xref: /qemu/blockdev.c (revision 47908a0f)
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 
51 static const char *const if_name[IF_COUNT] = {
52     [IF_NONE] = "none",
53     [IF_IDE] = "ide",
54     [IF_SCSI] = "scsi",
55     [IF_FLOPPY] = "floppy",
56     [IF_PFLASH] = "pflash",
57     [IF_MTD] = "mtd",
58     [IF_SD] = "sd",
59     [IF_VIRTIO] = "virtio",
60     [IF_XEN] = "xen",
61 };
62 
63 static const int if_max_devs[IF_COUNT] = {
64     /*
65      * Do not change these numbers!  They govern how drive option
66      * index maps to unit and bus.  That mapping is ABI.
67      *
68      * All controllers used to imlement if=T drives need to support
69      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
70      * Otherwise, some index values map to "impossible" bus, unit
71      * values.
72      *
73      * For instance, if you change [IF_SCSI] to 255, -drive
74      * if=scsi,index=12 no longer means bus=1,unit=5, but
75      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
76      * the drive can't be set up.  Regression.
77      */
78     [IF_IDE] = 2,
79     [IF_SCSI] = 7,
80 };
81 
82 /*
83  * We automatically delete the drive when a device using it gets
84  * unplugged.  Questionable feature, but we can't just drop it.
85  * Device models call blockdev_mark_auto_del() to schedule the
86  * automatic deletion, and generic qdev code calls blockdev_auto_del()
87  * when deletion is actually safe.
88  */
89 void blockdev_mark_auto_del(BlockDriverState *bs)
90 {
91     DriveInfo *dinfo = drive_get_by_blockdev(bs);
92 
93     if (dinfo && !dinfo->enable_auto_del) {
94         return;
95     }
96 
97     if (bs->job) {
98         block_job_cancel(bs->job);
99     }
100     if (dinfo) {
101         dinfo->auto_del = 1;
102     }
103 }
104 
105 void blockdev_auto_del(BlockDriverState *bs)
106 {
107     DriveInfo *dinfo = drive_get_by_blockdev(bs);
108 
109     if (dinfo && dinfo->auto_del) {
110         drive_put_ref(dinfo);
111     }
112 }
113 
114 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
115 {
116     int max_devs = if_max_devs[type];
117     return max_devs ? index / max_devs : 0;
118 }
119 
120 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
121 {
122     int max_devs = if_max_devs[type];
123     return max_devs ? index % max_devs : index;
124 }
125 
126 QemuOpts *drive_def(const char *optstr)
127 {
128     return qemu_opts_parse(qemu_find_opts("drive"), optstr, 0);
129 }
130 
131 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
132                     const char *optstr)
133 {
134     QemuOpts *opts;
135     char buf[32];
136 
137     opts = drive_def(optstr);
138     if (!opts) {
139         return NULL;
140     }
141     if (type != IF_DEFAULT) {
142         qemu_opt_set(opts, "if", if_name[type]);
143     }
144     if (index >= 0) {
145         snprintf(buf, sizeof(buf), "%d", index);
146         qemu_opt_set(opts, "index", buf);
147     }
148     if (file)
149         qemu_opt_set(opts, "file", file);
150     return opts;
151 }
152 
153 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
154 {
155     DriveInfo *dinfo;
156 
157     /* seek interface, bus and unit */
158 
159     QTAILQ_FOREACH(dinfo, &drives, next) {
160         if (dinfo->type == type &&
161 	    dinfo->bus == bus &&
162 	    dinfo->unit == unit)
163             return dinfo;
164     }
165 
166     return NULL;
167 }
168 
169 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
170 {
171     return drive_get(type,
172                      drive_index_to_bus_id(type, index),
173                      drive_index_to_unit_id(type, index));
174 }
175 
176 int drive_get_max_bus(BlockInterfaceType type)
177 {
178     int max_bus;
179     DriveInfo *dinfo;
180 
181     max_bus = -1;
182     QTAILQ_FOREACH(dinfo, &drives, next) {
183         if(dinfo->type == type &&
184            dinfo->bus > max_bus)
185             max_bus = dinfo->bus;
186     }
187     return max_bus;
188 }
189 
190 /* Get a block device.  This should only be used for single-drive devices
191    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
192    appropriate bus.  */
193 DriveInfo *drive_get_next(BlockInterfaceType type)
194 {
195     static int next_block_unit[IF_COUNT];
196 
197     return drive_get(type, 0, next_block_unit[type]++);
198 }
199 
200 DriveInfo *drive_get_by_blockdev(BlockDriverState *bs)
201 {
202     DriveInfo *dinfo;
203 
204     QTAILQ_FOREACH(dinfo, &drives, next) {
205         if (dinfo->bdrv == bs) {
206             return dinfo;
207         }
208     }
209     return NULL;
210 }
211 
212 static void bdrv_format_print(void *opaque, const char *name)
213 {
214     error_printf(" %s", name);
215 }
216 
217 static void drive_uninit(DriveInfo *dinfo)
218 {
219     if (dinfo->opts) {
220         qemu_opts_del(dinfo->opts);
221     }
222 
223     bdrv_unref(dinfo->bdrv);
224     g_free(dinfo->id);
225     QTAILQ_REMOVE(&drives, dinfo, next);
226     g_free(dinfo->serial);
227     g_free(dinfo);
228 }
229 
230 void drive_put_ref(DriveInfo *dinfo)
231 {
232     assert(dinfo->refcount);
233     if (--dinfo->refcount == 0) {
234         drive_uninit(dinfo);
235     }
236 }
237 
238 void drive_get_ref(DriveInfo *dinfo)
239 {
240     dinfo->refcount++;
241 }
242 
243 typedef struct {
244     QEMUBH *bh;
245     BlockDriverState *bs;
246 } BDRVPutRefBH;
247 
248 static void bdrv_put_ref_bh(void *opaque)
249 {
250     BDRVPutRefBH *s = opaque;
251 
252     bdrv_unref(s->bs);
253     qemu_bh_delete(s->bh);
254     g_free(s);
255 }
256 
257 /*
258  * Release a BDS reference in a BH
259  *
260  * It is not safe to use bdrv_unref() from a callback function when the callers
261  * still need the BlockDriverState.  In such cases we schedule a BH to release
262  * the reference.
263  */
264 static void bdrv_put_ref_bh_schedule(BlockDriverState *bs)
265 {
266     BDRVPutRefBH *s;
267 
268     s = g_new(BDRVPutRefBH, 1);
269     s->bh = qemu_bh_new(bdrv_put_ref_bh, s);
270     s->bs = bs;
271     qemu_bh_schedule(s->bh);
272 }
273 
274 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
275 {
276     if (!strcmp(buf, "ignore")) {
277         return BLOCKDEV_ON_ERROR_IGNORE;
278     } else if (!is_read && !strcmp(buf, "enospc")) {
279         return BLOCKDEV_ON_ERROR_ENOSPC;
280     } else if (!strcmp(buf, "stop")) {
281         return BLOCKDEV_ON_ERROR_STOP;
282     } else if (!strcmp(buf, "report")) {
283         return BLOCKDEV_ON_ERROR_REPORT;
284     } else {
285         error_setg(errp, "'%s' invalid %s error action",
286                    buf, is_read ? "read" : "write");
287         return -1;
288     }
289 }
290 
291 static bool check_throttle_config(ThrottleConfig *cfg, Error **errp)
292 {
293     if (throttle_conflicting(cfg)) {
294         error_setg(errp, "bps/iops/max total values and read/write values"
295                          " cannot be used at the same time");
296         return false;
297     }
298 
299     if (!throttle_is_valid(cfg)) {
300         error_setg(errp, "bps/iops/maxs values must be 0 or greater");
301         return false;
302     }
303 
304     return true;
305 }
306 
307 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
308 
309 /* Takes the ownership of bs_opts */
310 static DriveInfo *blockdev_init(QDict *bs_opts,
311                                 BlockInterfaceType type,
312                                 Error **errp)
313 {
314     const char *buf;
315     const char *file = NULL;
316     const char *serial;
317     int ro = 0;
318     int bdrv_flags = 0;
319     int on_read_error, on_write_error;
320     DriveInfo *dinfo;
321     ThrottleConfig cfg;
322     int snapshot = 0;
323     bool copy_on_read;
324     int ret;
325     Error *error = NULL;
326     QemuOpts *opts;
327     const char *id;
328     bool has_driver_specific_opts;
329     BlockDriver *drv = NULL;
330 
331     /* Check common options by copying from bs_opts to opts, all other options
332      * stay in bs_opts for processing by bdrv_open(). */
333     id = qdict_get_try_str(bs_opts, "id");
334     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
335     if (error_is_set(&error)) {
336         error_propagate(errp, error);
337         return NULL;
338     }
339 
340     qemu_opts_absorb_qdict(opts, bs_opts, &error);
341     if (error_is_set(&error)) {
342         error_propagate(errp, error);
343         goto early_err;
344     }
345 
346     if (id) {
347         qdict_del(bs_opts, "id");
348     }
349 
350     has_driver_specific_opts = !!qdict_size(bs_opts);
351 
352     /* extract parameters */
353     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
354     ro = qemu_opt_get_bool(opts, "read-only", 0);
355     copy_on_read = qemu_opt_get_bool(opts, "copy-on-read", false);
356 
357     file = qemu_opt_get(opts, "file");
358     serial = qemu_opt_get(opts, "serial");
359 
360     if ((buf = qemu_opt_get(opts, "discard")) != NULL) {
361         if (bdrv_parse_discard_flags(buf, &bdrv_flags) != 0) {
362             error_setg(errp, "invalid discard option");
363             goto early_err;
364         }
365     }
366 
367     if (qemu_opt_get_bool(opts, "cache.writeback", true)) {
368         bdrv_flags |= BDRV_O_CACHE_WB;
369     }
370     if (qemu_opt_get_bool(opts, "cache.direct", false)) {
371         bdrv_flags |= BDRV_O_NOCACHE;
372     }
373     if (qemu_opt_get_bool(opts, "cache.no-flush", false)) {
374         bdrv_flags |= BDRV_O_NO_FLUSH;
375     }
376 
377 #ifdef CONFIG_LINUX_AIO
378     if ((buf = qemu_opt_get(opts, "aio")) != NULL) {
379         if (!strcmp(buf, "native")) {
380             bdrv_flags |= BDRV_O_NATIVE_AIO;
381         } else if (!strcmp(buf, "threads")) {
382             /* this is the default */
383         } else {
384            error_setg(errp, "invalid aio option");
385            goto early_err;
386         }
387     }
388 #endif
389 
390     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
391         if (is_help_option(buf)) {
392             error_printf("Supported formats:");
393             bdrv_iterate_format(bdrv_format_print, NULL);
394             error_printf("\n");
395             goto early_err;
396         }
397 
398         drv = bdrv_find_format(buf);
399         if (!drv) {
400             error_setg(errp, "'%s' invalid format", buf);
401             goto early_err;
402         }
403     }
404 
405     /* disk I/O throttling */
406     memset(&cfg, 0, sizeof(cfg));
407     cfg.buckets[THROTTLE_BPS_TOTAL].avg =
408         qemu_opt_get_number(opts, "throttling.bps-total", 0);
409     cfg.buckets[THROTTLE_BPS_READ].avg  =
410         qemu_opt_get_number(opts, "throttling.bps-read", 0);
411     cfg.buckets[THROTTLE_BPS_WRITE].avg =
412         qemu_opt_get_number(opts, "throttling.bps-write", 0);
413     cfg.buckets[THROTTLE_OPS_TOTAL].avg =
414         qemu_opt_get_number(opts, "throttling.iops-total", 0);
415     cfg.buckets[THROTTLE_OPS_READ].avg =
416         qemu_opt_get_number(opts, "throttling.iops-read", 0);
417     cfg.buckets[THROTTLE_OPS_WRITE].avg =
418         qemu_opt_get_number(opts, "throttling.iops-write", 0);
419 
420     cfg.buckets[THROTTLE_BPS_TOTAL].max =
421         qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
422     cfg.buckets[THROTTLE_BPS_READ].max  =
423         qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
424     cfg.buckets[THROTTLE_BPS_WRITE].max =
425         qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
426     cfg.buckets[THROTTLE_OPS_TOTAL].max =
427         qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
428     cfg.buckets[THROTTLE_OPS_READ].max =
429         qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
430     cfg.buckets[THROTTLE_OPS_WRITE].max =
431         qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
432 
433     cfg.op_size = qemu_opt_get_number(opts, "throttling.iops-size", 0);
434 
435     if (!check_throttle_config(&cfg, &error)) {
436         error_propagate(errp, error);
437         goto early_err;
438     }
439 
440     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
441     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
442         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO && type != IF_NONE) {
443             error_setg(errp, "werror is not supported by this bus type");
444             goto early_err;
445         }
446 
447         on_write_error = parse_block_error_action(buf, 0, &error);
448         if (error_is_set(&error)) {
449             error_propagate(errp, error);
450             goto early_err;
451         }
452     }
453 
454     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
455     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
456         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI && type != IF_NONE) {
457             error_report("rerror is not supported by this bus type");
458             goto early_err;
459         }
460 
461         on_read_error = parse_block_error_action(buf, 1, &error);
462         if (error_is_set(&error)) {
463             error_propagate(errp, error);
464             goto early_err;
465         }
466     }
467 
468     /* init */
469     dinfo = g_malloc0(sizeof(*dinfo));
470     dinfo->id = g_strdup(qemu_opts_id(opts));
471     dinfo->bdrv = bdrv_new(dinfo->id);
472     dinfo->bdrv->open_flags = snapshot ? BDRV_O_SNAPSHOT : 0;
473     dinfo->bdrv->read_only = ro;
474     dinfo->type = type;
475     dinfo->refcount = 1;
476     if (serial != NULL) {
477         dinfo->serial = g_strdup(serial);
478     }
479     QTAILQ_INSERT_TAIL(&drives, dinfo, next);
480 
481     bdrv_set_on_error(dinfo->bdrv, on_read_error, on_write_error);
482 
483     /* disk I/O throttling */
484     if (throttle_enabled(&cfg)) {
485         bdrv_io_limits_enable(dinfo->bdrv);
486         bdrv_set_io_limits(dinfo->bdrv, &cfg);
487     }
488 
489     if (!file || !*file) {
490         if (has_driver_specific_opts) {
491             file = NULL;
492         } else {
493             QDECREF(bs_opts);
494             qemu_opts_del(opts);
495             return dinfo;
496         }
497     }
498     if (snapshot) {
499         /* always use cache=unsafe with snapshot */
500         bdrv_flags &= ~BDRV_O_CACHE_MASK;
501         bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
502     }
503 
504     if (copy_on_read) {
505         bdrv_flags |= BDRV_O_COPY_ON_READ;
506     }
507 
508     if (runstate_check(RUN_STATE_INMIGRATE)) {
509         bdrv_flags |= BDRV_O_INCOMING;
510     }
511 
512     bdrv_flags |= ro ? 0 : BDRV_O_RDWR;
513 
514     QINCREF(bs_opts);
515     ret = bdrv_open(dinfo->bdrv, file, bs_opts, bdrv_flags, drv, &error);
516 
517     if (ret < 0) {
518         error_setg(errp, "could not open disk image %s: %s",
519                    file ?: dinfo->id, error_get_pretty(error));
520         error_free(error);
521         goto err;
522     }
523 
524     if (bdrv_key_required(dinfo->bdrv))
525         autostart = 0;
526 
527     QDECREF(bs_opts);
528     qemu_opts_del(opts);
529 
530     return dinfo;
531 
532 err:
533     bdrv_unref(dinfo->bdrv);
534     g_free(dinfo->id);
535     QTAILQ_REMOVE(&drives, dinfo, next);
536     g_free(dinfo);
537 early_err:
538     QDECREF(bs_opts);
539     qemu_opts_del(opts);
540     return NULL;
541 }
542 
543 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to)
544 {
545     const char *value;
546 
547     value = qemu_opt_get(opts, from);
548     if (value) {
549         qemu_opt_set(opts, to, value);
550         qemu_opt_unset(opts, from);
551     }
552 }
553 
554 QemuOptsList qemu_legacy_drive_opts = {
555     .name = "drive",
556     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
557     .desc = {
558         {
559             .name = "bus",
560             .type = QEMU_OPT_NUMBER,
561             .help = "bus number",
562         },{
563             .name = "unit",
564             .type = QEMU_OPT_NUMBER,
565             .help = "unit number (i.e. lun for scsi)",
566         },{
567             .name = "index",
568             .type = QEMU_OPT_NUMBER,
569             .help = "index number",
570         },{
571             .name = "media",
572             .type = QEMU_OPT_STRING,
573             .help = "media type (disk, cdrom)",
574         },{
575             .name = "if",
576             .type = QEMU_OPT_STRING,
577             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
578         },{
579             .name = "cyls",
580             .type = QEMU_OPT_NUMBER,
581             .help = "number of cylinders (ide disk geometry)",
582         },{
583             .name = "heads",
584             .type = QEMU_OPT_NUMBER,
585             .help = "number of heads (ide disk geometry)",
586         },{
587             .name = "secs",
588             .type = QEMU_OPT_NUMBER,
589             .help = "number of sectors (ide disk geometry)",
590         },{
591             .name = "trans",
592             .type = QEMU_OPT_STRING,
593             .help = "chs translation (auto, lba, none)",
594         },{
595             .name = "boot",
596             .type = QEMU_OPT_BOOL,
597             .help = "(deprecated, ignored)",
598         },{
599             .name = "addr",
600             .type = QEMU_OPT_STRING,
601             .help = "pci address (virtio only)",
602         },
603 
604         /* Options that are passed on, but have special semantics with -drive */
605         {
606             .name = "read-only",
607             .type = QEMU_OPT_BOOL,
608             .help = "open drive file as read-only",
609         },{
610             .name = "copy-on-read",
611             .type = QEMU_OPT_BOOL,
612             .help = "copy read data from backing file into image file",
613         },
614 
615         { /* end of list */ }
616     },
617 };
618 
619 DriveInfo *drive_init(QemuOpts *all_opts, BlockInterfaceType block_default_type)
620 {
621     const char *value;
622     DriveInfo *dinfo = NULL;
623     QDict *bs_opts;
624     QemuOpts *legacy_opts;
625     DriveMediaType media = MEDIA_DISK;
626     BlockInterfaceType type;
627     int cyls, heads, secs, translation;
628     int max_devs, bus_id, unit_id, index;
629     const char *devaddr;
630     bool read_only = false;
631     bool copy_on_read;
632     Error *local_err = NULL;
633 
634     /* Change legacy command line options into QMP ones */
635     qemu_opt_rename(all_opts, "iops", "throttling.iops-total");
636     qemu_opt_rename(all_opts, "iops_rd", "throttling.iops-read");
637     qemu_opt_rename(all_opts, "iops_wr", "throttling.iops-write");
638 
639     qemu_opt_rename(all_opts, "bps", "throttling.bps-total");
640     qemu_opt_rename(all_opts, "bps_rd", "throttling.bps-read");
641     qemu_opt_rename(all_opts, "bps_wr", "throttling.bps-write");
642 
643     qemu_opt_rename(all_opts, "iops_max", "throttling.iops-total-max");
644     qemu_opt_rename(all_opts, "iops_rd_max", "throttling.iops-read-max");
645     qemu_opt_rename(all_opts, "iops_wr_max", "throttling.iops-write-max");
646 
647     qemu_opt_rename(all_opts, "bps_max", "throttling.bps-total-max");
648     qemu_opt_rename(all_opts, "bps_rd_max", "throttling.bps-read-max");
649     qemu_opt_rename(all_opts, "bps_wr_max", "throttling.bps-write-max");
650 
651     qemu_opt_rename(all_opts,
652                     "iops_size", "throttling.iops-size");
653 
654     qemu_opt_rename(all_opts, "readonly", "read-only");
655 
656     value = qemu_opt_get(all_opts, "cache");
657     if (value) {
658         int flags = 0;
659 
660         if (bdrv_parse_cache_flags(value, &flags) != 0) {
661             error_report("invalid cache option");
662             return NULL;
663         }
664 
665         /* Specific options take precedence */
666         if (!qemu_opt_get(all_opts, "cache.writeback")) {
667             qemu_opt_set_bool(all_opts, "cache.writeback",
668                               !!(flags & BDRV_O_CACHE_WB));
669         }
670         if (!qemu_opt_get(all_opts, "cache.direct")) {
671             qemu_opt_set_bool(all_opts, "cache.direct",
672                               !!(flags & BDRV_O_NOCACHE));
673         }
674         if (!qemu_opt_get(all_opts, "cache.no-flush")) {
675             qemu_opt_set_bool(all_opts, "cache.no-flush",
676                               !!(flags & BDRV_O_NO_FLUSH));
677         }
678         qemu_opt_unset(all_opts, "cache");
679     }
680 
681     /* Get a QDict for processing the options */
682     bs_opts = qdict_new();
683     qemu_opts_to_qdict(all_opts, bs_opts);
684 
685     legacy_opts = qemu_opts_create_nofail(&qemu_legacy_drive_opts);
686     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
687     if (error_is_set(&local_err)) {
688         qerror_report_err(local_err);
689         error_free(local_err);
690         goto fail;
691     }
692 
693     /* Deprecated option boot=[on|off] */
694     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
695         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
696                 "ignored. Future versions will reject this parameter. Please "
697                 "update your scripts.\n");
698     }
699 
700     /* Media type */
701     value = qemu_opt_get(legacy_opts, "media");
702     if (value) {
703         if (!strcmp(value, "disk")) {
704             media = MEDIA_DISK;
705         } else if (!strcmp(value, "cdrom")) {
706             media = MEDIA_CDROM;
707             read_only = true;
708         } else {
709             error_report("'%s' invalid media", value);
710             goto fail;
711         }
712     }
713 
714     /* copy-on-read is disabled with a warning for read-only devices */
715     read_only |= qemu_opt_get_bool(legacy_opts, "read-only", false);
716     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
717 
718     if (read_only && copy_on_read) {
719         error_report("warning: disabling copy-on-read on read-only drive");
720         copy_on_read = false;
721     }
722 
723     qdict_put(bs_opts, "read-only",
724               qstring_from_str(read_only ? "on" : "off"));
725     qdict_put(bs_opts, "copy-on-read",
726               qstring_from_str(copy_on_read ? "on" :"off"));
727 
728     /* Controller type */
729     value = qemu_opt_get(legacy_opts, "if");
730     if (value) {
731         for (type = 0;
732              type < IF_COUNT && strcmp(value, if_name[type]);
733              type++) {
734         }
735         if (type == IF_COUNT) {
736             error_report("unsupported bus type '%s'", value);
737             goto fail;
738         }
739     } else {
740         type = block_default_type;
741     }
742 
743     /* Geometry */
744     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
745     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
746     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
747 
748     if (cyls || heads || secs) {
749         if (cyls < 1) {
750             error_report("invalid physical cyls number");
751             goto fail;
752         }
753         if (heads < 1) {
754             error_report("invalid physical heads number");
755             goto fail;
756         }
757         if (secs < 1) {
758             error_report("invalid physical secs number");
759             goto fail;
760         }
761     }
762 
763     translation = BIOS_ATA_TRANSLATION_AUTO;
764     value = qemu_opt_get(legacy_opts, "trans");
765     if (value != NULL) {
766         if (!cyls) {
767             error_report("'%s' trans must be used with cyls, heads and secs",
768                          value);
769             goto fail;
770         }
771         if (!strcmp(value, "none")) {
772             translation = BIOS_ATA_TRANSLATION_NONE;
773         } else if (!strcmp(value, "lba")) {
774             translation = BIOS_ATA_TRANSLATION_LBA;
775         } else if (!strcmp(value, "auto")) {
776             translation = BIOS_ATA_TRANSLATION_AUTO;
777         } else {
778             error_report("'%s' invalid translation type", value);
779             goto fail;
780         }
781     }
782 
783     if (media == MEDIA_CDROM) {
784         if (cyls || secs || heads) {
785             error_report("CHS can't be set with media=cdrom");
786             goto fail;
787         }
788     }
789 
790     /* Device address specified by bus/unit or index.
791      * If none was specified, try to find the first free one. */
792     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
793     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
794     index   = qemu_opt_get_number(legacy_opts, "index", -1);
795 
796     max_devs = if_max_devs[type];
797 
798     if (index != -1) {
799         if (bus_id != 0 || unit_id != -1) {
800             error_report("index cannot be used with bus and unit");
801             goto fail;
802         }
803         bus_id = drive_index_to_bus_id(type, index);
804         unit_id = drive_index_to_unit_id(type, index);
805     }
806 
807     if (unit_id == -1) {
808        unit_id = 0;
809        while (drive_get(type, bus_id, unit_id) != NULL) {
810            unit_id++;
811            if (max_devs && unit_id >= max_devs) {
812                unit_id -= max_devs;
813                bus_id++;
814            }
815        }
816     }
817 
818     if (max_devs && unit_id >= max_devs) {
819         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
820         goto fail;
821     }
822 
823     if (drive_get(type, bus_id, unit_id) != NULL) {
824         error_report("drive with bus=%d, unit=%d (index=%d) exists",
825                      bus_id, unit_id, index);
826         goto fail;
827     }
828 
829     /* no id supplied -> create one */
830     if (qemu_opts_id(all_opts) == NULL) {
831         char *new_id;
832         const char *mediastr = "";
833         if (type == IF_IDE || type == IF_SCSI) {
834             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
835         }
836         if (max_devs) {
837             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
838                                      mediastr, unit_id);
839         } else {
840             new_id = g_strdup_printf("%s%s%i", if_name[type],
841                                      mediastr, unit_id);
842         }
843         qdict_put(bs_opts, "id", qstring_from_str(new_id));
844         g_free(new_id);
845     }
846 
847     /* Add virtio block device */
848     devaddr = qemu_opt_get(legacy_opts, "addr");
849     if (devaddr && type != IF_VIRTIO) {
850         error_report("addr is not supported by this bus type");
851         goto fail;
852     }
853 
854     if (type == IF_VIRTIO) {
855         QemuOpts *devopts;
856         devopts = qemu_opts_create_nofail(qemu_find_opts("device"));
857         if (arch_type == QEMU_ARCH_S390X) {
858             qemu_opt_set(devopts, "driver", "virtio-blk-s390");
859         } else {
860             qemu_opt_set(devopts, "driver", "virtio-blk-pci");
861         }
862         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"));
863         if (devaddr) {
864             qemu_opt_set(devopts, "addr", devaddr);
865         }
866     }
867 
868     /* Actual block device init: Functionality shared with blockdev-add */
869     dinfo = blockdev_init(bs_opts, type, &local_err);
870     if (dinfo == NULL) {
871         if (error_is_set(&local_err)) {
872             qerror_report_err(local_err);
873             error_free(local_err);
874         }
875         goto fail;
876     } else {
877         assert(!error_is_set(&local_err));
878     }
879 
880     /* Set legacy DriveInfo fields */
881     dinfo->enable_auto_del = true;
882     dinfo->opts = all_opts;
883 
884     dinfo->cyls = cyls;
885     dinfo->heads = heads;
886     dinfo->secs = secs;
887     dinfo->trans = translation;
888 
889     dinfo->bus = bus_id;
890     dinfo->unit = unit_id;
891     dinfo->devaddr = devaddr;
892 
893     switch(type) {
894     case IF_IDE:
895     case IF_SCSI:
896     case IF_XEN:
897     case IF_NONE:
898         dinfo->media_cd = media == MEDIA_CDROM;
899         break;
900     default:
901         break;
902     }
903 
904 fail:
905     qemu_opts_del(legacy_opts);
906     return dinfo;
907 }
908 
909 void do_commit(Monitor *mon, const QDict *qdict)
910 {
911     const char *device = qdict_get_str(qdict, "device");
912     BlockDriverState *bs;
913     int ret;
914 
915     if (!strcmp(device, "all")) {
916         ret = bdrv_commit_all();
917     } else {
918         bs = bdrv_find(device);
919         if (!bs) {
920             monitor_printf(mon, "Device '%s' not found\n", device);
921             return;
922         }
923         ret = bdrv_commit(bs);
924     }
925     if (ret < 0) {
926         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
927                        strerror(-ret));
928     }
929 }
930 
931 static void blockdev_do_action(int kind, void *data, Error **errp)
932 {
933     TransactionAction action;
934     TransactionActionList list;
935 
936     action.kind = kind;
937     action.data = data;
938     list.value = &action;
939     list.next = NULL;
940     qmp_transaction(&list, errp);
941 }
942 
943 void qmp_blockdev_snapshot_sync(const char *device, const char *snapshot_file,
944                                 bool has_format, const char *format,
945                                 bool has_mode, enum NewImageMode mode,
946                                 Error **errp)
947 {
948     BlockdevSnapshot snapshot = {
949         .device = (char *) device,
950         .snapshot_file = (char *) snapshot_file,
951         .has_format = has_format,
952         .format = (char *) format,
953         .has_mode = has_mode,
954         .mode = mode,
955     };
956     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
957                        &snapshot, errp);
958 }
959 
960 void qmp_blockdev_snapshot_internal_sync(const char *device,
961                                          const char *name,
962                                          Error **errp)
963 {
964     BlockdevSnapshotInternal snapshot = {
965         .device = (char *) device,
966         .name = (char *) name
967     };
968 
969     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
970                        &snapshot, errp);
971 }
972 
973 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
974                                                          bool has_id,
975                                                          const char *id,
976                                                          bool has_name,
977                                                          const char *name,
978                                                          Error **errp)
979 {
980     BlockDriverState *bs = bdrv_find(device);
981     QEMUSnapshotInfo sn;
982     Error *local_err = NULL;
983     SnapshotInfo *info = NULL;
984     int ret;
985 
986     if (!bs) {
987         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
988         return NULL;
989     }
990 
991     if (!has_id) {
992         id = NULL;
993     }
994 
995     if (!has_name) {
996         name = NULL;
997     }
998 
999     if (!id && !name) {
1000         error_setg(errp, "Name or id must be provided");
1001         return NULL;
1002     }
1003 
1004     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1005     if (error_is_set(&local_err)) {
1006         error_propagate(errp, local_err);
1007         return NULL;
1008     }
1009     if (!ret) {
1010         error_setg(errp,
1011                    "Snapshot with id '%s' and name '%s' does not exist on "
1012                    "device '%s'",
1013                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1014         return NULL;
1015     }
1016 
1017     bdrv_snapshot_delete(bs, id, name, &local_err);
1018     if (error_is_set(&local_err)) {
1019         error_propagate(errp, local_err);
1020         return NULL;
1021     }
1022 
1023     info = g_malloc0(sizeof(SnapshotInfo));
1024     info->id = g_strdup(sn.id_str);
1025     info->name = g_strdup(sn.name);
1026     info->date_nsec = sn.date_nsec;
1027     info->date_sec = sn.date_sec;
1028     info->vm_state_size = sn.vm_state_size;
1029     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1030     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1031 
1032     return info;
1033 }
1034 
1035 /* New and old BlockDriverState structs for group snapshots */
1036 
1037 typedef struct BlkTransactionState BlkTransactionState;
1038 
1039 /* Only prepare() may fail. In a single transaction, only one of commit() or
1040    abort() will be called, clean() will always be called if it present. */
1041 typedef struct BdrvActionOps {
1042     /* Size of state struct, in bytes. */
1043     size_t instance_size;
1044     /* Prepare the work, must NOT be NULL. */
1045     void (*prepare)(BlkTransactionState *common, Error **errp);
1046     /* Commit the changes, can be NULL. */
1047     void (*commit)(BlkTransactionState *common);
1048     /* Abort the changes on fail, can be NULL. */
1049     void (*abort)(BlkTransactionState *common);
1050     /* Clean up resource in the end, can be NULL. */
1051     void (*clean)(BlkTransactionState *common);
1052 } BdrvActionOps;
1053 
1054 /*
1055  * This structure must be arranged as first member in child type, assuming
1056  * that compiler will also arrange it to the same address with parent instance.
1057  * Later it will be used in free().
1058  */
1059 struct BlkTransactionState {
1060     TransactionAction *action;
1061     const BdrvActionOps *ops;
1062     QSIMPLEQ_ENTRY(BlkTransactionState) entry;
1063 };
1064 
1065 /* internal snapshot private data */
1066 typedef struct InternalSnapshotState {
1067     BlkTransactionState common;
1068     BlockDriverState *bs;
1069     QEMUSnapshotInfo sn;
1070 } InternalSnapshotState;
1071 
1072 static void internal_snapshot_prepare(BlkTransactionState *common,
1073                                       Error **errp)
1074 {
1075     const char *device;
1076     const char *name;
1077     BlockDriverState *bs;
1078     QEMUSnapshotInfo old_sn, *sn;
1079     bool ret;
1080     qemu_timeval tv;
1081     BlockdevSnapshotInternal *internal;
1082     InternalSnapshotState *state;
1083     int ret1;
1084 
1085     g_assert(common->action->kind ==
1086              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1087     internal = common->action->blockdev_snapshot_internal_sync;
1088     state = DO_UPCAST(InternalSnapshotState, common, common);
1089 
1090     /* 1. parse input */
1091     device = internal->device;
1092     name = internal->name;
1093 
1094     /* 2. check for validation */
1095     bs = bdrv_find(device);
1096     if (!bs) {
1097         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1098         return;
1099     }
1100 
1101     if (!bdrv_is_inserted(bs)) {
1102         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1103         return;
1104     }
1105 
1106     if (bdrv_is_read_only(bs)) {
1107         error_set(errp, QERR_DEVICE_IS_READ_ONLY, device);
1108         return;
1109     }
1110 
1111     if (!bdrv_can_snapshot(bs)) {
1112         error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1113                   bs->drv->format_name, device, "internal snapshot");
1114         return;
1115     }
1116 
1117     if (!strlen(name)) {
1118         error_setg(errp, "Name is empty");
1119         return;
1120     }
1121 
1122     /* check whether a snapshot with name exist */
1123     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn, errp);
1124     if (error_is_set(errp)) {
1125         return;
1126     } else if (ret) {
1127         error_setg(errp,
1128                    "Snapshot with name '%s' already exists on device '%s'",
1129                    name, device);
1130         return;
1131     }
1132 
1133     /* 3. take the snapshot */
1134     sn = &state->sn;
1135     pstrcpy(sn->name, sizeof(sn->name), name);
1136     qemu_gettimeofday(&tv);
1137     sn->date_sec = tv.tv_sec;
1138     sn->date_nsec = tv.tv_usec * 1000;
1139     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1140 
1141     ret1 = bdrv_snapshot_create(bs, sn);
1142     if (ret1 < 0) {
1143         error_setg_errno(errp, -ret1,
1144                          "Failed to create snapshot '%s' on device '%s'",
1145                          name, device);
1146         return;
1147     }
1148 
1149     /* 4. succeed, mark a snapshot is created */
1150     state->bs = bs;
1151 }
1152 
1153 static void internal_snapshot_abort(BlkTransactionState *common)
1154 {
1155     InternalSnapshotState *state =
1156                              DO_UPCAST(InternalSnapshotState, common, common);
1157     BlockDriverState *bs = state->bs;
1158     QEMUSnapshotInfo *sn = &state->sn;
1159     Error *local_error = NULL;
1160 
1161     if (!bs) {
1162         return;
1163     }
1164 
1165     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1166         error_report("Failed to delete snapshot with id '%s' and name '%s' on "
1167                      "device '%s' in abort: %s",
1168                      sn->id_str,
1169                      sn->name,
1170                      bdrv_get_device_name(bs),
1171                      error_get_pretty(local_error));
1172         error_free(local_error);
1173     }
1174 }
1175 
1176 /* external snapshot private data */
1177 typedef struct ExternalSnapshotState {
1178     BlkTransactionState common;
1179     BlockDriverState *old_bs;
1180     BlockDriverState *new_bs;
1181 } ExternalSnapshotState;
1182 
1183 static void external_snapshot_prepare(BlkTransactionState *common,
1184                                       Error **errp)
1185 {
1186     BlockDriver *drv;
1187     int flags, ret;
1188     Error *local_err = NULL;
1189     const char *device;
1190     const char *new_image_file;
1191     const char *format = "qcow2";
1192     enum NewImageMode mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1193     ExternalSnapshotState *state =
1194                              DO_UPCAST(ExternalSnapshotState, common, common);
1195     TransactionAction *action = common->action;
1196 
1197     /* get parameters */
1198     g_assert(action->kind == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC);
1199 
1200     device = action->blockdev_snapshot_sync->device;
1201     new_image_file = action->blockdev_snapshot_sync->snapshot_file;
1202     if (action->blockdev_snapshot_sync->has_format) {
1203         format = action->blockdev_snapshot_sync->format;
1204     }
1205     if (action->blockdev_snapshot_sync->has_mode) {
1206         mode = action->blockdev_snapshot_sync->mode;
1207     }
1208 
1209     /* start processing */
1210     drv = bdrv_find_format(format);
1211     if (!drv) {
1212         error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1213         return;
1214     }
1215 
1216     state->old_bs = bdrv_find(device);
1217     if (!state->old_bs) {
1218         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1219         return;
1220     }
1221 
1222     if (!bdrv_is_inserted(state->old_bs)) {
1223         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1224         return;
1225     }
1226 
1227     if (bdrv_in_use(state->old_bs)) {
1228         error_set(errp, QERR_DEVICE_IN_USE, device);
1229         return;
1230     }
1231 
1232     if (!bdrv_is_read_only(state->old_bs)) {
1233         if (bdrv_flush(state->old_bs)) {
1234             error_set(errp, QERR_IO_ERROR);
1235             return;
1236         }
1237     }
1238 
1239     if (bdrv_check_ext_snapshot(state->old_bs) != EXT_SNAPSHOT_ALLOWED) {
1240         error_set(errp, QERR_FEATURE_DISABLED, "snapshot");
1241         return;
1242     }
1243 
1244     flags = state->old_bs->open_flags;
1245 
1246     /* create new image w/backing file */
1247     if (mode != NEW_IMAGE_MODE_EXISTING) {
1248         bdrv_img_create(new_image_file, format,
1249                         state->old_bs->filename,
1250                         state->old_bs->drv->format_name,
1251                         NULL, -1, flags, &local_err, false);
1252         if (error_is_set(&local_err)) {
1253             error_propagate(errp, local_err);
1254             return;
1255         }
1256     }
1257 
1258     /* We will manually add the backing_hd field to the bs later */
1259     state->new_bs = bdrv_new("");
1260     /* TODO Inherit bs->options or only take explicit options with an
1261      * extended QMP command? */
1262     ret = bdrv_open(state->new_bs, new_image_file, NULL,
1263                     flags | BDRV_O_NO_BACKING, drv, &local_err);
1264     if (ret != 0) {
1265         error_propagate(errp, local_err);
1266     }
1267 }
1268 
1269 static void external_snapshot_commit(BlkTransactionState *common)
1270 {
1271     ExternalSnapshotState *state =
1272                              DO_UPCAST(ExternalSnapshotState, common, common);
1273 
1274     /* This removes our old bs and adds the new bs */
1275     bdrv_append(state->new_bs, state->old_bs);
1276     /* We don't need (or want) to use the transactional
1277      * bdrv_reopen_multiple() across all the entries at once, because we
1278      * don't want to abort all of them if one of them fails the reopen */
1279     bdrv_reopen(state->new_bs, state->new_bs->open_flags & ~BDRV_O_RDWR,
1280                 NULL);
1281 }
1282 
1283 static void external_snapshot_abort(BlkTransactionState *common)
1284 {
1285     ExternalSnapshotState *state =
1286                              DO_UPCAST(ExternalSnapshotState, common, common);
1287     if (state->new_bs) {
1288         bdrv_unref(state->new_bs);
1289     }
1290 }
1291 
1292 typedef struct DriveBackupState {
1293     BlkTransactionState common;
1294     BlockDriverState *bs;
1295     BlockJob *job;
1296 } DriveBackupState;
1297 
1298 static void drive_backup_prepare(BlkTransactionState *common, Error **errp)
1299 {
1300     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1301     DriveBackup *backup;
1302     Error *local_err = NULL;
1303 
1304     assert(common->action->kind == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1305     backup = common->action->drive_backup;
1306 
1307     qmp_drive_backup(backup->device, backup->target,
1308                      backup->has_format, backup->format,
1309                      backup->sync,
1310                      backup->has_mode, backup->mode,
1311                      backup->has_speed, backup->speed,
1312                      backup->has_on_source_error, backup->on_source_error,
1313                      backup->has_on_target_error, backup->on_target_error,
1314                      &local_err);
1315     if (error_is_set(&local_err)) {
1316         error_propagate(errp, local_err);
1317         state->bs = NULL;
1318         state->job = NULL;
1319         return;
1320     }
1321 
1322     state->bs = bdrv_find(backup->device);
1323     state->job = state->bs->job;
1324 }
1325 
1326 static void drive_backup_abort(BlkTransactionState *common)
1327 {
1328     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1329     BlockDriverState *bs = state->bs;
1330 
1331     /* Only cancel if it's the job we started */
1332     if (bs && bs->job && bs->job == state->job) {
1333         block_job_cancel_sync(bs->job);
1334     }
1335 }
1336 
1337 static void abort_prepare(BlkTransactionState *common, Error **errp)
1338 {
1339     error_setg(errp, "Transaction aborted using Abort action");
1340 }
1341 
1342 static void abort_commit(BlkTransactionState *common)
1343 {
1344     g_assert_not_reached(); /* this action never succeeds */
1345 }
1346 
1347 static const BdrvActionOps actions[] = {
1348     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
1349         .instance_size = sizeof(ExternalSnapshotState),
1350         .prepare  = external_snapshot_prepare,
1351         .commit   = external_snapshot_commit,
1352         .abort = external_snapshot_abort,
1353     },
1354     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
1355         .instance_size = sizeof(DriveBackupState),
1356         .prepare = drive_backup_prepare,
1357         .abort = drive_backup_abort,
1358     },
1359     [TRANSACTION_ACTION_KIND_ABORT] = {
1360         .instance_size = sizeof(BlkTransactionState),
1361         .prepare = abort_prepare,
1362         .commit = abort_commit,
1363     },
1364     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
1365         .instance_size = sizeof(InternalSnapshotState),
1366         .prepare  = internal_snapshot_prepare,
1367         .abort = internal_snapshot_abort,
1368     },
1369 };
1370 
1371 /*
1372  * 'Atomic' group snapshots.  The snapshots are taken as a set, and if any fail
1373  *  then we do not pivot any of the devices in the group, and abandon the
1374  *  snapshots
1375  */
1376 void qmp_transaction(TransactionActionList *dev_list, Error **errp)
1377 {
1378     TransactionActionList *dev_entry = dev_list;
1379     BlkTransactionState *state, *next;
1380     Error *local_err = NULL;
1381 
1382     QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionState) snap_bdrv_states;
1383     QSIMPLEQ_INIT(&snap_bdrv_states);
1384 
1385     /* drain all i/o before any snapshots */
1386     bdrv_drain_all();
1387 
1388     /* We don't do anything in this loop that commits us to the snapshot */
1389     while (NULL != dev_entry) {
1390         TransactionAction *dev_info = NULL;
1391         const BdrvActionOps *ops;
1392 
1393         dev_info = dev_entry->value;
1394         dev_entry = dev_entry->next;
1395 
1396         assert(dev_info->kind < ARRAY_SIZE(actions));
1397 
1398         ops = &actions[dev_info->kind];
1399         assert(ops->instance_size > 0);
1400 
1401         state = g_malloc0(ops->instance_size);
1402         state->ops = ops;
1403         state->action = dev_info;
1404         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
1405 
1406         state->ops->prepare(state, &local_err);
1407         if (error_is_set(&local_err)) {
1408             error_propagate(errp, local_err);
1409             goto delete_and_fail;
1410         }
1411     }
1412 
1413     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1414         if (state->ops->commit) {
1415             state->ops->commit(state);
1416         }
1417     }
1418 
1419     /* success */
1420     goto exit;
1421 
1422 delete_and_fail:
1423     /*
1424     * failure, and it is all-or-none; abandon each new bs, and keep using
1425     * the original bs for all images
1426     */
1427     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1428         if (state->ops->abort) {
1429             state->ops->abort(state);
1430         }
1431     }
1432 exit:
1433     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
1434         if (state->ops->clean) {
1435             state->ops->clean(state);
1436         }
1437         g_free(state);
1438     }
1439 }
1440 
1441 
1442 static void eject_device(BlockDriverState *bs, int force, Error **errp)
1443 {
1444     if (bdrv_in_use(bs)) {
1445         error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
1446         return;
1447     }
1448     if (!bdrv_dev_has_removable_media(bs)) {
1449         error_set(errp, QERR_DEVICE_NOT_REMOVABLE, bdrv_get_device_name(bs));
1450         return;
1451     }
1452 
1453     if (bdrv_dev_is_medium_locked(bs) && !bdrv_dev_is_tray_open(bs)) {
1454         bdrv_dev_eject_request(bs, force);
1455         if (!force) {
1456             error_set(errp, QERR_DEVICE_LOCKED, bdrv_get_device_name(bs));
1457             return;
1458         }
1459     }
1460 
1461     bdrv_close(bs);
1462 }
1463 
1464 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
1465 {
1466     BlockDriverState *bs;
1467 
1468     bs = bdrv_find(device);
1469     if (!bs) {
1470         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1471         return;
1472     }
1473 
1474     eject_device(bs, force, errp);
1475 }
1476 
1477 void qmp_block_passwd(const char *device, const char *password, Error **errp)
1478 {
1479     BlockDriverState *bs;
1480     int err;
1481 
1482     bs = bdrv_find(device);
1483     if (!bs) {
1484         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1485         return;
1486     }
1487 
1488     err = bdrv_set_key(bs, password);
1489     if (err == -EINVAL) {
1490         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
1491         return;
1492     } else if (err < 0) {
1493         error_set(errp, QERR_INVALID_PASSWORD);
1494         return;
1495     }
1496 }
1497 
1498 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename,
1499                                     int bdrv_flags, BlockDriver *drv,
1500                                     const char *password, Error **errp)
1501 {
1502     Error *local_err = NULL;
1503     int ret;
1504 
1505     ret = bdrv_open(bs, filename, NULL, bdrv_flags, drv, &local_err);
1506     if (ret < 0) {
1507         error_propagate(errp, local_err);
1508         return;
1509     }
1510 
1511     if (bdrv_key_required(bs)) {
1512         if (password) {
1513             if (bdrv_set_key(bs, password) < 0) {
1514                 error_set(errp, QERR_INVALID_PASSWORD);
1515             }
1516         } else {
1517             error_set(errp, QERR_DEVICE_ENCRYPTED, bdrv_get_device_name(bs),
1518                       bdrv_get_encrypted_filename(bs));
1519         }
1520     } else if (password) {
1521         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
1522     }
1523 }
1524 
1525 void qmp_change_blockdev(const char *device, const char *filename,
1526                          bool has_format, const char *format, Error **errp)
1527 {
1528     BlockDriverState *bs;
1529     BlockDriver *drv = NULL;
1530     int bdrv_flags;
1531     Error *err = NULL;
1532 
1533     bs = bdrv_find(device);
1534     if (!bs) {
1535         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1536         return;
1537     }
1538 
1539     if (format) {
1540         drv = bdrv_find_whitelisted_format(format, bs->read_only);
1541         if (!drv) {
1542             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1543             return;
1544         }
1545     }
1546 
1547     eject_device(bs, 0, &err);
1548     if (error_is_set(&err)) {
1549         error_propagate(errp, err);
1550         return;
1551     }
1552 
1553     bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
1554     bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
1555 
1556     qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp);
1557 }
1558 
1559 /* throttling disk I/O limits */
1560 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
1561                                int64_t bps_wr,
1562                                int64_t iops,
1563                                int64_t iops_rd,
1564                                int64_t iops_wr,
1565                                bool has_bps_max,
1566                                int64_t bps_max,
1567                                bool has_bps_rd_max,
1568                                int64_t bps_rd_max,
1569                                bool has_bps_wr_max,
1570                                int64_t bps_wr_max,
1571                                bool has_iops_max,
1572                                int64_t iops_max,
1573                                bool has_iops_rd_max,
1574                                int64_t iops_rd_max,
1575                                bool has_iops_wr_max,
1576                                int64_t iops_wr_max,
1577                                bool has_iops_size,
1578                                int64_t iops_size, Error **errp)
1579 {
1580     ThrottleConfig cfg;
1581     BlockDriverState *bs;
1582 
1583     bs = bdrv_find(device);
1584     if (!bs) {
1585         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1586         return;
1587     }
1588 
1589     memset(&cfg, 0, sizeof(cfg));
1590     cfg.buckets[THROTTLE_BPS_TOTAL].avg = bps;
1591     cfg.buckets[THROTTLE_BPS_READ].avg  = bps_rd;
1592     cfg.buckets[THROTTLE_BPS_WRITE].avg = bps_wr;
1593 
1594     cfg.buckets[THROTTLE_OPS_TOTAL].avg = iops;
1595     cfg.buckets[THROTTLE_OPS_READ].avg  = iops_rd;
1596     cfg.buckets[THROTTLE_OPS_WRITE].avg = iops_wr;
1597 
1598     if (has_bps_max) {
1599         cfg.buckets[THROTTLE_BPS_TOTAL].max = bps_max;
1600     }
1601     if (has_bps_rd_max) {
1602         cfg.buckets[THROTTLE_BPS_READ].max = bps_rd_max;
1603     }
1604     if (has_bps_wr_max) {
1605         cfg.buckets[THROTTLE_BPS_WRITE].max = bps_wr_max;
1606     }
1607     if (has_iops_max) {
1608         cfg.buckets[THROTTLE_OPS_TOTAL].max = iops_max;
1609     }
1610     if (has_iops_rd_max) {
1611         cfg.buckets[THROTTLE_OPS_READ].max = iops_rd_max;
1612     }
1613     if (has_iops_wr_max) {
1614         cfg.buckets[THROTTLE_OPS_WRITE].max = iops_wr_max;
1615     }
1616 
1617     if (has_iops_size) {
1618         cfg.op_size = iops_size;
1619     }
1620 
1621     if (!check_throttle_config(&cfg, errp)) {
1622         return;
1623     }
1624 
1625     if (!bs->io_limits_enabled && throttle_enabled(&cfg)) {
1626         bdrv_io_limits_enable(bs);
1627     } else if (bs->io_limits_enabled && !throttle_enabled(&cfg)) {
1628         bdrv_io_limits_disable(bs);
1629     }
1630 
1631     if (bs->io_limits_enabled) {
1632         bdrv_set_io_limits(bs, &cfg);
1633     }
1634 }
1635 
1636 int do_drive_del(Monitor *mon, const QDict *qdict, QObject **ret_data)
1637 {
1638     const char *id = qdict_get_str(qdict, "id");
1639     BlockDriverState *bs;
1640 
1641     bs = bdrv_find(id);
1642     if (!bs) {
1643         qerror_report(QERR_DEVICE_NOT_FOUND, id);
1644         return -1;
1645     }
1646     if (bdrv_in_use(bs)) {
1647         qerror_report(QERR_DEVICE_IN_USE, id);
1648         return -1;
1649     }
1650 
1651     /* quiesce block driver; prevent further io */
1652     bdrv_drain_all();
1653     bdrv_flush(bs);
1654     bdrv_close(bs);
1655 
1656     /* if we have a device attached to this BlockDriverState
1657      * then we need to make the drive anonymous until the device
1658      * can be removed.  If this is a drive with no device backing
1659      * then we can just get rid of the block driver state right here.
1660      */
1661     if (bdrv_get_attached_dev(bs)) {
1662         bdrv_make_anon(bs);
1663 
1664         /* Further I/O must not pause the guest */
1665         bdrv_set_on_error(bs, BLOCKDEV_ON_ERROR_REPORT,
1666                           BLOCKDEV_ON_ERROR_REPORT);
1667     } else {
1668         drive_uninit(drive_get_by_blockdev(bs));
1669     }
1670 
1671     return 0;
1672 }
1673 
1674 void qmp_block_resize(const char *device, int64_t size, Error **errp)
1675 {
1676     BlockDriverState *bs;
1677     int ret;
1678 
1679     bs = bdrv_find(device);
1680     if (!bs) {
1681         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1682         return;
1683     }
1684 
1685     if (size < 0) {
1686         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
1687         return;
1688     }
1689 
1690     /* complete all in-flight operations before resizing the device */
1691     bdrv_drain_all();
1692 
1693     ret = bdrv_truncate(bs, size);
1694     switch (ret) {
1695     case 0:
1696         break;
1697     case -ENOMEDIUM:
1698         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1699         break;
1700     case -ENOTSUP:
1701         error_set(errp, QERR_UNSUPPORTED);
1702         break;
1703     case -EACCES:
1704         error_set(errp, QERR_DEVICE_IS_READ_ONLY, device);
1705         break;
1706     case -EBUSY:
1707         error_set(errp, QERR_DEVICE_IN_USE, device);
1708         break;
1709     default:
1710         error_setg_errno(errp, -ret, "Could not resize");
1711         break;
1712     }
1713 }
1714 
1715 static void block_job_cb(void *opaque, int ret)
1716 {
1717     BlockDriverState *bs = opaque;
1718     QObject *obj;
1719 
1720     trace_block_job_cb(bs, bs->job, ret);
1721 
1722     assert(bs->job);
1723     obj = qobject_from_block_job(bs->job);
1724     if (ret < 0) {
1725         QDict *dict = qobject_to_qdict(obj);
1726         qdict_put(dict, "error", qstring_from_str(strerror(-ret)));
1727     }
1728 
1729     if (block_job_is_cancelled(bs->job)) {
1730         monitor_protocol_event(QEVENT_BLOCK_JOB_CANCELLED, obj);
1731     } else {
1732         monitor_protocol_event(QEVENT_BLOCK_JOB_COMPLETED, obj);
1733     }
1734     qobject_decref(obj);
1735 
1736     bdrv_put_ref_bh_schedule(bs);
1737 }
1738 
1739 void qmp_block_stream(const char *device, bool has_base,
1740                       const char *base, bool has_speed, int64_t speed,
1741                       bool has_on_error, BlockdevOnError on_error,
1742                       Error **errp)
1743 {
1744     BlockDriverState *bs;
1745     BlockDriverState *base_bs = NULL;
1746     Error *local_err = NULL;
1747 
1748     if (!has_on_error) {
1749         on_error = BLOCKDEV_ON_ERROR_REPORT;
1750     }
1751 
1752     bs = bdrv_find(device);
1753     if (!bs) {
1754         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1755         return;
1756     }
1757 
1758     if (base) {
1759         base_bs = bdrv_find_backing_image(bs, base);
1760         if (base_bs == NULL) {
1761             error_set(errp, QERR_BASE_NOT_FOUND, base);
1762             return;
1763         }
1764     }
1765 
1766     stream_start(bs, base_bs, base, has_speed ? speed : 0,
1767                  on_error, block_job_cb, bs, &local_err);
1768     if (error_is_set(&local_err)) {
1769         error_propagate(errp, local_err);
1770         return;
1771     }
1772 
1773     trace_qmp_block_stream(bs, bs->job);
1774 }
1775 
1776 void qmp_block_commit(const char *device,
1777                       bool has_base, const char *base, const char *top,
1778                       bool has_speed, int64_t speed,
1779                       Error **errp)
1780 {
1781     BlockDriverState *bs;
1782     BlockDriverState *base_bs, *top_bs;
1783     Error *local_err = NULL;
1784     /* This will be part of the QMP command, if/when the
1785      * BlockdevOnError change for blkmirror makes it in
1786      */
1787     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
1788 
1789     /* drain all i/o before commits */
1790     bdrv_drain_all();
1791 
1792     bs = bdrv_find(device);
1793     if (!bs) {
1794         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1795         return;
1796     }
1797 
1798     /* default top_bs is the active layer */
1799     top_bs = bs;
1800 
1801     if (top) {
1802         if (strcmp(bs->filename, top) != 0) {
1803             top_bs = bdrv_find_backing_image(bs, top);
1804         }
1805     }
1806 
1807     if (top_bs == NULL) {
1808         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
1809         return;
1810     }
1811 
1812     if (has_base && base) {
1813         base_bs = bdrv_find_backing_image(top_bs, base);
1814     } else {
1815         base_bs = bdrv_find_base(top_bs);
1816     }
1817 
1818     if (base_bs == NULL) {
1819         error_set(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
1820         return;
1821     }
1822 
1823     commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
1824                 &local_err);
1825     if (local_err != NULL) {
1826         error_propagate(errp, local_err);
1827         return;
1828     }
1829 }
1830 
1831 void qmp_drive_backup(const char *device, const char *target,
1832                       bool has_format, const char *format,
1833                       enum MirrorSyncMode sync,
1834                       bool has_mode, enum NewImageMode mode,
1835                       bool has_speed, int64_t speed,
1836                       bool has_on_source_error, BlockdevOnError on_source_error,
1837                       bool has_on_target_error, BlockdevOnError on_target_error,
1838                       Error **errp)
1839 {
1840     BlockDriverState *bs;
1841     BlockDriverState *target_bs;
1842     BlockDriverState *source = NULL;
1843     BlockDriver *drv = NULL;
1844     Error *local_err = NULL;
1845     int flags;
1846     int64_t size;
1847     int ret;
1848 
1849     if (!has_speed) {
1850         speed = 0;
1851     }
1852     if (!has_on_source_error) {
1853         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
1854     }
1855     if (!has_on_target_error) {
1856         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
1857     }
1858     if (!has_mode) {
1859         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1860     }
1861 
1862     bs = bdrv_find(device);
1863     if (!bs) {
1864         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1865         return;
1866     }
1867 
1868     if (!bdrv_is_inserted(bs)) {
1869         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1870         return;
1871     }
1872 
1873     if (!has_format) {
1874         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
1875     }
1876     if (format) {
1877         drv = bdrv_find_format(format);
1878         if (!drv) {
1879             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1880             return;
1881         }
1882     }
1883 
1884     if (bdrv_in_use(bs)) {
1885         error_set(errp, QERR_DEVICE_IN_USE, device);
1886         return;
1887     }
1888 
1889     flags = bs->open_flags | BDRV_O_RDWR;
1890 
1891     /* See if we have a backing HD we can use to create our new image
1892      * on top of. */
1893     if (sync == MIRROR_SYNC_MODE_TOP) {
1894         source = bs->backing_hd;
1895         if (!source) {
1896             sync = MIRROR_SYNC_MODE_FULL;
1897         }
1898     }
1899     if (sync == MIRROR_SYNC_MODE_NONE) {
1900         source = bs;
1901     }
1902 
1903     size = bdrv_getlength(bs);
1904     if (size < 0) {
1905         error_setg_errno(errp, -size, "bdrv_getlength failed");
1906         return;
1907     }
1908 
1909     if (mode != NEW_IMAGE_MODE_EXISTING) {
1910         assert(format && drv);
1911         if (source) {
1912             bdrv_img_create(target, format, source->filename,
1913                             source->drv->format_name, NULL,
1914                             size, flags, &local_err, false);
1915         } else {
1916             bdrv_img_create(target, format, NULL, NULL, NULL,
1917                             size, flags, &local_err, false);
1918         }
1919     }
1920 
1921     if (error_is_set(&local_err)) {
1922         error_propagate(errp, local_err);
1923         return;
1924     }
1925 
1926     target_bs = bdrv_new("");
1927     ret = bdrv_open(target_bs, target, NULL, flags, drv, &local_err);
1928     if (ret < 0) {
1929         bdrv_unref(target_bs);
1930         error_propagate(errp, local_err);
1931         return;
1932     }
1933 
1934     backup_start(bs, target_bs, speed, sync, on_source_error, on_target_error,
1935                  block_job_cb, bs, &local_err);
1936     if (local_err != NULL) {
1937         bdrv_unref(target_bs);
1938         error_propagate(errp, local_err);
1939         return;
1940     }
1941 }
1942 
1943 #define DEFAULT_MIRROR_BUF_SIZE   (10 << 20)
1944 
1945 void qmp_drive_mirror(const char *device, const char *target,
1946                       bool has_format, const char *format,
1947                       enum MirrorSyncMode sync,
1948                       bool has_mode, enum NewImageMode mode,
1949                       bool has_speed, int64_t speed,
1950                       bool has_granularity, uint32_t granularity,
1951                       bool has_buf_size, int64_t buf_size,
1952                       bool has_on_source_error, BlockdevOnError on_source_error,
1953                       bool has_on_target_error, BlockdevOnError on_target_error,
1954                       Error **errp)
1955 {
1956     BlockDriverState *bs;
1957     BlockDriverState *source, *target_bs;
1958     BlockDriver *drv = NULL;
1959     Error *local_err = NULL;
1960     int flags;
1961     int64_t size;
1962     int ret;
1963 
1964     if (!has_speed) {
1965         speed = 0;
1966     }
1967     if (!has_on_source_error) {
1968         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
1969     }
1970     if (!has_on_target_error) {
1971         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
1972     }
1973     if (!has_mode) {
1974         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1975     }
1976     if (!has_granularity) {
1977         granularity = 0;
1978     }
1979     if (!has_buf_size) {
1980         buf_size = DEFAULT_MIRROR_BUF_SIZE;
1981     }
1982 
1983     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
1984         error_set(errp, QERR_INVALID_PARAMETER, device);
1985         return;
1986     }
1987     if (granularity & (granularity - 1)) {
1988         error_set(errp, QERR_INVALID_PARAMETER, device);
1989         return;
1990     }
1991 
1992     bs = bdrv_find(device);
1993     if (!bs) {
1994         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1995         return;
1996     }
1997 
1998     if (!bdrv_is_inserted(bs)) {
1999         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2000         return;
2001     }
2002 
2003     if (!has_format) {
2004         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
2005     }
2006     if (format) {
2007         drv = bdrv_find_format(format);
2008         if (!drv) {
2009             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
2010             return;
2011         }
2012     }
2013 
2014     if (bdrv_in_use(bs)) {
2015         error_set(errp, QERR_DEVICE_IN_USE, device);
2016         return;
2017     }
2018 
2019     flags = bs->open_flags | BDRV_O_RDWR;
2020     source = bs->backing_hd;
2021     if (!source && sync == MIRROR_SYNC_MODE_TOP) {
2022         sync = MIRROR_SYNC_MODE_FULL;
2023     }
2024     if (sync == MIRROR_SYNC_MODE_NONE) {
2025         source = bs;
2026     }
2027 
2028     size = bdrv_getlength(bs);
2029     if (size < 0) {
2030         error_setg_errno(errp, -size, "bdrv_getlength failed");
2031         return;
2032     }
2033 
2034     if ((sync == MIRROR_SYNC_MODE_FULL || !source)
2035         && mode != NEW_IMAGE_MODE_EXISTING)
2036     {
2037         /* create new image w/o backing file */
2038         assert(format && drv);
2039         bdrv_img_create(target, format,
2040                         NULL, NULL, NULL, size, flags, &local_err, false);
2041     } else {
2042         switch (mode) {
2043         case NEW_IMAGE_MODE_EXISTING:
2044             break;
2045         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
2046             /* create new image with backing file */
2047             bdrv_img_create(target, format,
2048                             source->filename,
2049                             source->drv->format_name,
2050                             NULL, size, flags, &local_err, false);
2051             break;
2052         default:
2053             abort();
2054         }
2055     }
2056 
2057     if (error_is_set(&local_err)) {
2058         error_propagate(errp, local_err);
2059         return;
2060     }
2061 
2062     /* Mirroring takes care of copy-on-write using the source's backing
2063      * file.
2064      */
2065     target_bs = bdrv_new("");
2066     ret = bdrv_open(target_bs, target, NULL, flags | BDRV_O_NO_BACKING, drv,
2067                     &local_err);
2068     if (ret < 0) {
2069         bdrv_unref(target_bs);
2070         error_propagate(errp, local_err);
2071         return;
2072     }
2073 
2074     mirror_start(bs, target_bs, speed, granularity, buf_size, sync,
2075                  on_source_error, on_target_error,
2076                  block_job_cb, bs, &local_err);
2077     if (local_err != NULL) {
2078         bdrv_unref(target_bs);
2079         error_propagate(errp, local_err);
2080         return;
2081     }
2082 }
2083 
2084 static BlockJob *find_block_job(const char *device)
2085 {
2086     BlockDriverState *bs;
2087 
2088     bs = bdrv_find(device);
2089     if (!bs || !bs->job) {
2090         return NULL;
2091     }
2092     return bs->job;
2093 }
2094 
2095 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
2096 {
2097     BlockJob *job = find_block_job(device);
2098 
2099     if (!job) {
2100         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2101         return;
2102     }
2103 
2104     block_job_set_speed(job, speed, errp);
2105 }
2106 
2107 void qmp_block_job_cancel(const char *device,
2108                           bool has_force, bool force, Error **errp)
2109 {
2110     BlockJob *job = find_block_job(device);
2111 
2112     if (!has_force) {
2113         force = false;
2114     }
2115 
2116     if (!job) {
2117         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2118         return;
2119     }
2120     if (job->paused && !force) {
2121         error_set(errp, QERR_BLOCK_JOB_PAUSED, device);
2122         return;
2123     }
2124 
2125     trace_qmp_block_job_cancel(job);
2126     block_job_cancel(job);
2127 }
2128 
2129 void qmp_block_job_pause(const char *device, Error **errp)
2130 {
2131     BlockJob *job = find_block_job(device);
2132 
2133     if (!job) {
2134         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2135         return;
2136     }
2137 
2138     trace_qmp_block_job_pause(job);
2139     block_job_pause(job);
2140 }
2141 
2142 void qmp_block_job_resume(const char *device, Error **errp)
2143 {
2144     BlockJob *job = find_block_job(device);
2145 
2146     if (!job) {
2147         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2148         return;
2149     }
2150 
2151     trace_qmp_block_job_resume(job);
2152     block_job_resume(job);
2153 }
2154 
2155 void qmp_block_job_complete(const char *device, Error **errp)
2156 {
2157     BlockJob *job = find_block_job(device);
2158 
2159     if (!job) {
2160         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2161         return;
2162     }
2163 
2164     trace_qmp_block_job_complete(job);
2165     block_job_complete(job, errp);
2166 }
2167 
2168 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
2169 {
2170     QmpOutputVisitor *ov = qmp_output_visitor_new();
2171     QObject *obj;
2172     QDict *qdict;
2173     Error *local_err = NULL;
2174 
2175     /* Require an ID in the top level */
2176     if (!options->has_id) {
2177         error_setg(errp, "Block device needs an ID");
2178         goto fail;
2179     }
2180 
2181     /* TODO Sort it out in raw-posix and drive_init: Reject aio=native with
2182      * cache.direct=false instead of silently switching to aio=threads, except
2183      * if called from drive_init.
2184      *
2185      * For now, simply forbidding the combination for all drivers will do. */
2186     if (options->has_aio && options->aio == BLOCKDEV_AIO_OPTIONS_NATIVE) {
2187         bool direct = options->cache->has_direct && options->cache->direct;
2188         if (!options->has_cache && !direct) {
2189             error_setg(errp, "aio=native requires cache.direct=true");
2190             goto fail;
2191         }
2192     }
2193 
2194     visit_type_BlockdevOptions(qmp_output_get_visitor(ov),
2195                                &options, NULL, &local_err);
2196     if (error_is_set(&local_err)) {
2197         error_propagate(errp, local_err);
2198         goto fail;
2199     }
2200 
2201     obj = qmp_output_get_qobject(ov);
2202     qdict = qobject_to_qdict(obj);
2203 
2204     qdict_flatten(qdict);
2205 
2206     blockdev_init(qdict, IF_NONE, &local_err);
2207     if (error_is_set(&local_err)) {
2208         error_propagate(errp, local_err);
2209         goto fail;
2210     }
2211 
2212 fail:
2213     qmp_output_visitor_cleanup(ov);
2214 }
2215 
2216 static void do_qmp_query_block_jobs_one(void *opaque, BlockDriverState *bs)
2217 {
2218     BlockJobInfoList **prev = opaque;
2219     BlockJob *job = bs->job;
2220 
2221     if (job) {
2222         BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
2223         elem->value = block_job_query(bs->job);
2224         (*prev)->next = elem;
2225         *prev = elem;
2226     }
2227 }
2228 
2229 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
2230 {
2231     /* Dummy is a fake list element for holding the head pointer */
2232     BlockJobInfoList dummy = {};
2233     BlockJobInfoList *prev = &dummy;
2234     bdrv_iterate(do_qmp_query_block_jobs_one, &prev);
2235     return dummy.next;
2236 }
2237 
2238 QemuOptsList qemu_common_drive_opts = {
2239     .name = "drive",
2240     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
2241     .desc = {
2242         {
2243             .name = "snapshot",
2244             .type = QEMU_OPT_BOOL,
2245             .help = "enable/disable snapshot mode",
2246         },{
2247             .name = "file",
2248             .type = QEMU_OPT_STRING,
2249             .help = "disk image",
2250         },{
2251             .name = "discard",
2252             .type = QEMU_OPT_STRING,
2253             .help = "discard operation (ignore/off, unmap/on)",
2254         },{
2255             .name = "cache.writeback",
2256             .type = QEMU_OPT_BOOL,
2257             .help = "enables writeback mode for any caches",
2258         },{
2259             .name = "cache.direct",
2260             .type = QEMU_OPT_BOOL,
2261             .help = "enables use of O_DIRECT (bypass the host page cache)",
2262         },{
2263             .name = "cache.no-flush",
2264             .type = QEMU_OPT_BOOL,
2265             .help = "ignore any flush requests for the device",
2266         },{
2267             .name = "aio",
2268             .type = QEMU_OPT_STRING,
2269             .help = "host AIO implementation (threads, native)",
2270         },{
2271             .name = "format",
2272             .type = QEMU_OPT_STRING,
2273             .help = "disk format (raw, qcow2, ...)",
2274         },{
2275             .name = "serial",
2276             .type = QEMU_OPT_STRING,
2277             .help = "disk serial number",
2278         },{
2279             .name = "rerror",
2280             .type = QEMU_OPT_STRING,
2281             .help = "read error action",
2282         },{
2283             .name = "werror",
2284             .type = QEMU_OPT_STRING,
2285             .help = "write error action",
2286         },{
2287             .name = "read-only",
2288             .type = QEMU_OPT_BOOL,
2289             .help = "open drive file as read-only",
2290         },{
2291             .name = "throttling.iops-total",
2292             .type = QEMU_OPT_NUMBER,
2293             .help = "limit total I/O operations per second",
2294         },{
2295             .name = "throttling.iops-read",
2296             .type = QEMU_OPT_NUMBER,
2297             .help = "limit read operations per second",
2298         },{
2299             .name = "throttling.iops-write",
2300             .type = QEMU_OPT_NUMBER,
2301             .help = "limit write operations per second",
2302         },{
2303             .name = "throttling.bps-total",
2304             .type = QEMU_OPT_NUMBER,
2305             .help = "limit total bytes per second",
2306         },{
2307             .name = "throttling.bps-read",
2308             .type = QEMU_OPT_NUMBER,
2309             .help = "limit read bytes per second",
2310         },{
2311             .name = "throttling.bps-write",
2312             .type = QEMU_OPT_NUMBER,
2313             .help = "limit write bytes per second",
2314         },{
2315             .name = "throttling.iops-total-max",
2316             .type = QEMU_OPT_NUMBER,
2317             .help = "I/O operations burst",
2318         },{
2319             .name = "throttling.iops-read-max",
2320             .type = QEMU_OPT_NUMBER,
2321             .help = "I/O operations read burst",
2322         },{
2323             .name = "throttling.iops-write-max",
2324             .type = QEMU_OPT_NUMBER,
2325             .help = "I/O operations write burst",
2326         },{
2327             .name = "throttling.bps-total-max",
2328             .type = QEMU_OPT_NUMBER,
2329             .help = "total bytes burst",
2330         },{
2331             .name = "throttling.bps-read-max",
2332             .type = QEMU_OPT_NUMBER,
2333             .help = "total bytes read burst",
2334         },{
2335             .name = "throttling.bps-write-max",
2336             .type = QEMU_OPT_NUMBER,
2337             .help = "total bytes write burst",
2338         },{
2339             .name = "throttling.iops-size",
2340             .type = QEMU_OPT_NUMBER,
2341             .help = "when limiting by iops max size of an I/O in bytes",
2342         },{
2343             .name = "copy-on-read",
2344             .type = QEMU_OPT_BOOL,
2345             .help = "copy read data from backing file into image file",
2346         },
2347         { /* end of list */ }
2348     },
2349 };
2350 
2351 QemuOptsList qemu_drive_opts = {
2352     .name = "drive",
2353     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
2354     .desc = {
2355         /*
2356          * no elements => accept any params
2357          * validation will happen later
2358          */
2359         { /* end of list */ }
2360     },
2361 };
2362