xref: /openbmc/qemu/blockdev.c (revision d44f928a54497188c25357840a3224925d1b527b)
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/block-backend.h"
34 #include "sysemu/blockdev.h"
35 #include "hw/block/block.h"
36 #include "block/blockjob.h"
37 #include "block/throttle-groups.h"
38 #include "monitor/monitor.h"
39 #include "qemu/error-report.h"
40 #include "qemu/option.h"
41 #include "qemu/config-file.h"
42 #include "qapi/qmp/types.h"
43 #include "qapi-visit.h"
44 #include "qapi/qmp/qerror.h"
45 #include "qapi/qmp-output-visitor.h"
46 #include "qapi/util.h"
47 #include "sysemu/sysemu.h"
48 #include "block/block_int.h"
49 #include "qmp-commands.h"
50 #include "trace.h"
51 #include "sysemu/arch_init.h"
52 
53 static const char *const if_name[IF_COUNT] = {
54     [IF_NONE] = "none",
55     [IF_IDE] = "ide",
56     [IF_SCSI] = "scsi",
57     [IF_FLOPPY] = "floppy",
58     [IF_PFLASH] = "pflash",
59     [IF_MTD] = "mtd",
60     [IF_SD] = "sd",
61     [IF_VIRTIO] = "virtio",
62     [IF_XEN] = "xen",
63 };
64 
65 static int if_max_devs[IF_COUNT] = {
66     /*
67      * Do not change these numbers!  They govern how drive option
68      * index maps to unit and bus.  That mapping is ABI.
69      *
70      * All controllers used to imlement if=T drives need to support
71      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
72      * Otherwise, some index values map to "impossible" bus, unit
73      * values.
74      *
75      * For instance, if you change [IF_SCSI] to 255, -drive
76      * if=scsi,index=12 no longer means bus=1,unit=5, but
77      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
78      * the drive can't be set up.  Regression.
79      */
80     [IF_IDE] = 2,
81     [IF_SCSI] = 7,
82 };
83 
84 /**
85  * Boards may call this to offer board-by-board overrides
86  * of the default, global values.
87  */
88 void override_max_devs(BlockInterfaceType type, int max_devs)
89 {
90     BlockBackend *blk;
91     DriveInfo *dinfo;
92 
93     if (max_devs <= 0) {
94         return;
95     }
96 
97     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
98         dinfo = blk_legacy_dinfo(blk);
99         if (dinfo->type == type) {
100             fprintf(stderr, "Cannot override units-per-bus property of"
101                     " the %s interface, because a drive of that type has"
102                     " already been added.\n", if_name[type]);
103             g_assert_not_reached();
104         }
105     }
106 
107     if_max_devs[type] = max_devs;
108 }
109 
110 /*
111  * We automatically delete the drive when a device using it gets
112  * unplugged.  Questionable feature, but we can't just drop it.
113  * Device models call blockdev_mark_auto_del() to schedule the
114  * automatic deletion, and generic qdev code calls blockdev_auto_del()
115  * when deletion is actually safe.
116  */
117 void blockdev_mark_auto_del(BlockBackend *blk)
118 {
119     DriveInfo *dinfo = blk_legacy_dinfo(blk);
120     BlockDriverState *bs = blk_bs(blk);
121     AioContext *aio_context;
122 
123     if (!dinfo) {
124         return;
125     }
126 
127     aio_context = bdrv_get_aio_context(bs);
128     aio_context_acquire(aio_context);
129 
130     if (bs->job) {
131         block_job_cancel(bs->job);
132     }
133 
134     aio_context_release(aio_context);
135 
136     dinfo->auto_del = 1;
137 }
138 
139 void blockdev_auto_del(BlockBackend *blk)
140 {
141     DriveInfo *dinfo = blk_legacy_dinfo(blk);
142 
143     if (dinfo && dinfo->auto_del) {
144         blk_unref(blk);
145     }
146 }
147 
148 /**
149  * Returns the current mapping of how many units per bus
150  * a particular interface can support.
151  *
152  *  A positive integer indicates n units per bus.
153  *  0 implies the mapping has not been established.
154  * -1 indicates an invalid BlockInterfaceType was given.
155  */
156 int drive_get_max_devs(BlockInterfaceType type)
157 {
158     if (type >= IF_IDE && type < IF_COUNT) {
159         return if_max_devs[type];
160     }
161 
162     return -1;
163 }
164 
165 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
166 {
167     int max_devs = if_max_devs[type];
168     return max_devs ? index / max_devs : 0;
169 }
170 
171 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
172 {
173     int max_devs = if_max_devs[type];
174     return max_devs ? index % max_devs : index;
175 }
176 
177 QemuOpts *drive_def(const char *optstr)
178 {
179     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
180 }
181 
182 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
183                     const char *optstr)
184 {
185     QemuOpts *opts;
186 
187     opts = drive_def(optstr);
188     if (!opts) {
189         return NULL;
190     }
191     if (type != IF_DEFAULT) {
192         qemu_opt_set(opts, "if", if_name[type], &error_abort);
193     }
194     if (index >= 0) {
195         qemu_opt_set_number(opts, "index", index, &error_abort);
196     }
197     if (file)
198         qemu_opt_set(opts, "file", file, &error_abort);
199     return opts;
200 }
201 
202 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
203 {
204     BlockBackend *blk;
205     DriveInfo *dinfo;
206 
207     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
208         dinfo = blk_legacy_dinfo(blk);
209         if (dinfo && dinfo->type == type
210             && dinfo->bus == bus && dinfo->unit == unit) {
211             return dinfo;
212         }
213     }
214 
215     return NULL;
216 }
217 
218 bool drive_check_orphaned(void)
219 {
220     BlockBackend *blk;
221     DriveInfo *dinfo;
222     bool rs = false;
223 
224     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
225         dinfo = blk_legacy_dinfo(blk);
226         /* If dinfo->bdrv->dev is NULL, it has no device attached. */
227         /* Unless this is a default drive, this may be an oversight. */
228         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
229             dinfo->type != IF_NONE) {
230             fprintf(stderr, "Warning: Orphaned drive without device: "
231                     "id=%s,file=%s,if=%s,bus=%d,unit=%d\n",
232                     blk_name(blk), blk_bs(blk)->filename, if_name[dinfo->type],
233                     dinfo->bus, dinfo->unit);
234             rs = true;
235         }
236     }
237 
238     return rs;
239 }
240 
241 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
242 {
243     return drive_get(type,
244                      drive_index_to_bus_id(type, index),
245                      drive_index_to_unit_id(type, index));
246 }
247 
248 int drive_get_max_bus(BlockInterfaceType type)
249 {
250     int max_bus;
251     BlockBackend *blk;
252     DriveInfo *dinfo;
253 
254     max_bus = -1;
255     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
256         dinfo = blk_legacy_dinfo(blk);
257         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
258             max_bus = dinfo->bus;
259         }
260     }
261     return max_bus;
262 }
263 
264 /* Get a block device.  This should only be used for single-drive devices
265    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
266    appropriate bus.  */
267 DriveInfo *drive_get_next(BlockInterfaceType type)
268 {
269     static int next_block_unit[IF_COUNT];
270 
271     return drive_get(type, 0, next_block_unit[type]++);
272 }
273 
274 static void bdrv_format_print(void *opaque, const char *name)
275 {
276     error_printf(" %s", name);
277 }
278 
279 typedef struct {
280     QEMUBH *bh;
281     BlockDriverState *bs;
282 } BDRVPutRefBH;
283 
284 static void bdrv_put_ref_bh(void *opaque)
285 {
286     BDRVPutRefBH *s = opaque;
287 
288     bdrv_unref(s->bs);
289     qemu_bh_delete(s->bh);
290     g_free(s);
291 }
292 
293 /*
294  * Release a BDS reference in a BH
295  *
296  * It is not safe to use bdrv_unref() from a callback function when the callers
297  * still need the BlockDriverState.  In such cases we schedule a BH to release
298  * the reference.
299  */
300 static void bdrv_put_ref_bh_schedule(BlockDriverState *bs)
301 {
302     BDRVPutRefBH *s;
303 
304     s = g_new(BDRVPutRefBH, 1);
305     s->bh = qemu_bh_new(bdrv_put_ref_bh, s);
306     s->bs = bs;
307     qemu_bh_schedule(s->bh);
308 }
309 
310 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
311 {
312     if (!strcmp(buf, "ignore")) {
313         return BLOCKDEV_ON_ERROR_IGNORE;
314     } else if (!is_read && !strcmp(buf, "enospc")) {
315         return BLOCKDEV_ON_ERROR_ENOSPC;
316     } else if (!strcmp(buf, "stop")) {
317         return BLOCKDEV_ON_ERROR_STOP;
318     } else if (!strcmp(buf, "report")) {
319         return BLOCKDEV_ON_ERROR_REPORT;
320     } else {
321         error_setg(errp, "'%s' invalid %s error action",
322                    buf, is_read ? "read" : "write");
323         return -1;
324     }
325 }
326 
327 static bool check_throttle_config(ThrottleConfig *cfg, Error **errp)
328 {
329     if (throttle_conflicting(cfg)) {
330         error_setg(errp, "bps/iops/max total values and read/write values"
331                          " cannot be used at the same time");
332         return false;
333     }
334 
335     if (!throttle_is_valid(cfg)) {
336         error_setg(errp, "bps/iops/maxs values must be 0 or greater");
337         return false;
338     }
339 
340     if (throttle_max_is_missing_limit(cfg)) {
341         error_setg(errp, "bps_max/iops_max require corresponding"
342                          " bps/iops values");
343         return false;
344     }
345 
346     return true;
347 }
348 
349 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
350 
351 /* Takes the ownership of bs_opts */
352 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
353                                    Error **errp)
354 {
355     const char *buf;
356     int ro = 0;
357     int bdrv_flags = 0;
358     int on_read_error, on_write_error;
359     BlockBackend *blk;
360     BlockDriverState *bs;
361     ThrottleConfig cfg;
362     int snapshot = 0;
363     bool copy_on_read;
364     Error *error = NULL;
365     QemuOpts *opts;
366     const char *id;
367     bool has_driver_specific_opts;
368     BlockdevDetectZeroesOptions detect_zeroes;
369     const char *throttling_group;
370 
371     /* Check common options by copying from bs_opts to opts, all other options
372      * stay in bs_opts for processing by bdrv_open(). */
373     id = qdict_get_try_str(bs_opts, "id");
374     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
375     if (error) {
376         error_propagate(errp, error);
377         goto err_no_opts;
378     }
379 
380     qemu_opts_absorb_qdict(opts, bs_opts, &error);
381     if (error) {
382         error_propagate(errp, error);
383         goto early_err;
384     }
385 
386     if (id) {
387         qdict_del(bs_opts, "id");
388     }
389 
390     has_driver_specific_opts = !!qdict_size(bs_opts);
391 
392     /* extract parameters */
393     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
394     ro = qemu_opt_get_bool(opts, "read-only", 0);
395     copy_on_read = qemu_opt_get_bool(opts, "copy-on-read", false);
396 
397     if ((buf = qemu_opt_get(opts, "discard")) != NULL) {
398         if (bdrv_parse_discard_flags(buf, &bdrv_flags) != 0) {
399             error_setg(errp, "invalid discard option");
400             goto early_err;
401         }
402     }
403 
404     if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true)) {
405         bdrv_flags |= BDRV_O_CACHE_WB;
406     }
407     if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
408         bdrv_flags |= BDRV_O_NOCACHE;
409     }
410     if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
411         bdrv_flags |= BDRV_O_NO_FLUSH;
412     }
413 
414     if ((buf = qemu_opt_get(opts, "aio")) != NULL) {
415         if (!strcmp(buf, "native")) {
416             bdrv_flags |= BDRV_O_NATIVE_AIO;
417         } else if (!strcmp(buf, "threads")) {
418             /* this is the default */
419         } else {
420            error_setg(errp, "invalid aio option");
421            goto early_err;
422         }
423     }
424 
425     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
426         if (is_help_option(buf)) {
427             error_printf("Supported formats:");
428             bdrv_iterate_format(bdrv_format_print, NULL);
429             error_printf("\n");
430             goto early_err;
431         }
432 
433         if (qdict_haskey(bs_opts, "driver")) {
434             error_setg(errp, "Cannot specify both 'driver' and 'format'");
435             goto early_err;
436         }
437         qdict_put(bs_opts, "driver", qstring_from_str(buf));
438     }
439 
440     /* disk I/O throttling */
441     memset(&cfg, 0, sizeof(cfg));
442     cfg.buckets[THROTTLE_BPS_TOTAL].avg =
443         qemu_opt_get_number(opts, "throttling.bps-total", 0);
444     cfg.buckets[THROTTLE_BPS_READ].avg  =
445         qemu_opt_get_number(opts, "throttling.bps-read", 0);
446     cfg.buckets[THROTTLE_BPS_WRITE].avg =
447         qemu_opt_get_number(opts, "throttling.bps-write", 0);
448     cfg.buckets[THROTTLE_OPS_TOTAL].avg =
449         qemu_opt_get_number(opts, "throttling.iops-total", 0);
450     cfg.buckets[THROTTLE_OPS_READ].avg =
451         qemu_opt_get_number(opts, "throttling.iops-read", 0);
452     cfg.buckets[THROTTLE_OPS_WRITE].avg =
453         qemu_opt_get_number(opts, "throttling.iops-write", 0);
454 
455     cfg.buckets[THROTTLE_BPS_TOTAL].max =
456         qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
457     cfg.buckets[THROTTLE_BPS_READ].max  =
458         qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
459     cfg.buckets[THROTTLE_BPS_WRITE].max =
460         qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
461     cfg.buckets[THROTTLE_OPS_TOTAL].max =
462         qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
463     cfg.buckets[THROTTLE_OPS_READ].max =
464         qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
465     cfg.buckets[THROTTLE_OPS_WRITE].max =
466         qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
467 
468     cfg.op_size = qemu_opt_get_number(opts, "throttling.iops-size", 0);
469 
470     throttling_group = qemu_opt_get(opts, "throttling.group");
471 
472     if (!check_throttle_config(&cfg, &error)) {
473         error_propagate(errp, error);
474         goto early_err;
475     }
476 
477     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
478     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
479         on_write_error = parse_block_error_action(buf, 0, &error);
480         if (error) {
481             error_propagate(errp, error);
482             goto early_err;
483         }
484     }
485 
486     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
487     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
488         on_read_error = parse_block_error_action(buf, 1, &error);
489         if (error) {
490             error_propagate(errp, error);
491             goto early_err;
492         }
493     }
494 
495     detect_zeroes =
496         qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
497                         qemu_opt_get(opts, "detect-zeroes"),
498                         BLOCKDEV_DETECT_ZEROES_OPTIONS_MAX,
499                         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
500                         &error);
501     if (error) {
502         error_propagate(errp, error);
503         goto early_err;
504     }
505 
506     if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
507         !(bdrv_flags & BDRV_O_UNMAP)) {
508         error_setg(errp, "setting detect-zeroes to unmap is not allowed "
509                          "without setting discard operation to unmap");
510         goto early_err;
511     }
512 
513     /* init */
514     if ((!file || !*file) && !has_driver_specific_opts) {
515         blk = blk_new_with_bs(qemu_opts_id(opts), errp);
516         if (!blk) {
517             goto early_err;
518         }
519 
520         bs = blk_bs(blk);
521         bs->open_flags = snapshot ? BDRV_O_SNAPSHOT : 0;
522         bs->read_only = ro;
523 
524         QDECREF(bs_opts);
525     } else {
526         if (file && !*file) {
527             file = NULL;
528         }
529 
530         if (snapshot) {
531             /* always use cache=unsafe with snapshot */
532             bdrv_flags &= ~BDRV_O_CACHE_MASK;
533             bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
534         }
535 
536         if (copy_on_read) {
537             bdrv_flags |= BDRV_O_COPY_ON_READ;
538         }
539 
540         bdrv_flags |= ro ? 0 : BDRV_O_RDWR;
541 
542         blk = blk_new_open(qemu_opts_id(opts), file, NULL, bs_opts, bdrv_flags,
543                            errp);
544         if (!blk) {
545             goto err_no_bs_opts;
546         }
547         bs = blk_bs(blk);
548     }
549 
550     bs->detect_zeroes = detect_zeroes;
551 
552     bdrv_set_on_error(bs, on_read_error, on_write_error);
553 
554     /* disk I/O throttling */
555     if (throttle_enabled(&cfg)) {
556         if (!throttling_group) {
557             throttling_group = blk_name(blk);
558         }
559         bdrv_io_limits_enable(bs, throttling_group);
560         bdrv_set_io_limits(bs, &cfg);
561     }
562 
563     if (bdrv_key_required(bs)) {
564         autostart = 0;
565     }
566 
567 err_no_bs_opts:
568     qemu_opts_del(opts);
569     return blk;
570 
571 early_err:
572     qemu_opts_del(opts);
573 err_no_opts:
574     QDECREF(bs_opts);
575     return NULL;
576 }
577 
578 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
579                             Error **errp)
580 {
581     const char *value;
582 
583     value = qemu_opt_get(opts, from);
584     if (value) {
585         if (qemu_opt_find(opts, to)) {
586             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
587                        "same time", to, from);
588             return;
589         }
590     }
591 
592     /* rename all items in opts */
593     while ((value = qemu_opt_get(opts, from))) {
594         qemu_opt_set(opts, to, value, &error_abort);
595         qemu_opt_unset(opts, from);
596     }
597 }
598 
599 QemuOptsList qemu_legacy_drive_opts = {
600     .name = "drive",
601     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
602     .desc = {
603         {
604             .name = "bus",
605             .type = QEMU_OPT_NUMBER,
606             .help = "bus number",
607         },{
608             .name = "unit",
609             .type = QEMU_OPT_NUMBER,
610             .help = "unit number (i.e. lun for scsi)",
611         },{
612             .name = "index",
613             .type = QEMU_OPT_NUMBER,
614             .help = "index number",
615         },{
616             .name = "media",
617             .type = QEMU_OPT_STRING,
618             .help = "media type (disk, cdrom)",
619         },{
620             .name = "if",
621             .type = QEMU_OPT_STRING,
622             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
623         },{
624             .name = "cyls",
625             .type = QEMU_OPT_NUMBER,
626             .help = "number of cylinders (ide disk geometry)",
627         },{
628             .name = "heads",
629             .type = QEMU_OPT_NUMBER,
630             .help = "number of heads (ide disk geometry)",
631         },{
632             .name = "secs",
633             .type = QEMU_OPT_NUMBER,
634             .help = "number of sectors (ide disk geometry)",
635         },{
636             .name = "trans",
637             .type = QEMU_OPT_STRING,
638             .help = "chs translation (auto, lba, none)",
639         },{
640             .name = "boot",
641             .type = QEMU_OPT_BOOL,
642             .help = "(deprecated, ignored)",
643         },{
644             .name = "addr",
645             .type = QEMU_OPT_STRING,
646             .help = "pci address (virtio only)",
647         },{
648             .name = "serial",
649             .type = QEMU_OPT_STRING,
650             .help = "disk serial number",
651         },{
652             .name = "file",
653             .type = QEMU_OPT_STRING,
654             .help = "file name",
655         },
656 
657         /* Options that are passed on, but have special semantics with -drive */
658         {
659             .name = "read-only",
660             .type = QEMU_OPT_BOOL,
661             .help = "open drive file as read-only",
662         },{
663             .name = "rerror",
664             .type = QEMU_OPT_STRING,
665             .help = "read error action",
666         },{
667             .name = "werror",
668             .type = QEMU_OPT_STRING,
669             .help = "write error action",
670         },{
671             .name = "copy-on-read",
672             .type = QEMU_OPT_BOOL,
673             .help = "copy read data from backing file into image file",
674         },
675 
676         { /* end of list */ }
677     },
678 };
679 
680 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
681 {
682     const char *value;
683     BlockBackend *blk;
684     DriveInfo *dinfo = NULL;
685     QDict *bs_opts;
686     QemuOpts *legacy_opts;
687     DriveMediaType media = MEDIA_DISK;
688     BlockInterfaceType type;
689     int cyls, heads, secs, translation;
690     int max_devs, bus_id, unit_id, index;
691     const char *devaddr;
692     const char *werror, *rerror;
693     bool read_only = false;
694     bool copy_on_read;
695     const char *serial;
696     const char *filename;
697     Error *local_err = NULL;
698     int i;
699 
700     /* Change legacy command line options into QMP ones */
701     static const struct {
702         const char *from;
703         const char *to;
704     } opt_renames[] = {
705         { "iops",           "throttling.iops-total" },
706         { "iops_rd",        "throttling.iops-read" },
707         { "iops_wr",        "throttling.iops-write" },
708 
709         { "bps",            "throttling.bps-total" },
710         { "bps_rd",         "throttling.bps-read" },
711         { "bps_wr",         "throttling.bps-write" },
712 
713         { "iops_max",       "throttling.iops-total-max" },
714         { "iops_rd_max",    "throttling.iops-read-max" },
715         { "iops_wr_max",    "throttling.iops-write-max" },
716 
717         { "bps_max",        "throttling.bps-total-max" },
718         { "bps_rd_max",     "throttling.bps-read-max" },
719         { "bps_wr_max",     "throttling.bps-write-max" },
720 
721         { "iops_size",      "throttling.iops-size" },
722 
723         { "group",          "throttling.group" },
724 
725         { "readonly",       "read-only" },
726     };
727 
728     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
729         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
730                         &local_err);
731         if (local_err) {
732             error_report_err(local_err);
733             return NULL;
734         }
735     }
736 
737     value = qemu_opt_get(all_opts, "cache");
738     if (value) {
739         int flags = 0;
740 
741         if (bdrv_parse_cache_flags(value, &flags) != 0) {
742             error_report("invalid cache option");
743             return NULL;
744         }
745 
746         /* Specific options take precedence */
747         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
748             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
749                               !!(flags & BDRV_O_CACHE_WB), &error_abort);
750         }
751         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
752             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
753                               !!(flags & BDRV_O_NOCACHE), &error_abort);
754         }
755         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
756             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
757                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
758         }
759         qemu_opt_unset(all_opts, "cache");
760     }
761 
762     /* Get a QDict for processing the options */
763     bs_opts = qdict_new();
764     qemu_opts_to_qdict(all_opts, bs_opts);
765 
766     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
767                                    &error_abort);
768     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
769     if (local_err) {
770         error_report_err(local_err);
771         goto fail;
772     }
773 
774     /* Deprecated option boot=[on|off] */
775     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
776         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
777                 "ignored. Future versions will reject this parameter. Please "
778                 "update your scripts.\n");
779     }
780 
781     /* Media type */
782     value = qemu_opt_get(legacy_opts, "media");
783     if (value) {
784         if (!strcmp(value, "disk")) {
785             media = MEDIA_DISK;
786         } else if (!strcmp(value, "cdrom")) {
787             media = MEDIA_CDROM;
788             read_only = true;
789         } else {
790             error_report("'%s' invalid media", value);
791             goto fail;
792         }
793     }
794 
795     /* copy-on-read is disabled with a warning for read-only devices */
796     read_only |= qemu_opt_get_bool(legacy_opts, "read-only", false);
797     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
798 
799     if (read_only && copy_on_read) {
800         error_report("warning: disabling copy-on-read on read-only drive");
801         copy_on_read = false;
802     }
803 
804     qdict_put(bs_opts, "read-only",
805               qstring_from_str(read_only ? "on" : "off"));
806     qdict_put(bs_opts, "copy-on-read",
807               qstring_from_str(copy_on_read ? "on" :"off"));
808 
809     /* Controller type */
810     value = qemu_opt_get(legacy_opts, "if");
811     if (value) {
812         for (type = 0;
813              type < IF_COUNT && strcmp(value, if_name[type]);
814              type++) {
815         }
816         if (type == IF_COUNT) {
817             error_report("unsupported bus type '%s'", value);
818             goto fail;
819         }
820     } else {
821         type = block_default_type;
822     }
823 
824     /* Geometry */
825     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
826     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
827     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
828 
829     if (cyls || heads || secs) {
830         if (cyls < 1) {
831             error_report("invalid physical cyls number");
832             goto fail;
833         }
834         if (heads < 1) {
835             error_report("invalid physical heads number");
836             goto fail;
837         }
838         if (secs < 1) {
839             error_report("invalid physical secs number");
840             goto fail;
841         }
842     }
843 
844     translation = BIOS_ATA_TRANSLATION_AUTO;
845     value = qemu_opt_get(legacy_opts, "trans");
846     if (value != NULL) {
847         if (!cyls) {
848             error_report("'%s' trans must be used with cyls, heads and secs",
849                          value);
850             goto fail;
851         }
852         if (!strcmp(value, "none")) {
853             translation = BIOS_ATA_TRANSLATION_NONE;
854         } else if (!strcmp(value, "lba")) {
855             translation = BIOS_ATA_TRANSLATION_LBA;
856         } else if (!strcmp(value, "large")) {
857             translation = BIOS_ATA_TRANSLATION_LARGE;
858         } else if (!strcmp(value, "rechs")) {
859             translation = BIOS_ATA_TRANSLATION_RECHS;
860         } else if (!strcmp(value, "auto")) {
861             translation = BIOS_ATA_TRANSLATION_AUTO;
862         } else {
863             error_report("'%s' invalid translation type", value);
864             goto fail;
865         }
866     }
867 
868     if (media == MEDIA_CDROM) {
869         if (cyls || secs || heads) {
870             error_report("CHS can't be set with media=cdrom");
871             goto fail;
872         }
873     }
874 
875     /* Device address specified by bus/unit or index.
876      * If none was specified, try to find the first free one. */
877     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
878     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
879     index   = qemu_opt_get_number(legacy_opts, "index", -1);
880 
881     max_devs = if_max_devs[type];
882 
883     if (index != -1) {
884         if (bus_id != 0 || unit_id != -1) {
885             error_report("index cannot be used with bus and unit");
886             goto fail;
887         }
888         bus_id = drive_index_to_bus_id(type, index);
889         unit_id = drive_index_to_unit_id(type, index);
890     }
891 
892     if (unit_id == -1) {
893        unit_id = 0;
894        while (drive_get(type, bus_id, unit_id) != NULL) {
895            unit_id++;
896            if (max_devs && unit_id >= max_devs) {
897                unit_id -= max_devs;
898                bus_id++;
899            }
900        }
901     }
902 
903     if (max_devs && unit_id >= max_devs) {
904         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
905         goto fail;
906     }
907 
908     if (drive_get(type, bus_id, unit_id) != NULL) {
909         error_report("drive with bus=%d, unit=%d (index=%d) exists",
910                      bus_id, unit_id, index);
911         goto fail;
912     }
913 
914     /* Serial number */
915     serial = qemu_opt_get(legacy_opts, "serial");
916 
917     /* no id supplied -> create one */
918     if (qemu_opts_id(all_opts) == NULL) {
919         char *new_id;
920         const char *mediastr = "";
921         if (type == IF_IDE || type == IF_SCSI) {
922             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
923         }
924         if (max_devs) {
925             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
926                                      mediastr, unit_id);
927         } else {
928             new_id = g_strdup_printf("%s%s%i", if_name[type],
929                                      mediastr, unit_id);
930         }
931         qdict_put(bs_opts, "id", qstring_from_str(new_id));
932         g_free(new_id);
933     }
934 
935     /* Add virtio block device */
936     devaddr = qemu_opt_get(legacy_opts, "addr");
937     if (devaddr && type != IF_VIRTIO) {
938         error_report("addr is not supported by this bus type");
939         goto fail;
940     }
941 
942     if (type == IF_VIRTIO) {
943         QemuOpts *devopts;
944         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
945                                    &error_abort);
946         if (arch_type == QEMU_ARCH_S390X) {
947             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
948         } else {
949             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
950         }
951         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
952                      &error_abort);
953         if (devaddr) {
954             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
955         }
956     }
957 
958     filename = qemu_opt_get(legacy_opts, "file");
959 
960     /* Check werror/rerror compatibility with if=... */
961     werror = qemu_opt_get(legacy_opts, "werror");
962     if (werror != NULL) {
963         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
964             type != IF_NONE) {
965             error_report("werror is not supported by this bus type");
966             goto fail;
967         }
968         qdict_put(bs_opts, "werror", qstring_from_str(werror));
969     }
970 
971     rerror = qemu_opt_get(legacy_opts, "rerror");
972     if (rerror != NULL) {
973         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
974             type != IF_NONE) {
975             error_report("rerror is not supported by this bus type");
976             goto fail;
977         }
978         qdict_put(bs_opts, "rerror", qstring_from_str(rerror));
979     }
980 
981     /* Actual block device init: Functionality shared with blockdev-add */
982     blk = blockdev_init(filename, bs_opts, &local_err);
983     bs_opts = NULL;
984     if (!blk) {
985         if (local_err) {
986             error_report_err(local_err);
987         }
988         goto fail;
989     } else {
990         assert(!local_err);
991     }
992 
993     /* Create legacy DriveInfo */
994     dinfo = g_malloc0(sizeof(*dinfo));
995     dinfo->opts = all_opts;
996 
997     dinfo->cyls = cyls;
998     dinfo->heads = heads;
999     dinfo->secs = secs;
1000     dinfo->trans = translation;
1001 
1002     dinfo->type = type;
1003     dinfo->bus = bus_id;
1004     dinfo->unit = unit_id;
1005     dinfo->devaddr = devaddr;
1006     dinfo->serial = g_strdup(serial);
1007 
1008     blk_set_legacy_dinfo(blk, dinfo);
1009 
1010     switch(type) {
1011     case IF_IDE:
1012     case IF_SCSI:
1013     case IF_XEN:
1014     case IF_NONE:
1015         dinfo->media_cd = media == MEDIA_CDROM;
1016         break;
1017     default:
1018         break;
1019     }
1020 
1021 fail:
1022     qemu_opts_del(legacy_opts);
1023     QDECREF(bs_opts);
1024     return dinfo;
1025 }
1026 
1027 void hmp_commit(Monitor *mon, const QDict *qdict)
1028 {
1029     const char *device = qdict_get_str(qdict, "device");
1030     BlockBackend *blk;
1031     int ret;
1032 
1033     if (!strcmp(device, "all")) {
1034         ret = bdrv_commit_all();
1035     } else {
1036         blk = blk_by_name(device);
1037         if (!blk) {
1038             monitor_printf(mon, "Device '%s' not found\n", device);
1039             return;
1040         }
1041         ret = bdrv_commit(blk_bs(blk));
1042     }
1043     if (ret < 0) {
1044         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1045                        strerror(-ret));
1046     }
1047 }
1048 
1049 static void blockdev_do_action(int kind, void *data, Error **errp)
1050 {
1051     TransactionAction action;
1052     TransactionActionList list;
1053 
1054     action.kind = kind;
1055     action.data = data;
1056     list.value = &action;
1057     list.next = NULL;
1058     qmp_transaction(&list, errp);
1059 }
1060 
1061 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1062                                 bool has_node_name, const char *node_name,
1063                                 const char *snapshot_file,
1064                                 bool has_snapshot_node_name,
1065                                 const char *snapshot_node_name,
1066                                 bool has_format, const char *format,
1067                                 bool has_mode, NewImageMode mode, Error **errp)
1068 {
1069     BlockdevSnapshot snapshot = {
1070         .has_device = has_device,
1071         .device = (char *) device,
1072         .has_node_name = has_node_name,
1073         .node_name = (char *) node_name,
1074         .snapshot_file = (char *) snapshot_file,
1075         .has_snapshot_node_name = has_snapshot_node_name,
1076         .snapshot_node_name = (char *) snapshot_node_name,
1077         .has_format = has_format,
1078         .format = (char *) format,
1079         .has_mode = has_mode,
1080         .mode = mode,
1081     };
1082     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1083                        &snapshot, errp);
1084 }
1085 
1086 void qmp_blockdev_snapshot_internal_sync(const char *device,
1087                                          const char *name,
1088                                          Error **errp)
1089 {
1090     BlockdevSnapshotInternal snapshot = {
1091         .device = (char *) device,
1092         .name = (char *) name
1093     };
1094 
1095     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1096                        &snapshot, errp);
1097 }
1098 
1099 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1100                                                          bool has_id,
1101                                                          const char *id,
1102                                                          bool has_name,
1103                                                          const char *name,
1104                                                          Error **errp)
1105 {
1106     BlockDriverState *bs;
1107     BlockBackend *blk;
1108     AioContext *aio_context;
1109     QEMUSnapshotInfo sn;
1110     Error *local_err = NULL;
1111     SnapshotInfo *info = NULL;
1112     int ret;
1113 
1114     blk = blk_by_name(device);
1115     if (!blk) {
1116         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1117                   "Device '%s' not found", device);
1118         return NULL;
1119     }
1120     bs = blk_bs(blk);
1121 
1122     if (!has_id) {
1123         id = NULL;
1124     }
1125 
1126     if (!has_name) {
1127         name = NULL;
1128     }
1129 
1130     if (!id && !name) {
1131         error_setg(errp, "Name or id must be provided");
1132         return NULL;
1133     }
1134 
1135     aio_context = bdrv_get_aio_context(bs);
1136     aio_context_acquire(aio_context);
1137 
1138     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1139         goto out_aio_context;
1140     }
1141 
1142     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1143     if (local_err) {
1144         error_propagate(errp, local_err);
1145         goto out_aio_context;
1146     }
1147     if (!ret) {
1148         error_setg(errp,
1149                    "Snapshot with id '%s' and name '%s' does not exist on "
1150                    "device '%s'",
1151                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1152         goto out_aio_context;
1153     }
1154 
1155     bdrv_snapshot_delete(bs, id, name, &local_err);
1156     if (local_err) {
1157         error_propagate(errp, local_err);
1158         goto out_aio_context;
1159     }
1160 
1161     aio_context_release(aio_context);
1162 
1163     info = g_new0(SnapshotInfo, 1);
1164     info->id = g_strdup(sn.id_str);
1165     info->name = g_strdup(sn.name);
1166     info->date_nsec = sn.date_nsec;
1167     info->date_sec = sn.date_sec;
1168     info->vm_state_size = sn.vm_state_size;
1169     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1170     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1171 
1172     return info;
1173 
1174 out_aio_context:
1175     aio_context_release(aio_context);
1176     return NULL;
1177 }
1178 
1179 /**
1180  * block_dirty_bitmap_lookup:
1181  * Return a dirty bitmap (if present), after validating
1182  * the node reference and bitmap names.
1183  *
1184  * @node: The name of the BDS node to search for bitmaps
1185  * @name: The name of the bitmap to search for
1186  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1187  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1188  * @errp: Output pointer for error information. Can be NULL.
1189  *
1190  * @return: A bitmap object on success, or NULL on failure.
1191  */
1192 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1193                                                   const char *name,
1194                                                   BlockDriverState **pbs,
1195                                                   AioContext **paio,
1196                                                   Error **errp)
1197 {
1198     BlockDriverState *bs;
1199     BdrvDirtyBitmap *bitmap;
1200     AioContext *aio_context;
1201 
1202     if (!node) {
1203         error_setg(errp, "Node cannot be NULL");
1204         return NULL;
1205     }
1206     if (!name) {
1207         error_setg(errp, "Bitmap name cannot be NULL");
1208         return NULL;
1209     }
1210     bs = bdrv_lookup_bs(node, node, NULL);
1211     if (!bs) {
1212         error_setg(errp, "Node '%s' not found", node);
1213         return NULL;
1214     }
1215 
1216     aio_context = bdrv_get_aio_context(bs);
1217     aio_context_acquire(aio_context);
1218 
1219     bitmap = bdrv_find_dirty_bitmap(bs, name);
1220     if (!bitmap) {
1221         error_setg(errp, "Dirty bitmap '%s' not found", name);
1222         goto fail;
1223     }
1224 
1225     if (pbs) {
1226         *pbs = bs;
1227     }
1228     if (paio) {
1229         *paio = aio_context;
1230     } else {
1231         aio_context_release(aio_context);
1232     }
1233 
1234     return bitmap;
1235 
1236  fail:
1237     aio_context_release(aio_context);
1238     return NULL;
1239 }
1240 
1241 /* New and old BlockDriverState structs for atomic group operations */
1242 
1243 typedef struct BlkTransactionState BlkTransactionState;
1244 
1245 /* Only prepare() may fail. In a single transaction, only one of commit() or
1246    abort() will be called, clean() will always be called if it present. */
1247 typedef struct BdrvActionOps {
1248     /* Size of state struct, in bytes. */
1249     size_t instance_size;
1250     /* Prepare the work, must NOT be NULL. */
1251     void (*prepare)(BlkTransactionState *common, Error **errp);
1252     /* Commit the changes, can be NULL. */
1253     void (*commit)(BlkTransactionState *common);
1254     /* Abort the changes on fail, can be NULL. */
1255     void (*abort)(BlkTransactionState *common);
1256     /* Clean up resource in the end, can be NULL. */
1257     void (*clean)(BlkTransactionState *common);
1258 } BdrvActionOps;
1259 
1260 /*
1261  * This structure must be arranged as first member in child type, assuming
1262  * that compiler will also arrange it to the same address with parent instance.
1263  * Later it will be used in free().
1264  */
1265 struct BlkTransactionState {
1266     TransactionAction *action;
1267     const BdrvActionOps *ops;
1268     QSIMPLEQ_ENTRY(BlkTransactionState) entry;
1269 };
1270 
1271 /* internal snapshot private data */
1272 typedef struct InternalSnapshotState {
1273     BlkTransactionState common;
1274     BlockDriverState *bs;
1275     AioContext *aio_context;
1276     QEMUSnapshotInfo sn;
1277 } InternalSnapshotState;
1278 
1279 static void internal_snapshot_prepare(BlkTransactionState *common,
1280                                       Error **errp)
1281 {
1282     Error *local_err = NULL;
1283     const char *device;
1284     const char *name;
1285     BlockBackend *blk;
1286     BlockDriverState *bs;
1287     QEMUSnapshotInfo old_sn, *sn;
1288     bool ret;
1289     qemu_timeval tv;
1290     BlockdevSnapshotInternal *internal;
1291     InternalSnapshotState *state;
1292     int ret1;
1293 
1294     g_assert(common->action->kind ==
1295              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1296     internal = common->action->blockdev_snapshot_internal_sync;
1297     state = DO_UPCAST(InternalSnapshotState, common, common);
1298 
1299     /* 1. parse input */
1300     device = internal->device;
1301     name = internal->name;
1302 
1303     /* 2. check for validation */
1304     blk = blk_by_name(device);
1305     if (!blk) {
1306         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1307                   "Device '%s' not found", device);
1308         return;
1309     }
1310     bs = blk_bs(blk);
1311 
1312     /* AioContext is released in .clean() */
1313     state->aio_context = bdrv_get_aio_context(bs);
1314     aio_context_acquire(state->aio_context);
1315 
1316     if (!bdrv_is_inserted(bs)) {
1317         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1318         return;
1319     }
1320 
1321     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1322         return;
1323     }
1324 
1325     if (bdrv_is_read_only(bs)) {
1326         error_setg(errp, "Device '%s' is read only", device);
1327         return;
1328     }
1329 
1330     if (!bdrv_can_snapshot(bs)) {
1331         error_setg(errp, "Block format '%s' used by device '%s' "
1332                    "does not support internal snapshots",
1333                    bs->drv->format_name, device);
1334         return;
1335     }
1336 
1337     if (!strlen(name)) {
1338         error_setg(errp, "Name is empty");
1339         return;
1340     }
1341 
1342     /* check whether a snapshot with name exist */
1343     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1344                                             &local_err);
1345     if (local_err) {
1346         error_propagate(errp, local_err);
1347         return;
1348     } else if (ret) {
1349         error_setg(errp,
1350                    "Snapshot with name '%s' already exists on device '%s'",
1351                    name, device);
1352         return;
1353     }
1354 
1355     /* 3. take the snapshot */
1356     sn = &state->sn;
1357     pstrcpy(sn->name, sizeof(sn->name), name);
1358     qemu_gettimeofday(&tv);
1359     sn->date_sec = tv.tv_sec;
1360     sn->date_nsec = tv.tv_usec * 1000;
1361     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1362 
1363     ret1 = bdrv_snapshot_create(bs, sn);
1364     if (ret1 < 0) {
1365         error_setg_errno(errp, -ret1,
1366                          "Failed to create snapshot '%s' on device '%s'",
1367                          name, device);
1368         return;
1369     }
1370 
1371     /* 4. succeed, mark a snapshot is created */
1372     state->bs = bs;
1373 }
1374 
1375 static void internal_snapshot_abort(BlkTransactionState *common)
1376 {
1377     InternalSnapshotState *state =
1378                              DO_UPCAST(InternalSnapshotState, common, common);
1379     BlockDriverState *bs = state->bs;
1380     QEMUSnapshotInfo *sn = &state->sn;
1381     Error *local_error = NULL;
1382 
1383     if (!bs) {
1384         return;
1385     }
1386 
1387     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1388         error_report("Failed to delete snapshot with id '%s' and name '%s' on "
1389                      "device '%s' in abort: %s",
1390                      sn->id_str,
1391                      sn->name,
1392                      bdrv_get_device_name(bs),
1393                      error_get_pretty(local_error));
1394         error_free(local_error);
1395     }
1396 }
1397 
1398 static void internal_snapshot_clean(BlkTransactionState *common)
1399 {
1400     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1401                                              common, common);
1402 
1403     if (state->aio_context) {
1404         aio_context_release(state->aio_context);
1405     }
1406 }
1407 
1408 /* external snapshot private data */
1409 typedef struct ExternalSnapshotState {
1410     BlkTransactionState common;
1411     BlockDriverState *old_bs;
1412     BlockDriverState *new_bs;
1413     AioContext *aio_context;
1414 } ExternalSnapshotState;
1415 
1416 static void external_snapshot_prepare(BlkTransactionState *common,
1417                                       Error **errp)
1418 {
1419     int flags, ret;
1420     QDict *options;
1421     Error *local_err = NULL;
1422     bool has_device = false;
1423     const char *device;
1424     bool has_node_name = false;
1425     const char *node_name;
1426     bool has_snapshot_node_name = false;
1427     const char *snapshot_node_name;
1428     const char *new_image_file;
1429     const char *format = "qcow2";
1430     enum NewImageMode mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1431     ExternalSnapshotState *state =
1432                              DO_UPCAST(ExternalSnapshotState, common, common);
1433     TransactionAction *action = common->action;
1434 
1435     /* get parameters */
1436     g_assert(action->kind == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC);
1437 
1438     has_device = action->blockdev_snapshot_sync->has_device;
1439     device = action->blockdev_snapshot_sync->device;
1440     has_node_name = action->blockdev_snapshot_sync->has_node_name;
1441     node_name = action->blockdev_snapshot_sync->node_name;
1442     has_snapshot_node_name =
1443         action->blockdev_snapshot_sync->has_snapshot_node_name;
1444     snapshot_node_name = action->blockdev_snapshot_sync->snapshot_node_name;
1445 
1446     new_image_file = action->blockdev_snapshot_sync->snapshot_file;
1447     if (action->blockdev_snapshot_sync->has_format) {
1448         format = action->blockdev_snapshot_sync->format;
1449     }
1450     if (action->blockdev_snapshot_sync->has_mode) {
1451         mode = action->blockdev_snapshot_sync->mode;
1452     }
1453 
1454     /* start processing */
1455     state->old_bs = bdrv_lookup_bs(has_device ? device : NULL,
1456                                    has_node_name ? node_name : NULL,
1457                                    &local_err);
1458     if (local_err) {
1459         error_propagate(errp, local_err);
1460         return;
1461     }
1462 
1463     if (has_node_name && !has_snapshot_node_name) {
1464         error_setg(errp, "New snapshot node name missing");
1465         return;
1466     }
1467 
1468     if (has_snapshot_node_name && bdrv_find_node(snapshot_node_name)) {
1469         error_setg(errp, "New snapshot node name already existing");
1470         return;
1471     }
1472 
1473     /* Acquire AioContext now so any threads operating on old_bs stop */
1474     state->aio_context = bdrv_get_aio_context(state->old_bs);
1475     aio_context_acquire(state->aio_context);
1476 
1477     if (!bdrv_is_inserted(state->old_bs)) {
1478         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1479         return;
1480     }
1481 
1482     if (bdrv_op_is_blocked(state->old_bs,
1483                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1484         return;
1485     }
1486 
1487     if (!bdrv_is_read_only(state->old_bs)) {
1488         if (bdrv_flush(state->old_bs)) {
1489             error_setg(errp, QERR_IO_ERROR);
1490             return;
1491         }
1492     }
1493 
1494     if (!bdrv_is_first_non_filter(state->old_bs)) {
1495         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1496         return;
1497     }
1498 
1499     flags = state->old_bs->open_flags;
1500 
1501     /* create new image w/backing file */
1502     if (mode != NEW_IMAGE_MODE_EXISTING) {
1503         bdrv_img_create(new_image_file, format,
1504                         state->old_bs->filename,
1505                         state->old_bs->drv->format_name,
1506                         NULL, -1, flags, &local_err, false);
1507         if (local_err) {
1508             error_propagate(errp, local_err);
1509             return;
1510         }
1511     }
1512 
1513     options = qdict_new();
1514     if (has_snapshot_node_name) {
1515         qdict_put(options, "node-name",
1516                   qstring_from_str(snapshot_node_name));
1517     }
1518     qdict_put(options, "driver", qstring_from_str(format));
1519 
1520     /* TODO Inherit bs->options or only take explicit options with an
1521      * extended QMP command? */
1522     assert(state->new_bs == NULL);
1523     ret = bdrv_open(&state->new_bs, new_image_file, NULL, options,
1524                     flags | BDRV_O_NO_BACKING, &local_err);
1525     /* We will manually add the backing_hd field to the bs later */
1526     if (ret != 0) {
1527         error_propagate(errp, local_err);
1528     }
1529 }
1530 
1531 static void external_snapshot_commit(BlkTransactionState *common)
1532 {
1533     ExternalSnapshotState *state =
1534                              DO_UPCAST(ExternalSnapshotState, common, common);
1535 
1536     bdrv_set_aio_context(state->new_bs, state->aio_context);
1537 
1538     /* This removes our old bs and adds the new bs */
1539     bdrv_append(state->new_bs, state->old_bs);
1540     /* We don't need (or want) to use the transactional
1541      * bdrv_reopen_multiple() across all the entries at once, because we
1542      * don't want to abort all of them if one of them fails the reopen */
1543     bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1544                 NULL);
1545 
1546     aio_context_release(state->aio_context);
1547 }
1548 
1549 static void external_snapshot_abort(BlkTransactionState *common)
1550 {
1551     ExternalSnapshotState *state =
1552                              DO_UPCAST(ExternalSnapshotState, common, common);
1553     if (state->new_bs) {
1554         bdrv_unref(state->new_bs);
1555     }
1556     if (state->aio_context) {
1557         aio_context_release(state->aio_context);
1558     }
1559 }
1560 
1561 typedef struct DriveBackupState {
1562     BlkTransactionState common;
1563     BlockDriverState *bs;
1564     AioContext *aio_context;
1565     BlockJob *job;
1566 } DriveBackupState;
1567 
1568 static void drive_backup_prepare(BlkTransactionState *common, Error **errp)
1569 {
1570     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1571     BlockDriverState *bs;
1572     BlockBackend *blk;
1573     DriveBackup *backup;
1574     Error *local_err = NULL;
1575 
1576     assert(common->action->kind == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1577     backup = common->action->drive_backup;
1578 
1579     blk = blk_by_name(backup->device);
1580     if (!blk) {
1581         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1582                   "Device '%s' not found", backup->device);
1583         return;
1584     }
1585     bs = blk_bs(blk);
1586 
1587     /* AioContext is released in .clean() */
1588     state->aio_context = bdrv_get_aio_context(bs);
1589     aio_context_acquire(state->aio_context);
1590 
1591     qmp_drive_backup(backup->device, backup->target,
1592                      backup->has_format, backup->format,
1593                      backup->sync,
1594                      backup->has_mode, backup->mode,
1595                      backup->has_speed, backup->speed,
1596                      backup->has_bitmap, backup->bitmap,
1597                      backup->has_on_source_error, backup->on_source_error,
1598                      backup->has_on_target_error, backup->on_target_error,
1599                      &local_err);
1600     if (local_err) {
1601         error_propagate(errp, local_err);
1602         return;
1603     }
1604 
1605     state->bs = bs;
1606     state->job = state->bs->job;
1607 }
1608 
1609 static void drive_backup_abort(BlkTransactionState *common)
1610 {
1611     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1612     BlockDriverState *bs = state->bs;
1613 
1614     /* Only cancel if it's the job we started */
1615     if (bs && bs->job && bs->job == state->job) {
1616         block_job_cancel_sync(bs->job);
1617     }
1618 }
1619 
1620 static void drive_backup_clean(BlkTransactionState *common)
1621 {
1622     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1623 
1624     if (state->aio_context) {
1625         aio_context_release(state->aio_context);
1626     }
1627 }
1628 
1629 typedef struct BlockdevBackupState {
1630     BlkTransactionState common;
1631     BlockDriverState *bs;
1632     BlockJob *job;
1633     AioContext *aio_context;
1634 } BlockdevBackupState;
1635 
1636 static void blockdev_backup_prepare(BlkTransactionState *common, Error **errp)
1637 {
1638     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1639     BlockdevBackup *backup;
1640     BlockDriverState *bs, *target;
1641     BlockBackend *blk;
1642     Error *local_err = NULL;
1643 
1644     assert(common->action->kind == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1645     backup = common->action->blockdev_backup;
1646 
1647     blk = blk_by_name(backup->device);
1648     if (!blk) {
1649         error_setg(errp, "Device '%s' not found", backup->device);
1650         return;
1651     }
1652     bs = blk_bs(blk);
1653 
1654     blk = blk_by_name(backup->target);
1655     if (!blk) {
1656         error_setg(errp, "Device '%s' not found", backup->target);
1657         return;
1658     }
1659     target = blk_bs(blk);
1660 
1661     /* AioContext is released in .clean() */
1662     state->aio_context = bdrv_get_aio_context(bs);
1663     if (state->aio_context != bdrv_get_aio_context(target)) {
1664         state->aio_context = NULL;
1665         error_setg(errp, "Backup between two IO threads is not implemented");
1666         return;
1667     }
1668     aio_context_acquire(state->aio_context);
1669 
1670     qmp_blockdev_backup(backup->device, backup->target,
1671                         backup->sync,
1672                         backup->has_speed, backup->speed,
1673                         backup->has_on_source_error, backup->on_source_error,
1674                         backup->has_on_target_error, backup->on_target_error,
1675                         &local_err);
1676     if (local_err) {
1677         error_propagate(errp, local_err);
1678         return;
1679     }
1680 
1681     state->bs = bs;
1682     state->job = state->bs->job;
1683 }
1684 
1685 static void blockdev_backup_abort(BlkTransactionState *common)
1686 {
1687     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1688     BlockDriverState *bs = state->bs;
1689 
1690     /* Only cancel if it's the job we started */
1691     if (bs && bs->job && bs->job == state->job) {
1692         block_job_cancel_sync(bs->job);
1693     }
1694 }
1695 
1696 static void blockdev_backup_clean(BlkTransactionState *common)
1697 {
1698     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1699 
1700     if (state->aio_context) {
1701         aio_context_release(state->aio_context);
1702     }
1703 }
1704 
1705 static void abort_prepare(BlkTransactionState *common, Error **errp)
1706 {
1707     error_setg(errp, "Transaction aborted using Abort action");
1708 }
1709 
1710 static void abort_commit(BlkTransactionState *common)
1711 {
1712     g_assert_not_reached(); /* this action never succeeds */
1713 }
1714 
1715 static const BdrvActionOps actions[] = {
1716     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
1717         .instance_size = sizeof(ExternalSnapshotState),
1718         .prepare  = external_snapshot_prepare,
1719         .commit   = external_snapshot_commit,
1720         .abort = external_snapshot_abort,
1721     },
1722     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
1723         .instance_size = sizeof(DriveBackupState),
1724         .prepare = drive_backup_prepare,
1725         .abort = drive_backup_abort,
1726         .clean = drive_backup_clean,
1727     },
1728     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
1729         .instance_size = sizeof(BlockdevBackupState),
1730         .prepare = blockdev_backup_prepare,
1731         .abort = blockdev_backup_abort,
1732         .clean = blockdev_backup_clean,
1733     },
1734     [TRANSACTION_ACTION_KIND_ABORT] = {
1735         .instance_size = sizeof(BlkTransactionState),
1736         .prepare = abort_prepare,
1737         .commit = abort_commit,
1738     },
1739     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
1740         .instance_size = sizeof(InternalSnapshotState),
1741         .prepare  = internal_snapshot_prepare,
1742         .abort = internal_snapshot_abort,
1743         .clean = internal_snapshot_clean,
1744     },
1745 };
1746 
1747 /*
1748  * 'Atomic' group operations.  The operations are performed as a set, and if
1749  * any fail then we roll back all operations in the group.
1750  */
1751 void qmp_transaction(TransactionActionList *dev_list, Error **errp)
1752 {
1753     TransactionActionList *dev_entry = dev_list;
1754     BlkTransactionState *state, *next;
1755     Error *local_err = NULL;
1756 
1757     QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionState) snap_bdrv_states;
1758     QSIMPLEQ_INIT(&snap_bdrv_states);
1759 
1760     /* drain all i/o before any operations */
1761     bdrv_drain_all();
1762 
1763     /* We don't do anything in this loop that commits us to the operations */
1764     while (NULL != dev_entry) {
1765         TransactionAction *dev_info = NULL;
1766         const BdrvActionOps *ops;
1767 
1768         dev_info = dev_entry->value;
1769         dev_entry = dev_entry->next;
1770 
1771         assert(dev_info->kind < ARRAY_SIZE(actions));
1772 
1773         ops = &actions[dev_info->kind];
1774         assert(ops->instance_size > 0);
1775 
1776         state = g_malloc0(ops->instance_size);
1777         state->ops = ops;
1778         state->action = dev_info;
1779         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
1780 
1781         state->ops->prepare(state, &local_err);
1782         if (local_err) {
1783             error_propagate(errp, local_err);
1784             goto delete_and_fail;
1785         }
1786     }
1787 
1788     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1789         if (state->ops->commit) {
1790             state->ops->commit(state);
1791         }
1792     }
1793 
1794     /* success */
1795     goto exit;
1796 
1797 delete_and_fail:
1798     /* failure, and it is all-or-none; roll back all operations */
1799     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1800         if (state->ops->abort) {
1801             state->ops->abort(state);
1802         }
1803     }
1804 exit:
1805     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
1806         if (state->ops->clean) {
1807             state->ops->clean(state);
1808         }
1809         g_free(state);
1810     }
1811 }
1812 
1813 
1814 static void eject_device(BlockBackend *blk, int force, Error **errp)
1815 {
1816     BlockDriverState *bs = blk_bs(blk);
1817     AioContext *aio_context;
1818 
1819     aio_context = bdrv_get_aio_context(bs);
1820     aio_context_acquire(aio_context);
1821 
1822     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
1823         goto out;
1824     }
1825     if (!blk_dev_has_removable_media(blk)) {
1826         error_setg(errp, "Device '%s' is not removable",
1827                    bdrv_get_device_name(bs));
1828         goto out;
1829     }
1830 
1831     if (blk_dev_is_medium_locked(blk) && !blk_dev_is_tray_open(blk)) {
1832         blk_dev_eject_request(blk, force);
1833         if (!force) {
1834             error_setg(errp, "Device '%s' is locked",
1835                        bdrv_get_device_name(bs));
1836             goto out;
1837         }
1838     }
1839 
1840     bdrv_close(bs);
1841 
1842 out:
1843     aio_context_release(aio_context);
1844 }
1845 
1846 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
1847 {
1848     BlockBackend *blk;
1849 
1850     blk = blk_by_name(device);
1851     if (!blk) {
1852         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1853                   "Device '%s' not found", device);
1854         return;
1855     }
1856 
1857     eject_device(blk, force, errp);
1858 }
1859 
1860 void qmp_block_passwd(bool has_device, const char *device,
1861                       bool has_node_name, const char *node_name,
1862                       const char *password, Error **errp)
1863 {
1864     Error *local_err = NULL;
1865     BlockDriverState *bs;
1866     AioContext *aio_context;
1867 
1868     bs = bdrv_lookup_bs(has_device ? device : NULL,
1869                         has_node_name ? node_name : NULL,
1870                         &local_err);
1871     if (local_err) {
1872         error_propagate(errp, local_err);
1873         return;
1874     }
1875 
1876     aio_context = bdrv_get_aio_context(bs);
1877     aio_context_acquire(aio_context);
1878 
1879     bdrv_add_key(bs, password, errp);
1880 
1881     aio_context_release(aio_context);
1882 }
1883 
1884 /* Assumes AioContext is held */
1885 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename,
1886                                     int bdrv_flags, const char *format,
1887                                     const char *password, Error **errp)
1888 {
1889     Error *local_err = NULL;
1890     QDict *options = NULL;
1891     int ret;
1892 
1893     if (format) {
1894         options = qdict_new();
1895         qdict_put(options, "driver", qstring_from_str(format));
1896     }
1897 
1898     ret = bdrv_open(&bs, filename, NULL, options, bdrv_flags, &local_err);
1899     if (ret < 0) {
1900         error_propagate(errp, local_err);
1901         return;
1902     }
1903 
1904     bdrv_add_key(bs, password, errp);
1905 }
1906 
1907 void qmp_change_blockdev(const char *device, const char *filename,
1908                          const char *format, Error **errp)
1909 {
1910     BlockBackend *blk;
1911     BlockDriverState *bs;
1912     AioContext *aio_context;
1913     int bdrv_flags;
1914     Error *err = NULL;
1915 
1916     blk = blk_by_name(device);
1917     if (!blk) {
1918         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1919                   "Device '%s' not found", device);
1920         return;
1921     }
1922     bs = blk_bs(blk);
1923 
1924     aio_context = bdrv_get_aio_context(bs);
1925     aio_context_acquire(aio_context);
1926 
1927     eject_device(blk, 0, &err);
1928     if (err) {
1929         error_propagate(errp, err);
1930         goto out;
1931     }
1932 
1933     bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
1934     bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
1935 
1936     qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, format, NULL, errp);
1937 
1938 out:
1939     aio_context_release(aio_context);
1940 }
1941 
1942 /* throttling disk I/O limits */
1943 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
1944                                int64_t bps_wr,
1945                                int64_t iops,
1946                                int64_t iops_rd,
1947                                int64_t iops_wr,
1948                                bool has_bps_max,
1949                                int64_t bps_max,
1950                                bool has_bps_rd_max,
1951                                int64_t bps_rd_max,
1952                                bool has_bps_wr_max,
1953                                int64_t bps_wr_max,
1954                                bool has_iops_max,
1955                                int64_t iops_max,
1956                                bool has_iops_rd_max,
1957                                int64_t iops_rd_max,
1958                                bool has_iops_wr_max,
1959                                int64_t iops_wr_max,
1960                                bool has_iops_size,
1961                                int64_t iops_size,
1962                                bool has_group,
1963                                const char *group, Error **errp)
1964 {
1965     ThrottleConfig cfg;
1966     BlockDriverState *bs;
1967     BlockBackend *blk;
1968     AioContext *aio_context;
1969 
1970     blk = blk_by_name(device);
1971     if (!blk) {
1972         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1973                   "Device '%s' not found", device);
1974         return;
1975     }
1976     bs = blk_bs(blk);
1977 
1978     memset(&cfg, 0, sizeof(cfg));
1979     cfg.buckets[THROTTLE_BPS_TOTAL].avg = bps;
1980     cfg.buckets[THROTTLE_BPS_READ].avg  = bps_rd;
1981     cfg.buckets[THROTTLE_BPS_WRITE].avg = bps_wr;
1982 
1983     cfg.buckets[THROTTLE_OPS_TOTAL].avg = iops;
1984     cfg.buckets[THROTTLE_OPS_READ].avg  = iops_rd;
1985     cfg.buckets[THROTTLE_OPS_WRITE].avg = iops_wr;
1986 
1987     if (has_bps_max) {
1988         cfg.buckets[THROTTLE_BPS_TOTAL].max = bps_max;
1989     }
1990     if (has_bps_rd_max) {
1991         cfg.buckets[THROTTLE_BPS_READ].max = bps_rd_max;
1992     }
1993     if (has_bps_wr_max) {
1994         cfg.buckets[THROTTLE_BPS_WRITE].max = bps_wr_max;
1995     }
1996     if (has_iops_max) {
1997         cfg.buckets[THROTTLE_OPS_TOTAL].max = iops_max;
1998     }
1999     if (has_iops_rd_max) {
2000         cfg.buckets[THROTTLE_OPS_READ].max = iops_rd_max;
2001     }
2002     if (has_iops_wr_max) {
2003         cfg.buckets[THROTTLE_OPS_WRITE].max = iops_wr_max;
2004     }
2005 
2006     if (has_iops_size) {
2007         cfg.op_size = iops_size;
2008     }
2009 
2010     if (!check_throttle_config(&cfg, errp)) {
2011         return;
2012     }
2013 
2014     aio_context = bdrv_get_aio_context(bs);
2015     aio_context_acquire(aio_context);
2016 
2017     if (throttle_enabled(&cfg)) {
2018         /* Enable I/O limits if they're not enabled yet, otherwise
2019          * just update the throttling group. */
2020         if (!bs->io_limits_enabled) {
2021             bdrv_io_limits_enable(bs, has_group ? group : device);
2022         } else if (has_group) {
2023             bdrv_io_limits_update_group(bs, group);
2024         }
2025         /* Set the new throttling configuration */
2026         bdrv_set_io_limits(bs, &cfg);
2027     } else if (bs->io_limits_enabled) {
2028         /* If all throttling settings are set to 0, disable I/O limits */
2029         bdrv_io_limits_disable(bs);
2030     }
2031 
2032     aio_context_release(aio_context);
2033 }
2034 
2035 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2036                                 bool has_granularity, uint32_t granularity,
2037                                 Error **errp)
2038 {
2039     AioContext *aio_context;
2040     BlockDriverState *bs;
2041 
2042     if (!name || name[0] == '\0') {
2043         error_setg(errp, "Bitmap name cannot be empty");
2044         return;
2045     }
2046 
2047     bs = bdrv_lookup_bs(node, node, errp);
2048     if (!bs) {
2049         return;
2050     }
2051 
2052     aio_context = bdrv_get_aio_context(bs);
2053     aio_context_acquire(aio_context);
2054 
2055     if (has_granularity) {
2056         if (granularity < 512 || !is_power_of_2(granularity)) {
2057             error_setg(errp, "Granularity must be power of 2 "
2058                              "and at least 512");
2059             goto out;
2060         }
2061     } else {
2062         /* Default to cluster size, if available: */
2063         granularity = bdrv_get_default_bitmap_granularity(bs);
2064     }
2065 
2066     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2067 
2068  out:
2069     aio_context_release(aio_context);
2070 }
2071 
2072 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2073                                    Error **errp)
2074 {
2075     AioContext *aio_context;
2076     BlockDriverState *bs;
2077     BdrvDirtyBitmap *bitmap;
2078 
2079     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2080     if (!bitmap || !bs) {
2081         return;
2082     }
2083 
2084     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2085         error_setg(errp,
2086                    "Bitmap '%s' is currently frozen and cannot be removed",
2087                    name);
2088         goto out;
2089     }
2090     bdrv_dirty_bitmap_make_anon(bitmap);
2091     bdrv_release_dirty_bitmap(bs, bitmap);
2092 
2093  out:
2094     aio_context_release(aio_context);
2095 }
2096 
2097 /**
2098  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2099  * immediately after a full backup operation.
2100  */
2101 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2102                                   Error **errp)
2103 {
2104     AioContext *aio_context;
2105     BdrvDirtyBitmap *bitmap;
2106     BlockDriverState *bs;
2107 
2108     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2109     if (!bitmap || !bs) {
2110         return;
2111     }
2112 
2113     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2114         error_setg(errp,
2115                    "Bitmap '%s' is currently frozen and cannot be modified",
2116                    name);
2117         goto out;
2118     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2119         error_setg(errp,
2120                    "Bitmap '%s' is currently disabled and cannot be cleared",
2121                    name);
2122         goto out;
2123     }
2124 
2125     bdrv_clear_dirty_bitmap(bitmap);
2126 
2127  out:
2128     aio_context_release(aio_context);
2129 }
2130 
2131 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2132 {
2133     const char *id = qdict_get_str(qdict, "id");
2134     BlockBackend *blk;
2135     BlockDriverState *bs;
2136     AioContext *aio_context;
2137     Error *local_err = NULL;
2138 
2139     blk = blk_by_name(id);
2140     if (!blk) {
2141         error_report("Device '%s' not found", id);
2142         return;
2143     }
2144     bs = blk_bs(blk);
2145 
2146     if (!blk_legacy_dinfo(blk)) {
2147         error_report("Deleting device added with blockdev-add"
2148                      " is not supported");
2149         return;
2150     }
2151 
2152     aio_context = bdrv_get_aio_context(bs);
2153     aio_context_acquire(aio_context);
2154 
2155     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2156         error_report_err(local_err);
2157         aio_context_release(aio_context);
2158         return;
2159     }
2160 
2161     bdrv_close(bs);
2162 
2163     /* if we have a device attached to this BlockDriverState
2164      * then we need to make the drive anonymous until the device
2165      * can be removed.  If this is a drive with no device backing
2166      * then we can just get rid of the block driver state right here.
2167      */
2168     if (blk_get_attached_dev(blk)) {
2169         blk_hide_on_behalf_of_hmp_drive_del(blk);
2170         /* Further I/O must not pause the guest */
2171         bdrv_set_on_error(bs, BLOCKDEV_ON_ERROR_REPORT,
2172                           BLOCKDEV_ON_ERROR_REPORT);
2173     } else {
2174         blk_unref(blk);
2175     }
2176 
2177     aio_context_release(aio_context);
2178 }
2179 
2180 void qmp_block_resize(bool has_device, const char *device,
2181                       bool has_node_name, const char *node_name,
2182                       int64_t size, Error **errp)
2183 {
2184     Error *local_err = NULL;
2185     BlockDriverState *bs;
2186     AioContext *aio_context;
2187     int ret;
2188 
2189     bs = bdrv_lookup_bs(has_device ? device : NULL,
2190                         has_node_name ? node_name : NULL,
2191                         &local_err);
2192     if (local_err) {
2193         error_propagate(errp, local_err);
2194         return;
2195     }
2196 
2197     aio_context = bdrv_get_aio_context(bs);
2198     aio_context_acquire(aio_context);
2199 
2200     if (!bdrv_is_first_non_filter(bs)) {
2201         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2202         goto out;
2203     }
2204 
2205     if (size < 0) {
2206         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2207         goto out;
2208     }
2209 
2210     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2211         error_setg(errp, QERR_DEVICE_IN_USE, device);
2212         goto out;
2213     }
2214 
2215     /* complete all in-flight operations before resizing the device */
2216     bdrv_drain_all();
2217 
2218     ret = bdrv_truncate(bs, size);
2219     switch (ret) {
2220     case 0:
2221         break;
2222     case -ENOMEDIUM:
2223         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2224         break;
2225     case -ENOTSUP:
2226         error_setg(errp, QERR_UNSUPPORTED);
2227         break;
2228     case -EACCES:
2229         error_setg(errp, "Device '%s' is read only", device);
2230         break;
2231     case -EBUSY:
2232         error_setg(errp, QERR_DEVICE_IN_USE, device);
2233         break;
2234     default:
2235         error_setg_errno(errp, -ret, "Could not resize");
2236         break;
2237     }
2238 
2239 out:
2240     aio_context_release(aio_context);
2241 }
2242 
2243 static void block_job_cb(void *opaque, int ret)
2244 {
2245     /* Note that this function may be executed from another AioContext besides
2246      * the QEMU main loop.  If you need to access anything that assumes the
2247      * QEMU global mutex, use a BH or introduce a mutex.
2248      */
2249 
2250     BlockDriverState *bs = opaque;
2251     const char *msg = NULL;
2252 
2253     trace_block_job_cb(bs, bs->job, ret);
2254 
2255     assert(bs->job);
2256 
2257     if (ret < 0) {
2258         msg = strerror(-ret);
2259     }
2260 
2261     if (block_job_is_cancelled(bs->job)) {
2262         block_job_event_cancelled(bs->job);
2263     } else {
2264         block_job_event_completed(bs->job, msg);
2265     }
2266 
2267     bdrv_put_ref_bh_schedule(bs);
2268 }
2269 
2270 void qmp_block_stream(const char *device,
2271                       bool has_base, const char *base,
2272                       bool has_backing_file, const char *backing_file,
2273                       bool has_speed, int64_t speed,
2274                       bool has_on_error, BlockdevOnError on_error,
2275                       Error **errp)
2276 {
2277     BlockBackend *blk;
2278     BlockDriverState *bs;
2279     BlockDriverState *base_bs = NULL;
2280     AioContext *aio_context;
2281     Error *local_err = NULL;
2282     const char *base_name = NULL;
2283 
2284     if (!has_on_error) {
2285         on_error = BLOCKDEV_ON_ERROR_REPORT;
2286     }
2287 
2288     blk = blk_by_name(device);
2289     if (!blk) {
2290         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2291                   "Device '%s' not found", device);
2292         return;
2293     }
2294     bs = blk_bs(blk);
2295 
2296     aio_context = bdrv_get_aio_context(bs);
2297     aio_context_acquire(aio_context);
2298 
2299     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_STREAM, errp)) {
2300         goto out;
2301     }
2302 
2303     if (has_base) {
2304         base_bs = bdrv_find_backing_image(bs, base);
2305         if (base_bs == NULL) {
2306             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2307             goto out;
2308         }
2309         assert(bdrv_get_aio_context(base_bs) == aio_context);
2310         base_name = base;
2311     }
2312 
2313     /* if we are streaming the entire chain, the result will have no backing
2314      * file, and specifying one is therefore an error */
2315     if (base_bs == NULL && has_backing_file) {
2316         error_setg(errp, "backing file specified, but streaming the "
2317                          "entire chain");
2318         goto out;
2319     }
2320 
2321     /* backing_file string overrides base bs filename */
2322     base_name = has_backing_file ? backing_file : base_name;
2323 
2324     stream_start(bs, base_bs, base_name, has_speed ? speed : 0,
2325                  on_error, block_job_cb, bs, &local_err);
2326     if (local_err) {
2327         error_propagate(errp, local_err);
2328         goto out;
2329     }
2330 
2331     trace_qmp_block_stream(bs, bs->job);
2332 
2333 out:
2334     aio_context_release(aio_context);
2335 }
2336 
2337 void qmp_block_commit(const char *device,
2338                       bool has_base, const char *base,
2339                       bool has_top, const char *top,
2340                       bool has_backing_file, const char *backing_file,
2341                       bool has_speed, int64_t speed,
2342                       Error **errp)
2343 {
2344     BlockBackend *blk;
2345     BlockDriverState *bs;
2346     BlockDriverState *base_bs, *top_bs;
2347     AioContext *aio_context;
2348     Error *local_err = NULL;
2349     /* This will be part of the QMP command, if/when the
2350      * BlockdevOnError change for blkmirror makes it in
2351      */
2352     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
2353 
2354     if (!has_speed) {
2355         speed = 0;
2356     }
2357 
2358     /* Important Note:
2359      *  libvirt relies on the DeviceNotFound error class in order to probe for
2360      *  live commit feature versions; for this to work, we must make sure to
2361      *  perform the device lookup before any generic errors that may occur in a
2362      *  scenario in which all optional arguments are omitted. */
2363     blk = blk_by_name(device);
2364     if (!blk) {
2365         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2366                   "Device '%s' not found", device);
2367         return;
2368     }
2369     bs = blk_bs(blk);
2370 
2371     aio_context = bdrv_get_aio_context(bs);
2372     aio_context_acquire(aio_context);
2373 
2374     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2375         goto out;
2376     }
2377 
2378     /* default top_bs is the active layer */
2379     top_bs = bs;
2380 
2381     if (has_top && top) {
2382         if (strcmp(bs->filename, top) != 0) {
2383             top_bs = bdrv_find_backing_image(bs, top);
2384         }
2385     }
2386 
2387     if (top_bs == NULL) {
2388         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2389         goto out;
2390     }
2391 
2392     assert(bdrv_get_aio_context(top_bs) == aio_context);
2393 
2394     if (has_base && base) {
2395         base_bs = bdrv_find_backing_image(top_bs, base);
2396     } else {
2397         base_bs = bdrv_find_base(top_bs);
2398     }
2399 
2400     if (base_bs == NULL) {
2401         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
2402         goto out;
2403     }
2404 
2405     assert(bdrv_get_aio_context(base_bs) == aio_context);
2406 
2407     if (bdrv_op_is_blocked(base_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2408         goto out;
2409     }
2410 
2411     /* Do not allow attempts to commit an image into itself */
2412     if (top_bs == base_bs) {
2413         error_setg(errp, "cannot commit an image into itself");
2414         goto out;
2415     }
2416 
2417     if (top_bs == bs) {
2418         if (has_backing_file) {
2419             error_setg(errp, "'backing-file' specified,"
2420                              " but 'top' is the active layer");
2421             goto out;
2422         }
2423         commit_active_start(bs, base_bs, speed, on_error, block_job_cb,
2424                             bs, &local_err);
2425     } else {
2426         commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
2427                      has_backing_file ? backing_file : NULL, &local_err);
2428     }
2429     if (local_err != NULL) {
2430         error_propagate(errp, local_err);
2431         goto out;
2432     }
2433 
2434 out:
2435     aio_context_release(aio_context);
2436 }
2437 
2438 void qmp_drive_backup(const char *device, const char *target,
2439                       bool has_format, const char *format,
2440                       enum MirrorSyncMode sync,
2441                       bool has_mode, enum NewImageMode mode,
2442                       bool has_speed, int64_t speed,
2443                       bool has_bitmap, const char *bitmap,
2444                       bool has_on_source_error, BlockdevOnError on_source_error,
2445                       bool has_on_target_error, BlockdevOnError on_target_error,
2446                       Error **errp)
2447 {
2448     BlockBackend *blk;
2449     BlockDriverState *bs;
2450     BlockDriverState *target_bs;
2451     BlockDriverState *source = NULL;
2452     BdrvDirtyBitmap *bmap = NULL;
2453     AioContext *aio_context;
2454     QDict *options = NULL;
2455     Error *local_err = NULL;
2456     int flags;
2457     int64_t size;
2458     int ret;
2459 
2460     if (!has_speed) {
2461         speed = 0;
2462     }
2463     if (!has_on_source_error) {
2464         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2465     }
2466     if (!has_on_target_error) {
2467         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2468     }
2469     if (!has_mode) {
2470         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2471     }
2472 
2473     blk = blk_by_name(device);
2474     if (!blk) {
2475         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2476                   "Device '%s' not found", device);
2477         return;
2478     }
2479     bs = blk_bs(blk);
2480 
2481     aio_context = bdrv_get_aio_context(bs);
2482     aio_context_acquire(aio_context);
2483 
2484     /* Although backup_run has this check too, we need to use bs->drv below, so
2485      * do an early check redundantly. */
2486     if (!bdrv_is_inserted(bs)) {
2487         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2488         goto out;
2489     }
2490 
2491     if (!has_format) {
2492         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
2493     }
2494 
2495     /* Early check to avoid creating target */
2496     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
2497         goto out;
2498     }
2499 
2500     flags = bs->open_flags | BDRV_O_RDWR;
2501 
2502     /* See if we have a backing HD we can use to create our new image
2503      * on top of. */
2504     if (sync == MIRROR_SYNC_MODE_TOP) {
2505         source = backing_bs(bs);
2506         if (!source) {
2507             sync = MIRROR_SYNC_MODE_FULL;
2508         }
2509     }
2510     if (sync == MIRROR_SYNC_MODE_NONE) {
2511         source = bs;
2512     }
2513 
2514     size = bdrv_getlength(bs);
2515     if (size < 0) {
2516         error_setg_errno(errp, -size, "bdrv_getlength failed");
2517         goto out;
2518     }
2519 
2520     if (mode != NEW_IMAGE_MODE_EXISTING) {
2521         assert(format);
2522         if (source) {
2523             bdrv_img_create(target, format, source->filename,
2524                             source->drv->format_name, NULL,
2525                             size, flags, &local_err, false);
2526         } else {
2527             bdrv_img_create(target, format, NULL, NULL, NULL,
2528                             size, flags, &local_err, false);
2529         }
2530     }
2531 
2532     if (local_err) {
2533         error_propagate(errp, local_err);
2534         goto out;
2535     }
2536 
2537     if (format) {
2538         options = qdict_new();
2539         qdict_put(options, "driver", qstring_from_str(format));
2540     }
2541 
2542     target_bs = NULL;
2543     ret = bdrv_open(&target_bs, target, NULL, options, flags, &local_err);
2544     if (ret < 0) {
2545         error_propagate(errp, local_err);
2546         goto out;
2547     }
2548 
2549     bdrv_set_aio_context(target_bs, aio_context);
2550 
2551     if (has_bitmap) {
2552         bmap = bdrv_find_dirty_bitmap(bs, bitmap);
2553         if (!bmap) {
2554             error_setg(errp, "Bitmap '%s' could not be found", bitmap);
2555             goto out;
2556         }
2557     }
2558 
2559     backup_start(bs, target_bs, speed, sync, bmap,
2560                  on_source_error, on_target_error,
2561                  block_job_cb, bs, &local_err);
2562     if (local_err != NULL) {
2563         bdrv_unref(target_bs);
2564         error_propagate(errp, local_err);
2565         goto out;
2566     }
2567 
2568 out:
2569     aio_context_release(aio_context);
2570 }
2571 
2572 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
2573 {
2574     return bdrv_named_nodes_list(errp);
2575 }
2576 
2577 void qmp_blockdev_backup(const char *device, const char *target,
2578                          enum MirrorSyncMode sync,
2579                          bool has_speed, int64_t speed,
2580                          bool has_on_source_error,
2581                          BlockdevOnError on_source_error,
2582                          bool has_on_target_error,
2583                          BlockdevOnError on_target_error,
2584                          Error **errp)
2585 {
2586     BlockBackend *blk;
2587     BlockDriverState *bs;
2588     BlockDriverState *target_bs;
2589     Error *local_err = NULL;
2590     AioContext *aio_context;
2591 
2592     if (!has_speed) {
2593         speed = 0;
2594     }
2595     if (!has_on_source_error) {
2596         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2597     }
2598     if (!has_on_target_error) {
2599         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2600     }
2601 
2602     blk = blk_by_name(device);
2603     if (!blk) {
2604         error_setg(errp, "Device '%s' not found", device);
2605         return;
2606     }
2607     bs = blk_bs(blk);
2608 
2609     aio_context = bdrv_get_aio_context(bs);
2610     aio_context_acquire(aio_context);
2611 
2612     blk = blk_by_name(target);
2613     if (!blk) {
2614         error_setg(errp, "Device '%s' not found", target);
2615         goto out;
2616     }
2617     target_bs = blk_bs(blk);
2618 
2619     bdrv_ref(target_bs);
2620     bdrv_set_aio_context(target_bs, aio_context);
2621     backup_start(bs, target_bs, speed, sync, NULL, on_source_error,
2622                  on_target_error, block_job_cb, bs, &local_err);
2623     if (local_err != NULL) {
2624         bdrv_unref(target_bs);
2625         error_propagate(errp, local_err);
2626     }
2627 out:
2628     aio_context_release(aio_context);
2629 }
2630 
2631 void qmp_drive_mirror(const char *device, const char *target,
2632                       bool has_format, const char *format,
2633                       bool has_node_name, const char *node_name,
2634                       bool has_replaces, const char *replaces,
2635                       enum MirrorSyncMode sync,
2636                       bool has_mode, enum NewImageMode mode,
2637                       bool has_speed, int64_t speed,
2638                       bool has_granularity, uint32_t granularity,
2639                       bool has_buf_size, int64_t buf_size,
2640                       bool has_on_source_error, BlockdevOnError on_source_error,
2641                       bool has_on_target_error, BlockdevOnError on_target_error,
2642                       bool has_unmap, bool unmap,
2643                       Error **errp)
2644 {
2645     BlockBackend *blk;
2646     BlockDriverState *bs;
2647     BlockDriverState *source, *target_bs;
2648     AioContext *aio_context;
2649     Error *local_err = NULL;
2650     QDict *options;
2651     int flags;
2652     int64_t size;
2653     int ret;
2654 
2655     if (!has_speed) {
2656         speed = 0;
2657     }
2658     if (!has_on_source_error) {
2659         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2660     }
2661     if (!has_on_target_error) {
2662         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2663     }
2664     if (!has_mode) {
2665         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2666     }
2667     if (!has_granularity) {
2668         granularity = 0;
2669     }
2670     if (!has_buf_size) {
2671         buf_size = 0;
2672     }
2673     if (!has_unmap) {
2674         unmap = true;
2675     }
2676 
2677     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2678         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2679                    "a value in range [512B, 64MB]");
2680         return;
2681     }
2682     if (granularity & (granularity - 1)) {
2683         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2684                    "power of 2");
2685         return;
2686     }
2687 
2688     blk = blk_by_name(device);
2689     if (!blk) {
2690         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2691                   "Device '%s' not found", device);
2692         return;
2693     }
2694     bs = blk_bs(blk);
2695 
2696     aio_context = bdrv_get_aio_context(bs);
2697     aio_context_acquire(aio_context);
2698 
2699     if (!bdrv_is_inserted(bs)) {
2700         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2701         goto out;
2702     }
2703 
2704     if (!has_format) {
2705         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
2706     }
2707 
2708     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR, errp)) {
2709         goto out;
2710     }
2711 
2712     flags = bs->open_flags | BDRV_O_RDWR;
2713     source = backing_bs(bs);
2714     if (!source && sync == MIRROR_SYNC_MODE_TOP) {
2715         sync = MIRROR_SYNC_MODE_FULL;
2716     }
2717     if (sync == MIRROR_SYNC_MODE_NONE) {
2718         source = bs;
2719     }
2720 
2721     size = bdrv_getlength(bs);
2722     if (size < 0) {
2723         error_setg_errno(errp, -size, "bdrv_getlength failed");
2724         goto out;
2725     }
2726 
2727     if (has_replaces) {
2728         BlockDriverState *to_replace_bs;
2729         AioContext *replace_aio_context;
2730         int64_t replace_size;
2731 
2732         if (!has_node_name) {
2733             error_setg(errp, "a node-name must be provided when replacing a"
2734                              " named node of the graph");
2735             goto out;
2736         }
2737 
2738         to_replace_bs = check_to_replace_node(bs, replaces, &local_err);
2739 
2740         if (!to_replace_bs) {
2741             error_propagate(errp, local_err);
2742             goto out;
2743         }
2744 
2745         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
2746         aio_context_acquire(replace_aio_context);
2747         replace_size = bdrv_getlength(to_replace_bs);
2748         aio_context_release(replace_aio_context);
2749 
2750         if (size != replace_size) {
2751             error_setg(errp, "cannot replace image with a mirror image of "
2752                              "different size");
2753             goto out;
2754         }
2755     }
2756 
2757     if ((sync == MIRROR_SYNC_MODE_FULL || !source)
2758         && mode != NEW_IMAGE_MODE_EXISTING)
2759     {
2760         /* create new image w/o backing file */
2761         assert(format);
2762         bdrv_img_create(target, format,
2763                         NULL, NULL, NULL, size, flags, &local_err, false);
2764     } else {
2765         switch (mode) {
2766         case NEW_IMAGE_MODE_EXISTING:
2767             break;
2768         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
2769             /* create new image with backing file */
2770             bdrv_img_create(target, format,
2771                             source->filename,
2772                             source->drv->format_name,
2773                             NULL, size, flags, &local_err, false);
2774             break;
2775         default:
2776             abort();
2777         }
2778     }
2779 
2780     if (local_err) {
2781         error_propagate(errp, local_err);
2782         goto out;
2783     }
2784 
2785     options = qdict_new();
2786     if (has_node_name) {
2787         qdict_put(options, "node-name", qstring_from_str(node_name));
2788     }
2789     if (format) {
2790         qdict_put(options, "driver", qstring_from_str(format));
2791     }
2792 
2793     /* Mirroring takes care of copy-on-write using the source's backing
2794      * file.
2795      */
2796     target_bs = NULL;
2797     ret = bdrv_open(&target_bs, target, NULL, options,
2798                     flags | BDRV_O_NO_BACKING, &local_err);
2799     if (ret < 0) {
2800         error_propagate(errp, local_err);
2801         goto out;
2802     }
2803 
2804     bdrv_set_aio_context(target_bs, aio_context);
2805 
2806     /* pass the node name to replace to mirror start since it's loose coupling
2807      * and will allow to check whether the node still exist at mirror completion
2808      */
2809     mirror_start(bs, target_bs,
2810                  has_replaces ? replaces : NULL,
2811                  speed, granularity, buf_size, sync,
2812                  on_source_error, on_target_error,
2813                  unmap,
2814                  block_job_cb, bs, &local_err);
2815     if (local_err != NULL) {
2816         bdrv_unref(target_bs);
2817         error_propagate(errp, local_err);
2818         goto out;
2819     }
2820 
2821 out:
2822     aio_context_release(aio_context);
2823 }
2824 
2825 /* Get the block job for a given device name and acquire its AioContext */
2826 static BlockJob *find_block_job(const char *device, AioContext **aio_context,
2827                                 Error **errp)
2828 {
2829     BlockBackend *blk;
2830     BlockDriverState *bs;
2831 
2832     blk = blk_by_name(device);
2833     if (!blk) {
2834         goto notfound;
2835     }
2836     bs = blk_bs(blk);
2837 
2838     *aio_context = bdrv_get_aio_context(bs);
2839     aio_context_acquire(*aio_context);
2840 
2841     if (!bs->job) {
2842         aio_context_release(*aio_context);
2843         goto notfound;
2844     }
2845 
2846     return bs->job;
2847 
2848 notfound:
2849     error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
2850               "No active block job on device '%s'", device);
2851     *aio_context = NULL;
2852     return NULL;
2853 }
2854 
2855 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
2856 {
2857     AioContext *aio_context;
2858     BlockJob *job = find_block_job(device, &aio_context, errp);
2859 
2860     if (!job) {
2861         return;
2862     }
2863 
2864     block_job_set_speed(job, speed, errp);
2865     aio_context_release(aio_context);
2866 }
2867 
2868 void qmp_block_job_cancel(const char *device,
2869                           bool has_force, bool force, Error **errp)
2870 {
2871     AioContext *aio_context;
2872     BlockJob *job = find_block_job(device, &aio_context, errp);
2873 
2874     if (!job) {
2875         return;
2876     }
2877 
2878     if (!has_force) {
2879         force = false;
2880     }
2881 
2882     if (job->user_paused && !force) {
2883         error_setg(errp, "The block job for device '%s' is currently paused",
2884                    device);
2885         goto out;
2886     }
2887 
2888     trace_qmp_block_job_cancel(job);
2889     block_job_cancel(job);
2890 out:
2891     aio_context_release(aio_context);
2892 }
2893 
2894 void qmp_block_job_pause(const char *device, Error **errp)
2895 {
2896     AioContext *aio_context;
2897     BlockJob *job = find_block_job(device, &aio_context, errp);
2898 
2899     if (!job || job->user_paused) {
2900         return;
2901     }
2902 
2903     job->user_paused = true;
2904     trace_qmp_block_job_pause(job);
2905     block_job_pause(job);
2906     aio_context_release(aio_context);
2907 }
2908 
2909 void qmp_block_job_resume(const char *device, Error **errp)
2910 {
2911     AioContext *aio_context;
2912     BlockJob *job = find_block_job(device, &aio_context, errp);
2913 
2914     if (!job || !job->user_paused) {
2915         return;
2916     }
2917 
2918     job->user_paused = false;
2919     trace_qmp_block_job_resume(job);
2920     block_job_resume(job);
2921     aio_context_release(aio_context);
2922 }
2923 
2924 void qmp_block_job_complete(const char *device, Error **errp)
2925 {
2926     AioContext *aio_context;
2927     BlockJob *job = find_block_job(device, &aio_context, errp);
2928 
2929     if (!job) {
2930         return;
2931     }
2932 
2933     trace_qmp_block_job_complete(job);
2934     block_job_complete(job, errp);
2935     aio_context_release(aio_context);
2936 }
2937 
2938 void qmp_change_backing_file(const char *device,
2939                              const char *image_node_name,
2940                              const char *backing_file,
2941                              Error **errp)
2942 {
2943     BlockBackend *blk;
2944     BlockDriverState *bs = NULL;
2945     AioContext *aio_context;
2946     BlockDriverState *image_bs = NULL;
2947     Error *local_err = NULL;
2948     bool ro;
2949     int open_flags;
2950     int ret;
2951 
2952     blk = blk_by_name(device);
2953     if (!blk) {
2954         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2955                   "Device '%s' not found", device);
2956         return;
2957     }
2958     bs = blk_bs(blk);
2959 
2960     aio_context = bdrv_get_aio_context(bs);
2961     aio_context_acquire(aio_context);
2962 
2963     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
2964     if (local_err) {
2965         error_propagate(errp, local_err);
2966         goto out;
2967     }
2968 
2969     if (!image_bs) {
2970         error_setg(errp, "image file not found");
2971         goto out;
2972     }
2973 
2974     if (bdrv_find_base(image_bs) == image_bs) {
2975         error_setg(errp, "not allowing backing file change on an image "
2976                          "without a backing file");
2977         goto out;
2978     }
2979 
2980     /* even though we are not necessarily operating on bs, we need it to
2981      * determine if block ops are currently prohibited on the chain */
2982     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
2983         goto out;
2984     }
2985 
2986     /* final sanity check */
2987     if (!bdrv_chain_contains(bs, image_bs)) {
2988         error_setg(errp, "'%s' and image file are not in the same chain",
2989                    device);
2990         goto out;
2991     }
2992 
2993     /* if not r/w, reopen to make r/w */
2994     open_flags = image_bs->open_flags;
2995     ro = bdrv_is_read_only(image_bs);
2996 
2997     if (ro) {
2998         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
2999         if (local_err) {
3000             error_propagate(errp, local_err);
3001             goto out;
3002         }
3003     }
3004 
3005     ret = bdrv_change_backing_file(image_bs, backing_file,
3006                                image_bs->drv ? image_bs->drv->format_name : "");
3007 
3008     if (ret < 0) {
3009         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3010                          backing_file);
3011         /* don't exit here, so we can try to restore open flags if
3012          * appropriate */
3013     }
3014 
3015     if (ro) {
3016         bdrv_reopen(image_bs, open_flags, &local_err);
3017         if (local_err) {
3018             error_propagate(errp, local_err); /* will preserve prior errp */
3019         }
3020     }
3021 
3022 out:
3023     aio_context_release(aio_context);
3024 }
3025 
3026 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3027 {
3028     QmpOutputVisitor *ov = qmp_output_visitor_new();
3029     BlockBackend *blk;
3030     QObject *obj;
3031     QDict *qdict;
3032     Error *local_err = NULL;
3033 
3034     /* Require an ID in the top level */
3035     if (!options->has_id) {
3036         error_setg(errp, "Block device needs an ID");
3037         goto fail;
3038     }
3039 
3040     /* TODO Sort it out in raw-posix and drive_new(): Reject aio=native with
3041      * cache.direct=false instead of silently switching to aio=threads, except
3042      * when called from drive_new().
3043      *
3044      * For now, simply forbidding the combination for all drivers will do. */
3045     if (options->has_aio && options->aio == BLOCKDEV_AIO_OPTIONS_NATIVE) {
3046         bool direct = options->has_cache &&
3047                       options->cache->has_direct &&
3048                       options->cache->direct;
3049         if (!direct) {
3050             error_setg(errp, "aio=native requires cache.direct=true");
3051             goto fail;
3052         }
3053     }
3054 
3055     visit_type_BlockdevOptions(qmp_output_get_visitor(ov),
3056                                &options, NULL, &local_err);
3057     if (local_err) {
3058         error_propagate(errp, local_err);
3059         goto fail;
3060     }
3061 
3062     obj = qmp_output_get_qobject(ov);
3063     qdict = qobject_to_qdict(obj);
3064 
3065     qdict_flatten(qdict);
3066 
3067     blk = blockdev_init(NULL, qdict, &local_err);
3068     if (local_err) {
3069         error_propagate(errp, local_err);
3070         goto fail;
3071     }
3072 
3073     if (bdrv_key_required(blk_bs(blk))) {
3074         blk_unref(blk);
3075         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3076         goto fail;
3077     }
3078 
3079 fail:
3080     qmp_output_visitor_cleanup(ov);
3081 }
3082 
3083 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3084 {
3085     BlockJobInfoList *head = NULL, **p_next = &head;
3086     BlockDriverState *bs;
3087 
3088     for (bs = bdrv_next(NULL); bs; bs = bdrv_next(bs)) {
3089         AioContext *aio_context = bdrv_get_aio_context(bs);
3090 
3091         aio_context_acquire(aio_context);
3092 
3093         if (bs->job) {
3094             BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
3095             elem->value = block_job_query(bs->job);
3096             *p_next = elem;
3097             p_next = &elem->next;
3098         }
3099 
3100         aio_context_release(aio_context);
3101     }
3102 
3103     return head;
3104 }
3105 
3106 QemuOptsList qemu_common_drive_opts = {
3107     .name = "drive",
3108     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3109     .desc = {
3110         {
3111             .name = "snapshot",
3112             .type = QEMU_OPT_BOOL,
3113             .help = "enable/disable snapshot mode",
3114         },{
3115             .name = "discard",
3116             .type = QEMU_OPT_STRING,
3117             .help = "discard operation (ignore/off, unmap/on)",
3118         },{
3119             .name = BDRV_OPT_CACHE_WB,
3120             .type = QEMU_OPT_BOOL,
3121             .help = "enables writeback mode for any caches",
3122         },{
3123             .name = BDRV_OPT_CACHE_DIRECT,
3124             .type = QEMU_OPT_BOOL,
3125             .help = "enables use of O_DIRECT (bypass the host page cache)",
3126         },{
3127             .name = BDRV_OPT_CACHE_NO_FLUSH,
3128             .type = QEMU_OPT_BOOL,
3129             .help = "ignore any flush requests for the device",
3130         },{
3131             .name = "aio",
3132             .type = QEMU_OPT_STRING,
3133             .help = "host AIO implementation (threads, native)",
3134         },{
3135             .name = "format",
3136             .type = QEMU_OPT_STRING,
3137             .help = "disk format (raw, qcow2, ...)",
3138         },{
3139             .name = "rerror",
3140             .type = QEMU_OPT_STRING,
3141             .help = "read error action",
3142         },{
3143             .name = "werror",
3144             .type = QEMU_OPT_STRING,
3145             .help = "write error action",
3146         },{
3147             .name = "read-only",
3148             .type = QEMU_OPT_BOOL,
3149             .help = "open drive file as read-only",
3150         },{
3151             .name = "throttling.iops-total",
3152             .type = QEMU_OPT_NUMBER,
3153             .help = "limit total I/O operations per second",
3154         },{
3155             .name = "throttling.iops-read",
3156             .type = QEMU_OPT_NUMBER,
3157             .help = "limit read operations per second",
3158         },{
3159             .name = "throttling.iops-write",
3160             .type = QEMU_OPT_NUMBER,
3161             .help = "limit write operations per second",
3162         },{
3163             .name = "throttling.bps-total",
3164             .type = QEMU_OPT_NUMBER,
3165             .help = "limit total bytes per second",
3166         },{
3167             .name = "throttling.bps-read",
3168             .type = QEMU_OPT_NUMBER,
3169             .help = "limit read bytes per second",
3170         },{
3171             .name = "throttling.bps-write",
3172             .type = QEMU_OPT_NUMBER,
3173             .help = "limit write bytes per second",
3174         },{
3175             .name = "throttling.iops-total-max",
3176             .type = QEMU_OPT_NUMBER,
3177             .help = "I/O operations burst",
3178         },{
3179             .name = "throttling.iops-read-max",
3180             .type = QEMU_OPT_NUMBER,
3181             .help = "I/O operations read burst",
3182         },{
3183             .name = "throttling.iops-write-max",
3184             .type = QEMU_OPT_NUMBER,
3185             .help = "I/O operations write burst",
3186         },{
3187             .name = "throttling.bps-total-max",
3188             .type = QEMU_OPT_NUMBER,
3189             .help = "total bytes burst",
3190         },{
3191             .name = "throttling.bps-read-max",
3192             .type = QEMU_OPT_NUMBER,
3193             .help = "total bytes read burst",
3194         },{
3195             .name = "throttling.bps-write-max",
3196             .type = QEMU_OPT_NUMBER,
3197             .help = "total bytes write burst",
3198         },{
3199             .name = "throttling.iops-size",
3200             .type = QEMU_OPT_NUMBER,
3201             .help = "when limiting by iops max size of an I/O in bytes",
3202         },{
3203             .name = "throttling.group",
3204             .type = QEMU_OPT_STRING,
3205             .help = "name of the block throttling group",
3206         },{
3207             .name = "copy-on-read",
3208             .type = QEMU_OPT_BOOL,
3209             .help = "copy read data from backing file into image file",
3210         },{
3211             .name = "detect-zeroes",
3212             .type = QEMU_OPT_STRING,
3213             .help = "try to optimize zero writes (off, on, unmap)",
3214         },
3215         { /* end of list */ }
3216     },
3217 };
3218 
3219 QemuOptsList qemu_drive_opts = {
3220     .name = "drive",
3221     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3222     .desc = {
3223         /*
3224          * no elements => accept any params
3225          * validation will happen later
3226          */
3227         { /* end of list */ }
3228     },
3229 };
3230