xref: /qemu/blockdev.c (revision de4c2dcf7fedf5fa727113e1dec0d0e5dd0462a0)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "sysemu/blockdev.h"
34 #include "hw/block/block.h"
35 #include "block/blockjob.h"
36 #include "monitor/monitor.h"
37 #include "qapi/qmp/qerror.h"
38 #include "qemu/option.h"
39 #include "qemu/config-file.h"
40 #include "qapi/qmp/types.h"
41 #include "qapi-visit.h"
42 #include "qapi/qmp-output-visitor.h"
43 #include "sysemu/sysemu.h"
44 #include "block/block_int.h"
45 #include "qmp-commands.h"
46 #include "trace.h"
47 #include "sysemu/arch_init.h"
48 
49 static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives);
50 extern QemuOptsList qemu_common_drive_opts;
51 
52 static const char *const if_name[IF_COUNT] = {
53     [IF_NONE] = "none",
54     [IF_IDE] = "ide",
55     [IF_SCSI] = "scsi",
56     [IF_FLOPPY] = "floppy",
57     [IF_PFLASH] = "pflash",
58     [IF_MTD] = "mtd",
59     [IF_SD] = "sd",
60     [IF_VIRTIO] = "virtio",
61     [IF_XEN] = "xen",
62 };
63 
64 static const int if_max_devs[IF_COUNT] = {
65     /*
66      * Do not change these numbers!  They govern how drive option
67      * index maps to unit and bus.  That mapping is ABI.
68      *
69      * All controllers used to imlement if=T drives need to support
70      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
71      * Otherwise, some index values map to "impossible" bus, unit
72      * values.
73      *
74      * For instance, if you change [IF_SCSI] to 255, -drive
75      * if=scsi,index=12 no longer means bus=1,unit=5, but
76      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
77      * the drive can't be set up.  Regression.
78      */
79     [IF_IDE] = 2,
80     [IF_SCSI] = 7,
81 };
82 
83 /*
84  * We automatically delete the drive when a device using it gets
85  * unplugged.  Questionable feature, but we can't just drop it.
86  * Device models call blockdev_mark_auto_del() to schedule the
87  * automatic deletion, and generic qdev code calls blockdev_auto_del()
88  * when deletion is actually safe.
89  */
90 void blockdev_mark_auto_del(BlockDriverState *bs)
91 {
92     DriveInfo *dinfo = drive_get_by_blockdev(bs);
93 
94     if (dinfo && !dinfo->enable_auto_del) {
95         return;
96     }
97 
98     if (bs->job) {
99         block_job_cancel(bs->job);
100     }
101     if (dinfo) {
102         dinfo->auto_del = 1;
103     }
104 }
105 
106 void blockdev_auto_del(BlockDriverState *bs)
107 {
108     DriveInfo *dinfo = drive_get_by_blockdev(bs);
109 
110     if (dinfo && dinfo->auto_del) {
111         drive_put_ref(dinfo);
112     }
113 }
114 
115 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
116 {
117     int max_devs = if_max_devs[type];
118     return max_devs ? index / max_devs : 0;
119 }
120 
121 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
122 {
123     int max_devs = if_max_devs[type];
124     return max_devs ? index % max_devs : index;
125 }
126 
127 QemuOpts *drive_def(const char *optstr)
128 {
129     return qemu_opts_parse(qemu_find_opts("drive"), optstr, 0);
130 }
131 
132 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
133                     const char *optstr)
134 {
135     QemuOpts *opts;
136     char buf[32];
137 
138     opts = drive_def(optstr);
139     if (!opts) {
140         return NULL;
141     }
142     if (type != IF_DEFAULT) {
143         qemu_opt_set(opts, "if", if_name[type]);
144     }
145     if (index >= 0) {
146         snprintf(buf, sizeof(buf), "%d", index);
147         qemu_opt_set(opts, "index", buf);
148     }
149     if (file)
150         qemu_opt_set(opts, "file", file);
151     return opts;
152 }
153 
154 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
155 {
156     DriveInfo *dinfo;
157 
158     /* seek interface, bus and unit */
159 
160     QTAILQ_FOREACH(dinfo, &drives, next) {
161         if (dinfo->type == type &&
162 	    dinfo->bus == bus &&
163 	    dinfo->unit == unit)
164             return dinfo;
165     }
166 
167     return NULL;
168 }
169 
170 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
171 {
172     return drive_get(type,
173                      drive_index_to_bus_id(type, index),
174                      drive_index_to_unit_id(type, index));
175 }
176 
177 int drive_get_max_bus(BlockInterfaceType type)
178 {
179     int max_bus;
180     DriveInfo *dinfo;
181 
182     max_bus = -1;
183     QTAILQ_FOREACH(dinfo, &drives, next) {
184         if(dinfo->type == type &&
185            dinfo->bus > max_bus)
186             max_bus = dinfo->bus;
187     }
188     return max_bus;
189 }
190 
191 /* Get a block device.  This should only be used for single-drive devices
192    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
193    appropriate bus.  */
194 DriveInfo *drive_get_next(BlockInterfaceType type)
195 {
196     static int next_block_unit[IF_COUNT];
197 
198     return drive_get(type, 0, next_block_unit[type]++);
199 }
200 
201 DriveInfo *drive_get_by_blockdev(BlockDriverState *bs)
202 {
203     DriveInfo *dinfo;
204 
205     QTAILQ_FOREACH(dinfo, &drives, next) {
206         if (dinfo->bdrv == bs) {
207             return dinfo;
208         }
209     }
210     return NULL;
211 }
212 
213 static void bdrv_format_print(void *opaque, const char *name)
214 {
215     error_printf(" %s", name);
216 }
217 
218 static void drive_uninit(DriveInfo *dinfo)
219 {
220     if (dinfo->opts) {
221         qemu_opts_del(dinfo->opts);
222     }
223 
224     bdrv_unref(dinfo->bdrv);
225     g_free(dinfo->id);
226     QTAILQ_REMOVE(&drives, dinfo, next);
227     g_free(dinfo->serial);
228     g_free(dinfo);
229 }
230 
231 void drive_put_ref(DriveInfo *dinfo)
232 {
233     assert(dinfo->refcount);
234     if (--dinfo->refcount == 0) {
235         drive_uninit(dinfo);
236     }
237 }
238 
239 void drive_get_ref(DriveInfo *dinfo)
240 {
241     dinfo->refcount++;
242 }
243 
244 typedef struct {
245     QEMUBH *bh;
246     BlockDriverState *bs;
247 } BDRVPutRefBH;
248 
249 static void bdrv_put_ref_bh(void *opaque)
250 {
251     BDRVPutRefBH *s = opaque;
252 
253     bdrv_unref(s->bs);
254     qemu_bh_delete(s->bh);
255     g_free(s);
256 }
257 
258 /*
259  * Release a BDS reference in a BH
260  *
261  * It is not safe to use bdrv_unref() from a callback function when the callers
262  * still need the BlockDriverState.  In such cases we schedule a BH to release
263  * the reference.
264  */
265 static void bdrv_put_ref_bh_schedule(BlockDriverState *bs)
266 {
267     BDRVPutRefBH *s;
268 
269     s = g_new(BDRVPutRefBH, 1);
270     s->bh = qemu_bh_new(bdrv_put_ref_bh, s);
271     s->bs = bs;
272     qemu_bh_schedule(s->bh);
273 }
274 
275 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
276 {
277     if (!strcmp(buf, "ignore")) {
278         return BLOCKDEV_ON_ERROR_IGNORE;
279     } else if (!is_read && !strcmp(buf, "enospc")) {
280         return BLOCKDEV_ON_ERROR_ENOSPC;
281     } else if (!strcmp(buf, "stop")) {
282         return BLOCKDEV_ON_ERROR_STOP;
283     } else if (!strcmp(buf, "report")) {
284         return BLOCKDEV_ON_ERROR_REPORT;
285     } else {
286         error_setg(errp, "'%s' invalid %s error action",
287                    buf, is_read ? "read" : "write");
288         return -1;
289     }
290 }
291 
292 static bool check_throttle_config(ThrottleConfig *cfg, Error **errp)
293 {
294     if (throttle_conflicting(cfg)) {
295         error_setg(errp, "bps/iops/max total values and read/write values"
296                          " cannot be used at the same time");
297         return false;
298     }
299 
300     if (!throttle_is_valid(cfg)) {
301         error_setg(errp, "bps/iops/maxs values must be 0 or greater");
302         return false;
303     }
304 
305     return true;
306 }
307 
308 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
309 
310 /* Takes the ownership of bs_opts */
311 static DriveInfo *blockdev_init(QDict *bs_opts,
312                                 BlockInterfaceType type,
313                                 Error **errp)
314 {
315     const char *buf;
316     const char *file = NULL;
317     const char *serial;
318     int ro = 0;
319     int bdrv_flags = 0;
320     int on_read_error, on_write_error;
321     DriveInfo *dinfo;
322     ThrottleConfig cfg;
323     int snapshot = 0;
324     bool copy_on_read;
325     int ret;
326     Error *error = NULL;
327     QemuOpts *opts;
328     const char *id;
329     bool has_driver_specific_opts;
330     BlockDriver *drv = NULL;
331 
332     /* Check common options by copying from bs_opts to opts, all other options
333      * stay in bs_opts for processing by bdrv_open(). */
334     id = qdict_get_try_str(bs_opts, "id");
335     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
336     if (error_is_set(&error)) {
337         error_propagate(errp, error);
338         return NULL;
339     }
340 
341     qemu_opts_absorb_qdict(opts, bs_opts, &error);
342     if (error_is_set(&error)) {
343         error_propagate(errp, error);
344         return NULL;
345     }
346 
347     if (id) {
348         qdict_del(bs_opts, "id");
349     }
350 
351     has_driver_specific_opts = !!qdict_size(bs_opts);
352 
353     /* extract parameters */
354     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
355     ro = qemu_opt_get_bool(opts, "read-only", 0);
356     copy_on_read = qemu_opt_get_bool(opts, "copy-on-read", false);
357 
358     file = qemu_opt_get(opts, "file");
359     serial = qemu_opt_get(opts, "serial");
360 
361     if ((buf = qemu_opt_get(opts, "discard")) != NULL) {
362         if (bdrv_parse_discard_flags(buf, &bdrv_flags) != 0) {
363             error_setg(errp, "invalid discard option");
364             return NULL;
365         }
366     }
367 
368     if (qemu_opt_get_bool(opts, "cache.writeback", true)) {
369         bdrv_flags |= BDRV_O_CACHE_WB;
370     }
371     if (qemu_opt_get_bool(opts, "cache.direct", false)) {
372         bdrv_flags |= BDRV_O_NOCACHE;
373     }
374     if (qemu_opt_get_bool(opts, "cache.no-flush", false)) {
375         bdrv_flags |= BDRV_O_NO_FLUSH;
376     }
377 
378 #ifdef CONFIG_LINUX_AIO
379     if ((buf = qemu_opt_get(opts, "aio")) != NULL) {
380         if (!strcmp(buf, "native")) {
381             bdrv_flags |= BDRV_O_NATIVE_AIO;
382         } else if (!strcmp(buf, "threads")) {
383             /* this is the default */
384         } else {
385            error_setg(errp, "invalid aio option");
386            return NULL;
387         }
388     }
389 #endif
390 
391     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
392         if (is_help_option(buf)) {
393             error_printf("Supported formats:");
394             bdrv_iterate_format(bdrv_format_print, NULL);
395             error_printf("\n");
396             return NULL;
397         }
398 
399         drv = bdrv_find_format(buf);
400         if (!drv) {
401             error_setg(errp, "'%s' invalid format", buf);
402             return NULL;
403         }
404     }
405 
406     /* disk I/O throttling */
407     memset(&cfg, 0, sizeof(cfg));
408     cfg.buckets[THROTTLE_BPS_TOTAL].avg =
409         qemu_opt_get_number(opts, "throttling.bps-total", 0);
410     cfg.buckets[THROTTLE_BPS_READ].avg  =
411         qemu_opt_get_number(opts, "throttling.bps-read", 0);
412     cfg.buckets[THROTTLE_BPS_WRITE].avg =
413         qemu_opt_get_number(opts, "throttling.bps-write", 0);
414     cfg.buckets[THROTTLE_OPS_TOTAL].avg =
415         qemu_opt_get_number(opts, "throttling.iops-total", 0);
416     cfg.buckets[THROTTLE_OPS_READ].avg =
417         qemu_opt_get_number(opts, "throttling.iops-read", 0);
418     cfg.buckets[THROTTLE_OPS_WRITE].avg =
419         qemu_opt_get_number(opts, "throttling.iops-write", 0);
420 
421     cfg.buckets[THROTTLE_BPS_TOTAL].max =
422         qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
423     cfg.buckets[THROTTLE_BPS_READ].max  =
424         qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
425     cfg.buckets[THROTTLE_BPS_WRITE].max =
426         qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
427     cfg.buckets[THROTTLE_OPS_TOTAL].max =
428         qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
429     cfg.buckets[THROTTLE_OPS_READ].max =
430         qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
431     cfg.buckets[THROTTLE_OPS_WRITE].max =
432         qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
433 
434     cfg.op_size = qemu_opt_get_number(opts, "throttling.iops-size", 0);
435 
436     if (!check_throttle_config(&cfg, &error)) {
437         error_propagate(errp, error);
438         return NULL;
439     }
440 
441     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
442     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
443         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO && type != IF_NONE) {
444             error_setg(errp, "werror is not supported by this bus type");
445             return NULL;
446         }
447 
448         on_write_error = parse_block_error_action(buf, 0, &error);
449         if (error_is_set(&error)) {
450             error_propagate(errp, error);
451             return NULL;
452         }
453     }
454 
455     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
456     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
457         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI && type != IF_NONE) {
458             error_report("rerror is not supported by this bus type");
459             return NULL;
460         }
461 
462         on_read_error = parse_block_error_action(buf, 1, &error);
463         if (error_is_set(&error)) {
464             error_propagate(errp, error);
465             return NULL;
466         }
467     }
468 
469     /* init */
470     dinfo = g_malloc0(sizeof(*dinfo));
471     dinfo->id = g_strdup(qemu_opts_id(opts));
472     dinfo->bdrv = bdrv_new(dinfo->id);
473     dinfo->bdrv->open_flags = snapshot ? BDRV_O_SNAPSHOT : 0;
474     dinfo->bdrv->read_only = ro;
475     dinfo->type = type;
476     dinfo->refcount = 1;
477     if (serial != NULL) {
478         dinfo->serial = g_strdup(serial);
479     }
480     QTAILQ_INSERT_TAIL(&drives, dinfo, next);
481 
482     bdrv_set_on_error(dinfo->bdrv, on_read_error, on_write_error);
483 
484     /* disk I/O throttling */
485     if (throttle_enabled(&cfg)) {
486         bdrv_io_limits_enable(dinfo->bdrv);
487         bdrv_set_io_limits(dinfo->bdrv, &cfg);
488     }
489 
490     if (!file || !*file) {
491         if (has_driver_specific_opts) {
492             file = NULL;
493         } else {
494             return dinfo;
495         }
496     }
497     if (snapshot) {
498         /* always use cache=unsafe with snapshot */
499         bdrv_flags &= ~BDRV_O_CACHE_MASK;
500         bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
501     }
502 
503     if (copy_on_read) {
504         bdrv_flags |= BDRV_O_COPY_ON_READ;
505     }
506 
507     if (runstate_check(RUN_STATE_INMIGRATE)) {
508         bdrv_flags |= BDRV_O_INCOMING;
509     }
510 
511     bdrv_flags |= ro ? 0 : BDRV_O_RDWR;
512 
513     QINCREF(bs_opts);
514     ret = bdrv_open(dinfo->bdrv, file, bs_opts, bdrv_flags, drv, &error);
515 
516     if (ret < 0) {
517         error_setg(errp, "could not open disk image %s: %s",
518                    file ?: dinfo->id, error_get_pretty(error));
519         error_free(error);
520         goto err;
521     }
522 
523     if (bdrv_key_required(dinfo->bdrv))
524         autostart = 0;
525 
526     QDECREF(bs_opts);
527     qemu_opts_del(opts);
528 
529     return dinfo;
530 
531 err:
532     qemu_opts_del(opts);
533     QDECREF(bs_opts);
534     bdrv_unref(dinfo->bdrv);
535     g_free(dinfo->id);
536     QTAILQ_REMOVE(&drives, dinfo, next);
537     g_free(dinfo);
538     return NULL;
539 }
540 
541 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to)
542 {
543     const char *value;
544 
545     value = qemu_opt_get(opts, from);
546     if (value) {
547         qemu_opt_set(opts, to, value);
548         qemu_opt_unset(opts, from);
549     }
550 }
551 
552 QemuOptsList qemu_legacy_drive_opts = {
553     .name = "drive",
554     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
555     .desc = {
556         {
557             .name = "bus",
558             .type = QEMU_OPT_NUMBER,
559             .help = "bus number",
560         },{
561             .name = "unit",
562             .type = QEMU_OPT_NUMBER,
563             .help = "unit number (i.e. lun for scsi)",
564         },{
565             .name = "index",
566             .type = QEMU_OPT_NUMBER,
567             .help = "index number",
568         },{
569             .name = "media",
570             .type = QEMU_OPT_STRING,
571             .help = "media type (disk, cdrom)",
572         },{
573             .name = "if",
574             .type = QEMU_OPT_STRING,
575             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
576         },{
577             .name = "cyls",
578             .type = QEMU_OPT_NUMBER,
579             .help = "number of cylinders (ide disk geometry)",
580         },{
581             .name = "heads",
582             .type = QEMU_OPT_NUMBER,
583             .help = "number of heads (ide disk geometry)",
584         },{
585             .name = "secs",
586             .type = QEMU_OPT_NUMBER,
587             .help = "number of sectors (ide disk geometry)",
588         },{
589             .name = "trans",
590             .type = QEMU_OPT_STRING,
591             .help = "chs translation (auto, lba, none)",
592         },{
593             .name = "boot",
594             .type = QEMU_OPT_BOOL,
595             .help = "(deprecated, ignored)",
596         },{
597             .name = "addr",
598             .type = QEMU_OPT_STRING,
599             .help = "pci address (virtio only)",
600         },
601 
602         /* Options that are passed on, but have special semantics with -drive */
603         {
604             .name = "read-only",
605             .type = QEMU_OPT_BOOL,
606             .help = "open drive file as read-only",
607         },{
608             .name = "copy-on-read",
609             .type = QEMU_OPT_BOOL,
610             .help = "copy read data from backing file into image file",
611         },
612 
613         { /* end of list */ }
614     },
615 };
616 
617 DriveInfo *drive_init(QemuOpts *all_opts, BlockInterfaceType block_default_type)
618 {
619     const char *value;
620     DriveInfo *dinfo = NULL;
621     QDict *bs_opts;
622     QemuOpts *legacy_opts;
623     DriveMediaType media = MEDIA_DISK;
624     BlockInterfaceType type;
625     int cyls, heads, secs, translation;
626     int max_devs, bus_id, unit_id, index;
627     const char *devaddr;
628     bool read_only = false;
629     bool copy_on_read;
630     Error *local_err = NULL;
631 
632     /* Change legacy command line options into QMP ones */
633     qemu_opt_rename(all_opts, "iops", "throttling.iops-total");
634     qemu_opt_rename(all_opts, "iops_rd", "throttling.iops-read");
635     qemu_opt_rename(all_opts, "iops_wr", "throttling.iops-write");
636 
637     qemu_opt_rename(all_opts, "bps", "throttling.bps-total");
638     qemu_opt_rename(all_opts, "bps_rd", "throttling.bps-read");
639     qemu_opt_rename(all_opts, "bps_wr", "throttling.bps-write");
640 
641     qemu_opt_rename(all_opts, "iops_max", "throttling.iops-total-max");
642     qemu_opt_rename(all_opts, "iops_rd_max", "throttling.iops-read-max");
643     qemu_opt_rename(all_opts, "iops_wr_max", "throttling.iops-write-max");
644 
645     qemu_opt_rename(all_opts, "bps_max", "throttling.bps-total-max");
646     qemu_opt_rename(all_opts, "bps_rd_max", "throttling.bps-read-max");
647     qemu_opt_rename(all_opts, "bps_wr_max", "throttling.bps-write-max");
648 
649     qemu_opt_rename(all_opts,
650                     "iops_size", "throttling.iops-size");
651 
652     qemu_opt_rename(all_opts, "readonly", "read-only");
653 
654     value = qemu_opt_get(all_opts, "cache");
655     if (value) {
656         int flags = 0;
657 
658         if (bdrv_parse_cache_flags(value, &flags) != 0) {
659             error_report("invalid cache option");
660             return NULL;
661         }
662 
663         /* Specific options take precedence */
664         if (!qemu_opt_get(all_opts, "cache.writeback")) {
665             qemu_opt_set_bool(all_opts, "cache.writeback",
666                               !!(flags & BDRV_O_CACHE_WB));
667         }
668         if (!qemu_opt_get(all_opts, "cache.direct")) {
669             qemu_opt_set_bool(all_opts, "cache.direct",
670                               !!(flags & BDRV_O_NOCACHE));
671         }
672         if (!qemu_opt_get(all_opts, "cache.no-flush")) {
673             qemu_opt_set_bool(all_opts, "cache.no-flush",
674                               !!(flags & BDRV_O_NO_FLUSH));
675         }
676         qemu_opt_unset(all_opts, "cache");
677     }
678 
679     /* Get a QDict for processing the options */
680     bs_opts = qdict_new();
681     qemu_opts_to_qdict(all_opts, bs_opts);
682 
683     legacy_opts = qemu_opts_create_nofail(&qemu_legacy_drive_opts);
684     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
685     if (error_is_set(&local_err)) {
686         qerror_report_err(local_err);
687         error_free(local_err);
688         goto fail;
689     }
690 
691     /* Deprecated option boot=[on|off] */
692     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
693         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
694                 "ignored. Future versions will reject this parameter. Please "
695                 "update your scripts.\n");
696     }
697 
698     /* Media type */
699     value = qemu_opt_get(legacy_opts, "media");
700     if (value) {
701         if (!strcmp(value, "disk")) {
702             media = MEDIA_DISK;
703         } else if (!strcmp(value, "cdrom")) {
704             media = MEDIA_CDROM;
705             read_only = true;
706         } else {
707             error_report("'%s' invalid media", value);
708             goto fail;
709         }
710     }
711 
712     /* copy-on-read is disabled with a warning for read-only devices */
713     read_only |= qemu_opt_get_bool(legacy_opts, "read-only", false);
714     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
715 
716     if (read_only && copy_on_read) {
717         error_report("warning: disabling copy-on-read on read-only drive");
718         copy_on_read = false;
719     }
720 
721     qdict_put(bs_opts, "read-only",
722               qstring_from_str(read_only ? "on" : "off"));
723     qdict_put(bs_opts, "copy-on-read",
724               qstring_from_str(copy_on_read ? "on" :"off"));
725 
726     /* Controller type */
727     value = qemu_opt_get(legacy_opts, "if");
728     if (value) {
729         for (type = 0;
730              type < IF_COUNT && strcmp(value, if_name[type]);
731              type++) {
732         }
733         if (type == IF_COUNT) {
734             error_report("unsupported bus type '%s'", value);
735             goto fail;
736         }
737     } else {
738         type = block_default_type;
739     }
740 
741     /* Geometry */
742     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
743     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
744     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
745 
746     if (cyls || heads || secs) {
747         if (cyls < 1) {
748             error_report("invalid physical cyls number");
749             goto fail;
750         }
751         if (heads < 1) {
752             error_report("invalid physical heads number");
753             goto fail;
754         }
755         if (secs < 1) {
756             error_report("invalid physical secs number");
757             goto fail;
758         }
759     }
760 
761     translation = BIOS_ATA_TRANSLATION_AUTO;
762     value = qemu_opt_get(legacy_opts, "trans");
763     if (value != NULL) {
764         if (!cyls) {
765             error_report("'%s' trans must be used with cyls, heads and secs",
766                          value);
767             goto fail;
768         }
769         if (!strcmp(value, "none")) {
770             translation = BIOS_ATA_TRANSLATION_NONE;
771         } else if (!strcmp(value, "lba")) {
772             translation = BIOS_ATA_TRANSLATION_LBA;
773         } else if (!strcmp(value, "auto")) {
774             translation = BIOS_ATA_TRANSLATION_AUTO;
775         } else {
776             error_report("'%s' invalid translation type", value);
777             goto fail;
778         }
779     }
780 
781     if (media == MEDIA_CDROM) {
782         if (cyls || secs || heads) {
783             error_report("CHS can't be set with media=cdrom");
784             goto fail;
785         }
786     }
787 
788     /* Device address specified by bus/unit or index.
789      * If none was specified, try to find the first free one. */
790     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
791     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
792     index   = qemu_opt_get_number(legacy_opts, "index", -1);
793 
794     max_devs = if_max_devs[type];
795 
796     if (index != -1) {
797         if (bus_id != 0 || unit_id != -1) {
798             error_report("index cannot be used with bus and unit");
799             goto fail;
800         }
801         bus_id = drive_index_to_bus_id(type, index);
802         unit_id = drive_index_to_unit_id(type, index);
803     }
804 
805     if (unit_id == -1) {
806        unit_id = 0;
807        while (drive_get(type, bus_id, unit_id) != NULL) {
808            unit_id++;
809            if (max_devs && unit_id >= max_devs) {
810                unit_id -= max_devs;
811                bus_id++;
812            }
813        }
814     }
815 
816     if (max_devs && unit_id >= max_devs) {
817         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
818         goto fail;
819     }
820 
821     if (drive_get(type, bus_id, unit_id) != NULL) {
822         error_report("drive with bus=%d, unit=%d (index=%d) exists",
823                      bus_id, unit_id, index);
824         goto fail;
825     }
826 
827     /* no id supplied -> create one */
828     if (qemu_opts_id(all_opts) == NULL) {
829         char *new_id;
830         const char *mediastr = "";
831         if (type == IF_IDE || type == IF_SCSI) {
832             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
833         }
834         if (max_devs) {
835             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
836                                      mediastr, unit_id);
837         } else {
838             new_id = g_strdup_printf("%s%s%i", if_name[type],
839                                      mediastr, unit_id);
840         }
841         qdict_put(bs_opts, "id", qstring_from_str(new_id));
842         g_free(new_id);
843     }
844 
845     /* Add virtio block device */
846     devaddr = qemu_opt_get(legacy_opts, "addr");
847     if (devaddr && type != IF_VIRTIO) {
848         error_report("addr is not supported by this bus type");
849         goto fail;
850     }
851 
852     if (type == IF_VIRTIO) {
853         QemuOpts *devopts;
854         devopts = qemu_opts_create_nofail(qemu_find_opts("device"));
855         if (arch_type == QEMU_ARCH_S390X) {
856             qemu_opt_set(devopts, "driver", "virtio-blk-s390");
857         } else {
858             qemu_opt_set(devopts, "driver", "virtio-blk-pci");
859         }
860         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"));
861         if (devaddr) {
862             qemu_opt_set(devopts, "addr", devaddr);
863         }
864     }
865 
866     /* Actual block device init: Functionality shared with blockdev-add */
867     dinfo = blockdev_init(bs_opts, type, &local_err);
868     if (dinfo == NULL) {
869         if (error_is_set(&local_err)) {
870             qerror_report_err(local_err);
871             error_free(local_err);
872         }
873         goto fail;
874     } else {
875         assert(!error_is_set(&local_err));
876     }
877 
878     /* Set legacy DriveInfo fields */
879     dinfo->enable_auto_del = true;
880     dinfo->opts = all_opts;
881 
882     dinfo->cyls = cyls;
883     dinfo->heads = heads;
884     dinfo->secs = secs;
885     dinfo->trans = translation;
886 
887     dinfo->bus = bus_id;
888     dinfo->unit = unit_id;
889     dinfo->devaddr = devaddr;
890 
891     switch(type) {
892     case IF_IDE:
893     case IF_SCSI:
894     case IF_XEN:
895     case IF_NONE:
896         dinfo->media_cd = media == MEDIA_CDROM;
897         break;
898     default:
899         break;
900     }
901 
902 fail:
903     qemu_opts_del(legacy_opts);
904     return dinfo;
905 }
906 
907 void do_commit(Monitor *mon, const QDict *qdict)
908 {
909     const char *device = qdict_get_str(qdict, "device");
910     BlockDriverState *bs;
911     int ret;
912 
913     if (!strcmp(device, "all")) {
914         ret = bdrv_commit_all();
915     } else {
916         bs = bdrv_find(device);
917         if (!bs) {
918             monitor_printf(mon, "Device '%s' not found\n", device);
919             return;
920         }
921         ret = bdrv_commit(bs);
922     }
923     if (ret < 0) {
924         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
925                        strerror(-ret));
926     }
927 }
928 
929 static void blockdev_do_action(int kind, void *data, Error **errp)
930 {
931     TransactionAction action;
932     TransactionActionList list;
933 
934     action.kind = kind;
935     action.data = data;
936     list.value = &action;
937     list.next = NULL;
938     qmp_transaction(&list, errp);
939 }
940 
941 void qmp_blockdev_snapshot_sync(const char *device, const char *snapshot_file,
942                                 bool has_format, const char *format,
943                                 bool has_mode, enum NewImageMode mode,
944                                 Error **errp)
945 {
946     BlockdevSnapshot snapshot = {
947         .device = (char *) device,
948         .snapshot_file = (char *) snapshot_file,
949         .has_format = has_format,
950         .format = (char *) format,
951         .has_mode = has_mode,
952         .mode = mode,
953     };
954     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
955                        &snapshot, errp);
956 }
957 
958 void qmp_blockdev_snapshot_internal_sync(const char *device,
959                                          const char *name,
960                                          Error **errp)
961 {
962     BlockdevSnapshotInternal snapshot = {
963         .device = (char *) device,
964         .name = (char *) name
965     };
966 
967     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
968                        &snapshot, errp);
969 }
970 
971 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
972                                                          bool has_id,
973                                                          const char *id,
974                                                          bool has_name,
975                                                          const char *name,
976                                                          Error **errp)
977 {
978     BlockDriverState *bs = bdrv_find(device);
979     QEMUSnapshotInfo sn;
980     Error *local_err = NULL;
981     SnapshotInfo *info = NULL;
982     int ret;
983 
984     if (!bs) {
985         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
986         return NULL;
987     }
988 
989     if (!has_id) {
990         id = NULL;
991     }
992 
993     if (!has_name) {
994         name = NULL;
995     }
996 
997     if (!id && !name) {
998         error_setg(errp, "Name or id must be provided");
999         return NULL;
1000     }
1001 
1002     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1003     if (error_is_set(&local_err)) {
1004         error_propagate(errp, local_err);
1005         return NULL;
1006     }
1007     if (!ret) {
1008         error_setg(errp,
1009                    "Snapshot with id '%s' and name '%s' does not exist on "
1010                    "device '%s'",
1011                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1012         return NULL;
1013     }
1014 
1015     bdrv_snapshot_delete(bs, id, name, &local_err);
1016     if (error_is_set(&local_err)) {
1017         error_propagate(errp, local_err);
1018         return NULL;
1019     }
1020 
1021     info = g_malloc0(sizeof(SnapshotInfo));
1022     info->id = g_strdup(sn.id_str);
1023     info->name = g_strdup(sn.name);
1024     info->date_nsec = sn.date_nsec;
1025     info->date_sec = sn.date_sec;
1026     info->vm_state_size = sn.vm_state_size;
1027     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1028     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1029 
1030     return info;
1031 }
1032 
1033 /* New and old BlockDriverState structs for group snapshots */
1034 
1035 typedef struct BlkTransactionState BlkTransactionState;
1036 
1037 /* Only prepare() may fail. In a single transaction, only one of commit() or
1038    abort() will be called, clean() will always be called if it present. */
1039 typedef struct BdrvActionOps {
1040     /* Size of state struct, in bytes. */
1041     size_t instance_size;
1042     /* Prepare the work, must NOT be NULL. */
1043     void (*prepare)(BlkTransactionState *common, Error **errp);
1044     /* Commit the changes, can be NULL. */
1045     void (*commit)(BlkTransactionState *common);
1046     /* Abort the changes on fail, can be NULL. */
1047     void (*abort)(BlkTransactionState *common);
1048     /* Clean up resource in the end, can be NULL. */
1049     void (*clean)(BlkTransactionState *common);
1050 } BdrvActionOps;
1051 
1052 /*
1053  * This structure must be arranged as first member in child type, assuming
1054  * that compiler will also arrange it to the same address with parent instance.
1055  * Later it will be used in free().
1056  */
1057 struct BlkTransactionState {
1058     TransactionAction *action;
1059     const BdrvActionOps *ops;
1060     QSIMPLEQ_ENTRY(BlkTransactionState) entry;
1061 };
1062 
1063 /* internal snapshot private data */
1064 typedef struct InternalSnapshotState {
1065     BlkTransactionState common;
1066     BlockDriverState *bs;
1067     QEMUSnapshotInfo sn;
1068 } InternalSnapshotState;
1069 
1070 static void internal_snapshot_prepare(BlkTransactionState *common,
1071                                       Error **errp)
1072 {
1073     const char *device;
1074     const char *name;
1075     BlockDriverState *bs;
1076     QEMUSnapshotInfo old_sn, *sn;
1077     bool ret;
1078     qemu_timeval tv;
1079     BlockdevSnapshotInternal *internal;
1080     InternalSnapshotState *state;
1081     int ret1;
1082 
1083     g_assert(common->action->kind ==
1084              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1085     internal = common->action->blockdev_snapshot_internal_sync;
1086     state = DO_UPCAST(InternalSnapshotState, common, common);
1087 
1088     /* 1. parse input */
1089     device = internal->device;
1090     name = internal->name;
1091 
1092     /* 2. check for validation */
1093     bs = bdrv_find(device);
1094     if (!bs) {
1095         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1096         return;
1097     }
1098 
1099     if (!bdrv_is_inserted(bs)) {
1100         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1101         return;
1102     }
1103 
1104     if (bdrv_is_read_only(bs)) {
1105         error_set(errp, QERR_DEVICE_IS_READ_ONLY, device);
1106         return;
1107     }
1108 
1109     if (!bdrv_can_snapshot(bs)) {
1110         error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1111                   bs->drv->format_name, device, "internal snapshot");
1112         return;
1113     }
1114 
1115     if (!strlen(name)) {
1116         error_setg(errp, "Name is empty");
1117         return;
1118     }
1119 
1120     /* check whether a snapshot with name exist */
1121     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn, errp);
1122     if (error_is_set(errp)) {
1123         return;
1124     } else if (ret) {
1125         error_setg(errp,
1126                    "Snapshot with name '%s' already exists on device '%s'",
1127                    name, device);
1128         return;
1129     }
1130 
1131     /* 3. take the snapshot */
1132     sn = &state->sn;
1133     pstrcpy(sn->name, sizeof(sn->name), name);
1134     qemu_gettimeofday(&tv);
1135     sn->date_sec = tv.tv_sec;
1136     sn->date_nsec = tv.tv_usec * 1000;
1137     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1138 
1139     ret1 = bdrv_snapshot_create(bs, sn);
1140     if (ret1 < 0) {
1141         error_setg_errno(errp, -ret1,
1142                          "Failed to create snapshot '%s' on device '%s'",
1143                          name, device);
1144         return;
1145     }
1146 
1147     /* 4. succeed, mark a snapshot is created */
1148     state->bs = bs;
1149 }
1150 
1151 static void internal_snapshot_abort(BlkTransactionState *common)
1152 {
1153     InternalSnapshotState *state =
1154                              DO_UPCAST(InternalSnapshotState, common, common);
1155     BlockDriverState *bs = state->bs;
1156     QEMUSnapshotInfo *sn = &state->sn;
1157     Error *local_error = NULL;
1158 
1159     if (!bs) {
1160         return;
1161     }
1162 
1163     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1164         error_report("Failed to delete snapshot with id '%s' and name '%s' on "
1165                      "device '%s' in abort: %s",
1166                      sn->id_str,
1167                      sn->name,
1168                      bdrv_get_device_name(bs),
1169                      error_get_pretty(local_error));
1170         error_free(local_error);
1171     }
1172 }
1173 
1174 /* external snapshot private data */
1175 typedef struct ExternalSnapshotState {
1176     BlkTransactionState common;
1177     BlockDriverState *old_bs;
1178     BlockDriverState *new_bs;
1179 } ExternalSnapshotState;
1180 
1181 static void external_snapshot_prepare(BlkTransactionState *common,
1182                                       Error **errp)
1183 {
1184     BlockDriver *drv;
1185     int flags, ret;
1186     Error *local_err = NULL;
1187     const char *device;
1188     const char *new_image_file;
1189     const char *format = "qcow2";
1190     enum NewImageMode mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1191     ExternalSnapshotState *state =
1192                              DO_UPCAST(ExternalSnapshotState, common, common);
1193     TransactionAction *action = common->action;
1194 
1195     /* get parameters */
1196     g_assert(action->kind == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC);
1197 
1198     device = action->blockdev_snapshot_sync->device;
1199     new_image_file = action->blockdev_snapshot_sync->snapshot_file;
1200     if (action->blockdev_snapshot_sync->has_format) {
1201         format = action->blockdev_snapshot_sync->format;
1202     }
1203     if (action->blockdev_snapshot_sync->has_mode) {
1204         mode = action->blockdev_snapshot_sync->mode;
1205     }
1206 
1207     /* start processing */
1208     drv = bdrv_find_format(format);
1209     if (!drv) {
1210         error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1211         return;
1212     }
1213 
1214     state->old_bs = bdrv_find(device);
1215     if (!state->old_bs) {
1216         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1217         return;
1218     }
1219 
1220     if (!bdrv_is_inserted(state->old_bs)) {
1221         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1222         return;
1223     }
1224 
1225     if (bdrv_in_use(state->old_bs)) {
1226         error_set(errp, QERR_DEVICE_IN_USE, device);
1227         return;
1228     }
1229 
1230     if (!bdrv_is_read_only(state->old_bs)) {
1231         if (bdrv_flush(state->old_bs)) {
1232             error_set(errp, QERR_IO_ERROR);
1233             return;
1234         }
1235     }
1236 
1237     if (bdrv_check_ext_snapshot(state->old_bs) != EXT_SNAPSHOT_ALLOWED) {
1238         error_set(errp, QERR_FEATURE_DISABLED, "snapshot");
1239         return;
1240     }
1241 
1242     flags = state->old_bs->open_flags;
1243 
1244     /* create new image w/backing file */
1245     if (mode != NEW_IMAGE_MODE_EXISTING) {
1246         bdrv_img_create(new_image_file, format,
1247                         state->old_bs->filename,
1248                         state->old_bs->drv->format_name,
1249                         NULL, -1, flags, &local_err, false);
1250         if (error_is_set(&local_err)) {
1251             error_propagate(errp, local_err);
1252             return;
1253         }
1254     }
1255 
1256     /* We will manually add the backing_hd field to the bs later */
1257     state->new_bs = bdrv_new("");
1258     /* TODO Inherit bs->options or only take explicit options with an
1259      * extended QMP command? */
1260     ret = bdrv_open(state->new_bs, new_image_file, NULL,
1261                     flags | BDRV_O_NO_BACKING, drv, &local_err);
1262     if (ret != 0) {
1263         error_propagate(errp, local_err);
1264     }
1265 }
1266 
1267 static void external_snapshot_commit(BlkTransactionState *common)
1268 {
1269     ExternalSnapshotState *state =
1270                              DO_UPCAST(ExternalSnapshotState, common, common);
1271 
1272     /* This removes our old bs and adds the new bs */
1273     bdrv_append(state->new_bs, state->old_bs);
1274     /* We don't need (or want) to use the transactional
1275      * bdrv_reopen_multiple() across all the entries at once, because we
1276      * don't want to abort all of them if one of them fails the reopen */
1277     bdrv_reopen(state->new_bs, state->new_bs->open_flags & ~BDRV_O_RDWR,
1278                 NULL);
1279 }
1280 
1281 static void external_snapshot_abort(BlkTransactionState *common)
1282 {
1283     ExternalSnapshotState *state =
1284                              DO_UPCAST(ExternalSnapshotState, common, common);
1285     if (state->new_bs) {
1286         bdrv_unref(state->new_bs);
1287     }
1288 }
1289 
1290 typedef struct DriveBackupState {
1291     BlkTransactionState common;
1292     BlockDriverState *bs;
1293     BlockJob *job;
1294 } DriveBackupState;
1295 
1296 static void drive_backup_prepare(BlkTransactionState *common, Error **errp)
1297 {
1298     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1299     DriveBackup *backup;
1300     Error *local_err = NULL;
1301 
1302     assert(common->action->kind == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1303     backup = common->action->drive_backup;
1304 
1305     qmp_drive_backup(backup->device, backup->target,
1306                      backup->has_format, backup->format,
1307                      backup->sync,
1308                      backup->has_mode, backup->mode,
1309                      backup->has_speed, backup->speed,
1310                      backup->has_on_source_error, backup->on_source_error,
1311                      backup->has_on_target_error, backup->on_target_error,
1312                      &local_err);
1313     if (error_is_set(&local_err)) {
1314         error_propagate(errp, local_err);
1315         state->bs = NULL;
1316         state->job = NULL;
1317         return;
1318     }
1319 
1320     state->bs = bdrv_find(backup->device);
1321     state->job = state->bs->job;
1322 }
1323 
1324 static void drive_backup_abort(BlkTransactionState *common)
1325 {
1326     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1327     BlockDriverState *bs = state->bs;
1328 
1329     /* Only cancel if it's the job we started */
1330     if (bs && bs->job && bs->job == state->job) {
1331         block_job_cancel_sync(bs->job);
1332     }
1333 }
1334 
1335 static void abort_prepare(BlkTransactionState *common, Error **errp)
1336 {
1337     error_setg(errp, "Transaction aborted using Abort action");
1338 }
1339 
1340 static void abort_commit(BlkTransactionState *common)
1341 {
1342     g_assert_not_reached(); /* this action never succeeds */
1343 }
1344 
1345 static const BdrvActionOps actions[] = {
1346     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
1347         .instance_size = sizeof(ExternalSnapshotState),
1348         .prepare  = external_snapshot_prepare,
1349         .commit   = external_snapshot_commit,
1350         .abort = external_snapshot_abort,
1351     },
1352     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
1353         .instance_size = sizeof(DriveBackupState),
1354         .prepare = drive_backup_prepare,
1355         .abort = drive_backup_abort,
1356     },
1357     [TRANSACTION_ACTION_KIND_ABORT] = {
1358         .instance_size = sizeof(BlkTransactionState),
1359         .prepare = abort_prepare,
1360         .commit = abort_commit,
1361     },
1362     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
1363         .instance_size = sizeof(InternalSnapshotState),
1364         .prepare  = internal_snapshot_prepare,
1365         .abort = internal_snapshot_abort,
1366     },
1367 };
1368 
1369 /*
1370  * 'Atomic' group snapshots.  The snapshots are taken as a set, and if any fail
1371  *  then we do not pivot any of the devices in the group, and abandon the
1372  *  snapshots
1373  */
1374 void qmp_transaction(TransactionActionList *dev_list, Error **errp)
1375 {
1376     TransactionActionList *dev_entry = dev_list;
1377     BlkTransactionState *state, *next;
1378     Error *local_err = NULL;
1379 
1380     QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionState) snap_bdrv_states;
1381     QSIMPLEQ_INIT(&snap_bdrv_states);
1382 
1383     /* drain all i/o before any snapshots */
1384     bdrv_drain_all();
1385 
1386     /* We don't do anything in this loop that commits us to the snapshot */
1387     while (NULL != dev_entry) {
1388         TransactionAction *dev_info = NULL;
1389         const BdrvActionOps *ops;
1390 
1391         dev_info = dev_entry->value;
1392         dev_entry = dev_entry->next;
1393 
1394         assert(dev_info->kind < ARRAY_SIZE(actions));
1395 
1396         ops = &actions[dev_info->kind];
1397         assert(ops->instance_size > 0);
1398 
1399         state = g_malloc0(ops->instance_size);
1400         state->ops = ops;
1401         state->action = dev_info;
1402         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
1403 
1404         state->ops->prepare(state, &local_err);
1405         if (error_is_set(&local_err)) {
1406             error_propagate(errp, local_err);
1407             goto delete_and_fail;
1408         }
1409     }
1410 
1411     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1412         if (state->ops->commit) {
1413             state->ops->commit(state);
1414         }
1415     }
1416 
1417     /* success */
1418     goto exit;
1419 
1420 delete_and_fail:
1421     /*
1422     * failure, and it is all-or-none; abandon each new bs, and keep using
1423     * the original bs for all images
1424     */
1425     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1426         if (state->ops->abort) {
1427             state->ops->abort(state);
1428         }
1429     }
1430 exit:
1431     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
1432         if (state->ops->clean) {
1433             state->ops->clean(state);
1434         }
1435         g_free(state);
1436     }
1437 }
1438 
1439 
1440 static void eject_device(BlockDriverState *bs, int force, Error **errp)
1441 {
1442     if (bdrv_in_use(bs)) {
1443         error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
1444         return;
1445     }
1446     if (!bdrv_dev_has_removable_media(bs)) {
1447         error_set(errp, QERR_DEVICE_NOT_REMOVABLE, bdrv_get_device_name(bs));
1448         return;
1449     }
1450 
1451     if (bdrv_dev_is_medium_locked(bs) && !bdrv_dev_is_tray_open(bs)) {
1452         bdrv_dev_eject_request(bs, force);
1453         if (!force) {
1454             error_set(errp, QERR_DEVICE_LOCKED, bdrv_get_device_name(bs));
1455             return;
1456         }
1457     }
1458 
1459     bdrv_close(bs);
1460 }
1461 
1462 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
1463 {
1464     BlockDriverState *bs;
1465 
1466     bs = bdrv_find(device);
1467     if (!bs) {
1468         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1469         return;
1470     }
1471 
1472     eject_device(bs, force, errp);
1473 }
1474 
1475 void qmp_block_passwd(const char *device, const char *password, Error **errp)
1476 {
1477     BlockDriverState *bs;
1478     int err;
1479 
1480     bs = bdrv_find(device);
1481     if (!bs) {
1482         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1483         return;
1484     }
1485 
1486     err = bdrv_set_key(bs, password);
1487     if (err == -EINVAL) {
1488         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
1489         return;
1490     } else if (err < 0) {
1491         error_set(errp, QERR_INVALID_PASSWORD);
1492         return;
1493     }
1494 }
1495 
1496 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename,
1497                                     int bdrv_flags, BlockDriver *drv,
1498                                     const char *password, Error **errp)
1499 {
1500     Error *local_err = NULL;
1501     int ret;
1502 
1503     ret = bdrv_open(bs, filename, NULL, bdrv_flags, drv, &local_err);
1504     if (ret < 0) {
1505         error_propagate(errp, local_err);
1506         return;
1507     }
1508 
1509     if (bdrv_key_required(bs)) {
1510         if (password) {
1511             if (bdrv_set_key(bs, password) < 0) {
1512                 error_set(errp, QERR_INVALID_PASSWORD);
1513             }
1514         } else {
1515             error_set(errp, QERR_DEVICE_ENCRYPTED, bdrv_get_device_name(bs),
1516                       bdrv_get_encrypted_filename(bs));
1517         }
1518     } else if (password) {
1519         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
1520     }
1521 }
1522 
1523 void qmp_change_blockdev(const char *device, const char *filename,
1524                          bool has_format, const char *format, Error **errp)
1525 {
1526     BlockDriverState *bs;
1527     BlockDriver *drv = NULL;
1528     int bdrv_flags;
1529     Error *err = NULL;
1530 
1531     bs = bdrv_find(device);
1532     if (!bs) {
1533         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1534         return;
1535     }
1536 
1537     if (format) {
1538         drv = bdrv_find_whitelisted_format(format, bs->read_only);
1539         if (!drv) {
1540             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1541             return;
1542         }
1543     }
1544 
1545     eject_device(bs, 0, &err);
1546     if (error_is_set(&err)) {
1547         error_propagate(errp, err);
1548         return;
1549     }
1550 
1551     bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
1552     bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
1553 
1554     qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp);
1555 }
1556 
1557 /* throttling disk I/O limits */
1558 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
1559                                int64_t bps_wr,
1560                                int64_t iops,
1561                                int64_t iops_rd,
1562                                int64_t iops_wr,
1563                                bool has_bps_max,
1564                                int64_t bps_max,
1565                                bool has_bps_rd_max,
1566                                int64_t bps_rd_max,
1567                                bool has_bps_wr_max,
1568                                int64_t bps_wr_max,
1569                                bool has_iops_max,
1570                                int64_t iops_max,
1571                                bool has_iops_rd_max,
1572                                int64_t iops_rd_max,
1573                                bool has_iops_wr_max,
1574                                int64_t iops_wr_max,
1575                                bool has_iops_size,
1576                                int64_t iops_size, Error **errp)
1577 {
1578     ThrottleConfig cfg;
1579     BlockDriverState *bs;
1580 
1581     bs = bdrv_find(device);
1582     if (!bs) {
1583         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1584         return;
1585     }
1586 
1587     memset(&cfg, 0, sizeof(cfg));
1588     cfg.buckets[THROTTLE_BPS_TOTAL].avg = bps;
1589     cfg.buckets[THROTTLE_BPS_READ].avg  = bps_rd;
1590     cfg.buckets[THROTTLE_BPS_WRITE].avg = bps_wr;
1591 
1592     cfg.buckets[THROTTLE_OPS_TOTAL].avg = iops;
1593     cfg.buckets[THROTTLE_OPS_READ].avg  = iops_rd;
1594     cfg.buckets[THROTTLE_OPS_WRITE].avg = iops_wr;
1595 
1596     if (has_bps_max) {
1597         cfg.buckets[THROTTLE_BPS_TOTAL].max = bps_max;
1598     }
1599     if (has_bps_rd_max) {
1600         cfg.buckets[THROTTLE_BPS_READ].max = bps_rd_max;
1601     }
1602     if (has_bps_wr_max) {
1603         cfg.buckets[THROTTLE_BPS_WRITE].max = bps_wr_max;
1604     }
1605     if (has_iops_max) {
1606         cfg.buckets[THROTTLE_OPS_TOTAL].max = iops_max;
1607     }
1608     if (has_iops_rd_max) {
1609         cfg.buckets[THROTTLE_OPS_READ].max = iops_rd_max;
1610     }
1611     if (has_iops_wr_max) {
1612         cfg.buckets[THROTTLE_OPS_WRITE].max = iops_wr_max;
1613     }
1614 
1615     if (has_iops_size) {
1616         cfg.op_size = iops_size;
1617     }
1618 
1619     if (!check_throttle_config(&cfg, errp)) {
1620         return;
1621     }
1622 
1623     if (!bs->io_limits_enabled && throttle_enabled(&cfg)) {
1624         bdrv_io_limits_enable(bs);
1625     } else if (bs->io_limits_enabled && !throttle_enabled(&cfg)) {
1626         bdrv_io_limits_disable(bs);
1627     }
1628 
1629     if (bs->io_limits_enabled) {
1630         bdrv_set_io_limits(bs, &cfg);
1631     }
1632 }
1633 
1634 int do_drive_del(Monitor *mon, const QDict *qdict, QObject **ret_data)
1635 {
1636     const char *id = qdict_get_str(qdict, "id");
1637     BlockDriverState *bs;
1638 
1639     bs = bdrv_find(id);
1640     if (!bs) {
1641         qerror_report(QERR_DEVICE_NOT_FOUND, id);
1642         return -1;
1643     }
1644     if (bdrv_in_use(bs)) {
1645         qerror_report(QERR_DEVICE_IN_USE, id);
1646         return -1;
1647     }
1648 
1649     /* quiesce block driver; prevent further io */
1650     bdrv_drain_all();
1651     bdrv_flush(bs);
1652     bdrv_close(bs);
1653 
1654     /* if we have a device attached to this BlockDriverState
1655      * then we need to make the drive anonymous until the device
1656      * can be removed.  If this is a drive with no device backing
1657      * then we can just get rid of the block driver state right here.
1658      */
1659     if (bdrv_get_attached_dev(bs)) {
1660         bdrv_make_anon(bs);
1661 
1662         /* Further I/O must not pause the guest */
1663         bdrv_set_on_error(bs, BLOCKDEV_ON_ERROR_REPORT,
1664                           BLOCKDEV_ON_ERROR_REPORT);
1665     } else {
1666         drive_uninit(drive_get_by_blockdev(bs));
1667     }
1668 
1669     return 0;
1670 }
1671 
1672 void qmp_block_resize(const char *device, int64_t size, Error **errp)
1673 {
1674     BlockDriverState *bs;
1675     int ret;
1676 
1677     bs = bdrv_find(device);
1678     if (!bs) {
1679         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1680         return;
1681     }
1682 
1683     if (size < 0) {
1684         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
1685         return;
1686     }
1687 
1688     /* complete all in-flight operations before resizing the device */
1689     bdrv_drain_all();
1690 
1691     ret = bdrv_truncate(bs, size);
1692     switch (ret) {
1693     case 0:
1694         break;
1695     case -ENOMEDIUM:
1696         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1697         break;
1698     case -ENOTSUP:
1699         error_set(errp, QERR_UNSUPPORTED);
1700         break;
1701     case -EACCES:
1702         error_set(errp, QERR_DEVICE_IS_READ_ONLY, device);
1703         break;
1704     case -EBUSY:
1705         error_set(errp, QERR_DEVICE_IN_USE, device);
1706         break;
1707     default:
1708         error_setg_errno(errp, -ret, "Could not resize");
1709         break;
1710     }
1711 }
1712 
1713 static void block_job_cb(void *opaque, int ret)
1714 {
1715     BlockDriverState *bs = opaque;
1716     QObject *obj;
1717 
1718     trace_block_job_cb(bs, bs->job, ret);
1719 
1720     assert(bs->job);
1721     obj = qobject_from_block_job(bs->job);
1722     if (ret < 0) {
1723         QDict *dict = qobject_to_qdict(obj);
1724         qdict_put(dict, "error", qstring_from_str(strerror(-ret)));
1725     }
1726 
1727     if (block_job_is_cancelled(bs->job)) {
1728         monitor_protocol_event(QEVENT_BLOCK_JOB_CANCELLED, obj);
1729     } else {
1730         monitor_protocol_event(QEVENT_BLOCK_JOB_COMPLETED, obj);
1731     }
1732     qobject_decref(obj);
1733 
1734     bdrv_put_ref_bh_schedule(bs);
1735 }
1736 
1737 void qmp_block_stream(const char *device, bool has_base,
1738                       const char *base, bool has_speed, int64_t speed,
1739                       bool has_on_error, BlockdevOnError on_error,
1740                       Error **errp)
1741 {
1742     BlockDriverState *bs;
1743     BlockDriverState *base_bs = NULL;
1744     Error *local_err = NULL;
1745 
1746     if (!has_on_error) {
1747         on_error = BLOCKDEV_ON_ERROR_REPORT;
1748     }
1749 
1750     bs = bdrv_find(device);
1751     if (!bs) {
1752         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1753         return;
1754     }
1755 
1756     if (base) {
1757         base_bs = bdrv_find_backing_image(bs, base);
1758         if (base_bs == NULL) {
1759             error_set(errp, QERR_BASE_NOT_FOUND, base);
1760             return;
1761         }
1762     }
1763 
1764     stream_start(bs, base_bs, base, has_speed ? speed : 0,
1765                  on_error, block_job_cb, bs, &local_err);
1766     if (error_is_set(&local_err)) {
1767         error_propagate(errp, local_err);
1768         return;
1769     }
1770 
1771     trace_qmp_block_stream(bs, bs->job);
1772 }
1773 
1774 void qmp_block_commit(const char *device,
1775                       bool has_base, const char *base, const char *top,
1776                       bool has_speed, int64_t speed,
1777                       Error **errp)
1778 {
1779     BlockDriverState *bs;
1780     BlockDriverState *base_bs, *top_bs;
1781     Error *local_err = NULL;
1782     /* This will be part of the QMP command, if/when the
1783      * BlockdevOnError change for blkmirror makes it in
1784      */
1785     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
1786 
1787     /* drain all i/o before commits */
1788     bdrv_drain_all();
1789 
1790     bs = bdrv_find(device);
1791     if (!bs) {
1792         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1793         return;
1794     }
1795 
1796     /* default top_bs is the active layer */
1797     top_bs = bs;
1798 
1799     if (top) {
1800         if (strcmp(bs->filename, top) != 0) {
1801             top_bs = bdrv_find_backing_image(bs, top);
1802         }
1803     }
1804 
1805     if (top_bs == NULL) {
1806         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
1807         return;
1808     }
1809 
1810     if (has_base && base) {
1811         base_bs = bdrv_find_backing_image(top_bs, base);
1812     } else {
1813         base_bs = bdrv_find_base(top_bs);
1814     }
1815 
1816     if (base_bs == NULL) {
1817         error_set(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
1818         return;
1819     }
1820 
1821     commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
1822                 &local_err);
1823     if (local_err != NULL) {
1824         error_propagate(errp, local_err);
1825         return;
1826     }
1827 }
1828 
1829 void qmp_drive_backup(const char *device, const char *target,
1830                       bool has_format, const char *format,
1831                       enum MirrorSyncMode sync,
1832                       bool has_mode, enum NewImageMode mode,
1833                       bool has_speed, int64_t speed,
1834                       bool has_on_source_error, BlockdevOnError on_source_error,
1835                       bool has_on_target_error, BlockdevOnError on_target_error,
1836                       Error **errp)
1837 {
1838     BlockDriverState *bs;
1839     BlockDriverState *target_bs;
1840     BlockDriverState *source = NULL;
1841     BlockDriver *drv = NULL;
1842     Error *local_err = NULL;
1843     int flags;
1844     int64_t size;
1845     int ret;
1846 
1847     if (!has_speed) {
1848         speed = 0;
1849     }
1850     if (!has_on_source_error) {
1851         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
1852     }
1853     if (!has_on_target_error) {
1854         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
1855     }
1856     if (!has_mode) {
1857         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1858     }
1859 
1860     bs = bdrv_find(device);
1861     if (!bs) {
1862         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1863         return;
1864     }
1865 
1866     if (!bdrv_is_inserted(bs)) {
1867         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1868         return;
1869     }
1870 
1871     if (!has_format) {
1872         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
1873     }
1874     if (format) {
1875         drv = bdrv_find_format(format);
1876         if (!drv) {
1877             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1878             return;
1879         }
1880     }
1881 
1882     if (bdrv_in_use(bs)) {
1883         error_set(errp, QERR_DEVICE_IN_USE, device);
1884         return;
1885     }
1886 
1887     flags = bs->open_flags | BDRV_O_RDWR;
1888 
1889     /* See if we have a backing HD we can use to create our new image
1890      * on top of. */
1891     if (sync == MIRROR_SYNC_MODE_TOP) {
1892         source = bs->backing_hd;
1893         if (!source) {
1894             sync = MIRROR_SYNC_MODE_FULL;
1895         }
1896     }
1897     if (sync == MIRROR_SYNC_MODE_NONE) {
1898         source = bs;
1899     }
1900 
1901     size = bdrv_getlength(bs);
1902     if (size < 0) {
1903         error_setg_errno(errp, -size, "bdrv_getlength failed");
1904         return;
1905     }
1906 
1907     if (mode != NEW_IMAGE_MODE_EXISTING) {
1908         assert(format && drv);
1909         if (source) {
1910             bdrv_img_create(target, format, source->filename,
1911                             source->drv->format_name, NULL,
1912                             size, flags, &local_err, false);
1913         } else {
1914             bdrv_img_create(target, format, NULL, NULL, NULL,
1915                             size, flags, &local_err, false);
1916         }
1917     }
1918 
1919     if (error_is_set(&local_err)) {
1920         error_propagate(errp, local_err);
1921         return;
1922     }
1923 
1924     target_bs = bdrv_new("");
1925     ret = bdrv_open(target_bs, target, NULL, flags, drv, &local_err);
1926     if (ret < 0) {
1927         bdrv_unref(target_bs);
1928         error_propagate(errp, local_err);
1929         return;
1930     }
1931 
1932     backup_start(bs, target_bs, speed, sync, on_source_error, on_target_error,
1933                  block_job_cb, bs, &local_err);
1934     if (local_err != NULL) {
1935         bdrv_unref(target_bs);
1936         error_propagate(errp, local_err);
1937         return;
1938     }
1939 }
1940 
1941 #define DEFAULT_MIRROR_BUF_SIZE   (10 << 20)
1942 
1943 void qmp_drive_mirror(const char *device, const char *target,
1944                       bool has_format, const char *format,
1945                       enum MirrorSyncMode sync,
1946                       bool has_mode, enum NewImageMode mode,
1947                       bool has_speed, int64_t speed,
1948                       bool has_granularity, uint32_t granularity,
1949                       bool has_buf_size, int64_t buf_size,
1950                       bool has_on_source_error, BlockdevOnError on_source_error,
1951                       bool has_on_target_error, BlockdevOnError on_target_error,
1952                       Error **errp)
1953 {
1954     BlockDriverState *bs;
1955     BlockDriverState *source, *target_bs;
1956     BlockDriver *drv = NULL;
1957     Error *local_err = NULL;
1958     int flags;
1959     int64_t size;
1960     int ret;
1961 
1962     if (!has_speed) {
1963         speed = 0;
1964     }
1965     if (!has_on_source_error) {
1966         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
1967     }
1968     if (!has_on_target_error) {
1969         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
1970     }
1971     if (!has_mode) {
1972         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1973     }
1974     if (!has_granularity) {
1975         granularity = 0;
1976     }
1977     if (!has_buf_size) {
1978         buf_size = DEFAULT_MIRROR_BUF_SIZE;
1979     }
1980 
1981     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
1982         error_set(errp, QERR_INVALID_PARAMETER, device);
1983         return;
1984     }
1985     if (granularity & (granularity - 1)) {
1986         error_set(errp, QERR_INVALID_PARAMETER, device);
1987         return;
1988     }
1989 
1990     bs = bdrv_find(device);
1991     if (!bs) {
1992         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1993         return;
1994     }
1995 
1996     if (!bdrv_is_inserted(bs)) {
1997         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1998         return;
1999     }
2000 
2001     if (!has_format) {
2002         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
2003     }
2004     if (format) {
2005         drv = bdrv_find_format(format);
2006         if (!drv) {
2007             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
2008             return;
2009         }
2010     }
2011 
2012     if (bdrv_in_use(bs)) {
2013         error_set(errp, QERR_DEVICE_IN_USE, device);
2014         return;
2015     }
2016 
2017     flags = bs->open_flags | BDRV_O_RDWR;
2018     source = bs->backing_hd;
2019     if (!source && sync == MIRROR_SYNC_MODE_TOP) {
2020         sync = MIRROR_SYNC_MODE_FULL;
2021     }
2022 
2023     size = bdrv_getlength(bs);
2024     if (size < 0) {
2025         error_setg_errno(errp, -size, "bdrv_getlength failed");
2026         return;
2027     }
2028 
2029     if (sync == MIRROR_SYNC_MODE_FULL && mode != NEW_IMAGE_MODE_EXISTING) {
2030         /* create new image w/o backing file */
2031         assert(format && drv);
2032         bdrv_img_create(target, format,
2033                         NULL, NULL, NULL, size, flags, &local_err, false);
2034     } else {
2035         switch (mode) {
2036         case NEW_IMAGE_MODE_EXISTING:
2037             break;
2038         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
2039             /* create new image with backing file */
2040             bdrv_img_create(target, format,
2041                             source->filename,
2042                             source->drv->format_name,
2043                             NULL, size, flags, &local_err, false);
2044             break;
2045         default:
2046             abort();
2047         }
2048     }
2049 
2050     if (error_is_set(&local_err)) {
2051         error_propagate(errp, local_err);
2052         return;
2053     }
2054 
2055     /* Mirroring takes care of copy-on-write using the source's backing
2056      * file.
2057      */
2058     target_bs = bdrv_new("");
2059     ret = bdrv_open(target_bs, target, NULL, flags | BDRV_O_NO_BACKING, drv,
2060                     &local_err);
2061     if (ret < 0) {
2062         bdrv_unref(target_bs);
2063         error_propagate(errp, local_err);
2064         return;
2065     }
2066 
2067     mirror_start(bs, target_bs, speed, granularity, buf_size, sync,
2068                  on_source_error, on_target_error,
2069                  block_job_cb, bs, &local_err);
2070     if (local_err != NULL) {
2071         bdrv_unref(target_bs);
2072         error_propagate(errp, local_err);
2073         return;
2074     }
2075 }
2076 
2077 static BlockJob *find_block_job(const char *device)
2078 {
2079     BlockDriverState *bs;
2080 
2081     bs = bdrv_find(device);
2082     if (!bs || !bs->job) {
2083         return NULL;
2084     }
2085     return bs->job;
2086 }
2087 
2088 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
2089 {
2090     BlockJob *job = find_block_job(device);
2091 
2092     if (!job) {
2093         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2094         return;
2095     }
2096 
2097     block_job_set_speed(job, speed, errp);
2098 }
2099 
2100 void qmp_block_job_cancel(const char *device,
2101                           bool has_force, bool force, Error **errp)
2102 {
2103     BlockJob *job = find_block_job(device);
2104 
2105     if (!has_force) {
2106         force = false;
2107     }
2108 
2109     if (!job) {
2110         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2111         return;
2112     }
2113     if (job->paused && !force) {
2114         error_set(errp, QERR_BLOCK_JOB_PAUSED, device);
2115         return;
2116     }
2117 
2118     trace_qmp_block_job_cancel(job);
2119     block_job_cancel(job);
2120 }
2121 
2122 void qmp_block_job_pause(const char *device, Error **errp)
2123 {
2124     BlockJob *job = find_block_job(device);
2125 
2126     if (!job) {
2127         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2128         return;
2129     }
2130 
2131     trace_qmp_block_job_pause(job);
2132     block_job_pause(job);
2133 }
2134 
2135 void qmp_block_job_resume(const char *device, Error **errp)
2136 {
2137     BlockJob *job = find_block_job(device);
2138 
2139     if (!job) {
2140         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2141         return;
2142     }
2143 
2144     trace_qmp_block_job_resume(job);
2145     block_job_resume(job);
2146 }
2147 
2148 void qmp_block_job_complete(const char *device, Error **errp)
2149 {
2150     BlockJob *job = find_block_job(device);
2151 
2152     if (!job) {
2153         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2154         return;
2155     }
2156 
2157     trace_qmp_block_job_complete(job);
2158     block_job_complete(job, errp);
2159 }
2160 
2161 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
2162 {
2163     QmpOutputVisitor *ov = qmp_output_visitor_new();
2164     QObject *obj;
2165     QDict *qdict;
2166     Error *local_err = NULL;
2167 
2168     /* Require an ID in the top level */
2169     if (!options->has_id) {
2170         error_setg(errp, "Block device needs an ID");
2171         goto fail;
2172     }
2173 
2174     /* TODO Sort it out in raw-posix and drive_init: Reject aio=native with
2175      * cache.direct=false instead of silently switching to aio=threads, except
2176      * if called from drive_init.
2177      *
2178      * For now, simply forbidding the combination for all drivers will do. */
2179     if (options->has_aio && options->aio == BLOCKDEV_AIO_OPTIONS_NATIVE) {
2180         bool direct = options->cache->has_direct && options->cache->direct;
2181         if (!options->has_cache && !direct) {
2182             error_setg(errp, "aio=native requires cache.direct=true");
2183             goto fail;
2184         }
2185     }
2186 
2187     visit_type_BlockdevOptions(qmp_output_get_visitor(ov),
2188                                &options, NULL, &local_err);
2189     if (error_is_set(&local_err)) {
2190         error_propagate(errp, local_err);
2191         goto fail;
2192     }
2193 
2194     obj = qmp_output_get_qobject(ov);
2195     qdict = qobject_to_qdict(obj);
2196 
2197     qdict_flatten(qdict);
2198 
2199     blockdev_init(qdict, IF_NONE, &local_err);
2200     if (error_is_set(&local_err)) {
2201         error_propagate(errp, local_err);
2202         goto fail;
2203     }
2204 
2205 fail:
2206     qmp_output_visitor_cleanup(ov);
2207 }
2208 
2209 static void do_qmp_query_block_jobs_one(void *opaque, BlockDriverState *bs)
2210 {
2211     BlockJobInfoList **prev = opaque;
2212     BlockJob *job = bs->job;
2213 
2214     if (job) {
2215         BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
2216         elem->value = block_job_query(bs->job);
2217         (*prev)->next = elem;
2218         *prev = elem;
2219     }
2220 }
2221 
2222 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
2223 {
2224     /* Dummy is a fake list element for holding the head pointer */
2225     BlockJobInfoList dummy = {};
2226     BlockJobInfoList *prev = &dummy;
2227     bdrv_iterate(do_qmp_query_block_jobs_one, &prev);
2228     return dummy.next;
2229 }
2230 
2231 QemuOptsList qemu_common_drive_opts = {
2232     .name = "drive",
2233     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
2234     .desc = {
2235         {
2236             .name = "snapshot",
2237             .type = QEMU_OPT_BOOL,
2238             .help = "enable/disable snapshot mode",
2239         },{
2240             .name = "file",
2241             .type = QEMU_OPT_STRING,
2242             .help = "disk image",
2243         },{
2244             .name = "discard",
2245             .type = QEMU_OPT_STRING,
2246             .help = "discard operation (ignore/off, unmap/on)",
2247         },{
2248             .name = "cache.writeback",
2249             .type = QEMU_OPT_BOOL,
2250             .help = "enables writeback mode for any caches",
2251         },{
2252             .name = "cache.direct",
2253             .type = QEMU_OPT_BOOL,
2254             .help = "enables use of O_DIRECT (bypass the host page cache)",
2255         },{
2256             .name = "cache.no-flush",
2257             .type = QEMU_OPT_BOOL,
2258             .help = "ignore any flush requests for the device",
2259         },{
2260             .name = "aio",
2261             .type = QEMU_OPT_STRING,
2262             .help = "host AIO implementation (threads, native)",
2263         },{
2264             .name = "format",
2265             .type = QEMU_OPT_STRING,
2266             .help = "disk format (raw, qcow2, ...)",
2267         },{
2268             .name = "serial",
2269             .type = QEMU_OPT_STRING,
2270             .help = "disk serial number",
2271         },{
2272             .name = "rerror",
2273             .type = QEMU_OPT_STRING,
2274             .help = "read error action",
2275         },{
2276             .name = "werror",
2277             .type = QEMU_OPT_STRING,
2278             .help = "write error action",
2279         },{
2280             .name = "read-only",
2281             .type = QEMU_OPT_BOOL,
2282             .help = "open drive file as read-only",
2283         },{
2284             .name = "throttling.iops-total",
2285             .type = QEMU_OPT_NUMBER,
2286             .help = "limit total I/O operations per second",
2287         },{
2288             .name = "throttling.iops-read",
2289             .type = QEMU_OPT_NUMBER,
2290             .help = "limit read operations per second",
2291         },{
2292             .name = "throttling.iops-write",
2293             .type = QEMU_OPT_NUMBER,
2294             .help = "limit write operations per second",
2295         },{
2296             .name = "throttling.bps-total",
2297             .type = QEMU_OPT_NUMBER,
2298             .help = "limit total bytes per second",
2299         },{
2300             .name = "throttling.bps-read",
2301             .type = QEMU_OPT_NUMBER,
2302             .help = "limit read bytes per second",
2303         },{
2304             .name = "throttling.bps-write",
2305             .type = QEMU_OPT_NUMBER,
2306             .help = "limit write bytes per second",
2307         },{
2308             .name = "throttling.iops-total-max",
2309             .type = QEMU_OPT_NUMBER,
2310             .help = "I/O operations burst",
2311         },{
2312             .name = "throttling.iops-read-max",
2313             .type = QEMU_OPT_NUMBER,
2314             .help = "I/O operations read burst",
2315         },{
2316             .name = "throttling.iops-write-max",
2317             .type = QEMU_OPT_NUMBER,
2318             .help = "I/O operations write burst",
2319         },{
2320             .name = "throttling.bps-total-max",
2321             .type = QEMU_OPT_NUMBER,
2322             .help = "total bytes burst",
2323         },{
2324             .name = "throttling.bps-read-max",
2325             .type = QEMU_OPT_NUMBER,
2326             .help = "total bytes read burst",
2327         },{
2328             .name = "throttling.bps-write-max",
2329             .type = QEMU_OPT_NUMBER,
2330             .help = "total bytes write burst",
2331         },{
2332             .name = "throttling.iops-size",
2333             .type = QEMU_OPT_NUMBER,
2334             .help = "when limiting by iops max size of an I/O in bytes",
2335         },{
2336             .name = "copy-on-read",
2337             .type = QEMU_OPT_BOOL,
2338             .help = "copy read data from backing file into image file",
2339         },
2340         { /* end of list */ }
2341     },
2342 };
2343 
2344 QemuOptsList qemu_drive_opts = {
2345     .name = "drive",
2346     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
2347     .desc = {
2348         /*
2349          * no elements => accept any params
2350          * validation will happen later
2351          */
2352         { /* end of list */ }
2353     },
2354 };
2355