should not do anything else for this request.  Note that
438  * scsi_handle_rw_error always manages its reference counts, independent
439  * of the return value.
440  */
441 static bool scsi_handle_rw_error(SCSIDiskReq *r, int error, bool acct_failed)
442 {
443     bool is_read = (r->req.cmd.mode == SCSI_XFER_FROM_DEV);
444     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
445     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
446     BlockErrorAction action = blk_get_error_action(s->qdev.conf.blk,
447                                                    is_read, error);
448 
449     if (action == BLOCK_ERROR_ACTION_REPORT) {
450         if (acct_failed) {
451             block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
452         }
453         switch (error) {
454         case 0:
455             /* A passthrough command has run and has produced sense data; check
456              * whether the error has to be handled by the guest or should rather
457              * pause the host.
458              */
459             assert(r->status && *r->status);
460             if (scsi_sense_buf_is_guest_recoverable(r->req.sense, sizeof(r->req.sense))) {
461                 /* These errors are handled by guest. */
462                 sdc->update_sense(&r->req);
463                 scsi_req_complete(&r->req, *r->status);
464                 return true;
465             }
466             error = scsi_sense_buf_to_errno(r->req.sense, sizeof(r->req.sense));
467             break;
468         case ENOMEDIUM:
469             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
470             break;
471         case ENOMEM:
472             scsi_check_condition(r, SENSE_CODE(TARGET_FAILURE));
473             break;
474         case EINVAL:
475             scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
476             break;
477         case ENOSPC:
478             scsi_check_condition(r, SENSE_CODE(SPACE_ALLOC_FAILED));
479             break;
480         default:
481             scsi_check_condition(r, SENSE_CODE(IO_ERROR));
482             break;
483         }
484     }
485 
486     blk_error_action(s->qdev.conf.blk, action, is_read, error);
487     if (action == BLOCK_ERROR_ACTION_IGNORE) {
488         scsi_req_complete(&r->req, 0);
489         return true;
490     }
491 
492     if (action == BLOCK_ERROR_ACTION_STOP) {
493         scsi_req_retry(&r->req);
494     }
495     return true;
496 }
497 
498 static void scsi_write_complete_noio(SCSIDiskReq *r, int ret)
499 {
500     uint32_t n;
501 
502     assert (r->req.aiocb == NULL);
503     if (scsi_disk_req_check_error(r, ret, false)) {
504         goto done;
505     }
506 
507     n = r->qiov.size / 512;
508     r->sector += n;
509     r->sector_count -= n;
510     if (r->sector_count == 0) {
511         scsi_write_do_fua(r);
512         return;
513     } else {
514         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
515         trace_scsi_disk_write_complete_noio(r->req.tag, r->qiov.size);
516         scsi_req_data(&r->req, r->qiov.size);
517     }
518 
519 done:
520     scsi_req_unref(&r->req);
521 }
522 
523 static void scsi_write_complete(void * opaque, int ret)
524 {
525     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
526     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
527 
528     assert (r->req.aiocb != NULL);
529     r->req.aiocb = NULL;
530 
531     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
532     if (ret < 0) {
533         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
534     } else {
535         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
536     }
537     scsi_write_complete_noio(r, ret);
538     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
539 }
540 
541 static void scsi_write_data(SCSIRequest *req)
542 {
543     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
544     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
545     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
546 
547     /* No data transfer may already be in progress */
548     assert(r->req.aiocb == NULL);
549 
550     /* The request is used as the AIO opaque value, so add a ref.  */
551     scsi_req_ref(&r->req);
552     if (r->req.cmd.mode != SCSI_XFER_TO_DEV) {
553         trace_scsi_disk_write_data_invalid();
554         scsi_write_complete_noio(r, -EINVAL);
555         return;
556     }
557 
558     if (!r->req.sg && !r->qiov.size) {
559         /* Called for the first time.  Ask the driver to send us more data.  */
560         r->started = true;
561         scsi_write_complete_noio(r, 0);
562         return;
563     }
564     if (!blk_is_available(req->dev->conf.blk)) {
565         scsi_write_complete_noio(r, -ENOMEDIUM);
566         return;
567     }
568 
569     if (r->req.cmd.buf[0] == VERIFY_10 || r->req.cmd.buf[0] == VERIFY_12 ||
570         r->req.cmd.buf[0] == VERIFY_16) {
571         if (r->req.sg) {
572             scsi_dma_complete_noio(r, 0);
573         } else {
574             scsi_write_complete_noio(r, 0);
575         }
576         return;
577     }
578 
579     if (r->req.sg) {
580         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_WRITE);
581         r->req.resid -= r->req.sg->size;
582         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
583                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
584                                   BDRV_SECTOR_SIZE,
585                                   sdc->dma_writev, r, scsi_dma_complete, r,
586                                   DMA_DIRECTION_TO_DEVICE);
587     } else {
588         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
589                          r->qiov.size, BLOCK_ACCT_WRITE);
590         r->req.aiocb = sdc->dma_writev(r->sector << BDRV_SECTOR_BITS, &r->qiov,
591                                        scsi_write_complete, r, r);
592     }
593 }
594 
595 /* Return a pointer to the data buffer.  */
596 static uint8_t *scsi_get_buf(SCSIRequest *req)
597 {
598     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
599 
600     return (uint8_t *)r->iov.iov_base;
601 }
602 
603 static int scsi_disk_emulate_vpd_page(SCSIRequest *req, uint8_t *outbuf)
604 {
605     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
606     uint8_t page_code = req->cmd.buf[2];
607     int start, buflen = 0;
608 
609     outbuf[buflen++] = s->qdev.type & 0x1f;
610     outbuf[buflen++] = page_code;
611     outbuf[buflen++] = 0x00;
612     outbuf[buflen++] = 0x00;
613     start = buflen;
614 
615     switch (page_code) {
616     case 0x00: /* Supported page codes, mandatory */
617     {
618         trace_scsi_disk_emulate_vpd_page_00(req->cmd.xfer);
619         outbuf[buflen++] = 0x00; /* list of supported pages (this page) */
620         if (s->serial) {
621             outbuf[buflen++] = 0x80; /* unit serial number */
622         }
623         outbuf[buflen++] = 0x83; /* device identification */
624         if (s->qdev.type == TYPE_DISK) {
625             outbuf[buflen++] = 0xb0; /* block limits */
626             outbuf[buflen++] = 0xb1; /* block device characteristics */
627             outbuf[buflen++] = 0xb2; /* thin provisioning */
628         }
629         break;
630     }
631     case 0x80: /* Device serial number, optional */
632     {
633         int l;
634 
635         if (!s->serial) {
636             trace_scsi_disk_emulate_vpd_page_80_not_supported();
637             return -1;
638         }
639 
640         l = strlen(s->serial);
641         if (l > 36) {
642             l = 36;
643         }
644 
645         trace_scsi_disk_emulate_vpd_page_80(req->cmd.xfer);
646         memcpy(outbuf + buflen, s->serial, l);
647         buflen += l;
648         break;
649     }
650 
651     case 0x83: /* Device identification page, mandatory */
652     {
653         int id_len = s->device_id ? MIN(strlen(s->device_id), 255 - 8) : 0;
654 
655         trace_scsi_disk_emulate_vpd_page_83(req->cmd.xfer);
656 
657         if (id_len) {
658             outbuf[buflen++] = 0x2; /* ASCII */
659             outbuf[buflen++] = 0;   /* not officially assigned */
660             outbuf[buflen++] = 0;   /* reserved */
661             outbuf[buflen++] = id_len; /* length of data following */
662             memcpy(outbuf + buflen, s->device_id, id_len);
663             buflen += id_len;
664         }
665 
666         if (s->qdev.wwn) {
667             outbuf[buflen++] = 0x1; /* Binary */
668             outbuf[buflen++] = 0x3; /* NAA */
669             outbuf[buflen++] = 0;   /* reserved */
670             outbuf[buflen++] = 8;
671             stq_be_p(&outbuf[buflen], s->qdev.wwn);
672             buflen += 8;
673         }
674 
675         if (s->qdev.port_wwn) {
676             outbuf[buflen++] = 0x61; /* SAS / Binary */
677             outbuf[buflen++] = 0x93; /* PIV / Target port / NAA */
678             outbuf[buflen++] = 0;    /* reserved */
679             outbuf[buflen++] = 8;
680             stq_be_p(&outbuf[buflen], s->qdev.port_wwn);
681             buflen += 8;
682         }
683 
684         if (s->port_index) {
685             outbuf[buflen++] = 0x61; /* SAS / Binary */
686 
687             /* PIV/Target port/relative target port */
688             outbuf[buflen++] = 0x94;
689 
690             outbuf[buflen++] = 0;    /* reserved */
691             outbuf[buflen++] = 4;
692             stw_be_p(&outbuf[buflen + 2], s->port_index);
693             buflen += 4;
694         }
695         break;
696     }
697     case 0xb0: /* block limits */
698     {
699         SCSIBlockLimits bl = {};
700 
701         if (s->qdev.type == TYPE_ROM) {
702             trace_scsi_disk_emulate_vpd_page_b0_not_supported();
703             return -1;
704         }
705         bl.wsnz = 1;
706         bl.unmap_sectors =
707             s->qdev.conf.discard_granularity / s->qdev.blocksize;
708         bl.min_io_size =
709             s->qdev.conf.min_io_size / s->qdev.blocksize;
710         bl.opt_io_size =
711             s->qdev.conf.opt_io_size / s->qdev.blocksize;
712         bl.max_unmap_sectors =
713             s->max_unmap_size / s->qdev.blocksize;
714         bl.max_io_sectors =
715             s->max_io_size / s->qdev.blocksize;
716         /* 255 descriptors fit in 4 KiB with an 8-byte header */
717         bl.max_unmap_descr = 255;
718 
719         if (s->qdev.type == TYPE_DISK) {
720             int max_transfer_blk = blk_get_max_transfer(s->qdev.conf.blk);
721             int max_io_sectors_blk =
722                 max_transfer_blk / s->qdev.blocksize;
723 
724             bl.max_io_sectors =
725                 MIN_NON_ZERO(max_io_sectors_blk, bl.max_io_sectors);
726         }
727         buflen += scsi_emulate_block_limits(outbuf + buflen, &bl);
728         break;
729     }
730     case 0xb1: /* block device characteristics */
731     {
732         buflen = 0x40;
733         outbuf[4] = (s->rotation_rate >> 8) & 0xff;
734         outbuf[5] = s->rotation_rate & 0xff;
735         outbuf[6] = 0; /* PRODUCT TYPE */
736         outbuf[7] = 0; /* WABEREQ | WACEREQ | NOMINAL FORM FACTOR */
737         outbuf[8] = 0; /* VBULS */
738         break;
739     }
740     case 0xb2: /* thin provisioning */
741     {
742         buflen = 8;
743         outbuf[4] = 0;
744         outbuf[5] = 0xe0; /* unmap & write_same 10/16 all supported */
745         outbuf[6] = s->qdev.conf.discard_granularity ? 2 : 1;
746         outbuf[7] = 0;
747         break;
748     }
749     default:
750         return -1;
751     }
752     /* done with EVPD */
753     assert(buflen - start <= 255);
754     outbuf[start - 1] = buflen - start;
755     return buflen;
756 }
757 
758 static int scsi_disk_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf)
759 {
760     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
761     int buflen = 0;
762 
763     if (req->cmd.buf[1] & 0x1) {
764         /* Vital product data */
765         return scsi_disk_emulate_vpd_page(req, outbuf);
766     }
767 
768     /* Standard INQUIRY data */
769     if (req->cmd.buf[2] != 0) {
770         return -1;
771     }
772 
773     /* PAGE CODE == 0 */
774     buflen = req->cmd.xfer;
775     if (buflen > SCSI_MAX_INQUIRY_LEN) {
776         buflen = SCSI_MAX_INQUIRY_LEN;
777     }
778 
779     outbuf[0] = s->qdev.type & 0x1f;
780     outbuf[1] = (s->features & (1 << SCSI_DISK_F_REMOVABLE)) ? 0x80 : 0;
781 
782     strpadcpy((char *) &outbuf[16], 16, s->product, ' ');
783     strpadcpy((char *) &outbuf[8], 8, s->vendor, ' ');
784 
785     memset(&outbuf[32], 0, 4);
786     memcpy(&outbuf[32], s->version, MIN(4, strlen(s->version)));
787     /*
788      * We claim conformance to SPC-3, which is required for guests
789      * to ask for modern features like READ CAPACITY(16) or the
790      * block characteristics VPD page by default.  Not all of SPC-3
791      * is actually implemented, but we're good enough.
792      */
793     outbuf[2] = s->qdev.default_scsi_version;
794     outbuf[3] = 2 | 0x10; /* Format 2, HiSup */
795 
796     if (buflen > 36) {
797         outbuf[4] = buflen - 5; /* Additional Length = (Len - 1) - 4 */
798     } else {
799         /* If the allocation length of CDB is too small,
800                the additional length is not adjusted */
801         outbuf[4] = 36 - 5;
802     }
803 
804     /* Sync data transfer and TCQ.  */
805     outbuf[7] = 0x10 | (req->bus->info->tcq ? 0x02 : 0);
806     return buflen;
807 }
808 
809 static inline bool media_is_dvd(SCSIDiskState *s)
810 {
811     uint64_t nb_sectors;
812     if (s->qdev.type != TYPE_ROM) {
813         return false;
814     }
815     if (!blk_is_available(s->qdev.conf.blk)) {
816         return false;
817     }
818     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
819     return nb_sectors > CD_MAX_SECTORS;
820 }
821 
822 static inline bool media_is_cd(SCSIDiskState *s)
823 {
824     uint64_t nb_sectors;
825     if (s->qdev.type != TYPE_ROM) {
826         return false;
827     }
828     if (!blk_is_available(s->qdev.conf.blk)) {
829         return false;
830     }
831     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
832     return nb_sectors <= CD_MAX_SECTORS;
833 }
834 
835 static int scsi_read_disc_information(SCSIDiskState *s, SCSIDiskReq *r,
836                                       uint8_t *outbuf)
837 {
838     uint8_t type = r->req.cmd.buf[1] & 7;
839 
840     if (s->qdev.type != TYPE_ROM) {
841         return -1;
842     }
843 
844     /* Types 1/2 are only defined for Blu-Ray.  */
845     if (type != 0) {
846         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
847         return -1;
848     }
849 
850     memset(outbuf, 0, 34);
851     outbuf[1] = 32;
852     outbuf[2] = 0xe; /* last session complete, disc finalized */
853     outbuf[3] = 1;   /* first track on disc */
854     outbuf[4] = 1;   /* # of sessions */
855     outbuf[5] = 1;   /* first track of last session */
856     outbuf[6] = 1;   /* last track of last session */
857     outbuf[7] = 0x20; /* unrestricted use */
858     outbuf[8] = 0x00; /* CD-ROM or DVD-ROM */
859     /* 9-10-11: most significant byte corresponding bytes 4-5-6 */
860     /* 12-23: not meaningful for CD-ROM or DVD-ROM */
861     /* 24-31: disc bar code */
862     /* 32: disc application code */
863     /* 33: number of OPC tables */
864 
865     return 34;
866 }
867 
868 static int scsi_read_dvd_structure(SCSIDiskState *s, SCSIDiskReq *r,
869                                    uint8_t *outbuf)
870 {
871     static const int rds_caps_size[5] = {
872         [0] = 2048 + 4,
873         [1] = 4 + 4,
874         [3] = 188 + 4,
875         [4] = 2048 + 4,
876     };
877 
878     uint8_t media = r->req.cmd.buf[1];
879     uint8_t layer = r->req.cmd.buf[6];
880     uint8_t format = r->req.cmd.buf[7];
881     int size = -1;
882 
883     if (s->qdev.type != TYPE_ROM) {
884         return -1;
885     }
886     if (media != 0) {
887         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
888         return -1;
889     }
890 
891     if (format != 0xff) {
892         if (!blk_is_available(s->qdev.conf.blk)) {
893             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
894             return -1;
895         }
896         if (media_is_cd(s)) {
897             scsi_check_condition(r, SENSE_CODE(INCOMPATIBLE_FORMAT));
898             return -1;
899         }
900         if (format >= ARRAY_SIZE(rds_caps_size)) {
901             return -1;
902         }
903         size = rds_caps_size[format];
904         memset(outbuf, 0, size);
905     }
906 
907     switch (format) {
908     case 0x00: {
909         /* Physical format information */
910         uint64_t nb_sectors;
911         if (layer != 0) {
912             goto fail;
913         }
914         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
915 
916         outbuf[4] = 1;   /* DVD-ROM, part version 1 */
917         outbuf[5] = 0xf; /* 120mm disc, minimum rate unspecified */
918         outbuf[6] = 1;   /* one layer, read-only (per MMC-2 spec) */
919         outbuf[7] = 0;   /* default densities */
920 
921         stl_be_p(&outbuf[12], (nb_sectors >> 2) - 1); /* end sector */
922         stl_be_p(&outbuf[16], (nb_sectors >> 2) - 1); /* l0 end sector */
923         break;
924     }
925 
926     case 0x01: /* DVD copyright information, all zeros */
927         break;
928 
929     case 0x03: /* BCA information - invalid field for no BCA info */
930         return -1;
931 
932     case 0x04: /* DVD disc manufacturing information, all zeros */
933         break;
934 
935     case 0xff: { /* List capabilities */
936         int i;
937         size = 4;
938         for (i = 0; i < ARRAY_SIZE(rds_caps_size); i++) {
939             if (!rds_caps_size[i]) {
940                 continue;
941             }
942             outbuf[size] = i;
943             outbuf[size + 1] = 0x40; /* Not writable, readable */
944             stw_be_p(&outbuf[size + 2], rds_caps_size[i]);
945             size += 4;
946         }
947         break;
948      }
949 
950     default:
951         return -1;
952     }
953 
954     /* Size of buffer, not including 2 byte size field */
955     stw_be_p(outbuf, size - 2);
956     return size;
957 
958 fail:
959     return -1;
960 }
961 
962 static int scsi_event_status_media(SCSIDiskState *s, uint8_t *outbuf)
963 {
964     uint8_t event_code, media_status;
965 
966     media_status = 0;
967     if (s->tray_open) {
968         media_status = MS_TRAY_OPEN;
969     } else if (blk_is_inserted(s->qdev.conf.blk)) {
970         media_status = MS_MEDIA_PRESENT;
971     }
972 
973     /* Event notification descriptor */
974     event_code = MEC_NO_CHANGE;
975     if (media_status != MS_TRAY_OPEN) {
976         if (s->media_event) {
977             event_code = MEC_NEW_MEDIA;
978             s->media_event = false;
979         } else if (s->eject_request) {
980             event_code = MEC_EJECT_REQUESTED;
981             s->eject_request = false;
982         }
983     }
984 
985     outbuf[0] = event_code;
986     outbuf[1] = media_status;
987 
988     /* These fields are reserved, just clear them. */
989     outbuf[2] = 0;
990     outbuf[3] = 0;
991     return 4;
992 }
993 
994 static int scsi_get_event_status_notification(SCSIDiskState *s, SCSIDiskReq *r,
995                                               uint8_t *outbuf)
996 {
997     int size;
998     uint8_t *buf = r->req.cmd.buf;
999     uint8_t notification_class_request = buf[4];
1000     if (s->qdev.type != TYPE_ROM) {
1001         return -1;
1002     }
1003     if ((buf[1] & 1) == 0) {
1004         /* asynchronous */
1005         return -1;
1006     }
1007 
1008     size = 4;
1009     outbuf[0] = outbuf[1] = 0;
1010     outbuf[3] = 1 << GESN_MEDIA; /* supported events */
1011     if (notification_class_request & (1 << GESN_MEDIA)) {
1012         outbuf[2] = GESN_MEDIA;
1013         size += scsi_event_status_media(s, &outbuf[size]);
1014     } else {
1015         outbuf[2] = 0x80;
1016     }
1017     stw_be_p(outbuf, size - 4);
1018     return size;
1019 }
1020 
1021 static int scsi_get_configuration(SCSIDiskState *s, uint8_t *outbuf)
1022 {
1023     int current;
1024 
1025     if (s->qdev.type != TYPE_ROM) {
1026         return -1;
1027     }
1028 
1029     if (media_is_dvd(s)) {
1030         current = MMC_PROFILE_DVD_ROM;
1031     } else if (media_is_cd(s)) {
1032         current = MMC_PROFILE_CD_ROM;
1033     } else {
1034         current = MMC_PROFILE_NONE;
1035     }
1036 
1037     memset(outbuf, 0, 40);
1038     stl_be_p(&outbuf[0], 36); /* Bytes after the data length field */
1039     stw_be_p(&outbuf[6], current);
1040     /* outbuf[8] - outbuf[19]: Feature 0 - Profile list */
1041     outbuf[10] = 0x03; /* persistent, current */
1042     outbuf[11] = 8; /* two profiles */
1043     stw_be_p(&outbuf[12], MMC_PROFILE_DVD_ROM);
1044     outbuf[14] = (current == MMC_PROFILE_DVD_ROM);
1045     stw_be_p(&outbuf[16], MMC_PROFILE_CD_ROM);
1046     outbuf[18] = (current == MMC_PROFILE_CD_ROM);
1047     /* outbuf[20] - outbuf[31]: Feature 1 - Core feature */
1048     stw_be_p(&outbuf[20], 1);
1049     outbuf[22] = 0x08 | 0x03; /* version 2, persistent, current */
1050     outbuf[23] = 8;
1051     stl_be_p(&outbuf[24], 1); /* SCSI */
1052     outbuf[28] = 1; /* DBE = 1, mandatory */
1053     /* outbuf[32] - outbuf[39]: Feature 3 - Removable media feature */
1054     stw_be_p(&outbuf[32], 3);
1055     outbuf[34] = 0x08 | 0x03; /* version 2, persistent, current */
1056     outbuf[35] = 4;
1057     outbuf[36] = 0x39; /* tray, load=1, eject=1, unlocked at powerup, lock=1 */
1058     /* TODO: Random readable, CD read, DVD read, drive serial number,
1059        power management */
1060     return 40;
1061 }
1062 
1063 static int scsi_emulate_mechanism_status(SCSIDiskState *s, uint8_t *outbuf)
1064 {
1065     if (s->qdev.type != TYPE_ROM) {
1066         return -1;
1067     }
1068     memset(outbuf, 0, 8);
1069     outbuf[5] = 1; /* CD-ROM */
1070     return 8;
1071 }
1072 
1073 static int mode_sense_page(SCSIDiskState *s, int page, uint8_t **p_outbuf,
1074                            int page_control)
1075 {
1076     static const int mode_sense_valid[0x3f] = {
1077         [MODE_PAGE_HD_GEOMETRY]            = (1 << TYPE_DISK),
1078         [MODE_PAGE_FLEXIBLE_DISK_GEOMETRY] = (1 << TYPE_DISK),
1079         [MODE_PAGE_CACHING]                = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1080         [MODE_PAGE_R_W_ERROR]              = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1081         [MODE_PAGE_AUDIO_CTL]              = (1 << TYPE_ROM),
1082         [MODE_PAGE_CAPABILITIES]           = (1 << TYPE_ROM),
1083     };
1084 
1085     uint8_t *p = *p_outbuf + 2;
1086     int length;
1087 
1088     if ((mode_sense_valid[page] & (1 << s->qdev.type)) == 0) {
1089         return -1;
1090     }
1091 
1092     /*
1093      * If Changeable Values are requested, a mask denoting those mode parameters
1094      * that are changeable shall be returned. As we currently don't support
1095      * parameter changes via MODE_SELECT all bits are returned set to zero.
1096      * The buffer was already menset to zero by the caller of this function.
1097      *
1098      * The offsets here are off by two compared to the descriptions in the
1099      * SCSI specs, because those include a 2-byte header.  This is unfortunate,
1100      * but it is done so that offsets are consistent within our implementation
1101      * of MODE SENSE and MODE SELECT.  MODE SELECT has to deal with both
1102      * 2-byte and 4-byte headers.
1103      */
1104     switch (page) {
1105     case MODE_PAGE_HD_GEOMETRY:
1106         length = 0x16;
1107         if (page_control == 1) { /* Changeable Values */
1108             break;
1109         }
1110         /* if a geometry hint is available, use it */
1111         p[0] = (s->qdev.conf.cyls >> 16) & 0xff;
1112         p[1] = (s->qdev.conf.cyls >> 8) & 0xff;
1113         p[2] = s->qdev.conf.cyls & 0xff;
1114         p[3] = s->qdev.conf.heads & 0xff;
1115         /* Write precomp start cylinder, disabled */
1116         p[4] = (s->qdev.conf.cyls >> 16) & 0xff;
1117         p[5] = (s->qdev.conf.cyls >> 8) & 0xff;
1118         p[6] = s->qdev.conf.cyls & 0xff;
1119         /* Reduced current start cylinder, disabled */
1120         p[7] = (s->qdev.conf.cyls >> 16) & 0xff;
1121         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1122         p[9] = s->qdev.conf.cyls & 0xff;
1123         /* Device step rate [ns], 200ns */
1124         p[10] = 0;
1125         p[11] = 200;
1126         /* Landing zone cylinder */
1127         p[12] = 0xff;
1128         p[13] =  0xff;
1129         p[14] = 0xff;
1130         /* Medium rotation rate [rpm], 5400 rpm */
1131         p[18] = (5400 >> 8) & 0xff;
1132         p[19] = 5400 & 0xff;
1133         break;
1134 
1135     case MODE_PAGE_FLEXIBLE_DISK_GEOMETRY:
1136         length = 0x1e;
1137         if (page_control == 1) { /* Changeable Values */
1138             break;
1139         }
1140         /* Transfer rate [kbit/s], 5Mbit/s */
1141         p[0] = 5000 >> 8;
1142         p[1] = 5000 & 0xff;
1143         /* if a geometry hint is available, use it */
1144         p[2] = s->qdev.conf.heads & 0xff;
1145         p[3] = s->qdev.conf.secs & 0xff;
1146         p[4] = s->qdev.blocksize >> 8;
1147         p[6] = (s->qdev.conf.cyls >> 8) & 0xff;
1148         p[7] = s->qdev.conf.cyls & 0xff;
1149         /* Write precomp start cylinder, disabled */
1150         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1151         p[9] = s->qdev.conf.cyls & 0xff;
1152         /* Reduced current start cylinder, disabled */
1153         p[10] = (s->qdev.conf.cyls >> 8) & 0xff;
1154         p[11] = s->qdev.conf.cyls & 0xff;
1155         /* Device step rate [100us], 100us */
1156         p[12] = 0;
1157         p[13] = 1;
1158         /* Device step pulse width [us], 1us */
1159         p[14] = 1;
1160         /* Device head settle delay [100us], 100us */
1161         p[15] = 0;
1162         p[16] = 1;
1163         /* Motor on delay [0.1s], 0.1s */
1164         p[17] = 1;
1165         /* Motor off delay [0.1s], 0.1s */
1166         p[18] = 1;
1167         /* Medium rotation rate [rpm], 5400 rpm */
1168         p[26] = (5400 >> 8) & 0xff;
1169         p[27] = 5400 & 0xff;
1170         break;
1171 
1172     case MODE_PAGE_CACHING:
1173         length = 0x12;
1174         if (page_control == 1 || /* Changeable Values */
1175             blk_enable_write_cache(s->qdev.conf.blk)) {
1176             p[0] = 4; /* WCE */
1177         }
1178         break;
1179 
1180     case MODE_PAGE_R_W_ERROR:
1181         length = 10;
1182         if (page_control == 1) { /* Changeable Values */
1183             break;
1184         }
1185         p[0] = 0x80; /* Automatic Write Reallocation Enabled */
1186         if (s->qdev.type == TYPE_ROM) {
1187             p[1] = 0x20; /* Read Retry Count */
1188         }
1189         break;
1190 
1191     case MODE_PAGE_AUDIO_CTL:
1192         length = 14;
1193         break;
1194 
1195     case MODE_PAGE_CAPABILITIES:
1196         length = 0x14;
1197         if (page_control == 1) { /* Changeable Values */
1198             break;
1199         }
1200 
1201         p[0] = 0x3b; /* CD-R & CD-RW read */
1202         p[1] = 0; /* Writing not supported */
1203         p[2] = 0x7f; /* Audio, composite, digital out,
1204                         mode 2 form 1&2, multi session */
1205         p[3] = 0xff; /* CD DA, DA accurate, RW supported,
1206                         RW corrected, C2 errors, ISRC,
1207                         UPC, Bar code */
1208         p[4] = 0x2d | (s->tray_locked ? 2 : 0);
1209         /* Locking supported, jumper present, eject, tray */
1210         p[5] = 0; /* no volume & mute control, no
1211                      changer */
1212         p[6] = (50 * 176) >> 8; /* 50x read speed */
1213         p[7] = (50 * 176) & 0xff;
1214         p[8] = 2 >> 8; /* Two volume levels */
1215         p[9] = 2 & 0xff;
1216         p[10] = 2048 >> 8; /* 2M buffer */
1217         p[11] = 2048 & 0xff;
1218         p[12] = (16 * 176) >> 8; /* 16x read speed current */
1219         p[13] = (16 * 176) & 0xff;
1220         p[16] = (16 * 176) >> 8; /* 16x write speed */
1221         p[17] = (16 * 176) & 0xff;
1222         p[18] = (16 * 176) >> 8; /* 16x write speed current */
1223         p[19] = (16 * 176) & 0xff;
1224         break;
1225 
1226     default:
1227         return -1;
1228     }
1229 
1230     assert(length < 256);
1231     (*p_outbuf)[0] = page;
1232     (*p_outbuf)[1] = length;
1233     *p_outbuf += length + 2;
1234     return length + 2;
1235 }
1236 
1237 static int scsi_disk_emulate_mode_sense(SCSIDiskReq *r, uint8_t *outbuf)
1238 {
1239     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1240     uint64_t nb_sectors;
1241     bool dbd;
1242     int page, buflen, ret, page_control;
1243     uint8_t *p;
1244     uint8_t dev_specific_param;
1245 
1246     dbd = (r->req.cmd.buf[1] & 0x8) != 0;
1247     page = r->req.cmd.buf[2] & 0x3f;
1248     page_control = (r->req.cmd.buf[2] & 0xc0) >> 6;
1249 
1250     trace_scsi_disk_emulate_mode_sense((r->req.cmd.buf[0] == MODE_SENSE) ? 6 :
1251                                        10, page, r->req.cmd.xfer, page_control);
1252     memset(outbuf, 0, r->req.cmd.xfer);
1253     p = outbuf;
1254 
1255     if (s->qdev.type == TYPE_DISK) {
1256         dev_specific_param = s->features & (1 << SCSI_DISK_F_DPOFUA) ? 0x10 : 0;
1257         if (blk_is_read_only(s->qdev.conf.blk)) {
1258             dev_specific_param |= 0x80; /* Readonly.  */
1259         }
1260     } else {
1261         /* MMC prescribes that CD/DVD drives have no block descriptors,
1262          * and defines no device-specific parameter.  */
1263         dev_specific_param = 0x00;
1264         dbd = true;
1265     }
1266 
1267     if (r->req.cmd.buf[0] == MODE_SENSE) {
1268         p[1] = 0; /* Default media type.  */
1269         p[2] = dev_specific_param;
1270         p[3] = 0; /* Block descriptor length.  */
1271         p += 4;
1272     } else { /* MODE_SENSE_10 */
1273         p[2] = 0; /* Default media type.  */
1274         p[3] = dev_specific_param;
1275         p[6] = p[7] = 0; /* Block descriptor length.  */
1276         p += 8;
1277     }
1278 
1279     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1280     if (!dbd && nb_sectors) {
1281         if (r->req.cmd.buf[0] == MODE_SENSE) {
1282             outbuf[3] = 8; /* Block descriptor length  */
1283         } else { /* MODE_SENSE_10 */
1284             outbuf[7] = 8; /* Block descriptor length  */
1285         }
1286         nb_sectors /= (s->qdev.blocksize / 512);
1287         if (nb_sectors > 0xffffff) {
1288             nb_sectors = 0;
1289         }
1290         p[0] = 0; /* media density code */
1291         p[1] = (nb_sectors >> 16) & 0xff;
1292         p[2] = (nb_sectors >> 8) & 0xff;
1293         p[3] = nb_sectors & 0xff;
1294         p[4] = 0; /* reserved */
1295         p[5] = 0; /* bytes 5-7 are the sector size in bytes */
1296         p[6] = s->qdev.blocksize >> 8;
1297         p[7] = 0;
1298         p += 8;
1299     }
1300 
1301     if (page_control == 3) {
1302         /* Saved Values */
1303         scsi_check_condition(r, SENSE_CODE(SAVING_PARAMS_NOT_SUPPORTED));
1304         return -1;
1305     }
1306 
1307     if (page == 0x3f) {
1308         for (page = 0; page <= 0x3e; page++) {
1309             mode_sense_page(s, page, &p, page_control);
1310         }
1311     } else {
1312         ret = mode_sense_page(s, page, &p, page_control);
1313         if (ret == -1) {
1314             return -1;
1315         }
1316     }
1317 
1318     buflen = p - outbuf;
1319     /*
1320      * The mode data length field specifies the length in bytes of the
1321      * following data that is available to be transferred. The mode data
1322      * length does not include itself.
1323      */
1324     if (r->req.cmd.buf[0] == MODE_SENSE) {
1325         outbuf[0] = buflen - 1;
1326     } else { /* MODE_SENSE_10 */
1327         outbuf[0] = ((buflen - 2) >> 8) & 0xff;
1328         outbuf[1] = (buflen - 2) & 0xff;
1329     }
1330     return buflen;
1331 }
1332 
1333 static int scsi_disk_emulate_read_toc(SCSIRequest *req, uint8_t *outbuf)
1334 {
1335     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1336     int start_track, format, msf, toclen;
1337     uint64_t nb_sectors;
1338 
1339     msf = req->cmd.buf[1] & 2;
1340     format = req->cmd.buf[2] & 0xf;
1341     start_track = req->cmd.buf[6];
1342     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1343     trace_scsi_disk_emulate_read_toc(start_track, format, msf >> 1);
1344     nb_sectors /= s->qdev.blocksize / 512;
1345     switch (format) {
1346     case 0:
1347         toclen = cdrom_read_toc(nb_sectors, outbuf, msf, start_track);
1348         break;
1349     case 1:
1350         /* multi session : only a single session defined */
1351         toclen = 12;
1352         memset(outbuf, 0, 12);
1353         outbuf[1] = 0x0a;
1354         outbuf[2] = 0x01;
1355         outbuf[3] = 0x01;
1356         break;
1357     case 2:
1358         toclen = cdrom_read_toc_raw(nb_sectors, outbuf, msf, start_track);
1359         break;
1360     default:
1361         return -1;
1362     }
1363     return toclen;
1364 }
1365 
1366 static int scsi_disk_emulate_start_stop(SCSIDiskReq *r)
1367 {
1368     SCSIRequest *req = &r->req;
1369     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1370     bool start = req->cmd.buf[4] & 1;
1371     bool loej = req->cmd.buf[4] & 2; /* load on start, eject on !start */
1372     int pwrcnd = req->cmd.buf[4] & 0xf0;
1373 
1374     if (pwrcnd) {
1375         /* eject/load only happens for power condition == 0 */
1376         return 0;
1377     }
1378 
1379     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) && loej) {
1380         if (!start && !s->tray_open && s->tray_locked) {
1381             scsi_check_condition(r,
1382                                  blk_is_inserted(s->qdev.conf.blk)
1383                                  ? SENSE_CODE(ILLEGAL_REQ_REMOVAL_PREVENTED)
1384                                  : SENSE_CODE(NOT_READY_REMOVAL_PREVENTED));
1385             return -1;
1386         }
1387 
1388         if (s->tray_open != !start) {
1389             blk_eject(s->qdev.conf.blk, !start);
1390             s->tray_open = !start;
1391         }
1392     }
1393     return 0;
1394 }
1395 
1396 static void scsi_disk_emulate_read_data(SCSIRequest *req)
1397 {
1398     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1399     int buflen = r->iov.iov_len;
1400 
1401     if (buflen) {
1402         trace_scsi_disk_emulate_read_data(buflen);
1403         r->iov.iov_len = 0;
1404         r->started = true;
1405         scsi_req_data(&r->req, buflen);
1406         return;
1407     }
1408 
1409     /* This also clears the sense buffer for REQUEST SENSE.  */
1410     scsi_req_complete(&r->req, GOOD);
1411 }
1412 
1413 static int scsi_disk_check_mode_select(SCSIDiskState *s, int page,
1414                                        uint8_t *inbuf, int inlen)
1415 {
1416     uint8_t mode_current[SCSI_MAX_MODE_LEN];
1417     uint8_t mode_changeable[SCSI_MAX_MODE_LEN];
1418     uint8_t *p;
1419     int len, expected_len, changeable_len, i;
1420 
1421     /* The input buffer does not include the page header, so it is
1422      * off by 2 bytes.
1423      */
1424     expected_len = inlen + 2;
1425     if (expected_len > SCSI_MAX_MODE_LEN) {
1426         return -1;
1427     }
1428 
1429     p = mode_current;
1430     memset(mode_current, 0, inlen + 2);
1431     len = mode_sense_page(s, page, &p, 0);
1432     if (len < 0 || len != expected_len) {
1433         return -1;
1434     }
1435 
1436     p = mode_changeable;
1437     memset(mode_changeable, 0, inlen + 2);
1438     changeable_len = mode_sense_page(s, page, &p, 1);
1439     assert(changeable_len == len);
1440 
1441     /* Check that unchangeable bits are the same as what MODE SENSE
1442      * would return.
1443      */
1444     for (i = 2; i < len; i++) {
1445         if (((mode_current[i] ^ inbuf[i - 2]) & ~mode_changeable[i]) != 0) {
1446             return -1;
1447         }
1448     }
1449     return 0;
1450 }
1451 
1452 static void scsi_disk_apply_mode_select(SCSIDiskState *s, int page, uint8_t *p)
1453 {
1454     switch (page) {
1455     case MODE_PAGE_CACHING:
1456         blk_set_enable_write_cache(s->qdev.conf.blk, (p[0] & 4) != 0);
1457         break;
1458 
1459     default:
1460         break;
1461     }
1462 }
1463 
1464 static int mode_select_pages(SCSIDiskReq *r, uint8_t *p, int len, bool change)
1465 {
1466     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1467 
1468     while (len > 0) {
1469         int page, subpage, page_len;
1470 
1471         /* Parse both possible formats for the mode page headers.  */
1472         page = p[0] & 0x3f;
1473         if (p[0] & 0x40) {
1474             if (len < 4) {
1475                 goto invalid_param_len;
1476             }
1477             subpage = p[1];
1478             page_len = lduw_be_p(&p[2]);
1479             p += 4;
1480             len -= 4;
1481         } else {
1482             if (len < 2) {
1483                 goto invalid_param_len;
1484             }
1485             subpage = 0;
1486             page_len = p[1];
1487             p += 2;
1488             len -= 2;
1489         }
1490 
1491         if (subpage) {
1492             goto invalid_param;
1493         }
1494         if (page_len > len) {
1495             goto invalid_param_len;
1496         }
1497 
1498         if (!change) {
1499             if (scsi_disk_check_mode_select(s, page, p, page_len) < 0) {
1500                 goto invalid_param;
1501             }
1502         } else {
1503             scsi_disk_apply_mode_select(s, page, p);
1504         }
1505 
1506         p += page_len;
1507         len -= page_len;
1508     }
1509     return 0;
1510 
1511 invalid_param:
1512     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1513     return -1;
1514 
1515 invalid_param_len:
1516     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1517     return -1;
1518 }
1519 
1520 static void scsi_disk_emulate_mode_select(SCSIDiskReq *r, uint8_t *inbuf)
1521 {
1522     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1523     uint8_t *p = inbuf;
1524     int cmd = r->req.cmd.buf[0];
1525     int len = r->req.cmd.xfer;
1526     int hdr_len = (cmd == MODE_SELECT ? 4 : 8);
1527     int bd_len;
1528     int pass;
1529 
1530     /* We only support PF=1, SP=0.  */
1531     if ((r->req.cmd.buf[1] & 0x11) != 0x10) {
1532         goto invalid_field;
1533     }
1534 
1535     if (len < hdr_len) {
1536         goto invalid_param_len;
1537     }
1538 
1539     bd_len = (cmd == MODE_SELECT ? p[3] : lduw_be_p(&p[6]));
1540     len -= hdr_len;
1541     p += hdr_len;
1542     if (len < bd_len) {
1543         goto invalid_param_len;
1544     }
1545     if (bd_len != 0 && bd_len != 8) {
1546         goto invalid_param;
1547     }
1548 
1549     len -= bd_len;
1550     p += bd_len;
1551 
1552     /* Ensure no change is made if there is an error!  */
1553     for (pass = 0; pass < 2; pass++) {
1554         if (mode_select_pages(r, p, len, pass == 1) < 0) {
1555             assert(pass == 0);
1556             return;
1557         }
1558     }
1559     if (!blk_enable_write_cache(s->qdev.conf.blk)) {
1560         /* The request is used as the AIO opaque value, so add a ref.  */
1561         scsi_req_ref(&r->req);
1562         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
1563                          BLOCK_ACCT_FLUSH);
1564         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
1565         return;
1566     }
1567 
1568     scsi_req_complete(&r->req, GOOD);
1569     return;
1570 
1571 invalid_param:
1572     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1573     return;
1574 
1575 invalid_param_len:
1576     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1577     return;
1578 
1579 invalid_field:
1580     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1581 }
1582 
1583 static inline bool check_lba_range(SCSIDiskState *s,
1584                                    uint64_t sector_num, uint32_t nb_sectors)
1585 {
1586     /*
1587      * The first line tests that no overflow happens when computing the last
1588      * sector.  The second line tests that the last accessed sector is in
1589      * range.
1590      *
1591      * Careful, the computations should not underflow for nb_sectors == 0,
1592      * and a 0-block read to the first LBA beyond the end of device is
1593      * valid.
1594      */
1595     return (sector_num <= sector_num + nb_sectors &&
1596             sector_num + nb_sectors <= s->qdev.max_lba + 1);
1597 }
1598 
1599 typedef struct UnmapCBData {
1600     SCSIDiskReq *r;
1601     uint8_t *inbuf;
1602     int count;
1603 } UnmapCBData;
1604 
1605 static void scsi_unmap_complete(void *opaque, int ret);
1606 
1607 static void scsi_unmap_complete_noio(UnmapCBData *data, int ret)
1608 {
1609     SCSIDiskReq *r = data->r;
1610     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1611 
1612     assert(r->req.aiocb == NULL);
1613 
1614     if (data->count > 0) {
1615         r->sector = ldq_be_p(&data->inbuf[0])
1616             * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1617         r->sector_count = (ldl_be_p(&data->inbuf[8]) & 0xffffffffULL)
1618             * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1619         if (!check_lba_range(s, r->sector, r->sector_count)) {
1620             block_acct_invalid(blk_get_stats(s->qdev.conf.blk),
1621                                BLOCK_ACCT_UNMAP);
1622             scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1623             goto done;
1624         }
1625 
1626         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1627                          r->sector_count * BDRV_SECTOR_SIZE,
1628                          BLOCK_ACCT_UNMAP);
1629 
1630         r->req.aiocb = blk_aio_pdiscard(s->qdev.conf.blk,
1631                                         r->sector * BDRV_SECTOR_SIZE,
1632                                         r->sector_count * BDRV_SECTOR_SIZE,
1633                                         scsi_unmap_complete, data);
1634         data->count--;
1635         data->inbuf += 16;
1636         return;
1637     }
1638 
1639     scsi_req_complete(&r->req, GOOD);
1640 
1641 done:
1642     scsi_req_unref(&r->req);
1643     g_free(data);
1644 }
1645 
1646 static void scsi_unmap_complete(void *opaque, int ret)
1647 {
1648     UnmapCBData *data = opaque;
1649     SCSIDiskReq *r = data->r;
1650     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1651 
1652     assert(r->req.aiocb != NULL);
1653     r->req.aiocb = NULL;
1654 
1655     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
1656     if (scsi_disk_req_check_error(r, ret, true)) {
1657         scsi_req_unref(&r->req);
1658         g_free(data);
1659     } else {
1660         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1661         scsi_unmap_complete_noio(data, ret);
1662     }
1663     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1664 }
1665 
1666 static void scsi_disk_emulate_unmap(SCSIDiskReq *r, uint8_t *inbuf)
1667 {
1668     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1669     uint8_t *p = inbuf;
1670     int len = r->req.cmd.xfer;
1671     UnmapCBData *data;
1672 
1673     /* Reject ANCHOR=1.  */
1674     if (r->req.cmd.buf[1] & 0x1) {
1675         goto invalid_field;
1676     }
1677 
1678     if (len < 8) {
1679         goto invalid_param_len;
1680     }
1681     if (len < lduw_be_p(&p[0]) + 2) {
1682         goto invalid_param_len;
1683     }
1684     if (len < lduw_be_p(&p[2]) + 8) {
1685         goto invalid_param_len;
1686     }
1687     if (lduw_be_p(&p[2]) & 15) {
1688         goto invalid_param_len;
1689     }
1690 
1691     if (blk_is_read_only(s->qdev.conf.blk)) {
1692         block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1693         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1694         return;
1695     }
1696 
1697     data = g_new0(UnmapCBData, 1);
1698     data->r = r;
1699     data->inbuf = &p[8];
1700     data->count = lduw_be_p(&p[2]) >> 4;
1701 
1702     /* The matching unref is in scsi_unmap_complete, before data is freed.  */
1703     scsi_req_ref(&r->req);
1704     scsi_unmap_complete_noio(data, 0);
1705     return;
1706 
1707 invalid_param_len:
1708     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1709     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1710     return;
1711 
1712 invalid_field:
1713     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1714     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1715 }
1716 
1717 typedef struct WriteSameCBData {
1718     SCSIDiskReq *r;
1719     int64_t sector;
1720     int nb_sectors;
1721     QEMUIOVector qiov;
1722     struct iovec iov;
1723 } WriteSameCBData;
1724 
1725 static void scsi_write_same_complete(void *opaque, int ret)
1726 {
1727     WriteSameCBData *data = opaque;
1728     SCSIDiskReq *r = data->r;
1729     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1730 
1731     assert(r->req.aiocb != NULL);
1732     r->req.aiocb = NULL;
1733     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
1734     if (scsi_disk_req_check_error(r, ret, true)) {
1735         goto done;
1736     }
1737 
1738     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1739 
1740     data->nb_sectors -= data->iov.iov_len / 512;
1741     data->sector += data->iov.iov_len / 512;
1742     data->iov.iov_len = MIN(data->nb_sectors * 512, data->iov.iov_len);
1743     if (data->iov.iov_len) {
1744         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1745                          data->iov.iov_len, BLOCK_ACCT_WRITE);
1746         /* Reinitialize qiov, to handle unaligned WRITE SAME request
1747          * where final qiov may need smaller size */
1748         qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1749         r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1750                                        data->sector << BDRV_SECTOR_BITS,
1751                                        &data->qiov, 0,
1752                                        scsi_write_same_complete, data);
1753         aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1754         return;
1755     }
1756 
1757     scsi_req_complete(&r->req, GOOD);
1758 
1759 done:
1760     scsi_req_unref(&r->req);
1761     qemu_vfree(data->iov.iov_base);
1762     g_free(data);
1763     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1764 }
1765 
1766 static void scsi_disk_emulate_write_same(SCSIDiskReq *r, uint8_t *inbuf)
1767 {
1768     SCSIRequest *req = &r->req;
1769     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1770     uint32_t nb_sectors = scsi_data_cdb_xfer(r->req.cmd.buf);
1771     WriteSameCBData *data;
1772     uint8_t *buf;
1773     int i;
1774 
1775     /* Fail if PBDATA=1 or LBDATA=1 or ANCHOR=1.  */
1776     if (nb_sectors == 0 || (req->cmd.buf[1] & 0x16)) {
1777         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1778         return;
1779     }
1780 
1781     if (blk_is_read_only(s->qdev.conf.blk)) {
1782         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1783         return;
1784     }
1785     if (!check_lba_range(s, r->req.cmd.lba, nb_sectors)) {
1786         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1787         return;
1788     }
1789 
1790     if ((req->cmd.buf[1] & 0x1) || buffer_is_zero(inbuf, s->qdev.blocksize)) {
1791         int flags = (req->cmd.buf[1] & 0x8) ? BDRV_REQ_MAY_UNMAP : 0;
1792 
1793         /* The request is used as the AIO opaque value, so add a ref.  */
1794         scsi_req_ref(&r->req);
1795         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1796                          nb_sectors * s->qdev.blocksize,
1797                         BLOCK_ACCT_WRITE);
1798         r->req.aiocb = blk_aio_pwrite_zeroes(s->qdev.conf.blk,
1799                                 r->req.cmd.lba * s->qdev.blocksize,
1800                                 nb_sectors * s->qdev.blocksize,
1801                                 flags, scsi_aio_complete, r);
1802         return;
1803     }
1804 
1805     data = g_new0(WriteSameCBData, 1);
1806     data->r = r;
1807     data->sector = r->req.cmd.lba * (s->qdev.blocksize / 512);
1808     data->nb_sectors = nb_sectors * (s->qdev.blocksize / 512);
1809     data->iov.iov_len = MIN(data->nb_sectors * 512, SCSI_WRITE_SAME_MAX);
1810     data->iov.iov_base = buf = blk_blockalign(s->qdev.conf.blk,
1811                                               data->iov.iov_len);
1812     qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1813 
1814     for (i = 0; i < data->iov.iov_len; i += s->qdev.blocksize) {
1815         memcpy(&buf[i], inbuf, s->qdev.blocksize);
1816     }
1817 
1818     scsi_req_ref(&r->req);
1819     block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1820                      data->iov.iov_len, BLOCK_ACCT_WRITE);
1821     r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1822                                    data->sector << BDRV_SECTOR_BITS,
1823                                    &data->qiov, 0,
1824                                    scsi_write_same_complete, data);
1825 }
1826 
1827 static void scsi_disk_emulate_write_data(SCSIRequest *req)
1828 {
1829     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1830 
1831     if (r->iov.iov_len) {
1832         int buflen = r->iov.iov_len;
1833         trace_scsi_disk_emulate_write_data(buflen);
1834         r->iov.iov_len = 0;
1835         scsi_req_data(&r->req, buflen);
1836         return;
1837     }
1838 
1839     switch (req->cmd.buf[0]) {
1840     case MODE_SELECT:
1841     case MODE_SELECT_10:
1842         /* This also clears the sense buffer for REQUEST SENSE.  */
1843         scsi_disk_emulate_mode_select(r, r->iov.iov_base);
1844         break;
1845 
1846     case UNMAP:
1847         scsi_disk_emulate_unmap(r, r->iov.iov_base);
1848         break;
1849 
1850     case VERIFY_10:
1851     case VERIFY_12:
1852     case VERIFY_16:
1853         if (r->req.status == -1) {
1854             scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1855         }
1856         break;
1857 
1858     case WRITE_SAME_10:
1859     case WRITE_SAME_16:
1860         scsi_disk_emulate_write_same(r, r->iov.iov_base);
1861         break;
1862 
1863     default:
1864         abort();
1865     }
1866 }
1867 
1868 static int32_t scsi_disk_emulate_command(SCSIRequest *req, uint8_t *buf)
1869 {
1870     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1871     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1872     uint64_t nb_sectors;
1873     uint8_t *outbuf;
1874     int buflen;
1875 
1876     switch (req->cmd.buf[0]) {
1877     case INQUIRY:
1878     case MODE_SENSE:
1879     case MODE_SENSE_10:
1880     case RESERVE:
1881     case RESERVE_10:
1882     case RELEASE:
1883     case RELEASE_10:
1884     case START_STOP:
1885     case ALLOW_MEDIUM_REMOVAL:
1886     case GET_CONFIGURATION:
1887     case GET_EVENT_STATUS_NOTIFICATION:
1888     case MECHANISM_STATUS:
1889     case REQUEST_SENSE:
1890         break;
1891 
1892     default:
1893         if (!blk_is_available(s->qdev.conf.blk)) {
1894             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
1895             return 0;
1896         }
1897         break;
1898     }
1899 
1900     /*
1901      * FIXME: we shouldn't return anything bigger than 4k, but the code
1902      * requires the buffer to be as big as req->cmd.xfer in several
1903      * places.  So, do not allow CDBs with a very large ALLOCATION
1904      * LENGTH.  The real fix would be to modify scsi_read_data and
1905      * dma_buf_read, so that they return data beyond the buflen
1906      * as all zeros.
1907      */
1908     if (req->cmd.xfer > 65536) {
1909         goto illegal_request;
1910     }
1911     r->buflen = MAX(4096, req->cmd.xfer);
1912 
1913     if (!r->iov.iov_base) {
1914         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
1915     }
1916 
1917     buflen = req->cmd.xfer;
1918     outbuf = r->iov.iov_base;
1919     memset(outbuf, 0, r->buflen);
1920     switch (req->cmd.buf[0]) {
1921     case TEST_UNIT_READY:
1922         assert(blk_is_available(s->qdev.conf.blk));
1923         break;
1924     case INQUIRY:
1925         buflen = scsi_disk_emulate_inquiry(req, outbuf);
1926         if (buflen < 0) {
1927             goto illegal_request;
1928         }
1929         break;
1930     case MODE_SENSE:
1931     case MODE_SENSE_10:
1932         buflen = scsi_disk_emulate_mode_sense(r, outbuf);
1933         if (buflen < 0) {
1934             goto illegal_request;
1935         }
1936         break;
1937     case READ_TOC:
1938         buflen = scsi_disk_emulate_read_toc(req, outbuf);
1939         if (buflen < 0) {
1940             goto illegal_request;
1941         }
1942         break;
1943     case RESERVE:
1944         if (req->cmd.buf[1] & 1) {
1945             goto illegal_request;
1946         }
1947         break;
1948     case RESERVE_10:
1949         if (req->cmd.buf[1] & 3) {
1950             goto illegal_request;
1951         }
1952         break;
1953     case RELEASE:
1954         if (req->cmd.buf[1] & 1) {
1955             goto illegal_request;
1956         }
1957         break;
1958     case RELEASE_10:
1959         if (req->cmd.buf[1] & 3) {
1960             goto illegal_request;
1961         }
1962         break;
1963     case START_STOP:
1964         if (scsi_disk_emulate_start_stop(r) < 0) {
1965             return 0;
1966         }
1967         break;
1968     case ALLOW_MEDIUM_REMOVAL:
1969         s->tray_locked = req->cmd.buf[4] & 1;
1970         blk_lock_medium(s->qdev.conf.blk, req->cmd.buf[4] & 1);
1971         break;
1972     case READ_CAPACITY_10:
1973         /* The normal LEN field for this command is zero.  */
1974         memset(outbuf, 0, 8);
1975         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1976         if (!nb_sectors) {
1977             scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
1978             return 0;
1979         }
1980         if ((req->cmd.buf[8] & 1) == 0 && req->cmd.lba) {
1981             goto illegal_request;
1982         }
1983         nb_sectors /= s->qdev.blocksize / 512;
1984         /* Returned value is the address of the last sector.  */
1985         nb_sectors--;
1986         /* Remember the new size for read/write sanity checking. */
1987         s->qdev.max_lba = nb_sectors;
1988         /* Clip to 2TB, instead of returning capacity modulo 2TB. */
1989         if (nb_sectors > UINT32_MAX) {
1990             nb_sectors = UINT32_MAX;
1991         }
1992         outbuf[0] = (nb_sectors >> 24) & 0xff;
1993         outbuf[1] = (nb_sectors >> 16) & 0xff;
1994         outbuf[2] = (nb_sectors >> 8) & 0xff;
1995         outbuf[3] = nb_sectors & 0xff;
1996         outbuf[4] = 0;
1997         outbuf[5] = 0;
1998         outbuf[6] = s->qdev.blocksize >> 8;
1999         outbuf[7] = 0;
2000         break;
2001     case REQUEST_SENSE:
2002         /* Just return "NO SENSE".  */
2003         buflen = scsi_convert_sense(NULL, 0, outbuf, r->buflen,
2004                                     (req->cmd.buf[1] & 1) == 0);
2005         if (buflen < 0) {
2006             goto illegal_request;
2007         }
2008         break;
2009     case MECHANISM_STATUS:
2010         buflen = scsi_emulate_mechanism_status(s, outbuf);
2011         if (buflen < 0) {
2012             goto illegal_request;
2013         }
2014         break;
2015     case GET_CONFIGURATION:
2016         buflen = scsi_get_configuration(s, outbuf);
2017         if (buflen < 0) {
2018             goto illegal_request;
2019         }
2020         break;
2021     case GET_EVENT_STATUS_NOTIFICATION:
2022         buflen = scsi_get_event_status_notification(s, r, outbuf);
2023         if (buflen < 0) {
2024             goto illegal_request;
2025         }
2026         break;
2027     case READ_DISC_INFORMATION:
2028         buflen = scsi_read_disc_information(s, r, outbuf);
2029         if (buflen < 0) {
2030             goto illegal_request;
2031         }
2032         break;
2033     case READ_DVD_STRUCTURE:
2034         buflen = scsi_read_dvd_structure(s, r, outbuf);
2035         if (buflen < 0) {
2036             goto illegal_request;
2037         }
2038         break;
2039     case SERVICE_ACTION_IN_16:
2040         /* Service Action In subcommands. */
2041         if ((req->cmd.buf[1] & 31) == SAI_READ_CAPACITY_16) {
2042             trace_scsi_disk_emulate_command_SAI_16();
2043             memset(outbuf, 0, req->cmd.xfer);
2044             blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2045             if (!nb_sectors) {
2046                 scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
2047                 return 0;
2048             }
2049             if ((req->cmd.buf[14] & 1) == 0 && req->cmd.lba) {
2050                 goto illegal_request;
2051             }
2052             nb_sectors /= s->qdev.blocksize / 512;
2053             /* Returned value is the address of the last sector.  */
2054             nb_sectors--;
2055             /* Remember the new size for read/write sanity checking. */
2056             s->qdev.max_lba = nb_sectors;
2057             outbuf[0] = (nb_sectors >> 56) & 0xff;
2058             outbuf[1] = (nb_sectors >> 48) & 0xff;
2059             outbuf[2] = (nb_sectors >> 40) & 0xff;
2060             outbuf[3] = (nb_sectors >> 32) & 0xff;
2061             outbuf[4] = (nb_sectors >> 24) & 0xff;
2062             outbuf[5] = (nb_sectors >> 16) & 0xff;
2063             outbuf[6] = (nb_sectors >> 8) & 0xff;
2064             outbuf[7] = nb_sectors & 0xff;
2065             outbuf[8] = 0;
2066             outbuf[9] = 0;
2067             outbuf[10] = s->qdev.blocksize >> 8;
2068             outbuf[11] = 0;
2069             outbuf[12] = 0;
2070             outbuf[13] = get_physical_block_exp(&s->qdev.conf);
2071 
2072             /* set TPE bit if the format supports discard */
2073             if (s->qdev.conf.discard_granularity) {
2074                 outbuf[14] = 0x80;
2075             }
2076 
2077             /* Protection, exponent and lowest lba field left blank. */
2078             break;
2079         }
2080         trace_scsi_disk_emulate_command_SAI_unsupported();
2081         goto illegal_request;
2082     case SYNCHRONIZE_CACHE:
2083         /* The request is used as the AIO opaque value, so add a ref.  */
2084         scsi_req_ref(&r->req);
2085         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
2086                          BLOCK_ACCT_FLUSH);
2087         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
2088         return 0;
2089     case SEEK_10:
2090         trace_scsi_disk_emulate_command_SEEK_10(r->req.cmd.lba);
2091         if (r->req.cmd.lba > s->qdev.max_lba) {
2092             goto illegal_lba;
2093         }
2094         break;
2095     case MODE_SELECT:
2096         trace_scsi_disk_emulate_command_MODE_SELECT(r->req.cmd.xfer);
2097         break;
2098     case MODE_SELECT_10:
2099         trace_scsi_disk_emulate_command_MODE_SELECT_10(r->req.cmd.xfer);
2100         break;
2101     case UNMAP:
2102         trace_scsi_disk_emulate_command_UNMAP(r->req.cmd.xfer);
2103         break;
2104     case VERIFY_10:
2105     case VERIFY_12:
2106     case VERIFY_16:
2107         trace_scsi_disk_emulate_command_VERIFY((req->cmd.buf[1] >> 1) & 3);
2108         if (req->cmd.buf[1] & 6) {
2109             goto illegal_request;
2110         }
2111         break;
2112     case WRITE_SAME_10:
2113     case WRITE_SAME_16:
2114         trace_scsi_disk_emulate_command_WRITE_SAME(
2115                 req->cmd.buf[0] == WRITE_SAME_10 ? 10 : 16, r->req.cmd.xfer);
2116         break;
2117     default:
2118         trace_scsi_disk_emulate_command_UNKNOWN(buf[0],
2119                                                 scsi_command_name(buf[0]));
2120         scsi_check_condition(r, SENSE_CODE(INVALID_OPCODE));
2121         return 0;
2122     }
2123     assert(!r->req.aiocb);
2124     r->iov.iov_len = MIN(r->buflen, req->cmd.xfer);
2125     if (r->iov.iov_len == 0) {
2126         scsi_req_complete(&r->req, GOOD);
2127     }
2128     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2129         assert(r->iov.iov_len == req->cmd.xfer);
2130         return -r->iov.iov_len;
2131     } else {
2132         return r->iov.iov_len;
2133     }
2134 
2135 illegal_request:
2136     if (r->req.status == -1) {
2137         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2138     }
2139     return 0;
2140 
2141 illegal_lba:
2142     scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2143     return 0;
2144 }
2145 
2146 /* Execute a scsi command.  Returns the length of the data expected by the
2147    command.  This will be Positive for data transfers from the device
2148    (eg. disk reads), negative for transfers to the device (eg. disk writes),
2149    and zero if the command does not transfer any data.  */
2150 
2151 static int32_t scsi_disk_dma_command(SCSIRequest *req, uint8_t *buf)
2152 {
2153     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
2154     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2155     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
2156     uint32_t len;
2157     uint8_t command;
2158 
2159     command = buf[0];
2160 
2161     if (!blk_is_available(s->qdev.conf.blk)) {
2162         scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
2163         return 0;
2164     }
2165 
2166     len = scsi_data_cdb_xfer(r->req.cmd.buf);
2167     switch (command) {
2168     case READ_6:
2169     case READ_10:
2170     case READ_12:
2171     case READ_16:
2172         trace_scsi_disk_dma_command_READ(r->req.cmd.lba, len);
2173         /* Protection information is not supported.  For SCSI versions 2 and
2174          * older (as determined by snooping the guest's INQUIRY commands),
2175          * there is no RD/WR/VRPROTECT, so skip this check in these versions.
2176          */
2177         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2178             goto illegal_request;
2179         }
2180         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2181             goto illegal_lba;
2182         }
2183         r->sector = r->req.cmd.lba * (s->qdev.blocksize / 512);
2184         r->sector_count = len * (s->qdev.blocksize / 512);
2185         break;
2186     case WRITE_6:
2187     case WRITE_10:
2188     case WRITE_12:
2189     case WRITE_16:
2190     case WRITE_VERIFY_10:
2191     case WRITE_VERIFY_12:
2192     case WRITE_VERIFY_16:
2193         if (blk_is_read_only(s->qdev.conf.blk)) {
2194             scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
2195             return 0;
2196         }
2197         trace_scsi_disk_dma_command_WRITE(
2198                 (command & 0xe) == 0xe ? "And Verify " : "",
2199                 r->req.cmd.lba, len);
2200         /* fall through */
2201     case VERIFY_10:
2202     case VERIFY_12:
2203     case VERIFY_16:
2204         /* We get here only for BYTCHK == 0x01 and only for scsi-block.
2205          * As far as DMA is concerned, we can treat it the same as a write;
2206          * scsi_block_do_sgio will send VERIFY commands.
2207          */
2208         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2209             goto illegal_request;
2210         }
2211         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2212             goto illegal_lba;
2213         }
2214         r->sector = r->req.cmd.lba * (s->qdev.blocksize / 512);
2215         r->sector_count = len * (s->qdev.blocksize / 512);
2216         break;
2217     default:
2218         abort();
2219     illegal_request:
2220         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2221         return 0;
2222     illegal_lba:
2223         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2224         return 0;
2225     }
2226     r->need_fua_emulation = sdc->need_fua_emulation(&r->req.cmd);
2227     if (r->sector_count == 0) {
2228         scsi_req_complete(&r->req, GOOD);
2229     }
2230     assert(r->iov.iov_len == 0);
2231     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2232         return -r->sector_count * 512;
2233     } else {
2234         return r->sector_count * 512;
2235     }
2236 }
2237 
2238 static void scsi_disk_reset(DeviceState *dev)
2239 {
2240     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev.qdev, dev);
2241     uint64_t nb_sectors;
2242 
2243     scsi_device_purge_requests(&s->qdev, SENSE_CODE(RESET));
2244 
2245     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2246     nb_sectors /= s->qdev.blocksize / 512;
2247     if (nb_sectors) {
2248         nb_sectors--;
2249     }
2250     s->qdev.max_lba = nb_sectors;
2251     /* reset tray statuses */
2252     s->tray_locked = 0;
2253     s->tray_open = 0;
2254 
2255     s->qdev.scsi_version = s->qdev.default_scsi_version;
2256 }
2257 
2258 static void scsi_disk_resize_cb(void *opaque)
2259 {
2260     SCSIDiskState *s = opaque;
2261 
2262     /* SPC lists this sense code as available only for
2263      * direct-access devices.
2264      */
2265     if (s->qdev.type == TYPE_DISK) {
2266         scsi_device_report_change(&s->qdev, SENSE_CODE(CAPACITY_CHANGED));
2267     }
2268 }
2269 
2270 static void scsi_cd_change_media_cb(void *opaque, bool load, Error **errp)
2271 {
2272     SCSIDiskState *s = opaque;
2273 
2274     /*
2275      * When a CD gets changed, we have to report an ejected state and
2276      * then a loaded state to guests so that they detect tray
2277      * open/close and media change events.  Guests that do not use
2278      * GET_EVENT_STATUS_NOTIFICATION to detect such tray open/close
2279      * states rely on this behavior.
2280      *
2281      * media_changed governs the state machine used for unit attention
2282      * report.  media_event is used by GET EVENT STATUS NOTIFICATION.
2283      */
2284     s->media_changed = load;
2285     s->tray_open = !load;
2286     scsi_device_set_ua(&s->qdev, SENSE_CODE(UNIT_ATTENTION_NO_MEDIUM));
2287     s->media_event = true;
2288     s->eject_request = false;
2289 }
2290 
2291 static void scsi_cd_eject_request_cb(void *opaque, bool force)
2292 {
2293     SCSIDiskState *s = opaque;
2294 
2295     s->eject_request = true;
2296     if (force) {
2297         s->tray_locked = false;
2298     }
2299 }
2300 
2301 static bool scsi_cd_is_tray_open(void *opaque)
2302 {
2303     return ((SCSIDiskState *)opaque)->tray_open;
2304 }
2305 
2306 static bool scsi_cd_is_medium_locked(void *opaque)
2307 {
2308     return ((SCSIDiskState *)opaque)->tray_locked;
2309 }
2310 
2311 static const BlockDevOps scsi_disk_removable_block_ops = {
2312     .change_media_cb = scsi_cd_change_media_cb,
2313     .eject_request_cb = scsi_cd_eject_request_cb,
2314     .is_tray_open = scsi_cd_is_tray_open,
2315     .is_medium_locked = scsi_cd_is_medium_locked,
2316 
2317     .resize_cb = scsi_disk_resize_cb,
2318 };
2319 
2320 static const BlockDevOps scsi_disk_block_ops = {
2321     .resize_cb = scsi_disk_resize_cb,
2322 };
2323 
2324 static void scsi_disk_unit_attention_reported(SCSIDevice *dev)
2325 {
2326     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2327     if (s->media_changed) {
2328         s->media_changed = false;
2329         scsi_device_set_ua(&s->qdev, SENSE_CODE(MEDIUM_CHANGED));
2330     }
2331 }
2332 
2333 static void scsi_realize(SCSIDevice *dev, Error **errp)
2334 {
2335     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2336     bool read_only;
2337 
2338     if (!s->qdev.conf.blk) {
2339         error_setg(errp, "drive property not set");
2340         return;
2341     }
2342 
2343     if (!(s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2344         !blk_is_inserted(s->qdev.conf.blk)) {
2345         error_setg(errp, "Device needs media, but drive is empty");
2346         return;
2347     }
2348 
2349     blkconf_blocksizes(&s->qdev.conf);
2350 
2351     if (s->qdev.conf.logical_block_size >
2352         s->qdev.conf.physical_block_size) {
2353         error_setg(errp,
2354                    "logical_block_size > physical_block_size not supported");
2355         return;
2356     }
2357 
2358     if (blk_get_aio_context(s->qdev.conf.blk) != qemu_get_aio_context() &&
2359         !s->qdev.hba_supports_iothread)
2360     {
2361         error_setg(errp, "HBA does not support iothreads");
2362         return;
2363     }
2364 
2365     if (dev->type == TYPE_DISK) {
2366         if (!blkconf_geometry(&dev->conf, NULL, 65535, 255, 255, errp)) {
2367             return;
2368         }
2369     }
2370 
2371     read_only = blk_is_read_only(s->qdev.conf.blk);
2372     if (dev->type == TYPE_ROM) {
2373         read_only = true;
2374     }
2375 
2376     if (!blkconf_apply_backend_options(&dev->conf, read_only,
2377                                        dev->type == TYPE_DISK, errp)) {
2378         return;
2379     }
2380 
2381     if (s->qdev.conf.discard_granularity == -1) {
2382         s->qdev.conf.discard_granularity =
2383             MAX(s->qdev.conf.logical_block_size, DEFAULT_DISCARD_GRANULARITY);
2384     }
2385 
2386     if (!s->version) {
2387         s->version = g_strdup(qemu_hw_version());
2388     }
2389     if (!s->vendor) {
2390         s->vendor = g_strdup("QEMU");
2391     }
2392     if (!s->device_id) {
2393         if (s->serial) {
2394             s->device_id = g_strdup_printf("%.20s", s->serial);
2395         } else {
2396             const char *str = blk_name(s->qdev.conf.blk);
2397             if (str && *str) {
2398                 s->device_id = g_strdup(str);
2399             }
2400         }
2401     }
2402 
2403     if (blk_is_sg(s->qdev.conf.blk)) {
2404         error_setg(errp, "unwanted /dev/sg*");
2405         return;
2406     }
2407 
2408     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2409             !(s->features & (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS))) {
2410         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_removable_block_ops, s);
2411     } else {
2412         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_block_ops, s);
2413     }
2414     blk_set_guest_block_size(s->qdev.conf.blk, s->qdev.blocksize);
2415 
2416     blk_iostatus_enable(s->qdev.conf.blk);
2417 }
2418 
2419 static void scsi_hd_realize(SCSIDevice *dev, Error **errp)
2420 {
2421     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2422     AioContext *ctx = NULL;
2423     /* can happen for devices without drive. The error message for missing
2424      * backend will be issued in scsi_realize
2425      */
2426     if (s->qdev.conf.blk) {
2427         ctx = blk_get_aio_context(s->qdev.conf.blk);
2428         aio_context_acquire(ctx);
2429         blkconf_blocksizes(&s->qdev.conf);
2430     }
2431     s->qdev.blocksize = s->qdev.conf.logical_block_size;
2432     s->qdev.type = TYPE_DISK;
2433     if (!s->product) {
2434         s->product = g_strdup("QEMU HARDDISK");
2435     }
2436     scsi_realize(&s->qdev, errp);
2437     if (ctx) {
2438         aio_context_release(ctx);
2439     }
2440 }
2441 
2442 static void scsi_cd_realize(SCSIDevice *dev, Error **errp)
2443 {
2444     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2445     AioContext *ctx;
2446     int ret;
2447 
2448     if (!dev->conf.blk) {
2449         /* Anonymous BlockBackend for an empty drive. As we put it into
2450          * dev->conf, qdev takes care of detaching on unplug. */
2451         dev->conf.blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
2452         ret = blk_attach_dev(dev->conf.blk, &dev->qdev);
2453         assert(ret == 0);
2454     }
2455 
2456     ctx = blk_get_aio_context(dev->conf.blk);
2457     aio_context_acquire(ctx);
2458     s->qdev.blocksize = 2048;
2459     s->qdev.type = TYPE_ROM;
2460     s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2461     if (!s->product) {
2462         s->product = g_strdup("QEMU CD-ROM");
2463     }
2464     scsi_realize(&s->qdev, errp);
2465     aio_context_release(ctx);
2466 }
2467 
2468 static void scsi_disk_realize(SCSIDevice *dev, Error **errp)
2469 {
2470     DriveInfo *dinfo;
2471     Error *local_err = NULL;
2472 
2473     if (!dev->conf.blk) {
2474         scsi_realize(dev, &local_err);
2475         assert(local_err);
2476         error_propagate(errp, local_err);
2477         return;
2478     }
2479 
2480     dinfo = blk_legacy_dinfo(dev->conf.blk);
2481     if (dinfo && dinfo->media_cd) {
2482         scsi_cd_realize(dev, errp);
2483     } else {
2484         scsi_hd_realize(dev, errp);
2485     }
2486 }
2487 
2488 static const SCSIReqOps scsi_disk_emulate_reqops = {
2489     .size         = sizeof(SCSIDiskReq),
2490     .free_req     = scsi_free_request,
2491     .send_command = scsi_disk_emulate_command,
2492     .read_data    = scsi_disk_emulate_read_data,
2493     .write_data   = scsi_disk_emulate_write_data,
2494     .get_buf      = scsi_get_buf,
2495 };
2496 
2497 static const SCSIReqOps scsi_disk_dma_reqops = {
2498     .size         = sizeof(SCSIDiskReq),
2499     .free_req     = scsi_free_request,
2500     .send_command = scsi_disk_dma_command,
2501     .read_data    = scsi_read_data,
2502     .write_data   = scsi_write_data,
2503     .get_buf      = scsi_get_buf,
2504     .load_request = scsi_disk_load_request,
2505     .save_request = scsi_disk_save_request,
2506 };
2507 
2508 static const SCSIReqOps *const scsi_disk_reqops_dispatch[256] = {
2509     [TEST_UNIT_READY]                 = &scsi_disk_emulate_reqops,
2510     [INQUIRY]                         = &scsi_disk_emulate_reqops,
2511     [MODE_SENSE]                      = &scsi_disk_emulate_reqops,
2512     [MODE_SENSE_10]                   = &scsi_disk_emulate_reqops,
2513     [START_STOP]                      = &scsi_disk_emulate_reqops,
2514     [ALLOW_MEDIUM_REMOVAL]            = &scsi_disk_emulate_reqops,
2515     [READ_CAPACITY_10]                = &scsi_disk_emulate_reqops,
2516     [READ_TOC]                        = &scsi_disk_emulate_reqops,
2517     [READ_DVD_STRUCTURE]              = &scsi_disk_emulate_reqops,
2518     [READ_DISC_INFORMATION]           = &scsi_disk_emulate_reqops,
2519     [GET_CONFIGURATION]               = &scsi_disk_emulate_reqops,
2520     [GET_EVENT_STATUS_NOTIFICATION]   = &scsi_disk_emulate_reqops,
2521     [MECHANISM_STATUS]                = &scsi_disk_emulate_reqops,
2522     [SERVICE_ACTION_IN_16]            = &scsi_disk_emulate_reqops,
2523     [REQUEST_SENSE]                   = &scsi_disk_emulate_reqops,
2524     [SYNCHRONIZE_CACHE]               = &scsi_disk_emulate_reqops,
2525     [SEEK_10]                         = &scsi_disk_emulate_reqops,
2526     [MODE_SELECT]                     = &scsi_disk_emulate_reqops,
2527     [MODE_SELECT_10]                  = &scsi_disk_emulate_reqops,
2528     [UNMAP]                           = &scsi_disk_emulate_reqops,
2529     [WRITE_SAME_10]                   = &scsi_disk_emulate_reqops,
2530     [WRITE_SAME_16]                   = &scsi_disk_emulate_reqops,
2531     [VERIFY_10]                       = &scsi_disk_emulate_reqops,
2532     [VERIFY_12]                       = &scsi_disk_emulate_reqops,
2533     [VERIFY_16]                       = &scsi_disk_emulate_reqops,
2534 
2535     [READ_6]                          = &scsi_disk_dma_reqops,
2536     [READ_10]                         = &scsi_disk_dma_reqops,
2537     [READ_12]                         = &scsi_disk_dma_reqops,
2538     [READ_16]                         = &scsi_disk_dma_reqops,
2539     [WRITE_6]                         = &scsi_disk_dma_reqops,
2540     [WRITE_10]                        = &scsi_disk_dma_reqops,
2541     [WRITE_12]                        = &scsi_disk_dma_reqops,
2542     [WRITE_16]                        = &scsi_disk_dma_reqops,
2543     [WRITE_VERIFY_10]                 = &scsi_disk_dma_reqops,
2544     [WRITE_VERIFY_12]                 = &scsi_disk_dma_reqops,
2545     [WRITE_VERIFY_16]                 = &scsi_disk_dma_reqops,
2546 };
2547 
2548 static void scsi_disk_new_request_dump(uint32_t lun, uint32_t tag, uint8_t *buf)
2549 {
2550     int i;
2551     int len = scsi_cdb_length(buf);
2552     char *line_buffer, *p;
2553 
2554     line_buffer = g_malloc(len * 5 + 1);
2555 
2556     for (i = 0, p = line_buffer; i < len; i++) {
2557         p += sprintf(p, " 0x%02x", buf[i]);
2558     }
2559     trace_scsi_disk_new_request(lun, tag, line_buffer);
2560 
2561     g_free(line_buffer);
2562 }
2563 
2564 static SCSIRequest *scsi_new_request(SCSIDevice *d, uint32_t tag, uint32_t lun,
2565                                      uint8_t *buf, void *hba_private)
2566 {
2567     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2568     SCSIRequest *req;
2569     const SCSIReqOps *ops;
2570     uint8_t command;
2571 
2572     command = buf[0];
2573     ops = scsi_disk_reqops_dispatch[command];
2574     if (!ops) {
2575         ops = &scsi_disk_emulate_reqops;
2576     }
2577     req = scsi_req_alloc(ops, &s->qdev, tag, lun, hba_private);
2578 
2579     if (trace_event_get_state_backends(TRACE_SCSI_DISK_NEW_REQUEST)) {
2580         scsi_disk_new_request_dump(lun, tag, buf);
2581     }
2582 
2583     return req;
2584 }
2585 
2586 #ifdef __linux__
2587 static int get_device_type(SCSIDiskState *s)
2588 {
2589     uint8_t cmd[16];
2590     uint8_t buf[36];
2591     int ret;
2592 
2593     memset(cmd, 0, sizeof(cmd));
2594     memset(buf, 0, sizeof(buf));
2595     cmd[0] = INQUIRY;
2596     cmd[4] = sizeof(buf);
2597 
2598     ret = scsi_SG_IO_FROM_DEV(s->qdev.conf.blk, cmd, sizeof(cmd),
2599                               buf, sizeof(buf));
2600     if (ret < 0) {
2601         return -1;
2602     }
2603     s->qdev.type = buf[0];
2604     if (buf[1] & 0x80) {
2605         s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2606     }
2607     return 0;
2608 }
2609 
2610 static void scsi_block_realize(SCSIDevice *dev, Error **errp)
2611 {
2612     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2613     AioContext *ctx;
2614     int sg_version;
2615     int rc;
2616 
2617     if (!s->qdev.conf.blk) {
2618         error_setg(errp, "drive property not set");
2619         return;
2620     }
2621 
2622     if (s->rotation_rate) {
2623         error_report_once("rotation_rate is specified for scsi-block but is "
2624                           "not implemented. This option is deprecated and will "
2625                           "be removed in a future version");
2626     }
2627 
2628     ctx = blk_get_aio_context(s->qdev.conf.blk);
2629     aio_context_acquire(ctx);
2630 
2631     /* check we are using a driver managing SG_IO (version 3 and after) */
2632     rc = blk_ioctl(s->qdev.conf.blk, SG_GET_VERSION_NUM, &sg_version);
2633     if (rc < 0) {
2634         error_setg_errno(errp, -rc, "cannot get SG_IO version number");
2635         if (rc != -EPERM) {
2636             error_append_hint(errp, "Is this a SCSI device?\n");
2637         }
2638         goto out;
2639     }
2640     if (sg_version < 30000) {
2641         error_setg(errp, "scsi generic interface too old");
2642         goto out;
2643     }
2644 
2645     /* get device type from INQUIRY data */
2646     rc = get_device_type(s);
2647     if (rc < 0) {
2648         error_setg(errp, "INQUIRY failed");
2649         goto out;
2650     }
2651 
2652     /* Make a guess for the block size, we'll fix it when the guest sends.
2653      * READ CAPACITY.  If they don't, they likely would assume these sizes
2654      * anyway. (TODO: check in /sys).
2655      */
2656     if (s->qdev.type == TYPE_ROM || s->qdev.type == TYPE_WORM) {
2657         s->qdev.blocksize = 2048;
2658     } else {
2659         s->qdev.blocksize = 512;
2660     }
2661 
2662     /* Makes the scsi-block device not removable by using HMP and QMP eject
2663      * command.
2664      */
2665     s->features |= (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS);
2666 
2667     scsi_realize(&s->qdev, errp);
2668     scsi_generic_read_device_inquiry(&s->qdev);
2669 
2670 out:
2671     aio_context_release(ctx);
2672 }
2673 
2674 typedef struct SCSIBlockReq {
2675     SCSIDiskReq req;
2676     sg_io_hdr_t io_header;
2677 
2678     /* Selected bytes of the original CDB, copied into our own CDB.  */
2679     uint8_t cmd, cdb1, group_number;
2680 
2681     /* CDB passed to SG_IO.  */
2682     uint8_t cdb[16];
2683 } SCSIBlockReq;
2684 
2685 static BlockAIOCB *scsi_block_do_sgio(SCSIBlockReq *req,
2686                                       int64_t offset, QEMUIOVector *iov,
2687                                       int direction,
2688                                       BlockCompletionFunc *cb, void *opaque)
2689 {
2690     sg_io_hdr_t *io_header = &req->io_header;
2691     SCSIDiskReq *r = &req->req;
2692     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2693     int nb_logical_blocks;
2694     uint64_t lba;
2695     BlockAIOCB *aiocb;
2696 
2697     /* This is not supported yet.  It can only happen if the guest does
2698      * reads and writes that are not aligned to one logical sectors
2699      * _and_ cover multiple MemoryRegions.
2700      */
2701     assert(offset % s->qdev.blocksize == 0);
2702     assert(iov->size % s->qdev.blocksize == 0);
2703 
2704     io_header->interface_id = 'S';
2705 
2706     /* The data transfer comes from the QEMUIOVector.  */
2707     io_header->dxfer_direction = direction;
2708     io_header->dxfer_len = iov->size;
2709     io_header->dxferp = (void *)iov->iov;
2710     io_header->iovec_count = iov->niov;
2711     assert(io_header->iovec_count == iov->niov); /* no overflow! */
2712 
2713     /* Build a new CDB with the LBA and length patched in, in case
2714      * DMA helpers split the transfer in multiple segments.  Do not
2715      * build a CDB smaller than what the guest wanted, and only build
2716      * a larger one if strictly necessary.
2717      */
2718     io_header->cmdp = req->cdb;
2719     lba = offset / s->qdev.blocksize;
2720     nb_logical_blocks = io_header->dxfer_len / s->qdev.blocksize;
2721 
2722     if ((req->cmd >> 5) == 0 && lba <= 0x1ffff) {
2723         /* 6-byte CDB */
2724         stl_be_p(&req->cdb[0], lba | (req->cmd << 24));
2725         req->cdb[4] = nb_logical_blocks;
2726         req->cdb[5] = 0;
2727         io_header->cmd_len = 6;
2728     } else if ((req->cmd >> 5) <= 1 && lba <= 0xffffffffULL) {
2729         /* 10-byte CDB */
2730         req->cdb[0] = (req->cmd & 0x1f) | 0x20;
2731         req->cdb[1] = req->cdb1;
2732         stl_be_p(&req->cdb[2], lba);
2733         req->cdb[6] = req->group_number;
2734         stw_be_p(&req->cdb[7], nb_logical_blocks);
2735         req->cdb[9] = 0;
2736         io_header->cmd_len = 10;
2737     } else if ((req->cmd >> 5) != 4 && lba <= 0xffffffffULL) {
2738         /* 12-byte CDB */
2739         req->cdb[0] = (req->cmd & 0x1f) | 0xA0;
2740         req->cdb[1] = req->cdb1;
2741         stl_be_p(&req->cdb[2], lba);
2742         stl_be_p(&req->cdb[6], nb_logical_blocks);
2743         req->cdb[10] = req->group_number;
2744         req->cdb[11] = 0;
2745         io_header->cmd_len = 12;
2746     } else {
2747         /* 16-byte CDB */
2748         req->cdb[0] = (req->cmd & 0x1f) | 0x80;
2749         req->cdb[1] = req->cdb1;
2750         stq_be_p(&req->cdb[2], lba);
2751         stl_be_p(&req->cdb[10], nb_logical_blocks);
2752         req->cdb[14] = req->group_number;
2753         req->cdb[15] = 0;
2754         io_header->cmd_len = 16;
2755     }
2756 
2757     /* The rest is as in scsi-generic.c.  */
2758     io_header->mx_sb_len = sizeof(r->req.sense);
2759     io_header->sbp = r->req.sense;
2760     io_header->timeout = UINT_MAX;
2761     io_header->usr_ptr = r;
2762     io_header->flags |= SG_FLAG_DIRECT_IO;
2763 
2764     aiocb = blk_aio_ioctl(s->qdev.conf.blk, SG_IO, io_header, cb, opaque);
2765     assert(aiocb != NULL);
2766     return aiocb;
2767 }
2768 
2769 static bool scsi_block_no_fua(SCSICommand *cmd)
2770 {
2771     return false;
2772 }
2773 
2774 static BlockAIOCB *scsi_block_dma_readv(int64_t offset,
2775                                         QEMUIOVector *iov,
2776                                         BlockCompletionFunc *cb, void *cb_opaque,
2777                                         void *opaque)
2778 {
2779     SCSIBlockReq *r = opaque;
2780     return scsi_block_do_sgio(r, offset, iov,
2781                               SG_DXFER_FROM_DEV, cb, cb_opaque);
2782 }
2783 
2784 static BlockAIOCB *scsi_block_dma_writev(int64_t offset,
2785                                          QEMUIOVector *iov,
2786                                          BlockCompletionFunc *cb, void *cb_opaque,
2787                                          void *opaque)
2788 {
2789     SCSIBlockReq *r = opaque;
2790     return scsi_block_do_sgio(r, offset, iov,
2791                               SG_DXFER_TO_DEV, cb, cb_opaque);
2792 }
2793 
2794 static bool scsi_block_is_passthrough(SCSIDiskState *s, uint8_t *buf)
2795 {
2796     switch (buf[0]) {
2797     case VERIFY_10:
2798     case VERIFY_12:
2799     case VERIFY_16:
2800         /* Check if BYTCHK == 0x01 (data-out buffer contains data
2801          * for the number of logical blocks specified in the length
2802          * field).  For other modes, do not use scatter/gather operation.
2803          */
2804         if ((buf[1] & 6) == 2) {
2805             return false;
2806         }
2807         break;
2808 
2809     case READ_6:
2810     case READ_10:
2811     case READ_12:
2812     case READ_16:
2813     case WRITE_6:
2814     case WRITE_10:
2815     case WRITE_12:
2816     case WRITE_16:
2817     case WRITE_VERIFY_10:
2818     case WRITE_VERIFY_12:
2819     case WRITE_VERIFY_16:
2820         /* MMC writing cannot be done via DMA helpers, because it sometimes
2821          * involves writing beyond the maximum LBA or to negative LBA (lead-in).
2822          * We might use scsi_block_dma_reqops as long as no writing commands are
2823          * seen, but performance usually isn't paramount on optical media.  So,
2824          * just make scsi-block operate the same as scsi-generic for them.
2825          */
2826         if (s->qdev.type != TYPE_ROM) {
2827             return false;
2828         }
2829         break;
2830 
2831     default:
2832         break;
2833     }
2834 
2835     return true;
2836 }
2837 
2838 
2839 static int32_t scsi_block_dma_command(SCSIRequest *req, uint8_t *buf)
2840 {
2841     SCSIBlockReq *r = (SCSIBlockReq *)req;
2842     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2843 
2844     r->cmd = req->cmd.buf[0];
2845     switch (r->cmd >> 5) {
2846     case 0:
2847         /* 6-byte CDB.  */
2848         r->cdb1 = r->group_number = 0;
2849         break;
2850     case 1:
2851         /* 10-byte CDB.  */
2852         r->cdb1 = req->cmd.buf[1];
2853         r->group_number = req->cmd.buf[6];
2854         break;
2855     case 4:
2856         /* 12-byte CDB.  */
2857         r->cdb1 = req->cmd.buf[1];
2858         r->group_number = req->cmd.buf[10];
2859         break;
2860     case 5:
2861         /* 16-byte CDB.  */
2862         r->cdb1 = req->cmd.buf[1];
2863         r->group_number = req->cmd.buf[14];
2864         break;
2865     default:
2866         abort();
2867     }
2868 
2869     /* Protection information is not supported.  For SCSI versions 2 and
2870      * older (as determined by snooping the guest's INQUIRY commands),
2871      * there is no RD/WR/VRPROTECT, so skip this check in these versions.
2872      */
2873     if (s->qdev.scsi_version > 2 && (req->cmd.buf[1] & 0xe0)) {
2874         scsi_check_condition(&r->req, SENSE_CODE(INVALID_FIELD));
2875         return 0;
2876     }
2877 
2878     r->req.status = &r->io_header.status;
2879     return scsi_disk_dma_command(req, buf);
2880 }
2881 
2882 static const SCSIReqOps scsi_block_dma_reqops = {
2883     .size         = sizeof(SCSIBlockReq),
2884     .free_req     = scsi_free_request,
2885     .send_command = scsi_block_dma_command,
2886     .read_data    = scsi_read_data,
2887     .write_data   = scsi_write_data,
2888     .get_buf      = scsi_get_buf,
2889     .load_request = scsi_disk_load_request,
2890     .save_request = scsi_disk_save_request,
2891 };
2892 
2893 static SCSIRequest *scsi_block_new_request(SCSIDevice *d, uint32_t tag,
2894                                            uint32_t lun, uint8_t *buf,
2895                                            void *hba_private)
2896 {
2897     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2898 
2899     if (scsi_block_is_passthrough(s, buf)) {
2900         return scsi_req_alloc(&scsi_generic_req_ops, &s->qdev, tag, lun,
2901                               hba_private);
2902     } else {
2903         return scsi_req_alloc(&scsi_block_dma_reqops, &s->qdev, tag, lun,
2904                               hba_private);
2905     }
2906 }
2907 
2908 static int scsi_block_parse_cdb(SCSIDevice *d, SCSICommand *cmd,
2909                                   uint8_t *buf, void *hba_private)
2910 {
2911     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2912 
2913     if (scsi_block_is_passthrough(s, buf)) {
2914         return scsi_bus_parse_cdb(&s->qdev, cmd, buf, hba_private);
2915     } else {
2916         return scsi_req_parse_cdb(&s->qdev, cmd, buf);
2917     }
2918 }
2919 
2920 static void scsi_block_update_sense(SCSIRequest *req)
2921 {
2922     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
2923     SCSIBlockReq *br = DO_UPCAST(SCSIBlockReq, req, r);
2924     r->req.sense_len = MIN(br->io_header.sb_len_wr, sizeof(r->req.sense));
2925 }
2926 #endif
2927 
2928 static
2929 BlockAIOCB *scsi_dma_readv(int64_t offset, QEMUIOVector *iov,
2930                            BlockCompletionFunc *cb, void *cb_opaque,
2931                            void *opaque)
2932 {
2933     SCSIDiskReq *r = opaque;
2934     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2935     return blk_aio_preadv(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
2936 }
2937 
2938 static
2939 BlockAIOCB *scsi_dma_writev(int64_t offset, QEMUIOVector *iov,
2940                             BlockCompletionFunc *cb, void *cb_opaque,
2941                             void *opaque)
2942 {
2943     SCSIDiskReq *r = opaque;
2944     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2945     return blk_aio_pwritev(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
2946 }
2947 
2948 static void scsi_disk_base_class_initfn(ObjectClass *klass, void *data)
2949 {
2950     DeviceClass *dc = DEVICE_CLASS(klass);
2951     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
2952 
2953     dc->fw_name = "disk";
2954     dc->reset = scsi_disk_reset;
2955     sdc->dma_readv = scsi_dma_readv;
2956     sdc->dma_writev = scsi_dma_writev;
2957     sdc->need_fua_emulation = scsi_is_cmd_fua;
2958 }
2959 
2960 static const TypeInfo scsi_disk_base_info = {
2961     .name          = TYPE_SCSI_DISK_BASE,
2962     .parent        = TYPE_SCSI_DEVICE,
2963     .class_init    = scsi_disk_base_class_initfn,
2964     .instance_size = sizeof(SCSIDiskState),
2965     .class_size    = sizeof(SCSIDiskClass),
2966     .abstract      = true,
2967 };
2968 
2969 #define DEFINE_SCSI_DISK_PROPERTIES()                                   \
2970     DEFINE_PROP_DRIVE_IOTHREAD("drive", SCSIDiskState, qdev.conf.blk),  \
2971     DEFINE_BLOCK_PROPERTIES_BASE(SCSIDiskState, qdev.conf),             \
2972     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),            \
2973     DEFINE_PROP_STRING("ver", SCSIDiskState, version),                  \
2974     DEFINE_PROP_STRING("serial", SCSIDiskState, serial),                \
2975     DEFINE_PROP_STRING("vendor", SCSIDiskState, vendor),                \
2976     DEFINE_PROP_STRING("product", SCSIDiskState, product),              \
2977     DEFINE_PROP_STRING("device_id", SCSIDiskState, device_id)
2978 
2979 
2980 static Property scsi_hd_properties[] = {
2981     DEFINE_SCSI_DISK_PROPERTIES(),
2982     DEFINE_PROP_BIT("removable", SCSIDiskState, features,
2983                     SCSI_DISK_F_REMOVABLE, false),
2984     DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
2985                     SCSI_DISK_F_DPOFUA, false),
2986     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
2987     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
2988     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
2989     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
2990                        DEFAULT_MAX_UNMAP_SIZE),
2991     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
2992                        DEFAULT_MAX_IO_SIZE),
2993     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
2994     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
2995                       5),
2996     DEFINE_BLOCK_CHS_PROPERTIES(SCSIDiskState, qdev.conf),
2997     DEFINE_PROP_END_OF_LIST(),
2998 };
2999 
3000 static const VMStateDescription vmstate_scsi_disk_state = {
3001     .name = "scsi-disk",
3002     .version_id = 1,
3003     .minimum_version_id = 1,
3004     .fields = (VMStateField[]) {
3005         VMSTATE_SCSI_DEVICE(qdev, SCSIDiskState),
3006         VMSTATE_BOOL(media_changed, SCSIDiskState),
3007         VMSTATE_BOOL(media_event, SCSIDiskState),
3008         VMSTATE_BOOL(eject_request, SCSIDiskState),
3009         VMSTATE_BOOL(tray_open, SCSIDiskState),
3010         VMSTATE_BOOL(tray_locked, SCSIDiskState),
3011         VMSTATE_END_OF_LIST()
3012     }
3013 };
3014 
3015 static void scsi_hd_class_initfn(ObjectClass *klass, void *data)
3016 {
3017     DeviceClass *dc = DEVICE_CLASS(klass);
3018     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3019 
3020     sc->realize      = scsi_hd_realize;
3021     sc->alloc_req    = scsi_new_request;
3022     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3023     dc->desc = "virtual SCSI disk";
3024     dc->props = scsi_hd_properties;
3025     dc->vmsd  = &vmstate_scsi_disk_state;
3026 }
3027 
3028 static const TypeInfo scsi_hd_info = {
3029     .name          = "scsi-hd",
3030     .parent        = TYPE_SCSI_DISK_BASE,
3031     .class_init    = scsi_hd_class_initfn,
3032 };
3033 
3034 static Property scsi_cd_properties[] = {
3035     DEFINE_SCSI_DISK_PROPERTIES(),
3036     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3037     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3038     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3039     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3040                        DEFAULT_MAX_IO_SIZE),
3041     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3042                       5),
3043     DEFINE_PROP_END_OF_LIST(),
3044 };
3045 
3046 static void scsi_cd_class_initfn(ObjectClass *klass, void *data)
3047 {
3048     DeviceClass *dc = DEVICE_CLASS(klass);
3049     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3050 
3051     sc->realize      = scsi_cd_realize;
3052     sc->alloc_req    = scsi_new_request;
3053     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3054     dc->desc = "virtual SCSI CD-ROM";
3055     dc->props = scsi_cd_properties;
3056     dc->vmsd  = &vmstate_scsi_disk_state;
3057 }
3058 
3059 static const TypeInfo scsi_cd_info = {
3060     .name          = "scsi-cd",
3061     .parent        = TYPE_SCSI_DISK_BASE,
3062     .class_init    = scsi_cd_class_initfn,
3063 };
3064 
3065 #ifdef __linux__
3066 static Property scsi_block_properties[] = {
3067     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),         \
3068     DEFINE_PROP_DRIVE("drive", SCSIDiskState, qdev.conf.blk),
3069     DEFINE_PROP_BOOL("share-rw", SCSIDiskState, qdev.conf.share_rw, false),
3070     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3071     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3072                        DEFAULT_MAX_UNMAP_SIZE),
3073     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3074                        DEFAULT_MAX_IO_SIZE),
3075     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3076                       -1),
3077     DEFINE_PROP_END_OF_LIST(),
3078 };
3079 
3080 static void scsi_block_class_initfn(ObjectClass *klass, void *data)
3081 {
3082     DeviceClass *dc = DEVICE_CLASS(klass);
3083     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3084     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
3085 
3086     sc->realize      = scsi_block_realize;
3087     sc->alloc_req    = scsi_block_new_request;
3088     sc->parse_cdb    = scsi_block_parse_cdb;
3089     sdc->dma_readv   = scsi_block_dma_readv;
3090     sdc->dma_writev  = scsi_block_dma_writev;
3091     sdc->update_sense = scsi_block_update_sense;
3092     sdc->need_fua_emulation = scsi_block_no_fua;
3093     dc->desc = "SCSI block device passthrough";
3094     dc->props = scsi_block_properties;
3095     dc->vmsd  = &vmstate_scsi_disk_state;
3096 }
3097 
3098 static const TypeInfo scsi_block_info = {
3099     .name          = "scsi-block",
3100     .parent        = TYPE_SCSI_DISK_BASE,
3101     .class_init    = scsi_block_class_initfn,
3102 };
3103 #endif
3104 
3105 static Property scsi_disk_properties[] = {
3106     DEFINE_SCSI_DISK_PROPERTIES(),
3107     DEFINE_PROP_BIT("removable", SCSIDiskState, features,
3108                     SCSI_DISK_F_REMOVABLE, false),
3109     DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
3110                     SCSI_DISK_F_DPOFUA, false),
3111     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3112     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3113     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3114     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3115                        DEFAULT_MAX_UNMAP_SIZE),
3116     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3117                        DEFAULT_MAX_IO_SIZE),
3118     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3119                       5),
3120     DEFINE_PROP_END_OF_LIST(),
3121 };
3122 
3123 static void scsi_disk_class_initfn(ObjectClass *klass, void *data)
3124 {
3125     DeviceClass *dc = DEVICE_CLASS(klass);
3126     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3127 
3128     sc->realize      = scsi_disk_realize;
3129     sc->alloc_req    = scsi_new_request;
3130     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3131     dc->fw_name = "disk";
3132     dc->desc = "virtual SCSI disk or CD-ROM (legacy)";
3133     dc->reset = scsi_disk_reset;
3134     dc->props = scsi_disk_properties;
3135     dc->vmsd  = &vmstate_scsi_disk_state;
3136 }
3137 
3138 static const TypeInfo scsi_disk_info = {
3139     .name          = "scsi-disk",
3140     .parent        = TYPE_SCSI_DISK_BASE,
3141     .class_init    = scsi_disk_class_initfn,
3142 };
3143 
3144 static void scsi_disk_register_types(void)
3145 {
3146     type_register_static(&scsi_disk_base_info);
3147     type_register_static(&scsi_hd_info);
3148     type_register_static(&scsi_cd_info);
3149 #ifdef __linux__
3150     type_register_static(&scsi_block_info);
3151 #endif
3152     type_register_static(&scsi_disk_info);
3153 }
3154 
3155 type_init(scsi_disk_register_types)
3156