xref: /openbmc/qemu/hw/scsi/scsi-disk.c (revision effd60c8)
1 /*
2  * SCSI Device emulation
3  *
4  * Copyright (c) 2006 CodeSourcery.
5  * Based on code by Fabrice Bellard
6  *
7  * Written by Paul Brook
8  * Modifications:
9  *  2009-Dec-12 Artyom Tarasenko : implemented stamdard inquiry for the case
10  *                                 when the allocation length of CDB is smaller
11  *                                 than 36.
12  *  2009-Oct-13 Artyom Tarasenko : implemented the block descriptor in the
13  *                                 MODE SENSE response.
14  *
15  * This code is licensed under the LGPL.
16  *
17  * Note that this file only handles the SCSI architecture model and device
18  * commands.  Emulation of interface/link layer protocols is handled by
19  * the host adapter emulator.
20  */
21 
22 #include "qemu/osdep.h"
23 #include "qemu/units.h"
24 #include "qapi/error.h"
25 #include "qemu/error-report.h"
26 #include "qemu/main-loop.h"
27 #include "qemu/module.h"
28 #include "qemu/hw-version.h"
29 #include "qemu/memalign.h"
30 #include "hw/scsi/scsi.h"
31 #include "migration/qemu-file-types.h"
32 #include "migration/vmstate.h"
33 #include "hw/scsi/emulation.h"
34 #include "scsi/constants.h"
35 #include "sysemu/block-backend.h"
36 #include "sysemu/blockdev.h"
37 #include "hw/block/block.h"
38 #include "hw/qdev-properties.h"
39 #include "hw/qdev-properties-system.h"
40 #include "sysemu/dma.h"
41 #include "sysemu/sysemu.h"
42 #include "qemu/cutils.h"
43 #include "trace.h"
44 #include "qom/object.h"
45 
46 #ifdef __linux
47 #include <scsi/sg.h>
48 #endif
49 
50 #define SCSI_WRITE_SAME_MAX         (512 * KiB)
51 #define SCSI_DMA_BUF_SIZE           (128 * KiB)
52 #define SCSI_MAX_INQUIRY_LEN        256
53 #define SCSI_MAX_MODE_LEN           256
54 
55 #define DEFAULT_DISCARD_GRANULARITY (4 * KiB)
56 #define DEFAULT_MAX_UNMAP_SIZE      (1 * GiB)
57 #define DEFAULT_MAX_IO_SIZE         INT_MAX     /* 2 GB - 1 block */
58 
59 #define TYPE_SCSI_DISK_BASE         "scsi-disk-base"
60 
61 OBJECT_DECLARE_TYPE(SCSIDiskState, SCSIDiskClass, SCSI_DISK_BASE)
62 
63 struct SCSIDiskClass {
64     SCSIDeviceClass parent_class;
65     DMAIOFunc       *dma_readv;
66     DMAIOFunc       *dma_writev;
67     bool            (*need_fua_emulation)(SCSICommand *cmd);
68     void            (*update_sense)(SCSIRequest *r);
69 };
70 
71 typedef struct SCSIDiskReq {
72     SCSIRequest req;
73     /* Both sector and sector_count are in terms of BDRV_SECTOR_SIZE bytes.  */
74     uint64_t sector;
75     uint32_t sector_count;
76     uint32_t buflen;
77     bool started;
78     bool need_fua_emulation;
79     struct iovec iov;
80     QEMUIOVector qiov;
81     BlockAcctCookie acct;
82 } SCSIDiskReq;
83 
84 #define SCSI_DISK_F_REMOVABLE             0
85 #define SCSI_DISK_F_DPOFUA                1
86 #define SCSI_DISK_F_NO_REMOVABLE_DEVOPS   2
87 
88 struct SCSIDiskState {
89     SCSIDevice qdev;
90     uint32_t features;
91     bool media_changed;
92     bool media_event;
93     bool eject_request;
94     uint16_t port_index;
95     uint64_t max_unmap_size;
96     uint64_t max_io_size;
97     uint32_t quirks;
98     QEMUBH *bh;
99     char *version;
100     char *serial;
101     char *vendor;
102     char *product;
103     char *device_id;
104     bool tray_open;
105     bool tray_locked;
106     /*
107      * 0x0000        - rotation rate not reported
108      * 0x0001        - non-rotating medium (SSD)
109      * 0x0002-0x0400 - reserved
110      * 0x0401-0xffe  - rotations per minute
111      * 0xffff        - reserved
112      */
113     uint16_t rotation_rate;
114 };
115 
116 static void scsi_free_request(SCSIRequest *req)
117 {
118     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
119 
120     qemu_vfree(r->iov.iov_base);
121 }
122 
123 /* Helper function for command completion with sense.  */
124 static void scsi_check_condition(SCSIDiskReq *r, SCSISense sense)
125 {
126     trace_scsi_disk_check_condition(r->req.tag, sense.key, sense.asc,
127                                     sense.ascq);
128     scsi_req_build_sense(&r->req, sense);
129     scsi_req_complete(&r->req, CHECK_CONDITION);
130 }
131 
132 static void scsi_init_iovec(SCSIDiskReq *r, size_t size)
133 {
134     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
135 
136     if (!r->iov.iov_base) {
137         r->buflen = size;
138         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
139     }
140     r->iov.iov_len = MIN(r->sector_count * BDRV_SECTOR_SIZE, r->buflen);
141     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
142 }
143 
144 static void scsi_disk_save_request(QEMUFile *f, SCSIRequest *req)
145 {
146     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
147 
148     qemu_put_be64s(f, &r->sector);
149     qemu_put_be32s(f, &r->sector_count);
150     qemu_put_be32s(f, &r->buflen);
151     if (r->buflen) {
152         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
153             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
154         } else if (!req->retry) {
155             uint32_t len = r->iov.iov_len;
156             qemu_put_be32s(f, &len);
157             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
158         }
159     }
160 }
161 
162 static void scsi_disk_load_request(QEMUFile *f, SCSIRequest *req)
163 {
164     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
165 
166     qemu_get_be64s(f, &r->sector);
167     qemu_get_be32s(f, &r->sector_count);
168     qemu_get_be32s(f, &r->buflen);
169     if (r->buflen) {
170         scsi_init_iovec(r, r->buflen);
171         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
172             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
173         } else if (!r->req.retry) {
174             uint32_t len;
175             qemu_get_be32s(f, &len);
176             r->iov.iov_len = len;
177             assert(r->iov.iov_len <= r->buflen);
178             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
179         }
180     }
181 
182     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
183 }
184 
185 /*
186  * scsi_handle_rw_error has two return values.  False means that the error
187  * must be ignored, true means that the error has been processed and the
188  * caller should not do anything else for this request.  Note that
189  * scsi_handle_rw_error always manages its reference counts, independent
190  * of the return value.
191  */
192 static bool scsi_handle_rw_error(SCSIDiskReq *r, int ret, bool acct_failed)
193 {
194     bool is_read = (r->req.cmd.mode == SCSI_XFER_FROM_DEV);
195     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
196     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
197     SCSISense sense = SENSE_CODE(NO_SENSE);
198     int error = 0;
199     bool req_has_sense = false;
200     BlockErrorAction action;
201     int status;
202 
203     if (ret < 0) {
204         status = scsi_sense_from_errno(-ret, &sense);
205         error = -ret;
206     } else {
207         /* A passthrough command has completed with nonzero status.  */
208         status = ret;
209         if (status == CHECK_CONDITION) {
210             req_has_sense = true;
211             error = scsi_sense_buf_to_errno(r->req.sense, sizeof(r->req.sense));
212         } else {
213             error = EINVAL;
214         }
215     }
216 
217     /*
218      * Check whether the error has to be handled by the guest or should
219      * rather follow the rerror=/werror= settings.  Guest-handled errors
220      * are usually retried immediately, so do not post them to QMP and
221      * do not account them as failed I/O.
222      */
223     if (req_has_sense &&
224         scsi_sense_buf_is_guest_recoverable(r->req.sense, sizeof(r->req.sense))) {
225         action = BLOCK_ERROR_ACTION_REPORT;
226         acct_failed = false;
227     } else {
228         action = blk_get_error_action(s->qdev.conf.blk, is_read, error);
229         blk_error_action(s->qdev.conf.blk, action, is_read, error);
230     }
231 
232     switch (action) {
233     case BLOCK_ERROR_ACTION_REPORT:
234         if (acct_failed) {
235             block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
236         }
237         if (req_has_sense) {
238             sdc->update_sense(&r->req);
239         } else if (status == CHECK_CONDITION) {
240             scsi_req_build_sense(&r->req, sense);
241         }
242         scsi_req_complete(&r->req, status);
243         return true;
244 
245     case BLOCK_ERROR_ACTION_IGNORE:
246         return false;
247 
248     case BLOCK_ERROR_ACTION_STOP:
249         scsi_req_retry(&r->req);
250         return true;
251 
252     default:
253         g_assert_not_reached();
254     }
255 }
256 
257 static bool scsi_disk_req_check_error(SCSIDiskReq *r, int ret, bool acct_failed)
258 {
259     if (r->req.io_canceled) {
260         scsi_req_cancel_complete(&r->req);
261         return true;
262     }
263 
264     if (ret < 0) {
265         return scsi_handle_rw_error(r, ret, acct_failed);
266     }
267 
268     return false;
269 }
270 
271 static void scsi_aio_complete(void *opaque, int ret)
272 {
273     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
274     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
275 
276     /* The request must only run in the BlockBackend's AioContext */
277     assert(blk_get_aio_context(s->qdev.conf.blk) ==
278            qemu_get_current_aio_context());
279 
280     assert(r->req.aiocb != NULL);
281     r->req.aiocb = NULL;
282 
283     if (scsi_disk_req_check_error(r, ret, true)) {
284         goto done;
285     }
286 
287     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
288     scsi_req_complete(&r->req, GOOD);
289 
290 done:
291     scsi_req_unref(&r->req);
292 }
293 
294 static bool scsi_is_cmd_fua(SCSICommand *cmd)
295 {
296     switch (cmd->buf[0]) {
297     case READ_10:
298     case READ_12:
299     case READ_16:
300     case WRITE_10:
301     case WRITE_12:
302     case WRITE_16:
303         return (cmd->buf[1] & 8) != 0;
304 
305     case VERIFY_10:
306     case VERIFY_12:
307     case VERIFY_16:
308     case WRITE_VERIFY_10:
309     case WRITE_VERIFY_12:
310     case WRITE_VERIFY_16:
311         return true;
312 
313     case READ_6:
314     case WRITE_6:
315     default:
316         return false;
317     }
318 }
319 
320 static void scsi_write_do_fua(SCSIDiskReq *r)
321 {
322     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
323 
324     assert(r->req.aiocb == NULL);
325     assert(!r->req.io_canceled);
326 
327     if (r->need_fua_emulation) {
328         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
329                          BLOCK_ACCT_FLUSH);
330         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
331         return;
332     }
333 
334     scsi_req_complete(&r->req, GOOD);
335     scsi_req_unref(&r->req);
336 }
337 
338 static void scsi_dma_complete_noio(SCSIDiskReq *r, int ret)
339 {
340     assert(r->req.aiocb == NULL);
341     if (scsi_disk_req_check_error(r, ret, false)) {
342         goto done;
343     }
344 
345     r->sector += r->sector_count;
346     r->sector_count = 0;
347     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
348         scsi_write_do_fua(r);
349         return;
350     } else {
351         scsi_req_complete(&r->req, GOOD);
352     }
353 
354 done:
355     scsi_req_unref(&r->req);
356 }
357 
358 static void scsi_dma_complete(void *opaque, int ret)
359 {
360     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
361     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
362 
363     assert(r->req.aiocb != NULL);
364     r->req.aiocb = NULL;
365 
366     if (ret < 0) {
367         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
368     } else {
369         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
370     }
371     scsi_dma_complete_noio(r, ret);
372 }
373 
374 static void scsi_read_complete_noio(SCSIDiskReq *r, int ret)
375 {
376     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
377     uint32_t n;
378 
379     /* The request must only run in the BlockBackend's AioContext */
380     assert(blk_get_aio_context(s->qdev.conf.blk) ==
381            qemu_get_current_aio_context());
382 
383     assert(r->req.aiocb == NULL);
384     if (scsi_disk_req_check_error(r, ret, false)) {
385         goto done;
386     }
387 
388     n = r->qiov.size / BDRV_SECTOR_SIZE;
389     r->sector += n;
390     r->sector_count -= n;
391     scsi_req_data(&r->req, r->qiov.size);
392 
393 done:
394     scsi_req_unref(&r->req);
395 }
396 
397 static void scsi_read_complete(void *opaque, int ret)
398 {
399     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
400     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
401 
402     assert(r->req.aiocb != NULL);
403     r->req.aiocb = NULL;
404 
405     if (ret < 0) {
406         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
407     } else {
408         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
409         trace_scsi_disk_read_complete(r->req.tag, r->qiov.size);
410     }
411     scsi_read_complete_noio(r, ret);
412 }
413 
414 /* Actually issue a read to the block device.  */
415 static void scsi_do_read(SCSIDiskReq *r, int ret)
416 {
417     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
418     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
419 
420     assert (r->req.aiocb == NULL);
421     if (scsi_disk_req_check_error(r, ret, false)) {
422         goto done;
423     }
424 
425     /* The request is used as the AIO opaque value, so add a ref.  */
426     scsi_req_ref(&r->req);
427 
428     if (r->req.sg) {
429         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_READ);
430         r->req.residual -= r->req.sg->size;
431         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
432                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
433                                   BDRV_SECTOR_SIZE,
434                                   sdc->dma_readv, r, scsi_dma_complete, r,
435                                   DMA_DIRECTION_FROM_DEVICE);
436     } else {
437         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
438         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
439                          r->qiov.size, BLOCK_ACCT_READ);
440         r->req.aiocb = sdc->dma_readv(r->sector << BDRV_SECTOR_BITS, &r->qiov,
441                                       scsi_read_complete, r, r);
442     }
443 
444 done:
445     scsi_req_unref(&r->req);
446 }
447 
448 static void scsi_do_read_cb(void *opaque, int ret)
449 {
450     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
451     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
452 
453     assert (r->req.aiocb != NULL);
454     r->req.aiocb = NULL;
455 
456     if (ret < 0) {
457         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
458     } else {
459         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
460     }
461     scsi_do_read(opaque, ret);
462 }
463 
464 /* Read more data from scsi device into buffer.  */
465 static void scsi_read_data(SCSIRequest *req)
466 {
467     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
468     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
469     bool first;
470 
471     trace_scsi_disk_read_data_count(r->sector_count);
472     if (r->sector_count == 0) {
473         /* This also clears the sense buffer for REQUEST SENSE.  */
474         scsi_req_complete(&r->req, GOOD);
475         return;
476     }
477 
478     /* No data transfer may already be in progress */
479     assert(r->req.aiocb == NULL);
480 
481     /* The request is used as the AIO opaque value, so add a ref.  */
482     scsi_req_ref(&r->req);
483     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
484         trace_scsi_disk_read_data_invalid();
485         scsi_read_complete_noio(r, -EINVAL);
486         return;
487     }
488 
489     if (!blk_is_available(req->dev->conf.blk)) {
490         scsi_read_complete_noio(r, -ENOMEDIUM);
491         return;
492     }
493 
494     first = !r->started;
495     r->started = true;
496     if (first && r->need_fua_emulation) {
497         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
498                          BLOCK_ACCT_FLUSH);
499         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_do_read_cb, r);
500     } else {
501         scsi_do_read(r, 0);
502     }
503 }
504 
505 static void scsi_write_complete_noio(SCSIDiskReq *r, int ret)
506 {
507     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
508     uint32_t n;
509 
510     /* The request must only run in the BlockBackend's AioContext */
511     assert(blk_get_aio_context(s->qdev.conf.blk) ==
512            qemu_get_current_aio_context());
513 
514     assert (r->req.aiocb == NULL);
515     if (scsi_disk_req_check_error(r, ret, false)) {
516         goto done;
517     }
518 
519     n = r->qiov.size / BDRV_SECTOR_SIZE;
520     r->sector += n;
521     r->sector_count -= n;
522     if (r->sector_count == 0) {
523         scsi_write_do_fua(r);
524         return;
525     } else {
526         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
527         trace_scsi_disk_write_complete_noio(r->req.tag, r->qiov.size);
528         scsi_req_data(&r->req, r->qiov.size);
529     }
530 
531 done:
532     scsi_req_unref(&r->req);
533 }
534 
535 static void scsi_write_complete(void * opaque, int ret)
536 {
537     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
538     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
539 
540     assert (r->req.aiocb != NULL);
541     r->req.aiocb = NULL;
542 
543     if (ret < 0) {
544         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
545     } else {
546         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
547     }
548     scsi_write_complete_noio(r, ret);
549 }
550 
551 static void scsi_write_data(SCSIRequest *req)
552 {
553     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
554     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
555     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
556 
557     /* No data transfer may already be in progress */
558     assert(r->req.aiocb == NULL);
559 
560     /* The request is used as the AIO opaque value, so add a ref.  */
561     scsi_req_ref(&r->req);
562     if (r->req.cmd.mode != SCSI_XFER_TO_DEV) {
563         trace_scsi_disk_write_data_invalid();
564         scsi_write_complete_noio(r, -EINVAL);
565         return;
566     }
567 
568     if (!r->req.sg && !r->qiov.size) {
569         /* Called for the first time.  Ask the driver to send us more data.  */
570         r->started = true;
571         scsi_write_complete_noio(r, 0);
572         return;
573     }
574     if (!blk_is_available(req->dev->conf.blk)) {
575         scsi_write_complete_noio(r, -ENOMEDIUM);
576         return;
577     }
578 
579     if (r->req.cmd.buf[0] == VERIFY_10 || r->req.cmd.buf[0] == VERIFY_12 ||
580         r->req.cmd.buf[0] == VERIFY_16) {
581         if (r->req.sg) {
582             scsi_dma_complete_noio(r, 0);
583         } else {
584             scsi_write_complete_noio(r, 0);
585         }
586         return;
587     }
588 
589     if (r->req.sg) {
590         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_WRITE);
591         r->req.residual -= r->req.sg->size;
592         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
593                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
594                                   BDRV_SECTOR_SIZE,
595                                   sdc->dma_writev, r, scsi_dma_complete, r,
596                                   DMA_DIRECTION_TO_DEVICE);
597     } else {
598         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
599                          r->qiov.size, BLOCK_ACCT_WRITE);
600         r->req.aiocb = sdc->dma_writev(r->sector << BDRV_SECTOR_BITS, &r->qiov,
601                                        scsi_write_complete, r, r);
602     }
603 }
604 
605 /* Return a pointer to the data buffer.  */
606 static uint8_t *scsi_get_buf(SCSIRequest *req)
607 {
608     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
609 
610     return (uint8_t *)r->iov.iov_base;
611 }
612 
613 static int scsi_disk_emulate_vpd_page(SCSIRequest *req, uint8_t *outbuf)
614 {
615     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
616     uint8_t page_code = req->cmd.buf[2];
617     int start, buflen = 0;
618 
619     outbuf[buflen++] = s->qdev.type & 0x1f;
620     outbuf[buflen++] = page_code;
621     outbuf[buflen++] = 0x00;
622     outbuf[buflen++] = 0x00;
623     start = buflen;
624 
625     switch (page_code) {
626     case 0x00: /* Supported page codes, mandatory */
627     {
628         trace_scsi_disk_emulate_vpd_page_00(req->cmd.xfer);
629         outbuf[buflen++] = 0x00; /* list of supported pages (this page) */
630         if (s->serial) {
631             outbuf[buflen++] = 0x80; /* unit serial number */
632         }
633         outbuf[buflen++] = 0x83; /* device identification */
634         if (s->qdev.type == TYPE_DISK) {
635             outbuf[buflen++] = 0xb0; /* block limits */
636             outbuf[buflen++] = 0xb1; /* block device characteristics */
637             outbuf[buflen++] = 0xb2; /* thin provisioning */
638         }
639         break;
640     }
641     case 0x80: /* Device serial number, optional */
642     {
643         int l;
644 
645         if (!s->serial) {
646             trace_scsi_disk_emulate_vpd_page_80_not_supported();
647             return -1;
648         }
649 
650         l = strlen(s->serial);
651         if (l > 36) {
652             l = 36;
653         }
654 
655         trace_scsi_disk_emulate_vpd_page_80(req->cmd.xfer);
656         memcpy(outbuf + buflen, s->serial, l);
657         buflen += l;
658         break;
659     }
660 
661     case 0x83: /* Device identification page, mandatory */
662     {
663         int id_len = s->device_id ? MIN(strlen(s->device_id), 255 - 8) : 0;
664 
665         trace_scsi_disk_emulate_vpd_page_83(req->cmd.xfer);
666 
667         if (id_len) {
668             outbuf[buflen++] = 0x2; /* ASCII */
669             outbuf[buflen++] = 0;   /* not officially assigned */
670             outbuf[buflen++] = 0;   /* reserved */
671             outbuf[buflen++] = id_len; /* length of data following */
672             memcpy(outbuf + buflen, s->device_id, id_len);
673             buflen += id_len;
674         }
675 
676         if (s->qdev.wwn) {
677             outbuf[buflen++] = 0x1; /* Binary */
678             outbuf[buflen++] = 0x3; /* NAA */
679             outbuf[buflen++] = 0;   /* reserved */
680             outbuf[buflen++] = 8;
681             stq_be_p(&outbuf[buflen], s->qdev.wwn);
682             buflen += 8;
683         }
684 
685         if (s->qdev.port_wwn) {
686             outbuf[buflen++] = 0x61; /* SAS / Binary */
687             outbuf[buflen++] = 0x93; /* PIV / Target port / NAA */
688             outbuf[buflen++] = 0;    /* reserved */
689             outbuf[buflen++] = 8;
690             stq_be_p(&outbuf[buflen], s->qdev.port_wwn);
691             buflen += 8;
692         }
693 
694         if (s->port_index) {
695             outbuf[buflen++] = 0x61; /* SAS / Binary */
696 
697             /* PIV/Target port/relative target port */
698             outbuf[buflen++] = 0x94;
699 
700             outbuf[buflen++] = 0;    /* reserved */
701             outbuf[buflen++] = 4;
702             stw_be_p(&outbuf[buflen + 2], s->port_index);
703             buflen += 4;
704         }
705         break;
706     }
707     case 0xb0: /* block limits */
708     {
709         SCSIBlockLimits bl = {};
710 
711         if (s->qdev.type == TYPE_ROM) {
712             trace_scsi_disk_emulate_vpd_page_b0_not_supported();
713             return -1;
714         }
715         bl.wsnz = 1;
716         bl.unmap_sectors =
717             s->qdev.conf.discard_granularity / s->qdev.blocksize;
718         bl.min_io_size =
719             s->qdev.conf.min_io_size / s->qdev.blocksize;
720         bl.opt_io_size =
721             s->qdev.conf.opt_io_size / s->qdev.blocksize;
722         bl.max_unmap_sectors =
723             s->max_unmap_size / s->qdev.blocksize;
724         bl.max_io_sectors =
725             s->max_io_size / s->qdev.blocksize;
726         /* 255 descriptors fit in 4 KiB with an 8-byte header */
727         bl.max_unmap_descr = 255;
728 
729         if (s->qdev.type == TYPE_DISK) {
730             int max_transfer_blk = blk_get_max_transfer(s->qdev.conf.blk);
731             int max_io_sectors_blk =
732                 max_transfer_blk / s->qdev.blocksize;
733 
734             bl.max_io_sectors =
735                 MIN_NON_ZERO(max_io_sectors_blk, bl.max_io_sectors);
736         }
737         buflen += scsi_emulate_block_limits(outbuf + buflen, &bl);
738         break;
739     }
740     case 0xb1: /* block device characteristics */
741     {
742         buflen = 0x40;
743         outbuf[4] = (s->rotation_rate >> 8) & 0xff;
744         outbuf[5] = s->rotation_rate & 0xff;
745         outbuf[6] = 0; /* PRODUCT TYPE */
746         outbuf[7] = 0; /* WABEREQ | WACEREQ | NOMINAL FORM FACTOR */
747         outbuf[8] = 0; /* VBULS */
748         break;
749     }
750     case 0xb2: /* thin provisioning */
751     {
752         buflen = 8;
753         outbuf[4] = 0;
754         outbuf[5] = 0xe0; /* unmap & write_same 10/16 all supported */
755         outbuf[6] = s->qdev.conf.discard_granularity ? 2 : 1;
756         outbuf[7] = 0;
757         break;
758     }
759     default:
760         return -1;
761     }
762     /* done with EVPD */
763     assert(buflen - start <= 255);
764     outbuf[start - 1] = buflen - start;
765     return buflen;
766 }
767 
768 static int scsi_disk_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf)
769 {
770     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
771     int buflen = 0;
772 
773     if (req->cmd.buf[1] & 0x1) {
774         /* Vital product data */
775         return scsi_disk_emulate_vpd_page(req, outbuf);
776     }
777 
778     /* Standard INQUIRY data */
779     if (req->cmd.buf[2] != 0) {
780         return -1;
781     }
782 
783     /* PAGE CODE == 0 */
784     buflen = req->cmd.xfer;
785     if (buflen > SCSI_MAX_INQUIRY_LEN) {
786         buflen = SCSI_MAX_INQUIRY_LEN;
787     }
788 
789     outbuf[0] = s->qdev.type & 0x1f;
790     outbuf[1] = (s->features & (1 << SCSI_DISK_F_REMOVABLE)) ? 0x80 : 0;
791 
792     strpadcpy((char *) &outbuf[16], 16, s->product, ' ');
793     strpadcpy((char *) &outbuf[8], 8, s->vendor, ' ');
794 
795     memset(&outbuf[32], 0, 4);
796     memcpy(&outbuf[32], s->version, MIN(4, strlen(s->version)));
797     /*
798      * We claim conformance to SPC-3, which is required for guests
799      * to ask for modern features like READ CAPACITY(16) or the
800      * block characteristics VPD page by default.  Not all of SPC-3
801      * is actually implemented, but we're good enough.
802      */
803     outbuf[2] = s->qdev.default_scsi_version;
804     outbuf[3] = 2 | 0x10; /* Format 2, HiSup */
805 
806     if (buflen > 36) {
807         outbuf[4] = buflen - 5; /* Additional Length = (Len - 1) - 4 */
808     } else {
809         /* If the allocation length of CDB is too small,
810                the additional length is not adjusted */
811         outbuf[4] = 36 - 5;
812     }
813 
814     /* Sync data transfer and TCQ.  */
815     outbuf[7] = 0x10 | (req->bus->info->tcq ? 0x02 : 0);
816     return buflen;
817 }
818 
819 static inline bool media_is_dvd(SCSIDiskState *s)
820 {
821     uint64_t nb_sectors;
822     if (s->qdev.type != TYPE_ROM) {
823         return false;
824     }
825     if (!blk_is_available(s->qdev.conf.blk)) {
826         return false;
827     }
828     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
829     return nb_sectors > CD_MAX_SECTORS;
830 }
831 
832 static inline bool media_is_cd(SCSIDiskState *s)
833 {
834     uint64_t nb_sectors;
835     if (s->qdev.type != TYPE_ROM) {
836         return false;
837     }
838     if (!blk_is_available(s->qdev.conf.blk)) {
839         return false;
840     }
841     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
842     return nb_sectors <= CD_MAX_SECTORS;
843 }
844 
845 static int scsi_read_disc_information(SCSIDiskState *s, SCSIDiskReq *r,
846                                       uint8_t *outbuf)
847 {
848     uint8_t type = r->req.cmd.buf[1] & 7;
849 
850     if (s->qdev.type != TYPE_ROM) {
851         return -1;
852     }
853 
854     /* Types 1/2 are only defined for Blu-Ray.  */
855     if (type != 0) {
856         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
857         return -1;
858     }
859 
860     memset(outbuf, 0, 34);
861     outbuf[1] = 32;
862     outbuf[2] = 0xe; /* last session complete, disc finalized */
863     outbuf[3] = 1;   /* first track on disc */
864     outbuf[4] = 1;   /* # of sessions */
865     outbuf[5] = 1;   /* first track of last session */
866     outbuf[6] = 1;   /* last track of last session */
867     outbuf[7] = 0x20; /* unrestricted use */
868     outbuf[8] = 0x00; /* CD-ROM or DVD-ROM */
869     /* 9-10-11: most significant byte corresponding bytes 4-5-6 */
870     /* 12-23: not meaningful for CD-ROM or DVD-ROM */
871     /* 24-31: disc bar code */
872     /* 32: disc application code */
873     /* 33: number of OPC tables */
874 
875     return 34;
876 }
877 
878 static int scsi_read_dvd_structure(SCSIDiskState *s, SCSIDiskReq *r,
879                                    uint8_t *outbuf)
880 {
881     static const int rds_caps_size[5] = {
882         [0] = 2048 + 4,
883         [1] = 4 + 4,
884         [3] = 188 + 4,
885         [4] = 2048 + 4,
886     };
887 
888     uint8_t media = r->req.cmd.buf[1];
889     uint8_t layer = r->req.cmd.buf[6];
890     uint8_t format = r->req.cmd.buf[7];
891     int size = -1;
892 
893     if (s->qdev.type != TYPE_ROM) {
894         return -1;
895     }
896     if (media != 0) {
897         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
898         return -1;
899     }
900 
901     if (format != 0xff) {
902         if (!blk_is_available(s->qdev.conf.blk)) {
903             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
904             return -1;
905         }
906         if (media_is_cd(s)) {
907             scsi_check_condition(r, SENSE_CODE(INCOMPATIBLE_FORMAT));
908             return -1;
909         }
910         if (format >= ARRAY_SIZE(rds_caps_size)) {
911             return -1;
912         }
913         size = rds_caps_size[format];
914         memset(outbuf, 0, size);
915     }
916 
917     switch (format) {
918     case 0x00: {
919         /* Physical format information */
920         uint64_t nb_sectors;
921         if (layer != 0) {
922             goto fail;
923         }
924         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
925 
926         outbuf[4] = 1;   /* DVD-ROM, part version 1 */
927         outbuf[5] = 0xf; /* 120mm disc, minimum rate unspecified */
928         outbuf[6] = 1;   /* one layer, read-only (per MMC-2 spec) */
929         outbuf[7] = 0;   /* default densities */
930 
931         stl_be_p(&outbuf[12], (nb_sectors >> 2) - 1); /* end sector */
932         stl_be_p(&outbuf[16], (nb_sectors >> 2) - 1); /* l0 end sector */
933         break;
934     }
935 
936     case 0x01: /* DVD copyright information, all zeros */
937         break;
938 
939     case 0x03: /* BCA information - invalid field for no BCA info */
940         return -1;
941 
942     case 0x04: /* DVD disc manufacturing information, all zeros */
943         break;
944 
945     case 0xff: { /* List capabilities */
946         int i;
947         size = 4;
948         for (i = 0; i < ARRAY_SIZE(rds_caps_size); i++) {
949             if (!rds_caps_size[i]) {
950                 continue;
951             }
952             outbuf[size] = i;
953             outbuf[size + 1] = 0x40; /* Not writable, readable */
954             stw_be_p(&outbuf[size + 2], rds_caps_size[i]);
955             size += 4;
956         }
957         break;
958      }
959 
960     default:
961         return -1;
962     }
963 
964     /* Size of buffer, not including 2 byte size field */
965     stw_be_p(outbuf, size - 2);
966     return size;
967 
968 fail:
969     return -1;
970 }
971 
972 static int scsi_event_status_media(SCSIDiskState *s, uint8_t *outbuf)
973 {
974     uint8_t event_code, media_status;
975 
976     media_status = 0;
977     if (s->tray_open) {
978         media_status = MS_TRAY_OPEN;
979     } else if (blk_is_inserted(s->qdev.conf.blk)) {
980         media_status = MS_MEDIA_PRESENT;
981     }
982 
983     /* Event notification descriptor */
984     event_code = MEC_NO_CHANGE;
985     if (media_status != MS_TRAY_OPEN) {
986         if (s->media_event) {
987             event_code = MEC_NEW_MEDIA;
988             s->media_event = false;
989         } else if (s->eject_request) {
990             event_code = MEC_EJECT_REQUESTED;
991             s->eject_request = false;
992         }
993     }
994 
995     outbuf[0] = event_code;
996     outbuf[1] = media_status;
997 
998     /* These fields are reserved, just clear them. */
999     outbuf[2] = 0;
1000     outbuf[3] = 0;
1001     return 4;
1002 }
1003 
1004 static int scsi_get_event_status_notification(SCSIDiskState *s, SCSIDiskReq *r,
1005                                               uint8_t *outbuf)
1006 {
1007     int size;
1008     uint8_t *buf = r->req.cmd.buf;
1009     uint8_t notification_class_request = buf[4];
1010     if (s->qdev.type != TYPE_ROM) {
1011         return -1;
1012     }
1013     if ((buf[1] & 1) == 0) {
1014         /* asynchronous */
1015         return -1;
1016     }
1017 
1018     size = 4;
1019     outbuf[0] = outbuf[1] = 0;
1020     outbuf[3] = 1 << GESN_MEDIA; /* supported events */
1021     if (notification_class_request & (1 << GESN_MEDIA)) {
1022         outbuf[2] = GESN_MEDIA;
1023         size += scsi_event_status_media(s, &outbuf[size]);
1024     } else {
1025         outbuf[2] = 0x80;
1026     }
1027     stw_be_p(outbuf, size - 4);
1028     return size;
1029 }
1030 
1031 static int scsi_get_configuration(SCSIDiskState *s, uint8_t *outbuf)
1032 {
1033     int current;
1034 
1035     if (s->qdev.type != TYPE_ROM) {
1036         return -1;
1037     }
1038 
1039     if (media_is_dvd(s)) {
1040         current = MMC_PROFILE_DVD_ROM;
1041     } else if (media_is_cd(s)) {
1042         current = MMC_PROFILE_CD_ROM;
1043     } else {
1044         current = MMC_PROFILE_NONE;
1045     }
1046 
1047     memset(outbuf, 0, 40);
1048     stl_be_p(&outbuf[0], 36); /* Bytes after the data length field */
1049     stw_be_p(&outbuf[6], current);
1050     /* outbuf[8] - outbuf[19]: Feature 0 - Profile list */
1051     outbuf[10] = 0x03; /* persistent, current */
1052     outbuf[11] = 8; /* two profiles */
1053     stw_be_p(&outbuf[12], MMC_PROFILE_DVD_ROM);
1054     outbuf[14] = (current == MMC_PROFILE_DVD_ROM);
1055     stw_be_p(&outbuf[16], MMC_PROFILE_CD_ROM);
1056     outbuf[18] = (current == MMC_PROFILE_CD_ROM);
1057     /* outbuf[20] - outbuf[31]: Feature 1 - Core feature */
1058     stw_be_p(&outbuf[20], 1);
1059     outbuf[22] = 0x08 | 0x03; /* version 2, persistent, current */
1060     outbuf[23] = 8;
1061     stl_be_p(&outbuf[24], 1); /* SCSI */
1062     outbuf[28] = 1; /* DBE = 1, mandatory */
1063     /* outbuf[32] - outbuf[39]: Feature 3 - Removable media feature */
1064     stw_be_p(&outbuf[32], 3);
1065     outbuf[34] = 0x08 | 0x03; /* version 2, persistent, current */
1066     outbuf[35] = 4;
1067     outbuf[36] = 0x39; /* tray, load=1, eject=1, unlocked at powerup, lock=1 */
1068     /* TODO: Random readable, CD read, DVD read, drive serial number,
1069        power management */
1070     return 40;
1071 }
1072 
1073 static int scsi_emulate_mechanism_status(SCSIDiskState *s, uint8_t *outbuf)
1074 {
1075     if (s->qdev.type != TYPE_ROM) {
1076         return -1;
1077     }
1078     memset(outbuf, 0, 8);
1079     outbuf[5] = 1; /* CD-ROM */
1080     return 8;
1081 }
1082 
1083 static int mode_sense_page(SCSIDiskState *s, int page, uint8_t **p_outbuf,
1084                            int page_control)
1085 {
1086     static const int mode_sense_valid[0x3f] = {
1087         [MODE_PAGE_VENDOR_SPECIFIC]        = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1088         [MODE_PAGE_HD_GEOMETRY]            = (1 << TYPE_DISK),
1089         [MODE_PAGE_FLEXIBLE_DISK_GEOMETRY] = (1 << TYPE_DISK),
1090         [MODE_PAGE_CACHING]                = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1091         [MODE_PAGE_R_W_ERROR]              = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1092         [MODE_PAGE_AUDIO_CTL]              = (1 << TYPE_ROM),
1093         [MODE_PAGE_CAPABILITIES]           = (1 << TYPE_ROM),
1094         [MODE_PAGE_APPLE_VENDOR]           = (1 << TYPE_ROM),
1095     };
1096 
1097     uint8_t *p = *p_outbuf + 2;
1098     int length;
1099 
1100     assert(page < ARRAY_SIZE(mode_sense_valid));
1101     if ((mode_sense_valid[page] & (1 << s->qdev.type)) == 0) {
1102         return -1;
1103     }
1104 
1105     /*
1106      * If Changeable Values are requested, a mask denoting those mode parameters
1107      * that are changeable shall be returned. As we currently don't support
1108      * parameter changes via MODE_SELECT all bits are returned set to zero.
1109      * The buffer was already menset to zero by the caller of this function.
1110      *
1111      * The offsets here are off by two compared to the descriptions in the
1112      * SCSI specs, because those include a 2-byte header.  This is unfortunate,
1113      * but it is done so that offsets are consistent within our implementation
1114      * of MODE SENSE and MODE SELECT.  MODE SELECT has to deal with both
1115      * 2-byte and 4-byte headers.
1116      */
1117     switch (page) {
1118     case MODE_PAGE_HD_GEOMETRY:
1119         length = 0x16;
1120         if (page_control == 1) { /* Changeable Values */
1121             break;
1122         }
1123         /* if a geometry hint is available, use it */
1124         p[0] = (s->qdev.conf.cyls >> 16) & 0xff;
1125         p[1] = (s->qdev.conf.cyls >> 8) & 0xff;
1126         p[2] = s->qdev.conf.cyls & 0xff;
1127         p[3] = s->qdev.conf.heads & 0xff;
1128         /* Write precomp start cylinder, disabled */
1129         p[4] = (s->qdev.conf.cyls >> 16) & 0xff;
1130         p[5] = (s->qdev.conf.cyls >> 8) & 0xff;
1131         p[6] = s->qdev.conf.cyls & 0xff;
1132         /* Reduced current start cylinder, disabled */
1133         p[7] = (s->qdev.conf.cyls >> 16) & 0xff;
1134         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1135         p[9] = s->qdev.conf.cyls & 0xff;
1136         /* Device step rate [ns], 200ns */
1137         p[10] = 0;
1138         p[11] = 200;
1139         /* Landing zone cylinder */
1140         p[12] = 0xff;
1141         p[13] =  0xff;
1142         p[14] = 0xff;
1143         /* Medium rotation rate [rpm], 5400 rpm */
1144         p[18] = (5400 >> 8) & 0xff;
1145         p[19] = 5400 & 0xff;
1146         break;
1147 
1148     case MODE_PAGE_FLEXIBLE_DISK_GEOMETRY:
1149         length = 0x1e;
1150         if (page_control == 1) { /* Changeable Values */
1151             break;
1152         }
1153         /* Transfer rate [kbit/s], 5Mbit/s */
1154         p[0] = 5000 >> 8;
1155         p[1] = 5000 & 0xff;
1156         /* if a geometry hint is available, use it */
1157         p[2] = s->qdev.conf.heads & 0xff;
1158         p[3] = s->qdev.conf.secs & 0xff;
1159         p[4] = s->qdev.blocksize >> 8;
1160         p[6] = (s->qdev.conf.cyls >> 8) & 0xff;
1161         p[7] = s->qdev.conf.cyls & 0xff;
1162         /* Write precomp start cylinder, disabled */
1163         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1164         p[9] = s->qdev.conf.cyls & 0xff;
1165         /* Reduced current start cylinder, disabled */
1166         p[10] = (s->qdev.conf.cyls >> 8) & 0xff;
1167         p[11] = s->qdev.conf.cyls & 0xff;
1168         /* Device step rate [100us], 100us */
1169         p[12] = 0;
1170         p[13] = 1;
1171         /* Device step pulse width [us], 1us */
1172         p[14] = 1;
1173         /* Device head settle delay [100us], 100us */
1174         p[15] = 0;
1175         p[16] = 1;
1176         /* Motor on delay [0.1s], 0.1s */
1177         p[17] = 1;
1178         /* Motor off delay [0.1s], 0.1s */
1179         p[18] = 1;
1180         /* Medium rotation rate [rpm], 5400 rpm */
1181         p[26] = (5400 >> 8) & 0xff;
1182         p[27] = 5400 & 0xff;
1183         break;
1184 
1185     case MODE_PAGE_CACHING:
1186         length = 0x12;
1187         if (page_control == 1 || /* Changeable Values */
1188             blk_enable_write_cache(s->qdev.conf.blk)) {
1189             p[0] = 4; /* WCE */
1190         }
1191         break;
1192 
1193     case MODE_PAGE_R_W_ERROR:
1194         length = 10;
1195         if (page_control == 1) { /* Changeable Values */
1196             if (s->qdev.type == TYPE_ROM) {
1197                 /* Automatic Write Reallocation Enabled */
1198                 p[0] = 0x80;
1199             }
1200             break;
1201         }
1202         p[0] = 0x80; /* Automatic Write Reallocation Enabled */
1203         if (s->qdev.type == TYPE_ROM) {
1204             p[1] = 0x20; /* Read Retry Count */
1205         }
1206         break;
1207 
1208     case MODE_PAGE_AUDIO_CTL:
1209         length = 14;
1210         break;
1211 
1212     case MODE_PAGE_CAPABILITIES:
1213         length = 0x14;
1214         if (page_control == 1) { /* Changeable Values */
1215             break;
1216         }
1217 
1218         p[0] = 0x3b; /* CD-R & CD-RW read */
1219         p[1] = 0; /* Writing not supported */
1220         p[2] = 0x7f; /* Audio, composite, digital out,
1221                         mode 2 form 1&2, multi session */
1222         p[3] = 0xff; /* CD DA, DA accurate, RW supported,
1223                         RW corrected, C2 errors, ISRC,
1224                         UPC, Bar code */
1225         p[4] = 0x2d | (s->tray_locked ? 2 : 0);
1226         /* Locking supported, jumper present, eject, tray */
1227         p[5] = 0; /* no volume & mute control, no
1228                      changer */
1229         p[6] = (50 * 176) >> 8; /* 50x read speed */
1230         p[7] = (50 * 176) & 0xff;
1231         p[8] = 2 >> 8; /* Two volume levels */
1232         p[9] = 2 & 0xff;
1233         p[10] = 2048 >> 8; /* 2M buffer */
1234         p[11] = 2048 & 0xff;
1235         p[12] = (16 * 176) >> 8; /* 16x read speed current */
1236         p[13] = (16 * 176) & 0xff;
1237         p[16] = (16 * 176) >> 8; /* 16x write speed */
1238         p[17] = (16 * 176) & 0xff;
1239         p[18] = (16 * 176) >> 8; /* 16x write speed current */
1240         p[19] = (16 * 176) & 0xff;
1241         break;
1242 
1243      case MODE_PAGE_APPLE_VENDOR:
1244         if (s->quirks & (1 << SCSI_DISK_QUIRK_MODE_PAGE_APPLE_VENDOR)) {
1245             length = 0x1e;
1246             if (page_control == 1) { /* Changeable Values */
1247                 break;
1248             }
1249 
1250             memset(p, 0, length);
1251             strcpy((char *)p + 8, "APPLE COMPUTER, INC   ");
1252             break;
1253         } else {
1254             return -1;
1255         }
1256 
1257     case MODE_PAGE_VENDOR_SPECIFIC:
1258         if (s->qdev.type == TYPE_DISK && (s->quirks &
1259             (1 << SCSI_DISK_QUIRK_MODE_PAGE_VENDOR_SPECIFIC_APPLE))) {
1260             length = 0x2;
1261             if (page_control == 1) { /* Changeable Values */
1262                 p[0] = 0xff;
1263                 p[1] = 0xff;
1264                 break;
1265             }
1266             p[0] = 0;
1267             p[1] = 0;
1268             break;
1269         } else {
1270             return -1;
1271         }
1272 
1273     default:
1274         return -1;
1275     }
1276 
1277     assert(length < 256);
1278     (*p_outbuf)[0] = page;
1279     (*p_outbuf)[1] = length;
1280     *p_outbuf += length + 2;
1281     return length + 2;
1282 }
1283 
1284 static int scsi_disk_emulate_mode_sense(SCSIDiskReq *r, uint8_t *outbuf)
1285 {
1286     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1287     uint64_t nb_sectors;
1288     bool dbd;
1289     int page, buflen, ret, page_control;
1290     uint8_t *p;
1291     uint8_t dev_specific_param;
1292 
1293     dbd = (r->req.cmd.buf[1] & 0x8) != 0;
1294     page = r->req.cmd.buf[2] & 0x3f;
1295     page_control = (r->req.cmd.buf[2] & 0xc0) >> 6;
1296 
1297     trace_scsi_disk_emulate_mode_sense((r->req.cmd.buf[0] == MODE_SENSE) ? 6 :
1298                                        10, page, r->req.cmd.xfer, page_control);
1299     memset(outbuf, 0, r->req.cmd.xfer);
1300     p = outbuf;
1301 
1302     if (s->qdev.type == TYPE_DISK) {
1303         dev_specific_param = s->features & (1 << SCSI_DISK_F_DPOFUA) ? 0x10 : 0;
1304         if (!blk_is_writable(s->qdev.conf.blk)) {
1305             dev_specific_param |= 0x80; /* Readonly.  */
1306         }
1307     } else {
1308         if (s->quirks & (1 << SCSI_DISK_QUIRK_MODE_SENSE_ROM_USE_DBD)) {
1309             /* Use DBD from the request... */
1310             dev_specific_param = 0x00;
1311 
1312             /*
1313              * ... unless we receive a request for MODE_PAGE_APPLE_VENDOR
1314              * which should never return a block descriptor even though DBD is
1315              * not set, otherwise CDROM detection fails in MacOS
1316              */
1317             if (s->quirks & (1 << SCSI_DISK_QUIRK_MODE_PAGE_APPLE_VENDOR) &&
1318                 page == MODE_PAGE_APPLE_VENDOR) {
1319                 dbd = true;
1320             }
1321         } else {
1322             /*
1323              * MMC prescribes that CD/DVD drives have no block descriptors,
1324              * and defines no device-specific parameter.
1325              */
1326             dev_specific_param = 0x00;
1327             dbd = true;
1328         }
1329     }
1330 
1331     if (r->req.cmd.buf[0] == MODE_SENSE) {
1332         p[1] = 0; /* Default media type.  */
1333         p[2] = dev_specific_param;
1334         p[3] = 0; /* Block descriptor length.  */
1335         p += 4;
1336     } else { /* MODE_SENSE_10 */
1337         p[2] = 0; /* Default media type.  */
1338         p[3] = dev_specific_param;
1339         p[6] = p[7] = 0; /* Block descriptor length.  */
1340         p += 8;
1341     }
1342 
1343     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1344     if (!dbd && nb_sectors) {
1345         if (r->req.cmd.buf[0] == MODE_SENSE) {
1346             outbuf[3] = 8; /* Block descriptor length  */
1347         } else { /* MODE_SENSE_10 */
1348             outbuf[7] = 8; /* Block descriptor length  */
1349         }
1350         nb_sectors /= (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1351         if (nb_sectors > 0xffffff) {
1352             nb_sectors = 0;
1353         }
1354         p[0] = 0; /* media density code */
1355         p[1] = (nb_sectors >> 16) & 0xff;
1356         p[2] = (nb_sectors >> 8) & 0xff;
1357         p[3] = nb_sectors & 0xff;
1358         p[4] = 0; /* reserved */
1359         p[5] = 0; /* bytes 5-7 are the sector size in bytes */
1360         p[6] = s->qdev.blocksize >> 8;
1361         p[7] = 0;
1362         p += 8;
1363     }
1364 
1365     if (page_control == 3) {
1366         /* Saved Values */
1367         scsi_check_condition(r, SENSE_CODE(SAVING_PARAMS_NOT_SUPPORTED));
1368         return -1;
1369     }
1370 
1371     if (page == 0x3f) {
1372         for (page = 0; page <= 0x3e; page++) {
1373             mode_sense_page(s, page, &p, page_control);
1374         }
1375     } else {
1376         ret = mode_sense_page(s, page, &p, page_control);
1377         if (ret == -1) {
1378             return -1;
1379         }
1380     }
1381 
1382     buflen = p - outbuf;
1383     /*
1384      * The mode data length field specifies the length in bytes of the
1385      * following data that is available to be transferred. The mode data
1386      * length does not include itself.
1387      */
1388     if (r->req.cmd.buf[0] == MODE_SENSE) {
1389         outbuf[0] = buflen - 1;
1390     } else { /* MODE_SENSE_10 */
1391         outbuf[0] = ((buflen - 2) >> 8) & 0xff;
1392         outbuf[1] = (buflen - 2) & 0xff;
1393     }
1394     return buflen;
1395 }
1396 
1397 static int scsi_disk_emulate_read_toc(SCSIRequest *req, uint8_t *outbuf)
1398 {
1399     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1400     int start_track, format, msf, toclen;
1401     uint64_t nb_sectors;
1402 
1403     msf = req->cmd.buf[1] & 2;
1404     format = req->cmd.buf[2] & 0xf;
1405     start_track = req->cmd.buf[6];
1406     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1407     trace_scsi_disk_emulate_read_toc(start_track, format, msf >> 1);
1408     nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
1409     switch (format) {
1410     case 0:
1411         toclen = cdrom_read_toc(nb_sectors, outbuf, msf, start_track);
1412         break;
1413     case 1:
1414         /* multi session : only a single session defined */
1415         toclen = 12;
1416         memset(outbuf, 0, 12);
1417         outbuf[1] = 0x0a;
1418         outbuf[2] = 0x01;
1419         outbuf[3] = 0x01;
1420         break;
1421     case 2:
1422         toclen = cdrom_read_toc_raw(nb_sectors, outbuf, msf, start_track);
1423         break;
1424     default:
1425         return -1;
1426     }
1427     return toclen;
1428 }
1429 
1430 static int scsi_disk_emulate_start_stop(SCSIDiskReq *r)
1431 {
1432     SCSIRequest *req = &r->req;
1433     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1434     bool start = req->cmd.buf[4] & 1;
1435     bool loej = req->cmd.buf[4] & 2; /* load on start, eject on !start */
1436     int pwrcnd = req->cmd.buf[4] & 0xf0;
1437 
1438     if (pwrcnd) {
1439         /* eject/load only happens for power condition == 0 */
1440         return 0;
1441     }
1442 
1443     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) && loej) {
1444         if (!start && !s->tray_open && s->tray_locked) {
1445             scsi_check_condition(r,
1446                                  blk_is_inserted(s->qdev.conf.blk)
1447                                  ? SENSE_CODE(ILLEGAL_REQ_REMOVAL_PREVENTED)
1448                                  : SENSE_CODE(NOT_READY_REMOVAL_PREVENTED));
1449             return -1;
1450         }
1451 
1452         if (s->tray_open != !start) {
1453             blk_eject(s->qdev.conf.blk, !start);
1454             s->tray_open = !start;
1455         }
1456     }
1457     return 0;
1458 }
1459 
1460 static void scsi_disk_emulate_read_data(SCSIRequest *req)
1461 {
1462     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1463     int buflen = r->iov.iov_len;
1464 
1465     if (buflen) {
1466         trace_scsi_disk_emulate_read_data(buflen);
1467         r->iov.iov_len = 0;
1468         r->started = true;
1469         scsi_req_data(&r->req, buflen);
1470         return;
1471     }
1472 
1473     /* This also clears the sense buffer for REQUEST SENSE.  */
1474     scsi_req_complete(&r->req, GOOD);
1475 }
1476 
1477 static int scsi_disk_check_mode_select(SCSIDiskState *s, int page,
1478                                        uint8_t *inbuf, int inlen)
1479 {
1480     uint8_t mode_current[SCSI_MAX_MODE_LEN];
1481     uint8_t mode_changeable[SCSI_MAX_MODE_LEN];
1482     uint8_t *p;
1483     int len, expected_len, changeable_len, i;
1484 
1485     /* The input buffer does not include the page header, so it is
1486      * off by 2 bytes.
1487      */
1488     expected_len = inlen + 2;
1489     if (expected_len > SCSI_MAX_MODE_LEN) {
1490         return -1;
1491     }
1492 
1493     /* MODE_PAGE_ALLS is only valid for MODE SENSE commands */
1494     if (page == MODE_PAGE_ALLS) {
1495         return -1;
1496     }
1497 
1498     p = mode_current;
1499     memset(mode_current, 0, inlen + 2);
1500     len = mode_sense_page(s, page, &p, 0);
1501     if (len < 0 || len != expected_len) {
1502         return -1;
1503     }
1504 
1505     p = mode_changeable;
1506     memset(mode_changeable, 0, inlen + 2);
1507     changeable_len = mode_sense_page(s, page, &p, 1);
1508     assert(changeable_len == len);
1509 
1510     /* Check that unchangeable bits are the same as what MODE SENSE
1511      * would return.
1512      */
1513     for (i = 2; i < len; i++) {
1514         if (((mode_current[i] ^ inbuf[i - 2]) & ~mode_changeable[i]) != 0) {
1515             return -1;
1516         }
1517     }
1518     return 0;
1519 }
1520 
1521 static void scsi_disk_apply_mode_select(SCSIDiskState *s, int page, uint8_t *p)
1522 {
1523     switch (page) {
1524     case MODE_PAGE_CACHING:
1525         blk_set_enable_write_cache(s->qdev.conf.blk, (p[0] & 4) != 0);
1526         break;
1527 
1528     default:
1529         break;
1530     }
1531 }
1532 
1533 static int mode_select_pages(SCSIDiskReq *r, uint8_t *p, int len, bool change)
1534 {
1535     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1536 
1537     while (len > 0) {
1538         int page, subpage, page_len;
1539 
1540         /* Parse both possible formats for the mode page headers.  */
1541         page = p[0] & 0x3f;
1542         if (p[0] & 0x40) {
1543             if (len < 4) {
1544                 goto invalid_param_len;
1545             }
1546             subpage = p[1];
1547             page_len = lduw_be_p(&p[2]);
1548             p += 4;
1549             len -= 4;
1550         } else {
1551             if (len < 2) {
1552                 goto invalid_param_len;
1553             }
1554             subpage = 0;
1555             page_len = p[1];
1556             p += 2;
1557             len -= 2;
1558         }
1559 
1560         if (subpage) {
1561             goto invalid_param;
1562         }
1563         if (page_len > len) {
1564             if (!(s->quirks & SCSI_DISK_QUIRK_MODE_PAGE_TRUNCATED)) {
1565                 goto invalid_param_len;
1566             }
1567             trace_scsi_disk_mode_select_page_truncated(page, page_len, len);
1568         }
1569 
1570         if (!change) {
1571             if (scsi_disk_check_mode_select(s, page, p, page_len) < 0) {
1572                 goto invalid_param;
1573             }
1574         } else {
1575             scsi_disk_apply_mode_select(s, page, p);
1576         }
1577 
1578         p += page_len;
1579         len -= page_len;
1580     }
1581     return 0;
1582 
1583 invalid_param:
1584     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1585     return -1;
1586 
1587 invalid_param_len:
1588     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1589     return -1;
1590 }
1591 
1592 static void scsi_disk_emulate_mode_select(SCSIDiskReq *r, uint8_t *inbuf)
1593 {
1594     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1595     uint8_t *p = inbuf;
1596     int cmd = r->req.cmd.buf[0];
1597     int len = r->req.cmd.xfer;
1598     int hdr_len = (cmd == MODE_SELECT ? 4 : 8);
1599     int bd_len, bs;
1600     int pass;
1601 
1602     if ((r->req.cmd.buf[1] & 0x11) != 0x10) {
1603         if (!(s->quirks &
1604             (1 << SCSI_DISK_QUIRK_MODE_PAGE_VENDOR_SPECIFIC_APPLE))) {
1605             /* We only support PF=1, SP=0.  */
1606             goto invalid_field;
1607         }
1608     }
1609 
1610     if (len < hdr_len) {
1611         goto invalid_param_len;
1612     }
1613 
1614     bd_len = (cmd == MODE_SELECT ? p[3] : lduw_be_p(&p[6]));
1615     len -= hdr_len;
1616     p += hdr_len;
1617     if (len < bd_len) {
1618         goto invalid_param_len;
1619     }
1620     if (bd_len != 0 && bd_len != 8) {
1621         goto invalid_param;
1622     }
1623 
1624     /* Allow changing the block size */
1625     if (bd_len) {
1626         bs = p[5] << 16 | p[6] << 8 | p[7];
1627 
1628         /*
1629          * Since the existing code only checks/updates bits 8-15 of the block
1630          * size, restrict ourselves to the same requirement for now to ensure
1631          * that a block size set by a block descriptor and then read back by
1632          * a subsequent SCSI command will be the same. Also disallow a block
1633          * size of 256 since we cannot handle anything below BDRV_SECTOR_SIZE.
1634          */
1635         if (bs && !(bs & ~0xfe00) && bs != s->qdev.blocksize) {
1636             s->qdev.blocksize = bs;
1637             trace_scsi_disk_mode_select_set_blocksize(s->qdev.blocksize);
1638         }
1639     }
1640 
1641     len -= bd_len;
1642     p += bd_len;
1643 
1644     /* Ensure no change is made if there is an error!  */
1645     for (pass = 0; pass < 2; pass++) {
1646         if (mode_select_pages(r, p, len, pass == 1) < 0) {
1647             assert(pass == 0);
1648             return;
1649         }
1650     }
1651     if (!blk_enable_write_cache(s->qdev.conf.blk)) {
1652         /* The request is used as the AIO opaque value, so add a ref.  */
1653         scsi_req_ref(&r->req);
1654         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
1655                          BLOCK_ACCT_FLUSH);
1656         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
1657         return;
1658     }
1659 
1660     scsi_req_complete(&r->req, GOOD);
1661     return;
1662 
1663 invalid_param:
1664     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1665     return;
1666 
1667 invalid_param_len:
1668     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1669     return;
1670 
1671 invalid_field:
1672     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1673 }
1674 
1675 /* sector_num and nb_sectors expected to be in qdev blocksize */
1676 static inline bool check_lba_range(SCSIDiskState *s,
1677                                    uint64_t sector_num, uint32_t nb_sectors)
1678 {
1679     /*
1680      * The first line tests that no overflow happens when computing the last
1681      * sector.  The second line tests that the last accessed sector is in
1682      * range.
1683      *
1684      * Careful, the computations should not underflow for nb_sectors == 0,
1685      * and a 0-block read to the first LBA beyond the end of device is
1686      * valid.
1687      */
1688     return (sector_num <= sector_num + nb_sectors &&
1689             sector_num + nb_sectors <= s->qdev.max_lba + 1);
1690 }
1691 
1692 typedef struct UnmapCBData {
1693     SCSIDiskReq *r;
1694     uint8_t *inbuf;
1695     int count;
1696 } UnmapCBData;
1697 
1698 static void scsi_unmap_complete(void *opaque, int ret);
1699 
1700 static void scsi_unmap_complete_noio(UnmapCBData *data, int ret)
1701 {
1702     SCSIDiskReq *r = data->r;
1703     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1704 
1705     assert(r->req.aiocb == NULL);
1706 
1707     if (data->count > 0) {
1708         uint64_t sector_num = ldq_be_p(&data->inbuf[0]);
1709         uint32_t nb_sectors = ldl_be_p(&data->inbuf[8]) & 0xffffffffULL;
1710         r->sector = sector_num * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1711         r->sector_count = nb_sectors * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1712 
1713         if (!check_lba_range(s, sector_num, nb_sectors)) {
1714             block_acct_invalid(blk_get_stats(s->qdev.conf.blk),
1715                                BLOCK_ACCT_UNMAP);
1716             scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1717             goto done;
1718         }
1719 
1720         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1721                          r->sector_count * BDRV_SECTOR_SIZE,
1722                          BLOCK_ACCT_UNMAP);
1723 
1724         r->req.aiocb = blk_aio_pdiscard(s->qdev.conf.blk,
1725                                         r->sector * BDRV_SECTOR_SIZE,
1726                                         r->sector_count * BDRV_SECTOR_SIZE,
1727                                         scsi_unmap_complete, data);
1728         data->count--;
1729         data->inbuf += 16;
1730         return;
1731     }
1732 
1733     scsi_req_complete(&r->req, GOOD);
1734 
1735 done:
1736     scsi_req_unref(&r->req);
1737     g_free(data);
1738 }
1739 
1740 static void scsi_unmap_complete(void *opaque, int ret)
1741 {
1742     UnmapCBData *data = opaque;
1743     SCSIDiskReq *r = data->r;
1744     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1745 
1746     assert(r->req.aiocb != NULL);
1747     r->req.aiocb = NULL;
1748 
1749     if (scsi_disk_req_check_error(r, ret, true)) {
1750         scsi_req_unref(&r->req);
1751         g_free(data);
1752     } else {
1753         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1754         scsi_unmap_complete_noio(data, ret);
1755     }
1756 }
1757 
1758 static void scsi_disk_emulate_unmap(SCSIDiskReq *r, uint8_t *inbuf)
1759 {
1760     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1761     uint8_t *p = inbuf;
1762     int len = r->req.cmd.xfer;
1763     UnmapCBData *data;
1764 
1765     /* Reject ANCHOR=1.  */
1766     if (r->req.cmd.buf[1] & 0x1) {
1767         goto invalid_field;
1768     }
1769 
1770     if (len < 8) {
1771         goto invalid_param_len;
1772     }
1773     if (len < lduw_be_p(&p[0]) + 2) {
1774         goto invalid_param_len;
1775     }
1776     if (len < lduw_be_p(&p[2]) + 8) {
1777         goto invalid_param_len;
1778     }
1779     if (lduw_be_p(&p[2]) & 15) {
1780         goto invalid_param_len;
1781     }
1782 
1783     if (!blk_is_writable(s->qdev.conf.blk)) {
1784         block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1785         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1786         return;
1787     }
1788 
1789     data = g_new0(UnmapCBData, 1);
1790     data->r = r;
1791     data->inbuf = &p[8];
1792     data->count = lduw_be_p(&p[2]) >> 4;
1793 
1794     /* The matching unref is in scsi_unmap_complete, before data is freed.  */
1795     scsi_req_ref(&r->req);
1796     scsi_unmap_complete_noio(data, 0);
1797     return;
1798 
1799 invalid_param_len:
1800     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1801     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1802     return;
1803 
1804 invalid_field:
1805     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1806     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1807 }
1808 
1809 typedef struct WriteSameCBData {
1810     SCSIDiskReq *r;
1811     int64_t sector;
1812     int nb_sectors;
1813     QEMUIOVector qiov;
1814     struct iovec iov;
1815 } WriteSameCBData;
1816 
1817 static void scsi_write_same_complete(void *opaque, int ret)
1818 {
1819     WriteSameCBData *data = opaque;
1820     SCSIDiskReq *r = data->r;
1821     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1822 
1823     assert(r->req.aiocb != NULL);
1824     r->req.aiocb = NULL;
1825 
1826     if (scsi_disk_req_check_error(r, ret, true)) {
1827         goto done;
1828     }
1829 
1830     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1831 
1832     data->nb_sectors -= data->iov.iov_len / BDRV_SECTOR_SIZE;
1833     data->sector += data->iov.iov_len / BDRV_SECTOR_SIZE;
1834     data->iov.iov_len = MIN(data->nb_sectors * BDRV_SECTOR_SIZE,
1835                             data->iov.iov_len);
1836     if (data->iov.iov_len) {
1837         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1838                          data->iov.iov_len, BLOCK_ACCT_WRITE);
1839         /* Reinitialize qiov, to handle unaligned WRITE SAME request
1840          * where final qiov may need smaller size */
1841         qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1842         r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1843                                        data->sector << BDRV_SECTOR_BITS,
1844                                        &data->qiov, 0,
1845                                        scsi_write_same_complete, data);
1846         return;
1847     }
1848 
1849     scsi_req_complete(&r->req, GOOD);
1850 
1851 done:
1852     scsi_req_unref(&r->req);
1853     qemu_vfree(data->iov.iov_base);
1854     g_free(data);
1855 }
1856 
1857 static void scsi_disk_emulate_write_same(SCSIDiskReq *r, uint8_t *inbuf)
1858 {
1859     SCSIRequest *req = &r->req;
1860     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1861     uint32_t nb_sectors = scsi_data_cdb_xfer(r->req.cmd.buf);
1862     WriteSameCBData *data;
1863     uint8_t *buf;
1864     int i, l;
1865 
1866     /* Fail if PBDATA=1 or LBDATA=1 or ANCHOR=1.  */
1867     if (nb_sectors == 0 || (req->cmd.buf[1] & 0x16)) {
1868         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1869         return;
1870     }
1871 
1872     if (!blk_is_writable(s->qdev.conf.blk)) {
1873         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1874         return;
1875     }
1876     if (!check_lba_range(s, r->req.cmd.lba, nb_sectors)) {
1877         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1878         return;
1879     }
1880 
1881     if ((req->cmd.buf[1] & 0x1) || buffer_is_zero(inbuf, s->qdev.blocksize)) {
1882         int flags = (req->cmd.buf[1] & 0x8) ? BDRV_REQ_MAY_UNMAP : 0;
1883 
1884         /* The request is used as the AIO opaque value, so add a ref.  */
1885         scsi_req_ref(&r->req);
1886         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1887                          nb_sectors * s->qdev.blocksize,
1888                         BLOCK_ACCT_WRITE);
1889         r->req.aiocb = blk_aio_pwrite_zeroes(s->qdev.conf.blk,
1890                                 r->req.cmd.lba * s->qdev.blocksize,
1891                                 nb_sectors * s->qdev.blocksize,
1892                                 flags, scsi_aio_complete, r);
1893         return;
1894     }
1895 
1896     data = g_new0(WriteSameCBData, 1);
1897     data->r = r;
1898     data->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1899     data->nb_sectors = nb_sectors * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1900     data->iov.iov_len = MIN(data->nb_sectors * BDRV_SECTOR_SIZE,
1901                             SCSI_WRITE_SAME_MAX);
1902     data->iov.iov_base = buf = blk_blockalign(s->qdev.conf.blk,
1903                                               data->iov.iov_len);
1904     qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1905 
1906     for (i = 0; i < data->iov.iov_len; i += l) {
1907         l = MIN(s->qdev.blocksize, data->iov.iov_len - i);
1908         memcpy(&buf[i], inbuf, l);
1909     }
1910 
1911     scsi_req_ref(&r->req);
1912     block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1913                      data->iov.iov_len, BLOCK_ACCT_WRITE);
1914     r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1915                                    data->sector << BDRV_SECTOR_BITS,
1916                                    &data->qiov, 0,
1917                                    scsi_write_same_complete, data);
1918 }
1919 
1920 static void scsi_disk_emulate_write_data(SCSIRequest *req)
1921 {
1922     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1923 
1924     if (r->iov.iov_len) {
1925         int buflen = r->iov.iov_len;
1926         trace_scsi_disk_emulate_write_data(buflen);
1927         r->iov.iov_len = 0;
1928         scsi_req_data(&r->req, buflen);
1929         return;
1930     }
1931 
1932     switch (req->cmd.buf[0]) {
1933     case MODE_SELECT:
1934     case MODE_SELECT_10:
1935         /* This also clears the sense buffer for REQUEST SENSE.  */
1936         scsi_disk_emulate_mode_select(r, r->iov.iov_base);
1937         break;
1938 
1939     case UNMAP:
1940         scsi_disk_emulate_unmap(r, r->iov.iov_base);
1941         break;
1942 
1943     case VERIFY_10:
1944     case VERIFY_12:
1945     case VERIFY_16:
1946         if (r->req.status == -1) {
1947             scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1948         }
1949         break;
1950 
1951     case WRITE_SAME_10:
1952     case WRITE_SAME_16:
1953         scsi_disk_emulate_write_same(r, r->iov.iov_base);
1954         break;
1955 
1956     case FORMAT_UNIT:
1957         scsi_req_complete(&r->req, GOOD);
1958         break;
1959 
1960     default:
1961         abort();
1962     }
1963 }
1964 
1965 static int32_t scsi_disk_emulate_command(SCSIRequest *req, uint8_t *buf)
1966 {
1967     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1968     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1969     uint64_t nb_sectors;
1970     uint8_t *outbuf;
1971     int buflen;
1972 
1973     switch (req->cmd.buf[0]) {
1974     case INQUIRY:
1975     case MODE_SENSE:
1976     case MODE_SENSE_10:
1977     case RESERVE:
1978     case RESERVE_10:
1979     case RELEASE:
1980     case RELEASE_10:
1981     case START_STOP:
1982     case ALLOW_MEDIUM_REMOVAL:
1983     case GET_CONFIGURATION:
1984     case GET_EVENT_STATUS_NOTIFICATION:
1985     case MECHANISM_STATUS:
1986     case REQUEST_SENSE:
1987         break;
1988 
1989     default:
1990         if (!blk_is_available(s->qdev.conf.blk)) {
1991             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
1992             return 0;
1993         }
1994         break;
1995     }
1996 
1997     /*
1998      * FIXME: we shouldn't return anything bigger than 4k, but the code
1999      * requires the buffer to be as big as req->cmd.xfer in several
2000      * places.  So, do not allow CDBs with a very large ALLOCATION
2001      * LENGTH.  The real fix would be to modify scsi_read_data and
2002      * dma_buf_read, so that they return data beyond the buflen
2003      * as all zeros.
2004      */
2005     if (req->cmd.xfer > 65536) {
2006         goto illegal_request;
2007     }
2008     r->buflen = MAX(4096, req->cmd.xfer);
2009 
2010     if (!r->iov.iov_base) {
2011         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
2012     }
2013 
2014     outbuf = r->iov.iov_base;
2015     memset(outbuf, 0, r->buflen);
2016     switch (req->cmd.buf[0]) {
2017     case TEST_UNIT_READY:
2018         assert(blk_is_available(s->qdev.conf.blk));
2019         break;
2020     case INQUIRY:
2021         buflen = scsi_disk_emulate_inquiry(req, outbuf);
2022         if (buflen < 0) {
2023             goto illegal_request;
2024         }
2025         break;
2026     case MODE_SENSE:
2027     case MODE_SENSE_10:
2028         buflen = scsi_disk_emulate_mode_sense(r, outbuf);
2029         if (buflen < 0) {
2030             goto illegal_request;
2031         }
2032         break;
2033     case READ_TOC:
2034         buflen = scsi_disk_emulate_read_toc(req, outbuf);
2035         if (buflen < 0) {
2036             goto illegal_request;
2037         }
2038         break;
2039     case RESERVE:
2040         if (req->cmd.buf[1] & 1) {
2041             goto illegal_request;
2042         }
2043         break;
2044     case RESERVE_10:
2045         if (req->cmd.buf[1] & 3) {
2046             goto illegal_request;
2047         }
2048         break;
2049     case RELEASE:
2050         if (req->cmd.buf[1] & 1) {
2051             goto illegal_request;
2052         }
2053         break;
2054     case RELEASE_10:
2055         if (req->cmd.buf[1] & 3) {
2056             goto illegal_request;
2057         }
2058         break;
2059     case START_STOP:
2060         if (scsi_disk_emulate_start_stop(r) < 0) {
2061             return 0;
2062         }
2063         break;
2064     case ALLOW_MEDIUM_REMOVAL:
2065         s->tray_locked = req->cmd.buf[4] & 1;
2066         blk_lock_medium(s->qdev.conf.blk, req->cmd.buf[4] & 1);
2067         break;
2068     case READ_CAPACITY_10:
2069         /* The normal LEN field for this command is zero.  */
2070         memset(outbuf, 0, 8);
2071         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2072         if (!nb_sectors) {
2073             scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
2074             return 0;
2075         }
2076         if ((req->cmd.buf[8] & 1) == 0 && req->cmd.lba) {
2077             goto illegal_request;
2078         }
2079         nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2080         /* Returned value is the address of the last sector.  */
2081         nb_sectors--;
2082         /* Remember the new size for read/write sanity checking. */
2083         s->qdev.max_lba = nb_sectors;
2084         /* Clip to 2TB, instead of returning capacity modulo 2TB. */
2085         if (nb_sectors > UINT32_MAX) {
2086             nb_sectors = UINT32_MAX;
2087         }
2088         outbuf[0] = (nb_sectors >> 24) & 0xff;
2089         outbuf[1] = (nb_sectors >> 16) & 0xff;
2090         outbuf[2] = (nb_sectors >> 8) & 0xff;
2091         outbuf[3] = nb_sectors & 0xff;
2092         outbuf[4] = 0;
2093         outbuf[5] = 0;
2094         outbuf[6] = s->qdev.blocksize >> 8;
2095         outbuf[7] = 0;
2096         break;
2097     case REQUEST_SENSE:
2098         /* Just return "NO SENSE".  */
2099         buflen = scsi_convert_sense(NULL, 0, outbuf, r->buflen,
2100                                     (req->cmd.buf[1] & 1) == 0);
2101         if (buflen < 0) {
2102             goto illegal_request;
2103         }
2104         break;
2105     case MECHANISM_STATUS:
2106         buflen = scsi_emulate_mechanism_status(s, outbuf);
2107         if (buflen < 0) {
2108             goto illegal_request;
2109         }
2110         break;
2111     case GET_CONFIGURATION:
2112         buflen = scsi_get_configuration(s, outbuf);
2113         if (buflen < 0) {
2114             goto illegal_request;
2115         }
2116         break;
2117     case GET_EVENT_STATUS_NOTIFICATION:
2118         buflen = scsi_get_event_status_notification(s, r, outbuf);
2119         if (buflen < 0) {
2120             goto illegal_request;
2121         }
2122         break;
2123     case READ_DISC_INFORMATION:
2124         buflen = scsi_read_disc_information(s, r, outbuf);
2125         if (buflen < 0) {
2126             goto illegal_request;
2127         }
2128         break;
2129     case READ_DVD_STRUCTURE:
2130         buflen = scsi_read_dvd_structure(s, r, outbuf);
2131         if (buflen < 0) {
2132             goto illegal_request;
2133         }
2134         break;
2135     case SERVICE_ACTION_IN_16:
2136         /* Service Action In subcommands. */
2137         if ((req->cmd.buf[1] & 31) == SAI_READ_CAPACITY_16) {
2138             trace_scsi_disk_emulate_command_SAI_16();
2139             memset(outbuf, 0, req->cmd.xfer);
2140             blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2141             if (!nb_sectors) {
2142                 scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
2143                 return 0;
2144             }
2145             if ((req->cmd.buf[14] & 1) == 0 && req->cmd.lba) {
2146                 goto illegal_request;
2147             }
2148             nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2149             /* Returned value is the address of the last sector.  */
2150             nb_sectors--;
2151             /* Remember the new size for read/write sanity checking. */
2152             s->qdev.max_lba = nb_sectors;
2153             outbuf[0] = (nb_sectors >> 56) & 0xff;
2154             outbuf[1] = (nb_sectors >> 48) & 0xff;
2155             outbuf[2] = (nb_sectors >> 40) & 0xff;
2156             outbuf[3] = (nb_sectors >> 32) & 0xff;
2157             outbuf[4] = (nb_sectors >> 24) & 0xff;
2158             outbuf[5] = (nb_sectors >> 16) & 0xff;
2159             outbuf[6] = (nb_sectors >> 8) & 0xff;
2160             outbuf[7] = nb_sectors & 0xff;
2161             outbuf[8] = 0;
2162             outbuf[9] = 0;
2163             outbuf[10] = s->qdev.blocksize >> 8;
2164             outbuf[11] = 0;
2165             outbuf[12] = 0;
2166             outbuf[13] = get_physical_block_exp(&s->qdev.conf);
2167 
2168             /* set TPE bit if the format supports discard */
2169             if (s->qdev.conf.discard_granularity) {
2170                 outbuf[14] = 0x80;
2171             }
2172 
2173             /* Protection, exponent and lowest lba field left blank. */
2174             break;
2175         }
2176         trace_scsi_disk_emulate_command_SAI_unsupported();
2177         goto illegal_request;
2178     case SYNCHRONIZE_CACHE:
2179         /* The request is used as the AIO opaque value, so add a ref.  */
2180         scsi_req_ref(&r->req);
2181         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
2182                          BLOCK_ACCT_FLUSH);
2183         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
2184         return 0;
2185     case SEEK_10:
2186         trace_scsi_disk_emulate_command_SEEK_10(r->req.cmd.lba);
2187         if (r->req.cmd.lba > s->qdev.max_lba) {
2188             goto illegal_lba;
2189         }
2190         break;
2191     case MODE_SELECT:
2192         trace_scsi_disk_emulate_command_MODE_SELECT(r->req.cmd.xfer);
2193         break;
2194     case MODE_SELECT_10:
2195         trace_scsi_disk_emulate_command_MODE_SELECT_10(r->req.cmd.xfer);
2196         break;
2197     case UNMAP:
2198         trace_scsi_disk_emulate_command_UNMAP(r->req.cmd.xfer);
2199         break;
2200     case VERIFY_10:
2201     case VERIFY_12:
2202     case VERIFY_16:
2203         trace_scsi_disk_emulate_command_VERIFY((req->cmd.buf[1] >> 1) & 3);
2204         if (req->cmd.buf[1] & 6) {
2205             goto illegal_request;
2206         }
2207         break;
2208     case WRITE_SAME_10:
2209     case WRITE_SAME_16:
2210         trace_scsi_disk_emulate_command_WRITE_SAME(
2211                 req->cmd.buf[0] == WRITE_SAME_10 ? 10 : 16, r->req.cmd.xfer);
2212         break;
2213     case FORMAT_UNIT:
2214         trace_scsi_disk_emulate_command_FORMAT_UNIT(r->req.cmd.xfer);
2215         break;
2216     default:
2217         trace_scsi_disk_emulate_command_UNKNOWN(buf[0],
2218                                                 scsi_command_name(buf[0]));
2219         scsi_check_condition(r, SENSE_CODE(INVALID_OPCODE));
2220         return 0;
2221     }
2222     assert(!r->req.aiocb);
2223     r->iov.iov_len = MIN(r->buflen, req->cmd.xfer);
2224     if (r->iov.iov_len == 0) {
2225         scsi_req_complete(&r->req, GOOD);
2226     }
2227     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2228         assert(r->iov.iov_len == req->cmd.xfer);
2229         return -r->iov.iov_len;
2230     } else {
2231         return r->iov.iov_len;
2232     }
2233 
2234 illegal_request:
2235     if (r->req.status == -1) {
2236         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2237     }
2238     return 0;
2239 
2240 illegal_lba:
2241     scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2242     return 0;
2243 }
2244 
2245 /* Execute a scsi command.  Returns the length of the data expected by the
2246    command.  This will be Positive for data transfers from the device
2247    (eg. disk reads), negative for transfers to the device (eg. disk writes),
2248    and zero if the command does not transfer any data.  */
2249 
2250 static int32_t scsi_disk_dma_command(SCSIRequest *req, uint8_t *buf)
2251 {
2252     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
2253     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2254     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
2255     uint32_t len;
2256     uint8_t command;
2257 
2258     command = buf[0];
2259 
2260     if (!blk_is_available(s->qdev.conf.blk)) {
2261         scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
2262         return 0;
2263     }
2264 
2265     len = scsi_data_cdb_xfer(r->req.cmd.buf);
2266     switch (command) {
2267     case READ_6:
2268     case READ_10:
2269     case READ_12:
2270     case READ_16:
2271         trace_scsi_disk_dma_command_READ(r->req.cmd.lba, len);
2272         /* Protection information is not supported.  For SCSI versions 2 and
2273          * older (as determined by snooping the guest's INQUIRY commands),
2274          * there is no RD/WR/VRPROTECT, so skip this check in these versions.
2275          */
2276         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2277             goto illegal_request;
2278         }
2279         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2280             goto illegal_lba;
2281         }
2282         r->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2283         r->sector_count = len * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2284         break;
2285     case WRITE_6:
2286     case WRITE_10:
2287     case WRITE_12:
2288     case WRITE_16:
2289     case WRITE_VERIFY_10:
2290     case WRITE_VERIFY_12:
2291     case WRITE_VERIFY_16:
2292         if (!blk_is_writable(s->qdev.conf.blk)) {
2293             scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
2294             return 0;
2295         }
2296         trace_scsi_disk_dma_command_WRITE(
2297                 (command & 0xe) == 0xe ? "And Verify " : "",
2298                 r->req.cmd.lba, len);
2299         /* fall through */
2300     case VERIFY_10:
2301     case VERIFY_12:
2302     case VERIFY_16:
2303         /* We get here only for BYTCHK == 0x01 and only for scsi-block.
2304          * As far as DMA is concerned, we can treat it the same as a write;
2305          * scsi_block_do_sgio will send VERIFY commands.
2306          */
2307         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2308             goto illegal_request;
2309         }
2310         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2311             goto illegal_lba;
2312         }
2313         r->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2314         r->sector_count = len * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2315         break;
2316     default:
2317         abort();
2318     illegal_request:
2319         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2320         return 0;
2321     illegal_lba:
2322         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2323         return 0;
2324     }
2325     r->need_fua_emulation = sdc->need_fua_emulation(&r->req.cmd);
2326     if (r->sector_count == 0) {
2327         scsi_req_complete(&r->req, GOOD);
2328     }
2329     assert(r->iov.iov_len == 0);
2330     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2331         return -r->sector_count * BDRV_SECTOR_SIZE;
2332     } else {
2333         return r->sector_count * BDRV_SECTOR_SIZE;
2334     }
2335 }
2336 
2337 static void scsi_disk_reset(DeviceState *dev)
2338 {
2339     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev.qdev, dev);
2340     uint64_t nb_sectors;
2341 
2342     scsi_device_purge_requests(&s->qdev, SENSE_CODE(RESET));
2343 
2344     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2345 
2346     nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2347     if (nb_sectors) {
2348         nb_sectors--;
2349     }
2350     s->qdev.max_lba = nb_sectors;
2351     /* reset tray statuses */
2352     s->tray_locked = 0;
2353     s->tray_open = 0;
2354 
2355     s->qdev.scsi_version = s->qdev.default_scsi_version;
2356 }
2357 
2358 static void scsi_disk_drained_begin(void *opaque)
2359 {
2360     SCSIDiskState *s = opaque;
2361 
2362     scsi_device_drained_begin(&s->qdev);
2363 }
2364 
2365 static void scsi_disk_drained_end(void *opaque)
2366 {
2367     SCSIDiskState *s = opaque;
2368 
2369     scsi_device_drained_end(&s->qdev);
2370 }
2371 
2372 static void scsi_disk_resize_cb(void *opaque)
2373 {
2374     SCSIDiskState *s = opaque;
2375 
2376     /* SPC lists this sense code as available only for
2377      * direct-access devices.
2378      */
2379     if (s->qdev.type == TYPE_DISK) {
2380         scsi_device_report_change(&s->qdev, SENSE_CODE(CAPACITY_CHANGED));
2381     }
2382 }
2383 
2384 static void scsi_cd_change_media_cb(void *opaque, bool load, Error **errp)
2385 {
2386     SCSIDiskState *s = opaque;
2387 
2388     /*
2389      * When a CD gets changed, we have to report an ejected state and
2390      * then a loaded state to guests so that they detect tray
2391      * open/close and media change events.  Guests that do not use
2392      * GET_EVENT_STATUS_NOTIFICATION to detect such tray open/close
2393      * states rely on this behavior.
2394      *
2395      * media_changed governs the state machine used for unit attention
2396      * report.  media_event is used by GET EVENT STATUS NOTIFICATION.
2397      */
2398     s->media_changed = load;
2399     s->tray_open = !load;
2400     scsi_device_set_ua(&s->qdev, SENSE_CODE(UNIT_ATTENTION_NO_MEDIUM));
2401     s->media_event = true;
2402     s->eject_request = false;
2403 }
2404 
2405 static void scsi_cd_eject_request_cb(void *opaque, bool force)
2406 {
2407     SCSIDiskState *s = opaque;
2408 
2409     s->eject_request = true;
2410     if (force) {
2411         s->tray_locked = false;
2412     }
2413 }
2414 
2415 static bool scsi_cd_is_tray_open(void *opaque)
2416 {
2417     return ((SCSIDiskState *)opaque)->tray_open;
2418 }
2419 
2420 static bool scsi_cd_is_medium_locked(void *opaque)
2421 {
2422     return ((SCSIDiskState *)opaque)->tray_locked;
2423 }
2424 
2425 static const BlockDevOps scsi_disk_removable_block_ops = {
2426     .change_media_cb  = scsi_cd_change_media_cb,
2427     .drained_begin    = scsi_disk_drained_begin,
2428     .drained_end      = scsi_disk_drained_end,
2429     .eject_request_cb = scsi_cd_eject_request_cb,
2430     .is_medium_locked = scsi_cd_is_medium_locked,
2431     .is_tray_open     = scsi_cd_is_tray_open,
2432     .resize_cb        = scsi_disk_resize_cb,
2433 };
2434 
2435 static const BlockDevOps scsi_disk_block_ops = {
2436     .drained_begin = scsi_disk_drained_begin,
2437     .drained_end   = scsi_disk_drained_end,
2438     .resize_cb     = scsi_disk_resize_cb,
2439 };
2440 
2441 static void scsi_disk_unit_attention_reported(SCSIDevice *dev)
2442 {
2443     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2444     if (s->media_changed) {
2445         s->media_changed = false;
2446         scsi_device_set_ua(&s->qdev, SENSE_CODE(MEDIUM_CHANGED));
2447     }
2448 }
2449 
2450 static void scsi_realize(SCSIDevice *dev, Error **errp)
2451 {
2452     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2453     bool read_only;
2454 
2455     if (!s->qdev.conf.blk) {
2456         error_setg(errp, "drive property not set");
2457         return;
2458     }
2459 
2460     if (!(s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2461         !blk_is_inserted(s->qdev.conf.blk)) {
2462         error_setg(errp, "Device needs media, but drive is empty");
2463         return;
2464     }
2465 
2466     if (!blkconf_blocksizes(&s->qdev.conf, errp)) {
2467         return;
2468     }
2469 
2470     if (blk_get_aio_context(s->qdev.conf.blk) != qemu_get_aio_context() &&
2471         !s->qdev.hba_supports_iothread)
2472     {
2473         error_setg(errp, "HBA does not support iothreads");
2474         return;
2475     }
2476 
2477     if (dev->type == TYPE_DISK) {
2478         if (!blkconf_geometry(&dev->conf, NULL, 65535, 255, 255, errp)) {
2479             return;
2480         }
2481     }
2482 
2483     read_only = !blk_supports_write_perm(s->qdev.conf.blk);
2484     if (dev->type == TYPE_ROM) {
2485         read_only = true;
2486     }
2487 
2488     if (!blkconf_apply_backend_options(&dev->conf, read_only,
2489                                        dev->type == TYPE_DISK, errp)) {
2490         return;
2491     }
2492 
2493     if (s->qdev.conf.discard_granularity == -1) {
2494         s->qdev.conf.discard_granularity =
2495             MAX(s->qdev.conf.logical_block_size, DEFAULT_DISCARD_GRANULARITY);
2496     }
2497 
2498     if (!s->version) {
2499         s->version = g_strdup(qemu_hw_version());
2500     }
2501     if (!s->vendor) {
2502         s->vendor = g_strdup("QEMU");
2503     }
2504     if (!s->device_id) {
2505         if (s->serial) {
2506             s->device_id = g_strdup_printf("%.20s", s->serial);
2507         } else {
2508             const char *str = blk_name(s->qdev.conf.blk);
2509             if (str && *str) {
2510                 s->device_id = g_strdup(str);
2511             }
2512         }
2513     }
2514 
2515     if (blk_is_sg(s->qdev.conf.blk)) {
2516         error_setg(errp, "unwanted /dev/sg*");
2517         return;
2518     }
2519 
2520     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2521             !(s->features & (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS))) {
2522         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_removable_block_ops, s);
2523     } else {
2524         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_block_ops, s);
2525     }
2526 
2527     blk_iostatus_enable(s->qdev.conf.blk);
2528 
2529     add_boot_device_lchs(&dev->qdev, NULL,
2530                          dev->conf.lcyls,
2531                          dev->conf.lheads,
2532                          dev->conf.lsecs);
2533 }
2534 
2535 static void scsi_unrealize(SCSIDevice *dev)
2536 {
2537     del_boot_device_lchs(&dev->qdev, NULL);
2538 }
2539 
2540 static void scsi_hd_realize(SCSIDevice *dev, Error **errp)
2541 {
2542     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2543 
2544     /* can happen for devices without drive. The error message for missing
2545      * backend will be issued in scsi_realize
2546      */
2547     if (s->qdev.conf.blk) {
2548         if (!blkconf_blocksizes(&s->qdev.conf, errp)) {
2549             return;
2550         }
2551     }
2552     s->qdev.blocksize = s->qdev.conf.logical_block_size;
2553     s->qdev.type = TYPE_DISK;
2554     if (!s->product) {
2555         s->product = g_strdup("QEMU HARDDISK");
2556     }
2557     scsi_realize(&s->qdev, errp);
2558 }
2559 
2560 static void scsi_cd_realize(SCSIDevice *dev, Error **errp)
2561 {
2562     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2563     int ret;
2564     uint32_t blocksize = 2048;
2565 
2566     if (!dev->conf.blk) {
2567         /* Anonymous BlockBackend for an empty drive. As we put it into
2568          * dev->conf, qdev takes care of detaching on unplug. */
2569         dev->conf.blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
2570         ret = blk_attach_dev(dev->conf.blk, &dev->qdev);
2571         assert(ret == 0);
2572     }
2573 
2574     if (dev->conf.physical_block_size != 0) {
2575         blocksize = dev->conf.physical_block_size;
2576     }
2577 
2578     s->qdev.blocksize = blocksize;
2579     s->qdev.type = TYPE_ROM;
2580     s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2581     if (!s->product) {
2582         s->product = g_strdup("QEMU CD-ROM");
2583     }
2584     scsi_realize(&s->qdev, errp);
2585 }
2586 
2587 
2588 static const SCSIReqOps scsi_disk_emulate_reqops = {
2589     .size         = sizeof(SCSIDiskReq),
2590     .free_req     = scsi_free_request,
2591     .send_command = scsi_disk_emulate_command,
2592     .read_data    = scsi_disk_emulate_read_data,
2593     .write_data   = scsi_disk_emulate_write_data,
2594     .get_buf      = scsi_get_buf,
2595 };
2596 
2597 static const SCSIReqOps scsi_disk_dma_reqops = {
2598     .size         = sizeof(SCSIDiskReq),
2599     .free_req     = scsi_free_request,
2600     .send_command = scsi_disk_dma_command,
2601     .read_data    = scsi_read_data,
2602     .write_data   = scsi_write_data,
2603     .get_buf      = scsi_get_buf,
2604     .load_request = scsi_disk_load_request,
2605     .save_request = scsi_disk_save_request,
2606 };
2607 
2608 static const SCSIReqOps *const scsi_disk_reqops_dispatch[256] = {
2609     [TEST_UNIT_READY]                 = &scsi_disk_emulate_reqops,
2610     [INQUIRY]                         = &scsi_disk_emulate_reqops,
2611     [MODE_SENSE]                      = &scsi_disk_emulate_reqops,
2612     [MODE_SENSE_10]                   = &scsi_disk_emulate_reqops,
2613     [START_STOP]                      = &scsi_disk_emulate_reqops,
2614     [ALLOW_MEDIUM_REMOVAL]            = &scsi_disk_emulate_reqops,
2615     [READ_CAPACITY_10]                = &scsi_disk_emulate_reqops,
2616     [READ_TOC]                        = &scsi_disk_emulate_reqops,
2617     [READ_DVD_STRUCTURE]              = &scsi_disk_emulate_reqops,
2618     [READ_DISC_INFORMATION]           = &scsi_disk_emulate_reqops,
2619     [GET_CONFIGURATION]               = &scsi_disk_emulate_reqops,
2620     [GET_EVENT_STATUS_NOTIFICATION]   = &scsi_disk_emulate_reqops,
2621     [MECHANISM_STATUS]                = &scsi_disk_emulate_reqops,
2622     [SERVICE_ACTION_IN_16]            = &scsi_disk_emulate_reqops,
2623     [REQUEST_SENSE]                   = &scsi_disk_emulate_reqops,
2624     [SYNCHRONIZE_CACHE]               = &scsi_disk_emulate_reqops,
2625     [SEEK_10]                         = &scsi_disk_emulate_reqops,
2626     [MODE_SELECT]                     = &scsi_disk_emulate_reqops,
2627     [MODE_SELECT_10]                  = &scsi_disk_emulate_reqops,
2628     [UNMAP]                           = &scsi_disk_emulate_reqops,
2629     [WRITE_SAME_10]                   = &scsi_disk_emulate_reqops,
2630     [WRITE_SAME_16]                   = &scsi_disk_emulate_reqops,
2631     [VERIFY_10]                       = &scsi_disk_emulate_reqops,
2632     [VERIFY_12]                       = &scsi_disk_emulate_reqops,
2633     [VERIFY_16]                       = &scsi_disk_emulate_reqops,
2634     [FORMAT_UNIT]                     = &scsi_disk_emulate_reqops,
2635 
2636     [READ_6]                          = &scsi_disk_dma_reqops,
2637     [READ_10]                         = &scsi_disk_dma_reqops,
2638     [READ_12]                         = &scsi_disk_dma_reqops,
2639     [READ_16]                         = &scsi_disk_dma_reqops,
2640     [WRITE_6]                         = &scsi_disk_dma_reqops,
2641     [WRITE_10]                        = &scsi_disk_dma_reqops,
2642     [WRITE_12]                        = &scsi_disk_dma_reqops,
2643     [WRITE_16]                        = &scsi_disk_dma_reqops,
2644     [WRITE_VERIFY_10]                 = &scsi_disk_dma_reqops,
2645     [WRITE_VERIFY_12]                 = &scsi_disk_dma_reqops,
2646     [WRITE_VERIFY_16]                 = &scsi_disk_dma_reqops,
2647 };
2648 
2649 static void scsi_disk_new_request_dump(uint32_t lun, uint32_t tag, uint8_t *buf)
2650 {
2651     int i;
2652     int len = scsi_cdb_length(buf);
2653     char *line_buffer, *p;
2654 
2655     assert(len > 0 && len <= 16);
2656     line_buffer = g_malloc(len * 5 + 1);
2657 
2658     for (i = 0, p = line_buffer; i < len; i++) {
2659         p += sprintf(p, " 0x%02x", buf[i]);
2660     }
2661     trace_scsi_disk_new_request(lun, tag, line_buffer);
2662 
2663     g_free(line_buffer);
2664 }
2665 
2666 static SCSIRequest *scsi_new_request(SCSIDevice *d, uint32_t tag, uint32_t lun,
2667                                      uint8_t *buf, void *hba_private)
2668 {
2669     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2670     SCSIRequest *req;
2671     const SCSIReqOps *ops;
2672     uint8_t command;
2673 
2674     command = buf[0];
2675     ops = scsi_disk_reqops_dispatch[command];
2676     if (!ops) {
2677         ops = &scsi_disk_emulate_reqops;
2678     }
2679     req = scsi_req_alloc(ops, &s->qdev, tag, lun, hba_private);
2680 
2681     if (trace_event_get_state_backends(TRACE_SCSI_DISK_NEW_REQUEST)) {
2682         scsi_disk_new_request_dump(lun, tag, buf);
2683     }
2684 
2685     return req;
2686 }
2687 
2688 #ifdef __linux__
2689 static int get_device_type(SCSIDiskState *s)
2690 {
2691     uint8_t cmd[16];
2692     uint8_t buf[36];
2693     int ret;
2694 
2695     memset(cmd, 0, sizeof(cmd));
2696     memset(buf, 0, sizeof(buf));
2697     cmd[0] = INQUIRY;
2698     cmd[4] = sizeof(buf);
2699 
2700     ret = scsi_SG_IO_FROM_DEV(s->qdev.conf.blk, cmd, sizeof(cmd),
2701                               buf, sizeof(buf), s->qdev.io_timeout);
2702     if (ret < 0) {
2703         return -1;
2704     }
2705     s->qdev.type = buf[0];
2706     if (buf[1] & 0x80) {
2707         s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2708     }
2709     return 0;
2710 }
2711 
2712 static void scsi_block_realize(SCSIDevice *dev, Error **errp)
2713 {
2714     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2715     int sg_version;
2716     int rc;
2717 
2718     if (!s->qdev.conf.blk) {
2719         error_setg(errp, "drive property not set");
2720         return;
2721     }
2722 
2723     if (s->rotation_rate) {
2724         error_report_once("rotation_rate is specified for scsi-block but is "
2725                           "not implemented. This option is deprecated and will "
2726                           "be removed in a future version");
2727     }
2728 
2729     /* check we are using a driver managing SG_IO (version 3 and after) */
2730     rc = blk_ioctl(s->qdev.conf.blk, SG_GET_VERSION_NUM, &sg_version);
2731     if (rc < 0) {
2732         error_setg_errno(errp, -rc, "cannot get SG_IO version number");
2733         if (rc != -EPERM) {
2734             error_append_hint(errp, "Is this a SCSI device?\n");
2735         }
2736         return;
2737     }
2738     if (sg_version < 30000) {
2739         error_setg(errp, "scsi generic interface too old");
2740         return;
2741     }
2742 
2743     /* get device type from INQUIRY data */
2744     rc = get_device_type(s);
2745     if (rc < 0) {
2746         error_setg(errp, "INQUIRY failed");
2747         return;
2748     }
2749 
2750     /* Make a guess for the block size, we'll fix it when the guest sends.
2751      * READ CAPACITY.  If they don't, they likely would assume these sizes
2752      * anyway. (TODO: check in /sys).
2753      */
2754     if (s->qdev.type == TYPE_ROM || s->qdev.type == TYPE_WORM) {
2755         s->qdev.blocksize = 2048;
2756     } else {
2757         s->qdev.blocksize = 512;
2758     }
2759 
2760     /* Makes the scsi-block device not removable by using HMP and QMP eject
2761      * command.
2762      */
2763     s->features |= (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS);
2764 
2765     scsi_realize(&s->qdev, errp);
2766     scsi_generic_read_device_inquiry(&s->qdev);
2767 }
2768 
2769 typedef struct SCSIBlockReq {
2770     SCSIDiskReq req;
2771     sg_io_hdr_t io_header;
2772 
2773     /* Selected bytes of the original CDB, copied into our own CDB.  */
2774     uint8_t cmd, cdb1, group_number;
2775 
2776     /* CDB passed to SG_IO.  */
2777     uint8_t cdb[16];
2778     BlockCompletionFunc *cb;
2779     void *cb_opaque;
2780 } SCSIBlockReq;
2781 
2782 static void scsi_block_sgio_complete(void *opaque, int ret)
2783 {
2784     SCSIBlockReq *req = (SCSIBlockReq *)opaque;
2785     SCSIDiskReq *r = &req->req;
2786     sg_io_hdr_t *io_hdr = &req->io_header;
2787 
2788     if (ret == 0) {
2789         if (io_hdr->host_status != SCSI_HOST_OK) {
2790             scsi_req_complete_failed(&r->req, io_hdr->host_status);
2791             scsi_req_unref(&r->req);
2792             return;
2793         }
2794 
2795         if (io_hdr->driver_status & SG_ERR_DRIVER_TIMEOUT) {
2796             ret = BUSY;
2797         } else {
2798             ret = io_hdr->status;
2799         }
2800 
2801         if (ret > 0) {
2802             if (scsi_handle_rw_error(r, ret, true)) {
2803                 scsi_req_unref(&r->req);
2804                 return;
2805             }
2806 
2807             /* Ignore error.  */
2808             ret = 0;
2809         }
2810     }
2811 
2812     req->cb(req->cb_opaque, ret);
2813 }
2814 
2815 static BlockAIOCB *scsi_block_do_sgio(SCSIBlockReq *req,
2816                                       int64_t offset, QEMUIOVector *iov,
2817                                       int direction,
2818                                       BlockCompletionFunc *cb, void *opaque)
2819 {
2820     sg_io_hdr_t *io_header = &req->io_header;
2821     SCSIDiskReq *r = &req->req;
2822     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2823     int nb_logical_blocks;
2824     uint64_t lba;
2825     BlockAIOCB *aiocb;
2826 
2827     /* This is not supported yet.  It can only happen if the guest does
2828      * reads and writes that are not aligned to one logical sectors
2829      * _and_ cover multiple MemoryRegions.
2830      */
2831     assert(offset % s->qdev.blocksize == 0);
2832     assert(iov->size % s->qdev.blocksize == 0);
2833 
2834     io_header->interface_id = 'S';
2835 
2836     /* The data transfer comes from the QEMUIOVector.  */
2837     io_header->dxfer_direction = direction;
2838     io_header->dxfer_len = iov->size;
2839     io_header->dxferp = (void *)iov->iov;
2840     io_header->iovec_count = iov->niov;
2841     assert(io_header->iovec_count == iov->niov); /* no overflow! */
2842 
2843     /* Build a new CDB with the LBA and length patched in, in case
2844      * DMA helpers split the transfer in multiple segments.  Do not
2845      * build a CDB smaller than what the guest wanted, and only build
2846      * a larger one if strictly necessary.
2847      */
2848     io_header->cmdp = req->cdb;
2849     lba = offset / s->qdev.blocksize;
2850     nb_logical_blocks = io_header->dxfer_len / s->qdev.blocksize;
2851 
2852     if ((req->cmd >> 5) == 0 && lba <= 0x1ffff) {
2853         /* 6-byte CDB */
2854         stl_be_p(&req->cdb[0], lba | (req->cmd << 24));
2855         req->cdb[4] = nb_logical_blocks;
2856         req->cdb[5] = 0;
2857         io_header->cmd_len = 6;
2858     } else if ((req->cmd >> 5) <= 1 && lba <= 0xffffffffULL) {
2859         /* 10-byte CDB */
2860         req->cdb[0] = (req->cmd & 0x1f) | 0x20;
2861         req->cdb[1] = req->cdb1;
2862         stl_be_p(&req->cdb[2], lba);
2863         req->cdb[6] = req->group_number;
2864         stw_be_p(&req->cdb[7], nb_logical_blocks);
2865         req->cdb[9] = 0;
2866         io_header->cmd_len = 10;
2867     } else if ((req->cmd >> 5) != 4 && lba <= 0xffffffffULL) {
2868         /* 12-byte CDB */
2869         req->cdb[0] = (req->cmd & 0x1f) | 0xA0;
2870         req->cdb[1] = req->cdb1;
2871         stl_be_p(&req->cdb[2], lba);
2872         stl_be_p(&req->cdb[6], nb_logical_blocks);
2873         req->cdb[10] = req->group_number;
2874         req->cdb[11] = 0;
2875         io_header->cmd_len = 12;
2876     } else {
2877         /* 16-byte CDB */
2878         req->cdb[0] = (req->cmd & 0x1f) | 0x80;
2879         req->cdb[1] = req->cdb1;
2880         stq_be_p(&req->cdb[2], lba);
2881         stl_be_p(&req->cdb[10], nb_logical_blocks);
2882         req->cdb[14] = req->group_number;
2883         req->cdb[15] = 0;
2884         io_header->cmd_len = 16;
2885     }
2886 
2887     /* The rest is as in scsi-generic.c.  */
2888     io_header->mx_sb_len = sizeof(r->req.sense);
2889     io_header->sbp = r->req.sense;
2890     io_header->timeout = s->qdev.io_timeout * 1000;
2891     io_header->usr_ptr = r;
2892     io_header->flags |= SG_FLAG_DIRECT_IO;
2893     req->cb = cb;
2894     req->cb_opaque = opaque;
2895     trace_scsi_disk_aio_sgio_command(r->req.tag, req->cdb[0], lba,
2896                                      nb_logical_blocks, io_header->timeout);
2897     aiocb = blk_aio_ioctl(s->qdev.conf.blk, SG_IO, io_header, scsi_block_sgio_complete, req);
2898     assert(aiocb != NULL);
2899     return aiocb;
2900 }
2901 
2902 static bool scsi_block_no_fua(SCSICommand *cmd)
2903 {
2904     return false;
2905 }
2906 
2907 static BlockAIOCB *scsi_block_dma_readv(int64_t offset,
2908                                         QEMUIOVector *iov,
2909                                         BlockCompletionFunc *cb, void *cb_opaque,
2910                                         void *opaque)
2911 {
2912     SCSIBlockReq *r = opaque;
2913     return scsi_block_do_sgio(r, offset, iov,
2914                               SG_DXFER_FROM_DEV, cb, cb_opaque);
2915 }
2916 
2917 static BlockAIOCB *scsi_block_dma_writev(int64_t offset,
2918                                          QEMUIOVector *iov,
2919                                          BlockCompletionFunc *cb, void *cb_opaque,
2920                                          void *opaque)
2921 {
2922     SCSIBlockReq *r = opaque;
2923     return scsi_block_do_sgio(r, offset, iov,
2924                               SG_DXFER_TO_DEV, cb, cb_opaque);
2925 }
2926 
2927 static bool scsi_block_is_passthrough(SCSIDiskState *s, uint8_t *buf)
2928 {
2929     switch (buf[0]) {
2930     case VERIFY_10:
2931     case VERIFY_12:
2932     case VERIFY_16:
2933         /* Check if BYTCHK == 0x01 (data-out buffer contains data
2934          * for the number of logical blocks specified in the length
2935          * field).  For other modes, do not use scatter/gather operation.
2936          */
2937         if ((buf[1] & 6) == 2) {
2938             return false;
2939         }
2940         break;
2941 
2942     case READ_6:
2943     case READ_10:
2944     case READ_12:
2945     case READ_16:
2946     case WRITE_6:
2947     case WRITE_10:
2948     case WRITE_12:
2949     case WRITE_16:
2950     case WRITE_VERIFY_10:
2951     case WRITE_VERIFY_12:
2952     case WRITE_VERIFY_16:
2953         /* MMC writing cannot be done via DMA helpers, because it sometimes
2954          * involves writing beyond the maximum LBA or to negative LBA (lead-in).
2955          * We might use scsi_block_dma_reqops as long as no writing commands are
2956          * seen, but performance usually isn't paramount on optical media.  So,
2957          * just make scsi-block operate the same as scsi-generic for them.
2958          */
2959         if (s->qdev.type != TYPE_ROM) {
2960             return false;
2961         }
2962         break;
2963 
2964     default:
2965         break;
2966     }
2967 
2968     return true;
2969 }
2970 
2971 
2972 static int32_t scsi_block_dma_command(SCSIRequest *req, uint8_t *buf)
2973 {
2974     SCSIBlockReq *r = (SCSIBlockReq *)req;
2975     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2976 
2977     r->cmd = req->cmd.buf[0];
2978     switch (r->cmd >> 5) {
2979     case 0:
2980         /* 6-byte CDB.  */
2981         r->cdb1 = r->group_number = 0;
2982         break;
2983     case 1:
2984         /* 10-byte CDB.  */
2985         r->cdb1 = req->cmd.buf[1];
2986         r->group_number = req->cmd.buf[6];
2987         break;
2988     case 4:
2989         /* 12-byte CDB.  */
2990         r->cdb1 = req->cmd.buf[1];
2991         r->group_number = req->cmd.buf[10];
2992         break;
2993     case 5:
2994         /* 16-byte CDB.  */
2995         r->cdb1 = req->cmd.buf[1];
2996         r->group_number = req->cmd.buf[14];
2997         break;
2998     default:
2999         abort();
3000     }
3001 
3002     /* Protection information is not supported.  For SCSI versions 2 and
3003      * older (as determined by snooping the guest's INQUIRY commands),
3004      * there is no RD/WR/VRPROTECT, so skip this check in these versions.
3005      */
3006     if (s->qdev.scsi_version > 2 && (req->cmd.buf[1] & 0xe0)) {
3007         scsi_check_condition(&r->req, SENSE_CODE(INVALID_FIELD));
3008         return 0;
3009     }
3010 
3011     return scsi_disk_dma_command(req, buf);
3012 }
3013 
3014 static const SCSIReqOps scsi_block_dma_reqops = {
3015     .size         = sizeof(SCSIBlockReq),
3016     .free_req     = scsi_free_request,
3017     .send_command = scsi_block_dma_command,
3018     .read_data    = scsi_read_data,
3019     .write_data   = scsi_write_data,
3020     .get_buf      = scsi_get_buf,
3021     .load_request = scsi_disk_load_request,
3022     .save_request = scsi_disk_save_request,
3023 };
3024 
3025 static SCSIRequest *scsi_block_new_request(SCSIDevice *d, uint32_t tag,
3026                                            uint32_t lun, uint8_t *buf,
3027                                            void *hba_private)
3028 {
3029     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
3030 
3031     if (scsi_block_is_passthrough(s, buf)) {
3032         return scsi_req_alloc(&scsi_generic_req_ops, &s->qdev, tag, lun,
3033                               hba_private);
3034     } else {
3035         return scsi_req_alloc(&scsi_block_dma_reqops, &s->qdev, tag, lun,
3036                               hba_private);
3037     }
3038 }
3039 
3040 static int scsi_block_parse_cdb(SCSIDevice *d, SCSICommand *cmd,
3041                                   uint8_t *buf, size_t buf_len,
3042                                   void *hba_private)
3043 {
3044     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
3045 
3046     if (scsi_block_is_passthrough(s, buf)) {
3047         return scsi_bus_parse_cdb(&s->qdev, cmd, buf, buf_len, hba_private);
3048     } else {
3049         return scsi_req_parse_cdb(&s->qdev, cmd, buf, buf_len);
3050     }
3051 }
3052 
3053 static void scsi_block_update_sense(SCSIRequest *req)
3054 {
3055     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
3056     SCSIBlockReq *br = DO_UPCAST(SCSIBlockReq, req, r);
3057     r->req.sense_len = MIN(br->io_header.sb_len_wr, sizeof(r->req.sense));
3058 }
3059 #endif
3060 
3061 static
3062 BlockAIOCB *scsi_dma_readv(int64_t offset, QEMUIOVector *iov,
3063                            BlockCompletionFunc *cb, void *cb_opaque,
3064                            void *opaque)
3065 {
3066     SCSIDiskReq *r = opaque;
3067     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
3068     return blk_aio_preadv(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
3069 }
3070 
3071 static
3072 BlockAIOCB *scsi_dma_writev(int64_t offset, QEMUIOVector *iov,
3073                             BlockCompletionFunc *cb, void *cb_opaque,
3074                             void *opaque)
3075 {
3076     SCSIDiskReq *r = opaque;
3077     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
3078     return blk_aio_pwritev(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
3079 }
3080 
3081 static void scsi_disk_base_class_initfn(ObjectClass *klass, void *data)
3082 {
3083     DeviceClass *dc = DEVICE_CLASS(klass);
3084     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
3085 
3086     dc->fw_name = "disk";
3087     dc->reset = scsi_disk_reset;
3088     sdc->dma_readv = scsi_dma_readv;
3089     sdc->dma_writev = scsi_dma_writev;
3090     sdc->need_fua_emulation = scsi_is_cmd_fua;
3091 }
3092 
3093 static const TypeInfo scsi_disk_base_info = {
3094     .name          = TYPE_SCSI_DISK_BASE,
3095     .parent        = TYPE_SCSI_DEVICE,
3096     .class_init    = scsi_disk_base_class_initfn,
3097     .instance_size = sizeof(SCSIDiskState),
3098     .class_size    = sizeof(SCSIDiskClass),
3099     .abstract      = true,
3100 };
3101 
3102 #define DEFINE_SCSI_DISK_PROPERTIES()                                   \
3103     DEFINE_PROP_DRIVE_IOTHREAD("drive", SCSIDiskState, qdev.conf.blk),  \
3104     DEFINE_BLOCK_PROPERTIES_BASE(SCSIDiskState, qdev.conf),             \
3105     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),            \
3106     DEFINE_PROP_STRING("ver", SCSIDiskState, version),                  \
3107     DEFINE_PROP_STRING("serial", SCSIDiskState, serial),                \
3108     DEFINE_PROP_STRING("vendor", SCSIDiskState, vendor),                \
3109     DEFINE_PROP_STRING("product", SCSIDiskState, product),              \
3110     DEFINE_PROP_STRING("device_id", SCSIDiskState, device_id)
3111 
3112 
3113 static Property scsi_hd_properties[] = {
3114     DEFINE_SCSI_DISK_PROPERTIES(),
3115     DEFINE_PROP_BIT("removable", SCSIDiskState, features,
3116                     SCSI_DISK_F_REMOVABLE, false),
3117     DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
3118                     SCSI_DISK_F_DPOFUA, false),
3119     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3120     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3121     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3122     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3123                        DEFAULT_MAX_UNMAP_SIZE),
3124     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3125                        DEFAULT_MAX_IO_SIZE),
3126     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3127     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3128                       5),
3129     DEFINE_PROP_BIT("quirk_mode_page_vendor_specific_apple", SCSIDiskState,
3130                     quirks, SCSI_DISK_QUIRK_MODE_PAGE_VENDOR_SPECIFIC_APPLE,
3131                     0),
3132     DEFINE_BLOCK_CHS_PROPERTIES(SCSIDiskState, qdev.conf),
3133     DEFINE_PROP_END_OF_LIST(),
3134 };
3135 
3136 static const VMStateDescription vmstate_scsi_disk_state = {
3137     .name = "scsi-disk",
3138     .version_id = 1,
3139     .minimum_version_id = 1,
3140     .fields = (const VMStateField[]) {
3141         VMSTATE_SCSI_DEVICE(qdev, SCSIDiskState),
3142         VMSTATE_BOOL(media_changed, SCSIDiskState),
3143         VMSTATE_BOOL(media_event, SCSIDiskState),
3144         VMSTATE_BOOL(eject_request, SCSIDiskState),
3145         VMSTATE_BOOL(tray_open, SCSIDiskState),
3146         VMSTATE_BOOL(tray_locked, SCSIDiskState),
3147         VMSTATE_END_OF_LIST()
3148     }
3149 };
3150 
3151 static void scsi_hd_class_initfn(ObjectClass *klass, void *data)
3152 {
3153     DeviceClass *dc = DEVICE_CLASS(klass);
3154     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3155 
3156     sc->realize      = scsi_hd_realize;
3157     sc->unrealize    = scsi_unrealize;
3158     sc->alloc_req    = scsi_new_request;
3159     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3160     dc->desc = "virtual SCSI disk";
3161     device_class_set_props(dc, scsi_hd_properties);
3162     dc->vmsd  = &vmstate_scsi_disk_state;
3163 }
3164 
3165 static const TypeInfo scsi_hd_info = {
3166     .name          = "scsi-hd",
3167     .parent        = TYPE_SCSI_DISK_BASE,
3168     .class_init    = scsi_hd_class_initfn,
3169 };
3170 
3171 static Property scsi_cd_properties[] = {
3172     DEFINE_SCSI_DISK_PROPERTIES(),
3173     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3174     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3175     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3176     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3177                        DEFAULT_MAX_IO_SIZE),
3178     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3179                       5),
3180     DEFINE_PROP_BIT("quirk_mode_page_apple_vendor", SCSIDiskState, quirks,
3181                     SCSI_DISK_QUIRK_MODE_PAGE_APPLE_VENDOR, 0),
3182     DEFINE_PROP_BIT("quirk_mode_sense_rom_use_dbd", SCSIDiskState, quirks,
3183                     SCSI_DISK_QUIRK_MODE_SENSE_ROM_USE_DBD, 0),
3184     DEFINE_PROP_BIT("quirk_mode_page_vendor_specific_apple", SCSIDiskState,
3185                     quirks, SCSI_DISK_QUIRK_MODE_PAGE_VENDOR_SPECIFIC_APPLE,
3186                     0),
3187     DEFINE_PROP_BIT("quirk_mode_page_truncated", SCSIDiskState, quirks,
3188                     SCSI_DISK_QUIRK_MODE_PAGE_TRUNCATED, 0),
3189     DEFINE_PROP_END_OF_LIST(),
3190 };
3191 
3192 static void scsi_cd_class_initfn(ObjectClass *klass, void *data)
3193 {
3194     DeviceClass *dc = DEVICE_CLASS(klass);
3195     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3196 
3197     sc->realize      = scsi_cd_realize;
3198     sc->alloc_req    = scsi_new_request;
3199     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3200     dc->desc = "virtual SCSI CD-ROM";
3201     device_class_set_props(dc, scsi_cd_properties);
3202     dc->vmsd  = &vmstate_scsi_disk_state;
3203 }
3204 
3205 static const TypeInfo scsi_cd_info = {
3206     .name          = "scsi-cd",
3207     .parent        = TYPE_SCSI_DISK_BASE,
3208     .class_init    = scsi_cd_class_initfn,
3209 };
3210 
3211 #ifdef __linux__
3212 static Property scsi_block_properties[] = {
3213     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),
3214     DEFINE_PROP_DRIVE("drive", SCSIDiskState, qdev.conf.blk),
3215     DEFINE_PROP_BOOL("share-rw", SCSIDiskState, qdev.conf.share_rw, false),
3216     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3217     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3218                        DEFAULT_MAX_UNMAP_SIZE),
3219     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3220                        DEFAULT_MAX_IO_SIZE),
3221     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3222                       -1),
3223     DEFINE_PROP_UINT32("io_timeout", SCSIDiskState, qdev.io_timeout,
3224                        DEFAULT_IO_TIMEOUT),
3225     DEFINE_PROP_END_OF_LIST(),
3226 };
3227 
3228 static void scsi_block_class_initfn(ObjectClass *klass, void *data)
3229 {
3230     DeviceClass *dc = DEVICE_CLASS(klass);
3231     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3232     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
3233 
3234     sc->realize      = scsi_block_realize;
3235     sc->alloc_req    = scsi_block_new_request;
3236     sc->parse_cdb    = scsi_block_parse_cdb;
3237     sdc->dma_readv   = scsi_block_dma_readv;
3238     sdc->dma_writev  = scsi_block_dma_writev;
3239     sdc->update_sense = scsi_block_update_sense;
3240     sdc->need_fua_emulation = scsi_block_no_fua;
3241     dc->desc = "SCSI block device passthrough";
3242     device_class_set_props(dc, scsi_block_properties);
3243     dc->vmsd  = &vmstate_scsi_disk_state;
3244 }
3245 
3246 static const TypeInfo scsi_block_info = {
3247     .name          = "scsi-block",
3248     .parent        = TYPE_SCSI_DISK_BASE,
3249     .class_init    = scsi_block_class_initfn,
3250 };
3251 #endif
3252 
3253 static void scsi_disk_register_types(void)
3254 {
3255     type_register_static(&scsi_disk_base_info);
3256     type_register_static(&scsi_hd_info);
3257     type_register_static(&scsi_cd_info);
3258 #ifdef __linux__
3259     type_register_static(&scsi_block_info);
3260 #endif
3261 }
3262 
3263 type_init(scsi_disk_register_types)
3264