xref: /openbmc/qemu/hw/block/xen-block.c (revision 06831001)
1 /*
2  * Copyright (c) 2018  Citrix Systems Inc.
3  *
4  * This work is licensed under the terms of the GNU GPL, version 2 or later.
5  * See the COPYING file in the top-level directory.
6  */
7 
8 #include "qemu/osdep.h"
9 #include "qemu/cutils.h"
10 #include "qemu/main-loop.h"
11 #include "qemu/module.h"
12 #include "qemu/option.h"
13 #include "qapi/error.h"
14 #include "qapi/qapi-commands-block-core.h"
15 #include "qapi/qapi-commands-qom.h"
16 #include "qapi/qapi-visit-block-core.h"
17 #include "qapi/qobject-input-visitor.h"
18 #include "qapi/visitor.h"
19 #include "qapi/qmp/qdict.h"
20 #include "qapi/qmp/qstring.h"
21 #include "qom/object_interfaces.h"
22 #include "hw/block/xen_blkif.h"
23 #include "hw/qdev-properties.h"
24 #include "hw/xen/xen-block.h"
25 #include "hw/xen/xen-backend.h"
26 #include "sysemu/blockdev.h"
27 #include "sysemu/block-backend.h"
28 #include "sysemu/iothread.h"
29 #include "dataplane/xen-block.h"
30 #include "trace.h"
31 
32 static char *xen_block_get_name(XenDevice *xendev, Error **errp)
33 {
34     XenBlockDevice *blockdev = XEN_BLOCK_DEVICE(xendev);
35     XenBlockVdev *vdev = &blockdev->props.vdev;
36 
37     return g_strdup_printf("%lu", vdev->number);
38 }
39 
40 static void xen_block_disconnect(XenDevice *xendev, Error **errp)
41 {
42     XenBlockDevice *blockdev = XEN_BLOCK_DEVICE(xendev);
43     const char *type = object_get_typename(OBJECT(blockdev));
44     XenBlockVdev *vdev = &blockdev->props.vdev;
45 
46     trace_xen_block_disconnect(type, vdev->disk, vdev->partition);
47 
48     xen_block_dataplane_stop(blockdev->dataplane);
49 }
50 
51 static void xen_block_connect(XenDevice *xendev, Error **errp)
52 {
53     XenBlockDevice *blockdev = XEN_BLOCK_DEVICE(xendev);
54     const char *type = object_get_typename(OBJECT(blockdev));
55     XenBlockVdev *vdev = &blockdev->props.vdev;
56     BlockConf *conf = &blockdev->props.conf;
57     unsigned int feature_large_sector_size;
58     unsigned int order, nr_ring_ref, *ring_ref, event_channel, protocol;
59     char *str;
60 
61     trace_xen_block_connect(type, vdev->disk, vdev->partition);
62 
63     if (xen_device_frontend_scanf(xendev, "feature-large-sector-size", "%u",
64                                   &feature_large_sector_size) != 1) {
65         feature_large_sector_size = 0;
66     }
67 
68     if (feature_large_sector_size != 1 &&
69         conf->logical_block_size != XEN_BLKIF_SECTOR_SIZE) {
70         error_setg(errp, "logical_block_size != %u not supported by frontend",
71                    XEN_BLKIF_SECTOR_SIZE);
72         return;
73     }
74 
75     if (xen_device_frontend_scanf(xendev, "ring-page-order", "%u",
76                                   &order) != 1) {
77         nr_ring_ref = 1;
78         ring_ref = g_new(unsigned int, nr_ring_ref);
79 
80         if (xen_device_frontend_scanf(xendev, "ring-ref", "%u",
81                                       &ring_ref[0]) != 1) {
82             error_setg(errp, "failed to read ring-ref");
83             g_free(ring_ref);
84             return;
85         }
86     } else if (qemu_xen_gnttab_can_map_multi() &&
87                order <= blockdev->props.max_ring_page_order) {
88         unsigned int i;
89 
90         nr_ring_ref = 1 << order;
91         ring_ref = g_new(unsigned int, nr_ring_ref);
92 
93         for (i = 0; i < nr_ring_ref; i++) {
94             const char *key = g_strdup_printf("ring-ref%u", i);
95 
96             if (xen_device_frontend_scanf(xendev, key, "%u",
97                                           &ring_ref[i]) != 1) {
98                 error_setg(errp, "failed to read %s", key);
99                 g_free((gpointer)key);
100                 g_free(ring_ref);
101                 return;
102             }
103 
104             g_free((gpointer)key);
105         }
106     } else {
107         error_setg(errp, "invalid ring-page-order (%d)", order);
108         return;
109     }
110 
111     if (xen_device_frontend_scanf(xendev, "event-channel", "%u",
112                                   &event_channel) != 1) {
113         error_setg(errp, "failed to read event-channel");
114         g_free(ring_ref);
115         return;
116     }
117 
118     if (xen_device_frontend_scanf(xendev, "protocol", "%ms",
119                                   &str) != 1) {
120         protocol = BLKIF_PROTOCOL_NATIVE;
121     } else {
122         if (strcmp(str, XEN_IO_PROTO_ABI_X86_32) == 0) {
123             protocol = BLKIF_PROTOCOL_X86_32;
124         } else if (strcmp(str, XEN_IO_PROTO_ABI_X86_64) == 0) {
125             protocol = BLKIF_PROTOCOL_X86_64;
126         } else {
127             protocol = BLKIF_PROTOCOL_NATIVE;
128         }
129 
130         free(str);
131     }
132 
133     xen_block_dataplane_start(blockdev->dataplane, ring_ref, nr_ring_ref,
134                               event_channel, protocol, errp);
135 
136     g_free(ring_ref);
137 }
138 
139 static void xen_block_unrealize(XenDevice *xendev)
140 {
141     XenBlockDevice *blockdev = XEN_BLOCK_DEVICE(xendev);
142     XenBlockDeviceClass *blockdev_class =
143         XEN_BLOCK_DEVICE_GET_CLASS(xendev);
144     const char *type = object_get_typename(OBJECT(blockdev));
145     XenBlockVdev *vdev = &blockdev->props.vdev;
146 
147     if (vdev->type == XEN_BLOCK_VDEV_TYPE_INVALID) {
148         return;
149     }
150 
151     trace_xen_block_unrealize(type, vdev->disk, vdev->partition);
152 
153     /* Disconnect from the frontend in case this has not already happened */
154     xen_block_disconnect(xendev, NULL);
155 
156     xen_block_dataplane_destroy(blockdev->dataplane);
157     blockdev->dataplane = NULL;
158 
159     if (blockdev_class->unrealize) {
160         blockdev_class->unrealize(blockdev);
161     }
162 }
163 
164 static void xen_block_set_size(XenBlockDevice *blockdev)
165 {
166     const char *type = object_get_typename(OBJECT(blockdev));
167     XenBlockVdev *vdev = &blockdev->props.vdev;
168     BlockConf *conf = &blockdev->props.conf;
169     int64_t sectors = blk_getlength(conf->blk) / conf->logical_block_size;
170     XenDevice *xendev = XEN_DEVICE(blockdev);
171 
172     trace_xen_block_size(type, vdev->disk, vdev->partition, sectors);
173 
174     xen_device_backend_printf(xendev, "sectors", "%"PRIi64, sectors);
175 }
176 
177 static void xen_block_resize_cb(void *opaque)
178 {
179     XenBlockDevice *blockdev = opaque;
180     XenDevice *xendev = XEN_DEVICE(blockdev);
181     enum xenbus_state state = xen_device_backend_get_state(xendev);
182 
183     xen_block_set_size(blockdev);
184 
185     /*
186      * Mimic the behaviour of Linux xen-blkback and re-write the state
187      * to trigger the frontend watch.
188      */
189     xen_device_backend_printf(xendev, "state", "%u", state);
190 }
191 
192 /* Suspend request handling */
193 static void xen_block_drained_begin(void *opaque)
194 {
195     XenBlockDevice *blockdev = opaque;
196 
197     xen_block_dataplane_detach(blockdev->dataplane);
198 }
199 
200 /* Resume request handling */
201 static void xen_block_drained_end(void *opaque)
202 {
203     XenBlockDevice *blockdev = opaque;
204 
205     xen_block_dataplane_attach(blockdev->dataplane);
206 }
207 
208 static const BlockDevOps xen_block_dev_ops = {
209     .resize_cb     = xen_block_resize_cb,
210     .drained_begin = xen_block_drained_begin,
211     .drained_end   = xen_block_drained_end,
212 };
213 
214 static void xen_block_realize(XenDevice *xendev, Error **errp)
215 {
216     ERRP_GUARD();
217     XenBlockDevice *blockdev = XEN_BLOCK_DEVICE(xendev);
218     XenBlockDeviceClass *blockdev_class =
219         XEN_BLOCK_DEVICE_GET_CLASS(xendev);
220     const char *type = object_get_typename(OBJECT(blockdev));
221     XenBlockVdev *vdev = &blockdev->props.vdev;
222     BlockConf *conf = &blockdev->props.conf;
223     BlockBackend *blk = conf->blk;
224 
225     if (vdev->type == XEN_BLOCK_VDEV_TYPE_INVALID) {
226         error_setg(errp, "vdev property not set");
227         return;
228     }
229 
230     trace_xen_block_realize(type, vdev->disk, vdev->partition);
231 
232     if (blockdev_class->realize) {
233         blockdev_class->realize(blockdev, errp);
234         if (*errp) {
235             return;
236         }
237     }
238 
239     /*
240      * The blkif protocol does not deal with removable media, so it must
241      * always be present, even for CDRom devices.
242      */
243     assert(blk);
244     if (!blk_is_inserted(blk)) {
245         error_setg(errp, "device needs media, but drive is empty");
246         return;
247     }
248 
249     if (!blkconf_apply_backend_options(conf, blockdev->info & VDISK_READONLY,
250                                        true, errp)) {
251         return;
252     }
253 
254     if (!(blockdev->info & VDISK_CDROM) &&
255         !blkconf_geometry(conf, NULL, 65535, 255, 255, errp)) {
256         return;
257     }
258 
259     if (!blkconf_blocksizes(conf, errp)) {
260         return;
261     }
262 
263     if (conf->discard_granularity == -1) {
264         conf->discard_granularity = conf->physical_block_size;
265     }
266 
267     if (blk_get_flags(blk) & BDRV_O_UNMAP) {
268         xen_device_backend_printf(xendev, "feature-discard", "%u", 1);
269         xen_device_backend_printf(xendev, "discard-granularity", "%u",
270                                   conf->discard_granularity);
271         xen_device_backend_printf(xendev, "discard-alignment", "%u", 0);
272     }
273 
274     xen_device_backend_printf(xendev, "feature-flush-cache", "%u", 1);
275 
276     if (qemu_xen_gnttab_can_map_multi()) {
277         xen_device_backend_printf(xendev, "max-ring-page-order", "%u",
278                                   blockdev->props.max_ring_page_order);
279     }
280 
281     xen_device_backend_printf(xendev, "info", "%u", blockdev->info);
282 
283     xen_device_frontend_printf(xendev, "virtual-device", "%lu",
284                                vdev->number);
285     xen_device_frontend_printf(xendev, "device-type", "%s",
286                                blockdev->device_type);
287 
288     xen_device_backend_printf(xendev, "sector-size", "%u",
289                               conf->logical_block_size);
290 
291     xen_block_set_size(blockdev);
292 
293     blockdev->dataplane =
294         xen_block_dataplane_create(xendev, blk, conf->logical_block_size,
295                                    blockdev->props.iothread);
296 
297     blk_set_dev_ops(blk, &xen_block_dev_ops, blockdev);
298 }
299 
300 static void xen_block_frontend_changed(XenDevice *xendev,
301                                        enum xenbus_state frontend_state,
302                                        Error **errp)
303 {
304     ERRP_GUARD();
305     enum xenbus_state backend_state = xen_device_backend_get_state(xendev);
306 
307     switch (frontend_state) {
308     case XenbusStateInitialised:
309     case XenbusStateConnected:
310         if (backend_state == XenbusStateConnected) {
311             break;
312         }
313 
314         xen_block_disconnect(xendev, errp);
315         if (*errp) {
316             break;
317         }
318 
319         xen_block_connect(xendev, errp);
320         if (*errp) {
321             break;
322         }
323 
324         xen_device_backend_set_state(xendev, XenbusStateConnected);
325         break;
326 
327     case XenbusStateClosing:
328         xen_device_backend_set_state(xendev, XenbusStateClosing);
329         break;
330 
331     case XenbusStateClosed:
332     case XenbusStateUnknown:
333         xen_block_disconnect(xendev, errp);
334         if (*errp) {
335             break;
336         }
337 
338         xen_device_backend_set_state(xendev, XenbusStateClosed);
339         break;
340 
341     default:
342         break;
343     }
344 }
345 
346 static char *disk_to_vbd_name(unsigned int disk)
347 {
348     char *name, *prefix = (disk >= 26) ?
349         disk_to_vbd_name((disk / 26) - 1) : g_strdup("");
350 
351     name = g_strdup_printf("%s%c", prefix, 'a' + disk % 26);
352     g_free(prefix);
353 
354     return name;
355 }
356 
357 static void xen_block_get_vdev(Object *obj, Visitor *v, const char *name,
358                                void *opaque, Error **errp)
359 {
360     Property *prop = opaque;
361     XenBlockVdev *vdev = object_field_prop_ptr(obj, prop);
362     char *str;
363 
364     switch (vdev->type) {
365     case XEN_BLOCK_VDEV_TYPE_DP:
366         str = g_strdup_printf("d%lup%lu", vdev->disk, vdev->partition);
367         break;
368 
369     case XEN_BLOCK_VDEV_TYPE_XVD:
370     case XEN_BLOCK_VDEV_TYPE_HD:
371     case XEN_BLOCK_VDEV_TYPE_SD: {
372         char *name = disk_to_vbd_name(vdev->disk);
373 
374         str = g_strdup_printf("%s%s%lu",
375                               (vdev->type == XEN_BLOCK_VDEV_TYPE_XVD) ?
376                               "xvd" :
377                               (vdev->type == XEN_BLOCK_VDEV_TYPE_HD) ?
378                               "hd" :
379                               "sd",
380                               name, vdev->partition);
381         g_free(name);
382         break;
383     }
384     default:
385         error_setg(errp, "invalid vdev type");
386         return;
387     }
388 
389     visit_type_str(v, name, &str, errp);
390     g_free(str);
391 }
392 
393 static int vbd_name_to_disk(const char *name, const char **endp,
394                             unsigned long *disk)
395 {
396     unsigned int n = 0;
397 
398     while (*name != '\0') {
399         if (!g_ascii_isalpha(*name) || !g_ascii_islower(*name)) {
400             break;
401         }
402 
403         n *= 26;
404         n += *name++ - 'a' + 1;
405     }
406     *endp = name;
407 
408     if (!n) {
409         return -1;
410     }
411 
412     *disk = n - 1;
413 
414     return 0;
415 }
416 
417 static void xen_block_set_vdev(Object *obj, Visitor *v, const char *name,
418                                void *opaque, Error **errp)
419 {
420     Property *prop = opaque;
421     XenBlockVdev *vdev = object_field_prop_ptr(obj, prop);
422     char *str, *p;
423     const char *end;
424 
425     if (!visit_type_str(v, name, &str, errp)) {
426         return;
427     }
428 
429     p = strchr(str, 'd');
430     if (!p) {
431         goto invalid;
432     }
433 
434     *p++ = '\0';
435     if (*str == '\0') {
436         vdev->type = XEN_BLOCK_VDEV_TYPE_DP;
437     } else if (strcmp(str, "xv") == 0) {
438         vdev->type = XEN_BLOCK_VDEV_TYPE_XVD;
439     } else if (strcmp(str, "h") == 0) {
440         vdev->type = XEN_BLOCK_VDEV_TYPE_HD;
441     } else if (strcmp(str, "s") == 0) {
442         vdev->type = XEN_BLOCK_VDEV_TYPE_SD;
443     } else {
444         goto invalid;
445     }
446 
447     if (vdev->type == XEN_BLOCK_VDEV_TYPE_DP) {
448         if (qemu_strtoul(p, &end, 10, &vdev->disk)) {
449             goto invalid;
450         }
451 
452         if (*end == 'p') {
453             if (*(++end) == '\0') {
454                 goto invalid;
455             }
456         }
457     } else {
458         if (vbd_name_to_disk(p, &end, &vdev->disk)) {
459             goto invalid;
460         }
461     }
462 
463     if (*end != '\0') {
464         p = (char *)end;
465 
466         if (qemu_strtoul(p, &end, 10, &vdev->partition)) {
467             goto invalid;
468         }
469 
470         if (*end != '\0') {
471             goto invalid;
472         }
473     } else {
474         vdev->partition = 0;
475     }
476 
477     switch (vdev->type) {
478     case XEN_BLOCK_VDEV_TYPE_DP:
479     case XEN_BLOCK_VDEV_TYPE_XVD:
480         if (vdev->disk < (1 << 4) && vdev->partition < (1 << 4)) {
481             vdev->number = (202 << 8) | (vdev->disk << 4) |
482                 vdev->partition;
483         } else if (vdev->disk < (1 << 20) && vdev->partition < (1 << 8)) {
484             vdev->number = (1 << 28) | (vdev->disk << 8) |
485                 vdev->partition;
486         } else {
487             goto invalid;
488         }
489         break;
490 
491     case XEN_BLOCK_VDEV_TYPE_HD:
492         if ((vdev->disk == 0 || vdev->disk == 1) &&
493             vdev->partition < (1 << 6)) {
494             vdev->number = (3 << 8) | (vdev->disk << 6) | vdev->partition;
495         } else if ((vdev->disk == 2 || vdev->disk == 3) &&
496                    vdev->partition < (1 << 6)) {
497             vdev->number = (22 << 8) | ((vdev->disk - 2) << 6) |
498                 vdev->partition;
499         } else {
500             goto invalid;
501         }
502         break;
503 
504     case XEN_BLOCK_VDEV_TYPE_SD:
505         if (vdev->disk < (1 << 4) && vdev->partition < (1 << 4)) {
506             vdev->number = (8 << 8) | (vdev->disk << 4) | vdev->partition;
507         } else {
508             goto invalid;
509         }
510         break;
511 
512     default:
513         goto invalid;
514     }
515 
516     g_free(str);
517     return;
518 
519 invalid:
520     error_setg(errp, "invalid virtual disk specifier");
521 
522     vdev->type = XEN_BLOCK_VDEV_TYPE_INVALID;
523     g_free(str);
524 }
525 
526 /*
527  * This property deals with 'vdev' names adhering to the Xen VBD naming
528  * scheme described in:
529  *
530  * https://xenbits.xen.org/docs/unstable/man/xen-vbd-interface.7.html
531  */
532 const PropertyInfo xen_block_prop_vdev = {
533     .name  = "str",
534     .description = "Virtual Disk specifier: d*p*/xvd*/hd*/sd*",
535     .get = xen_block_get_vdev,
536     .set = xen_block_set_vdev,
537 };
538 
539 static Property xen_block_props[] = {
540     DEFINE_PROP("vdev", XenBlockDevice, props.vdev,
541                 xen_block_prop_vdev, XenBlockVdev),
542     DEFINE_BLOCK_PROPERTIES(XenBlockDevice, props.conf),
543     DEFINE_PROP_UINT32("max-ring-page-order", XenBlockDevice,
544                        props.max_ring_page_order, 4),
545     DEFINE_PROP_LINK("iothread", XenBlockDevice, props.iothread,
546                      TYPE_IOTHREAD, IOThread *),
547     DEFINE_PROP_END_OF_LIST()
548 };
549 
550 static void xen_block_class_init(ObjectClass *class, void *data)
551 {
552     DeviceClass *dev_class = DEVICE_CLASS(class);
553     XenDeviceClass *xendev_class = XEN_DEVICE_CLASS(class);
554 
555     xendev_class->backend = "qdisk";
556     xendev_class->device = "vbd";
557     xendev_class->get_name = xen_block_get_name;
558     xendev_class->realize = xen_block_realize;
559     xendev_class->frontend_changed = xen_block_frontend_changed;
560     xendev_class->unrealize = xen_block_unrealize;
561 
562     device_class_set_props(dev_class, xen_block_props);
563 }
564 
565 static const TypeInfo xen_block_type_info = {
566     .name = TYPE_XEN_BLOCK_DEVICE,
567     .parent = TYPE_XEN_DEVICE,
568     .instance_size = sizeof(XenBlockDevice),
569     .abstract = true,
570     .class_size = sizeof(XenBlockDeviceClass),
571     .class_init = xen_block_class_init,
572 };
573 
574 static void xen_disk_unrealize(XenBlockDevice *blockdev)
575 {
576     trace_xen_disk_unrealize();
577 }
578 
579 static void xen_disk_realize(XenBlockDevice *blockdev, Error **errp)
580 {
581     BlockConf *conf = &blockdev->props.conf;
582 
583     trace_xen_disk_realize();
584 
585     blockdev->device_type = "disk";
586 
587     if (!conf->blk) {
588         error_setg(errp, "drive property not set");
589         return;
590     }
591 
592     blockdev->info = blk_supports_write_perm(conf->blk) ? 0 : VDISK_READONLY;
593 }
594 
595 static void xen_disk_class_init(ObjectClass *class, void *data)
596 {
597     DeviceClass *dev_class = DEVICE_CLASS(class);
598     XenBlockDeviceClass *blockdev_class = XEN_BLOCK_DEVICE_CLASS(class);
599 
600     blockdev_class->realize = xen_disk_realize;
601     blockdev_class->unrealize = xen_disk_unrealize;
602 
603     dev_class->desc = "Xen Disk Device";
604 }
605 
606 static const TypeInfo xen_disk_type_info = {
607     .name = TYPE_XEN_DISK_DEVICE,
608     .parent = TYPE_XEN_BLOCK_DEVICE,
609     .instance_size = sizeof(XenDiskDevice),
610     .class_init = xen_disk_class_init,
611 };
612 
613 static void xen_cdrom_unrealize(XenBlockDevice *blockdev)
614 {
615     trace_xen_cdrom_unrealize();
616 }
617 
618 static void xen_cdrom_realize(XenBlockDevice *blockdev, Error **errp)
619 {
620     BlockConf *conf = &blockdev->props.conf;
621 
622     trace_xen_cdrom_realize();
623 
624     blockdev->device_type = "cdrom";
625 
626     if (!conf->blk) {
627         int rc;
628 
629         /* Set up an empty drive */
630         conf->blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
631 
632         rc = blk_attach_dev(conf->blk, DEVICE(blockdev));
633         if (!rc) {
634             error_setg_errno(errp, -rc, "failed to create drive");
635             return;
636         }
637     }
638 
639     blockdev->info = VDISK_READONLY | VDISK_CDROM;
640 }
641 
642 static void xen_cdrom_class_init(ObjectClass *class, void *data)
643 {
644     DeviceClass *dev_class = DEVICE_CLASS(class);
645     XenBlockDeviceClass *blockdev_class = XEN_BLOCK_DEVICE_CLASS(class);
646 
647     blockdev_class->realize = xen_cdrom_realize;
648     blockdev_class->unrealize = xen_cdrom_unrealize;
649 
650     dev_class->desc = "Xen CD-ROM Device";
651 }
652 
653 static const TypeInfo xen_cdrom_type_info = {
654     .name = TYPE_XEN_CDROM_DEVICE,
655     .parent = TYPE_XEN_BLOCK_DEVICE,
656     .instance_size = sizeof(XenCDRomDevice),
657     .class_init = xen_cdrom_class_init,
658 };
659 
660 static void xen_block_register_types(void)
661 {
662     type_register_static(&xen_block_type_info);
663     type_register_static(&xen_disk_type_info);
664     type_register_static(&xen_cdrom_type_info);
665 }
666 
667 type_init(xen_block_register_types)
668 
669 static void xen_block_blockdev_del(const char *node_name, Error **errp)
670 {
671     trace_xen_block_blockdev_del(node_name);
672 
673     qmp_blockdev_del(node_name, errp);
674 }
675 
676 static char *xen_block_blockdev_add(const char *id, QDict *qdict,
677                                     Error **errp)
678 {
679     ERRP_GUARD();
680     const char *driver = qdict_get_try_str(qdict, "driver");
681     BlockdevOptions *options = NULL;
682     char *node_name;
683     Visitor *v;
684 
685     if (!driver) {
686         error_setg(errp, "no 'driver' parameter");
687         return NULL;
688     }
689 
690     node_name = g_strdup_printf("%s-%s", id, driver);
691     qdict_put_str(qdict, "node-name", node_name);
692 
693     trace_xen_block_blockdev_add(node_name);
694 
695     v = qobject_input_visitor_new(QOBJECT(qdict));
696     visit_type_BlockdevOptions(v, NULL, &options, errp);
697     visit_free(v);
698     if (!options) {
699         goto fail;
700     }
701 
702     qmp_blockdev_add(options, errp);
703 
704     if (*errp) {
705         goto fail;
706     }
707 
708     qapi_free_BlockdevOptions(options);
709 
710     return node_name;
711 
712 fail:
713     if (options) {
714         qapi_free_BlockdevOptions(options);
715     }
716     g_free(node_name);
717 
718     return NULL;
719 }
720 
721 static void xen_block_drive_destroy(XenBlockDrive *drive, Error **errp)
722 {
723     ERRP_GUARD();
724     char *node_name = drive->node_name;
725 
726     if (node_name) {
727         xen_block_blockdev_del(node_name, errp);
728         if (*errp) {
729             return;
730         }
731         g_free(node_name);
732         drive->node_name = NULL;
733     }
734     g_free(drive->id);
735     g_free(drive);
736 }
737 
738 static XenBlockDrive *xen_block_drive_create(const char *id,
739                                              const char *device_type,
740                                              QDict *opts, Error **errp)
741 {
742     ERRP_GUARD();
743     const char *params = qdict_get_try_str(opts, "params");
744     const char *mode = qdict_get_try_str(opts, "mode");
745     const char *direct_io_safe = qdict_get_try_str(opts, "direct-io-safe");
746     const char *discard_enable = qdict_get_try_str(opts, "discard-enable");
747     char *driver = NULL;
748     char *filename = NULL;
749     XenBlockDrive *drive = NULL;
750     QDict *file_layer;
751     QDict *driver_layer;
752     struct stat st;
753     int rc;
754 
755     if (params) {
756         char **v = g_strsplit(params, ":", 2);
757 
758         if (v[1] == NULL) {
759             filename = g_strdup(v[0]);
760             driver = g_strdup("raw");
761         } else {
762             if (strcmp(v[0], "aio") == 0) {
763                 driver = g_strdup("raw");
764             } else if (strcmp(v[0], "vhd") == 0) {
765                 driver = g_strdup("vpc");
766             } else {
767                 driver = g_strdup(v[0]);
768             }
769             filename = g_strdup(v[1]);
770         }
771 
772         g_strfreev(v);
773     } else {
774         error_setg(errp, "no params");
775         goto done;
776     }
777 
778     assert(filename);
779     assert(driver);
780 
781     drive = g_new0(XenBlockDrive, 1);
782     drive->id = g_strdup(id);
783 
784     file_layer = qdict_new();
785     driver_layer = qdict_new();
786 
787     rc = stat(filename, &st);
788     if (rc) {
789         error_setg_errno(errp, errno, "Could not stat file '%s'", filename);
790         goto done;
791     }
792     if (S_ISBLK(st.st_mode)) {
793         qdict_put_str(file_layer, "driver", "host_device");
794     } else {
795         qdict_put_str(file_layer, "driver", "file");
796     }
797 
798     qdict_put_str(file_layer, "filename", filename);
799     g_free(filename);
800 
801     if (mode && *mode != 'w') {
802         qdict_put_bool(file_layer, "read-only", true);
803     }
804 
805     if (direct_io_safe) {
806         unsigned long value;
807 
808         if (!qemu_strtoul(direct_io_safe, NULL, 2, &value) && !!value) {
809             QDict *cache_qdict = qdict_new();
810 
811             qdict_put_bool(cache_qdict, "direct", true);
812             qdict_put(file_layer, "cache", cache_qdict);
813 
814             qdict_put_str(file_layer, "aio", "native");
815         }
816     }
817 
818     if (discard_enable) {
819         unsigned long value;
820 
821         if (!qemu_strtoul(discard_enable, NULL, 2, &value) && !!value) {
822             qdict_put_str(file_layer, "discard", "unmap");
823             qdict_put_str(driver_layer, "discard", "unmap");
824         }
825     }
826 
827     /*
828      * It is necessary to turn file locking off as an emulated device
829      * may have already opened the same image file.
830      */
831     qdict_put_str(file_layer, "locking", "off");
832 
833     qdict_put_str(driver_layer, "driver", driver);
834     g_free(driver);
835 
836     qdict_put(driver_layer, "file", file_layer);
837 
838     g_assert(!drive->node_name);
839     drive->node_name = xen_block_blockdev_add(drive->id, driver_layer,
840                                               errp);
841 
842     qobject_unref(driver_layer);
843 
844 done:
845     if (*errp) {
846         xen_block_drive_destroy(drive, NULL);
847         return NULL;
848     }
849 
850     return drive;
851 }
852 
853 static const char *xen_block_drive_get_node_name(XenBlockDrive *drive)
854 {
855     return drive->node_name ? drive->node_name : "";
856 }
857 
858 static void xen_block_iothread_destroy(XenBlockIOThread *iothread,
859                                        Error **errp)
860 {
861     qmp_object_del(iothread->id, errp);
862 
863     g_free(iothread->id);
864     g_free(iothread);
865 }
866 
867 static XenBlockIOThread *xen_block_iothread_create(const char *id,
868                                                    Error **errp)
869 {
870     ERRP_GUARD();
871     XenBlockIOThread *iothread = g_new(XenBlockIOThread, 1);
872     ObjectOptions *opts;
873 
874     iothread->id = g_strdup(id);
875 
876     opts = g_new(ObjectOptions, 1);
877     *opts = (ObjectOptions) {
878         .qom_type = OBJECT_TYPE_IOTHREAD,
879         .id = g_strdup(id),
880     };
881     qmp_object_add(opts, errp);
882     qapi_free_ObjectOptions(opts);
883 
884     if (*errp) {
885         g_free(iothread->id);
886         g_free(iothread);
887         return NULL;
888     }
889 
890     return iothread;
891 }
892 
893 static void xen_block_device_create(XenBackendInstance *backend,
894                                     QDict *opts, Error **errp)
895 {
896     ERRP_GUARD();
897     XenBus *xenbus = xen_backend_get_bus(backend);
898     const char *name = xen_backend_get_name(backend);
899     unsigned long number;
900     const char *vdev, *device_type;
901     XenBlockDrive *drive = NULL;
902     XenBlockIOThread *iothread = NULL;
903     XenDevice *xendev = NULL;
904     const char *type;
905     XenBlockDevice *blockdev;
906 
907     if (qemu_strtoul(name, NULL, 10, &number)) {
908         error_setg(errp, "failed to parse name '%s'", name);
909         goto fail;
910     }
911 
912     trace_xen_block_device_create(number);
913 
914     vdev = qdict_get_try_str(opts, "dev");
915     if (!vdev) {
916         error_setg(errp, "no dev parameter");
917         goto fail;
918     }
919 
920     device_type = qdict_get_try_str(opts, "device-type");
921     if (!device_type) {
922         error_setg(errp, "no device-type parameter");
923         goto fail;
924     }
925 
926     if (!strcmp(device_type, "disk")) {
927         type = TYPE_XEN_DISK_DEVICE;
928     } else if (!strcmp(device_type, "cdrom")) {
929         type = TYPE_XEN_CDROM_DEVICE;
930     } else {
931         error_setg(errp, "invalid device-type parameter '%s'", device_type);
932         goto fail;
933     }
934 
935     drive = xen_block_drive_create(vdev, device_type, opts, errp);
936     if (!drive) {
937         error_prepend(errp, "failed to create drive: ");
938         goto fail;
939     }
940 
941     iothread = xen_block_iothread_create(vdev, errp);
942     if (*errp) {
943         error_prepend(errp, "failed to create iothread: ");
944         goto fail;
945     }
946 
947     xendev = XEN_DEVICE(qdev_new(type));
948     blockdev = XEN_BLOCK_DEVICE(xendev);
949 
950     if (!object_property_set_str(OBJECT(xendev), "vdev", vdev,
951                                  errp)) {
952         error_prepend(errp, "failed to set 'vdev': ");
953         goto fail;
954     }
955 
956     if (!object_property_set_str(OBJECT(xendev), "drive",
957                                  xen_block_drive_get_node_name(drive),
958                                  errp)) {
959         error_prepend(errp, "failed to set 'drive': ");
960         goto fail;
961     }
962 
963     if (!object_property_set_str(OBJECT(xendev), "iothread", iothread->id,
964                                  errp)) {
965         error_prepend(errp, "failed to set 'iothread': ");
966         goto fail;
967     }
968 
969     blockdev->iothread = iothread;
970     blockdev->drive = drive;
971 
972     if (!qdev_realize_and_unref(DEVICE(xendev), BUS(xenbus), errp)) {
973         error_prepend(errp, "realization of device %s failed: ", type);
974         goto fail;
975     }
976 
977     xen_backend_set_device(backend, xendev);
978     return;
979 
980 fail:
981     if (xendev) {
982         object_unparent(OBJECT(xendev));
983     }
984 
985     if (iothread) {
986         xen_block_iothread_destroy(iothread, NULL);
987     }
988 
989     if (drive) {
990         xen_block_drive_destroy(drive, NULL);
991     }
992 }
993 
994 static void xen_block_device_destroy(XenBackendInstance *backend,
995                                      Error **errp)
996 {
997     ERRP_GUARD();
998     XenDevice *xendev = xen_backend_get_device(backend);
999     XenBlockDevice *blockdev = XEN_BLOCK_DEVICE(xendev);
1000     XenBlockVdev *vdev = &blockdev->props.vdev;
1001     XenBlockDrive *drive = blockdev->drive;
1002     XenBlockIOThread *iothread = blockdev->iothread;
1003 
1004     trace_xen_block_device_destroy(vdev->number);
1005 
1006     object_unparent(OBJECT(xendev));
1007 
1008     /*
1009      * Drain all pending RCU callbacks as object_unparent() frees `xendev'
1010      * in a RCU callback.
1011      * And due to the property "drive" still existing in `xendev', we
1012      * can't destroy the XenBlockDrive associated with `xendev' with
1013      * xen_block_drive_destroy() below.
1014      */
1015     drain_call_rcu();
1016 
1017     if (iothread) {
1018         xen_block_iothread_destroy(iothread, errp);
1019         if (*errp) {
1020             error_prepend(errp, "failed to destroy iothread: ");
1021             return;
1022         }
1023     }
1024 
1025     if (drive) {
1026         xen_block_drive_destroy(drive, errp);
1027         if (*errp) {
1028             error_prepend(errp, "failed to destroy drive: ");
1029             return;
1030         }
1031     }
1032 }
1033 
1034 static const XenBackendInfo xen_block_backend_info = {
1035     .type = "qdisk",
1036     .create = xen_block_device_create,
1037     .destroy = xen_block_device_destroy,
1038 };
1039 
1040 static void xen_block_register_backend(void)
1041 {
1042     xen_backend_register(&xen_block_backend_info);
1043 }
1044 
1045 xen_backend_init(xen_block_register_backend);
1046