xref: /openbmc/qemu/hw/scsi/scsi-disk.c (revision 071a6dba)
1 /*
2  * SCSI Device emulation
3  *
4  * Copyright (c) 2006 CodeSourcery.
5  * Based on code by Fabrice Bellard
6  *
7  * Written by Paul Brook
8  * Modifications:
9  *  2009-Dec-12 Artyom Tarasenko : implemented stamdard inquiry for the case
10  *                                 when the allocation length of CDB is smaller
11  *                                 than 36.
12  *  2009-Oct-13 Artyom Tarasenko : implemented the block descriptor in the
13  *                                 MODE SENSE response.
14  *
15  * This code is licensed under the LGPL.
16  *
17  * Note that this file only handles the SCSI architecture model and device
18  * commands.  Emulation of interface/link layer protocols is handled by
19  * the host adapter emulator.
20  */
21 
22 #include "qemu/osdep.h"
23 #include "qemu/units.h"
24 #include "qapi/error.h"
25 #include "qemu/error-report.h"
26 #include "qemu/main-loop.h"
27 #include "qemu/module.h"
28 #include "hw/scsi/scsi.h"
29 #include "migration/qemu-file-types.h"
30 #include "migration/vmstate.h"
31 #include "hw/scsi/emulation.h"
32 #include "scsi/constants.h"
33 #include "sysemu/block-backend.h"
34 #include "sysemu/blockdev.h"
35 #include "hw/block/block.h"
36 #include "hw/qdev-properties.h"
37 #include "sysemu/dma.h"
38 #include "sysemu/sysemu.h"
39 #include "qemu/cutils.h"
40 #include "trace.h"
41 
42 #ifdef __linux
43 #include <scsi/sg.h>
44 #endif
45 
46 #define SCSI_WRITE_SAME_MAX         (512 * KiB)
47 #define SCSI_DMA_BUF_SIZE           (128 * KiB)
48 #define SCSI_MAX_INQUIRY_LEN        256
49 #define SCSI_MAX_MODE_LEN           256
50 
51 #define DEFAULT_DISCARD_GRANULARITY (4 * KiB)
52 #define DEFAULT_MAX_UNMAP_SIZE      (1 * GiB)
53 #define DEFAULT_MAX_IO_SIZE         INT_MAX     /* 2 GB - 1 block */
54 
55 #define TYPE_SCSI_DISK_BASE         "scsi-disk-base"
56 
57 #define SCSI_DISK_BASE(obj) \
58      OBJECT_CHECK(SCSIDiskState, (obj), TYPE_SCSI_DISK_BASE)
59 #define SCSI_DISK_BASE_CLASS(klass) \
60      OBJECT_CLASS_CHECK(SCSIDiskClass, (klass), TYPE_SCSI_DISK_BASE)
61 #define SCSI_DISK_BASE_GET_CLASS(obj) \
62      OBJECT_GET_CLASS(SCSIDiskClass, (obj), TYPE_SCSI_DISK_BASE)
63 
64 typedef struct SCSIDiskClass {
65     SCSIDeviceClass parent_class;
66     DMAIOFunc       *dma_readv;
67     DMAIOFunc       *dma_writev;
68     bool            (*need_fua_emulation)(SCSICommand *cmd);
69     void            (*update_sense)(SCSIRequest *r);
70 } SCSIDiskClass;
71 
72 typedef struct SCSIDiskReq {
73     SCSIRequest req;
74     /* Both sector and sector_count are in terms of BDRV_SECTOR_SIZE bytes.  */
75     uint64_t sector;
76     uint32_t sector_count;
77     uint32_t buflen;
78     bool started;
79     bool need_fua_emulation;
80     struct iovec iov;
81     QEMUIOVector qiov;
82     BlockAcctCookie acct;
83     unsigned char *status;
84 } SCSIDiskReq;
85 
86 #define SCSI_DISK_F_REMOVABLE             0
87 #define SCSI_DISK_F_DPOFUA                1
88 #define SCSI_DISK_F_NO_REMOVABLE_DEVOPS   2
89 
90 typedef struct SCSIDiskState
91 {
92     SCSIDevice qdev;
93     uint32_t features;
94     bool media_changed;
95     bool media_event;
96     bool eject_request;
97     uint16_t port_index;
98     uint64_t max_unmap_size;
99     uint64_t max_io_size;
100     QEMUBH *bh;
101     char *version;
102     char *serial;
103     char *vendor;
104     char *product;
105     char *device_id;
106     bool tray_open;
107     bool tray_locked;
108     /*
109      * 0x0000        - rotation rate not reported
110      * 0x0001        - non-rotating medium (SSD)
111      * 0x0002-0x0400 - reserved
112      * 0x0401-0xffe  - rotations per minute
113      * 0xffff        - reserved
114      */
115     uint16_t rotation_rate;
116 } SCSIDiskState;
117 
118 static bool scsi_handle_rw_error(SCSIDiskReq *r, int error, bool acct_failed);
119 
120 static void scsi_free_request(SCSIRequest *req)
121 {
122     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
123 
124     qemu_vfree(r->iov.iov_base);
125 }
126 
127 /* Helper function for command completion with sense.  */
128 static void scsi_check_condition(SCSIDiskReq *r, SCSISense sense)
129 {
130     trace_scsi_disk_check_condition(r->req.tag, sense.key, sense.asc,
131                                     sense.ascq);
132     scsi_req_build_sense(&r->req, sense);
133     scsi_req_complete(&r->req, CHECK_CONDITION);
134 }
135 
136 static void scsi_init_iovec(SCSIDiskReq *r, size_t size)
137 {
138     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
139 
140     if (!r->iov.iov_base) {
141         r->buflen = size;
142         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
143     }
144     r->iov.iov_len = MIN(r->sector_count * BDRV_SECTOR_SIZE, r->buflen);
145     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
146 }
147 
148 static void scsi_disk_save_request(QEMUFile *f, SCSIRequest *req)
149 {
150     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
151 
152     qemu_put_be64s(f, &r->sector);
153     qemu_put_be32s(f, &r->sector_count);
154     qemu_put_be32s(f, &r->buflen);
155     if (r->buflen) {
156         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
157             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
158         } else if (!req->retry) {
159             uint32_t len = r->iov.iov_len;
160             qemu_put_be32s(f, &len);
161             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
162         }
163     }
164 }
165 
166 static void scsi_disk_load_request(QEMUFile *f, SCSIRequest *req)
167 {
168     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
169 
170     qemu_get_be64s(f, &r->sector);
171     qemu_get_be32s(f, &r->sector_count);
172     qemu_get_be32s(f, &r->buflen);
173     if (r->buflen) {
174         scsi_init_iovec(r, r->buflen);
175         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
176             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
177         } else if (!r->req.retry) {
178             uint32_t len;
179             qemu_get_be32s(f, &len);
180             r->iov.iov_len = len;
181             assert(r->iov.iov_len <= r->buflen);
182             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
183         }
184     }
185 
186     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
187 }
188 
189 static bool scsi_disk_req_check_error(SCSIDiskReq *r, int ret, bool acct_failed)
190 {
191     if (r->req.io_canceled) {
192         scsi_req_cancel_complete(&r->req);
193         return true;
194     }
195 
196     if (ret < 0 || (r->status && *r->status)) {
197         return scsi_handle_rw_error(r, -ret, acct_failed);
198     }
199 
200     return false;
201 }
202 
203 static void scsi_aio_complete(void *opaque, int ret)
204 {
205     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
206     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
207 
208     assert(r->req.aiocb != NULL);
209     r->req.aiocb = NULL;
210     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
211     if (scsi_disk_req_check_error(r, ret, true)) {
212         goto done;
213     }
214 
215     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
216     scsi_req_complete(&r->req, GOOD);
217 
218 done:
219     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
220     scsi_req_unref(&r->req);
221 }
222 
223 static bool scsi_is_cmd_fua(SCSICommand *cmd)
224 {
225     switch (cmd->buf[0]) {
226     case READ_10:
227     case READ_12:
228     case READ_16:
229     case WRITE_10:
230     case WRITE_12:
231     case WRITE_16:
232         return (cmd->buf[1] & 8) != 0;
233 
234     case VERIFY_10:
235     case VERIFY_12:
236     case VERIFY_16:
237     case WRITE_VERIFY_10:
238     case WRITE_VERIFY_12:
239     case WRITE_VERIFY_16:
240         return true;
241 
242     case READ_6:
243     case WRITE_6:
244     default:
245         return false;
246     }
247 }
248 
249 static void scsi_write_do_fua(SCSIDiskReq *r)
250 {
251     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
252 
253     assert(r->req.aiocb == NULL);
254     assert(!r->req.io_canceled);
255 
256     if (r->need_fua_emulation) {
257         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
258                          BLOCK_ACCT_FLUSH);
259         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
260         return;
261     }
262 
263     scsi_req_complete(&r->req, GOOD);
264     scsi_req_unref(&r->req);
265 }
266 
267 static void scsi_dma_complete_noio(SCSIDiskReq *r, int ret)
268 {
269     assert(r->req.aiocb == NULL);
270     if (scsi_disk_req_check_error(r, ret, false)) {
271         goto done;
272     }
273 
274     r->sector += r->sector_count;
275     r->sector_count = 0;
276     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
277         scsi_write_do_fua(r);
278         return;
279     } else {
280         scsi_req_complete(&r->req, GOOD);
281     }
282 
283 done:
284     scsi_req_unref(&r->req);
285 }
286 
287 static void scsi_dma_complete(void *opaque, int ret)
288 {
289     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
290     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
291 
292     assert(r->req.aiocb != NULL);
293     r->req.aiocb = NULL;
294 
295     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
296     if (ret < 0) {
297         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
298     } else {
299         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
300     }
301     scsi_dma_complete_noio(r, ret);
302     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
303 }
304 
305 static void scsi_read_complete_noio(SCSIDiskReq *r, int ret)
306 {
307     uint32_t n;
308 
309     assert(r->req.aiocb == NULL);
310     if (scsi_disk_req_check_error(r, ret, false)) {
311         goto done;
312     }
313 
314     n = r->qiov.size / BDRV_SECTOR_SIZE;
315     r->sector += n;
316     r->sector_count -= n;
317     scsi_req_data(&r->req, r->qiov.size);
318 
319 done:
320     scsi_req_unref(&r->req);
321 }
322 
323 static void scsi_read_complete(void *opaque, int ret)
324 {
325     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
326     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
327 
328     assert(r->req.aiocb != NULL);
329     r->req.aiocb = NULL;
330 
331     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
332     if (ret < 0) {
333         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
334     } else {
335         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
336         trace_scsi_disk_read_complete(r->req.tag, r->qiov.size);
337     }
338     scsi_read_complete_noio(r, ret);
339     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
340 }
341 
342 /* Actually issue a read to the block device.  */
343 static void scsi_do_read(SCSIDiskReq *r, int ret)
344 {
345     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
346     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
347 
348     assert (r->req.aiocb == NULL);
349     if (scsi_disk_req_check_error(r, ret, false)) {
350         goto done;
351     }
352 
353     /* The request is used as the AIO opaque value, so add a ref.  */
354     scsi_req_ref(&r->req);
355 
356     if (r->req.sg) {
357         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_READ);
358         r->req.resid -= r->req.sg->size;
359         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
360                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
361                                   BDRV_SECTOR_SIZE,
362                                   sdc->dma_readv, r, scsi_dma_complete, r,
363                                   DMA_DIRECTION_FROM_DEVICE);
364     } else {
365         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
366         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
367                          r->qiov.size, BLOCK_ACCT_READ);
368         r->req.aiocb = sdc->dma_readv(r->sector << BDRV_SECTOR_BITS, &r->qiov,
369                                       scsi_read_complete, r, r);
370     }
371 
372 done:
373     scsi_req_unref(&r->req);
374 }
375 
376 static void scsi_do_read_cb(void *opaque, int ret)
377 {
378     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
379     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
380 
381     assert (r->req.aiocb != NULL);
382     r->req.aiocb = NULL;
383 
384     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
385     if (ret < 0) {
386         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
387     } else {
388         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
389     }
390     scsi_do_read(opaque, ret);
391     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
392 }
393 
394 /* Read more data from scsi device into buffer.  */
395 static void scsi_read_data(SCSIRequest *req)
396 {
397     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
398     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
399     bool first;
400 
401     trace_scsi_disk_read_data_count(r->sector_count);
402     if (r->sector_count == 0) {
403         /* This also clears the sense buffer for REQUEST SENSE.  */
404         scsi_req_complete(&r->req, GOOD);
405         return;
406     }
407 
408     /* No data transfer may already be in progress */
409     assert(r->req.aiocb == NULL);
410 
411     /* The request is used as the AIO opaque value, so add a ref.  */
412     scsi_req_ref(&r->req);
413     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
414         trace_scsi_disk_read_data_invalid();
415         scsi_read_complete_noio(r, -EINVAL);
416         return;
417     }
418 
419     if (!blk_is_available(req->dev->conf.blk)) {
420         scsi_read_complete_noio(r, -ENOMEDIUM);
421         return;
422     }
423 
424     first = !r->started;
425     r->started = true;
426     if (first && r->need_fua_emulation) {
427         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
428                          BLOCK_ACCT_FLUSH);
429         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_do_read_cb, r);
430     } else {
431         scsi_do_read(r, 0);
432     }
433 }
434 
435 /*
436  * scsi_handle_rw_error has two return values.  False means that the error
437  * must be ignored, true means that the error has been processed and the
438  * caller should not do anything else for this request.  Note that
439  * scsi_handle_rw_error always manages its reference counts, independent
440  * of the return value.
441  */
442 static bool scsi_handle_rw_error(SCSIDiskReq *r, int error, bool acct_failed)
443 {
444     bool is_read = (r->req.cmd.mode == SCSI_XFER_FROM_DEV);
445     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
446     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
447     BlockErrorAction action = blk_get_error_action(s->qdev.conf.blk,
448                                                    is_read, error);
449 
450     if (action == BLOCK_ERROR_ACTION_REPORT) {
451         if (acct_failed) {
452             block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
453         }
454         switch (error) {
455         case 0:
456             /* A passthrough command has run and has produced sense data; check
457              * whether the error has to be handled by the guest or should rather
458              * pause the host.
459              */
460             assert(r->status && *r->status);
461             if (scsi_sense_buf_is_guest_recoverable(r->req.sense, sizeof(r->req.sense))) {
462                 /* These errors are handled by guest. */
463                 sdc->update_sense(&r->req);
464                 scsi_req_complete(&r->req, *r->status);
465                 return true;
466             }
467             error = scsi_sense_buf_to_errno(r->req.sense, sizeof(r->req.sense));
468             break;
469         case ENOMEDIUM:
470             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
471             break;
472         case ENOMEM:
473             scsi_check_condition(r, SENSE_CODE(TARGET_FAILURE));
474             break;
475         case EINVAL:
476             scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
477             break;
478         case ENOSPC:
479             scsi_check_condition(r, SENSE_CODE(SPACE_ALLOC_FAILED));
480             break;
481         default:
482             scsi_check_condition(r, SENSE_CODE(IO_ERROR));
483             break;
484         }
485     }
486 
487     blk_error_action(s->qdev.conf.blk, action, is_read, error);
488     if (action == BLOCK_ERROR_ACTION_IGNORE) {
489         scsi_req_complete(&r->req, 0);
490         return true;
491     }
492 
493     if (action == BLOCK_ERROR_ACTION_STOP) {
494         scsi_req_retry(&r->req);
495     }
496     return true;
497 }
498 
499 static void scsi_write_complete_noio(SCSIDiskReq *r, int ret)
500 {
501     uint32_t n;
502 
503     assert (r->req.aiocb == NULL);
504     if (scsi_disk_req_check_error(r, ret, false)) {
505         goto done;
506     }
507 
508     n = r->qiov.size / BDRV_SECTOR_SIZE;
509     r->sector += n;
510     r->sector_count -= n;
511     if (r->sector_count == 0) {
512         scsi_write_do_fua(r);
513         return;
514     } else {
515         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
516         trace_scsi_disk_write_complete_noio(r->req.tag, r->qiov.size);
517         scsi_req_data(&r->req, r->qiov.size);
518     }
519 
520 done:
521     scsi_req_unref(&r->req);
522 }
523 
524 static void scsi_write_complete(void * opaque, int ret)
525 {
526     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
527     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
528 
529     assert (r->req.aiocb != NULL);
530     r->req.aiocb = NULL;
531 
532     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
533     if (ret < 0) {
534         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
535     } else {
536         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
537     }
538     scsi_write_complete_noio(r, ret);
539     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
540 }
541 
542 static void scsi_write_data(SCSIRequest *req)
543 {
544     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
545     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
546     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
547 
548     /* No data transfer may already be in progress */
549     assert(r->req.aiocb == NULL);
550 
551     /* The request is used as the AIO opaque value, so add a ref.  */
552     scsi_req_ref(&r->req);
553     if (r->req.cmd.mode != SCSI_XFER_TO_DEV) {
554         trace_scsi_disk_write_data_invalid();
555         scsi_write_complete_noio(r, -EINVAL);
556         return;
557     }
558 
559     if (!r->req.sg && !r->qiov.size) {
560         /* Called for the first time.  Ask the driver to send us more data.  */
561         r->started = true;
562         scsi_write_complete_noio(r, 0);
563         return;
564     }
565     if (!blk_is_available(req->dev->conf.blk)) {
566         scsi_write_complete_noio(r, -ENOMEDIUM);
567         return;
568     }
569 
570     if (r->req.cmd.buf[0] == VERIFY_10 || r->req.cmd.buf[0] == VERIFY_12 ||
571         r->req.cmd.buf[0] == VERIFY_16) {
572         if (r->req.sg) {
573             scsi_dma_complete_noio(r, 0);
574         } else {
575             scsi_write_complete_noio(r, 0);
576         }
577         return;
578     }
579 
580     if (r->req.sg) {
581         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_WRITE);
582         r->req.resid -= r->req.sg->size;
583         r->req.aiocb = dma_blk_io(blk_get_aio_context(s->qdev.conf.blk),
584                                   r->req.sg, r->sector << BDRV_SECTOR_BITS,
585                                   BDRV_SECTOR_SIZE,
586                                   sdc->dma_writev, r, scsi_dma_complete, r,
587                                   DMA_DIRECTION_TO_DEVICE);
588     } else {
589         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
590                          r->qiov.size, BLOCK_ACCT_WRITE);
591         r->req.aiocb = sdc->dma_writev(r->sector << BDRV_SECTOR_BITS, &r->qiov,
592                                        scsi_write_complete, r, r);
593     }
594 }
595 
596 /* Return a pointer to the data buffer.  */
597 static uint8_t *scsi_get_buf(SCSIRequest *req)
598 {
599     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
600 
601     return (uint8_t *)r->iov.iov_base;
602 }
603 
604 static int scsi_disk_emulate_vpd_page(SCSIRequest *req, uint8_t *outbuf)
605 {
606     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
607     uint8_t page_code = req->cmd.buf[2];
608     int start, buflen = 0;
609 
610     outbuf[buflen++] = s->qdev.type & 0x1f;
611     outbuf[buflen++] = page_code;
612     outbuf[buflen++] = 0x00;
613     outbuf[buflen++] = 0x00;
614     start = buflen;
615 
616     switch (page_code) {
617     case 0x00: /* Supported page codes, mandatory */
618     {
619         trace_scsi_disk_emulate_vpd_page_00(req->cmd.xfer);
620         outbuf[buflen++] = 0x00; /* list of supported pages (this page) */
621         if (s->serial) {
622             outbuf[buflen++] = 0x80; /* unit serial number */
623         }
624         outbuf[buflen++] = 0x83; /* device identification */
625         if (s->qdev.type == TYPE_DISK) {
626             outbuf[buflen++] = 0xb0; /* block limits */
627             outbuf[buflen++] = 0xb1; /* block device characteristics */
628             outbuf[buflen++] = 0xb2; /* thin provisioning */
629         }
630         break;
631     }
632     case 0x80: /* Device serial number, optional */
633     {
634         int l;
635 
636         if (!s->serial) {
637             trace_scsi_disk_emulate_vpd_page_80_not_supported();
638             return -1;
639         }
640 
641         l = strlen(s->serial);
642         if (l > 36) {
643             l = 36;
644         }
645 
646         trace_scsi_disk_emulate_vpd_page_80(req->cmd.xfer);
647         memcpy(outbuf + buflen, s->serial, l);
648         buflen += l;
649         break;
650     }
651 
652     case 0x83: /* Device identification page, mandatory */
653     {
654         int id_len = s->device_id ? MIN(strlen(s->device_id), 255 - 8) : 0;
655 
656         trace_scsi_disk_emulate_vpd_page_83(req->cmd.xfer);
657 
658         if (id_len) {
659             outbuf[buflen++] = 0x2; /* ASCII */
660             outbuf[buflen++] = 0;   /* not officially assigned */
661             outbuf[buflen++] = 0;   /* reserved */
662             outbuf[buflen++] = id_len; /* length of data following */
663             memcpy(outbuf + buflen, s->device_id, id_len);
664             buflen += id_len;
665         }
666 
667         if (s->qdev.wwn) {
668             outbuf[buflen++] = 0x1; /* Binary */
669             outbuf[buflen++] = 0x3; /* NAA */
670             outbuf[buflen++] = 0;   /* reserved */
671             outbuf[buflen++] = 8;
672             stq_be_p(&outbuf[buflen], s->qdev.wwn);
673             buflen += 8;
674         }
675 
676         if (s->qdev.port_wwn) {
677             outbuf[buflen++] = 0x61; /* SAS / Binary */
678             outbuf[buflen++] = 0x93; /* PIV / Target port / NAA */
679             outbuf[buflen++] = 0;    /* reserved */
680             outbuf[buflen++] = 8;
681             stq_be_p(&outbuf[buflen], s->qdev.port_wwn);
682             buflen += 8;
683         }
684 
685         if (s->port_index) {
686             outbuf[buflen++] = 0x61; /* SAS / Binary */
687 
688             /* PIV/Target port/relative target port */
689             outbuf[buflen++] = 0x94;
690 
691             outbuf[buflen++] = 0;    /* reserved */
692             outbuf[buflen++] = 4;
693             stw_be_p(&outbuf[buflen + 2], s->port_index);
694             buflen += 4;
695         }
696         break;
697     }
698     case 0xb0: /* block limits */
699     {
700         SCSIBlockLimits bl = {};
701 
702         if (s->qdev.type == TYPE_ROM) {
703             trace_scsi_disk_emulate_vpd_page_b0_not_supported();
704             return -1;
705         }
706         bl.wsnz = 1;
707         bl.unmap_sectors =
708             s->qdev.conf.discard_granularity / s->qdev.blocksize;
709         bl.min_io_size =
710             s->qdev.conf.min_io_size / s->qdev.blocksize;
711         bl.opt_io_size =
712             s->qdev.conf.opt_io_size / s->qdev.blocksize;
713         bl.max_unmap_sectors =
714             s->max_unmap_size / s->qdev.blocksize;
715         bl.max_io_sectors =
716             s->max_io_size / s->qdev.blocksize;
717         /* 255 descriptors fit in 4 KiB with an 8-byte header */
718         bl.max_unmap_descr = 255;
719 
720         if (s->qdev.type == TYPE_DISK) {
721             int max_transfer_blk = blk_get_max_transfer(s->qdev.conf.blk);
722             int max_io_sectors_blk =
723                 max_transfer_blk / s->qdev.blocksize;
724 
725             bl.max_io_sectors =
726                 MIN_NON_ZERO(max_io_sectors_blk, bl.max_io_sectors);
727         }
728         buflen += scsi_emulate_block_limits(outbuf + buflen, &bl);
729         break;
730     }
731     case 0xb1: /* block device characteristics */
732     {
733         buflen = 0x40;
734         outbuf[4] = (s->rotation_rate >> 8) & 0xff;
735         outbuf[5] = s->rotation_rate & 0xff;
736         outbuf[6] = 0; /* PRODUCT TYPE */
737         outbuf[7] = 0; /* WABEREQ | WACEREQ | NOMINAL FORM FACTOR */
738         outbuf[8] = 0; /* VBULS */
739         break;
740     }
741     case 0xb2: /* thin provisioning */
742     {
743         buflen = 8;
744         outbuf[4] = 0;
745         outbuf[5] = 0xe0; /* unmap & write_same 10/16 all supported */
746         outbuf[6] = s->qdev.conf.discard_granularity ? 2 : 1;
747         outbuf[7] = 0;
748         break;
749     }
750     default:
751         return -1;
752     }
753     /* done with EVPD */
754     assert(buflen - start <= 255);
755     outbuf[start - 1] = buflen - start;
756     return buflen;
757 }
758 
759 static int scsi_disk_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf)
760 {
761     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
762     int buflen = 0;
763 
764     if (req->cmd.buf[1] & 0x1) {
765         /* Vital product data */
766         return scsi_disk_emulate_vpd_page(req, outbuf);
767     }
768 
769     /* Standard INQUIRY data */
770     if (req->cmd.buf[2] != 0) {
771         return -1;
772     }
773 
774     /* PAGE CODE == 0 */
775     buflen = req->cmd.xfer;
776     if (buflen > SCSI_MAX_INQUIRY_LEN) {
777         buflen = SCSI_MAX_INQUIRY_LEN;
778     }
779 
780     outbuf[0] = s->qdev.type & 0x1f;
781     outbuf[1] = (s->features & (1 << SCSI_DISK_F_REMOVABLE)) ? 0x80 : 0;
782 
783     strpadcpy((char *) &outbuf[16], 16, s->product, ' ');
784     strpadcpy((char *) &outbuf[8], 8, s->vendor, ' ');
785 
786     memset(&outbuf[32], 0, 4);
787     memcpy(&outbuf[32], s->version, MIN(4, strlen(s->version)));
788     /*
789      * We claim conformance to SPC-3, which is required for guests
790      * to ask for modern features like READ CAPACITY(16) or the
791      * block characteristics VPD page by default.  Not all of SPC-3
792      * is actually implemented, but we're good enough.
793      */
794     outbuf[2] = s->qdev.default_scsi_version;
795     outbuf[3] = 2 | 0x10; /* Format 2, HiSup */
796 
797     if (buflen > 36) {
798         outbuf[4] = buflen - 5; /* Additional Length = (Len - 1) - 4 */
799     } else {
800         /* If the allocation length of CDB is too small,
801                the additional length is not adjusted */
802         outbuf[4] = 36 - 5;
803     }
804 
805     /* Sync data transfer and TCQ.  */
806     outbuf[7] = 0x10 | (req->bus->info->tcq ? 0x02 : 0);
807     return buflen;
808 }
809 
810 static inline bool media_is_dvd(SCSIDiskState *s)
811 {
812     uint64_t nb_sectors;
813     if (s->qdev.type != TYPE_ROM) {
814         return false;
815     }
816     if (!blk_is_available(s->qdev.conf.blk)) {
817         return false;
818     }
819     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
820     return nb_sectors > CD_MAX_SECTORS;
821 }
822 
823 static inline bool media_is_cd(SCSIDiskState *s)
824 {
825     uint64_t nb_sectors;
826     if (s->qdev.type != TYPE_ROM) {
827         return false;
828     }
829     if (!blk_is_available(s->qdev.conf.blk)) {
830         return false;
831     }
832     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
833     return nb_sectors <= CD_MAX_SECTORS;
834 }
835 
836 static int scsi_read_disc_information(SCSIDiskState *s, SCSIDiskReq *r,
837                                       uint8_t *outbuf)
838 {
839     uint8_t type = r->req.cmd.buf[1] & 7;
840 
841     if (s->qdev.type != TYPE_ROM) {
842         return -1;
843     }
844 
845     /* Types 1/2 are only defined for Blu-Ray.  */
846     if (type != 0) {
847         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
848         return -1;
849     }
850 
851     memset(outbuf, 0, 34);
852     outbuf[1] = 32;
853     outbuf[2] = 0xe; /* last session complete, disc finalized */
854     outbuf[3] = 1;   /* first track on disc */
855     outbuf[4] = 1;   /* # of sessions */
856     outbuf[5] = 1;   /* first track of last session */
857     outbuf[6] = 1;   /* last track of last session */
858     outbuf[7] = 0x20; /* unrestricted use */
859     outbuf[8] = 0x00; /* CD-ROM or DVD-ROM */
860     /* 9-10-11: most significant byte corresponding bytes 4-5-6 */
861     /* 12-23: not meaningful for CD-ROM or DVD-ROM */
862     /* 24-31: disc bar code */
863     /* 32: disc application code */
864     /* 33: number of OPC tables */
865 
866     return 34;
867 }
868 
869 static int scsi_read_dvd_structure(SCSIDiskState *s, SCSIDiskReq *r,
870                                    uint8_t *outbuf)
871 {
872     static const int rds_caps_size[5] = {
873         [0] = 2048 + 4,
874         [1] = 4 + 4,
875         [3] = 188 + 4,
876         [4] = 2048 + 4,
877     };
878 
879     uint8_t media = r->req.cmd.buf[1];
880     uint8_t layer = r->req.cmd.buf[6];
881     uint8_t format = r->req.cmd.buf[7];
882     int size = -1;
883 
884     if (s->qdev.type != TYPE_ROM) {
885         return -1;
886     }
887     if (media != 0) {
888         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
889         return -1;
890     }
891 
892     if (format != 0xff) {
893         if (!blk_is_available(s->qdev.conf.blk)) {
894             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
895             return -1;
896         }
897         if (media_is_cd(s)) {
898             scsi_check_condition(r, SENSE_CODE(INCOMPATIBLE_FORMAT));
899             return -1;
900         }
901         if (format >= ARRAY_SIZE(rds_caps_size)) {
902             return -1;
903         }
904         size = rds_caps_size[format];
905         memset(outbuf, 0, size);
906     }
907 
908     switch (format) {
909     case 0x00: {
910         /* Physical format information */
911         uint64_t nb_sectors;
912         if (layer != 0) {
913             goto fail;
914         }
915         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
916 
917         outbuf[4] = 1;   /* DVD-ROM, part version 1 */
918         outbuf[5] = 0xf; /* 120mm disc, minimum rate unspecified */
919         outbuf[6] = 1;   /* one layer, read-only (per MMC-2 spec) */
920         outbuf[7] = 0;   /* default densities */
921 
922         stl_be_p(&outbuf[12], (nb_sectors >> 2) - 1); /* end sector */
923         stl_be_p(&outbuf[16], (nb_sectors >> 2) - 1); /* l0 end sector */
924         break;
925     }
926 
927     case 0x01: /* DVD copyright information, all zeros */
928         break;
929 
930     case 0x03: /* BCA information - invalid field for no BCA info */
931         return -1;
932 
933     case 0x04: /* DVD disc manufacturing information, all zeros */
934         break;
935 
936     case 0xff: { /* List capabilities */
937         int i;
938         size = 4;
939         for (i = 0; i < ARRAY_SIZE(rds_caps_size); i++) {
940             if (!rds_caps_size[i]) {
941                 continue;
942             }
943             outbuf[size] = i;
944             outbuf[size + 1] = 0x40; /* Not writable, readable */
945             stw_be_p(&outbuf[size + 2], rds_caps_size[i]);
946             size += 4;
947         }
948         break;
949      }
950 
951     default:
952         return -1;
953     }
954 
955     /* Size of buffer, not including 2 byte size field */
956     stw_be_p(outbuf, size - 2);
957     return size;
958 
959 fail:
960     return -1;
961 }
962 
963 static int scsi_event_status_media(SCSIDiskState *s, uint8_t *outbuf)
964 {
965     uint8_t event_code, media_status;
966 
967     media_status = 0;
968     if (s->tray_open) {
969         media_status = MS_TRAY_OPEN;
970     } else if (blk_is_inserted(s->qdev.conf.blk)) {
971         media_status = MS_MEDIA_PRESENT;
972     }
973 
974     /* Event notification descriptor */
975     event_code = MEC_NO_CHANGE;
976     if (media_status != MS_TRAY_OPEN) {
977         if (s->media_event) {
978             event_code = MEC_NEW_MEDIA;
979             s->media_event = false;
980         } else if (s->eject_request) {
981             event_code = MEC_EJECT_REQUESTED;
982             s->eject_request = false;
983         }
984     }
985 
986     outbuf[0] = event_code;
987     outbuf[1] = media_status;
988 
989     /* These fields are reserved, just clear them. */
990     outbuf[2] = 0;
991     outbuf[3] = 0;
992     return 4;
993 }
994 
995 static int scsi_get_event_status_notification(SCSIDiskState *s, SCSIDiskReq *r,
996                                               uint8_t *outbuf)
997 {
998     int size;
999     uint8_t *buf = r->req.cmd.buf;
1000     uint8_t notification_class_request = buf[4];
1001     if (s->qdev.type != TYPE_ROM) {
1002         return -1;
1003     }
1004     if ((buf[1] & 1) == 0) {
1005         /* asynchronous */
1006         return -1;
1007     }
1008 
1009     size = 4;
1010     outbuf[0] = outbuf[1] = 0;
1011     outbuf[3] = 1 << GESN_MEDIA; /* supported events */
1012     if (notification_class_request & (1 << GESN_MEDIA)) {
1013         outbuf[2] = GESN_MEDIA;
1014         size += scsi_event_status_media(s, &outbuf[size]);
1015     } else {
1016         outbuf[2] = 0x80;
1017     }
1018     stw_be_p(outbuf, size - 4);
1019     return size;
1020 }
1021 
1022 static int scsi_get_configuration(SCSIDiskState *s, uint8_t *outbuf)
1023 {
1024     int current;
1025 
1026     if (s->qdev.type != TYPE_ROM) {
1027         return -1;
1028     }
1029 
1030     if (media_is_dvd(s)) {
1031         current = MMC_PROFILE_DVD_ROM;
1032     } else if (media_is_cd(s)) {
1033         current = MMC_PROFILE_CD_ROM;
1034     } else {
1035         current = MMC_PROFILE_NONE;
1036     }
1037 
1038     memset(outbuf, 0, 40);
1039     stl_be_p(&outbuf[0], 36); /* Bytes after the data length field */
1040     stw_be_p(&outbuf[6], current);
1041     /* outbuf[8] - outbuf[19]: Feature 0 - Profile list */
1042     outbuf[10] = 0x03; /* persistent, current */
1043     outbuf[11] = 8; /* two profiles */
1044     stw_be_p(&outbuf[12], MMC_PROFILE_DVD_ROM);
1045     outbuf[14] = (current == MMC_PROFILE_DVD_ROM);
1046     stw_be_p(&outbuf[16], MMC_PROFILE_CD_ROM);
1047     outbuf[18] = (current == MMC_PROFILE_CD_ROM);
1048     /* outbuf[20] - outbuf[31]: Feature 1 - Core feature */
1049     stw_be_p(&outbuf[20], 1);
1050     outbuf[22] = 0x08 | 0x03; /* version 2, persistent, current */
1051     outbuf[23] = 8;
1052     stl_be_p(&outbuf[24], 1); /* SCSI */
1053     outbuf[28] = 1; /* DBE = 1, mandatory */
1054     /* outbuf[32] - outbuf[39]: Feature 3 - Removable media feature */
1055     stw_be_p(&outbuf[32], 3);
1056     outbuf[34] = 0x08 | 0x03; /* version 2, persistent, current */
1057     outbuf[35] = 4;
1058     outbuf[36] = 0x39; /* tray, load=1, eject=1, unlocked at powerup, lock=1 */
1059     /* TODO: Random readable, CD read, DVD read, drive serial number,
1060        power management */
1061     return 40;
1062 }
1063 
1064 static int scsi_emulate_mechanism_status(SCSIDiskState *s, uint8_t *outbuf)
1065 {
1066     if (s->qdev.type != TYPE_ROM) {
1067         return -1;
1068     }
1069     memset(outbuf, 0, 8);
1070     outbuf[5] = 1; /* CD-ROM */
1071     return 8;
1072 }
1073 
1074 static int mode_sense_page(SCSIDiskState *s, int page, uint8_t **p_outbuf,
1075                            int page_control)
1076 {
1077     static const int mode_sense_valid[0x3f] = {
1078         [MODE_PAGE_HD_GEOMETRY]            = (1 << TYPE_DISK),
1079         [MODE_PAGE_FLEXIBLE_DISK_GEOMETRY] = (1 << TYPE_DISK),
1080         [MODE_PAGE_CACHING]                = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1081         [MODE_PAGE_R_W_ERROR]              = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1082         [MODE_PAGE_AUDIO_CTL]              = (1 << TYPE_ROM),
1083         [MODE_PAGE_CAPABILITIES]           = (1 << TYPE_ROM),
1084     };
1085 
1086     uint8_t *p = *p_outbuf + 2;
1087     int length;
1088 
1089     if ((mode_sense_valid[page] & (1 << s->qdev.type)) == 0) {
1090         return -1;
1091     }
1092 
1093     /*
1094      * If Changeable Values are requested, a mask denoting those mode parameters
1095      * that are changeable shall be returned. As we currently don't support
1096      * parameter changes via MODE_SELECT all bits are returned set to zero.
1097      * The buffer was already menset to zero by the caller of this function.
1098      *
1099      * The offsets here are off by two compared to the descriptions in the
1100      * SCSI specs, because those include a 2-byte header.  This is unfortunate,
1101      * but it is done so that offsets are consistent within our implementation
1102      * of MODE SENSE and MODE SELECT.  MODE SELECT has to deal with both
1103      * 2-byte and 4-byte headers.
1104      */
1105     switch (page) {
1106     case MODE_PAGE_HD_GEOMETRY:
1107         length = 0x16;
1108         if (page_control == 1) { /* Changeable Values */
1109             break;
1110         }
1111         /* if a geometry hint is available, use it */
1112         p[0] = (s->qdev.conf.cyls >> 16) & 0xff;
1113         p[1] = (s->qdev.conf.cyls >> 8) & 0xff;
1114         p[2] = s->qdev.conf.cyls & 0xff;
1115         p[3] = s->qdev.conf.heads & 0xff;
1116         /* Write precomp start cylinder, disabled */
1117         p[4] = (s->qdev.conf.cyls >> 16) & 0xff;
1118         p[5] = (s->qdev.conf.cyls >> 8) & 0xff;
1119         p[6] = s->qdev.conf.cyls & 0xff;
1120         /* Reduced current start cylinder, disabled */
1121         p[7] = (s->qdev.conf.cyls >> 16) & 0xff;
1122         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1123         p[9] = s->qdev.conf.cyls & 0xff;
1124         /* Device step rate [ns], 200ns */
1125         p[10] = 0;
1126         p[11] = 200;
1127         /* Landing zone cylinder */
1128         p[12] = 0xff;
1129         p[13] =  0xff;
1130         p[14] = 0xff;
1131         /* Medium rotation rate [rpm], 5400 rpm */
1132         p[18] = (5400 >> 8) & 0xff;
1133         p[19] = 5400 & 0xff;
1134         break;
1135 
1136     case MODE_PAGE_FLEXIBLE_DISK_GEOMETRY:
1137         length = 0x1e;
1138         if (page_control == 1) { /* Changeable Values */
1139             break;
1140         }
1141         /* Transfer rate [kbit/s], 5Mbit/s */
1142         p[0] = 5000 >> 8;
1143         p[1] = 5000 & 0xff;
1144         /* if a geometry hint is available, use it */
1145         p[2] = s->qdev.conf.heads & 0xff;
1146         p[3] = s->qdev.conf.secs & 0xff;
1147         p[4] = s->qdev.blocksize >> 8;
1148         p[6] = (s->qdev.conf.cyls >> 8) & 0xff;
1149         p[7] = s->qdev.conf.cyls & 0xff;
1150         /* Write precomp start cylinder, disabled */
1151         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1152         p[9] = s->qdev.conf.cyls & 0xff;
1153         /* Reduced current start cylinder, disabled */
1154         p[10] = (s->qdev.conf.cyls >> 8) & 0xff;
1155         p[11] = s->qdev.conf.cyls & 0xff;
1156         /* Device step rate [100us], 100us */
1157         p[12] = 0;
1158         p[13] = 1;
1159         /* Device step pulse width [us], 1us */
1160         p[14] = 1;
1161         /* Device head settle delay [100us], 100us */
1162         p[15] = 0;
1163         p[16] = 1;
1164         /* Motor on delay [0.1s], 0.1s */
1165         p[17] = 1;
1166         /* Motor off delay [0.1s], 0.1s */
1167         p[18] = 1;
1168         /* Medium rotation rate [rpm], 5400 rpm */
1169         p[26] = (5400 >> 8) & 0xff;
1170         p[27] = 5400 & 0xff;
1171         break;
1172 
1173     case MODE_PAGE_CACHING:
1174         length = 0x12;
1175         if (page_control == 1 || /* Changeable Values */
1176             blk_enable_write_cache(s->qdev.conf.blk)) {
1177             p[0] = 4; /* WCE */
1178         }
1179         break;
1180 
1181     case MODE_PAGE_R_W_ERROR:
1182         length = 10;
1183         if (page_control == 1) { /* Changeable Values */
1184             break;
1185         }
1186         p[0] = 0x80; /* Automatic Write Reallocation Enabled */
1187         if (s->qdev.type == TYPE_ROM) {
1188             p[1] = 0x20; /* Read Retry Count */
1189         }
1190         break;
1191 
1192     case MODE_PAGE_AUDIO_CTL:
1193         length = 14;
1194         break;
1195 
1196     case MODE_PAGE_CAPABILITIES:
1197         length = 0x14;
1198         if (page_control == 1) { /* Changeable Values */
1199             break;
1200         }
1201 
1202         p[0] = 0x3b; /* CD-R & CD-RW read */
1203         p[1] = 0; /* Writing not supported */
1204         p[2] = 0x7f; /* Audio, composite, digital out,
1205                         mode 2 form 1&2, multi session */
1206         p[3] = 0xff; /* CD DA, DA accurate, RW supported,
1207                         RW corrected, C2 errors, ISRC,
1208                         UPC, Bar code */
1209         p[4] = 0x2d | (s->tray_locked ? 2 : 0);
1210         /* Locking supported, jumper present, eject, tray */
1211         p[5] = 0; /* no volume & mute control, no
1212                      changer */
1213         p[6] = (50 * 176) >> 8; /* 50x read speed */
1214         p[7] = (50 * 176) & 0xff;
1215         p[8] = 2 >> 8; /* Two volume levels */
1216         p[9] = 2 & 0xff;
1217         p[10] = 2048 >> 8; /* 2M buffer */
1218         p[11] = 2048 & 0xff;
1219         p[12] = (16 * 176) >> 8; /* 16x read speed current */
1220         p[13] = (16 * 176) & 0xff;
1221         p[16] = (16 * 176) >> 8; /* 16x write speed */
1222         p[17] = (16 * 176) & 0xff;
1223         p[18] = (16 * 176) >> 8; /* 16x write speed current */
1224         p[19] = (16 * 176) & 0xff;
1225         break;
1226 
1227     default:
1228         return -1;
1229     }
1230 
1231     assert(length < 256);
1232     (*p_outbuf)[0] = page;
1233     (*p_outbuf)[1] = length;
1234     *p_outbuf += length + 2;
1235     return length + 2;
1236 }
1237 
1238 static int scsi_disk_emulate_mode_sense(SCSIDiskReq *r, uint8_t *outbuf)
1239 {
1240     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1241     uint64_t nb_sectors;
1242     bool dbd;
1243     int page, buflen, ret, page_control;
1244     uint8_t *p;
1245     uint8_t dev_specific_param;
1246 
1247     dbd = (r->req.cmd.buf[1] & 0x8) != 0;
1248     page = r->req.cmd.buf[2] & 0x3f;
1249     page_control = (r->req.cmd.buf[2] & 0xc0) >> 6;
1250 
1251     trace_scsi_disk_emulate_mode_sense((r->req.cmd.buf[0] == MODE_SENSE) ? 6 :
1252                                        10, page, r->req.cmd.xfer, page_control);
1253     memset(outbuf, 0, r->req.cmd.xfer);
1254     p = outbuf;
1255 
1256     if (s->qdev.type == TYPE_DISK) {
1257         dev_specific_param = s->features & (1 << SCSI_DISK_F_DPOFUA) ? 0x10 : 0;
1258         if (blk_is_read_only(s->qdev.conf.blk)) {
1259             dev_specific_param |= 0x80; /* Readonly.  */
1260         }
1261     } else {
1262         /* MMC prescribes that CD/DVD drives have no block descriptors,
1263          * and defines no device-specific parameter.  */
1264         dev_specific_param = 0x00;
1265         dbd = true;
1266     }
1267 
1268     if (r->req.cmd.buf[0] == MODE_SENSE) {
1269         p[1] = 0; /* Default media type.  */
1270         p[2] = dev_specific_param;
1271         p[3] = 0; /* Block descriptor length.  */
1272         p += 4;
1273     } else { /* MODE_SENSE_10 */
1274         p[2] = 0; /* Default media type.  */
1275         p[3] = dev_specific_param;
1276         p[6] = p[7] = 0; /* Block descriptor length.  */
1277         p += 8;
1278     }
1279 
1280     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1281     if (!dbd && nb_sectors) {
1282         if (r->req.cmd.buf[0] == MODE_SENSE) {
1283             outbuf[3] = 8; /* Block descriptor length  */
1284         } else { /* MODE_SENSE_10 */
1285             outbuf[7] = 8; /* Block descriptor length  */
1286         }
1287         nb_sectors /= (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1288         if (nb_sectors > 0xffffff) {
1289             nb_sectors = 0;
1290         }
1291         p[0] = 0; /* media density code */
1292         p[1] = (nb_sectors >> 16) & 0xff;
1293         p[2] = (nb_sectors >> 8) & 0xff;
1294         p[3] = nb_sectors & 0xff;
1295         p[4] = 0; /* reserved */
1296         p[5] = 0; /* bytes 5-7 are the sector size in bytes */
1297         p[6] = s->qdev.blocksize >> 8;
1298         p[7] = 0;
1299         p += 8;
1300     }
1301 
1302     if (page_control == 3) {
1303         /* Saved Values */
1304         scsi_check_condition(r, SENSE_CODE(SAVING_PARAMS_NOT_SUPPORTED));
1305         return -1;
1306     }
1307 
1308     if (page == 0x3f) {
1309         for (page = 0; page <= 0x3e; page++) {
1310             mode_sense_page(s, page, &p, page_control);
1311         }
1312     } else {
1313         ret = mode_sense_page(s, page, &p, page_control);
1314         if (ret == -1) {
1315             return -1;
1316         }
1317     }
1318 
1319     buflen = p - outbuf;
1320     /*
1321      * The mode data length field specifies the length in bytes of the
1322      * following data that is available to be transferred. The mode data
1323      * length does not include itself.
1324      */
1325     if (r->req.cmd.buf[0] == MODE_SENSE) {
1326         outbuf[0] = buflen - 1;
1327     } else { /* MODE_SENSE_10 */
1328         outbuf[0] = ((buflen - 2) >> 8) & 0xff;
1329         outbuf[1] = (buflen - 2) & 0xff;
1330     }
1331     return buflen;
1332 }
1333 
1334 static int scsi_disk_emulate_read_toc(SCSIRequest *req, uint8_t *outbuf)
1335 {
1336     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1337     int start_track, format, msf, toclen;
1338     uint64_t nb_sectors;
1339 
1340     msf = req->cmd.buf[1] & 2;
1341     format = req->cmd.buf[2] & 0xf;
1342     start_track = req->cmd.buf[6];
1343     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1344     trace_scsi_disk_emulate_read_toc(start_track, format, msf >> 1);
1345     nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
1346     switch (format) {
1347     case 0:
1348         toclen = cdrom_read_toc(nb_sectors, outbuf, msf, start_track);
1349         break;
1350     case 1:
1351         /* multi session : only a single session defined */
1352         toclen = 12;
1353         memset(outbuf, 0, 12);
1354         outbuf[1] = 0x0a;
1355         outbuf[2] = 0x01;
1356         outbuf[3] = 0x01;
1357         break;
1358     case 2:
1359         toclen = cdrom_read_toc_raw(nb_sectors, outbuf, msf, start_track);
1360         break;
1361     default:
1362         return -1;
1363     }
1364     return toclen;
1365 }
1366 
1367 static int scsi_disk_emulate_start_stop(SCSIDiskReq *r)
1368 {
1369     SCSIRequest *req = &r->req;
1370     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1371     bool start = req->cmd.buf[4] & 1;
1372     bool loej = req->cmd.buf[4] & 2; /* load on start, eject on !start */
1373     int pwrcnd = req->cmd.buf[4] & 0xf0;
1374 
1375     if (pwrcnd) {
1376         /* eject/load only happens for power condition == 0 */
1377         return 0;
1378     }
1379 
1380     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) && loej) {
1381         if (!start && !s->tray_open && s->tray_locked) {
1382             scsi_check_condition(r,
1383                                  blk_is_inserted(s->qdev.conf.blk)
1384                                  ? SENSE_CODE(ILLEGAL_REQ_REMOVAL_PREVENTED)
1385                                  : SENSE_CODE(NOT_READY_REMOVAL_PREVENTED));
1386             return -1;
1387         }
1388 
1389         if (s->tray_open != !start) {
1390             blk_eject(s->qdev.conf.blk, !start);
1391             s->tray_open = !start;
1392         }
1393     }
1394     return 0;
1395 }
1396 
1397 static void scsi_disk_emulate_read_data(SCSIRequest *req)
1398 {
1399     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1400     int buflen = r->iov.iov_len;
1401 
1402     if (buflen) {
1403         trace_scsi_disk_emulate_read_data(buflen);
1404         r->iov.iov_len = 0;
1405         r->started = true;
1406         scsi_req_data(&r->req, buflen);
1407         return;
1408     }
1409 
1410     /* This also clears the sense buffer for REQUEST SENSE.  */
1411     scsi_req_complete(&r->req, GOOD);
1412 }
1413 
1414 static int scsi_disk_check_mode_select(SCSIDiskState *s, int page,
1415                                        uint8_t *inbuf, int inlen)
1416 {
1417     uint8_t mode_current[SCSI_MAX_MODE_LEN];
1418     uint8_t mode_changeable[SCSI_MAX_MODE_LEN];
1419     uint8_t *p;
1420     int len, expected_len, changeable_len, i;
1421 
1422     /* The input buffer does not include the page header, so it is
1423      * off by 2 bytes.
1424      */
1425     expected_len = inlen + 2;
1426     if (expected_len > SCSI_MAX_MODE_LEN) {
1427         return -1;
1428     }
1429 
1430     p = mode_current;
1431     memset(mode_current, 0, inlen + 2);
1432     len = mode_sense_page(s, page, &p, 0);
1433     if (len < 0 || len != expected_len) {
1434         return -1;
1435     }
1436 
1437     p = mode_changeable;
1438     memset(mode_changeable, 0, inlen + 2);
1439     changeable_len = mode_sense_page(s, page, &p, 1);
1440     assert(changeable_len == len);
1441 
1442     /* Check that unchangeable bits are the same as what MODE SENSE
1443      * would return.
1444      */
1445     for (i = 2; i < len; i++) {
1446         if (((mode_current[i] ^ inbuf[i - 2]) & ~mode_changeable[i]) != 0) {
1447             return -1;
1448         }
1449     }
1450     return 0;
1451 }
1452 
1453 static void scsi_disk_apply_mode_select(SCSIDiskState *s, int page, uint8_t *p)
1454 {
1455     switch (page) {
1456     case MODE_PAGE_CACHING:
1457         blk_set_enable_write_cache(s->qdev.conf.blk, (p[0] & 4) != 0);
1458         break;
1459 
1460     default:
1461         break;
1462     }
1463 }
1464 
1465 static int mode_select_pages(SCSIDiskReq *r, uint8_t *p, int len, bool change)
1466 {
1467     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1468 
1469     while (len > 0) {
1470         int page, subpage, page_len;
1471 
1472         /* Parse both possible formats for the mode page headers.  */
1473         page = p[0] & 0x3f;
1474         if (p[0] & 0x40) {
1475             if (len < 4) {
1476                 goto invalid_param_len;
1477             }
1478             subpage = p[1];
1479             page_len = lduw_be_p(&p[2]);
1480             p += 4;
1481             len -= 4;
1482         } else {
1483             if (len < 2) {
1484                 goto invalid_param_len;
1485             }
1486             subpage = 0;
1487             page_len = p[1];
1488             p += 2;
1489             len -= 2;
1490         }
1491 
1492         if (subpage) {
1493             goto invalid_param;
1494         }
1495         if (page_len > len) {
1496             goto invalid_param_len;
1497         }
1498 
1499         if (!change) {
1500             if (scsi_disk_check_mode_select(s, page, p, page_len) < 0) {
1501                 goto invalid_param;
1502             }
1503         } else {
1504             scsi_disk_apply_mode_select(s, page, p);
1505         }
1506 
1507         p += page_len;
1508         len -= page_len;
1509     }
1510     return 0;
1511 
1512 invalid_param:
1513     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1514     return -1;
1515 
1516 invalid_param_len:
1517     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1518     return -1;
1519 }
1520 
1521 static void scsi_disk_emulate_mode_select(SCSIDiskReq *r, uint8_t *inbuf)
1522 {
1523     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1524     uint8_t *p = inbuf;
1525     int cmd = r->req.cmd.buf[0];
1526     int len = r->req.cmd.xfer;
1527     int hdr_len = (cmd == MODE_SELECT ? 4 : 8);
1528     int bd_len;
1529     int pass;
1530 
1531     /* We only support PF=1, SP=0.  */
1532     if ((r->req.cmd.buf[1] & 0x11) != 0x10) {
1533         goto invalid_field;
1534     }
1535 
1536     if (len < hdr_len) {
1537         goto invalid_param_len;
1538     }
1539 
1540     bd_len = (cmd == MODE_SELECT ? p[3] : lduw_be_p(&p[6]));
1541     len -= hdr_len;
1542     p += hdr_len;
1543     if (len < bd_len) {
1544         goto invalid_param_len;
1545     }
1546     if (bd_len != 0 && bd_len != 8) {
1547         goto invalid_param;
1548     }
1549 
1550     len -= bd_len;
1551     p += bd_len;
1552 
1553     /* Ensure no change is made if there is an error!  */
1554     for (pass = 0; pass < 2; pass++) {
1555         if (mode_select_pages(r, p, len, pass == 1) < 0) {
1556             assert(pass == 0);
1557             return;
1558         }
1559     }
1560     if (!blk_enable_write_cache(s->qdev.conf.blk)) {
1561         /* The request is used as the AIO opaque value, so add a ref.  */
1562         scsi_req_ref(&r->req);
1563         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
1564                          BLOCK_ACCT_FLUSH);
1565         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
1566         return;
1567     }
1568 
1569     scsi_req_complete(&r->req, GOOD);
1570     return;
1571 
1572 invalid_param:
1573     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1574     return;
1575 
1576 invalid_param_len:
1577     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1578     return;
1579 
1580 invalid_field:
1581     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1582 }
1583 
1584 static inline bool check_lba_range(SCSIDiskState *s,
1585                                    uint64_t sector_num, uint32_t nb_sectors)
1586 {
1587     /*
1588      * The first line tests that no overflow happens when computing the last
1589      * sector.  The second line tests that the last accessed sector is in
1590      * range.
1591      *
1592      * Careful, the computations should not underflow for nb_sectors == 0,
1593      * and a 0-block read to the first LBA beyond the end of device is
1594      * valid.
1595      */
1596     return (sector_num <= sector_num + nb_sectors &&
1597             sector_num + nb_sectors <= s->qdev.max_lba + 1);
1598 }
1599 
1600 typedef struct UnmapCBData {
1601     SCSIDiskReq *r;
1602     uint8_t *inbuf;
1603     int count;
1604 } UnmapCBData;
1605 
1606 static void scsi_unmap_complete(void *opaque, int ret);
1607 
1608 static void scsi_unmap_complete_noio(UnmapCBData *data, int ret)
1609 {
1610     SCSIDiskReq *r = data->r;
1611     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1612 
1613     assert(r->req.aiocb == NULL);
1614 
1615     if (data->count > 0) {
1616         r->sector = ldq_be_p(&data->inbuf[0])
1617             * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1618         r->sector_count = (ldl_be_p(&data->inbuf[8]) & 0xffffffffULL)
1619             * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1620         if (!check_lba_range(s, r->sector, r->sector_count)) {
1621             block_acct_invalid(blk_get_stats(s->qdev.conf.blk),
1622                                BLOCK_ACCT_UNMAP);
1623             scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1624             goto done;
1625         }
1626 
1627         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1628                          r->sector_count * BDRV_SECTOR_SIZE,
1629                          BLOCK_ACCT_UNMAP);
1630 
1631         r->req.aiocb = blk_aio_pdiscard(s->qdev.conf.blk,
1632                                         r->sector * BDRV_SECTOR_SIZE,
1633                                         r->sector_count * BDRV_SECTOR_SIZE,
1634                                         scsi_unmap_complete, data);
1635         data->count--;
1636         data->inbuf += 16;
1637         return;
1638     }
1639 
1640     scsi_req_complete(&r->req, GOOD);
1641 
1642 done:
1643     scsi_req_unref(&r->req);
1644     g_free(data);
1645 }
1646 
1647 static void scsi_unmap_complete(void *opaque, int ret)
1648 {
1649     UnmapCBData *data = opaque;
1650     SCSIDiskReq *r = data->r;
1651     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1652 
1653     assert(r->req.aiocb != NULL);
1654     r->req.aiocb = NULL;
1655 
1656     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
1657     if (scsi_disk_req_check_error(r, ret, true)) {
1658         scsi_req_unref(&r->req);
1659         g_free(data);
1660     } else {
1661         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1662         scsi_unmap_complete_noio(data, ret);
1663     }
1664     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1665 }
1666 
1667 static void scsi_disk_emulate_unmap(SCSIDiskReq *r, uint8_t *inbuf)
1668 {
1669     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1670     uint8_t *p = inbuf;
1671     int len = r->req.cmd.xfer;
1672     UnmapCBData *data;
1673 
1674     /* Reject ANCHOR=1.  */
1675     if (r->req.cmd.buf[1] & 0x1) {
1676         goto invalid_field;
1677     }
1678 
1679     if (len < 8) {
1680         goto invalid_param_len;
1681     }
1682     if (len < lduw_be_p(&p[0]) + 2) {
1683         goto invalid_param_len;
1684     }
1685     if (len < lduw_be_p(&p[2]) + 8) {
1686         goto invalid_param_len;
1687     }
1688     if (lduw_be_p(&p[2]) & 15) {
1689         goto invalid_param_len;
1690     }
1691 
1692     if (blk_is_read_only(s->qdev.conf.blk)) {
1693         block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1694         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1695         return;
1696     }
1697 
1698     data = g_new0(UnmapCBData, 1);
1699     data->r = r;
1700     data->inbuf = &p[8];
1701     data->count = lduw_be_p(&p[2]) >> 4;
1702 
1703     /* The matching unref is in scsi_unmap_complete, before data is freed.  */
1704     scsi_req_ref(&r->req);
1705     scsi_unmap_complete_noio(data, 0);
1706     return;
1707 
1708 invalid_param_len:
1709     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1710     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1711     return;
1712 
1713 invalid_field:
1714     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1715     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1716 }
1717 
1718 typedef struct WriteSameCBData {
1719     SCSIDiskReq *r;
1720     int64_t sector;
1721     int nb_sectors;
1722     QEMUIOVector qiov;
1723     struct iovec iov;
1724 } WriteSameCBData;
1725 
1726 static void scsi_write_same_complete(void *opaque, int ret)
1727 {
1728     WriteSameCBData *data = opaque;
1729     SCSIDiskReq *r = data->r;
1730     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1731 
1732     assert(r->req.aiocb != NULL);
1733     r->req.aiocb = NULL;
1734     aio_context_acquire(blk_get_aio_context(s->qdev.conf.blk));
1735     if (scsi_disk_req_check_error(r, ret, true)) {
1736         goto done;
1737     }
1738 
1739     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1740 
1741     data->nb_sectors -= data->iov.iov_len / BDRV_SECTOR_SIZE;
1742     data->sector += data->iov.iov_len / BDRV_SECTOR_SIZE;
1743     data->iov.iov_len = MIN(data->nb_sectors * BDRV_SECTOR_SIZE,
1744                             data->iov.iov_len);
1745     if (data->iov.iov_len) {
1746         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1747                          data->iov.iov_len, BLOCK_ACCT_WRITE);
1748         /* Reinitialize qiov, to handle unaligned WRITE SAME request
1749          * where final qiov may need smaller size */
1750         qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1751         r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1752                                        data->sector << BDRV_SECTOR_BITS,
1753                                        &data->qiov, 0,
1754                                        scsi_write_same_complete, data);
1755         aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1756         return;
1757     }
1758 
1759     scsi_req_complete(&r->req, GOOD);
1760 
1761 done:
1762     scsi_req_unref(&r->req);
1763     qemu_vfree(data->iov.iov_base);
1764     g_free(data);
1765     aio_context_release(blk_get_aio_context(s->qdev.conf.blk));
1766 }
1767 
1768 static void scsi_disk_emulate_write_same(SCSIDiskReq *r, uint8_t *inbuf)
1769 {
1770     SCSIRequest *req = &r->req;
1771     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1772     uint32_t nb_sectors = scsi_data_cdb_xfer(r->req.cmd.buf);
1773     WriteSameCBData *data;
1774     uint8_t *buf;
1775     int i;
1776 
1777     /* Fail if PBDATA=1 or LBDATA=1 or ANCHOR=1.  */
1778     if (nb_sectors == 0 || (req->cmd.buf[1] & 0x16)) {
1779         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1780         return;
1781     }
1782 
1783     if (blk_is_read_only(s->qdev.conf.blk)) {
1784         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1785         return;
1786     }
1787     if (!check_lba_range(s, r->req.cmd.lba, nb_sectors)) {
1788         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1789         return;
1790     }
1791 
1792     if ((req->cmd.buf[1] & 0x1) || buffer_is_zero(inbuf, s->qdev.blocksize)) {
1793         int flags = (req->cmd.buf[1] & 0x8) ? BDRV_REQ_MAY_UNMAP : 0;
1794 
1795         /* The request is used as the AIO opaque value, so add a ref.  */
1796         scsi_req_ref(&r->req);
1797         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1798                          nb_sectors * s->qdev.blocksize,
1799                         BLOCK_ACCT_WRITE);
1800         r->req.aiocb = blk_aio_pwrite_zeroes(s->qdev.conf.blk,
1801                                 r->req.cmd.lba * s->qdev.blocksize,
1802                                 nb_sectors * s->qdev.blocksize,
1803                                 flags, scsi_aio_complete, r);
1804         return;
1805     }
1806 
1807     data = g_new0(WriteSameCBData, 1);
1808     data->r = r;
1809     data->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1810     data->nb_sectors = nb_sectors * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1811     data->iov.iov_len = MIN(data->nb_sectors * BDRV_SECTOR_SIZE,
1812                             SCSI_WRITE_SAME_MAX);
1813     data->iov.iov_base = buf = blk_blockalign(s->qdev.conf.blk,
1814                                               data->iov.iov_len);
1815     qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1816 
1817     for (i = 0; i < data->iov.iov_len; i += s->qdev.blocksize) {
1818         memcpy(&buf[i], inbuf, s->qdev.blocksize);
1819     }
1820 
1821     scsi_req_ref(&r->req);
1822     block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1823                      data->iov.iov_len, BLOCK_ACCT_WRITE);
1824     r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1825                                    data->sector << BDRV_SECTOR_BITS,
1826                                    &data->qiov, 0,
1827                                    scsi_write_same_complete, data);
1828 }
1829 
1830 static void scsi_disk_emulate_write_data(SCSIRequest *req)
1831 {
1832     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1833 
1834     if (r->iov.iov_len) {
1835         int buflen = r->iov.iov_len;
1836         trace_scsi_disk_emulate_write_data(buflen);
1837         r->iov.iov_len = 0;
1838         scsi_req_data(&r->req, buflen);
1839         return;
1840     }
1841 
1842     switch (req->cmd.buf[0]) {
1843     case MODE_SELECT:
1844     case MODE_SELECT_10:
1845         /* This also clears the sense buffer for REQUEST SENSE.  */
1846         scsi_disk_emulate_mode_select(r, r->iov.iov_base);
1847         break;
1848 
1849     case UNMAP:
1850         scsi_disk_emulate_unmap(r, r->iov.iov_base);
1851         break;
1852 
1853     case VERIFY_10:
1854     case VERIFY_12:
1855     case VERIFY_16:
1856         if (r->req.status == -1) {
1857             scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1858         }
1859         break;
1860 
1861     case WRITE_SAME_10:
1862     case WRITE_SAME_16:
1863         scsi_disk_emulate_write_same(r, r->iov.iov_base);
1864         break;
1865 
1866     default:
1867         abort();
1868     }
1869 }
1870 
1871 static int32_t scsi_disk_emulate_command(SCSIRequest *req, uint8_t *buf)
1872 {
1873     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1874     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1875     uint64_t nb_sectors;
1876     uint8_t *outbuf;
1877     int buflen;
1878 
1879     switch (req->cmd.buf[0]) {
1880     case INQUIRY:
1881     case MODE_SENSE:
1882     case MODE_SENSE_10:
1883     case RESERVE:
1884     case RESERVE_10:
1885     case RELEASE:
1886     case RELEASE_10:
1887     case START_STOP:
1888     case ALLOW_MEDIUM_REMOVAL:
1889     case GET_CONFIGURATION:
1890     case GET_EVENT_STATUS_NOTIFICATION:
1891     case MECHANISM_STATUS:
1892     case REQUEST_SENSE:
1893         break;
1894 
1895     default:
1896         if (!blk_is_available(s->qdev.conf.blk)) {
1897             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
1898             return 0;
1899         }
1900         break;
1901     }
1902 
1903     /*
1904      * FIXME: we shouldn't return anything bigger than 4k, but the code
1905      * requires the buffer to be as big as req->cmd.xfer in several
1906      * places.  So, do not allow CDBs with a very large ALLOCATION
1907      * LENGTH.  The real fix would be to modify scsi_read_data and
1908      * dma_buf_read, so that they return data beyond the buflen
1909      * as all zeros.
1910      */
1911     if (req->cmd.xfer > 65536) {
1912         goto illegal_request;
1913     }
1914     r->buflen = MAX(4096, req->cmd.xfer);
1915 
1916     if (!r->iov.iov_base) {
1917         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
1918     }
1919 
1920     outbuf = r->iov.iov_base;
1921     memset(outbuf, 0, r->buflen);
1922     switch (req->cmd.buf[0]) {
1923     case TEST_UNIT_READY:
1924         assert(blk_is_available(s->qdev.conf.blk));
1925         break;
1926     case INQUIRY:
1927         buflen = scsi_disk_emulate_inquiry(req, outbuf);
1928         if (buflen < 0) {
1929             goto illegal_request;
1930         }
1931         break;
1932     case MODE_SENSE:
1933     case MODE_SENSE_10:
1934         buflen = scsi_disk_emulate_mode_sense(r, outbuf);
1935         if (buflen < 0) {
1936             goto illegal_request;
1937         }
1938         break;
1939     case READ_TOC:
1940         buflen = scsi_disk_emulate_read_toc(req, outbuf);
1941         if (buflen < 0) {
1942             goto illegal_request;
1943         }
1944         break;
1945     case RESERVE:
1946         if (req->cmd.buf[1] & 1) {
1947             goto illegal_request;
1948         }
1949         break;
1950     case RESERVE_10:
1951         if (req->cmd.buf[1] & 3) {
1952             goto illegal_request;
1953         }
1954         break;
1955     case RELEASE:
1956         if (req->cmd.buf[1] & 1) {
1957             goto illegal_request;
1958         }
1959         break;
1960     case RELEASE_10:
1961         if (req->cmd.buf[1] & 3) {
1962             goto illegal_request;
1963         }
1964         break;
1965     case START_STOP:
1966         if (scsi_disk_emulate_start_stop(r) < 0) {
1967             return 0;
1968         }
1969         break;
1970     case ALLOW_MEDIUM_REMOVAL:
1971         s->tray_locked = req->cmd.buf[4] & 1;
1972         blk_lock_medium(s->qdev.conf.blk, req->cmd.buf[4] & 1);
1973         break;
1974     case READ_CAPACITY_10:
1975         /* The normal LEN field for this command is zero.  */
1976         memset(outbuf, 0, 8);
1977         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1978         if (!nb_sectors) {
1979             scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
1980             return 0;
1981         }
1982         if ((req->cmd.buf[8] & 1) == 0 && req->cmd.lba) {
1983             goto illegal_request;
1984         }
1985         nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
1986         /* Returned value is the address of the last sector.  */
1987         nb_sectors--;
1988         /* Remember the new size for read/write sanity checking. */
1989         s->qdev.max_lba = nb_sectors;
1990         /* Clip to 2TB, instead of returning capacity modulo 2TB. */
1991         if (nb_sectors > UINT32_MAX) {
1992             nb_sectors = UINT32_MAX;
1993         }
1994         outbuf[0] = (nb_sectors >> 24) & 0xff;
1995         outbuf[1] = (nb_sectors >> 16) & 0xff;
1996         outbuf[2] = (nb_sectors >> 8) & 0xff;
1997         outbuf[3] = nb_sectors & 0xff;
1998         outbuf[4] = 0;
1999         outbuf[5] = 0;
2000         outbuf[6] = s->qdev.blocksize >> 8;
2001         outbuf[7] = 0;
2002         break;
2003     case REQUEST_SENSE:
2004         /* Just return "NO SENSE".  */
2005         buflen = scsi_convert_sense(NULL, 0, outbuf, r->buflen,
2006                                     (req->cmd.buf[1] & 1) == 0);
2007         if (buflen < 0) {
2008             goto illegal_request;
2009         }
2010         break;
2011     case MECHANISM_STATUS:
2012         buflen = scsi_emulate_mechanism_status(s, outbuf);
2013         if (buflen < 0) {
2014             goto illegal_request;
2015         }
2016         break;
2017     case GET_CONFIGURATION:
2018         buflen = scsi_get_configuration(s, outbuf);
2019         if (buflen < 0) {
2020             goto illegal_request;
2021         }
2022         break;
2023     case GET_EVENT_STATUS_NOTIFICATION:
2024         buflen = scsi_get_event_status_notification(s, r, outbuf);
2025         if (buflen < 0) {
2026             goto illegal_request;
2027         }
2028         break;
2029     case READ_DISC_INFORMATION:
2030         buflen = scsi_read_disc_information(s, r, outbuf);
2031         if (buflen < 0) {
2032             goto illegal_request;
2033         }
2034         break;
2035     case READ_DVD_STRUCTURE:
2036         buflen = scsi_read_dvd_structure(s, r, outbuf);
2037         if (buflen < 0) {
2038             goto illegal_request;
2039         }
2040         break;
2041     case SERVICE_ACTION_IN_16:
2042         /* Service Action In subcommands. */
2043         if ((req->cmd.buf[1] & 31) == SAI_READ_CAPACITY_16) {
2044             trace_scsi_disk_emulate_command_SAI_16();
2045             memset(outbuf, 0, req->cmd.xfer);
2046             blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2047             if (!nb_sectors) {
2048                 scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
2049                 return 0;
2050             }
2051             if ((req->cmd.buf[14] & 1) == 0 && req->cmd.lba) {
2052                 goto illegal_request;
2053             }
2054             nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2055             /* Returned value is the address of the last sector.  */
2056             nb_sectors--;
2057             /* Remember the new size for read/write sanity checking. */
2058             s->qdev.max_lba = nb_sectors;
2059             outbuf[0] = (nb_sectors >> 56) & 0xff;
2060             outbuf[1] = (nb_sectors >> 48) & 0xff;
2061             outbuf[2] = (nb_sectors >> 40) & 0xff;
2062             outbuf[3] = (nb_sectors >> 32) & 0xff;
2063             outbuf[4] = (nb_sectors >> 24) & 0xff;
2064             outbuf[5] = (nb_sectors >> 16) & 0xff;
2065             outbuf[6] = (nb_sectors >> 8) & 0xff;
2066             outbuf[7] = nb_sectors & 0xff;
2067             outbuf[8] = 0;
2068             outbuf[9] = 0;
2069             outbuf[10] = s->qdev.blocksize >> 8;
2070             outbuf[11] = 0;
2071             outbuf[12] = 0;
2072             outbuf[13] = get_physical_block_exp(&s->qdev.conf);
2073 
2074             /* set TPE bit if the format supports discard */
2075             if (s->qdev.conf.discard_granularity) {
2076                 outbuf[14] = 0x80;
2077             }
2078 
2079             /* Protection, exponent and lowest lba field left blank. */
2080             break;
2081         }
2082         trace_scsi_disk_emulate_command_SAI_unsupported();
2083         goto illegal_request;
2084     case SYNCHRONIZE_CACHE:
2085         /* The request is used as the AIO opaque value, so add a ref.  */
2086         scsi_req_ref(&r->req);
2087         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
2088                          BLOCK_ACCT_FLUSH);
2089         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
2090         return 0;
2091     case SEEK_10:
2092         trace_scsi_disk_emulate_command_SEEK_10(r->req.cmd.lba);
2093         if (r->req.cmd.lba > s->qdev.max_lba) {
2094             goto illegal_lba;
2095         }
2096         break;
2097     case MODE_SELECT:
2098         trace_scsi_disk_emulate_command_MODE_SELECT(r->req.cmd.xfer);
2099         break;
2100     case MODE_SELECT_10:
2101         trace_scsi_disk_emulate_command_MODE_SELECT_10(r->req.cmd.xfer);
2102         break;
2103     case UNMAP:
2104         trace_scsi_disk_emulate_command_UNMAP(r->req.cmd.xfer);
2105         break;
2106     case VERIFY_10:
2107     case VERIFY_12:
2108     case VERIFY_16:
2109         trace_scsi_disk_emulate_command_VERIFY((req->cmd.buf[1] >> 1) & 3);
2110         if (req->cmd.buf[1] & 6) {
2111             goto illegal_request;
2112         }
2113         break;
2114     case WRITE_SAME_10:
2115     case WRITE_SAME_16:
2116         trace_scsi_disk_emulate_command_WRITE_SAME(
2117                 req->cmd.buf[0] == WRITE_SAME_10 ? 10 : 16, r->req.cmd.xfer);
2118         break;
2119     default:
2120         trace_scsi_disk_emulate_command_UNKNOWN(buf[0],
2121                                                 scsi_command_name(buf[0]));
2122         scsi_check_condition(r, SENSE_CODE(INVALID_OPCODE));
2123         return 0;
2124     }
2125     assert(!r->req.aiocb);
2126     r->iov.iov_len = MIN(r->buflen, req->cmd.xfer);
2127     if (r->iov.iov_len == 0) {
2128         scsi_req_complete(&r->req, GOOD);
2129     }
2130     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2131         assert(r->iov.iov_len == req->cmd.xfer);
2132         return -r->iov.iov_len;
2133     } else {
2134         return r->iov.iov_len;
2135     }
2136 
2137 illegal_request:
2138     if (r->req.status == -1) {
2139         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2140     }
2141     return 0;
2142 
2143 illegal_lba:
2144     scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2145     return 0;
2146 }
2147 
2148 /* Execute a scsi command.  Returns the length of the data expected by the
2149    command.  This will be Positive for data transfers from the device
2150    (eg. disk reads), negative for transfers to the device (eg. disk writes),
2151    and zero if the command does not transfer any data.  */
2152 
2153 static int32_t scsi_disk_dma_command(SCSIRequest *req, uint8_t *buf)
2154 {
2155     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
2156     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2157     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
2158     uint32_t len;
2159     uint8_t command;
2160 
2161     command = buf[0];
2162 
2163     if (!blk_is_available(s->qdev.conf.blk)) {
2164         scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
2165         return 0;
2166     }
2167 
2168     len = scsi_data_cdb_xfer(r->req.cmd.buf);
2169     switch (command) {
2170     case READ_6:
2171     case READ_10:
2172     case READ_12:
2173     case READ_16:
2174         trace_scsi_disk_dma_command_READ(r->req.cmd.lba, len);
2175         /* Protection information is not supported.  For SCSI versions 2 and
2176          * older (as determined by snooping the guest's INQUIRY commands),
2177          * there is no RD/WR/VRPROTECT, so skip this check in these versions.
2178          */
2179         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2180             goto illegal_request;
2181         }
2182         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2183             goto illegal_lba;
2184         }
2185         r->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2186         r->sector_count = len * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2187         break;
2188     case WRITE_6:
2189     case WRITE_10:
2190     case WRITE_12:
2191     case WRITE_16:
2192     case WRITE_VERIFY_10:
2193     case WRITE_VERIFY_12:
2194     case WRITE_VERIFY_16:
2195         if (blk_is_read_only(s->qdev.conf.blk)) {
2196             scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
2197             return 0;
2198         }
2199         trace_scsi_disk_dma_command_WRITE(
2200                 (command & 0xe) == 0xe ? "And Verify " : "",
2201                 r->req.cmd.lba, len);
2202         /* fall through */
2203     case VERIFY_10:
2204     case VERIFY_12:
2205     case VERIFY_16:
2206         /* We get here only for BYTCHK == 0x01 and only for scsi-block.
2207          * As far as DMA is concerned, we can treat it the same as a write;
2208          * scsi_block_do_sgio will send VERIFY commands.
2209          */
2210         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2211             goto illegal_request;
2212         }
2213         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2214             goto illegal_lba;
2215         }
2216         r->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2217         r->sector_count = len * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2218         break;
2219     default:
2220         abort();
2221     illegal_request:
2222         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2223         return 0;
2224     illegal_lba:
2225         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2226         return 0;
2227     }
2228     r->need_fua_emulation = sdc->need_fua_emulation(&r->req.cmd);
2229     if (r->sector_count == 0) {
2230         scsi_req_complete(&r->req, GOOD);
2231     }
2232     assert(r->iov.iov_len == 0);
2233     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2234         return -r->sector_count * BDRV_SECTOR_SIZE;
2235     } else {
2236         return r->sector_count * BDRV_SECTOR_SIZE;
2237     }
2238 }
2239 
2240 static void scsi_disk_reset(DeviceState *dev)
2241 {
2242     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev.qdev, dev);
2243     uint64_t nb_sectors;
2244 
2245     scsi_device_purge_requests(&s->qdev, SENSE_CODE(RESET));
2246 
2247     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2248     nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2249     if (nb_sectors) {
2250         nb_sectors--;
2251     }
2252     s->qdev.max_lba = nb_sectors;
2253     /* reset tray statuses */
2254     s->tray_locked = 0;
2255     s->tray_open = 0;
2256 
2257     s->qdev.scsi_version = s->qdev.default_scsi_version;
2258 }
2259 
2260 static void scsi_disk_resize_cb(void *opaque)
2261 {
2262     SCSIDiskState *s = opaque;
2263 
2264     /* SPC lists this sense code as available only for
2265      * direct-access devices.
2266      */
2267     if (s->qdev.type == TYPE_DISK) {
2268         scsi_device_report_change(&s->qdev, SENSE_CODE(CAPACITY_CHANGED));
2269     }
2270 }
2271 
2272 static void scsi_cd_change_media_cb(void *opaque, bool load, Error **errp)
2273 {
2274     SCSIDiskState *s = opaque;
2275 
2276     /*
2277      * When a CD gets changed, we have to report an ejected state and
2278      * then a loaded state to guests so that they detect tray
2279      * open/close and media change events.  Guests that do not use
2280      * GET_EVENT_STATUS_NOTIFICATION to detect such tray open/close
2281      * states rely on this behavior.
2282      *
2283      * media_changed governs the state machine used for unit attention
2284      * report.  media_event is used by GET EVENT STATUS NOTIFICATION.
2285      */
2286     s->media_changed = load;
2287     s->tray_open = !load;
2288     scsi_device_set_ua(&s->qdev, SENSE_CODE(UNIT_ATTENTION_NO_MEDIUM));
2289     s->media_event = true;
2290     s->eject_request = false;
2291 }
2292 
2293 static void scsi_cd_eject_request_cb(void *opaque, bool force)
2294 {
2295     SCSIDiskState *s = opaque;
2296 
2297     s->eject_request = true;
2298     if (force) {
2299         s->tray_locked = false;
2300     }
2301 }
2302 
2303 static bool scsi_cd_is_tray_open(void *opaque)
2304 {
2305     return ((SCSIDiskState *)opaque)->tray_open;
2306 }
2307 
2308 static bool scsi_cd_is_medium_locked(void *opaque)
2309 {
2310     return ((SCSIDiskState *)opaque)->tray_locked;
2311 }
2312 
2313 static const BlockDevOps scsi_disk_removable_block_ops = {
2314     .change_media_cb = scsi_cd_change_media_cb,
2315     .eject_request_cb = scsi_cd_eject_request_cb,
2316     .is_tray_open = scsi_cd_is_tray_open,
2317     .is_medium_locked = scsi_cd_is_medium_locked,
2318 
2319     .resize_cb = scsi_disk_resize_cb,
2320 };
2321 
2322 static const BlockDevOps scsi_disk_block_ops = {
2323     .resize_cb = scsi_disk_resize_cb,
2324 };
2325 
2326 static void scsi_disk_unit_attention_reported(SCSIDevice *dev)
2327 {
2328     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2329     if (s->media_changed) {
2330         s->media_changed = false;
2331         scsi_device_set_ua(&s->qdev, SENSE_CODE(MEDIUM_CHANGED));
2332     }
2333 }
2334 
2335 static void scsi_realize(SCSIDevice *dev, Error **errp)
2336 {
2337     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2338     bool read_only;
2339 
2340     if (!s->qdev.conf.blk) {
2341         error_setg(errp, "drive property not set");
2342         return;
2343     }
2344 
2345     if (!(s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2346         !blk_is_inserted(s->qdev.conf.blk)) {
2347         error_setg(errp, "Device needs media, but drive is empty");
2348         return;
2349     }
2350 
2351     if (!blkconf_blocksizes(&s->qdev.conf, errp)) {
2352         return;
2353     }
2354 
2355     if (blk_get_aio_context(s->qdev.conf.blk) != qemu_get_aio_context() &&
2356         !s->qdev.hba_supports_iothread)
2357     {
2358         error_setg(errp, "HBA does not support iothreads");
2359         return;
2360     }
2361 
2362     if (dev->type == TYPE_DISK) {
2363         if (!blkconf_geometry(&dev->conf, NULL, 65535, 255, 255, errp)) {
2364             return;
2365         }
2366     }
2367 
2368     read_only = blk_is_read_only(s->qdev.conf.blk);
2369     if (dev->type == TYPE_ROM) {
2370         read_only = true;
2371     }
2372 
2373     if (!blkconf_apply_backend_options(&dev->conf, read_only,
2374                                        dev->type == TYPE_DISK, errp)) {
2375         return;
2376     }
2377 
2378     if (s->qdev.conf.discard_granularity == -1) {
2379         s->qdev.conf.discard_granularity =
2380             MAX(s->qdev.conf.logical_block_size, DEFAULT_DISCARD_GRANULARITY);
2381     }
2382 
2383     if (!s->version) {
2384         s->version = g_strdup(qemu_hw_version());
2385     }
2386     if (!s->vendor) {
2387         s->vendor = g_strdup("QEMU");
2388     }
2389     if (!s->device_id) {
2390         if (s->serial) {
2391             s->device_id = g_strdup_printf("%.20s", s->serial);
2392         } else {
2393             const char *str = blk_name(s->qdev.conf.blk);
2394             if (str && *str) {
2395                 s->device_id = g_strdup(str);
2396             }
2397         }
2398     }
2399 
2400     if (blk_is_sg(s->qdev.conf.blk)) {
2401         error_setg(errp, "unwanted /dev/sg*");
2402         return;
2403     }
2404 
2405     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2406             !(s->features & (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS))) {
2407         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_removable_block_ops, s);
2408     } else {
2409         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_block_ops, s);
2410     }
2411     blk_set_guest_block_size(s->qdev.conf.blk, s->qdev.blocksize);
2412 
2413     blk_iostatus_enable(s->qdev.conf.blk);
2414 
2415     add_boot_device_lchs(&dev->qdev, NULL,
2416                          dev->conf.lcyls,
2417                          dev->conf.lheads,
2418                          dev->conf.lsecs);
2419 }
2420 
2421 static void scsi_unrealize(SCSIDevice *dev)
2422 {
2423     del_boot_device_lchs(&dev->qdev, NULL);
2424 }
2425 
2426 static void scsi_hd_realize(SCSIDevice *dev, Error **errp)
2427 {
2428     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2429     AioContext *ctx = NULL;
2430     /* can happen for devices without drive. The error message for missing
2431      * backend will be issued in scsi_realize
2432      */
2433     if (s->qdev.conf.blk) {
2434         ctx = blk_get_aio_context(s->qdev.conf.blk);
2435         aio_context_acquire(ctx);
2436         if (!blkconf_blocksizes(&s->qdev.conf, errp)) {
2437             goto out;
2438         }
2439     }
2440     s->qdev.blocksize = s->qdev.conf.logical_block_size;
2441     s->qdev.type = TYPE_DISK;
2442     if (!s->product) {
2443         s->product = g_strdup("QEMU HARDDISK");
2444     }
2445     scsi_realize(&s->qdev, errp);
2446 out:
2447     if (ctx) {
2448         aio_context_release(ctx);
2449     }
2450 }
2451 
2452 static void scsi_cd_realize(SCSIDevice *dev, Error **errp)
2453 {
2454     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2455     AioContext *ctx;
2456     int ret;
2457 
2458     if (!dev->conf.blk) {
2459         /* Anonymous BlockBackend for an empty drive. As we put it into
2460          * dev->conf, qdev takes care of detaching on unplug. */
2461         dev->conf.blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
2462         ret = blk_attach_dev(dev->conf.blk, &dev->qdev);
2463         assert(ret == 0);
2464     }
2465 
2466     ctx = blk_get_aio_context(dev->conf.blk);
2467     aio_context_acquire(ctx);
2468     s->qdev.blocksize = 2048;
2469     s->qdev.type = TYPE_ROM;
2470     s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2471     if (!s->product) {
2472         s->product = g_strdup("QEMU CD-ROM");
2473     }
2474     scsi_realize(&s->qdev, errp);
2475     aio_context_release(ctx);
2476 }
2477 
2478 static void scsi_disk_realize(SCSIDevice *dev, Error **errp)
2479 {
2480     DriveInfo *dinfo;
2481     Error *local_err = NULL;
2482 
2483     warn_report("'scsi-disk' is deprecated, "
2484                 "please use 'scsi-hd' or 'scsi-cd' instead");
2485 
2486     if (!dev->conf.blk) {
2487         scsi_realize(dev, &local_err);
2488         assert(local_err);
2489         error_propagate(errp, local_err);
2490         return;
2491     }
2492 
2493     dinfo = blk_legacy_dinfo(dev->conf.blk);
2494     if (dinfo && dinfo->media_cd) {
2495         scsi_cd_realize(dev, errp);
2496     } else {
2497         scsi_hd_realize(dev, errp);
2498     }
2499 }
2500 
2501 static const SCSIReqOps scsi_disk_emulate_reqops = {
2502     .size         = sizeof(SCSIDiskReq),
2503     .free_req     = scsi_free_request,
2504     .send_command = scsi_disk_emulate_command,
2505     .read_data    = scsi_disk_emulate_read_data,
2506     .write_data   = scsi_disk_emulate_write_data,
2507     .get_buf      = scsi_get_buf,
2508 };
2509 
2510 static const SCSIReqOps scsi_disk_dma_reqops = {
2511     .size         = sizeof(SCSIDiskReq),
2512     .free_req     = scsi_free_request,
2513     .send_command = scsi_disk_dma_command,
2514     .read_data    = scsi_read_data,
2515     .write_data   = scsi_write_data,
2516     .get_buf      = scsi_get_buf,
2517     .load_request = scsi_disk_load_request,
2518     .save_request = scsi_disk_save_request,
2519 };
2520 
2521 static const SCSIReqOps *const scsi_disk_reqops_dispatch[256] = {
2522     [TEST_UNIT_READY]                 = &scsi_disk_emulate_reqops,
2523     [INQUIRY]                         = &scsi_disk_emulate_reqops,
2524     [MODE_SENSE]                      = &scsi_disk_emulate_reqops,
2525     [MODE_SENSE_10]                   = &scsi_disk_emulate_reqops,
2526     [START_STOP]                      = &scsi_disk_emulate_reqops,
2527     [ALLOW_MEDIUM_REMOVAL]            = &scsi_disk_emulate_reqops,
2528     [READ_CAPACITY_10]                = &scsi_disk_emulate_reqops,
2529     [READ_TOC]                        = &scsi_disk_emulate_reqops,
2530     [READ_DVD_STRUCTURE]              = &scsi_disk_emulate_reqops,
2531     [READ_DISC_INFORMATION]           = &scsi_disk_emulate_reqops,
2532     [GET_CONFIGURATION]               = &scsi_disk_emulate_reqops,
2533     [GET_EVENT_STATUS_NOTIFICATION]   = &scsi_disk_emulate_reqops,
2534     [MECHANISM_STATUS]                = &scsi_disk_emulate_reqops,
2535     [SERVICE_ACTION_IN_16]            = &scsi_disk_emulate_reqops,
2536     [REQUEST_SENSE]                   = &scsi_disk_emulate_reqops,
2537     [SYNCHRONIZE_CACHE]               = &scsi_disk_emulate_reqops,
2538     [SEEK_10]                         = &scsi_disk_emulate_reqops,
2539     [MODE_SELECT]                     = &scsi_disk_emulate_reqops,
2540     [MODE_SELECT_10]                  = &scsi_disk_emulate_reqops,
2541     [UNMAP]                           = &scsi_disk_emulate_reqops,
2542     [WRITE_SAME_10]                   = &scsi_disk_emulate_reqops,
2543     [WRITE_SAME_16]                   = &scsi_disk_emulate_reqops,
2544     [VERIFY_10]                       = &scsi_disk_emulate_reqops,
2545     [VERIFY_12]                       = &scsi_disk_emulate_reqops,
2546     [VERIFY_16]                       = &scsi_disk_emulate_reqops,
2547 
2548     [READ_6]                          = &scsi_disk_dma_reqops,
2549     [READ_10]                         = &scsi_disk_dma_reqops,
2550     [READ_12]                         = &scsi_disk_dma_reqops,
2551     [READ_16]                         = &scsi_disk_dma_reqops,
2552     [WRITE_6]                         = &scsi_disk_dma_reqops,
2553     [WRITE_10]                        = &scsi_disk_dma_reqops,
2554     [WRITE_12]                        = &scsi_disk_dma_reqops,
2555     [WRITE_16]                        = &scsi_disk_dma_reqops,
2556     [WRITE_VERIFY_10]                 = &scsi_disk_dma_reqops,
2557     [WRITE_VERIFY_12]                 = &scsi_disk_dma_reqops,
2558     [WRITE_VERIFY_16]                 = &scsi_disk_dma_reqops,
2559 };
2560 
2561 static void scsi_disk_new_request_dump(uint32_t lun, uint32_t tag, uint8_t *buf)
2562 {
2563     int i;
2564     int len = scsi_cdb_length(buf);
2565     char *line_buffer, *p;
2566 
2567     line_buffer = g_malloc(len * 5 + 1);
2568 
2569     for (i = 0, p = line_buffer; i < len; i++) {
2570         p += sprintf(p, " 0x%02x", buf[i]);
2571     }
2572     trace_scsi_disk_new_request(lun, tag, line_buffer);
2573 
2574     g_free(line_buffer);
2575 }
2576 
2577 static SCSIRequest *scsi_new_request(SCSIDevice *d, uint32_t tag, uint32_t lun,
2578                                      uint8_t *buf, void *hba_private)
2579 {
2580     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2581     SCSIRequest *req;
2582     const SCSIReqOps *ops;
2583     uint8_t command;
2584 
2585     command = buf[0];
2586     ops = scsi_disk_reqops_dispatch[command];
2587     if (!ops) {
2588         ops = &scsi_disk_emulate_reqops;
2589     }
2590     req = scsi_req_alloc(ops, &s->qdev, tag, lun, hba_private);
2591 
2592     if (trace_event_get_state_backends(TRACE_SCSI_DISK_NEW_REQUEST)) {
2593         scsi_disk_new_request_dump(lun, tag, buf);
2594     }
2595 
2596     return req;
2597 }
2598 
2599 #ifdef __linux__
2600 static int get_device_type(SCSIDiskState *s)
2601 {
2602     uint8_t cmd[16];
2603     uint8_t buf[36];
2604     int ret;
2605 
2606     memset(cmd, 0, sizeof(cmd));
2607     memset(buf, 0, sizeof(buf));
2608     cmd[0] = INQUIRY;
2609     cmd[4] = sizeof(buf);
2610 
2611     ret = scsi_SG_IO_FROM_DEV(s->qdev.conf.blk, cmd, sizeof(cmd),
2612                               buf, sizeof(buf));
2613     if (ret < 0) {
2614         return -1;
2615     }
2616     s->qdev.type = buf[0];
2617     if (buf[1] & 0x80) {
2618         s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2619     }
2620     return 0;
2621 }
2622 
2623 static void scsi_block_realize(SCSIDevice *dev, Error **errp)
2624 {
2625     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2626     AioContext *ctx;
2627     int sg_version;
2628     int rc;
2629 
2630     if (!s->qdev.conf.blk) {
2631         error_setg(errp, "drive property not set");
2632         return;
2633     }
2634 
2635     if (s->rotation_rate) {
2636         error_report_once("rotation_rate is specified for scsi-block but is "
2637                           "not implemented. This option is deprecated and will "
2638                           "be removed in a future version");
2639     }
2640 
2641     ctx = blk_get_aio_context(s->qdev.conf.blk);
2642     aio_context_acquire(ctx);
2643 
2644     /* check we are using a driver managing SG_IO (version 3 and after) */
2645     rc = blk_ioctl(s->qdev.conf.blk, SG_GET_VERSION_NUM, &sg_version);
2646     if (rc < 0) {
2647         error_setg_errno(errp, -rc, "cannot get SG_IO version number");
2648         if (rc != -EPERM) {
2649             error_append_hint(errp, "Is this a SCSI device?\n");
2650         }
2651         goto out;
2652     }
2653     if (sg_version < 30000) {
2654         error_setg(errp, "scsi generic interface too old");
2655         goto out;
2656     }
2657 
2658     /* get device type from INQUIRY data */
2659     rc = get_device_type(s);
2660     if (rc < 0) {
2661         error_setg(errp, "INQUIRY failed");
2662         goto out;
2663     }
2664 
2665     /* Make a guess for the block size, we'll fix it when the guest sends.
2666      * READ CAPACITY.  If they don't, they likely would assume these sizes
2667      * anyway. (TODO: check in /sys).
2668      */
2669     if (s->qdev.type == TYPE_ROM || s->qdev.type == TYPE_WORM) {
2670         s->qdev.blocksize = 2048;
2671     } else {
2672         s->qdev.blocksize = 512;
2673     }
2674 
2675     /* Makes the scsi-block device not removable by using HMP and QMP eject
2676      * command.
2677      */
2678     s->features |= (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS);
2679 
2680     scsi_realize(&s->qdev, errp);
2681     scsi_generic_read_device_inquiry(&s->qdev);
2682 
2683 out:
2684     aio_context_release(ctx);
2685 }
2686 
2687 typedef struct SCSIBlockReq {
2688     SCSIDiskReq req;
2689     sg_io_hdr_t io_header;
2690 
2691     /* Selected bytes of the original CDB, copied into our own CDB.  */
2692     uint8_t cmd, cdb1, group_number;
2693 
2694     /* CDB passed to SG_IO.  */
2695     uint8_t cdb[16];
2696 } SCSIBlockReq;
2697 
2698 static BlockAIOCB *scsi_block_do_sgio(SCSIBlockReq *req,
2699                                       int64_t offset, QEMUIOVector *iov,
2700                                       int direction,
2701                                       BlockCompletionFunc *cb, void *opaque)
2702 {
2703     sg_io_hdr_t *io_header = &req->io_header;
2704     SCSIDiskReq *r = &req->req;
2705     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2706     int nb_logical_blocks;
2707     uint64_t lba;
2708     BlockAIOCB *aiocb;
2709 
2710     /* This is not supported yet.  It can only happen if the guest does
2711      * reads and writes that are not aligned to one logical sectors
2712      * _and_ cover multiple MemoryRegions.
2713      */
2714     assert(offset % s->qdev.blocksize == 0);
2715     assert(iov->size % s->qdev.blocksize == 0);
2716 
2717     io_header->interface_id = 'S';
2718 
2719     /* The data transfer comes from the QEMUIOVector.  */
2720     io_header->dxfer_direction = direction;
2721     io_header->dxfer_len = iov->size;
2722     io_header->dxferp = (void *)iov->iov;
2723     io_header->iovec_count = iov->niov;
2724     assert(io_header->iovec_count == iov->niov); /* no overflow! */
2725 
2726     /* Build a new CDB with the LBA and length patched in, in case
2727      * DMA helpers split the transfer in multiple segments.  Do not
2728      * build a CDB smaller than what the guest wanted, and only build
2729      * a larger one if strictly necessary.
2730      */
2731     io_header->cmdp = req->cdb;
2732     lba = offset / s->qdev.blocksize;
2733     nb_logical_blocks = io_header->dxfer_len / s->qdev.blocksize;
2734 
2735     if ((req->cmd >> 5) == 0 && lba <= 0x1ffff) {
2736         /* 6-byte CDB */
2737         stl_be_p(&req->cdb[0], lba | (req->cmd << 24));
2738         req->cdb[4] = nb_logical_blocks;
2739         req->cdb[5] = 0;
2740         io_header->cmd_len = 6;
2741     } else if ((req->cmd >> 5) <= 1 && lba <= 0xffffffffULL) {
2742         /* 10-byte CDB */
2743         req->cdb[0] = (req->cmd & 0x1f) | 0x20;
2744         req->cdb[1] = req->cdb1;
2745         stl_be_p(&req->cdb[2], lba);
2746         req->cdb[6] = req->group_number;
2747         stw_be_p(&req->cdb[7], nb_logical_blocks);
2748         req->cdb[9] = 0;
2749         io_header->cmd_len = 10;
2750     } else if ((req->cmd >> 5) != 4 && lba <= 0xffffffffULL) {
2751         /* 12-byte CDB */
2752         req->cdb[0] = (req->cmd & 0x1f) | 0xA0;
2753         req->cdb[1] = req->cdb1;
2754         stl_be_p(&req->cdb[2], lba);
2755         stl_be_p(&req->cdb[6], nb_logical_blocks);
2756         req->cdb[10] = req->group_number;
2757         req->cdb[11] = 0;
2758         io_header->cmd_len = 12;
2759     } else {
2760         /* 16-byte CDB */
2761         req->cdb[0] = (req->cmd & 0x1f) | 0x80;
2762         req->cdb[1] = req->cdb1;
2763         stq_be_p(&req->cdb[2], lba);
2764         stl_be_p(&req->cdb[10], nb_logical_blocks);
2765         req->cdb[14] = req->group_number;
2766         req->cdb[15] = 0;
2767         io_header->cmd_len = 16;
2768     }
2769 
2770     /* The rest is as in scsi-generic.c.  */
2771     io_header->mx_sb_len = sizeof(r->req.sense);
2772     io_header->sbp = r->req.sense;
2773     io_header->timeout = UINT_MAX;
2774     io_header->usr_ptr = r;
2775     io_header->flags |= SG_FLAG_DIRECT_IO;
2776 
2777     aiocb = blk_aio_ioctl(s->qdev.conf.blk, SG_IO, io_header, cb, opaque);
2778     assert(aiocb != NULL);
2779     return aiocb;
2780 }
2781 
2782 static bool scsi_block_no_fua(SCSICommand *cmd)
2783 {
2784     return false;
2785 }
2786 
2787 static BlockAIOCB *scsi_block_dma_readv(int64_t offset,
2788                                         QEMUIOVector *iov,
2789                                         BlockCompletionFunc *cb, void *cb_opaque,
2790                                         void *opaque)
2791 {
2792     SCSIBlockReq *r = opaque;
2793     return scsi_block_do_sgio(r, offset, iov,
2794                               SG_DXFER_FROM_DEV, cb, cb_opaque);
2795 }
2796 
2797 static BlockAIOCB *scsi_block_dma_writev(int64_t offset,
2798                                          QEMUIOVector *iov,
2799                                          BlockCompletionFunc *cb, void *cb_opaque,
2800                                          void *opaque)
2801 {
2802     SCSIBlockReq *r = opaque;
2803     return scsi_block_do_sgio(r, offset, iov,
2804                               SG_DXFER_TO_DEV, cb, cb_opaque);
2805 }
2806 
2807 static bool scsi_block_is_passthrough(SCSIDiskState *s, uint8_t *buf)
2808 {
2809     switch (buf[0]) {
2810     case VERIFY_10:
2811     case VERIFY_12:
2812     case VERIFY_16:
2813         /* Check if BYTCHK == 0x01 (data-out buffer contains data
2814          * for the number of logical blocks specified in the length
2815          * field).  For other modes, do not use scatter/gather operation.
2816          */
2817         if ((buf[1] & 6) == 2) {
2818             return false;
2819         }
2820         break;
2821 
2822     case READ_6:
2823     case READ_10:
2824     case READ_12:
2825     case READ_16:
2826     case WRITE_6:
2827     case WRITE_10:
2828     case WRITE_12:
2829     case WRITE_16:
2830     case WRITE_VERIFY_10:
2831     case WRITE_VERIFY_12:
2832     case WRITE_VERIFY_16:
2833         /* MMC writing cannot be done via DMA helpers, because it sometimes
2834          * involves writing beyond the maximum LBA or to negative LBA (lead-in).
2835          * We might use scsi_block_dma_reqops as long as no writing commands are
2836          * seen, but performance usually isn't paramount on optical media.  So,
2837          * just make scsi-block operate the same as scsi-generic for them.
2838          */
2839         if (s->qdev.type != TYPE_ROM) {
2840             return false;
2841         }
2842         break;
2843 
2844     default:
2845         break;
2846     }
2847 
2848     return true;
2849 }
2850 
2851 
2852 static int32_t scsi_block_dma_command(SCSIRequest *req, uint8_t *buf)
2853 {
2854     SCSIBlockReq *r = (SCSIBlockReq *)req;
2855     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2856 
2857     r->cmd = req->cmd.buf[0];
2858     switch (r->cmd >> 5) {
2859     case 0:
2860         /* 6-byte CDB.  */
2861         r->cdb1 = r->group_number = 0;
2862         break;
2863     case 1:
2864         /* 10-byte CDB.  */
2865         r->cdb1 = req->cmd.buf[1];
2866         r->group_number = req->cmd.buf[6];
2867         break;
2868     case 4:
2869         /* 12-byte CDB.  */
2870         r->cdb1 = req->cmd.buf[1];
2871         r->group_number = req->cmd.buf[10];
2872         break;
2873     case 5:
2874         /* 16-byte CDB.  */
2875         r->cdb1 = req->cmd.buf[1];
2876         r->group_number = req->cmd.buf[14];
2877         break;
2878     default:
2879         abort();
2880     }
2881 
2882     /* Protection information is not supported.  For SCSI versions 2 and
2883      * older (as determined by snooping the guest's INQUIRY commands),
2884      * there is no RD/WR/VRPROTECT, so skip this check in these versions.
2885      */
2886     if (s->qdev.scsi_version > 2 && (req->cmd.buf[1] & 0xe0)) {
2887         scsi_check_condition(&r->req, SENSE_CODE(INVALID_FIELD));
2888         return 0;
2889     }
2890 
2891     r->req.status = &r->io_header.status;
2892     return scsi_disk_dma_command(req, buf);
2893 }
2894 
2895 static const SCSIReqOps scsi_block_dma_reqops = {
2896     .size         = sizeof(SCSIBlockReq),
2897     .free_req     = scsi_free_request,
2898     .send_command = scsi_block_dma_command,
2899     .read_data    = scsi_read_data,
2900     .write_data   = scsi_write_data,
2901     .get_buf      = scsi_get_buf,
2902     .load_request = scsi_disk_load_request,
2903     .save_request = scsi_disk_save_request,
2904 };
2905 
2906 static SCSIRequest *scsi_block_new_request(SCSIDevice *d, uint32_t tag,
2907                                            uint32_t lun, uint8_t *buf,
2908                                            void *hba_private)
2909 {
2910     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2911 
2912     if (scsi_block_is_passthrough(s, buf)) {
2913         return scsi_req_alloc(&scsi_generic_req_ops, &s->qdev, tag, lun,
2914                               hba_private);
2915     } else {
2916         return scsi_req_alloc(&scsi_block_dma_reqops, &s->qdev, tag, lun,
2917                               hba_private);
2918     }
2919 }
2920 
2921 static int scsi_block_parse_cdb(SCSIDevice *d, SCSICommand *cmd,
2922                                   uint8_t *buf, void *hba_private)
2923 {
2924     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2925 
2926     if (scsi_block_is_passthrough(s, buf)) {
2927         return scsi_bus_parse_cdb(&s->qdev, cmd, buf, hba_private);
2928     } else {
2929         return scsi_req_parse_cdb(&s->qdev, cmd, buf);
2930     }
2931 }
2932 
2933 static void scsi_block_update_sense(SCSIRequest *req)
2934 {
2935     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
2936     SCSIBlockReq *br = DO_UPCAST(SCSIBlockReq, req, r);
2937     r->req.sense_len = MIN(br->io_header.sb_len_wr, sizeof(r->req.sense));
2938 }
2939 #endif
2940 
2941 static
2942 BlockAIOCB *scsi_dma_readv(int64_t offset, QEMUIOVector *iov,
2943                            BlockCompletionFunc *cb, void *cb_opaque,
2944                            void *opaque)
2945 {
2946     SCSIDiskReq *r = opaque;
2947     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2948     return blk_aio_preadv(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
2949 }
2950 
2951 static
2952 BlockAIOCB *scsi_dma_writev(int64_t offset, QEMUIOVector *iov,
2953                             BlockCompletionFunc *cb, void *cb_opaque,
2954                             void *opaque)
2955 {
2956     SCSIDiskReq *r = opaque;
2957     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2958     return blk_aio_pwritev(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
2959 }
2960 
2961 static void scsi_disk_base_class_initfn(ObjectClass *klass, void *data)
2962 {
2963     DeviceClass *dc = DEVICE_CLASS(klass);
2964     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
2965 
2966     dc->fw_name = "disk";
2967     dc->reset = scsi_disk_reset;
2968     sdc->dma_readv = scsi_dma_readv;
2969     sdc->dma_writev = scsi_dma_writev;
2970     sdc->need_fua_emulation = scsi_is_cmd_fua;
2971 }
2972 
2973 static const TypeInfo scsi_disk_base_info = {
2974     .name          = TYPE_SCSI_DISK_BASE,
2975     .parent        = TYPE_SCSI_DEVICE,
2976     .class_init    = scsi_disk_base_class_initfn,
2977     .instance_size = sizeof(SCSIDiskState),
2978     .class_size    = sizeof(SCSIDiskClass),
2979     .abstract      = true,
2980 };
2981 
2982 #define DEFINE_SCSI_DISK_PROPERTIES()                                   \
2983     DEFINE_PROP_DRIVE_IOTHREAD("drive", SCSIDiskState, qdev.conf.blk),  \
2984     DEFINE_BLOCK_PROPERTIES_BASE(SCSIDiskState, qdev.conf),             \
2985     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),            \
2986     DEFINE_PROP_STRING("ver", SCSIDiskState, version),                  \
2987     DEFINE_PROP_STRING("serial", SCSIDiskState, serial),                \
2988     DEFINE_PROP_STRING("vendor", SCSIDiskState, vendor),                \
2989     DEFINE_PROP_STRING("product", SCSIDiskState, product),              \
2990     DEFINE_PROP_STRING("device_id", SCSIDiskState, device_id)
2991 
2992 
2993 static Property scsi_hd_properties[] = {
2994     DEFINE_SCSI_DISK_PROPERTIES(),
2995     DEFINE_PROP_BIT("removable", SCSIDiskState, features,
2996                     SCSI_DISK_F_REMOVABLE, false),
2997     DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
2998                     SCSI_DISK_F_DPOFUA, false),
2999     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3000     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3001     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3002     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3003                        DEFAULT_MAX_UNMAP_SIZE),
3004     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3005                        DEFAULT_MAX_IO_SIZE),
3006     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3007     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3008                       5),
3009     DEFINE_BLOCK_CHS_PROPERTIES(SCSIDiskState, qdev.conf),
3010     DEFINE_PROP_END_OF_LIST(),
3011 };
3012 
3013 static const VMStateDescription vmstate_scsi_disk_state = {
3014     .name = "scsi-disk",
3015     .version_id = 1,
3016     .minimum_version_id = 1,
3017     .fields = (VMStateField[]) {
3018         VMSTATE_SCSI_DEVICE(qdev, SCSIDiskState),
3019         VMSTATE_BOOL(media_changed, SCSIDiskState),
3020         VMSTATE_BOOL(media_event, SCSIDiskState),
3021         VMSTATE_BOOL(eject_request, SCSIDiskState),
3022         VMSTATE_BOOL(tray_open, SCSIDiskState),
3023         VMSTATE_BOOL(tray_locked, SCSIDiskState),
3024         VMSTATE_END_OF_LIST()
3025     }
3026 };
3027 
3028 static void scsi_hd_class_initfn(ObjectClass *klass, void *data)
3029 {
3030     DeviceClass *dc = DEVICE_CLASS(klass);
3031     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3032 
3033     sc->realize      = scsi_hd_realize;
3034     sc->unrealize    = scsi_unrealize;
3035     sc->alloc_req    = scsi_new_request;
3036     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3037     dc->desc = "virtual SCSI disk";
3038     device_class_set_props(dc, scsi_hd_properties);
3039     dc->vmsd  = &vmstate_scsi_disk_state;
3040 }
3041 
3042 static const TypeInfo scsi_hd_info = {
3043     .name          = "scsi-hd",
3044     .parent        = TYPE_SCSI_DISK_BASE,
3045     .class_init    = scsi_hd_class_initfn,
3046 };
3047 
3048 static Property scsi_cd_properties[] = {
3049     DEFINE_SCSI_DISK_PROPERTIES(),
3050     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3051     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3052     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3053     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3054                        DEFAULT_MAX_IO_SIZE),
3055     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3056                       5),
3057     DEFINE_PROP_END_OF_LIST(),
3058 };
3059 
3060 static void scsi_cd_class_initfn(ObjectClass *klass, void *data)
3061 {
3062     DeviceClass *dc = DEVICE_CLASS(klass);
3063     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3064 
3065     sc->realize      = scsi_cd_realize;
3066     sc->alloc_req    = scsi_new_request;
3067     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3068     dc->desc = "virtual SCSI CD-ROM";
3069     device_class_set_props(dc, scsi_cd_properties);
3070     dc->vmsd  = &vmstate_scsi_disk_state;
3071 }
3072 
3073 static const TypeInfo scsi_cd_info = {
3074     .name          = "scsi-cd",
3075     .parent        = TYPE_SCSI_DISK_BASE,
3076     .class_init    = scsi_cd_class_initfn,
3077 };
3078 
3079 #ifdef __linux__
3080 static Property scsi_block_properties[] = {
3081     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),
3082     DEFINE_PROP_DRIVE("drive", SCSIDiskState, qdev.conf.blk),
3083     DEFINE_PROP_BOOL("share-rw", SCSIDiskState, qdev.conf.share_rw, false),
3084     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3085     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3086                        DEFAULT_MAX_UNMAP_SIZE),
3087     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3088                        DEFAULT_MAX_IO_SIZE),
3089     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3090                       -1),
3091     DEFINE_PROP_END_OF_LIST(),
3092 };
3093 
3094 static void scsi_block_class_initfn(ObjectClass *klass, void *data)
3095 {
3096     DeviceClass *dc = DEVICE_CLASS(klass);
3097     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3098     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
3099 
3100     sc->realize      = scsi_block_realize;
3101     sc->alloc_req    = scsi_block_new_request;
3102     sc->parse_cdb    = scsi_block_parse_cdb;
3103     sdc->dma_readv   = scsi_block_dma_readv;
3104     sdc->dma_writev  = scsi_block_dma_writev;
3105     sdc->update_sense = scsi_block_update_sense;
3106     sdc->need_fua_emulation = scsi_block_no_fua;
3107     dc->desc = "SCSI block device passthrough";
3108     device_class_set_props(dc, scsi_block_properties);
3109     dc->vmsd  = &vmstate_scsi_disk_state;
3110 }
3111 
3112 static const TypeInfo scsi_block_info = {
3113     .name          = "scsi-block",
3114     .parent        = TYPE_SCSI_DISK_BASE,
3115     .class_init    = scsi_block_class_initfn,
3116 };
3117 #endif
3118 
3119 static Property scsi_disk_properties[] = {
3120     DEFINE_SCSI_DISK_PROPERTIES(),
3121     DEFINE_PROP_BIT("removable", SCSIDiskState, features,
3122                     SCSI_DISK_F_REMOVABLE, false),
3123     DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
3124                     SCSI_DISK_F_DPOFUA, false),
3125     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3126     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3127     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3128     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3129                        DEFAULT_MAX_UNMAP_SIZE),
3130     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3131                        DEFAULT_MAX_IO_SIZE),
3132     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3133                       5),
3134     DEFINE_PROP_END_OF_LIST(),
3135 };
3136 
3137 static void scsi_disk_class_initfn(ObjectClass *klass, void *data)
3138 {
3139     DeviceClass *dc = DEVICE_CLASS(klass);
3140     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3141 
3142     sc->realize      = scsi_disk_realize;
3143     sc->alloc_req    = scsi_new_request;
3144     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3145     dc->fw_name = "disk";
3146     dc->desc = "virtual SCSI disk or CD-ROM (legacy)";
3147     dc->reset = scsi_disk_reset;
3148     device_class_set_props(dc, scsi_disk_properties);
3149     dc->vmsd  = &vmstate_scsi_disk_state;
3150 }
3151 
3152 static const TypeInfo scsi_disk_info = {
3153     .name          = "scsi-disk",
3154     .parent        = TYPE_SCSI_DISK_BASE,
3155     .class_init    = scsi_disk_class_initfn,
3156 };
3157 
3158 static void scsi_disk_register_types(void)
3159 {
3160     type_register_static(&scsi_disk_base_info);
3161     type_register_static(&scsi_hd_info);
3162     type_register_static(&scsi_cd_info);
3163 #ifdef __linux__
3164     type_register_static(&scsi_block_info);
3165 #endif
3166     type_register_static(&scsi_disk_info);
3167 }
3168 
3169 type_init(scsi_disk_register_types)
3170