xref: /qemu/hw/scsi/scsi-disk.c (revision 668df86ee8076152320345d8e36be7c95ec0a09a)
1 /*
2  * SCSI Device emulation
3  *
4  * Copyright (c) 2006 CodeSourcery.
5  * Based on code by Fabrice Bellard
6  *
7  * Written by Paul Brook
8  * Modifications:
9  *  2009-Dec-12 Artyom Tarasenko : implemented stamdard inquiry for the case
10  *                                 when the allocation length of CDB is smaller
11  *                                 than 36.
12  *  2009-Oct-13 Artyom Tarasenko : implemented the block descriptor in the
13  *                                 MODE SENSE response.
14  *
15  * This code is licensed under the LGPL.
16  *
17  * Note that this file only handles the SCSI architecture model and device
18  * commands.  Emulation of interface/link layer protocols is handled by
19  * the host adapter emulator.
20  */
21 
22 #include "qemu/osdep.h"
23 #include "qemu/units.h"
24 #include "qapi/error.h"
25 #include "qemu/error-report.h"
26 #include "qemu/main-loop.h"
27 #include "qemu/module.h"
28 #include "qemu/hw-version.h"
29 #include "qemu/memalign.h"
30 #include "hw/scsi/scsi.h"
31 #include "migration/qemu-file-types.h"
32 #include "migration/vmstate.h"
33 #include "hw/scsi/emulation.h"
34 #include "scsi/constants.h"
35 #include "system/arch_init.h"
36 #include "system/block-backend.h"
37 #include "system/blockdev.h"
38 #include "hw/block/block.h"
39 #include "hw/qdev-properties.h"
40 #include "hw/qdev-properties-system.h"
41 #include "system/dma.h"
42 #include "system/system.h"
43 #include "qemu/cutils.h"
44 #include "trace.h"
45 #include "qom/object.h"
46 
47 #ifdef __linux
48 #include <scsi/sg.h>
49 #endif
50 
51 #define SCSI_WRITE_SAME_MAX         (512 * KiB)
52 #define SCSI_DMA_BUF_SIZE           (128 * KiB)
53 #define SCSI_MAX_INQUIRY_LEN        256
54 #define SCSI_MAX_MODE_LEN           256
55 
56 #define DEFAULT_DISCARD_GRANULARITY (4 * KiB)
57 #define DEFAULT_MAX_UNMAP_SIZE      (1 * GiB)
58 #define DEFAULT_MAX_IO_SIZE         INT_MAX     /* 2 GB - 1 block */
59 
60 #define TYPE_SCSI_DISK_BASE         "scsi-disk-base"
61 
62 #define MAX_SERIAL_LEN              36
63 #define MAX_SERIAL_LEN_FOR_DEVID    20
64 
65 OBJECT_DECLARE_TYPE(SCSIDiskState, SCSIDiskClass, SCSI_DISK_BASE)
66 
67 struct SCSIDiskClass {
68     SCSIDeviceClass parent_class;
69     /*
70      * Callbacks receive ret == 0 for success. Errors are represented either as
71      * negative errno values, or as positive SAM status codes. For host_status
72      * errors, the function passes ret == -ENODEV and sets the host_status field
73      * of the SCSIRequest.
74      */
75     DMAIOFunc       *dma_readv;
76     DMAIOFunc       *dma_writev;
77     bool            (*need_fua)(SCSICommand *cmd);
78     void            (*update_sense)(SCSIRequest *r);
79 };
80 
81 typedef struct SCSIDiskReq {
82     SCSIRequest req;
83     /* Both sector and sector_count are in terms of BDRV_SECTOR_SIZE bytes.  */
84     uint64_t sector;
85     uint32_t sector_count;
86     uint32_t buflen;
87     bool started;
88     bool need_fua;
89     struct iovec iov;
90     QEMUIOVector qiov;
91     BlockAcctCookie acct;
92 } SCSIDiskReq;
93 
94 #define SCSI_DISK_F_REMOVABLE             0
95 #define SCSI_DISK_F_DPOFUA                1
96 #define SCSI_DISK_F_NO_REMOVABLE_DEVOPS   2
97 
98 struct SCSIDiskState {
99     SCSIDevice qdev;
100     uint32_t features;
101     bool media_changed;
102     bool media_event;
103     bool eject_request;
104     uint16_t port_index;
105     uint64_t max_unmap_size;
106     uint64_t max_io_size;
107     uint32_t quirks;
108     char *version;
109     char *serial;
110     char *vendor;
111     char *product;
112     char *device_id;
113     char *loadparm;     /* only for s390x */
114     bool tray_open;
115     bool tray_locked;
116     /*
117      * 0x0000        - rotation rate not reported
118      * 0x0001        - non-rotating medium (SSD)
119      * 0x0002-0x0400 - reserved
120      * 0x0401-0xffe  - rotations per minute
121      * 0xffff        - reserved
122      */
123     uint16_t rotation_rate;
124     bool migrate_emulated_scsi_request;
125 };
126 
scsi_free_request(SCSIRequest * req)127 static void scsi_free_request(SCSIRequest *req)
128 {
129     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
130 
131     qemu_vfree(r->iov.iov_base);
132 }
133 
134 /* Helper function for command completion with sense.  */
scsi_check_condition(SCSIDiskReq * r,SCSISense sense)135 static void scsi_check_condition(SCSIDiskReq *r, SCSISense sense)
136 {
137     trace_scsi_disk_check_condition(r->req.tag, sense.key, sense.asc,
138                                     sense.ascq);
139     scsi_req_build_sense(&r->req, sense);
140     scsi_req_complete(&r->req, CHECK_CONDITION);
141 }
142 
scsi_init_iovec(SCSIDiskReq * r,size_t size)143 static void scsi_init_iovec(SCSIDiskReq *r, size_t size)
144 {
145     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
146 
147     if (!r->iov.iov_base) {
148         r->buflen = size;
149         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
150     }
151     r->iov.iov_len = MIN(r->sector_count * BDRV_SECTOR_SIZE, r->buflen);
152     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
153 }
154 
scsi_disk_save_request(QEMUFile * f,SCSIRequest * req)155 static void scsi_disk_save_request(QEMUFile *f, SCSIRequest *req)
156 {
157     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
158 
159     qemu_put_be64s(f, &r->sector);
160     qemu_put_be32s(f, &r->sector_count);
161     qemu_put_be32s(f, &r->buflen);
162     if (r->buflen) {
163         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
164             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
165         } else if (!req->retry) {
166             uint32_t len = r->iov.iov_len;
167             qemu_put_be32s(f, &len);
168             qemu_put_buffer(f, r->iov.iov_base, r->iov.iov_len);
169         }
170     }
171 }
172 
scsi_disk_emulate_save_request(QEMUFile * f,SCSIRequest * req)173 static void scsi_disk_emulate_save_request(QEMUFile *f, SCSIRequest *req)
174 {
175     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
176 
177     if (s->migrate_emulated_scsi_request) {
178         scsi_disk_save_request(f, req);
179     }
180 }
181 
scsi_disk_load_request(QEMUFile * f,SCSIRequest * req)182 static void scsi_disk_load_request(QEMUFile *f, SCSIRequest *req)
183 {
184     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
185 
186     qemu_get_be64s(f, &r->sector);
187     qemu_get_be32s(f, &r->sector_count);
188     qemu_get_be32s(f, &r->buflen);
189     if (r->buflen) {
190         scsi_init_iovec(r, r->buflen);
191         if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
192             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
193         } else if (!r->req.retry) {
194             uint32_t len;
195             qemu_get_be32s(f, &len);
196             r->iov.iov_len = len;
197             assert(r->iov.iov_len <= r->buflen);
198             qemu_get_buffer(f, r->iov.iov_base, r->iov.iov_len);
199         }
200     }
201 
202     qemu_iovec_init_external(&r->qiov, &r->iov, 1);
203 }
204 
scsi_disk_emulate_load_request(QEMUFile * f,SCSIRequest * req)205 static void scsi_disk_emulate_load_request(QEMUFile *f, SCSIRequest *req)
206 {
207     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
208 
209     if (s->migrate_emulated_scsi_request) {
210         scsi_disk_load_request(f, req);
211     }
212 }
213 
214 /*
215  * scsi_handle_rw_error has two return values.  False means that the error
216  * must be ignored, true means that the error has been processed and the
217  * caller should not do anything else for this request.  Note that
218  * scsi_handle_rw_error always manages its reference counts, independent
219  * of the return value.
220  */
scsi_handle_rw_error(SCSIDiskReq * r,int ret,bool acct_failed)221 static bool scsi_handle_rw_error(SCSIDiskReq *r, int ret, bool acct_failed)
222 {
223     bool is_read = (r->req.cmd.mode == SCSI_XFER_FROM_DEV);
224     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
225     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
226     SCSISense sense = SENSE_CODE(NO_SENSE);
227     int16_t host_status;
228     int error;
229     bool req_has_sense = false;
230     BlockErrorAction action;
231     int status;
232 
233     /*
234      * host_status should only be set for SG_IO requests that came back with a
235      * host_status error in scsi_block_sgio_complete(). This error path passes
236      * -ENODEV as the return value.
237      *
238      * Reset host_status in the request because we may still want to complete
239      * the request successfully with the 'stop' or 'ignore' error policy.
240      */
241     host_status = r->req.host_status;
242     if (host_status != -1) {
243         assert(ret == -ENODEV);
244         r->req.host_status = -1;
245     }
246 
247     if (ret < 0) {
248         status = scsi_sense_from_errno(-ret, &sense);
249         error = -ret;
250     } else {
251         /* A passthrough command has completed with nonzero status.  */
252         status = ret;
253         switch (status) {
254         case CHECK_CONDITION:
255             req_has_sense = true;
256             error = scsi_sense_buf_to_errno(r->req.sense, sizeof(r->req.sense));
257             break;
258         case RESERVATION_CONFLICT:
259             /*
260              * Don't apply the error policy, always report to the guest.
261              *
262              * This is a passthrough code path, so it's not a backend error, but
263              * a response to an invalid guest request.
264              *
265              * Windows Failover Cluster validation intentionally sends invalid
266              * requests to verify that reservations work as intended. It is
267              * crucial that it sees the resulting errors.
268              *
269              * Treating a reservation conflict as a guest-side error is obvious
270              * when a pr-manager is in use. Without one, the situation is less
271              * clear, but there might be nothing that can be fixed on the host
272              * (like in the above example), and we don't want to be stuck in a
273              * loop where resuming the VM and retrying the request immediately
274              * stops it again. So always reporting is still the safer option in
275              * this case, too.
276              */
277             error = 0;
278             break;
279         default:
280             error = EINVAL;
281             break;
282         }
283     }
284 
285     /*
286      * Check whether the error has to be handled by the guest or should
287      * rather follow the rerror=/werror= settings.  Guest-handled errors
288      * are usually retried immediately, so do not post them to QMP and
289      * do not account them as failed I/O.
290      */
291     if (!error || (req_has_sense &&
292                    scsi_sense_buf_is_guest_recoverable(r->req.sense,
293                                                        sizeof(r->req.sense)))) {
294         action = BLOCK_ERROR_ACTION_REPORT;
295         acct_failed = false;
296     } else {
297         action = blk_get_error_action(s->qdev.conf.blk, is_read, error);
298         blk_error_action(s->qdev.conf.blk, action, is_read, error);
299     }
300 
301     switch (action) {
302     case BLOCK_ERROR_ACTION_REPORT:
303         if (acct_failed) {
304             block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
305         }
306         if (host_status != -1) {
307             scsi_req_complete_failed(&r->req, host_status);
308             return true;
309         }
310         if (req_has_sense) {
311             sdc->update_sense(&r->req);
312         } else if (status == CHECK_CONDITION) {
313             scsi_req_build_sense(&r->req, sense);
314         }
315         scsi_req_complete(&r->req, status);
316         return true;
317 
318     case BLOCK_ERROR_ACTION_IGNORE:
319         return false;
320 
321     case BLOCK_ERROR_ACTION_STOP:
322         scsi_req_retry(&r->req);
323         return true;
324 
325     default:
326         g_assert_not_reached();
327     }
328 }
329 
scsi_disk_req_check_error(SCSIDiskReq * r,int ret,bool acct_failed)330 static bool scsi_disk_req_check_error(SCSIDiskReq *r, int ret, bool acct_failed)
331 {
332     if (r->req.io_canceled) {
333         scsi_req_cancel_complete(&r->req);
334         return true;
335     }
336 
337     if (ret != 0) {
338         return scsi_handle_rw_error(r, ret, acct_failed);
339     }
340 
341     return false;
342 }
343 
scsi_aio_complete(void * opaque,int ret)344 static void scsi_aio_complete(void *opaque, int ret)
345 {
346     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
347     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
348 
349     /* The request must run in its AioContext */
350     assert(r->req.ctx == qemu_get_current_aio_context());
351 
352     assert(r->req.aiocb != NULL);
353     r->req.aiocb = NULL;
354 
355     if (scsi_disk_req_check_error(r, ret, true)) {
356         goto done;
357     }
358 
359     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
360     scsi_req_complete(&r->req, GOOD);
361 
362 done:
363     scsi_req_unref(&r->req);
364 }
365 
scsi_is_cmd_fua(SCSICommand * cmd)366 static bool scsi_is_cmd_fua(SCSICommand *cmd)
367 {
368     switch (cmd->buf[0]) {
369     case READ_10:
370     case READ_12:
371     case READ_16:
372     case WRITE_10:
373     case WRITE_12:
374     case WRITE_16:
375         return (cmd->buf[1] & 8) != 0;
376 
377     case VERIFY_10:
378     case VERIFY_12:
379     case VERIFY_16:
380     case WRITE_VERIFY_10:
381     case WRITE_VERIFY_12:
382     case WRITE_VERIFY_16:
383         return true;
384 
385     case READ_6:
386     case WRITE_6:
387     default:
388         return false;
389     }
390 }
391 
scsi_dma_complete_noio(SCSIDiskReq * r,int ret)392 static void scsi_dma_complete_noio(SCSIDiskReq *r, int ret)
393 {
394     assert(r->req.aiocb == NULL);
395     if (scsi_disk_req_check_error(r, ret, ret > 0)) {
396         goto done;
397     }
398 
399     r->sector += r->sector_count;
400     r->sector_count = 0;
401     scsi_req_complete(&r->req, GOOD);
402 
403 done:
404     scsi_req_unref(&r->req);
405 }
406 
scsi_dma_complete(void * opaque,int ret)407 static void scsi_dma_complete(void *opaque, int ret)
408 {
409     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
410     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
411 
412     assert(r->req.aiocb != NULL);
413     r->req.aiocb = NULL;
414 
415     /* ret > 0 is accounted for in scsi_disk_req_check_error() */
416     if (ret < 0) {
417         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
418     } else if (ret == 0) {
419         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
420     }
421     scsi_dma_complete_noio(r, ret);
422 }
423 
scsi_read_complete_noio(SCSIDiskReq * r,int ret)424 static void scsi_read_complete_noio(SCSIDiskReq *r, int ret)
425 {
426     uint32_t n;
427 
428     /* The request must run in its AioContext */
429     assert(r->req.ctx == qemu_get_current_aio_context());
430 
431     assert(r->req.aiocb == NULL);
432     if (scsi_disk_req_check_error(r, ret, ret > 0)) {
433         goto done;
434     }
435 
436     n = r->qiov.size / BDRV_SECTOR_SIZE;
437     r->sector += n;
438     r->sector_count -= n;
439     scsi_req_data(&r->req, r->qiov.size);
440 
441 done:
442     scsi_req_unref(&r->req);
443 }
444 
scsi_read_complete(void * opaque,int ret)445 static void scsi_read_complete(void *opaque, int ret)
446 {
447     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
448     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
449 
450     assert(r->req.aiocb != NULL);
451     r->req.aiocb = NULL;
452 
453     /* ret > 0 is accounted for in scsi_disk_req_check_error() */
454     if (ret < 0) {
455         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
456     } else if (ret == 0) {
457         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
458         trace_scsi_disk_read_complete(r->req.tag, r->qiov.size);
459     }
460     scsi_read_complete_noio(r, ret);
461 }
462 
463 /* Actually issue a read to the block device.  */
scsi_do_read(SCSIDiskReq * r,int ret)464 static void scsi_do_read(SCSIDiskReq *r, int ret)
465 {
466     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
467     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
468 
469     assert (r->req.aiocb == NULL);
470     if (scsi_disk_req_check_error(r, ret, false)) {
471         goto done;
472     }
473 
474     /* The request is used as the AIO opaque value, so add a ref.  */
475     scsi_req_ref(&r->req);
476 
477     if (r->req.sg) {
478         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_READ);
479         r->req.residual -= r->req.sg->size;
480         r->req.aiocb = dma_blk_io(r->req.sg, r->sector << BDRV_SECTOR_BITS,
481                                   BDRV_SECTOR_SIZE,
482                                   sdc->dma_readv, r, scsi_dma_complete, r,
483                                   DMA_DIRECTION_FROM_DEVICE);
484     } else {
485         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
486         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
487                          r->qiov.size, BLOCK_ACCT_READ);
488         r->req.aiocb = sdc->dma_readv(r->sector << BDRV_SECTOR_BITS, &r->qiov,
489                                       scsi_read_complete, r, r);
490     }
491 
492 done:
493     scsi_req_unref(&r->req);
494 }
495 
scsi_do_read_cb(void * opaque,int ret)496 static void scsi_do_read_cb(void *opaque, int ret)
497 {
498     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
499     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
500 
501     assert (r->req.aiocb != NULL);
502     r->req.aiocb = NULL;
503 
504     if (ret < 0) {
505         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
506     } else {
507         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
508     }
509     scsi_do_read(opaque, ret);
510 }
511 
512 /* Read more data from scsi device into buffer.  */
scsi_read_data(SCSIRequest * req)513 static void scsi_read_data(SCSIRequest *req)
514 {
515     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
516     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
517     bool first;
518 
519     trace_scsi_disk_read_data_count(r->sector_count);
520     if (r->sector_count == 0) {
521         /* This also clears the sense buffer for REQUEST SENSE.  */
522         scsi_req_complete(&r->req, GOOD);
523         return;
524     }
525 
526     /* No data transfer may already be in progress */
527     assert(r->req.aiocb == NULL);
528 
529     /* The request is used as the AIO opaque value, so add a ref.  */
530     scsi_req_ref(&r->req);
531     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
532         trace_scsi_disk_read_data_invalid();
533         scsi_read_complete_noio(r, -EINVAL);
534         return;
535     }
536 
537     if (!blk_is_available(req->dev->conf.blk)) {
538         scsi_read_complete_noio(r, -ENOMEDIUM);
539         return;
540     }
541 
542     first = !r->started;
543     r->started = true;
544     if (first && r->need_fua) {
545         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
546                          BLOCK_ACCT_FLUSH);
547         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_do_read_cb, r);
548     } else {
549         scsi_do_read(r, 0);
550     }
551 }
552 
scsi_write_complete_noio(SCSIDiskReq * r,int ret)553 static void scsi_write_complete_noio(SCSIDiskReq *r, int ret)
554 {
555     uint32_t n;
556 
557     /* The request must run in its AioContext */
558     assert(r->req.ctx == qemu_get_current_aio_context());
559 
560     assert (r->req.aiocb == NULL);
561     if (scsi_disk_req_check_error(r, ret, ret > 0)) {
562         goto done;
563     }
564 
565     n = r->qiov.size / BDRV_SECTOR_SIZE;
566     r->sector += n;
567     r->sector_count -= n;
568     if (r->sector_count == 0) {
569         scsi_req_complete(&r->req, GOOD);
570     } else {
571         scsi_init_iovec(r, SCSI_DMA_BUF_SIZE);
572         trace_scsi_disk_write_complete_noio(r->req.tag, r->qiov.size);
573         scsi_req_data(&r->req, r->qiov.size);
574     }
575 
576 done:
577     scsi_req_unref(&r->req);
578 }
579 
scsi_write_complete(void * opaque,int ret)580 static void scsi_write_complete(void * opaque, int ret)
581 {
582     SCSIDiskReq *r = (SCSIDiskReq *)opaque;
583     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
584 
585     assert (r->req.aiocb != NULL);
586     r->req.aiocb = NULL;
587 
588     /* ret > 0 is accounted for in scsi_disk_req_check_error() */
589     if (ret < 0) {
590         block_acct_failed(blk_get_stats(s->qdev.conf.blk), &r->acct);
591     } else if (ret == 0) {
592         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
593     }
594     scsi_write_complete_noio(r, ret);
595 }
596 
scsi_write_data(SCSIRequest * req)597 static void scsi_write_data(SCSIRequest *req)
598 {
599     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
600     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
601     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
602     BlockCompletionFunc *cb;
603 
604     /* No data transfer may already be in progress */
605     assert(r->req.aiocb == NULL);
606 
607     /* The request is used as the AIO opaque value, so add a ref.  */
608     scsi_req_ref(&r->req);
609     if (r->req.cmd.mode != SCSI_XFER_TO_DEV) {
610         trace_scsi_disk_write_data_invalid();
611         scsi_write_complete_noio(r, -EINVAL);
612         return;
613     }
614 
615     if (!r->req.sg && !r->qiov.size) {
616         /* Called for the first time.  Ask the driver to send us more data.  */
617         r->started = true;
618         scsi_write_complete_noio(r, 0);
619         return;
620     }
621     if (!blk_is_available(req->dev->conf.blk)) {
622         scsi_write_complete_noio(r, -ENOMEDIUM);
623         return;
624     }
625 
626     if (r->req.cmd.buf[0] == VERIFY_10 || r->req.cmd.buf[0] == VERIFY_12 ||
627         r->req.cmd.buf[0] == VERIFY_16) {
628         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
629                          BLOCK_ACCT_FLUSH);
630         cb = r->req.sg ? scsi_dma_complete : scsi_write_complete;
631         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, cb, r);
632         return;
633     }
634 
635     if (r->req.sg) {
636         dma_acct_start(s->qdev.conf.blk, &r->acct, r->req.sg, BLOCK_ACCT_WRITE);
637         r->req.residual -= r->req.sg->size;
638         r->req.aiocb = dma_blk_io(r->req.sg, r->sector << BDRV_SECTOR_BITS,
639                                   BDRV_SECTOR_SIZE,
640                                   sdc->dma_writev, r, scsi_dma_complete, r,
641                                   DMA_DIRECTION_TO_DEVICE);
642     } else {
643         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
644                          r->qiov.size, BLOCK_ACCT_WRITE);
645         r->req.aiocb = sdc->dma_writev(r->sector << BDRV_SECTOR_BITS, &r->qiov,
646                                        scsi_write_complete, r, r);
647     }
648 }
649 
650 /* Return a pointer to the data buffer.  */
scsi_get_buf(SCSIRequest * req)651 static uint8_t *scsi_get_buf(SCSIRequest *req)
652 {
653     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
654 
655     return (uint8_t *)r->iov.iov_base;
656 }
657 
scsi_disk_emulate_vpd_page(SCSIRequest * req,uint8_t * outbuf)658 static int scsi_disk_emulate_vpd_page(SCSIRequest *req, uint8_t *outbuf)
659 {
660     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
661     uint8_t page_code = req->cmd.buf[2];
662     int start, buflen = 0;
663 
664     outbuf[buflen++] = s->qdev.type & 0x1f;
665     outbuf[buflen++] = page_code;
666     outbuf[buflen++] = 0x00;
667     outbuf[buflen++] = 0x00;
668     start = buflen;
669 
670     switch (page_code) {
671     case 0x00: /* Supported page codes, mandatory */
672     {
673         trace_scsi_disk_emulate_vpd_page_00(req->cmd.xfer);
674         outbuf[buflen++] = 0x00; /* list of supported pages (this page) */
675         if (s->serial) {
676             outbuf[buflen++] = 0x80; /* unit serial number */
677         }
678         outbuf[buflen++] = 0x83; /* device identification */
679         if (s->qdev.type == TYPE_DISK) {
680             outbuf[buflen++] = 0xb0; /* block limits */
681             outbuf[buflen++] = 0xb1; /* block device characteristics */
682             outbuf[buflen++] = 0xb2; /* thin provisioning */
683         }
684         break;
685     }
686     case 0x80: /* Device serial number, optional */
687     {
688         int l;
689 
690         if (!s->serial) {
691             trace_scsi_disk_emulate_vpd_page_80_not_supported();
692             return -1;
693         }
694 
695         l = strlen(s->serial);
696         if (l > MAX_SERIAL_LEN) {
697             l = MAX_SERIAL_LEN;
698         }
699 
700         trace_scsi_disk_emulate_vpd_page_80(req->cmd.xfer);
701         memcpy(outbuf + buflen, s->serial, l);
702         buflen += l;
703         break;
704     }
705 
706     case 0x83: /* Device identification page, mandatory */
707     {
708         int id_len = s->device_id ? MIN(strlen(s->device_id), 255 - 8) : 0;
709 
710         trace_scsi_disk_emulate_vpd_page_83(req->cmd.xfer);
711 
712         if (id_len) {
713             outbuf[buflen++] = 0x2; /* ASCII */
714             outbuf[buflen++] = 0;   /* not officially assigned */
715             outbuf[buflen++] = 0;   /* reserved */
716             outbuf[buflen++] = id_len; /* length of data following */
717             memcpy(outbuf + buflen, s->device_id, id_len);
718             buflen += id_len;
719         }
720 
721         if (s->qdev.wwn) {
722             outbuf[buflen++] = 0x1; /* Binary */
723             outbuf[buflen++] = 0x3; /* NAA */
724             outbuf[buflen++] = 0;   /* reserved */
725             outbuf[buflen++] = 8;
726             stq_be_p(&outbuf[buflen], s->qdev.wwn);
727             buflen += 8;
728         }
729 
730         if (s->qdev.port_wwn) {
731             outbuf[buflen++] = 0x61; /* SAS / Binary */
732             outbuf[buflen++] = 0x93; /* PIV / Target port / NAA */
733             outbuf[buflen++] = 0;    /* reserved */
734             outbuf[buflen++] = 8;
735             stq_be_p(&outbuf[buflen], s->qdev.port_wwn);
736             buflen += 8;
737         }
738 
739         if (s->port_index) {
740             outbuf[buflen++] = 0x61; /* SAS / Binary */
741 
742             /* PIV/Target port/relative target port */
743             outbuf[buflen++] = 0x94;
744 
745             outbuf[buflen++] = 0;    /* reserved */
746             outbuf[buflen++] = 4;
747             stw_be_p(&outbuf[buflen + 2], s->port_index);
748             buflen += 4;
749         }
750         break;
751     }
752     case 0xb0: /* block limits */
753     {
754         SCSIBlockLimits bl = {};
755 
756         if (s->qdev.type == TYPE_ROM) {
757             trace_scsi_disk_emulate_vpd_page_b0_not_supported();
758             return -1;
759         }
760         bl.wsnz = 1;
761         bl.unmap_sectors =
762             s->qdev.conf.discard_granularity / s->qdev.blocksize;
763         bl.min_io_size =
764             s->qdev.conf.min_io_size / s->qdev.blocksize;
765         bl.opt_io_size =
766             s->qdev.conf.opt_io_size / s->qdev.blocksize;
767         bl.max_unmap_sectors =
768             s->max_unmap_size / s->qdev.blocksize;
769         bl.max_io_sectors =
770             s->max_io_size / s->qdev.blocksize;
771         /* 255 descriptors fit in 4 KiB with an 8-byte header */
772         bl.max_unmap_descr = 255;
773 
774         if (s->qdev.type == TYPE_DISK) {
775             int max_transfer_blk = blk_get_max_transfer(s->qdev.conf.blk);
776             int max_io_sectors_blk =
777                 max_transfer_blk / s->qdev.blocksize;
778 
779             bl.max_io_sectors =
780                 MIN_NON_ZERO(max_io_sectors_blk, bl.max_io_sectors);
781         }
782         buflen += scsi_emulate_block_limits(outbuf + buflen, &bl);
783         break;
784     }
785     case 0xb1: /* block device characteristics */
786     {
787         buflen = 0x40;
788         outbuf[4] = (s->rotation_rate >> 8) & 0xff;
789         outbuf[5] = s->rotation_rate & 0xff;
790         outbuf[6] = 0; /* PRODUCT TYPE */
791         outbuf[7] = 0; /* WABEREQ | WACEREQ | NOMINAL FORM FACTOR */
792         outbuf[8] = 0; /* VBULS */
793         break;
794     }
795     case 0xb2: /* thin provisioning */
796     {
797         buflen = 8;
798         outbuf[4] = 0;
799         outbuf[5] = 0xe0; /* unmap & write_same 10/16 all supported */
800         outbuf[6] = s->qdev.conf.discard_granularity ? 2 : 1;
801         outbuf[7] = 0;
802         break;
803     }
804     default:
805         return -1;
806     }
807     /* done with EVPD */
808     assert(buflen - start <= 255);
809     outbuf[start - 1] = buflen - start;
810     return buflen;
811 }
812 
scsi_disk_emulate_inquiry(SCSIRequest * req,uint8_t * outbuf)813 static int scsi_disk_emulate_inquiry(SCSIRequest *req, uint8_t *outbuf)
814 {
815     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
816     int buflen = 0;
817 
818     if (req->cmd.buf[1] & 0x1) {
819         /* Vital product data */
820         return scsi_disk_emulate_vpd_page(req, outbuf);
821     }
822 
823     /* Standard INQUIRY data */
824     if (req->cmd.buf[2] != 0) {
825         return -1;
826     }
827 
828     /* PAGE CODE == 0 */
829     buflen = req->cmd.xfer;
830     if (buflen > SCSI_MAX_INQUIRY_LEN) {
831         buflen = SCSI_MAX_INQUIRY_LEN;
832     }
833 
834     outbuf[0] = s->qdev.type & 0x1f;
835     outbuf[1] = (s->features & (1 << SCSI_DISK_F_REMOVABLE)) ? 0x80 : 0;
836 
837     strpadcpy((char *) &outbuf[16], 16, s->product, ' ');
838     strpadcpy((char *) &outbuf[8], 8, s->vendor, ' ');
839 
840     memset(&outbuf[32], 0, 4);
841     memcpy(&outbuf[32], s->version, MIN(4, strlen(s->version)));
842     /*
843      * We claim conformance to SPC-3, which is required for guests
844      * to ask for modern features like READ CAPACITY(16) or the
845      * block characteristics VPD page by default.  Not all of SPC-3
846      * is actually implemented, but we're good enough.
847      */
848     outbuf[2] = s->qdev.default_scsi_version;
849     outbuf[3] = 2 | 0x10; /* Format 2, HiSup */
850 
851     if (buflen > 36) {
852         outbuf[4] = buflen - 5; /* Additional Length = (Len - 1) - 4 */
853     } else {
854         /* If the allocation length of CDB is too small,
855                the additional length is not adjusted */
856         outbuf[4] = 36 - 5;
857     }
858 
859     /* Sync data transfer and TCQ.  */
860     outbuf[7] = 0x10 | (req->bus->info->tcq ? 0x02 : 0);
861     return buflen;
862 }
863 
media_is_dvd(SCSIDiskState * s)864 static inline bool media_is_dvd(SCSIDiskState *s)
865 {
866     uint64_t nb_sectors;
867     if (s->qdev.type != TYPE_ROM) {
868         return false;
869     }
870     if (!blk_is_available(s->qdev.conf.blk)) {
871         return false;
872     }
873     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
874     return nb_sectors > CD_MAX_SECTORS;
875 }
876 
media_is_cd(SCSIDiskState * s)877 static inline bool media_is_cd(SCSIDiskState *s)
878 {
879     uint64_t nb_sectors;
880     if (s->qdev.type != TYPE_ROM) {
881         return false;
882     }
883     if (!blk_is_available(s->qdev.conf.blk)) {
884         return false;
885     }
886     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
887     return nb_sectors <= CD_MAX_SECTORS;
888 }
889 
scsi_read_disc_information(SCSIDiskState * s,SCSIDiskReq * r,uint8_t * outbuf)890 static int scsi_read_disc_information(SCSIDiskState *s, SCSIDiskReq *r,
891                                       uint8_t *outbuf)
892 {
893     uint8_t type = r->req.cmd.buf[1] & 7;
894 
895     if (s->qdev.type != TYPE_ROM) {
896         return -1;
897     }
898 
899     /* Types 1/2 are only defined for Blu-Ray.  */
900     if (type != 0) {
901         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
902         return -1;
903     }
904 
905     memset(outbuf, 0, 34);
906     outbuf[1] = 32;
907     outbuf[2] = 0xe; /* last session complete, disc finalized */
908     outbuf[3] = 1;   /* first track on disc */
909     outbuf[4] = 1;   /* # of sessions */
910     outbuf[5] = 1;   /* first track of last session */
911     outbuf[6] = 1;   /* last track of last session */
912     outbuf[7] = 0x20; /* unrestricted use */
913     outbuf[8] = 0x00; /* CD-ROM or DVD-ROM */
914     /* 9-10-11: most significant byte corresponding bytes 4-5-6 */
915     /* 12-23: not meaningful for CD-ROM or DVD-ROM */
916     /* 24-31: disc bar code */
917     /* 32: disc application code */
918     /* 33: number of OPC tables */
919 
920     return 34;
921 }
922 
scsi_read_dvd_structure(SCSIDiskState * s,SCSIDiskReq * r,uint8_t * outbuf)923 static int scsi_read_dvd_structure(SCSIDiskState *s, SCSIDiskReq *r,
924                                    uint8_t *outbuf)
925 {
926     static const int rds_caps_size[5] = {
927         [0] = 2048 + 4,
928         [1] = 4 + 4,
929         [3] = 188 + 4,
930         [4] = 2048 + 4,
931     };
932 
933     uint8_t media = r->req.cmd.buf[1];
934     uint8_t layer = r->req.cmd.buf[6];
935     uint8_t format = r->req.cmd.buf[7];
936     int size = -1;
937 
938     if (s->qdev.type != TYPE_ROM) {
939         return -1;
940     }
941     if (media != 0) {
942         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
943         return -1;
944     }
945 
946     if (format != 0xff) {
947         if (!blk_is_available(s->qdev.conf.blk)) {
948             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
949             return -1;
950         }
951         if (media_is_cd(s)) {
952             scsi_check_condition(r, SENSE_CODE(INCOMPATIBLE_FORMAT));
953             return -1;
954         }
955         if (format >= ARRAY_SIZE(rds_caps_size)) {
956             return -1;
957         }
958         size = rds_caps_size[format];
959         memset(outbuf, 0, size);
960     }
961 
962     switch (format) {
963     case 0x00: {
964         /* Physical format information */
965         uint64_t nb_sectors;
966         if (layer != 0) {
967             goto fail;
968         }
969         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
970 
971         outbuf[4] = 1;   /* DVD-ROM, part version 1 */
972         outbuf[5] = 0xf; /* 120mm disc, minimum rate unspecified */
973         outbuf[6] = 1;   /* one layer, read-only (per MMC-2 spec) */
974         outbuf[7] = 0;   /* default densities */
975 
976         stl_be_p(&outbuf[12], (nb_sectors >> 2) - 1); /* end sector */
977         stl_be_p(&outbuf[16], (nb_sectors >> 2) - 1); /* l0 end sector */
978         break;
979     }
980 
981     case 0x01: /* DVD copyright information, all zeros */
982         break;
983 
984     case 0x03: /* BCA information - invalid field for no BCA info */
985         return -1;
986 
987     case 0x04: /* DVD disc manufacturing information, all zeros */
988         break;
989 
990     case 0xff: { /* List capabilities */
991         int i;
992         size = 4;
993         for (i = 0; i < ARRAY_SIZE(rds_caps_size); i++) {
994             if (!rds_caps_size[i]) {
995                 continue;
996             }
997             outbuf[size] = i;
998             outbuf[size + 1] = 0x40; /* Not writable, readable */
999             stw_be_p(&outbuf[size + 2], rds_caps_size[i]);
1000             size += 4;
1001         }
1002         break;
1003      }
1004 
1005     default:
1006         return -1;
1007     }
1008 
1009     /* Size of buffer, not including 2 byte size field */
1010     stw_be_p(outbuf, size - 2);
1011     return size;
1012 
1013 fail:
1014     return -1;
1015 }
1016 
scsi_event_status_media(SCSIDiskState * s,uint8_t * outbuf)1017 static int scsi_event_status_media(SCSIDiskState *s, uint8_t *outbuf)
1018 {
1019     uint8_t event_code, media_status;
1020 
1021     media_status = 0;
1022     if (s->tray_open) {
1023         media_status = MS_TRAY_OPEN;
1024     } else if (blk_is_inserted(s->qdev.conf.blk)) {
1025         media_status = MS_MEDIA_PRESENT;
1026     }
1027 
1028     /* Event notification descriptor */
1029     event_code = MEC_NO_CHANGE;
1030     if (media_status != MS_TRAY_OPEN) {
1031         if (s->media_event) {
1032             event_code = MEC_NEW_MEDIA;
1033             s->media_event = false;
1034         } else if (s->eject_request) {
1035             event_code = MEC_EJECT_REQUESTED;
1036             s->eject_request = false;
1037         }
1038     }
1039 
1040     outbuf[0] = event_code;
1041     outbuf[1] = media_status;
1042 
1043     /* These fields are reserved, just clear them. */
1044     outbuf[2] = 0;
1045     outbuf[3] = 0;
1046     return 4;
1047 }
1048 
scsi_get_event_status_notification(SCSIDiskState * s,SCSIDiskReq * r,uint8_t * outbuf)1049 static int scsi_get_event_status_notification(SCSIDiskState *s, SCSIDiskReq *r,
1050                                               uint8_t *outbuf)
1051 {
1052     int size;
1053     uint8_t *buf = r->req.cmd.buf;
1054     uint8_t notification_class_request = buf[4];
1055     if (s->qdev.type != TYPE_ROM) {
1056         return -1;
1057     }
1058     if ((buf[1] & 1) == 0) {
1059         /* asynchronous */
1060         return -1;
1061     }
1062 
1063     size = 4;
1064     outbuf[0] = outbuf[1] = 0;
1065     outbuf[3] = 1 << GESN_MEDIA; /* supported events */
1066     if (notification_class_request & (1 << GESN_MEDIA)) {
1067         outbuf[2] = GESN_MEDIA;
1068         size += scsi_event_status_media(s, &outbuf[size]);
1069     } else {
1070         outbuf[2] = 0x80;
1071     }
1072     stw_be_p(outbuf, size - 4);
1073     return size;
1074 }
1075 
scsi_get_configuration(SCSIDiskState * s,uint8_t * outbuf)1076 static int scsi_get_configuration(SCSIDiskState *s, uint8_t *outbuf)
1077 {
1078     int current;
1079 
1080     if (s->qdev.type != TYPE_ROM) {
1081         return -1;
1082     }
1083 
1084     if (media_is_dvd(s)) {
1085         current = MMC_PROFILE_DVD_ROM;
1086     } else if (media_is_cd(s)) {
1087         current = MMC_PROFILE_CD_ROM;
1088     } else {
1089         current = MMC_PROFILE_NONE;
1090     }
1091 
1092     memset(outbuf, 0, 40);
1093     stl_be_p(&outbuf[0], 36); /* Bytes after the data length field */
1094     stw_be_p(&outbuf[6], current);
1095     /* outbuf[8] - outbuf[19]: Feature 0 - Profile list */
1096     outbuf[10] = 0x03; /* persistent, current */
1097     outbuf[11] = 8; /* two profiles */
1098     stw_be_p(&outbuf[12], MMC_PROFILE_DVD_ROM);
1099     outbuf[14] = (current == MMC_PROFILE_DVD_ROM);
1100     stw_be_p(&outbuf[16], MMC_PROFILE_CD_ROM);
1101     outbuf[18] = (current == MMC_PROFILE_CD_ROM);
1102     /* outbuf[20] - outbuf[31]: Feature 1 - Core feature */
1103     stw_be_p(&outbuf[20], 1);
1104     outbuf[22] = 0x08 | 0x03; /* version 2, persistent, current */
1105     outbuf[23] = 8;
1106     stl_be_p(&outbuf[24], 1); /* SCSI */
1107     outbuf[28] = 1; /* DBE = 1, mandatory */
1108     /* outbuf[32] - outbuf[39]: Feature 3 - Removable media feature */
1109     stw_be_p(&outbuf[32], 3);
1110     outbuf[34] = 0x08 | 0x03; /* version 2, persistent, current */
1111     outbuf[35] = 4;
1112     outbuf[36] = 0x39; /* tray, load=1, eject=1, unlocked at powerup, lock=1 */
1113     /* TODO: Random readable, CD read, DVD read, drive serial number,
1114        power management */
1115     return 40;
1116 }
1117 
scsi_emulate_mechanism_status(SCSIDiskState * s,uint8_t * outbuf)1118 static int scsi_emulate_mechanism_status(SCSIDiskState *s, uint8_t *outbuf)
1119 {
1120     if (s->qdev.type != TYPE_ROM) {
1121         return -1;
1122     }
1123     memset(outbuf, 0, 8);
1124     outbuf[5] = 1; /* CD-ROM */
1125     return 8;
1126 }
1127 
mode_sense_page(SCSIDiskState * s,int page,uint8_t ** p_outbuf,int page_control)1128 static int mode_sense_page(SCSIDiskState *s, int page, uint8_t **p_outbuf,
1129                            int page_control)
1130 {
1131     static const int mode_sense_valid[0x3f] = {
1132         [MODE_PAGE_VENDOR_SPECIFIC]        = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1133         [MODE_PAGE_HD_GEOMETRY]            = (1 << TYPE_DISK),
1134         [MODE_PAGE_FLEXIBLE_DISK_GEOMETRY] = (1 << TYPE_DISK),
1135         [MODE_PAGE_CACHING]                = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1136         [MODE_PAGE_R_W_ERROR]              = (1 << TYPE_DISK) | (1 << TYPE_ROM),
1137         [MODE_PAGE_AUDIO_CTL]              = (1 << TYPE_ROM),
1138         [MODE_PAGE_CAPABILITIES]           = (1 << TYPE_ROM),
1139         [MODE_PAGE_APPLE_VENDOR]           = (1 << TYPE_ROM),
1140     };
1141 
1142     uint8_t *p = *p_outbuf + 2;
1143     int length;
1144 
1145     assert(page < ARRAY_SIZE(mode_sense_valid));
1146     if ((mode_sense_valid[page] & (1 << s->qdev.type)) == 0) {
1147         return -1;
1148     }
1149 
1150     /*
1151      * If Changeable Values are requested, a mask denoting those mode parameters
1152      * that are changeable shall be returned. As we currently don't support
1153      * parameter changes via MODE_SELECT all bits are returned set to zero.
1154      * The buffer was already menset to zero by the caller of this function.
1155      *
1156      * The offsets here are off by two compared to the descriptions in the
1157      * SCSI specs, because those include a 2-byte header.  This is unfortunate,
1158      * but it is done so that offsets are consistent within our implementation
1159      * of MODE SENSE and MODE SELECT.  MODE SELECT has to deal with both
1160      * 2-byte and 4-byte headers.
1161      */
1162     switch (page) {
1163     case MODE_PAGE_HD_GEOMETRY:
1164         length = 0x16;
1165         if (page_control == 1) { /* Changeable Values */
1166             break;
1167         }
1168         /* if a geometry hint is available, use it */
1169         p[0] = (s->qdev.conf.cyls >> 16) & 0xff;
1170         p[1] = (s->qdev.conf.cyls >> 8) & 0xff;
1171         p[2] = s->qdev.conf.cyls & 0xff;
1172         p[3] = s->qdev.conf.heads & 0xff;
1173         /* Write precomp start cylinder, disabled */
1174         p[4] = (s->qdev.conf.cyls >> 16) & 0xff;
1175         p[5] = (s->qdev.conf.cyls >> 8) & 0xff;
1176         p[6] = s->qdev.conf.cyls & 0xff;
1177         /* Reduced current start cylinder, disabled */
1178         p[7] = (s->qdev.conf.cyls >> 16) & 0xff;
1179         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1180         p[9] = s->qdev.conf.cyls & 0xff;
1181         /* Device step rate [ns], 200ns */
1182         p[10] = 0;
1183         p[11] = 200;
1184         /* Landing zone cylinder */
1185         p[12] = 0xff;
1186         p[13] =  0xff;
1187         p[14] = 0xff;
1188         /* Medium rotation rate [rpm], 5400 rpm */
1189         p[18] = (5400 >> 8) & 0xff;
1190         p[19] = 5400 & 0xff;
1191         break;
1192 
1193     case MODE_PAGE_FLEXIBLE_DISK_GEOMETRY:
1194         length = 0x1e;
1195         if (page_control == 1) { /* Changeable Values */
1196             break;
1197         }
1198         /* Transfer rate [kbit/s], 5Mbit/s */
1199         p[0] = 5000 >> 8;
1200         p[1] = 5000 & 0xff;
1201         /* if a geometry hint is available, use it */
1202         p[2] = s->qdev.conf.heads & 0xff;
1203         p[3] = s->qdev.conf.secs & 0xff;
1204         p[4] = s->qdev.blocksize >> 8;
1205         p[6] = (s->qdev.conf.cyls >> 8) & 0xff;
1206         p[7] = s->qdev.conf.cyls & 0xff;
1207         /* Write precomp start cylinder, disabled */
1208         p[8] = (s->qdev.conf.cyls >> 8) & 0xff;
1209         p[9] = s->qdev.conf.cyls & 0xff;
1210         /* Reduced current start cylinder, disabled */
1211         p[10] = (s->qdev.conf.cyls >> 8) & 0xff;
1212         p[11] = s->qdev.conf.cyls & 0xff;
1213         /* Device step rate [100us], 100us */
1214         p[12] = 0;
1215         p[13] = 1;
1216         /* Device step pulse width [us], 1us */
1217         p[14] = 1;
1218         /* Device head settle delay [100us], 100us */
1219         p[15] = 0;
1220         p[16] = 1;
1221         /* Motor on delay [0.1s], 0.1s */
1222         p[17] = 1;
1223         /* Motor off delay [0.1s], 0.1s */
1224         p[18] = 1;
1225         /* Medium rotation rate [rpm], 5400 rpm */
1226         p[26] = (5400 >> 8) & 0xff;
1227         p[27] = 5400 & 0xff;
1228         break;
1229 
1230     case MODE_PAGE_CACHING:
1231         length = 0x12;
1232         if (page_control == 1 || /* Changeable Values */
1233             blk_enable_write_cache(s->qdev.conf.blk)) {
1234             p[0] = 4; /* WCE */
1235         }
1236         break;
1237 
1238     case MODE_PAGE_R_W_ERROR:
1239         length = 10;
1240         if (page_control == 1) { /* Changeable Values */
1241             if (s->qdev.type == TYPE_ROM) {
1242                 /* Automatic Write Reallocation Enabled */
1243                 p[0] = 0x80;
1244             }
1245             break;
1246         }
1247         p[0] = 0x80; /* Automatic Write Reallocation Enabled */
1248         if (s->qdev.type == TYPE_ROM) {
1249             p[1] = 0x20; /* Read Retry Count */
1250         }
1251         break;
1252 
1253     case MODE_PAGE_AUDIO_CTL:
1254         length = 14;
1255         break;
1256 
1257     case MODE_PAGE_CAPABILITIES:
1258         length = 0x14;
1259         if (page_control == 1) { /* Changeable Values */
1260             break;
1261         }
1262 
1263         p[0] = 0x3b; /* CD-R & CD-RW read */
1264         p[1] = 0; /* Writing not supported */
1265         p[2] = 0x7f; /* Audio, composite, digital out,
1266                         mode 2 form 1&2, multi session */
1267         p[3] = 0xff; /* CD DA, DA accurate, RW supported,
1268                         RW corrected, C2 errors, ISRC,
1269                         UPC, Bar code */
1270         p[4] = 0x2d | (s->tray_locked ? 2 : 0);
1271         /* Locking supported, jumper present, eject, tray */
1272         p[5] = 0; /* no volume & mute control, no
1273                      changer */
1274         p[6] = (50 * 176) >> 8; /* 50x read speed */
1275         p[7] = (50 * 176) & 0xff;
1276         p[8] = 2 >> 8; /* Two volume levels */
1277         p[9] = 2 & 0xff;
1278         p[10] = 2048 >> 8; /* 2M buffer */
1279         p[11] = 2048 & 0xff;
1280         p[12] = (16 * 176) >> 8; /* 16x read speed current */
1281         p[13] = (16 * 176) & 0xff;
1282         p[16] = (16 * 176) >> 8; /* 16x write speed */
1283         p[17] = (16 * 176) & 0xff;
1284         p[18] = (16 * 176) >> 8; /* 16x write speed current */
1285         p[19] = (16 * 176) & 0xff;
1286         break;
1287 
1288      case MODE_PAGE_APPLE_VENDOR:
1289         if (s->quirks & (1 << SCSI_DISK_QUIRK_MODE_PAGE_APPLE_VENDOR)) {
1290             length = 0x1e;
1291             if (page_control == 1) { /* Changeable Values */
1292                 break;
1293             }
1294 
1295             memset(p, 0, length);
1296             strcpy((char *)p + 8, "APPLE COMPUTER, INC   ");
1297             break;
1298         } else {
1299             return -1;
1300         }
1301 
1302     case MODE_PAGE_VENDOR_SPECIFIC:
1303         if (s->qdev.type == TYPE_DISK && (s->quirks &
1304             (1 << SCSI_DISK_QUIRK_MODE_PAGE_VENDOR_SPECIFIC_APPLE))) {
1305             length = 0x2;
1306             if (page_control == 1) { /* Changeable Values */
1307                 p[0] = 0xff;
1308                 p[1] = 0xff;
1309                 break;
1310             }
1311             p[0] = 0;
1312             p[1] = 0;
1313             break;
1314         } else {
1315             return -1;
1316         }
1317 
1318     default:
1319         return -1;
1320     }
1321 
1322     assert(length < 256);
1323     (*p_outbuf)[0] = page;
1324     (*p_outbuf)[1] = length;
1325     *p_outbuf += length + 2;
1326     return length + 2;
1327 }
1328 
scsi_disk_emulate_mode_sense(SCSIDiskReq * r,uint8_t * outbuf)1329 static int scsi_disk_emulate_mode_sense(SCSIDiskReq *r, uint8_t *outbuf)
1330 {
1331     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1332     uint64_t nb_sectors;
1333     bool dbd;
1334     int page, buflen, ret, page_control;
1335     uint8_t *p;
1336     uint8_t dev_specific_param;
1337 
1338     dbd = (r->req.cmd.buf[1] & 0x8) != 0;
1339     page = r->req.cmd.buf[2] & 0x3f;
1340     page_control = (r->req.cmd.buf[2] & 0xc0) >> 6;
1341 
1342     trace_scsi_disk_emulate_mode_sense((r->req.cmd.buf[0] == MODE_SENSE) ? 6 :
1343                                        10, page, r->req.cmd.xfer, page_control);
1344     memset(outbuf, 0, r->req.cmd.xfer);
1345     p = outbuf;
1346 
1347     if (s->qdev.type == TYPE_DISK) {
1348         dev_specific_param = s->features & (1 << SCSI_DISK_F_DPOFUA) ? 0x10 : 0;
1349         if (!blk_is_writable(s->qdev.conf.blk)) {
1350             dev_specific_param |= 0x80; /* Readonly.  */
1351         }
1352     } else {
1353         if (s->quirks & (1 << SCSI_DISK_QUIRK_MODE_SENSE_ROM_USE_DBD)) {
1354             /* Use DBD from the request... */
1355             dev_specific_param = 0x00;
1356 
1357             /*
1358              * ... unless we receive a request for MODE_PAGE_APPLE_VENDOR
1359              * which should never return a block descriptor even though DBD is
1360              * not set, otherwise CDROM detection fails in MacOS
1361              */
1362             if (s->quirks & (1 << SCSI_DISK_QUIRK_MODE_PAGE_APPLE_VENDOR) &&
1363                 page == MODE_PAGE_APPLE_VENDOR) {
1364                 dbd = true;
1365             }
1366         } else {
1367             /*
1368              * MMC prescribes that CD/DVD drives have no block descriptors,
1369              * and defines no device-specific parameter.
1370              */
1371             dev_specific_param = 0x00;
1372             dbd = true;
1373         }
1374     }
1375 
1376     if (r->req.cmd.buf[0] == MODE_SENSE) {
1377         p[1] = 0; /* Default media type.  */
1378         p[2] = dev_specific_param;
1379         p[3] = 0; /* Block descriptor length.  */
1380         p += 4;
1381     } else { /* MODE_SENSE_10 */
1382         p[2] = 0; /* Default media type.  */
1383         p[3] = dev_specific_param;
1384         p[6] = p[7] = 0; /* Block descriptor length.  */
1385         p += 8;
1386     }
1387 
1388     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1389     if (!dbd && nb_sectors) {
1390         if (r->req.cmd.buf[0] == MODE_SENSE) {
1391             outbuf[3] = 8; /* Block descriptor length  */
1392         } else { /* MODE_SENSE_10 */
1393             outbuf[7] = 8; /* Block descriptor length  */
1394         }
1395         nb_sectors /= (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1396         if (nb_sectors > 0xffffff) {
1397             nb_sectors = 0;
1398         }
1399         p[0] = 0; /* media density code */
1400         p[1] = (nb_sectors >> 16) & 0xff;
1401         p[2] = (nb_sectors >> 8) & 0xff;
1402         p[3] = nb_sectors & 0xff;
1403         p[4] = 0; /* reserved */
1404         p[5] = 0; /* bytes 5-7 are the sector size in bytes */
1405         p[6] = s->qdev.blocksize >> 8;
1406         p[7] = 0;
1407         p += 8;
1408     }
1409 
1410     if (page_control == 3) {
1411         /* Saved Values */
1412         scsi_check_condition(r, SENSE_CODE(SAVING_PARAMS_NOT_SUPPORTED));
1413         return -1;
1414     }
1415 
1416     if (page == 0x3f) {
1417         for (page = 0; page <= 0x3e; page++) {
1418             mode_sense_page(s, page, &p, page_control);
1419         }
1420     } else {
1421         ret = mode_sense_page(s, page, &p, page_control);
1422         if (ret == -1) {
1423             return -1;
1424         }
1425     }
1426 
1427     buflen = p - outbuf;
1428     /*
1429      * The mode data length field specifies the length in bytes of the
1430      * following data that is available to be transferred. The mode data
1431      * length does not include itself.
1432      */
1433     if (r->req.cmd.buf[0] == MODE_SENSE) {
1434         outbuf[0] = buflen - 1;
1435     } else { /* MODE_SENSE_10 */
1436         outbuf[0] = ((buflen - 2) >> 8) & 0xff;
1437         outbuf[1] = (buflen - 2) & 0xff;
1438     }
1439     return buflen;
1440 }
1441 
scsi_disk_emulate_read_toc(SCSIRequest * req,uint8_t * outbuf)1442 static int scsi_disk_emulate_read_toc(SCSIRequest *req, uint8_t *outbuf)
1443 {
1444     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1445     int start_track, format, msf, toclen;
1446     uint64_t nb_sectors;
1447 
1448     msf = req->cmd.buf[1] & 2;
1449     format = req->cmd.buf[2] & 0xf;
1450     start_track = req->cmd.buf[6];
1451     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
1452     trace_scsi_disk_emulate_read_toc(start_track, format, msf >> 1);
1453     nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
1454     switch (format) {
1455     case 0:
1456         toclen = cdrom_read_toc(nb_sectors, outbuf, msf, start_track);
1457         break;
1458     case 1:
1459         /* multi session : only a single session defined */
1460         toclen = 12;
1461         memset(outbuf, 0, 12);
1462         outbuf[1] = 0x0a;
1463         outbuf[2] = 0x01;
1464         outbuf[3] = 0x01;
1465         break;
1466     case 2:
1467         toclen = cdrom_read_toc_raw(nb_sectors, outbuf, msf, start_track);
1468         break;
1469     default:
1470         return -1;
1471     }
1472     return toclen;
1473 }
1474 
scsi_disk_emulate_start_stop(SCSIDiskReq * r)1475 static int scsi_disk_emulate_start_stop(SCSIDiskReq *r)
1476 {
1477     SCSIRequest *req = &r->req;
1478     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1479     bool start = req->cmd.buf[4] & 1;
1480     bool loej = req->cmd.buf[4] & 2; /* load on start, eject on !start */
1481     int pwrcnd = req->cmd.buf[4] & 0xf0;
1482 
1483     if (pwrcnd) {
1484         /* eject/load only happens for power condition == 0 */
1485         return 0;
1486     }
1487 
1488     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) && loej) {
1489         if (!start && !s->tray_open && s->tray_locked) {
1490             scsi_check_condition(r,
1491                                  blk_is_inserted(s->qdev.conf.blk)
1492                                  ? SENSE_CODE(ILLEGAL_REQ_REMOVAL_PREVENTED)
1493                                  : SENSE_CODE(NOT_READY_REMOVAL_PREVENTED));
1494             return -1;
1495         }
1496 
1497         if (s->tray_open != !start) {
1498             blk_eject(s->qdev.conf.blk, !start);
1499             s->tray_open = !start;
1500         }
1501     }
1502     return 0;
1503 }
1504 
scsi_disk_emulate_read_data(SCSIRequest * req)1505 static void scsi_disk_emulate_read_data(SCSIRequest *req)
1506 {
1507     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1508     int buflen = r->iov.iov_len;
1509 
1510     if (buflen) {
1511         trace_scsi_disk_emulate_read_data(buflen);
1512         r->iov.iov_len = 0;
1513         r->started = true;
1514         scsi_req_data(&r->req, buflen);
1515         return;
1516     }
1517 
1518     /* This also clears the sense buffer for REQUEST SENSE.  */
1519     scsi_req_complete(&r->req, GOOD);
1520 }
1521 
scsi_disk_check_mode_select(SCSIDiskState * s,int page,uint8_t * inbuf,int inlen)1522 static int scsi_disk_check_mode_select(SCSIDiskState *s, int page,
1523                                        uint8_t *inbuf, int inlen)
1524 {
1525     uint8_t mode_current[SCSI_MAX_MODE_LEN];
1526     uint8_t mode_changeable[SCSI_MAX_MODE_LEN];
1527     uint8_t *p;
1528     int len, expected_len, changeable_len, i;
1529 
1530     /* The input buffer does not include the page header, so it is
1531      * off by 2 bytes.
1532      */
1533     expected_len = inlen + 2;
1534     if (expected_len > SCSI_MAX_MODE_LEN) {
1535         return -1;
1536     }
1537 
1538     /* MODE_PAGE_ALLS is only valid for MODE SENSE commands */
1539     if (page == MODE_PAGE_ALLS) {
1540         return -1;
1541     }
1542 
1543     p = mode_current;
1544     memset(mode_current, 0, inlen + 2);
1545     len = mode_sense_page(s, page, &p, 0);
1546     if (len < 0 || len != expected_len) {
1547         return -1;
1548     }
1549 
1550     p = mode_changeable;
1551     memset(mode_changeable, 0, inlen + 2);
1552     changeable_len = mode_sense_page(s, page, &p, 1);
1553     assert(changeable_len == len);
1554 
1555     /* Check that unchangeable bits are the same as what MODE SENSE
1556      * would return.
1557      */
1558     for (i = 2; i < len; i++) {
1559         if (((mode_current[i] ^ inbuf[i - 2]) & ~mode_changeable[i]) != 0) {
1560             return -1;
1561         }
1562     }
1563     return 0;
1564 }
1565 
scsi_disk_apply_mode_select(SCSIDiskState * s,int page,uint8_t * p)1566 static void scsi_disk_apply_mode_select(SCSIDiskState *s, int page, uint8_t *p)
1567 {
1568     switch (page) {
1569     case MODE_PAGE_CACHING:
1570         blk_set_enable_write_cache(s->qdev.conf.blk, (p[0] & 4) != 0);
1571         break;
1572 
1573     default:
1574         break;
1575     }
1576 }
1577 
mode_select_pages(SCSIDiskReq * r,uint8_t * p,int len,bool change)1578 static int mode_select_pages(SCSIDiskReq *r, uint8_t *p, int len, bool change)
1579 {
1580     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1581 
1582     while (len > 0) {
1583         int page, subpage, page_len;
1584 
1585         /* Parse both possible formats for the mode page headers.  */
1586         page = p[0] & 0x3f;
1587         if (p[0] & 0x40) {
1588             if (len < 4) {
1589                 goto invalid_param_len;
1590             }
1591             subpage = p[1];
1592             page_len = lduw_be_p(&p[2]);
1593             p += 4;
1594             len -= 4;
1595         } else {
1596             if (len < 2) {
1597                 goto invalid_param_len;
1598             }
1599             subpage = 0;
1600             page_len = p[1];
1601             p += 2;
1602             len -= 2;
1603         }
1604 
1605         if (subpage) {
1606             goto invalid_param;
1607         }
1608         if (page_len > len) {
1609             if (!(s->quirks & SCSI_DISK_QUIRK_MODE_PAGE_TRUNCATED)) {
1610                 goto invalid_param_len;
1611             }
1612             trace_scsi_disk_mode_select_page_truncated(page, page_len, len);
1613         }
1614 
1615         if (!change) {
1616             if (scsi_disk_check_mode_select(s, page, p, page_len) < 0) {
1617                 goto invalid_param;
1618             }
1619         } else {
1620             scsi_disk_apply_mode_select(s, page, p);
1621         }
1622 
1623         p += page_len;
1624         len -= page_len;
1625     }
1626     return 0;
1627 
1628 invalid_param:
1629     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1630     return -1;
1631 
1632 invalid_param_len:
1633     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1634     return -1;
1635 }
1636 
scsi_disk_emulate_mode_select(SCSIDiskReq * r,uint8_t * inbuf)1637 static void scsi_disk_emulate_mode_select(SCSIDiskReq *r, uint8_t *inbuf)
1638 {
1639     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1640     uint8_t *p = inbuf;
1641     int cmd = r->req.cmd.buf[0];
1642     int len = r->req.cmd.xfer;
1643     int hdr_len = (cmd == MODE_SELECT ? 4 : 8);
1644     int bd_len, bs;
1645     int pass;
1646 
1647     if ((r->req.cmd.buf[1] & 0x11) != 0x10) {
1648         if (!(s->quirks &
1649             (1 << SCSI_DISK_QUIRK_MODE_PAGE_VENDOR_SPECIFIC_APPLE))) {
1650             /* We only support PF=1, SP=0.  */
1651             goto invalid_field;
1652         }
1653     }
1654 
1655     if (len < hdr_len) {
1656         goto invalid_param_len;
1657     }
1658 
1659     bd_len = (cmd == MODE_SELECT ? p[3] : lduw_be_p(&p[6]));
1660     len -= hdr_len;
1661     p += hdr_len;
1662     if (len < bd_len) {
1663         goto invalid_param_len;
1664     }
1665     if (bd_len != 0 && bd_len != 8) {
1666         goto invalid_param;
1667     }
1668 
1669     /* Allow changing the block size */
1670     if (bd_len) {
1671         bs = p[5] << 16 | p[6] << 8 | p[7];
1672 
1673         /*
1674          * Since the existing code only checks/updates bits 8-15 of the block
1675          * size, restrict ourselves to the same requirement for now to ensure
1676          * that a block size set by a block descriptor and then read back by
1677          * a subsequent SCSI command will be the same. Also disallow a block
1678          * size of 256 since we cannot handle anything below BDRV_SECTOR_SIZE.
1679          */
1680         if (bs && !(bs & ~0xfe00) && bs != s->qdev.blocksize) {
1681             s->qdev.blocksize = bs;
1682             trace_scsi_disk_mode_select_set_blocksize(s->qdev.blocksize);
1683         }
1684     }
1685 
1686     len -= bd_len;
1687     p += bd_len;
1688 
1689     /* Ensure no change is made if there is an error!  */
1690     for (pass = 0; pass < 2; pass++) {
1691         if (mode_select_pages(r, p, len, pass == 1) < 0) {
1692             assert(pass == 0);
1693             return;
1694         }
1695     }
1696     if (!blk_enable_write_cache(s->qdev.conf.blk)) {
1697         /* The request is used as the AIO opaque value, so add a ref.  */
1698         scsi_req_ref(&r->req);
1699         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
1700                          BLOCK_ACCT_FLUSH);
1701         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
1702         return;
1703     }
1704 
1705     scsi_req_complete(&r->req, GOOD);
1706     return;
1707 
1708 invalid_param:
1709     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM));
1710     return;
1711 
1712 invalid_param_len:
1713     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1714     return;
1715 
1716 invalid_field:
1717     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1718 }
1719 
1720 /* sector_num and nb_sectors expected to be in qdev blocksize */
check_lba_range(SCSIDiskState * s,uint64_t sector_num,uint32_t nb_sectors)1721 static inline bool check_lba_range(SCSIDiskState *s,
1722                                    uint64_t sector_num, uint32_t nb_sectors)
1723 {
1724     /*
1725      * The first line tests that no overflow happens when computing the last
1726      * sector.  The second line tests that the last accessed sector is in
1727      * range.
1728      *
1729      * Careful, the computations should not underflow for nb_sectors == 0,
1730      * and a 0-block read to the first LBA beyond the end of device is
1731      * valid.
1732      */
1733     return (sector_num <= sector_num + nb_sectors &&
1734             sector_num + nb_sectors <= s->qdev.max_lba + 1);
1735 }
1736 
1737 typedef struct UnmapCBData {
1738     SCSIDiskReq *r;
1739     uint8_t *inbuf;
1740     int count;
1741 } UnmapCBData;
1742 
1743 static void scsi_unmap_complete(void *opaque, int ret);
1744 
scsi_unmap_complete_noio(UnmapCBData * data,int ret)1745 static void scsi_unmap_complete_noio(UnmapCBData *data, int ret)
1746 {
1747     SCSIDiskReq *r = data->r;
1748     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1749 
1750     assert(r->req.aiocb == NULL);
1751 
1752     if (data->count > 0) {
1753         uint64_t sector_num = ldq_be_p(&data->inbuf[0]);
1754         uint32_t nb_sectors = ldl_be_p(&data->inbuf[8]) & 0xffffffffULL;
1755         r->sector = sector_num * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1756         r->sector_count = nb_sectors * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1757 
1758         if (!check_lba_range(s, sector_num, nb_sectors)) {
1759             block_acct_invalid(blk_get_stats(s->qdev.conf.blk),
1760                                BLOCK_ACCT_UNMAP);
1761             scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1762             goto done;
1763         }
1764 
1765         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1766                          r->sector_count * BDRV_SECTOR_SIZE,
1767                          BLOCK_ACCT_UNMAP);
1768 
1769         r->req.aiocb = blk_aio_pdiscard(s->qdev.conf.blk,
1770                                         r->sector * BDRV_SECTOR_SIZE,
1771                                         r->sector_count * BDRV_SECTOR_SIZE,
1772                                         scsi_unmap_complete, data);
1773         data->count--;
1774         data->inbuf += 16;
1775         return;
1776     }
1777 
1778     scsi_req_complete(&r->req, GOOD);
1779 
1780 done:
1781     scsi_req_unref(&r->req);
1782     g_free(data);
1783 }
1784 
scsi_unmap_complete(void * opaque,int ret)1785 static void scsi_unmap_complete(void *opaque, int ret)
1786 {
1787     UnmapCBData *data = opaque;
1788     SCSIDiskReq *r = data->r;
1789     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1790 
1791     assert(r->req.aiocb != NULL);
1792     r->req.aiocb = NULL;
1793 
1794     if (scsi_disk_req_check_error(r, ret, true)) {
1795         scsi_req_unref(&r->req);
1796         g_free(data);
1797     } else {
1798         block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1799         scsi_unmap_complete_noio(data, ret);
1800     }
1801 }
1802 
scsi_disk_emulate_unmap(SCSIDiskReq * r,uint8_t * inbuf)1803 static void scsi_disk_emulate_unmap(SCSIDiskReq *r, uint8_t *inbuf)
1804 {
1805     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1806     uint8_t *p = inbuf;
1807     int len = r->req.cmd.xfer;
1808     UnmapCBData *data;
1809 
1810     /* Reject ANCHOR=1.  */
1811     if (r->req.cmd.buf[1] & 0x1) {
1812         goto invalid_field;
1813     }
1814 
1815     if (len < 8) {
1816         goto invalid_param_len;
1817     }
1818     if (len < lduw_be_p(&p[0]) + 2) {
1819         goto invalid_param_len;
1820     }
1821     if (len < lduw_be_p(&p[2]) + 8) {
1822         goto invalid_param_len;
1823     }
1824     if (lduw_be_p(&p[2]) & 15) {
1825         goto invalid_param_len;
1826     }
1827 
1828     if (!blk_is_writable(s->qdev.conf.blk)) {
1829         block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1830         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1831         return;
1832     }
1833 
1834     data = g_new0(UnmapCBData, 1);
1835     data->r = r;
1836     data->inbuf = &p[8];
1837     data->count = lduw_be_p(&p[2]) >> 4;
1838 
1839     /* The matching unref is in scsi_unmap_complete, before data is freed.  */
1840     scsi_req_ref(&r->req);
1841     scsi_unmap_complete_noio(data, 0);
1842     return;
1843 
1844 invalid_param_len:
1845     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1846     scsi_check_condition(r, SENSE_CODE(INVALID_PARAM_LEN));
1847     return;
1848 
1849 invalid_field:
1850     block_acct_invalid(blk_get_stats(s->qdev.conf.blk), BLOCK_ACCT_UNMAP);
1851     scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1852 }
1853 
1854 typedef struct WriteSameCBData {
1855     SCSIDiskReq *r;
1856     int64_t sector;
1857     int nb_sectors;
1858     QEMUIOVector qiov;
1859     struct iovec iov;
1860 } WriteSameCBData;
1861 
scsi_write_same_complete(void * opaque,int ret)1862 static void scsi_write_same_complete(void *opaque, int ret)
1863 {
1864     WriteSameCBData *data = opaque;
1865     SCSIDiskReq *r = data->r;
1866     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
1867 
1868     assert(r->req.aiocb != NULL);
1869     r->req.aiocb = NULL;
1870 
1871     if (scsi_disk_req_check_error(r, ret, true)) {
1872         goto done;
1873     }
1874 
1875     block_acct_done(blk_get_stats(s->qdev.conf.blk), &r->acct);
1876 
1877     data->nb_sectors -= data->iov.iov_len / BDRV_SECTOR_SIZE;
1878     data->sector += data->iov.iov_len / BDRV_SECTOR_SIZE;
1879     data->iov.iov_len = MIN(data->nb_sectors * BDRV_SECTOR_SIZE,
1880                             data->iov.iov_len);
1881     if (data->iov.iov_len) {
1882         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1883                          data->iov.iov_len, BLOCK_ACCT_WRITE);
1884         /* Reinitialize qiov, to handle unaligned WRITE SAME request
1885          * where final qiov may need smaller size */
1886         qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1887         r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1888                                        data->sector << BDRV_SECTOR_BITS,
1889                                        &data->qiov, 0,
1890                                        scsi_write_same_complete, data);
1891         return;
1892     }
1893 
1894     scsi_req_complete(&r->req, GOOD);
1895 
1896 done:
1897     scsi_req_unref(&r->req);
1898     qemu_vfree(data->iov.iov_base);
1899     g_free(data);
1900 }
1901 
scsi_disk_emulate_write_same(SCSIDiskReq * r,uint8_t * inbuf)1902 static void scsi_disk_emulate_write_same(SCSIDiskReq *r, uint8_t *inbuf)
1903 {
1904     SCSIRequest *req = &r->req;
1905     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
1906     uint32_t nb_sectors = scsi_data_cdb_xfer(r->req.cmd.buf);
1907     WriteSameCBData *data;
1908     uint8_t *buf;
1909     int i, l;
1910 
1911     /* Fail if PBDATA=1 or LBDATA=1 or ANCHOR=1.  */
1912     if (nb_sectors == 0 || (req->cmd.buf[1] & 0x16)) {
1913         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1914         return;
1915     }
1916 
1917     if (!blk_is_writable(s->qdev.conf.blk)) {
1918         scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
1919         return;
1920     }
1921     if (!check_lba_range(s, r->req.cmd.lba, nb_sectors)) {
1922         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
1923         return;
1924     }
1925 
1926     if ((req->cmd.buf[1] & 0x1) || buffer_is_zero(inbuf, s->qdev.blocksize)) {
1927         int flags = (req->cmd.buf[1] & 0x8) ? BDRV_REQ_MAY_UNMAP : 0;
1928 
1929         /* The request is used as the AIO opaque value, so add a ref.  */
1930         scsi_req_ref(&r->req);
1931         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1932                          nb_sectors * s->qdev.blocksize,
1933                         BLOCK_ACCT_WRITE);
1934         r->req.aiocb = blk_aio_pwrite_zeroes(s->qdev.conf.blk,
1935                                 r->req.cmd.lba * s->qdev.blocksize,
1936                                 nb_sectors * s->qdev.blocksize,
1937                                 flags, scsi_aio_complete, r);
1938         return;
1939     }
1940 
1941     data = g_new0(WriteSameCBData, 1);
1942     data->r = r;
1943     data->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1944     data->nb_sectors = nb_sectors * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
1945     data->iov.iov_len = MIN(data->nb_sectors * BDRV_SECTOR_SIZE,
1946                             SCSI_WRITE_SAME_MAX);
1947     data->iov.iov_base = buf = blk_blockalign(s->qdev.conf.blk,
1948                                               data->iov.iov_len);
1949     qemu_iovec_init_external(&data->qiov, &data->iov, 1);
1950 
1951     for (i = 0; i < data->iov.iov_len; i += l) {
1952         l = MIN(s->qdev.blocksize, data->iov.iov_len - i);
1953         memcpy(&buf[i], inbuf, l);
1954     }
1955 
1956     scsi_req_ref(&r->req);
1957     block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct,
1958                      data->iov.iov_len, BLOCK_ACCT_WRITE);
1959     r->req.aiocb = blk_aio_pwritev(s->qdev.conf.blk,
1960                                    data->sector << BDRV_SECTOR_BITS,
1961                                    &data->qiov, 0,
1962                                    scsi_write_same_complete, data);
1963 }
1964 
scsi_disk_emulate_write_data(SCSIRequest * req)1965 static void scsi_disk_emulate_write_data(SCSIRequest *req)
1966 {
1967     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
1968 
1969     if (r->iov.iov_len) {
1970         int buflen = r->iov.iov_len;
1971         trace_scsi_disk_emulate_write_data(buflen);
1972         r->iov.iov_len = 0;
1973         scsi_req_data(&r->req, buflen);
1974         return;
1975     }
1976 
1977     switch (req->cmd.buf[0]) {
1978     case MODE_SELECT:
1979     case MODE_SELECT_10:
1980         /* This also clears the sense buffer for REQUEST SENSE.  */
1981         scsi_disk_emulate_mode_select(r, r->iov.iov_base);
1982         break;
1983 
1984     case UNMAP:
1985         scsi_disk_emulate_unmap(r, r->iov.iov_base);
1986         break;
1987 
1988     case VERIFY_10:
1989     case VERIFY_12:
1990     case VERIFY_16:
1991         if (r->req.status == -1) {
1992             scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
1993         }
1994         break;
1995 
1996     case WRITE_SAME_10:
1997     case WRITE_SAME_16:
1998         scsi_disk_emulate_write_same(r, r->iov.iov_base);
1999         break;
2000 
2001     case FORMAT_UNIT:
2002         scsi_req_complete(&r->req, GOOD);
2003         break;
2004 
2005     default:
2006         abort();
2007     }
2008 }
2009 
scsi_disk_emulate_command(SCSIRequest * req,uint8_t * buf)2010 static int32_t scsi_disk_emulate_command(SCSIRequest *req, uint8_t *buf)
2011 {
2012     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
2013     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2014     uint64_t nb_sectors;
2015     uint8_t *outbuf;
2016     int buflen;
2017 
2018     switch (req->cmd.buf[0]) {
2019     case INQUIRY:
2020     case MODE_SENSE:
2021     case MODE_SENSE_10:
2022     case RESERVE:
2023     case RESERVE_10:
2024     case RELEASE:
2025     case RELEASE_10:
2026     case START_STOP:
2027     case ALLOW_MEDIUM_REMOVAL:
2028     case GET_CONFIGURATION:
2029     case GET_EVENT_STATUS_NOTIFICATION:
2030     case MECHANISM_STATUS:
2031     case REQUEST_SENSE:
2032         break;
2033 
2034     default:
2035         if (!blk_is_available(s->qdev.conf.blk)) {
2036             scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
2037             return 0;
2038         }
2039         break;
2040     }
2041 
2042     /*
2043      * FIXME: we shouldn't return anything bigger than 4k, but the code
2044      * requires the buffer to be as big as req->cmd.xfer in several
2045      * places.  So, do not allow CDBs with a very large ALLOCATION
2046      * LENGTH.  The real fix would be to modify scsi_read_data and
2047      * dma_buf_read, so that they return data beyond the buflen
2048      * as all zeros.
2049      */
2050     if (req->cmd.xfer > 65536) {
2051         goto illegal_request;
2052     }
2053     r->buflen = MAX(4096, req->cmd.xfer);
2054 
2055     if (!r->iov.iov_base) {
2056         r->iov.iov_base = blk_blockalign(s->qdev.conf.blk, r->buflen);
2057     }
2058 
2059     outbuf = r->iov.iov_base;
2060     memset(outbuf, 0, r->buflen);
2061     switch (req->cmd.buf[0]) {
2062     case TEST_UNIT_READY:
2063         assert(blk_is_available(s->qdev.conf.blk));
2064         break;
2065     case INQUIRY:
2066         buflen = scsi_disk_emulate_inquiry(req, outbuf);
2067         if (buflen < 0) {
2068             goto illegal_request;
2069         }
2070         break;
2071     case MODE_SENSE:
2072     case MODE_SENSE_10:
2073         buflen = scsi_disk_emulate_mode_sense(r, outbuf);
2074         if (buflen < 0) {
2075             goto illegal_request;
2076         }
2077         break;
2078     case READ_TOC:
2079         buflen = scsi_disk_emulate_read_toc(req, outbuf);
2080         if (buflen < 0) {
2081             goto illegal_request;
2082         }
2083         break;
2084     case RESERVE:
2085         if (req->cmd.buf[1] & 1) {
2086             goto illegal_request;
2087         }
2088         break;
2089     case RESERVE_10:
2090         if (req->cmd.buf[1] & 3) {
2091             goto illegal_request;
2092         }
2093         break;
2094     case RELEASE:
2095         if (req->cmd.buf[1] & 1) {
2096             goto illegal_request;
2097         }
2098         break;
2099     case RELEASE_10:
2100         if (req->cmd.buf[1] & 3) {
2101             goto illegal_request;
2102         }
2103         break;
2104     case START_STOP:
2105         if (scsi_disk_emulate_start_stop(r) < 0) {
2106             return 0;
2107         }
2108         break;
2109     case ALLOW_MEDIUM_REMOVAL:
2110         s->tray_locked = req->cmd.buf[4] & 1;
2111         blk_lock_medium(s->qdev.conf.blk, req->cmd.buf[4] & 1);
2112         break;
2113     case READ_CAPACITY_10:
2114         /* The normal LEN field for this command is zero.  */
2115         memset(outbuf, 0, 8);
2116         blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2117         if (!nb_sectors) {
2118             scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
2119             return 0;
2120         }
2121         if ((req->cmd.buf[8] & 1) == 0 && req->cmd.lba) {
2122             goto illegal_request;
2123         }
2124         nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2125         /* Returned value is the address of the last sector.  */
2126         nb_sectors--;
2127         /* Remember the new size for read/write sanity checking. */
2128         s->qdev.max_lba = nb_sectors;
2129         /* Clip to 2TB, instead of returning capacity modulo 2TB. */
2130         if (nb_sectors > UINT32_MAX) {
2131             nb_sectors = UINT32_MAX;
2132         }
2133         outbuf[0] = (nb_sectors >> 24) & 0xff;
2134         outbuf[1] = (nb_sectors >> 16) & 0xff;
2135         outbuf[2] = (nb_sectors >> 8) & 0xff;
2136         outbuf[3] = nb_sectors & 0xff;
2137         outbuf[4] = 0;
2138         outbuf[5] = 0;
2139         outbuf[6] = s->qdev.blocksize >> 8;
2140         outbuf[7] = 0;
2141         break;
2142     case REQUEST_SENSE:
2143         /* Just return "NO SENSE".  */
2144         buflen = scsi_convert_sense(NULL, 0, outbuf, r->buflen,
2145                                     (req->cmd.buf[1] & 1) == 0);
2146         if (buflen < 0) {
2147             goto illegal_request;
2148         }
2149         break;
2150     case MECHANISM_STATUS:
2151         buflen = scsi_emulate_mechanism_status(s, outbuf);
2152         if (buflen < 0) {
2153             goto illegal_request;
2154         }
2155         break;
2156     case GET_CONFIGURATION:
2157         buflen = scsi_get_configuration(s, outbuf);
2158         if (buflen < 0) {
2159             goto illegal_request;
2160         }
2161         break;
2162     case GET_EVENT_STATUS_NOTIFICATION:
2163         buflen = scsi_get_event_status_notification(s, r, outbuf);
2164         if (buflen < 0) {
2165             goto illegal_request;
2166         }
2167         break;
2168     case READ_DISC_INFORMATION:
2169         buflen = scsi_read_disc_information(s, r, outbuf);
2170         if (buflen < 0) {
2171             goto illegal_request;
2172         }
2173         break;
2174     case READ_DVD_STRUCTURE:
2175         buflen = scsi_read_dvd_structure(s, r, outbuf);
2176         if (buflen < 0) {
2177             goto illegal_request;
2178         }
2179         break;
2180     case SERVICE_ACTION_IN_16:
2181         /* Service Action In subcommands. */
2182         if ((req->cmd.buf[1] & 31) == SAI_READ_CAPACITY_16) {
2183             trace_scsi_disk_emulate_command_SAI_16();
2184             memset(outbuf, 0, req->cmd.xfer);
2185             blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2186             if (!nb_sectors) {
2187                 scsi_check_condition(r, SENSE_CODE(LUN_NOT_READY));
2188                 return 0;
2189             }
2190             if ((req->cmd.buf[14] & 1) == 0 && req->cmd.lba) {
2191                 goto illegal_request;
2192             }
2193             nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2194             /* Returned value is the address of the last sector.  */
2195             nb_sectors--;
2196             /* Remember the new size for read/write sanity checking. */
2197             s->qdev.max_lba = nb_sectors;
2198             outbuf[0] = (nb_sectors >> 56) & 0xff;
2199             outbuf[1] = (nb_sectors >> 48) & 0xff;
2200             outbuf[2] = (nb_sectors >> 40) & 0xff;
2201             outbuf[3] = (nb_sectors >> 32) & 0xff;
2202             outbuf[4] = (nb_sectors >> 24) & 0xff;
2203             outbuf[5] = (nb_sectors >> 16) & 0xff;
2204             outbuf[6] = (nb_sectors >> 8) & 0xff;
2205             outbuf[7] = nb_sectors & 0xff;
2206             outbuf[8] = 0;
2207             outbuf[9] = 0;
2208             outbuf[10] = s->qdev.blocksize >> 8;
2209             outbuf[11] = 0;
2210             outbuf[12] = 0;
2211             outbuf[13] = get_physical_block_exp(&s->qdev.conf);
2212 
2213             /* set TPE bit if the format supports discard */
2214             if (s->qdev.conf.discard_granularity) {
2215                 outbuf[14] = 0x80;
2216             }
2217 
2218             /* Protection, exponent and lowest lba field left blank. */
2219             break;
2220         }
2221         trace_scsi_disk_emulate_command_SAI_unsupported();
2222         goto illegal_request;
2223     case SYNCHRONIZE_CACHE:
2224         /* The request is used as the AIO opaque value, so add a ref.  */
2225         scsi_req_ref(&r->req);
2226         block_acct_start(blk_get_stats(s->qdev.conf.blk), &r->acct, 0,
2227                          BLOCK_ACCT_FLUSH);
2228         r->req.aiocb = blk_aio_flush(s->qdev.conf.blk, scsi_aio_complete, r);
2229         return 0;
2230     case SEEK_10:
2231         trace_scsi_disk_emulate_command_SEEK_10(r->req.cmd.lba);
2232         if (r->req.cmd.lba > s->qdev.max_lba) {
2233             goto illegal_lba;
2234         }
2235         break;
2236     case MODE_SELECT:
2237         trace_scsi_disk_emulate_command_MODE_SELECT(r->req.cmd.xfer);
2238         break;
2239     case MODE_SELECT_10:
2240         trace_scsi_disk_emulate_command_MODE_SELECT_10(r->req.cmd.xfer);
2241         break;
2242     case UNMAP:
2243         trace_scsi_disk_emulate_command_UNMAP(r->req.cmd.xfer);
2244         break;
2245     case VERIFY_10:
2246     case VERIFY_12:
2247     case VERIFY_16:
2248         trace_scsi_disk_emulate_command_VERIFY((req->cmd.buf[1] >> 1) & 3);
2249         if (req->cmd.buf[1] & 6) {
2250             goto illegal_request;
2251         }
2252         break;
2253     case WRITE_SAME_10:
2254     case WRITE_SAME_16:
2255         trace_scsi_disk_emulate_command_WRITE_SAME(
2256                 req->cmd.buf[0] == WRITE_SAME_10 ? 10 : 16, r->req.cmd.xfer);
2257         break;
2258     case FORMAT_UNIT:
2259         trace_scsi_disk_emulate_command_FORMAT_UNIT(r->req.cmd.xfer);
2260         break;
2261     default:
2262         trace_scsi_disk_emulate_command_UNKNOWN(buf[0],
2263                                                 scsi_command_name(buf[0]));
2264         scsi_check_condition(r, SENSE_CODE(INVALID_OPCODE));
2265         return 0;
2266     }
2267     assert(!r->req.aiocb);
2268     r->iov.iov_len = MIN(r->buflen, req->cmd.xfer);
2269     if (r->iov.iov_len == 0) {
2270         scsi_req_complete(&r->req, GOOD);
2271     }
2272     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2273         assert(r->iov.iov_len == req->cmd.xfer);
2274         return -r->iov.iov_len;
2275     } else {
2276         return r->iov.iov_len;
2277     }
2278 
2279 illegal_request:
2280     if (r->req.status == -1) {
2281         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2282     }
2283     return 0;
2284 
2285 illegal_lba:
2286     scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2287     return 0;
2288 }
2289 
2290 /* Execute a scsi command.  Returns the length of the data expected by the
2291    command.  This will be Positive for data transfers from the device
2292    (eg. disk reads), negative for transfers to the device (eg. disk writes),
2293    and zero if the command does not transfer any data.  */
2294 
scsi_disk_dma_command(SCSIRequest * req,uint8_t * buf)2295 static int32_t scsi_disk_dma_command(SCSIRequest *req, uint8_t *buf)
2296 {
2297     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
2298     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
2299     SCSIDiskClass *sdc = (SCSIDiskClass *) object_get_class(OBJECT(s));
2300     uint32_t len;
2301     uint8_t command;
2302 
2303     command = buf[0];
2304 
2305     if (!blk_is_available(s->qdev.conf.blk)) {
2306         scsi_check_condition(r, SENSE_CODE(NO_MEDIUM));
2307         return 0;
2308     }
2309 
2310     len = scsi_data_cdb_xfer(r->req.cmd.buf);
2311     switch (command) {
2312     case READ_6:
2313     case READ_10:
2314     case READ_12:
2315     case READ_16:
2316         trace_scsi_disk_dma_command_READ(r->req.cmd.lba, len);
2317         /* Protection information is not supported.  For SCSI versions 2 and
2318          * older (as determined by snooping the guest's INQUIRY commands),
2319          * there is no RD/WR/VRPROTECT, so skip this check in these versions.
2320          */
2321         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2322             goto illegal_request;
2323         }
2324         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2325             goto illegal_lba;
2326         }
2327         r->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2328         r->sector_count = len * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2329         break;
2330     case WRITE_6:
2331     case WRITE_10:
2332     case WRITE_12:
2333     case WRITE_16:
2334     case WRITE_VERIFY_10:
2335     case WRITE_VERIFY_12:
2336     case WRITE_VERIFY_16:
2337         if (!blk_is_writable(s->qdev.conf.blk)) {
2338             scsi_check_condition(r, SENSE_CODE(WRITE_PROTECTED));
2339             return 0;
2340         }
2341         trace_scsi_disk_dma_command_WRITE(
2342                 (command & 0xe) == 0xe ? "And Verify " : "",
2343                 r->req.cmd.lba, len);
2344         /* fall through */
2345     case VERIFY_10:
2346     case VERIFY_12:
2347     case VERIFY_16:
2348         /* We get here only for BYTCHK == 0x01 and only for scsi-block.
2349          * As far as DMA is concerned, we can treat it the same as a write;
2350          * scsi_block_do_sgio will send VERIFY commands.
2351          */
2352         if (s->qdev.scsi_version > 2 && (r->req.cmd.buf[1] & 0xe0)) {
2353             goto illegal_request;
2354         }
2355         if (!check_lba_range(s, r->req.cmd.lba, len)) {
2356             goto illegal_lba;
2357         }
2358         r->sector = r->req.cmd.lba * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2359         r->sector_count = len * (s->qdev.blocksize / BDRV_SECTOR_SIZE);
2360         break;
2361     default:
2362         abort();
2363     illegal_request:
2364         scsi_check_condition(r, SENSE_CODE(INVALID_FIELD));
2365         return 0;
2366     illegal_lba:
2367         scsi_check_condition(r, SENSE_CODE(LBA_OUT_OF_RANGE));
2368         return 0;
2369     }
2370     r->need_fua = sdc->need_fua(&r->req.cmd);
2371     if (r->sector_count == 0) {
2372         scsi_req_complete(&r->req, GOOD);
2373     }
2374     assert(r->iov.iov_len == 0);
2375     if (r->req.cmd.mode == SCSI_XFER_TO_DEV) {
2376         return -r->sector_count * BDRV_SECTOR_SIZE;
2377     } else {
2378         return r->sector_count * BDRV_SECTOR_SIZE;
2379     }
2380 }
2381 
scsi_disk_reset(DeviceState * dev)2382 static void scsi_disk_reset(DeviceState *dev)
2383 {
2384     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev.qdev, dev);
2385     uint64_t nb_sectors;
2386 
2387     scsi_device_purge_requests(&s->qdev, SENSE_CODE(RESET));
2388 
2389     blk_get_geometry(s->qdev.conf.blk, &nb_sectors);
2390 
2391     nb_sectors /= s->qdev.blocksize / BDRV_SECTOR_SIZE;
2392     if (nb_sectors) {
2393         nb_sectors--;
2394     }
2395     s->qdev.max_lba = nb_sectors;
2396     /* reset tray statuses */
2397     s->tray_locked = 0;
2398     s->tray_open = 0;
2399 
2400     s->qdev.scsi_version = s->qdev.default_scsi_version;
2401 }
2402 
scsi_disk_drained_begin(void * opaque)2403 static void scsi_disk_drained_begin(void *opaque)
2404 {
2405     SCSIDiskState *s = opaque;
2406 
2407     scsi_device_drained_begin(&s->qdev);
2408 }
2409 
scsi_disk_drained_end(void * opaque)2410 static void scsi_disk_drained_end(void *opaque)
2411 {
2412     SCSIDiskState *s = opaque;
2413 
2414     scsi_device_drained_end(&s->qdev);
2415 }
2416 
scsi_disk_resize_cb(void * opaque)2417 static void scsi_disk_resize_cb(void *opaque)
2418 {
2419     SCSIDiskState *s = opaque;
2420 
2421     /* SPC lists this sense code as available only for
2422      * direct-access devices.
2423      */
2424     if (s->qdev.type == TYPE_DISK) {
2425         scsi_device_report_change(&s->qdev, SENSE_CODE(CAPACITY_CHANGED));
2426     }
2427 }
2428 
scsi_cd_change_media_cb(void * opaque,bool load,Error ** errp)2429 static void scsi_cd_change_media_cb(void *opaque, bool load, Error **errp)
2430 {
2431     SCSIDiskState *s = opaque;
2432 
2433     /*
2434      * When a CD gets changed, we have to report an ejected state and
2435      * then a loaded state to guests so that they detect tray
2436      * open/close and media change events.  Guests that do not use
2437      * GET_EVENT_STATUS_NOTIFICATION to detect such tray open/close
2438      * states rely on this behavior.
2439      *
2440      * media_changed governs the state machine used for unit attention
2441      * report.  media_event is used by GET EVENT STATUS NOTIFICATION.
2442      */
2443     s->media_changed = load;
2444     s->tray_open = !load;
2445     scsi_device_set_ua(&s->qdev, SENSE_CODE(UNIT_ATTENTION_NO_MEDIUM));
2446     s->media_event = true;
2447     s->eject_request = false;
2448 }
2449 
scsi_cd_eject_request_cb(void * opaque,bool force)2450 static void scsi_cd_eject_request_cb(void *opaque, bool force)
2451 {
2452     SCSIDiskState *s = opaque;
2453 
2454     s->eject_request = true;
2455     if (force) {
2456         s->tray_locked = false;
2457     }
2458 }
2459 
scsi_cd_is_tray_open(void * opaque)2460 static bool scsi_cd_is_tray_open(void *opaque)
2461 {
2462     return ((SCSIDiskState *)opaque)->tray_open;
2463 }
2464 
scsi_cd_is_medium_locked(void * opaque)2465 static bool scsi_cd_is_medium_locked(void *opaque)
2466 {
2467     return ((SCSIDiskState *)opaque)->tray_locked;
2468 }
2469 
2470 static const BlockDevOps scsi_disk_removable_block_ops = {
2471     .change_media_cb  = scsi_cd_change_media_cb,
2472     .drained_begin    = scsi_disk_drained_begin,
2473     .drained_end      = scsi_disk_drained_end,
2474     .eject_request_cb = scsi_cd_eject_request_cb,
2475     .is_medium_locked = scsi_cd_is_medium_locked,
2476     .is_tray_open     = scsi_cd_is_tray_open,
2477     .resize_cb        = scsi_disk_resize_cb,
2478 };
2479 
2480 static const BlockDevOps scsi_disk_block_ops = {
2481     .drained_begin = scsi_disk_drained_begin,
2482     .drained_end   = scsi_disk_drained_end,
2483     .resize_cb     = scsi_disk_resize_cb,
2484 };
2485 
scsi_disk_unit_attention_reported(SCSIDevice * dev)2486 static void scsi_disk_unit_attention_reported(SCSIDevice *dev)
2487 {
2488     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2489     if (s->media_changed) {
2490         s->media_changed = false;
2491         scsi_device_set_ua(&s->qdev, SENSE_CODE(MEDIUM_CHANGED));
2492     }
2493 }
2494 
scsi_realize(SCSIDevice * dev,Error ** errp)2495 static void scsi_realize(SCSIDevice *dev, Error **errp)
2496 {
2497     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2498     bool read_only;
2499 
2500     if (!s->qdev.conf.blk) {
2501         error_setg(errp, "drive property not set");
2502         return;
2503     }
2504 
2505     if (!(s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2506         !blk_is_inserted(s->qdev.conf.blk)) {
2507         error_setg(errp, "Device needs media, but drive is empty");
2508         return;
2509     }
2510 
2511     if (!blkconf_blocksizes(&s->qdev.conf, errp)) {
2512         return;
2513     }
2514 
2515     if (blk_get_aio_context(s->qdev.conf.blk) != qemu_get_aio_context() &&
2516         !s->qdev.hba_supports_iothread)
2517     {
2518         error_setg(errp, "HBA does not support iothreads");
2519         return;
2520     }
2521 
2522     if (dev->type == TYPE_DISK) {
2523         if (!blkconf_geometry(&dev->conf, NULL, 65535, 255, 255, errp)) {
2524             return;
2525         }
2526     }
2527 
2528     read_only = !blk_supports_write_perm(s->qdev.conf.blk);
2529     if (dev->type == TYPE_ROM) {
2530         read_only = true;
2531     }
2532 
2533     if (!blkconf_apply_backend_options(&dev->conf, read_only,
2534                                        dev->type == TYPE_DISK, errp)) {
2535         return;
2536     }
2537 
2538     if (s->qdev.conf.discard_granularity == -1) {
2539         s->qdev.conf.discard_granularity =
2540             MAX(s->qdev.conf.logical_block_size, DEFAULT_DISCARD_GRANULARITY);
2541     }
2542 
2543     if (!s->version) {
2544         s->version = g_strdup(qemu_hw_version());
2545     }
2546     if (!s->vendor) {
2547         s->vendor = g_strdup("QEMU");
2548     }
2549     if (s->serial && strlen(s->serial) > MAX_SERIAL_LEN) {
2550         error_setg(errp, "The serial number can't be longer than %d characters",
2551                    MAX_SERIAL_LEN);
2552         return;
2553     }
2554     if (!s->device_id) {
2555         if (s->serial) {
2556             if (strlen(s->serial) > MAX_SERIAL_LEN_FOR_DEVID) {
2557                 error_setg(errp, "The serial number can't be longer than %d "
2558                            "characters when it is also used as the default for "
2559                            "device_id", MAX_SERIAL_LEN_FOR_DEVID);
2560                 return;
2561             }
2562             s->device_id = g_strdup(s->serial);
2563         } else {
2564             const char *str = blk_name(s->qdev.conf.blk);
2565             if (str && *str) {
2566                 s->device_id = g_strdup(str);
2567             }
2568         }
2569     }
2570 
2571     if (blk_is_sg(s->qdev.conf.blk)) {
2572         error_setg(errp, "unwanted /dev/sg*");
2573         return;
2574     }
2575 
2576     if ((s->features & (1 << SCSI_DISK_F_REMOVABLE)) &&
2577             !(s->features & (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS))) {
2578         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_removable_block_ops, s);
2579     } else {
2580         blk_set_dev_ops(s->qdev.conf.blk, &scsi_disk_block_ops, s);
2581     }
2582 
2583     blk_iostatus_enable(s->qdev.conf.blk);
2584 
2585     add_boot_device_lchs(&dev->qdev, NULL,
2586                          dev->conf.lcyls,
2587                          dev->conf.lheads,
2588                          dev->conf.lsecs);
2589 }
2590 
scsi_unrealize(SCSIDevice * dev)2591 static void scsi_unrealize(SCSIDevice *dev)
2592 {
2593     del_boot_device_lchs(&dev->qdev, NULL);
2594 }
2595 
scsi_hd_realize(SCSIDevice * dev,Error ** errp)2596 static void scsi_hd_realize(SCSIDevice *dev, Error **errp)
2597 {
2598     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2599 
2600     /* can happen for devices without drive. The error message for missing
2601      * backend will be issued in scsi_realize
2602      */
2603     if (s->qdev.conf.blk) {
2604         if (!blkconf_blocksizes(&s->qdev.conf, errp)) {
2605             return;
2606         }
2607     }
2608     s->qdev.blocksize = s->qdev.conf.logical_block_size;
2609     s->qdev.type = TYPE_DISK;
2610     if (!s->product) {
2611         s->product = g_strdup("QEMU HARDDISK");
2612     }
2613     scsi_realize(&s->qdev, errp);
2614 }
2615 
scsi_cd_realize(SCSIDevice * dev,Error ** errp)2616 static void scsi_cd_realize(SCSIDevice *dev, Error **errp)
2617 {
2618     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2619     int ret;
2620     uint32_t blocksize = 2048;
2621 
2622     if (!dev->conf.blk) {
2623         /* Anonymous BlockBackend for an empty drive. As we put it into
2624          * dev->conf, qdev takes care of detaching on unplug. */
2625         dev->conf.blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
2626         ret = blk_attach_dev(dev->conf.blk, &dev->qdev);
2627         assert(ret == 0);
2628     }
2629 
2630     if (dev->conf.physical_block_size != 0) {
2631         blocksize = dev->conf.physical_block_size;
2632     }
2633 
2634     s->qdev.blocksize = blocksize;
2635     s->qdev.type = TYPE_ROM;
2636     s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2637     if (!s->product) {
2638         s->product = g_strdup("QEMU CD-ROM");
2639     }
2640     scsi_realize(&s->qdev, errp);
2641 }
2642 
2643 
2644 static const SCSIReqOps scsi_disk_emulate_reqops = {
2645     .size         = sizeof(SCSIDiskReq),
2646     .free_req     = scsi_free_request,
2647     .send_command = scsi_disk_emulate_command,
2648     .read_data    = scsi_disk_emulate_read_data,
2649     .write_data   = scsi_disk_emulate_write_data,
2650     .get_buf      = scsi_get_buf,
2651     .load_request = scsi_disk_emulate_load_request,
2652     .save_request = scsi_disk_emulate_save_request,
2653 };
2654 
2655 static const SCSIReqOps scsi_disk_dma_reqops = {
2656     .size         = sizeof(SCSIDiskReq),
2657     .free_req     = scsi_free_request,
2658     .send_command = scsi_disk_dma_command,
2659     .read_data    = scsi_read_data,
2660     .write_data   = scsi_write_data,
2661     .get_buf      = scsi_get_buf,
2662     .load_request = scsi_disk_load_request,
2663     .save_request = scsi_disk_save_request,
2664 };
2665 
2666 static const SCSIReqOps *const scsi_disk_reqops_dispatch[256] = {
2667     [TEST_UNIT_READY]                 = &scsi_disk_emulate_reqops,
2668     [INQUIRY]                         = &scsi_disk_emulate_reqops,
2669     [MODE_SENSE]                      = &scsi_disk_emulate_reqops,
2670     [MODE_SENSE_10]                   = &scsi_disk_emulate_reqops,
2671     [START_STOP]                      = &scsi_disk_emulate_reqops,
2672     [ALLOW_MEDIUM_REMOVAL]            = &scsi_disk_emulate_reqops,
2673     [READ_CAPACITY_10]                = &scsi_disk_emulate_reqops,
2674     [READ_TOC]                        = &scsi_disk_emulate_reqops,
2675     [READ_DVD_STRUCTURE]              = &scsi_disk_emulate_reqops,
2676     [READ_DISC_INFORMATION]           = &scsi_disk_emulate_reqops,
2677     [GET_CONFIGURATION]               = &scsi_disk_emulate_reqops,
2678     [GET_EVENT_STATUS_NOTIFICATION]   = &scsi_disk_emulate_reqops,
2679     [MECHANISM_STATUS]                = &scsi_disk_emulate_reqops,
2680     [SERVICE_ACTION_IN_16]            = &scsi_disk_emulate_reqops,
2681     [REQUEST_SENSE]                   = &scsi_disk_emulate_reqops,
2682     [SYNCHRONIZE_CACHE]               = &scsi_disk_emulate_reqops,
2683     [SEEK_10]                         = &scsi_disk_emulate_reqops,
2684     [MODE_SELECT]                     = &scsi_disk_emulate_reqops,
2685     [MODE_SELECT_10]                  = &scsi_disk_emulate_reqops,
2686     [UNMAP]                           = &scsi_disk_emulate_reqops,
2687     [WRITE_SAME_10]                   = &scsi_disk_emulate_reqops,
2688     [WRITE_SAME_16]                   = &scsi_disk_emulate_reqops,
2689     [VERIFY_10]                       = &scsi_disk_emulate_reqops,
2690     [VERIFY_12]                       = &scsi_disk_emulate_reqops,
2691     [VERIFY_16]                       = &scsi_disk_emulate_reqops,
2692     [FORMAT_UNIT]                     = &scsi_disk_emulate_reqops,
2693 
2694     [READ_6]                          = &scsi_disk_dma_reqops,
2695     [READ_10]                         = &scsi_disk_dma_reqops,
2696     [READ_12]                         = &scsi_disk_dma_reqops,
2697     [READ_16]                         = &scsi_disk_dma_reqops,
2698     [WRITE_6]                         = &scsi_disk_dma_reqops,
2699     [WRITE_10]                        = &scsi_disk_dma_reqops,
2700     [WRITE_12]                        = &scsi_disk_dma_reqops,
2701     [WRITE_16]                        = &scsi_disk_dma_reqops,
2702     [WRITE_VERIFY_10]                 = &scsi_disk_dma_reqops,
2703     [WRITE_VERIFY_12]                 = &scsi_disk_dma_reqops,
2704     [WRITE_VERIFY_16]                 = &scsi_disk_dma_reqops,
2705 };
2706 
scsi_disk_new_request_dump(uint32_t lun,uint32_t tag,uint8_t * buf)2707 static void scsi_disk_new_request_dump(uint32_t lun, uint32_t tag, uint8_t *buf)
2708 {
2709     int len = scsi_cdb_length(buf);
2710     g_autoptr(GString) str = NULL;
2711 
2712     assert(len > 0 && len <= 16);
2713     str = qemu_hexdump_line(NULL, buf, len, 1, 0);
2714     trace_scsi_disk_new_request(lun, tag, str->str);
2715 }
2716 
scsi_new_request(SCSIDevice * d,uint32_t tag,uint32_t lun,uint8_t * buf,void * hba_private)2717 static SCSIRequest *scsi_new_request(SCSIDevice *d, uint32_t tag, uint32_t lun,
2718                                      uint8_t *buf, void *hba_private)
2719 {
2720     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
2721     SCSIRequest *req;
2722     const SCSIReqOps *ops;
2723     uint8_t command;
2724 
2725     command = buf[0];
2726     ops = scsi_disk_reqops_dispatch[command];
2727     if (!ops) {
2728         ops = &scsi_disk_emulate_reqops;
2729     }
2730     req = scsi_req_alloc(ops, &s->qdev, tag, lun, hba_private);
2731 
2732     if (trace_event_get_state_backends(TRACE_SCSI_DISK_NEW_REQUEST)) {
2733         scsi_disk_new_request_dump(lun, tag, buf);
2734     }
2735 
2736     return req;
2737 }
2738 
2739 #ifdef __linux__
get_device_type(SCSIDiskState * s)2740 static int get_device_type(SCSIDiskState *s)
2741 {
2742     uint8_t cmd[16];
2743     uint8_t buf[36];
2744     int ret;
2745 
2746     memset(cmd, 0, sizeof(cmd));
2747     memset(buf, 0, sizeof(buf));
2748     cmd[0] = INQUIRY;
2749     cmd[4] = sizeof(buf);
2750 
2751     ret = scsi_SG_IO_FROM_DEV(s->qdev.conf.blk, cmd, sizeof(cmd),
2752                               buf, sizeof(buf), s->qdev.io_timeout);
2753     if (ret < 0) {
2754         return -1;
2755     }
2756     s->qdev.type = buf[0];
2757     if (buf[1] & 0x80) {
2758         s->features |= 1 << SCSI_DISK_F_REMOVABLE;
2759     }
2760     return 0;
2761 }
2762 
scsi_block_realize(SCSIDevice * dev,Error ** errp)2763 static void scsi_block_realize(SCSIDevice *dev, Error **errp)
2764 {
2765     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, dev);
2766     int sg_version;
2767     int rc;
2768 
2769     if (!s->qdev.conf.blk) {
2770         error_setg(errp, "drive property not set");
2771         return;
2772     }
2773 
2774     if (s->rotation_rate) {
2775         error_report_once("rotation_rate is specified for scsi-block but is "
2776                           "not implemented. This option is deprecated and will "
2777                           "be removed in a future version");
2778     }
2779 
2780     /* check we are using a driver managing SG_IO (version 3 and after) */
2781     rc = blk_ioctl(s->qdev.conf.blk, SG_GET_VERSION_NUM, &sg_version);
2782     if (rc < 0) {
2783         error_setg_errno(errp, -rc, "cannot get SG_IO version number");
2784         if (rc != -EPERM) {
2785             error_append_hint(errp, "Is this a SCSI device?\n");
2786         }
2787         return;
2788     }
2789     if (sg_version < 30000) {
2790         error_setg(errp, "scsi generic interface too old");
2791         return;
2792     }
2793 
2794     /* get device type from INQUIRY data */
2795     rc = get_device_type(s);
2796     if (rc < 0) {
2797         error_setg(errp, "INQUIRY failed");
2798         return;
2799     }
2800 
2801     /* Make a guess for the block size, we'll fix it when the guest sends.
2802      * READ CAPACITY.  If they don't, they likely would assume these sizes
2803      * anyway. (TODO: check in /sys).
2804      */
2805     if (s->qdev.type == TYPE_ROM || s->qdev.type == TYPE_WORM) {
2806         s->qdev.blocksize = 2048;
2807     } else {
2808         s->qdev.blocksize = 512;
2809     }
2810 
2811     /* Makes the scsi-block device not removable by using HMP and QMP eject
2812      * command.
2813      */
2814     s->features |= (1 << SCSI_DISK_F_NO_REMOVABLE_DEVOPS);
2815 
2816     scsi_realize(&s->qdev, errp);
2817     scsi_generic_read_device_inquiry(&s->qdev);
2818 }
2819 
2820 typedef struct SCSIBlockReq {
2821     SCSIDiskReq req;
2822     sg_io_hdr_t io_header;
2823 
2824     /* Selected bytes of the original CDB, copied into our own CDB.  */
2825     uint8_t cmd, cdb1, group_number;
2826 
2827     /* CDB passed to SG_IO.  */
2828     uint8_t cdb[16];
2829     BlockCompletionFunc *cb;
2830     void *cb_opaque;
2831 } SCSIBlockReq;
2832 
scsi_block_sgio_complete(void * opaque,int ret)2833 static void scsi_block_sgio_complete(void *opaque, int ret)
2834 {
2835     SCSIBlockReq *req = (SCSIBlockReq *)opaque;
2836     SCSIDiskReq *r = &req->req;
2837     sg_io_hdr_t *io_hdr = &req->io_header;
2838 
2839     if (ret == 0) {
2840         if (io_hdr->host_status != SCSI_HOST_OK) {
2841             r->req.host_status = io_hdr->host_status;
2842             ret = -ENODEV;
2843         } else if (io_hdr->driver_status & SG_ERR_DRIVER_TIMEOUT) {
2844             ret = BUSY;
2845         } else {
2846             ret = io_hdr->status;
2847         }
2848     }
2849 
2850     req->cb(req->cb_opaque, ret);
2851 }
2852 
scsi_block_do_sgio(SCSIBlockReq * req,int64_t offset,QEMUIOVector * iov,int direction,BlockCompletionFunc * cb,void * opaque)2853 static BlockAIOCB *scsi_block_do_sgio(SCSIBlockReq *req,
2854                                       int64_t offset, QEMUIOVector *iov,
2855                                       int direction,
2856                                       BlockCompletionFunc *cb, void *opaque)
2857 {
2858     sg_io_hdr_t *io_header = &req->io_header;
2859     SCSIDiskReq *r = &req->req;
2860     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
2861     int nb_logical_blocks;
2862     uint64_t lba;
2863     BlockAIOCB *aiocb;
2864 
2865     /* This is not supported yet.  It can only happen if the guest does
2866      * reads and writes that are not aligned to one logical sectors
2867      * _and_ cover multiple MemoryRegions.
2868      */
2869     assert(offset % s->qdev.blocksize == 0);
2870     assert(iov->size % s->qdev.blocksize == 0);
2871 
2872     io_header->interface_id = 'S';
2873 
2874     /* The data transfer comes from the QEMUIOVector.  */
2875     io_header->dxfer_direction = direction;
2876     io_header->dxfer_len = iov->size;
2877     io_header->dxferp = (void *)iov->iov;
2878     io_header->iovec_count = iov->niov;
2879     assert(io_header->iovec_count == iov->niov); /* no overflow! */
2880 
2881     /* Build a new CDB with the LBA and length patched in, in case
2882      * DMA helpers split the transfer in multiple segments.  Do not
2883      * build a CDB smaller than what the guest wanted, and only build
2884      * a larger one if strictly necessary.
2885      */
2886     io_header->cmdp = req->cdb;
2887     lba = offset / s->qdev.blocksize;
2888     nb_logical_blocks = io_header->dxfer_len / s->qdev.blocksize;
2889 
2890     if ((req->cmd >> 5) == 0 && lba <= 0x1ffff) {
2891         /* 6-byte CDB */
2892         stl_be_p(&req->cdb[0], lba | (req->cmd << 24));
2893         req->cdb[4] = nb_logical_blocks;
2894         req->cdb[5] = 0;
2895         io_header->cmd_len = 6;
2896     } else if ((req->cmd >> 5) <= 1 && lba <= 0xffffffffULL) {
2897         /* 10-byte CDB */
2898         req->cdb[0] = (req->cmd & 0x1f) | 0x20;
2899         req->cdb[1] = req->cdb1;
2900         stl_be_p(&req->cdb[2], lba);
2901         req->cdb[6] = req->group_number;
2902         stw_be_p(&req->cdb[7], nb_logical_blocks);
2903         req->cdb[9] = 0;
2904         io_header->cmd_len = 10;
2905     } else if ((req->cmd >> 5) != 4 && lba <= 0xffffffffULL) {
2906         /* 12-byte CDB */
2907         req->cdb[0] = (req->cmd & 0x1f) | 0xA0;
2908         req->cdb[1] = req->cdb1;
2909         stl_be_p(&req->cdb[2], lba);
2910         stl_be_p(&req->cdb[6], nb_logical_blocks);
2911         req->cdb[10] = req->group_number;
2912         req->cdb[11] = 0;
2913         io_header->cmd_len = 12;
2914     } else {
2915         /* 16-byte CDB */
2916         req->cdb[0] = (req->cmd & 0x1f) | 0x80;
2917         req->cdb[1] = req->cdb1;
2918         stq_be_p(&req->cdb[2], lba);
2919         stl_be_p(&req->cdb[10], nb_logical_blocks);
2920         req->cdb[14] = req->group_number;
2921         req->cdb[15] = 0;
2922         io_header->cmd_len = 16;
2923     }
2924 
2925     /* The rest is as in scsi-generic.c.  */
2926     io_header->mx_sb_len = sizeof(r->req.sense);
2927     io_header->sbp = r->req.sense;
2928     io_header->timeout = s->qdev.io_timeout * 1000;
2929     io_header->usr_ptr = r;
2930     io_header->flags |= SG_FLAG_DIRECT_IO;
2931     req->cb = cb;
2932     req->cb_opaque = opaque;
2933     trace_scsi_disk_aio_sgio_command(r->req.tag, req->cdb[0], lba,
2934                                      nb_logical_blocks, io_header->timeout);
2935     aiocb = blk_aio_ioctl(s->qdev.conf.blk, SG_IO, io_header, scsi_block_sgio_complete, req);
2936     assert(aiocb != NULL);
2937     return aiocb;
2938 }
2939 
scsi_block_no_fua(SCSICommand * cmd)2940 static bool scsi_block_no_fua(SCSICommand *cmd)
2941 {
2942     return false;
2943 }
2944 
scsi_block_dma_readv(int64_t offset,QEMUIOVector * iov,BlockCompletionFunc * cb,void * cb_opaque,void * opaque)2945 static BlockAIOCB *scsi_block_dma_readv(int64_t offset,
2946                                         QEMUIOVector *iov,
2947                                         BlockCompletionFunc *cb, void *cb_opaque,
2948                                         void *opaque)
2949 {
2950     SCSIBlockReq *r = opaque;
2951     return scsi_block_do_sgio(r, offset, iov,
2952                               SG_DXFER_FROM_DEV, cb, cb_opaque);
2953 }
2954 
scsi_block_dma_writev(int64_t offset,QEMUIOVector * iov,BlockCompletionFunc * cb,void * cb_opaque,void * opaque)2955 static BlockAIOCB *scsi_block_dma_writev(int64_t offset,
2956                                          QEMUIOVector *iov,
2957                                          BlockCompletionFunc *cb, void *cb_opaque,
2958                                          void *opaque)
2959 {
2960     SCSIBlockReq *r = opaque;
2961     return scsi_block_do_sgio(r, offset, iov,
2962                               SG_DXFER_TO_DEV, cb, cb_opaque);
2963 }
2964 
scsi_block_is_passthrough(SCSIDiskState * s,uint8_t * buf)2965 static bool scsi_block_is_passthrough(SCSIDiskState *s, uint8_t *buf)
2966 {
2967     switch (buf[0]) {
2968     case VERIFY_10:
2969     case VERIFY_12:
2970     case VERIFY_16:
2971         /* Check if BYTCHK == 0x01 (data-out buffer contains data
2972          * for the number of logical blocks specified in the length
2973          * field).  For other modes, do not use scatter/gather operation.
2974          */
2975         if ((buf[1] & 6) == 2) {
2976             return false;
2977         }
2978         break;
2979 
2980     case READ_6:
2981     case READ_10:
2982     case READ_12:
2983     case READ_16:
2984     case WRITE_6:
2985     case WRITE_10:
2986     case WRITE_12:
2987     case WRITE_16:
2988     case WRITE_VERIFY_10:
2989     case WRITE_VERIFY_12:
2990     case WRITE_VERIFY_16:
2991         /* MMC writing cannot be done via DMA helpers, because it sometimes
2992          * involves writing beyond the maximum LBA or to negative LBA (lead-in).
2993          * We might use scsi_block_dma_reqops as long as no writing commands are
2994          * seen, but performance usually isn't paramount on optical media.  So,
2995          * just make scsi-block operate the same as scsi-generic for them.
2996          */
2997         if (s->qdev.type != TYPE_ROM) {
2998             return false;
2999         }
3000         break;
3001 
3002     default:
3003         break;
3004     }
3005 
3006     return true;
3007 }
3008 
3009 
scsi_block_dma_command(SCSIRequest * req,uint8_t * buf)3010 static int32_t scsi_block_dma_command(SCSIRequest *req, uint8_t *buf)
3011 {
3012     SCSIBlockReq *r = (SCSIBlockReq *)req;
3013     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, req->dev);
3014 
3015     r->cmd = req->cmd.buf[0];
3016     switch (r->cmd >> 5) {
3017     case 0:
3018         /* 6-byte CDB.  */
3019         r->cdb1 = r->group_number = 0;
3020         break;
3021     case 1:
3022         /* 10-byte CDB.  */
3023         r->cdb1 = req->cmd.buf[1];
3024         r->group_number = req->cmd.buf[6];
3025         break;
3026     case 4:
3027         /* 12-byte CDB.  */
3028         r->cdb1 = req->cmd.buf[1];
3029         r->group_number = req->cmd.buf[10];
3030         break;
3031     case 5:
3032         /* 16-byte CDB.  */
3033         r->cdb1 = req->cmd.buf[1];
3034         r->group_number = req->cmd.buf[14];
3035         break;
3036     default:
3037         abort();
3038     }
3039 
3040     /* Protection information is not supported.  For SCSI versions 2 and
3041      * older (as determined by snooping the guest's INQUIRY commands),
3042      * there is no RD/WR/VRPROTECT, so skip this check in these versions.
3043      */
3044     if (s->qdev.scsi_version > 2 && (req->cmd.buf[1] & 0xe0)) {
3045         scsi_check_condition(&r->req, SENSE_CODE(INVALID_FIELD));
3046         return 0;
3047     }
3048 
3049     return scsi_disk_dma_command(req, buf);
3050 }
3051 
3052 static const SCSIReqOps scsi_block_dma_reqops = {
3053     .size         = sizeof(SCSIBlockReq),
3054     .free_req     = scsi_free_request,
3055     .send_command = scsi_block_dma_command,
3056     .read_data    = scsi_read_data,
3057     .write_data   = scsi_write_data,
3058     .get_buf      = scsi_get_buf,
3059     .load_request = scsi_disk_load_request,
3060     .save_request = scsi_disk_save_request,
3061 };
3062 
scsi_block_new_request(SCSIDevice * d,uint32_t tag,uint32_t lun,uint8_t * buf,void * hba_private)3063 static SCSIRequest *scsi_block_new_request(SCSIDevice *d, uint32_t tag,
3064                                            uint32_t lun, uint8_t *buf,
3065                                            void *hba_private)
3066 {
3067     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
3068 
3069     if (scsi_block_is_passthrough(s, buf)) {
3070         return scsi_req_alloc(&scsi_generic_req_ops, &s->qdev, tag, lun,
3071                               hba_private);
3072     } else {
3073         return scsi_req_alloc(&scsi_block_dma_reqops, &s->qdev, tag, lun,
3074                               hba_private);
3075     }
3076 }
3077 
scsi_block_parse_cdb(SCSIDevice * d,SCSICommand * cmd,uint8_t * buf,size_t buf_len,void * hba_private)3078 static int scsi_block_parse_cdb(SCSIDevice *d, SCSICommand *cmd,
3079                                   uint8_t *buf, size_t buf_len,
3080                                   void *hba_private)
3081 {
3082     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, d);
3083 
3084     if (scsi_block_is_passthrough(s, buf)) {
3085         return scsi_bus_parse_cdb(&s->qdev, cmd, buf, buf_len, hba_private);
3086     } else {
3087         return scsi_req_parse_cdb(&s->qdev, cmd, buf, buf_len);
3088     }
3089 }
3090 
scsi_block_update_sense(SCSIRequest * req)3091 static void scsi_block_update_sense(SCSIRequest *req)
3092 {
3093     SCSIDiskReq *r = DO_UPCAST(SCSIDiskReq, req, req);
3094     SCSIBlockReq *br = DO_UPCAST(SCSIBlockReq, req, r);
3095     r->req.sense_len = MIN(br->io_header.sb_len_wr, sizeof(r->req.sense));
3096 }
3097 #endif
3098 
3099 static
scsi_dma_readv(int64_t offset,QEMUIOVector * iov,BlockCompletionFunc * cb,void * cb_opaque,void * opaque)3100 BlockAIOCB *scsi_dma_readv(int64_t offset, QEMUIOVector *iov,
3101                            BlockCompletionFunc *cb, void *cb_opaque,
3102                            void *opaque)
3103 {
3104     SCSIDiskReq *r = opaque;
3105     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
3106     return blk_aio_preadv(s->qdev.conf.blk, offset, iov, 0, cb, cb_opaque);
3107 }
3108 
3109 static
scsi_dma_writev(int64_t offset,QEMUIOVector * iov,BlockCompletionFunc * cb,void * cb_opaque,void * opaque)3110 BlockAIOCB *scsi_dma_writev(int64_t offset, QEMUIOVector *iov,
3111                             BlockCompletionFunc *cb, void *cb_opaque,
3112                             void *opaque)
3113 {
3114     SCSIDiskReq *r = opaque;
3115     SCSIDiskState *s = DO_UPCAST(SCSIDiskState, qdev, r->req.dev);
3116     int flags = r->need_fua ? BDRV_REQ_FUA : 0;
3117     return blk_aio_pwritev(s->qdev.conf.blk, offset, iov, flags, cb, cb_opaque);
3118 }
3119 
scsi_property_get_loadparm(Object * obj,Error ** errp)3120 static char *scsi_property_get_loadparm(Object *obj, Error **errp)
3121 {
3122     return g_strdup(SCSI_DISK_BASE(obj)->loadparm);
3123 }
3124 
scsi_property_set_loadparm(Object * obj,const char * value,Error ** errp)3125 static void scsi_property_set_loadparm(Object *obj, const char *value,
3126                                        Error **errp)
3127 {
3128     void *lp_str;
3129 
3130     if (object_property_get_int(obj, "bootindex", NULL) < 0) {
3131         error_setg(errp, "'loadparm' is only valid for boot devices");
3132         return;
3133     }
3134 
3135     lp_str = g_malloc0(strlen(value) + 1);
3136     if (!qdev_prop_sanitize_s390x_loadparm(lp_str, value, errp)) {
3137         g_free(lp_str);
3138         return;
3139     }
3140     SCSI_DISK_BASE(obj)->loadparm = lp_str;
3141 }
3142 
scsi_property_add_specifics(DeviceClass * dc)3143 static void scsi_property_add_specifics(DeviceClass *dc)
3144 {
3145     ObjectClass *oc = OBJECT_CLASS(dc);
3146 
3147     /* The loadparm property is only supported on s390x */
3148     if (qemu_arch_available(QEMU_ARCH_S390X)) {
3149         object_class_property_add_str(oc, "loadparm",
3150                                       scsi_property_get_loadparm,
3151                                       scsi_property_set_loadparm);
3152         object_class_property_set_description(oc, "loadparm",
3153                                               "load parameter (s390x only)");
3154     }
3155 }
3156 
scsi_disk_base_class_initfn(ObjectClass * klass,const void * data)3157 static void scsi_disk_base_class_initfn(ObjectClass *klass, const void *data)
3158 {
3159     DeviceClass *dc = DEVICE_CLASS(klass);
3160     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
3161 
3162     dc->fw_name = "disk";
3163     device_class_set_legacy_reset(dc, scsi_disk_reset);
3164     sdc->dma_readv = scsi_dma_readv;
3165     sdc->dma_writev = scsi_dma_writev;
3166     sdc->need_fua  = scsi_is_cmd_fua;
3167 }
3168 
3169 static const TypeInfo scsi_disk_base_info = {
3170     .name          = TYPE_SCSI_DISK_BASE,
3171     .parent        = TYPE_SCSI_DEVICE,
3172     .class_init    = scsi_disk_base_class_initfn,
3173     .instance_size = sizeof(SCSIDiskState),
3174     .class_size    = sizeof(SCSIDiskClass),
3175     .abstract      = true,
3176 };
3177 
3178 #define DEFINE_SCSI_DISK_PROPERTIES()                                   \
3179     DEFINE_PROP_DRIVE_IOTHREAD("drive", SCSIDiskState, qdev.conf.blk),  \
3180     DEFINE_BLOCK_PROPERTIES_BASE(SCSIDiskState, qdev.conf),             \
3181     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),            \
3182     DEFINE_PROP_STRING("ver", SCSIDiskState, version),                  \
3183     DEFINE_PROP_STRING("serial", SCSIDiskState, serial),                \
3184     DEFINE_PROP_STRING("vendor", SCSIDiskState, vendor),                \
3185     DEFINE_PROP_STRING("product", SCSIDiskState, product),              \
3186     DEFINE_PROP_STRING("device_id", SCSIDiskState, device_id),          \
3187     DEFINE_PROP_BOOL("migrate-emulated-scsi-request", SCSIDiskState, migrate_emulated_scsi_request, true)
3188 
3189 
3190 static const Property scsi_hd_properties[] = {
3191     DEFINE_SCSI_DISK_PROPERTIES(),
3192     DEFINE_PROP_BIT("removable", SCSIDiskState, features,
3193                     SCSI_DISK_F_REMOVABLE, false),
3194     DEFINE_PROP_BIT("dpofua", SCSIDiskState, features,
3195                     SCSI_DISK_F_DPOFUA, true),
3196     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3197     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3198     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3199     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3200                        DEFAULT_MAX_UNMAP_SIZE),
3201     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3202                        DEFAULT_MAX_IO_SIZE),
3203     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3204     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3205                       5),
3206     DEFINE_PROP_BIT("quirk_mode_page_vendor_specific_apple", SCSIDiskState,
3207                     quirks, SCSI_DISK_QUIRK_MODE_PAGE_VENDOR_SPECIFIC_APPLE,
3208                     0),
3209     DEFINE_BLOCK_CHS_PROPERTIES(SCSIDiskState, qdev.conf),
3210 };
3211 
3212 static const VMStateDescription vmstate_scsi_disk_state = {
3213     .name = "scsi-disk",
3214     .version_id = 1,
3215     .minimum_version_id = 1,
3216     .fields = (const VMStateField[]) {
3217         VMSTATE_SCSI_DEVICE(qdev, SCSIDiskState),
3218         VMSTATE_BOOL(media_changed, SCSIDiskState),
3219         VMSTATE_BOOL(media_event, SCSIDiskState),
3220         VMSTATE_BOOL(eject_request, SCSIDiskState),
3221         VMSTATE_BOOL(tray_open, SCSIDiskState),
3222         VMSTATE_BOOL(tray_locked, SCSIDiskState),
3223         VMSTATE_END_OF_LIST()
3224     }
3225 };
3226 
scsi_hd_class_initfn(ObjectClass * klass,const void * data)3227 static void scsi_hd_class_initfn(ObjectClass *klass, const void *data)
3228 {
3229     DeviceClass *dc = DEVICE_CLASS(klass);
3230     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3231 
3232     sc->realize      = scsi_hd_realize;
3233     sc->unrealize    = scsi_unrealize;
3234     sc->alloc_req    = scsi_new_request;
3235     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3236     dc->desc = "virtual SCSI disk";
3237     device_class_set_props(dc, scsi_hd_properties);
3238     dc->vmsd  = &vmstate_scsi_disk_state;
3239 
3240     scsi_property_add_specifics(dc);
3241 }
3242 
3243 static const TypeInfo scsi_hd_info = {
3244     .name          = "scsi-hd",
3245     .parent        = TYPE_SCSI_DISK_BASE,
3246     .class_init    = scsi_hd_class_initfn,
3247 };
3248 
3249 static const Property scsi_cd_properties[] = {
3250     DEFINE_SCSI_DISK_PROPERTIES(),
3251     DEFINE_PROP_UINT64("wwn", SCSIDiskState, qdev.wwn, 0),
3252     DEFINE_PROP_UINT64("port_wwn", SCSIDiskState, qdev.port_wwn, 0),
3253     DEFINE_PROP_UINT16("port_index", SCSIDiskState, port_index, 0),
3254     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3255                        DEFAULT_MAX_IO_SIZE),
3256     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3257                       5),
3258     DEFINE_PROP_BIT("quirk_mode_page_apple_vendor", SCSIDiskState, quirks,
3259                     SCSI_DISK_QUIRK_MODE_PAGE_APPLE_VENDOR, 0),
3260     DEFINE_PROP_BIT("quirk_mode_sense_rom_use_dbd", SCSIDiskState, quirks,
3261                     SCSI_DISK_QUIRK_MODE_SENSE_ROM_USE_DBD, 0),
3262     DEFINE_PROP_BIT("quirk_mode_page_vendor_specific_apple", SCSIDiskState,
3263                     quirks, SCSI_DISK_QUIRK_MODE_PAGE_VENDOR_SPECIFIC_APPLE,
3264                     0),
3265     DEFINE_PROP_BIT("quirk_mode_page_truncated", SCSIDiskState, quirks,
3266                     SCSI_DISK_QUIRK_MODE_PAGE_TRUNCATED, 0),
3267 };
3268 
scsi_cd_class_initfn(ObjectClass * klass,const void * data)3269 static void scsi_cd_class_initfn(ObjectClass *klass, const void *data)
3270 {
3271     DeviceClass *dc = DEVICE_CLASS(klass);
3272     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3273 
3274     sc->realize      = scsi_cd_realize;
3275     sc->alloc_req    = scsi_new_request;
3276     sc->unit_attention_reported = scsi_disk_unit_attention_reported;
3277     dc->desc = "virtual SCSI CD-ROM";
3278     device_class_set_props(dc, scsi_cd_properties);
3279     dc->vmsd  = &vmstate_scsi_disk_state;
3280 
3281     scsi_property_add_specifics(dc);
3282 }
3283 
3284 static const TypeInfo scsi_cd_info = {
3285     .name          = "scsi-cd",
3286     .parent        = TYPE_SCSI_DISK_BASE,
3287     .class_init    = scsi_cd_class_initfn,
3288 };
3289 
3290 #ifdef __linux__
3291 static const Property scsi_block_properties[] = {
3292     DEFINE_BLOCK_ERROR_PROPERTIES(SCSIDiskState, qdev.conf),
3293     DEFINE_PROP_DRIVE("drive", SCSIDiskState, qdev.conf.blk),
3294     DEFINE_PROP_BOOL("share-rw", SCSIDiskState, qdev.conf.share_rw, false),
3295     DEFINE_PROP_UINT16("rotation_rate", SCSIDiskState, rotation_rate, 0),
3296     DEFINE_PROP_UINT64("max_unmap_size", SCSIDiskState, max_unmap_size,
3297                        DEFAULT_MAX_UNMAP_SIZE),
3298     DEFINE_PROP_UINT64("max_io_size", SCSIDiskState, max_io_size,
3299                        DEFAULT_MAX_IO_SIZE),
3300     DEFINE_PROP_INT32("scsi_version", SCSIDiskState, qdev.default_scsi_version,
3301                       -1),
3302     DEFINE_PROP_UINT32("io_timeout", SCSIDiskState, qdev.io_timeout,
3303                        DEFAULT_IO_TIMEOUT),
3304 };
3305 
scsi_block_class_initfn(ObjectClass * klass,const void * data)3306 static void scsi_block_class_initfn(ObjectClass *klass, const void *data)
3307 {
3308     DeviceClass *dc = DEVICE_CLASS(klass);
3309     SCSIDeviceClass *sc = SCSI_DEVICE_CLASS(klass);
3310     SCSIDiskClass *sdc = SCSI_DISK_BASE_CLASS(klass);
3311 
3312     sc->realize      = scsi_block_realize;
3313     sc->alloc_req    = scsi_block_new_request;
3314     sc->parse_cdb    = scsi_block_parse_cdb;
3315     sdc->dma_readv   = scsi_block_dma_readv;
3316     sdc->dma_writev  = scsi_block_dma_writev;
3317     sdc->update_sense = scsi_block_update_sense;
3318     sdc->need_fua    = scsi_block_no_fua;
3319     dc->desc = "SCSI block device passthrough";
3320     device_class_set_props(dc, scsi_block_properties);
3321     dc->vmsd  = &vmstate_scsi_disk_state;
3322 }
3323 
3324 static const TypeInfo scsi_block_info = {
3325     .name          = "scsi-block",
3326     .parent        = TYPE_SCSI_DISK_BASE,
3327     .class_init    = scsi_block_class_initfn,
3328 };
3329 #endif
3330 
scsi_disk_register_types(void)3331 static void scsi_disk_register_types(void)
3332 {
3333     type_register_static(&scsi_disk_base_info);
3334     type_register_static(&scsi_hd_info);
3335     type_register_static(&scsi_cd_info);
3336 #ifdef __linux__
3337     type_register_static(&scsi_block_info);
3338 #endif
3339 }
3340 
3341 type_init(scsi_disk_register_types)
3342