xref: /openbmc/qemu/blockdev.c (revision 87ea75d5e135c0527c6a9dbac4317913409f28c7)
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(&qemu_legacy_drive_opts, NULL, 0,
686                                    &error_abort);
687     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
688     if (error_is_set(&local_err)) {
689         qerror_report_err(local_err);
690         error_free(local_err);
691         goto fail;
692     }
693 
694     /* Deprecated option boot=[on|off] */
695     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
696         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
697                 "ignored. Future versions will reject this parameter. Please "
698                 "update your scripts.\n");
699     }
700 
701     /* Media type */
702     value = qemu_opt_get(legacy_opts, "media");
703     if (value) {
704         if (!strcmp(value, "disk")) {
705             media = MEDIA_DISK;
706         } else if (!strcmp(value, "cdrom")) {
707             media = MEDIA_CDROM;
708             read_only = true;
709         } else {
710             error_report("'%s' invalid media", value);
711             goto fail;
712         }
713     }
714 
715     /* copy-on-read is disabled with a warning for read-only devices */
716     read_only |= qemu_opt_get_bool(legacy_opts, "read-only", false);
717     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
718 
719     if (read_only && copy_on_read) {
720         error_report("warning: disabling copy-on-read on read-only drive");
721         copy_on_read = false;
722     }
723 
724     qdict_put(bs_opts, "read-only",
725               qstring_from_str(read_only ? "on" : "off"));
726     qdict_put(bs_opts, "copy-on-read",
727               qstring_from_str(copy_on_read ? "on" :"off"));
728 
729     /* Controller type */
730     value = qemu_opt_get(legacy_opts, "if");
731     if (value) {
732         for (type = 0;
733              type < IF_COUNT && strcmp(value, if_name[type]);
734              type++) {
735         }
736         if (type == IF_COUNT) {
737             error_report("unsupported bus type '%s'", value);
738             goto fail;
739         }
740     } else {
741         type = block_default_type;
742     }
743 
744     /* Geometry */
745     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
746     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
747     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
748 
749     if (cyls || heads || secs) {
750         if (cyls < 1) {
751             error_report("invalid physical cyls number");
752             goto fail;
753         }
754         if (heads < 1) {
755             error_report("invalid physical heads number");
756             goto fail;
757         }
758         if (secs < 1) {
759             error_report("invalid physical secs number");
760             goto fail;
761         }
762     }
763 
764     translation = BIOS_ATA_TRANSLATION_AUTO;
765     value = qemu_opt_get(legacy_opts, "trans");
766     if (value != NULL) {
767         if (!cyls) {
768             error_report("'%s' trans must be used with cyls, heads and secs",
769                          value);
770             goto fail;
771         }
772         if (!strcmp(value, "none")) {
773             translation = BIOS_ATA_TRANSLATION_NONE;
774         } else if (!strcmp(value, "lba")) {
775             translation = BIOS_ATA_TRANSLATION_LBA;
776         } else if (!strcmp(value, "auto")) {
777             translation = BIOS_ATA_TRANSLATION_AUTO;
778         } else {
779             error_report("'%s' invalid translation type", value);
780             goto fail;
781         }
782     }
783 
784     if (media == MEDIA_CDROM) {
785         if (cyls || secs || heads) {
786             error_report("CHS can't be set with media=cdrom");
787             goto fail;
788         }
789     }
790 
791     /* Device address specified by bus/unit or index.
792      * If none was specified, try to find the first free one. */
793     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
794     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
795     index   = qemu_opt_get_number(legacy_opts, "index", -1);
796 
797     max_devs = if_max_devs[type];
798 
799     if (index != -1) {
800         if (bus_id != 0 || unit_id != -1) {
801             error_report("index cannot be used with bus and unit");
802             goto fail;
803         }
804         bus_id = drive_index_to_bus_id(type, index);
805         unit_id = drive_index_to_unit_id(type, index);
806     }
807 
808     if (unit_id == -1) {
809        unit_id = 0;
810        while (drive_get(type, bus_id, unit_id) != NULL) {
811            unit_id++;
812            if (max_devs && unit_id >= max_devs) {
813                unit_id -= max_devs;
814                bus_id++;
815            }
816        }
817     }
818 
819     if (max_devs && unit_id >= max_devs) {
820         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
821         goto fail;
822     }
823 
824     if (drive_get(type, bus_id, unit_id) != NULL) {
825         error_report("drive with bus=%d, unit=%d (index=%d) exists",
826                      bus_id, unit_id, index);
827         goto fail;
828     }
829 
830     /* no id supplied -> create one */
831     if (qemu_opts_id(all_opts) == NULL) {
832         char *new_id;
833         const char *mediastr = "";
834         if (type == IF_IDE || type == IF_SCSI) {
835             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
836         }
837         if (max_devs) {
838             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
839                                      mediastr, unit_id);
840         } else {
841             new_id = g_strdup_printf("%s%s%i", if_name[type],
842                                      mediastr, unit_id);
843         }
844         qdict_put(bs_opts, "id", qstring_from_str(new_id));
845         g_free(new_id);
846     }
847 
848     /* Add virtio block device */
849     devaddr = qemu_opt_get(legacy_opts, "addr");
850     if (devaddr && type != IF_VIRTIO) {
851         error_report("addr is not supported by this bus type");
852         goto fail;
853     }
854 
855     if (type == IF_VIRTIO) {
856         QemuOpts *devopts;
857         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
858                                    &error_abort);
859         if (arch_type == QEMU_ARCH_S390X) {
860             qemu_opt_set(devopts, "driver", "virtio-blk-s390");
861         } else {
862             qemu_opt_set(devopts, "driver", "virtio-blk-pci");
863         }
864         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"));
865         if (devaddr) {
866             qemu_opt_set(devopts, "addr", devaddr);
867         }
868     }
869 
870     /* Actual block device init: Functionality shared with blockdev-add */
871     dinfo = blockdev_init(bs_opts, type, &local_err);
872     if (dinfo == NULL) {
873         if (error_is_set(&local_err)) {
874             qerror_report_err(local_err);
875             error_free(local_err);
876         }
877         goto fail;
878     } else {
879         assert(!error_is_set(&local_err));
880     }
881 
882     /* Set legacy DriveInfo fields */
883     dinfo->enable_auto_del = true;
884     dinfo->opts = all_opts;
885 
886     dinfo->cyls = cyls;
887     dinfo->heads = heads;
888     dinfo->secs = secs;
889     dinfo->trans = translation;
890 
891     dinfo->bus = bus_id;
892     dinfo->unit = unit_id;
893     dinfo->devaddr = devaddr;
894 
895     switch(type) {
896     case IF_IDE:
897     case IF_SCSI:
898     case IF_XEN:
899     case IF_NONE:
900         dinfo->media_cd = media == MEDIA_CDROM;
901         break;
902     default:
903         break;
904     }
905 
906 fail:
907     qemu_opts_del(legacy_opts);
908     return dinfo;
909 }
910 
911 void do_commit(Monitor *mon, const QDict *qdict)
912 {
913     const char *device = qdict_get_str(qdict, "device");
914     BlockDriverState *bs;
915     int ret;
916 
917     if (!strcmp(device, "all")) {
918         ret = bdrv_commit_all();
919     } else {
920         bs = bdrv_find(device);
921         if (!bs) {
922             monitor_printf(mon, "Device '%s' not found\n", device);
923             return;
924         }
925         ret = bdrv_commit(bs);
926     }
927     if (ret < 0) {
928         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
929                        strerror(-ret));
930     }
931 }
932 
933 static void blockdev_do_action(int kind, void *data, Error **errp)
934 {
935     TransactionAction action;
936     TransactionActionList list;
937 
938     action.kind = kind;
939     action.data = data;
940     list.value = &action;
941     list.next = NULL;
942     qmp_transaction(&list, errp);
943 }
944 
945 void qmp_blockdev_snapshot_sync(const char *device, const char *snapshot_file,
946                                 bool has_format, const char *format,
947                                 bool has_mode, enum NewImageMode mode,
948                                 Error **errp)
949 {
950     BlockdevSnapshot snapshot = {
951         .device = (char *) device,
952         .snapshot_file = (char *) snapshot_file,
953         .has_format = has_format,
954         .format = (char *) format,
955         .has_mode = has_mode,
956         .mode = mode,
957     };
958     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
959                        &snapshot, errp);
960 }
961 
962 void qmp_blockdev_snapshot_internal_sync(const char *device,
963                                          const char *name,
964                                          Error **errp)
965 {
966     BlockdevSnapshotInternal snapshot = {
967         .device = (char *) device,
968         .name = (char *) name
969     };
970 
971     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
972                        &snapshot, errp);
973 }
974 
975 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
976                                                          bool has_id,
977                                                          const char *id,
978                                                          bool has_name,
979                                                          const char *name,
980                                                          Error **errp)
981 {
982     BlockDriverState *bs = bdrv_find(device);
983     QEMUSnapshotInfo sn;
984     Error *local_err = NULL;
985     SnapshotInfo *info = NULL;
986     int ret;
987 
988     if (!bs) {
989         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
990         return NULL;
991     }
992 
993     if (!has_id) {
994         id = NULL;
995     }
996 
997     if (!has_name) {
998         name = NULL;
999     }
1000 
1001     if (!id && !name) {
1002         error_setg(errp, "Name or id must be provided");
1003         return NULL;
1004     }
1005 
1006     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1007     if (error_is_set(&local_err)) {
1008         error_propagate(errp, local_err);
1009         return NULL;
1010     }
1011     if (!ret) {
1012         error_setg(errp,
1013                    "Snapshot with id '%s' and name '%s' does not exist on "
1014                    "device '%s'",
1015                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1016         return NULL;
1017     }
1018 
1019     bdrv_snapshot_delete(bs, id, name, &local_err);
1020     if (error_is_set(&local_err)) {
1021         error_propagate(errp, local_err);
1022         return NULL;
1023     }
1024 
1025     info = g_malloc0(sizeof(SnapshotInfo));
1026     info->id = g_strdup(sn.id_str);
1027     info->name = g_strdup(sn.name);
1028     info->date_nsec = sn.date_nsec;
1029     info->date_sec = sn.date_sec;
1030     info->vm_state_size = sn.vm_state_size;
1031     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1032     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1033 
1034     return info;
1035 }
1036 
1037 /* New and old BlockDriverState structs for group snapshots */
1038 
1039 typedef struct BlkTransactionState BlkTransactionState;
1040 
1041 /* Only prepare() may fail. In a single transaction, only one of commit() or
1042    abort() will be called, clean() will always be called if it present. */
1043 typedef struct BdrvActionOps {
1044     /* Size of state struct, in bytes. */
1045     size_t instance_size;
1046     /* Prepare the work, must NOT be NULL. */
1047     void (*prepare)(BlkTransactionState *common, Error **errp);
1048     /* Commit the changes, can be NULL. */
1049     void (*commit)(BlkTransactionState *common);
1050     /* Abort the changes on fail, can be NULL. */
1051     void (*abort)(BlkTransactionState *common);
1052     /* Clean up resource in the end, can be NULL. */
1053     void (*clean)(BlkTransactionState *common);
1054 } BdrvActionOps;
1055 
1056 /*
1057  * This structure must be arranged as first member in child type, assuming
1058  * that compiler will also arrange it to the same address with parent instance.
1059  * Later it will be used in free().
1060  */
1061 struct BlkTransactionState {
1062     TransactionAction *action;
1063     const BdrvActionOps *ops;
1064     QSIMPLEQ_ENTRY(BlkTransactionState) entry;
1065 };
1066 
1067 /* internal snapshot private data */
1068 typedef struct InternalSnapshotState {
1069     BlkTransactionState common;
1070     BlockDriverState *bs;
1071     QEMUSnapshotInfo sn;
1072 } InternalSnapshotState;
1073 
1074 static void internal_snapshot_prepare(BlkTransactionState *common,
1075                                       Error **errp)
1076 {
1077     const char *device;
1078     const char *name;
1079     BlockDriverState *bs;
1080     QEMUSnapshotInfo old_sn, *sn;
1081     bool ret;
1082     qemu_timeval tv;
1083     BlockdevSnapshotInternal *internal;
1084     InternalSnapshotState *state;
1085     int ret1;
1086 
1087     g_assert(common->action->kind ==
1088              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1089     internal = common->action->blockdev_snapshot_internal_sync;
1090     state = DO_UPCAST(InternalSnapshotState, common, common);
1091 
1092     /* 1. parse input */
1093     device = internal->device;
1094     name = internal->name;
1095 
1096     /* 2. check for validation */
1097     bs = bdrv_find(device);
1098     if (!bs) {
1099         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1100         return;
1101     }
1102 
1103     if (!bdrv_is_inserted(bs)) {
1104         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1105         return;
1106     }
1107 
1108     if (bdrv_is_read_only(bs)) {
1109         error_set(errp, QERR_DEVICE_IS_READ_ONLY, device);
1110         return;
1111     }
1112 
1113     if (!bdrv_can_snapshot(bs)) {
1114         error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1115                   bs->drv->format_name, device, "internal snapshot");
1116         return;
1117     }
1118 
1119     if (!strlen(name)) {
1120         error_setg(errp, "Name is empty");
1121         return;
1122     }
1123 
1124     /* check whether a snapshot with name exist */
1125     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn, errp);
1126     if (error_is_set(errp)) {
1127         return;
1128     } else if (ret) {
1129         error_setg(errp,
1130                    "Snapshot with name '%s' already exists on device '%s'",
1131                    name, device);
1132         return;
1133     }
1134 
1135     /* 3. take the snapshot */
1136     sn = &state->sn;
1137     pstrcpy(sn->name, sizeof(sn->name), name);
1138     qemu_gettimeofday(&tv);
1139     sn->date_sec = tv.tv_sec;
1140     sn->date_nsec = tv.tv_usec * 1000;
1141     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1142 
1143     ret1 = bdrv_snapshot_create(bs, sn);
1144     if (ret1 < 0) {
1145         error_setg_errno(errp, -ret1,
1146                          "Failed to create snapshot '%s' on device '%s'",
1147                          name, device);
1148         return;
1149     }
1150 
1151     /* 4. succeed, mark a snapshot is created */
1152     state->bs = bs;
1153 }
1154 
1155 static void internal_snapshot_abort(BlkTransactionState *common)
1156 {
1157     InternalSnapshotState *state =
1158                              DO_UPCAST(InternalSnapshotState, common, common);
1159     BlockDriverState *bs = state->bs;
1160     QEMUSnapshotInfo *sn = &state->sn;
1161     Error *local_error = NULL;
1162 
1163     if (!bs) {
1164         return;
1165     }
1166 
1167     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1168         error_report("Failed to delete snapshot with id '%s' and name '%s' on "
1169                      "device '%s' in abort: %s",
1170                      sn->id_str,
1171                      sn->name,
1172                      bdrv_get_device_name(bs),
1173                      error_get_pretty(local_error));
1174         error_free(local_error);
1175     }
1176 }
1177 
1178 /* external snapshot private data */
1179 typedef struct ExternalSnapshotState {
1180     BlkTransactionState common;
1181     BlockDriverState *old_bs;
1182     BlockDriverState *new_bs;
1183 } ExternalSnapshotState;
1184 
1185 static void external_snapshot_prepare(BlkTransactionState *common,
1186                                       Error **errp)
1187 {
1188     BlockDriver *drv;
1189     int flags, ret;
1190     Error *local_err = NULL;
1191     const char *device;
1192     const char *new_image_file;
1193     const char *format = "qcow2";
1194     enum NewImageMode mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1195     ExternalSnapshotState *state =
1196                              DO_UPCAST(ExternalSnapshotState, common, common);
1197     TransactionAction *action = common->action;
1198 
1199     /* get parameters */
1200     g_assert(action->kind == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC);
1201 
1202     device = action->blockdev_snapshot_sync->device;
1203     new_image_file = action->blockdev_snapshot_sync->snapshot_file;
1204     if (action->blockdev_snapshot_sync->has_format) {
1205         format = action->blockdev_snapshot_sync->format;
1206     }
1207     if (action->blockdev_snapshot_sync->has_mode) {
1208         mode = action->blockdev_snapshot_sync->mode;
1209     }
1210 
1211     /* start processing */
1212     drv = bdrv_find_format(format);
1213     if (!drv) {
1214         error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1215         return;
1216     }
1217 
1218     state->old_bs = bdrv_find(device);
1219     if (!state->old_bs) {
1220         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1221         return;
1222     }
1223 
1224     if (!bdrv_is_inserted(state->old_bs)) {
1225         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1226         return;
1227     }
1228 
1229     if (bdrv_in_use(state->old_bs)) {
1230         error_set(errp, QERR_DEVICE_IN_USE, device);
1231         return;
1232     }
1233 
1234     if (!bdrv_is_read_only(state->old_bs)) {
1235         if (bdrv_flush(state->old_bs)) {
1236             error_set(errp, QERR_IO_ERROR);
1237             return;
1238         }
1239     }
1240 
1241     if (bdrv_check_ext_snapshot(state->old_bs) != EXT_SNAPSHOT_ALLOWED) {
1242         error_set(errp, QERR_FEATURE_DISABLED, "snapshot");
1243         return;
1244     }
1245 
1246     flags = state->old_bs->open_flags;
1247 
1248     /* create new image w/backing file */
1249     if (mode != NEW_IMAGE_MODE_EXISTING) {
1250         bdrv_img_create(new_image_file, format,
1251                         state->old_bs->filename,
1252                         state->old_bs->drv->format_name,
1253                         NULL, -1, flags, &local_err, false);
1254         if (error_is_set(&local_err)) {
1255             error_propagate(errp, local_err);
1256             return;
1257         }
1258     }
1259 
1260     /* We will manually add the backing_hd field to the bs later */
1261     state->new_bs = bdrv_new("");
1262     /* TODO Inherit bs->options or only take explicit options with an
1263      * extended QMP command? */
1264     ret = bdrv_open(state->new_bs, new_image_file, NULL,
1265                     flags | BDRV_O_NO_BACKING, drv, &local_err);
1266     if (ret != 0) {
1267         error_propagate(errp, local_err);
1268     }
1269 }
1270 
1271 static void external_snapshot_commit(BlkTransactionState *common)
1272 {
1273     ExternalSnapshotState *state =
1274                              DO_UPCAST(ExternalSnapshotState, common, common);
1275 
1276     /* This removes our old bs and adds the new bs */
1277     bdrv_append(state->new_bs, state->old_bs);
1278     /* We don't need (or want) to use the transactional
1279      * bdrv_reopen_multiple() across all the entries at once, because we
1280      * don't want to abort all of them if one of them fails the reopen */
1281     bdrv_reopen(state->new_bs, state->new_bs->open_flags & ~BDRV_O_RDWR,
1282                 NULL);
1283 }
1284 
1285 static void external_snapshot_abort(BlkTransactionState *common)
1286 {
1287     ExternalSnapshotState *state =
1288                              DO_UPCAST(ExternalSnapshotState, common, common);
1289     if (state->new_bs) {
1290         bdrv_unref(state->new_bs);
1291     }
1292 }
1293 
1294 typedef struct DriveBackupState {
1295     BlkTransactionState common;
1296     BlockDriverState *bs;
1297     BlockJob *job;
1298 } DriveBackupState;
1299 
1300 static void drive_backup_prepare(BlkTransactionState *common, Error **errp)
1301 {
1302     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1303     DriveBackup *backup;
1304     Error *local_err = NULL;
1305 
1306     assert(common->action->kind == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1307     backup = common->action->drive_backup;
1308 
1309     qmp_drive_backup(backup->device, backup->target,
1310                      backup->has_format, backup->format,
1311                      backup->sync,
1312                      backup->has_mode, backup->mode,
1313                      backup->has_speed, backup->speed,
1314                      backup->has_on_source_error, backup->on_source_error,
1315                      backup->has_on_target_error, backup->on_target_error,
1316                      &local_err);
1317     if (error_is_set(&local_err)) {
1318         error_propagate(errp, local_err);
1319         state->bs = NULL;
1320         state->job = NULL;
1321         return;
1322     }
1323 
1324     state->bs = bdrv_find(backup->device);
1325     state->job = state->bs->job;
1326 }
1327 
1328 static void drive_backup_abort(BlkTransactionState *common)
1329 {
1330     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1331     BlockDriverState *bs = state->bs;
1332 
1333     /* Only cancel if it's the job we started */
1334     if (bs && bs->job && bs->job == state->job) {
1335         block_job_cancel_sync(bs->job);
1336     }
1337 }
1338 
1339 static void abort_prepare(BlkTransactionState *common, Error **errp)
1340 {
1341     error_setg(errp, "Transaction aborted using Abort action");
1342 }
1343 
1344 static void abort_commit(BlkTransactionState *common)
1345 {
1346     g_assert_not_reached(); /* this action never succeeds */
1347 }
1348 
1349 static const BdrvActionOps actions[] = {
1350     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
1351         .instance_size = sizeof(ExternalSnapshotState),
1352         .prepare  = external_snapshot_prepare,
1353         .commit   = external_snapshot_commit,
1354         .abort = external_snapshot_abort,
1355     },
1356     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
1357         .instance_size = sizeof(DriveBackupState),
1358         .prepare = drive_backup_prepare,
1359         .abort = drive_backup_abort,
1360     },
1361     [TRANSACTION_ACTION_KIND_ABORT] = {
1362         .instance_size = sizeof(BlkTransactionState),
1363         .prepare = abort_prepare,
1364         .commit = abort_commit,
1365     },
1366     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
1367         .instance_size = sizeof(InternalSnapshotState),
1368         .prepare  = internal_snapshot_prepare,
1369         .abort = internal_snapshot_abort,
1370     },
1371 };
1372 
1373 /*
1374  * 'Atomic' group snapshots.  The snapshots are taken as a set, and if any fail
1375  *  then we do not pivot any of the devices in the group, and abandon the
1376  *  snapshots
1377  */
1378 void qmp_transaction(TransactionActionList *dev_list, Error **errp)
1379 {
1380     TransactionActionList *dev_entry = dev_list;
1381     BlkTransactionState *state, *next;
1382     Error *local_err = NULL;
1383 
1384     QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionState) snap_bdrv_states;
1385     QSIMPLEQ_INIT(&snap_bdrv_states);
1386 
1387     /* drain all i/o before any snapshots */
1388     bdrv_drain_all();
1389 
1390     /* We don't do anything in this loop that commits us to the snapshot */
1391     while (NULL != dev_entry) {
1392         TransactionAction *dev_info = NULL;
1393         const BdrvActionOps *ops;
1394 
1395         dev_info = dev_entry->value;
1396         dev_entry = dev_entry->next;
1397 
1398         assert(dev_info->kind < ARRAY_SIZE(actions));
1399 
1400         ops = &actions[dev_info->kind];
1401         assert(ops->instance_size > 0);
1402 
1403         state = g_malloc0(ops->instance_size);
1404         state->ops = ops;
1405         state->action = dev_info;
1406         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
1407 
1408         state->ops->prepare(state, &local_err);
1409         if (error_is_set(&local_err)) {
1410             error_propagate(errp, local_err);
1411             goto delete_and_fail;
1412         }
1413     }
1414 
1415     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1416         if (state->ops->commit) {
1417             state->ops->commit(state);
1418         }
1419     }
1420 
1421     /* success */
1422     goto exit;
1423 
1424 delete_and_fail:
1425     /*
1426     * failure, and it is all-or-none; abandon each new bs, and keep using
1427     * the original bs for all images
1428     */
1429     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1430         if (state->ops->abort) {
1431             state->ops->abort(state);
1432         }
1433     }
1434 exit:
1435     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
1436         if (state->ops->clean) {
1437             state->ops->clean(state);
1438         }
1439         g_free(state);
1440     }
1441 }
1442 
1443 
1444 static void eject_device(BlockDriverState *bs, int force, Error **errp)
1445 {
1446     if (bdrv_in_use(bs)) {
1447         error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
1448         return;
1449     }
1450     if (!bdrv_dev_has_removable_media(bs)) {
1451         error_set(errp, QERR_DEVICE_NOT_REMOVABLE, bdrv_get_device_name(bs));
1452         return;
1453     }
1454 
1455     if (bdrv_dev_is_medium_locked(bs) && !bdrv_dev_is_tray_open(bs)) {
1456         bdrv_dev_eject_request(bs, force);
1457         if (!force) {
1458             error_set(errp, QERR_DEVICE_LOCKED, bdrv_get_device_name(bs));
1459             return;
1460         }
1461     }
1462 
1463     bdrv_close(bs);
1464 }
1465 
1466 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
1467 {
1468     BlockDriverState *bs;
1469 
1470     bs = bdrv_find(device);
1471     if (!bs) {
1472         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1473         return;
1474     }
1475 
1476     eject_device(bs, force, errp);
1477 }
1478 
1479 void qmp_block_passwd(const char *device, const char *password, Error **errp)
1480 {
1481     BlockDriverState *bs;
1482     int err;
1483 
1484     bs = bdrv_find(device);
1485     if (!bs) {
1486         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1487         return;
1488     }
1489 
1490     err = bdrv_set_key(bs, password);
1491     if (err == -EINVAL) {
1492         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
1493         return;
1494     } else if (err < 0) {
1495         error_set(errp, QERR_INVALID_PASSWORD);
1496         return;
1497     }
1498 }
1499 
1500 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename,
1501                                     int bdrv_flags, BlockDriver *drv,
1502                                     const char *password, Error **errp)
1503 {
1504     Error *local_err = NULL;
1505     int ret;
1506 
1507     ret = bdrv_open(bs, filename, NULL, bdrv_flags, drv, &local_err);
1508     if (ret < 0) {
1509         error_propagate(errp, local_err);
1510         return;
1511     }
1512 
1513     if (bdrv_key_required(bs)) {
1514         if (password) {
1515             if (bdrv_set_key(bs, password) < 0) {
1516                 error_set(errp, QERR_INVALID_PASSWORD);
1517             }
1518         } else {
1519             error_set(errp, QERR_DEVICE_ENCRYPTED, bdrv_get_device_name(bs),
1520                       bdrv_get_encrypted_filename(bs));
1521         }
1522     } else if (password) {
1523         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
1524     }
1525 }
1526 
1527 void qmp_change_blockdev(const char *device, const char *filename,
1528                          const char *format, Error **errp)
1529 {
1530     BlockDriverState *bs;
1531     BlockDriver *drv = NULL;
1532     int bdrv_flags;
1533     Error *err = NULL;
1534 
1535     bs = bdrv_find(device);
1536     if (!bs) {
1537         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1538         return;
1539     }
1540 
1541     if (format) {
1542         drv = bdrv_find_whitelisted_format(format, bs->read_only);
1543         if (!drv) {
1544             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1545             return;
1546         }
1547     }
1548 
1549     eject_device(bs, 0, &err);
1550     if (error_is_set(&err)) {
1551         error_propagate(errp, err);
1552         return;
1553     }
1554 
1555     bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
1556     bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
1557 
1558     qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp);
1559 }
1560 
1561 /* throttling disk I/O limits */
1562 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
1563                                int64_t bps_wr,
1564                                int64_t iops,
1565                                int64_t iops_rd,
1566                                int64_t iops_wr,
1567                                bool has_bps_max,
1568                                int64_t bps_max,
1569                                bool has_bps_rd_max,
1570                                int64_t bps_rd_max,
1571                                bool has_bps_wr_max,
1572                                int64_t bps_wr_max,
1573                                bool has_iops_max,
1574                                int64_t iops_max,
1575                                bool has_iops_rd_max,
1576                                int64_t iops_rd_max,
1577                                bool has_iops_wr_max,
1578                                int64_t iops_wr_max,
1579                                bool has_iops_size,
1580                                int64_t iops_size, Error **errp)
1581 {
1582     ThrottleConfig cfg;
1583     BlockDriverState *bs;
1584 
1585     bs = bdrv_find(device);
1586     if (!bs) {
1587         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1588         return;
1589     }
1590 
1591     memset(&cfg, 0, sizeof(cfg));
1592     cfg.buckets[THROTTLE_BPS_TOTAL].avg = bps;
1593     cfg.buckets[THROTTLE_BPS_READ].avg  = bps_rd;
1594     cfg.buckets[THROTTLE_BPS_WRITE].avg = bps_wr;
1595 
1596     cfg.buckets[THROTTLE_OPS_TOTAL].avg = iops;
1597     cfg.buckets[THROTTLE_OPS_READ].avg  = iops_rd;
1598     cfg.buckets[THROTTLE_OPS_WRITE].avg = iops_wr;
1599 
1600     if (has_bps_max) {
1601         cfg.buckets[THROTTLE_BPS_TOTAL].max = bps_max;
1602     }
1603     if (has_bps_rd_max) {
1604         cfg.buckets[THROTTLE_BPS_READ].max = bps_rd_max;
1605     }
1606     if (has_bps_wr_max) {
1607         cfg.buckets[THROTTLE_BPS_WRITE].max = bps_wr_max;
1608     }
1609     if (has_iops_max) {
1610         cfg.buckets[THROTTLE_OPS_TOTAL].max = iops_max;
1611     }
1612     if (has_iops_rd_max) {
1613         cfg.buckets[THROTTLE_OPS_READ].max = iops_rd_max;
1614     }
1615     if (has_iops_wr_max) {
1616         cfg.buckets[THROTTLE_OPS_WRITE].max = iops_wr_max;
1617     }
1618 
1619     if (has_iops_size) {
1620         cfg.op_size = iops_size;
1621     }
1622 
1623     if (!check_throttle_config(&cfg, errp)) {
1624         return;
1625     }
1626 
1627     if (!bs->io_limits_enabled && throttle_enabled(&cfg)) {
1628         bdrv_io_limits_enable(bs);
1629     } else if (bs->io_limits_enabled && !throttle_enabled(&cfg)) {
1630         bdrv_io_limits_disable(bs);
1631     }
1632 
1633     if (bs->io_limits_enabled) {
1634         bdrv_set_io_limits(bs, &cfg);
1635     }
1636 }
1637 
1638 int do_drive_del(Monitor *mon, const QDict *qdict, QObject **ret_data)
1639 {
1640     const char *id = qdict_get_str(qdict, "id");
1641     BlockDriverState *bs;
1642 
1643     bs = bdrv_find(id);
1644     if (!bs) {
1645         qerror_report(QERR_DEVICE_NOT_FOUND, id);
1646         return -1;
1647     }
1648     if (bdrv_in_use(bs)) {
1649         qerror_report(QERR_DEVICE_IN_USE, id);
1650         return -1;
1651     }
1652 
1653     /* quiesce block driver; prevent further io */
1654     bdrv_drain_all();
1655     bdrv_flush(bs);
1656     bdrv_close(bs);
1657 
1658     /* if we have a device attached to this BlockDriverState
1659      * then we need to make the drive anonymous until the device
1660      * can be removed.  If this is a drive with no device backing
1661      * then we can just get rid of the block driver state right here.
1662      */
1663     if (bdrv_get_attached_dev(bs)) {
1664         bdrv_make_anon(bs);
1665 
1666         /* Further I/O must not pause the guest */
1667         bdrv_set_on_error(bs, BLOCKDEV_ON_ERROR_REPORT,
1668                           BLOCKDEV_ON_ERROR_REPORT);
1669     } else {
1670         drive_uninit(drive_get_by_blockdev(bs));
1671     }
1672 
1673     return 0;
1674 }
1675 
1676 void qmp_block_resize(const char *device, int64_t size, Error **errp)
1677 {
1678     BlockDriverState *bs;
1679     int ret;
1680 
1681     bs = bdrv_find(device);
1682     if (!bs) {
1683         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1684         return;
1685     }
1686 
1687     if (size < 0) {
1688         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
1689         return;
1690     }
1691 
1692     /* complete all in-flight operations before resizing the device */
1693     bdrv_drain_all();
1694 
1695     ret = bdrv_truncate(bs, size);
1696     switch (ret) {
1697     case 0:
1698         break;
1699     case -ENOMEDIUM:
1700         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1701         break;
1702     case -ENOTSUP:
1703         error_set(errp, QERR_UNSUPPORTED);
1704         break;
1705     case -EACCES:
1706         error_set(errp, QERR_DEVICE_IS_READ_ONLY, device);
1707         break;
1708     case -EBUSY:
1709         error_set(errp, QERR_DEVICE_IN_USE, device);
1710         break;
1711     default:
1712         error_setg_errno(errp, -ret, "Could not resize");
1713         break;
1714     }
1715 }
1716 
1717 static void block_job_cb(void *opaque, int ret)
1718 {
1719     BlockDriverState *bs = opaque;
1720     QObject *obj;
1721 
1722     trace_block_job_cb(bs, bs->job, ret);
1723 
1724     assert(bs->job);
1725     obj = qobject_from_block_job(bs->job);
1726     if (ret < 0) {
1727         QDict *dict = qobject_to_qdict(obj);
1728         qdict_put(dict, "error", qstring_from_str(strerror(-ret)));
1729     }
1730 
1731     if (block_job_is_cancelled(bs->job)) {
1732         monitor_protocol_event(QEVENT_BLOCK_JOB_CANCELLED, obj);
1733     } else {
1734         monitor_protocol_event(QEVENT_BLOCK_JOB_COMPLETED, obj);
1735     }
1736     qobject_decref(obj);
1737 
1738     bdrv_put_ref_bh_schedule(bs);
1739 }
1740 
1741 void qmp_block_stream(const char *device, bool has_base,
1742                       const char *base, bool has_speed, int64_t speed,
1743                       bool has_on_error, BlockdevOnError on_error,
1744                       Error **errp)
1745 {
1746     BlockDriverState *bs;
1747     BlockDriverState *base_bs = NULL;
1748     Error *local_err = NULL;
1749 
1750     if (!has_on_error) {
1751         on_error = BLOCKDEV_ON_ERROR_REPORT;
1752     }
1753 
1754     bs = bdrv_find(device);
1755     if (!bs) {
1756         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1757         return;
1758     }
1759 
1760     if (base) {
1761         base_bs = bdrv_find_backing_image(bs, base);
1762         if (base_bs == NULL) {
1763             error_set(errp, QERR_BASE_NOT_FOUND, base);
1764             return;
1765         }
1766     }
1767 
1768     stream_start(bs, base_bs, base, has_speed ? speed : 0,
1769                  on_error, block_job_cb, bs, &local_err);
1770     if (error_is_set(&local_err)) {
1771         error_propagate(errp, local_err);
1772         return;
1773     }
1774 
1775     trace_qmp_block_stream(bs, bs->job);
1776 }
1777 
1778 void qmp_block_commit(const char *device,
1779                       bool has_base, const char *base, const char *top,
1780                       bool has_speed, int64_t speed,
1781                       Error **errp)
1782 {
1783     BlockDriverState *bs;
1784     BlockDriverState *base_bs, *top_bs;
1785     Error *local_err = NULL;
1786     /* This will be part of the QMP command, if/when the
1787      * BlockdevOnError change for blkmirror makes it in
1788      */
1789     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
1790 
1791     /* drain all i/o before commits */
1792     bdrv_drain_all();
1793 
1794     bs = bdrv_find(device);
1795     if (!bs) {
1796         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1797         return;
1798     }
1799 
1800     /* default top_bs is the active layer */
1801     top_bs = bs;
1802 
1803     if (top) {
1804         if (strcmp(bs->filename, top) != 0) {
1805             top_bs = bdrv_find_backing_image(bs, top);
1806         }
1807     }
1808 
1809     if (top_bs == NULL) {
1810         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
1811         return;
1812     }
1813 
1814     if (has_base && base) {
1815         base_bs = bdrv_find_backing_image(top_bs, base);
1816     } else {
1817         base_bs = bdrv_find_base(top_bs);
1818     }
1819 
1820     if (base_bs == NULL) {
1821         error_set(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
1822         return;
1823     }
1824 
1825     commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
1826                 &local_err);
1827     if (local_err != NULL) {
1828         error_propagate(errp, local_err);
1829         return;
1830     }
1831 }
1832 
1833 void qmp_drive_backup(const char *device, const char *target,
1834                       bool has_format, const char *format,
1835                       enum MirrorSyncMode sync,
1836                       bool has_mode, enum NewImageMode mode,
1837                       bool has_speed, int64_t speed,
1838                       bool has_on_source_error, BlockdevOnError on_source_error,
1839                       bool has_on_target_error, BlockdevOnError on_target_error,
1840                       Error **errp)
1841 {
1842     BlockDriverState *bs;
1843     BlockDriverState *target_bs;
1844     BlockDriverState *source = NULL;
1845     BlockDriver *drv = NULL;
1846     Error *local_err = NULL;
1847     int flags;
1848     int64_t size;
1849     int ret;
1850 
1851     if (!has_speed) {
1852         speed = 0;
1853     }
1854     if (!has_on_source_error) {
1855         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
1856     }
1857     if (!has_on_target_error) {
1858         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
1859     }
1860     if (!has_mode) {
1861         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1862     }
1863 
1864     bs = bdrv_find(device);
1865     if (!bs) {
1866         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1867         return;
1868     }
1869 
1870     if (!bdrv_is_inserted(bs)) {
1871         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1872         return;
1873     }
1874 
1875     if (!has_format) {
1876         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
1877     }
1878     if (format) {
1879         drv = bdrv_find_format(format);
1880         if (!drv) {
1881             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1882             return;
1883         }
1884     }
1885 
1886     if (bdrv_in_use(bs)) {
1887         error_set(errp, QERR_DEVICE_IN_USE, device);
1888         return;
1889     }
1890 
1891     flags = bs->open_flags | BDRV_O_RDWR;
1892 
1893     /* See if we have a backing HD we can use to create our new image
1894      * on top of. */
1895     if (sync == MIRROR_SYNC_MODE_TOP) {
1896         source = bs->backing_hd;
1897         if (!source) {
1898             sync = MIRROR_SYNC_MODE_FULL;
1899         }
1900     }
1901     if (sync == MIRROR_SYNC_MODE_NONE) {
1902         source = bs;
1903     }
1904 
1905     size = bdrv_getlength(bs);
1906     if (size < 0) {
1907         error_setg_errno(errp, -size, "bdrv_getlength failed");
1908         return;
1909     }
1910 
1911     if (mode != NEW_IMAGE_MODE_EXISTING) {
1912         assert(format && drv);
1913         if (source) {
1914             bdrv_img_create(target, format, source->filename,
1915                             source->drv->format_name, NULL,
1916                             size, flags, &local_err, false);
1917         } else {
1918             bdrv_img_create(target, format, NULL, NULL, NULL,
1919                             size, flags, &local_err, false);
1920         }
1921     }
1922 
1923     if (error_is_set(&local_err)) {
1924         error_propagate(errp, local_err);
1925         return;
1926     }
1927 
1928     target_bs = bdrv_new("");
1929     ret = bdrv_open(target_bs, target, NULL, flags, drv, &local_err);
1930     if (ret < 0) {
1931         bdrv_unref(target_bs);
1932         error_propagate(errp, local_err);
1933         return;
1934     }
1935 
1936     backup_start(bs, target_bs, speed, sync, on_source_error, on_target_error,
1937                  block_job_cb, bs, &local_err);
1938     if (local_err != NULL) {
1939         bdrv_unref(target_bs);
1940         error_propagate(errp, local_err);
1941         return;
1942     }
1943 }
1944 
1945 #define DEFAULT_MIRROR_BUF_SIZE   (10 << 20)
1946 
1947 void qmp_drive_mirror(const char *device, const char *target,
1948                       bool has_format, const char *format,
1949                       enum MirrorSyncMode sync,
1950                       bool has_mode, enum NewImageMode mode,
1951                       bool has_speed, int64_t speed,
1952                       bool has_granularity, uint32_t granularity,
1953                       bool has_buf_size, int64_t buf_size,
1954                       bool has_on_source_error, BlockdevOnError on_source_error,
1955                       bool has_on_target_error, BlockdevOnError on_target_error,
1956                       Error **errp)
1957 {
1958     BlockDriverState *bs;
1959     BlockDriverState *source, *target_bs;
1960     BlockDriver *drv = NULL;
1961     Error *local_err = NULL;
1962     int flags;
1963     int64_t size;
1964     int ret;
1965 
1966     if (!has_speed) {
1967         speed = 0;
1968     }
1969     if (!has_on_source_error) {
1970         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
1971     }
1972     if (!has_on_target_error) {
1973         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
1974     }
1975     if (!has_mode) {
1976         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1977     }
1978     if (!has_granularity) {
1979         granularity = 0;
1980     }
1981     if (!has_buf_size) {
1982         buf_size = DEFAULT_MIRROR_BUF_SIZE;
1983     }
1984 
1985     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
1986         error_set(errp, QERR_INVALID_PARAMETER, device);
1987         return;
1988     }
1989     if (granularity & (granularity - 1)) {
1990         error_set(errp, QERR_INVALID_PARAMETER, device);
1991         return;
1992     }
1993 
1994     bs = bdrv_find(device);
1995     if (!bs) {
1996         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1997         return;
1998     }
1999 
2000     if (!bdrv_is_inserted(bs)) {
2001         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2002         return;
2003     }
2004 
2005     if (!has_format) {
2006         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
2007     }
2008     if (format) {
2009         drv = bdrv_find_format(format);
2010         if (!drv) {
2011             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
2012             return;
2013         }
2014     }
2015 
2016     if (bdrv_in_use(bs)) {
2017         error_set(errp, QERR_DEVICE_IN_USE, device);
2018         return;
2019     }
2020 
2021     flags = bs->open_flags | BDRV_O_RDWR;
2022     source = bs->backing_hd;
2023     if (!source && sync == MIRROR_SYNC_MODE_TOP) {
2024         sync = MIRROR_SYNC_MODE_FULL;
2025     }
2026     if (sync == MIRROR_SYNC_MODE_NONE) {
2027         source = bs;
2028     }
2029 
2030     size = bdrv_getlength(bs);
2031     if (size < 0) {
2032         error_setg_errno(errp, -size, "bdrv_getlength failed");
2033         return;
2034     }
2035 
2036     if ((sync == MIRROR_SYNC_MODE_FULL || !source)
2037         && mode != NEW_IMAGE_MODE_EXISTING)
2038     {
2039         /* create new image w/o backing file */
2040         assert(format && drv);
2041         bdrv_img_create(target, format,
2042                         NULL, NULL, NULL, size, flags, &local_err, false);
2043     } else {
2044         switch (mode) {
2045         case NEW_IMAGE_MODE_EXISTING:
2046             break;
2047         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
2048             /* create new image with backing file */
2049             bdrv_img_create(target, format,
2050                             source->filename,
2051                             source->drv->format_name,
2052                             NULL, size, flags, &local_err, false);
2053             break;
2054         default:
2055             abort();
2056         }
2057     }
2058 
2059     if (error_is_set(&local_err)) {
2060         error_propagate(errp, local_err);
2061         return;
2062     }
2063 
2064     /* Mirroring takes care of copy-on-write using the source's backing
2065      * file.
2066      */
2067     target_bs = bdrv_new("");
2068     ret = bdrv_open(target_bs, target, NULL, flags | BDRV_O_NO_BACKING, drv,
2069                     &local_err);
2070     if (ret < 0) {
2071         bdrv_unref(target_bs);
2072         error_propagate(errp, local_err);
2073         return;
2074     }
2075 
2076     mirror_start(bs, target_bs, speed, granularity, buf_size, sync,
2077                  on_source_error, on_target_error,
2078                  block_job_cb, bs, &local_err);
2079     if (local_err != NULL) {
2080         bdrv_unref(target_bs);
2081         error_propagate(errp, local_err);
2082         return;
2083     }
2084 }
2085 
2086 static BlockJob *find_block_job(const char *device)
2087 {
2088     BlockDriverState *bs;
2089 
2090     bs = bdrv_find(device);
2091     if (!bs || !bs->job) {
2092         return NULL;
2093     }
2094     return bs->job;
2095 }
2096 
2097 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
2098 {
2099     BlockJob *job = find_block_job(device);
2100 
2101     if (!job) {
2102         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2103         return;
2104     }
2105 
2106     block_job_set_speed(job, speed, errp);
2107 }
2108 
2109 void qmp_block_job_cancel(const char *device,
2110                           bool has_force, bool force, Error **errp)
2111 {
2112     BlockJob *job = find_block_job(device);
2113 
2114     if (!has_force) {
2115         force = false;
2116     }
2117 
2118     if (!job) {
2119         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2120         return;
2121     }
2122     if (job->paused && !force) {
2123         error_set(errp, QERR_BLOCK_JOB_PAUSED, device);
2124         return;
2125     }
2126 
2127     trace_qmp_block_job_cancel(job);
2128     block_job_cancel(job);
2129 }
2130 
2131 void qmp_block_job_pause(const char *device, Error **errp)
2132 {
2133     BlockJob *job = find_block_job(device);
2134 
2135     if (!job) {
2136         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2137         return;
2138     }
2139 
2140     trace_qmp_block_job_pause(job);
2141     block_job_pause(job);
2142 }
2143 
2144 void qmp_block_job_resume(const char *device, Error **errp)
2145 {
2146     BlockJob *job = find_block_job(device);
2147 
2148     if (!job) {
2149         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2150         return;
2151     }
2152 
2153     trace_qmp_block_job_resume(job);
2154     block_job_resume(job);
2155 }
2156 
2157 void qmp_block_job_complete(const char *device, Error **errp)
2158 {
2159     BlockJob *job = find_block_job(device);
2160 
2161     if (!job) {
2162         error_set(errp, QERR_BLOCK_JOB_NOT_ACTIVE, device);
2163         return;
2164     }
2165 
2166     trace_qmp_block_job_complete(job);
2167     block_job_complete(job, errp);
2168 }
2169 
2170 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
2171 {
2172     QmpOutputVisitor *ov = qmp_output_visitor_new();
2173     QObject *obj;
2174     QDict *qdict;
2175     Error *local_err = NULL;
2176 
2177     /* Require an ID in the top level */
2178     if (!options->has_id) {
2179         error_setg(errp, "Block device needs an ID");
2180         goto fail;
2181     }
2182 
2183     /* TODO Sort it out in raw-posix and drive_init: Reject aio=native with
2184      * cache.direct=false instead of silently switching to aio=threads, except
2185      * if called from drive_init.
2186      *
2187      * For now, simply forbidding the combination for all drivers will do. */
2188     if (options->has_aio && options->aio == BLOCKDEV_AIO_OPTIONS_NATIVE) {
2189         bool direct = options->cache->has_direct && options->cache->direct;
2190         if (!options->has_cache && !direct) {
2191             error_setg(errp, "aio=native requires cache.direct=true");
2192             goto fail;
2193         }
2194     }
2195 
2196     visit_type_BlockdevOptions(qmp_output_get_visitor(ov),
2197                                &options, NULL, &local_err);
2198     if (error_is_set(&local_err)) {
2199         error_propagate(errp, local_err);
2200         goto fail;
2201     }
2202 
2203     obj = qmp_output_get_qobject(ov);
2204     qdict = qobject_to_qdict(obj);
2205 
2206     qdict_flatten(qdict);
2207 
2208     blockdev_init(qdict, IF_NONE, &local_err);
2209     if (error_is_set(&local_err)) {
2210         error_propagate(errp, local_err);
2211         goto fail;
2212     }
2213 
2214 fail:
2215     qmp_output_visitor_cleanup(ov);
2216 }
2217 
2218 static void do_qmp_query_block_jobs_one(void *opaque, BlockDriverState *bs)
2219 {
2220     BlockJobInfoList **prev = opaque;
2221     BlockJob *job = bs->job;
2222 
2223     if (job) {
2224         BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
2225         elem->value = block_job_query(bs->job);
2226         (*prev)->next = elem;
2227         *prev = elem;
2228     }
2229 }
2230 
2231 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
2232 {
2233     /* Dummy is a fake list element for holding the head pointer */
2234     BlockJobInfoList dummy = {};
2235     BlockJobInfoList *prev = &dummy;
2236     bdrv_iterate(do_qmp_query_block_jobs_one, &prev);
2237     return dummy.next;
2238 }
2239 
2240 QemuOptsList qemu_common_drive_opts = {
2241     .name = "drive",
2242     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
2243     .desc = {
2244         {
2245             .name = "snapshot",
2246             .type = QEMU_OPT_BOOL,
2247             .help = "enable/disable snapshot mode",
2248         },{
2249             .name = "file",
2250             .type = QEMU_OPT_STRING,
2251             .help = "disk image",
2252         },{
2253             .name = "discard",
2254             .type = QEMU_OPT_STRING,
2255             .help = "discard operation (ignore/off, unmap/on)",
2256         },{
2257             .name = "cache.writeback",
2258             .type = QEMU_OPT_BOOL,
2259             .help = "enables writeback mode for any caches",
2260         },{
2261             .name = "cache.direct",
2262             .type = QEMU_OPT_BOOL,
2263             .help = "enables use of O_DIRECT (bypass the host page cache)",
2264         },{
2265             .name = "cache.no-flush",
2266             .type = QEMU_OPT_BOOL,
2267             .help = "ignore any flush requests for the device",
2268         },{
2269             .name = "aio",
2270             .type = QEMU_OPT_STRING,
2271             .help = "host AIO implementation (threads, native)",
2272         },{
2273             .name = "format",
2274             .type = QEMU_OPT_STRING,
2275             .help = "disk format (raw, qcow2, ...)",
2276         },{
2277             .name = "serial",
2278             .type = QEMU_OPT_STRING,
2279             .help = "disk serial number",
2280         },{
2281             .name = "rerror",
2282             .type = QEMU_OPT_STRING,
2283             .help = "read error action",
2284         },{
2285             .name = "werror",
2286             .type = QEMU_OPT_STRING,
2287             .help = "write error action",
2288         },{
2289             .name = "read-only",
2290             .type = QEMU_OPT_BOOL,
2291             .help = "open drive file as read-only",
2292         },{
2293             .name = "throttling.iops-total",
2294             .type = QEMU_OPT_NUMBER,
2295             .help = "limit total I/O operations per second",
2296         },{
2297             .name = "throttling.iops-read",
2298             .type = QEMU_OPT_NUMBER,
2299             .help = "limit read operations per second",
2300         },{
2301             .name = "throttling.iops-write",
2302             .type = QEMU_OPT_NUMBER,
2303             .help = "limit write operations per second",
2304         },{
2305             .name = "throttling.bps-total",
2306             .type = QEMU_OPT_NUMBER,
2307             .help = "limit total bytes per second",
2308         },{
2309             .name = "throttling.bps-read",
2310             .type = QEMU_OPT_NUMBER,
2311             .help = "limit read bytes per second",
2312         },{
2313             .name = "throttling.bps-write",
2314             .type = QEMU_OPT_NUMBER,
2315             .help = "limit write bytes per second",
2316         },{
2317             .name = "throttling.iops-total-max",
2318             .type = QEMU_OPT_NUMBER,
2319             .help = "I/O operations burst",
2320         },{
2321             .name = "throttling.iops-read-max",
2322             .type = QEMU_OPT_NUMBER,
2323             .help = "I/O operations read burst",
2324         },{
2325             .name = "throttling.iops-write-max",
2326             .type = QEMU_OPT_NUMBER,
2327             .help = "I/O operations write burst",
2328         },{
2329             .name = "throttling.bps-total-max",
2330             .type = QEMU_OPT_NUMBER,
2331             .help = "total bytes burst",
2332         },{
2333             .name = "throttling.bps-read-max",
2334             .type = QEMU_OPT_NUMBER,
2335             .help = "total bytes read burst",
2336         },{
2337             .name = "throttling.bps-write-max",
2338             .type = QEMU_OPT_NUMBER,
2339             .help = "total bytes write burst",
2340         },{
2341             .name = "throttling.iops-size",
2342             .type = QEMU_OPT_NUMBER,
2343             .help = "when limiting by iops max size of an I/O in bytes",
2344         },{
2345             .name = "copy-on-read",
2346             .type = QEMU_OPT_BOOL,
2347             .help = "copy read data from backing file into image file",
2348         },
2349         { /* end of list */ }
2350     },
2351 };
2352 
2353 QemuOptsList qemu_drive_opts = {
2354     .name = "drive",
2355     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
2356     .desc = {
2357         /*
2358          * no elements => accept any params
2359          * validation will happen later
2360          */
2361         { /* end of list */ }
2362     },
2363 };
2364