xref: /openbmc/qemu/blockdev.c (revision a36e458cdda0196911c1cbe7cfe6f9530f2280e3)
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 "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/throttle-groups.h"
39 #include "monitor/monitor.h"
40 #include "qemu/error-report.h"
41 #include "qemu/option.h"
42 #include "qemu/config-file.h"
43 #include "qapi/qmp/types.h"
44 #include "qapi-visit.h"
45 #include "qapi/qmp/qerror.h"
46 #include "qapi/qobject-output-visitor.h"
47 #include "sysemu/sysemu.h"
48 #include "block/block_int.h"
49 #include "qmp-commands.h"
50 #include "block/trace.h"
51 #include "sysemu/arch_init.h"
52 #include "sysemu/qtest.h"
53 #include "qemu/cutils.h"
54 #include "qemu/help_option.h"
55 #include "qemu/throttle-options.h"
56 
57 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
58     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
59 
60 static int do_open_tray(const char *blk_name, const char *qdev_id,
61                         bool force, Error **errp);
62 
63 static const char *const if_name[IF_COUNT] = {
64     [IF_NONE] = "none",
65     [IF_IDE] = "ide",
66     [IF_SCSI] = "scsi",
67     [IF_FLOPPY] = "floppy",
68     [IF_PFLASH] = "pflash",
69     [IF_MTD] = "mtd",
70     [IF_SD] = "sd",
71     [IF_VIRTIO] = "virtio",
72     [IF_XEN] = "xen",
73 };
74 
75 static int if_max_devs[IF_COUNT] = {
76     /*
77      * Do not change these numbers!  They govern how drive option
78      * index maps to unit and bus.  That mapping is ABI.
79      *
80      * All controllers used to implement if=T drives need to support
81      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
82      * Otherwise, some index values map to "impossible" bus, unit
83      * values.
84      *
85      * For instance, if you change [IF_SCSI] to 255, -drive
86      * if=scsi,index=12 no longer means bus=1,unit=5, but
87      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
88      * the drive can't be set up.  Regression.
89      */
90     [IF_IDE] = 2,
91     [IF_SCSI] = 7,
92 };
93 
94 /**
95  * Boards may call this to offer board-by-board overrides
96  * of the default, global values.
97  */
98 void override_max_devs(BlockInterfaceType type, int max_devs)
99 {
100     BlockBackend *blk;
101     DriveInfo *dinfo;
102 
103     if (max_devs <= 0) {
104         return;
105     }
106 
107     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
108         dinfo = blk_legacy_dinfo(blk);
109         if (dinfo->type == type) {
110             fprintf(stderr, "Cannot override units-per-bus property of"
111                     " the %s interface, because a drive of that type has"
112                     " already been added.\n", if_name[type]);
113             g_assert_not_reached();
114         }
115     }
116 
117     if_max_devs[type] = max_devs;
118 }
119 
120 /*
121  * We automatically delete the drive when a device using it gets
122  * unplugged.  Questionable feature, but we can't just drop it.
123  * Device models call blockdev_mark_auto_del() to schedule the
124  * automatic deletion, and generic qdev code calls blockdev_auto_del()
125  * when deletion is actually safe.
126  */
127 void blockdev_mark_auto_del(BlockBackend *blk)
128 {
129     DriveInfo *dinfo = blk_legacy_dinfo(blk);
130     BlockDriverState *bs = blk_bs(blk);
131     AioContext *aio_context;
132 
133     if (!dinfo) {
134         return;
135     }
136 
137     if (bs) {
138         aio_context = bdrv_get_aio_context(bs);
139         aio_context_acquire(aio_context);
140 
141         if (bs->job) {
142             block_job_cancel(bs->job);
143         }
144 
145         aio_context_release(aio_context);
146     }
147 
148     dinfo->auto_del = 1;
149 }
150 
151 void blockdev_auto_del(BlockBackend *blk)
152 {
153     DriveInfo *dinfo = blk_legacy_dinfo(blk);
154 
155     if (dinfo && dinfo->auto_del) {
156         monitor_remove_blk(blk);
157         blk_unref(blk);
158     }
159 }
160 
161 /**
162  * Returns the current mapping of how many units per bus
163  * a particular interface can support.
164  *
165  *  A positive integer indicates n units per bus.
166  *  0 implies the mapping has not been established.
167  * -1 indicates an invalid BlockInterfaceType was given.
168  */
169 int drive_get_max_devs(BlockInterfaceType type)
170 {
171     if (type >= IF_IDE && type < IF_COUNT) {
172         return if_max_devs[type];
173     }
174 
175     return -1;
176 }
177 
178 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
179 {
180     int max_devs = if_max_devs[type];
181     return max_devs ? index / max_devs : 0;
182 }
183 
184 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
185 {
186     int max_devs = if_max_devs[type];
187     return max_devs ? index % max_devs : index;
188 }
189 
190 QemuOpts *drive_def(const char *optstr)
191 {
192     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
193 }
194 
195 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
196                     const char *optstr)
197 {
198     QemuOpts *opts;
199 
200     opts = drive_def(optstr);
201     if (!opts) {
202         return NULL;
203     }
204     if (type != IF_DEFAULT) {
205         qemu_opt_set(opts, "if", if_name[type], &error_abort);
206     }
207     if (index >= 0) {
208         qemu_opt_set_number(opts, "index", index, &error_abort);
209     }
210     if (file)
211         qemu_opt_set(opts, "file", file, &error_abort);
212     return opts;
213 }
214 
215 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
216 {
217     BlockBackend *blk;
218     DriveInfo *dinfo;
219 
220     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
221         dinfo = blk_legacy_dinfo(blk);
222         if (dinfo && dinfo->type == type
223             && dinfo->bus == bus && dinfo->unit == unit) {
224             return dinfo;
225         }
226     }
227 
228     return NULL;
229 }
230 
231 void drive_check_orphaned(void)
232 {
233     BlockBackend *blk;
234     DriveInfo *dinfo;
235     Location loc;
236     bool orphans = false;
237 
238     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
239         dinfo = blk_legacy_dinfo(blk);
240         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
241             dinfo->type != IF_NONE) {
242             loc_push_none(&loc);
243             qemu_opts_loc_restore(dinfo->opts);
244             error_report("machine type does not support"
245                          " if=%s,bus=%d,unit=%d",
246                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
247             loc_pop(&loc);
248             orphans = true;
249         }
250     }
251 
252     if (orphans) {
253         exit(1);
254     }
255 }
256 
257 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
258 {
259     return drive_get(type,
260                      drive_index_to_bus_id(type, index),
261                      drive_index_to_unit_id(type, index));
262 }
263 
264 int drive_get_max_bus(BlockInterfaceType type)
265 {
266     int max_bus;
267     BlockBackend *blk;
268     DriveInfo *dinfo;
269 
270     max_bus = -1;
271     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
272         dinfo = blk_legacy_dinfo(blk);
273         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
274             max_bus = dinfo->bus;
275         }
276     }
277     return max_bus;
278 }
279 
280 /* Get a block device.  This should only be used for single-drive devices
281    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
282    appropriate bus.  */
283 DriveInfo *drive_get_next(BlockInterfaceType type)
284 {
285     static int next_block_unit[IF_COUNT];
286 
287     return drive_get(type, 0, next_block_unit[type]++);
288 }
289 
290 static void bdrv_format_print(void *opaque, const char *name)
291 {
292     error_printf(" %s", name);
293 }
294 
295 typedef struct {
296     QEMUBH *bh;
297     BlockDriverState *bs;
298 } BDRVPutRefBH;
299 
300 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
301 {
302     if (!strcmp(buf, "ignore")) {
303         return BLOCKDEV_ON_ERROR_IGNORE;
304     } else if (!is_read && !strcmp(buf, "enospc")) {
305         return BLOCKDEV_ON_ERROR_ENOSPC;
306     } else if (!strcmp(buf, "stop")) {
307         return BLOCKDEV_ON_ERROR_STOP;
308     } else if (!strcmp(buf, "report")) {
309         return BLOCKDEV_ON_ERROR_REPORT;
310     } else {
311         error_setg(errp, "'%s' invalid %s error action",
312                    buf, is_read ? "read" : "write");
313         return -1;
314     }
315 }
316 
317 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
318                                   Error **errp)
319 {
320     const QListEntry *entry;
321     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
322         switch (qobject_type(entry->value)) {
323 
324         case QTYPE_QSTRING: {
325             unsigned long long length;
326             const char *str = qstring_get_str(qobject_to_qstring(entry->value));
327             if (parse_uint_full(str, &length, 10) == 0 &&
328                 length > 0 && length <= UINT_MAX) {
329                 block_acct_add_interval(stats, (unsigned) length);
330             } else {
331                 error_setg(errp, "Invalid interval length: %s", str);
332                 return false;
333             }
334             break;
335         }
336 
337         case QTYPE_QNUM: {
338             int64_t length = qnum_get_int(qobject_to_qnum(entry->value));
339 
340             if (length > 0 && length <= UINT_MAX) {
341                 block_acct_add_interval(stats, (unsigned) length);
342             } else {
343                 error_setg(errp, "Invalid interval length: %" PRId64, length);
344                 return false;
345             }
346             break;
347         }
348 
349         default:
350             error_setg(errp, "The specification of stats-intervals is invalid");
351             return false;
352         }
353     }
354     return true;
355 }
356 
357 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
358 
359 /* All parameters but @opts are optional and may be set to NULL. */
360 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
361     const char **throttling_group, ThrottleConfig *throttle_cfg,
362     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
363 {
364     Error *local_error = NULL;
365     const char *aio;
366 
367     if (bdrv_flags) {
368         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
369             *bdrv_flags |= BDRV_O_COPY_ON_READ;
370         }
371 
372         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
373             if (!strcmp(aio, "native")) {
374                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
375             } else if (!strcmp(aio, "threads")) {
376                 /* this is the default */
377             } else {
378                error_setg(errp, "invalid aio option");
379                return;
380             }
381         }
382     }
383 
384     /* disk I/O throttling */
385     if (throttling_group) {
386         *throttling_group = qemu_opt_get(opts, "throttling.group");
387     }
388 
389     if (throttle_cfg) {
390         throttle_config_init(throttle_cfg);
391         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
392             qemu_opt_get_number(opts, "throttling.bps-total", 0);
393         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
394             qemu_opt_get_number(opts, "throttling.bps-read", 0);
395         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
396             qemu_opt_get_number(opts, "throttling.bps-write", 0);
397         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
398             qemu_opt_get_number(opts, "throttling.iops-total", 0);
399         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
400             qemu_opt_get_number(opts, "throttling.iops-read", 0);
401         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
402             qemu_opt_get_number(opts, "throttling.iops-write", 0);
403 
404         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
405             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
406         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
407             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
408         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
409             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
410         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
411             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
412         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
413             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
414         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
415             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
416 
417         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
418             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
419         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
420             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
421         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
422             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
423         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
424             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
425         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
426             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
427         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
428             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
429 
430         throttle_cfg->op_size =
431             qemu_opt_get_number(opts, "throttling.iops-size", 0);
432 
433         if (!throttle_is_valid(throttle_cfg, errp)) {
434             return;
435         }
436     }
437 
438     if (detect_zeroes) {
439         *detect_zeroes =
440             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
441                             qemu_opt_get(opts, "detect-zeroes"),
442                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
443                             &local_error);
444         if (local_error) {
445             error_propagate(errp, local_error);
446             return;
447         }
448     }
449 }
450 
451 /* Takes the ownership of bs_opts */
452 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
453                                    Error **errp)
454 {
455     const char *buf;
456     int bdrv_flags = 0;
457     int on_read_error, on_write_error;
458     bool account_invalid, account_failed;
459     bool writethrough, read_only;
460     BlockBackend *blk;
461     BlockDriverState *bs;
462     ThrottleConfig cfg;
463     int snapshot = 0;
464     Error *error = NULL;
465     QemuOpts *opts;
466     QDict *interval_dict = NULL;
467     QList *interval_list = NULL;
468     const char *id;
469     BlockdevDetectZeroesOptions detect_zeroes =
470         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
471     const char *throttling_group = NULL;
472 
473     /* Check common options by copying from bs_opts to opts, all other options
474      * stay in bs_opts for processing by bdrv_open(). */
475     id = qdict_get_try_str(bs_opts, "id");
476     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
477     if (error) {
478         error_propagate(errp, error);
479         goto err_no_opts;
480     }
481 
482     qemu_opts_absorb_qdict(opts, bs_opts, &error);
483     if (error) {
484         error_propagate(errp, error);
485         goto early_err;
486     }
487 
488     if (id) {
489         qdict_del(bs_opts, "id");
490     }
491 
492     /* extract parameters */
493     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
494 
495     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
496     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
497 
498     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
499 
500     id = qemu_opts_id(opts);
501 
502     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
503     qdict_array_split(interval_dict, &interval_list);
504 
505     if (qdict_size(interval_dict) != 0) {
506         error_setg(errp, "Invalid option stats-intervals.%s",
507                    qdict_first(interval_dict)->key);
508         goto early_err;
509     }
510 
511     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
512                                     &detect_zeroes, &error);
513     if (error) {
514         error_propagate(errp, error);
515         goto early_err;
516     }
517 
518     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
519         if (is_help_option(buf)) {
520             error_printf("Supported formats:");
521             bdrv_iterate_format(bdrv_format_print, NULL);
522             error_printf("\n");
523             goto early_err;
524         }
525 
526         if (qdict_haskey(bs_opts, "driver")) {
527             error_setg(errp, "Cannot specify both 'driver' and 'format'");
528             goto early_err;
529         }
530         qdict_put_str(bs_opts, "driver", buf);
531     }
532 
533     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
534     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
535         on_write_error = parse_block_error_action(buf, 0, &error);
536         if (error) {
537             error_propagate(errp, error);
538             goto early_err;
539         }
540     }
541 
542     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
543     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
544         on_read_error = parse_block_error_action(buf, 1, &error);
545         if (error) {
546             error_propagate(errp, error);
547             goto early_err;
548         }
549     }
550 
551     if (snapshot) {
552         bdrv_flags |= BDRV_O_SNAPSHOT;
553     }
554 
555     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
556 
557     /* init */
558     if ((!file || !*file) && !qdict_size(bs_opts)) {
559         BlockBackendRootState *blk_rs;
560 
561         blk = blk_new(0, BLK_PERM_ALL);
562         blk_rs = blk_get_root_state(blk);
563         blk_rs->open_flags    = bdrv_flags;
564         blk_rs->read_only     = read_only;
565         blk_rs->detect_zeroes = detect_zeroes;
566 
567         QDECREF(bs_opts);
568     } else {
569         if (file && !*file) {
570             file = NULL;
571         }
572 
573         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
574          * with other callers) rather than what we want as the real defaults.
575          * Apply the defaults here instead. */
576         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
577         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
578         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
579                               read_only ? "on" : "off");
580         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
581 
582         if (runstate_check(RUN_STATE_INMIGRATE)) {
583             bdrv_flags |= BDRV_O_INACTIVE;
584         }
585 
586         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
587         if (!blk) {
588             goto err_no_bs_opts;
589         }
590         bs = blk_bs(blk);
591 
592         bs->detect_zeroes = detect_zeroes;
593 
594         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
595 
596         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
597             blk_unref(blk);
598             blk = NULL;
599             goto err_no_bs_opts;
600         }
601     }
602 
603     /* disk I/O throttling */
604     if (throttle_enabled(&cfg)) {
605         if (!throttling_group) {
606             throttling_group = id;
607         }
608         blk_io_limits_enable(blk, throttling_group);
609         blk_set_io_limits(blk, &cfg);
610     }
611 
612     blk_set_enable_write_cache(blk, !writethrough);
613     blk_set_on_error(blk, on_read_error, on_write_error);
614 
615     if (!monitor_add_blk(blk, id, errp)) {
616         blk_unref(blk);
617         blk = NULL;
618         goto err_no_bs_opts;
619     }
620 
621 err_no_bs_opts:
622     qemu_opts_del(opts);
623     QDECREF(interval_dict);
624     QDECREF(interval_list);
625     return blk;
626 
627 early_err:
628     qemu_opts_del(opts);
629     QDECREF(interval_dict);
630     QDECREF(interval_list);
631 err_no_opts:
632     QDECREF(bs_opts);
633     return NULL;
634 }
635 
636 /* Takes the ownership of bs_opts */
637 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
638 {
639     int bdrv_flags = 0;
640 
641     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
642      * with other callers) rather than what we want as the real defaults.
643      * Apply the defaults here instead. */
644     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
645     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
646     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
647 
648     if (runstate_check(RUN_STATE_INMIGRATE)) {
649         bdrv_flags |= BDRV_O_INACTIVE;
650     }
651 
652     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
653 }
654 
655 void blockdev_close_all_bdrv_states(void)
656 {
657     BlockDriverState *bs, *next_bs;
658 
659     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
660         AioContext *ctx = bdrv_get_aio_context(bs);
661 
662         aio_context_acquire(ctx);
663         bdrv_unref(bs);
664         aio_context_release(ctx);
665     }
666 }
667 
668 /* Iterates over the list of monitor-owned BlockDriverStates */
669 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
670 {
671     return bs ? QTAILQ_NEXT(bs, monitor_list)
672               : QTAILQ_FIRST(&monitor_bdrv_states);
673 }
674 
675 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
676                             Error **errp)
677 {
678     const char *value;
679 
680     value = qemu_opt_get(opts, from);
681     if (value) {
682         if (qemu_opt_find(opts, to)) {
683             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
684                        "same time", to, from);
685             return;
686         }
687     }
688 
689     /* rename all items in opts */
690     while ((value = qemu_opt_get(opts, from))) {
691         qemu_opt_set(opts, to, value, &error_abort);
692         qemu_opt_unset(opts, from);
693     }
694 }
695 
696 QemuOptsList qemu_legacy_drive_opts = {
697     .name = "drive",
698     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
699     .desc = {
700         {
701             .name = "bus",
702             .type = QEMU_OPT_NUMBER,
703             .help = "bus number",
704         },{
705             .name = "unit",
706             .type = QEMU_OPT_NUMBER,
707             .help = "unit number (i.e. lun for scsi)",
708         },{
709             .name = "index",
710             .type = QEMU_OPT_NUMBER,
711             .help = "index number",
712         },{
713             .name = "media",
714             .type = QEMU_OPT_STRING,
715             .help = "media type (disk, cdrom)",
716         },{
717             .name = "if",
718             .type = QEMU_OPT_STRING,
719             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
720         },{
721             .name = "cyls",
722             .type = QEMU_OPT_NUMBER,
723             .help = "number of cylinders (ide disk geometry)",
724         },{
725             .name = "heads",
726             .type = QEMU_OPT_NUMBER,
727             .help = "number of heads (ide disk geometry)",
728         },{
729             .name = "secs",
730             .type = QEMU_OPT_NUMBER,
731             .help = "number of sectors (ide disk geometry)",
732         },{
733             .name = "trans",
734             .type = QEMU_OPT_STRING,
735             .help = "chs translation (auto, lba, none)",
736         },{
737             .name = "boot",
738             .type = QEMU_OPT_BOOL,
739             .help = "(deprecated, ignored)",
740         },{
741             .name = "addr",
742             .type = QEMU_OPT_STRING,
743             .help = "pci address (virtio only)",
744         },{
745             .name = "serial",
746             .type = QEMU_OPT_STRING,
747             .help = "disk serial number",
748         },{
749             .name = "file",
750             .type = QEMU_OPT_STRING,
751             .help = "file name",
752         },
753 
754         /* Options that are passed on, but have special semantics with -drive */
755         {
756             .name = BDRV_OPT_READ_ONLY,
757             .type = QEMU_OPT_BOOL,
758             .help = "open drive file as read-only",
759         },{
760             .name = "rerror",
761             .type = QEMU_OPT_STRING,
762             .help = "read error action",
763         },{
764             .name = "werror",
765             .type = QEMU_OPT_STRING,
766             .help = "write error action",
767         },{
768             .name = "copy-on-read",
769             .type = QEMU_OPT_BOOL,
770             .help = "copy read data from backing file into image file",
771         },
772 
773         { /* end of list */ }
774     },
775 };
776 
777 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
778 {
779     const char *value;
780     BlockBackend *blk;
781     DriveInfo *dinfo = NULL;
782     QDict *bs_opts;
783     QemuOpts *legacy_opts;
784     DriveMediaType media = MEDIA_DISK;
785     BlockInterfaceType type;
786     int cyls, heads, secs, translation;
787     int max_devs, bus_id, unit_id, index;
788     const char *devaddr;
789     const char *werror, *rerror;
790     bool read_only = false;
791     bool copy_on_read;
792     const char *serial;
793     const char *filename;
794     Error *local_err = NULL;
795     int i;
796     const char *deprecated[] = {
797         "serial", "trans", "secs", "heads", "cyls", "addr"
798     };
799 
800     /* Change legacy command line options into QMP ones */
801     static const struct {
802         const char *from;
803         const char *to;
804     } opt_renames[] = {
805         { "iops",           "throttling.iops-total" },
806         { "iops_rd",        "throttling.iops-read" },
807         { "iops_wr",        "throttling.iops-write" },
808 
809         { "bps",            "throttling.bps-total" },
810         { "bps_rd",         "throttling.bps-read" },
811         { "bps_wr",         "throttling.bps-write" },
812 
813         { "iops_max",       "throttling.iops-total-max" },
814         { "iops_rd_max",    "throttling.iops-read-max" },
815         { "iops_wr_max",    "throttling.iops-write-max" },
816 
817         { "bps_max",        "throttling.bps-total-max" },
818         { "bps_rd_max",     "throttling.bps-read-max" },
819         { "bps_wr_max",     "throttling.bps-write-max" },
820 
821         { "iops_size",      "throttling.iops-size" },
822 
823         { "group",          "throttling.group" },
824 
825         { "readonly",       BDRV_OPT_READ_ONLY },
826     };
827 
828     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
829         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
830                         &local_err);
831         if (local_err) {
832             error_report_err(local_err);
833             return NULL;
834         }
835     }
836 
837     value = qemu_opt_get(all_opts, "cache");
838     if (value) {
839         int flags = 0;
840         bool writethrough;
841 
842         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
843             error_report("invalid cache option");
844             return NULL;
845         }
846 
847         /* Specific options take precedence */
848         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
849             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
850                               !writethrough, &error_abort);
851         }
852         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
853             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
854                               !!(flags & BDRV_O_NOCACHE), &error_abort);
855         }
856         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
857             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
858                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
859         }
860         qemu_opt_unset(all_opts, "cache");
861     }
862 
863     /* Get a QDict for processing the options */
864     bs_opts = qdict_new();
865     qemu_opts_to_qdict(all_opts, bs_opts);
866 
867     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
868                                    &error_abort);
869     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
870     if (local_err) {
871         error_report_err(local_err);
872         goto fail;
873     }
874 
875     /* Deprecated option boot=[on|off] */
876     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
877         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
878                 "ignored. Future versions will reject this parameter. Please "
879                 "update your scripts.\n");
880     }
881 
882     /* Other deprecated options */
883     if (!qtest_enabled()) {
884         for (i = 0; i < ARRAY_SIZE(deprecated); i++) {
885             if (qemu_opt_get(legacy_opts, deprecated[i]) != NULL) {
886                 error_report("'%s' is deprecated, please use the corresponding "
887                              "option of '-device' instead", deprecated[i]);
888             }
889         }
890     }
891 
892     /* Media type */
893     value = qemu_opt_get(legacy_opts, "media");
894     if (value) {
895         if (!strcmp(value, "disk")) {
896             media = MEDIA_DISK;
897         } else if (!strcmp(value, "cdrom")) {
898             media = MEDIA_CDROM;
899             read_only = true;
900         } else {
901             error_report("'%s' invalid media", value);
902             goto fail;
903         }
904     }
905 
906     /* copy-on-read is disabled with a warning for read-only devices */
907     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
908     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
909 
910     if (read_only && copy_on_read) {
911         warn_report("disabling copy-on-read on read-only drive");
912         copy_on_read = false;
913     }
914 
915     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
916     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
917 
918     /* Controller type */
919     value = qemu_opt_get(legacy_opts, "if");
920     if (value) {
921         for (type = 0;
922              type < IF_COUNT && strcmp(value, if_name[type]);
923              type++) {
924         }
925         if (type == IF_COUNT) {
926             error_report("unsupported bus type '%s'", value);
927             goto fail;
928         }
929     } else {
930         type = block_default_type;
931     }
932 
933     /* Geometry */
934     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
935     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
936     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
937 
938     if (cyls || heads || secs) {
939         if (cyls < 1) {
940             error_report("invalid physical cyls number");
941             goto fail;
942         }
943         if (heads < 1) {
944             error_report("invalid physical heads number");
945             goto fail;
946         }
947         if (secs < 1) {
948             error_report("invalid physical secs number");
949             goto fail;
950         }
951     }
952 
953     translation = BIOS_ATA_TRANSLATION_AUTO;
954     value = qemu_opt_get(legacy_opts, "trans");
955     if (value != NULL) {
956         if (!cyls) {
957             error_report("'%s' trans must be used with cyls, heads and secs",
958                          value);
959             goto fail;
960         }
961         if (!strcmp(value, "none")) {
962             translation = BIOS_ATA_TRANSLATION_NONE;
963         } else if (!strcmp(value, "lba")) {
964             translation = BIOS_ATA_TRANSLATION_LBA;
965         } else if (!strcmp(value, "large")) {
966             translation = BIOS_ATA_TRANSLATION_LARGE;
967         } else if (!strcmp(value, "rechs")) {
968             translation = BIOS_ATA_TRANSLATION_RECHS;
969         } else if (!strcmp(value, "auto")) {
970             translation = BIOS_ATA_TRANSLATION_AUTO;
971         } else {
972             error_report("'%s' invalid translation type", value);
973             goto fail;
974         }
975     }
976 
977     if (media == MEDIA_CDROM) {
978         if (cyls || secs || heads) {
979             error_report("CHS can't be set with media=cdrom");
980             goto fail;
981         }
982     }
983 
984     /* Device address specified by bus/unit or index.
985      * If none was specified, try to find the first free one. */
986     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
987     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
988     index   = qemu_opt_get_number(legacy_opts, "index", -1);
989 
990     max_devs = if_max_devs[type];
991 
992     if (index != -1) {
993         if (bus_id != 0 || unit_id != -1) {
994             error_report("index cannot be used with bus and unit");
995             goto fail;
996         }
997         bus_id = drive_index_to_bus_id(type, index);
998         unit_id = drive_index_to_unit_id(type, index);
999     }
1000 
1001     if (unit_id == -1) {
1002        unit_id = 0;
1003        while (drive_get(type, bus_id, unit_id) != NULL) {
1004            unit_id++;
1005            if (max_devs && unit_id >= max_devs) {
1006                unit_id -= max_devs;
1007                bus_id++;
1008            }
1009        }
1010     }
1011 
1012     if (max_devs && unit_id >= max_devs) {
1013         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1014         goto fail;
1015     }
1016 
1017     if (drive_get(type, bus_id, unit_id) != NULL) {
1018         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1019                      bus_id, unit_id, index);
1020         goto fail;
1021     }
1022 
1023     /* Serial number */
1024     serial = qemu_opt_get(legacy_opts, "serial");
1025 
1026     /* no id supplied -> create one */
1027     if (qemu_opts_id(all_opts) == NULL) {
1028         char *new_id;
1029         const char *mediastr = "";
1030         if (type == IF_IDE || type == IF_SCSI) {
1031             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1032         }
1033         if (max_devs) {
1034             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1035                                      mediastr, unit_id);
1036         } else {
1037             new_id = g_strdup_printf("%s%s%i", if_name[type],
1038                                      mediastr, unit_id);
1039         }
1040         qdict_put_str(bs_opts, "id", new_id);
1041         g_free(new_id);
1042     }
1043 
1044     /* Add virtio block device */
1045     devaddr = qemu_opt_get(legacy_opts, "addr");
1046     if (devaddr && type != IF_VIRTIO) {
1047         error_report("addr is not supported by this bus type");
1048         goto fail;
1049     }
1050 
1051     if (type == IF_VIRTIO) {
1052         QemuOpts *devopts;
1053         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1054                                    &error_abort);
1055         if (arch_type == QEMU_ARCH_S390X) {
1056             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1057         } else {
1058             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1059         }
1060         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1061                      &error_abort);
1062         if (devaddr) {
1063             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1064         }
1065     }
1066 
1067     filename = qemu_opt_get(legacy_opts, "file");
1068 
1069     /* Check werror/rerror compatibility with if=... */
1070     werror = qemu_opt_get(legacy_opts, "werror");
1071     if (werror != NULL) {
1072         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1073             type != IF_NONE) {
1074             error_report("werror is not supported by this bus type");
1075             goto fail;
1076         }
1077         qdict_put_str(bs_opts, "werror", werror);
1078     }
1079 
1080     rerror = qemu_opt_get(legacy_opts, "rerror");
1081     if (rerror != NULL) {
1082         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1083             type != IF_NONE) {
1084             error_report("rerror is not supported by this bus type");
1085             goto fail;
1086         }
1087         qdict_put_str(bs_opts, "rerror", rerror);
1088     }
1089 
1090     /* Actual block device init: Functionality shared with blockdev-add */
1091     blk = blockdev_init(filename, bs_opts, &local_err);
1092     bs_opts = NULL;
1093     if (!blk) {
1094         if (local_err) {
1095             error_report_err(local_err);
1096         }
1097         goto fail;
1098     } else {
1099         assert(!local_err);
1100     }
1101 
1102     /* Create legacy DriveInfo */
1103     dinfo = g_malloc0(sizeof(*dinfo));
1104     dinfo->opts = all_opts;
1105 
1106     dinfo->cyls = cyls;
1107     dinfo->heads = heads;
1108     dinfo->secs = secs;
1109     dinfo->trans = translation;
1110 
1111     dinfo->type = type;
1112     dinfo->bus = bus_id;
1113     dinfo->unit = unit_id;
1114     dinfo->devaddr = devaddr;
1115     dinfo->serial = g_strdup(serial);
1116 
1117     blk_set_legacy_dinfo(blk, dinfo);
1118 
1119     switch(type) {
1120     case IF_IDE:
1121     case IF_SCSI:
1122     case IF_XEN:
1123     case IF_NONE:
1124         dinfo->media_cd = media == MEDIA_CDROM;
1125         break;
1126     default:
1127         break;
1128     }
1129 
1130 fail:
1131     qemu_opts_del(legacy_opts);
1132     QDECREF(bs_opts);
1133     return dinfo;
1134 }
1135 
1136 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1137 {
1138     BlockDriverState *bs;
1139 
1140     bs = bdrv_lookup_bs(name, name, errp);
1141     if (bs == NULL) {
1142         return NULL;
1143     }
1144 
1145     if (!bdrv_is_root_node(bs)) {
1146         error_setg(errp, "Need a root block node");
1147         return NULL;
1148     }
1149 
1150     if (!bdrv_is_inserted(bs)) {
1151         error_setg(errp, "Device has no medium");
1152         return NULL;
1153     }
1154 
1155     return bs;
1156 }
1157 
1158 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1159                                  Error **errp)
1160 {
1161     BlockBackend *blk;
1162 
1163     if (!blk_name == !qdev_id) {
1164         error_setg(errp, "Need exactly one of 'device' and 'id'");
1165         return NULL;
1166     }
1167 
1168     if (qdev_id) {
1169         blk = blk_by_qdev_id(qdev_id, errp);
1170     } else {
1171         blk = blk_by_name(blk_name);
1172         if (blk == NULL) {
1173             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1174                       "Device '%s' not found", blk_name);
1175         }
1176     }
1177 
1178     return blk;
1179 }
1180 
1181 void hmp_commit(Monitor *mon, const QDict *qdict)
1182 {
1183     const char *device = qdict_get_str(qdict, "device");
1184     BlockBackend *blk;
1185     int ret;
1186 
1187     if (!strcmp(device, "all")) {
1188         ret = blk_commit_all();
1189     } else {
1190         BlockDriverState *bs;
1191         AioContext *aio_context;
1192 
1193         blk = blk_by_name(device);
1194         if (!blk) {
1195             monitor_printf(mon, "Device '%s' not found\n", device);
1196             return;
1197         }
1198         if (!blk_is_available(blk)) {
1199             monitor_printf(mon, "Device '%s' has no medium\n", device);
1200             return;
1201         }
1202 
1203         bs = blk_bs(blk);
1204         aio_context = bdrv_get_aio_context(bs);
1205         aio_context_acquire(aio_context);
1206 
1207         ret = bdrv_commit(bs);
1208 
1209         aio_context_release(aio_context);
1210     }
1211     if (ret < 0) {
1212         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1213                        strerror(-ret));
1214     }
1215 }
1216 
1217 static void blockdev_do_action(TransactionAction *action, Error **errp)
1218 {
1219     TransactionActionList list;
1220 
1221     list.value = action;
1222     list.next = NULL;
1223     qmp_transaction(&list, false, NULL, errp);
1224 }
1225 
1226 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1227                                 bool has_node_name, const char *node_name,
1228                                 const char *snapshot_file,
1229                                 bool has_snapshot_node_name,
1230                                 const char *snapshot_node_name,
1231                                 bool has_format, const char *format,
1232                                 bool has_mode, NewImageMode mode, Error **errp)
1233 {
1234     BlockdevSnapshotSync snapshot = {
1235         .has_device = has_device,
1236         .device = (char *) device,
1237         .has_node_name = has_node_name,
1238         .node_name = (char *) node_name,
1239         .snapshot_file = (char *) snapshot_file,
1240         .has_snapshot_node_name = has_snapshot_node_name,
1241         .snapshot_node_name = (char *) snapshot_node_name,
1242         .has_format = has_format,
1243         .format = (char *) format,
1244         .has_mode = has_mode,
1245         .mode = mode,
1246     };
1247     TransactionAction action = {
1248         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1249         .u.blockdev_snapshot_sync.data = &snapshot,
1250     };
1251     blockdev_do_action(&action, errp);
1252 }
1253 
1254 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1255                            Error **errp)
1256 {
1257     BlockdevSnapshot snapshot_data = {
1258         .node = (char *) node,
1259         .overlay = (char *) overlay
1260     };
1261     TransactionAction action = {
1262         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1263         .u.blockdev_snapshot.data = &snapshot_data,
1264     };
1265     blockdev_do_action(&action, errp);
1266 }
1267 
1268 void qmp_blockdev_snapshot_internal_sync(const char *device,
1269                                          const char *name,
1270                                          Error **errp)
1271 {
1272     BlockdevSnapshotInternal snapshot = {
1273         .device = (char *) device,
1274         .name = (char *) name
1275     };
1276     TransactionAction action = {
1277         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1278         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1279     };
1280     blockdev_do_action(&action, errp);
1281 }
1282 
1283 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1284                                                          bool has_id,
1285                                                          const char *id,
1286                                                          bool has_name,
1287                                                          const char *name,
1288                                                          Error **errp)
1289 {
1290     BlockDriverState *bs;
1291     AioContext *aio_context;
1292     QEMUSnapshotInfo sn;
1293     Error *local_err = NULL;
1294     SnapshotInfo *info = NULL;
1295     int ret;
1296 
1297     bs = qmp_get_root_bs(device, errp);
1298     if (!bs) {
1299         return NULL;
1300     }
1301     aio_context = bdrv_get_aio_context(bs);
1302     aio_context_acquire(aio_context);
1303 
1304     if (!has_id) {
1305         id = NULL;
1306     }
1307 
1308     if (!has_name) {
1309         name = NULL;
1310     }
1311 
1312     if (!id && !name) {
1313         error_setg(errp, "Name or id must be provided");
1314         goto out_aio_context;
1315     }
1316 
1317     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1318         goto out_aio_context;
1319     }
1320 
1321     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1322     if (local_err) {
1323         error_propagate(errp, local_err);
1324         goto out_aio_context;
1325     }
1326     if (!ret) {
1327         error_setg(errp,
1328                    "Snapshot with id '%s' and name '%s' does not exist on "
1329                    "device '%s'",
1330                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1331         goto out_aio_context;
1332     }
1333 
1334     bdrv_snapshot_delete(bs, id, name, &local_err);
1335     if (local_err) {
1336         error_propagate(errp, local_err);
1337         goto out_aio_context;
1338     }
1339 
1340     aio_context_release(aio_context);
1341 
1342     info = g_new0(SnapshotInfo, 1);
1343     info->id = g_strdup(sn.id_str);
1344     info->name = g_strdup(sn.name);
1345     info->date_nsec = sn.date_nsec;
1346     info->date_sec = sn.date_sec;
1347     info->vm_state_size = sn.vm_state_size;
1348     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1349     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1350 
1351     return info;
1352 
1353 out_aio_context:
1354     aio_context_release(aio_context);
1355     return NULL;
1356 }
1357 
1358 /**
1359  * block_dirty_bitmap_lookup:
1360  * Return a dirty bitmap (if present), after validating
1361  * the node reference and bitmap names.
1362  *
1363  * @node: The name of the BDS node to search for bitmaps
1364  * @name: The name of the bitmap to search for
1365  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1366  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1367  * @errp: Output pointer for error information. Can be NULL.
1368  *
1369  * @return: A bitmap object on success, or NULL on failure.
1370  */
1371 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1372                                                   const char *name,
1373                                                   BlockDriverState **pbs,
1374                                                   Error **errp)
1375 {
1376     BlockDriverState *bs;
1377     BdrvDirtyBitmap *bitmap;
1378 
1379     if (!node) {
1380         error_setg(errp, "Node cannot be NULL");
1381         return NULL;
1382     }
1383     if (!name) {
1384         error_setg(errp, "Bitmap name cannot be NULL");
1385         return NULL;
1386     }
1387     bs = bdrv_lookup_bs(node, node, NULL);
1388     if (!bs) {
1389         error_setg(errp, "Node '%s' not found", node);
1390         return NULL;
1391     }
1392 
1393     bitmap = bdrv_find_dirty_bitmap(bs, name);
1394     if (!bitmap) {
1395         error_setg(errp, "Dirty bitmap '%s' not found", name);
1396         return NULL;
1397     }
1398 
1399     if (pbs) {
1400         *pbs = bs;
1401     }
1402 
1403     return bitmap;
1404 }
1405 
1406 /* New and old BlockDriverState structs for atomic group operations */
1407 
1408 typedef struct BlkActionState BlkActionState;
1409 
1410 /**
1411  * BlkActionOps:
1412  * Table of operations that define an Action.
1413  *
1414  * @instance_size: Size of state struct, in bytes.
1415  * @prepare: Prepare the work, must NOT be NULL.
1416  * @commit: Commit the changes, can be NULL.
1417  * @abort: Abort the changes on fail, can be NULL.
1418  * @clean: Clean up resources after all transaction actions have called
1419  *         commit() or abort(). Can be NULL.
1420  *
1421  * Only prepare() may fail. In a single transaction, only one of commit() or
1422  * abort() will be called. clean() will always be called if it is present.
1423  */
1424 typedef struct BlkActionOps {
1425     size_t instance_size;
1426     void (*prepare)(BlkActionState *common, Error **errp);
1427     void (*commit)(BlkActionState *common);
1428     void (*abort)(BlkActionState *common);
1429     void (*clean)(BlkActionState *common);
1430 } BlkActionOps;
1431 
1432 /**
1433  * BlkActionState:
1434  * Describes one Action's state within a Transaction.
1435  *
1436  * @action: QAPI-defined enum identifying which Action to perform.
1437  * @ops: Table of ActionOps this Action can perform.
1438  * @block_job_txn: Transaction which this action belongs to.
1439  * @entry: List membership for all Actions in this Transaction.
1440  *
1441  * This structure must be arranged as first member in a subclassed type,
1442  * assuming that the compiler will also arrange it to the same offsets as the
1443  * base class.
1444  */
1445 struct BlkActionState {
1446     TransactionAction *action;
1447     const BlkActionOps *ops;
1448     BlockJobTxn *block_job_txn;
1449     TransactionProperties *txn_props;
1450     QSIMPLEQ_ENTRY(BlkActionState) entry;
1451 };
1452 
1453 /* internal snapshot private data */
1454 typedef struct InternalSnapshotState {
1455     BlkActionState common;
1456     BlockDriverState *bs;
1457     QEMUSnapshotInfo sn;
1458     bool created;
1459 } InternalSnapshotState;
1460 
1461 
1462 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1463 {
1464     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1465         error_setg(errp,
1466                    "Action '%s' does not support Transaction property "
1467                    "completion-mode = %s",
1468                    TransactionActionKind_str(s->action->type),
1469                    ActionCompletionMode_str(s->txn_props->completion_mode));
1470         return -1;
1471     }
1472     return 0;
1473 }
1474 
1475 static void internal_snapshot_prepare(BlkActionState *common,
1476                                       Error **errp)
1477 {
1478     Error *local_err = NULL;
1479     const char *device;
1480     const char *name;
1481     BlockDriverState *bs;
1482     QEMUSnapshotInfo old_sn, *sn;
1483     bool ret;
1484     qemu_timeval tv;
1485     BlockdevSnapshotInternal *internal;
1486     InternalSnapshotState *state;
1487     AioContext *aio_context;
1488     int ret1;
1489 
1490     g_assert(common->action->type ==
1491              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1492     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1493     state = DO_UPCAST(InternalSnapshotState, common, common);
1494 
1495     /* 1. parse input */
1496     device = internal->device;
1497     name = internal->name;
1498 
1499     /* 2. check for validation */
1500     if (action_check_completion_mode(common, errp) < 0) {
1501         return;
1502     }
1503 
1504     bs = qmp_get_root_bs(device, errp);
1505     if (!bs) {
1506         return;
1507     }
1508 
1509     aio_context = bdrv_get_aio_context(bs);
1510     aio_context_acquire(aio_context);
1511 
1512     state->bs = bs;
1513 
1514     /* Paired with .clean() */
1515     bdrv_drained_begin(bs);
1516 
1517     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1518         goto out;
1519     }
1520 
1521     if (bdrv_is_read_only(bs)) {
1522         error_setg(errp, "Device '%s' is read only", device);
1523         goto out;
1524     }
1525 
1526     if (!bdrv_can_snapshot(bs)) {
1527         error_setg(errp, "Block format '%s' used by device '%s' "
1528                    "does not support internal snapshots",
1529                    bs->drv->format_name, device);
1530         goto out;
1531     }
1532 
1533     if (!strlen(name)) {
1534         error_setg(errp, "Name is empty");
1535         goto out;
1536     }
1537 
1538     /* check whether a snapshot with name exist */
1539     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1540                                             &local_err);
1541     if (local_err) {
1542         error_propagate(errp, local_err);
1543         goto out;
1544     } else if (ret) {
1545         error_setg(errp,
1546                    "Snapshot with name '%s' already exists on device '%s'",
1547                    name, device);
1548         goto out;
1549     }
1550 
1551     /* 3. take the snapshot */
1552     sn = &state->sn;
1553     pstrcpy(sn->name, sizeof(sn->name), name);
1554     qemu_gettimeofday(&tv);
1555     sn->date_sec = tv.tv_sec;
1556     sn->date_nsec = tv.tv_usec * 1000;
1557     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1558 
1559     ret1 = bdrv_snapshot_create(bs, sn);
1560     if (ret1 < 0) {
1561         error_setg_errno(errp, -ret1,
1562                          "Failed to create snapshot '%s' on device '%s'",
1563                          name, device);
1564         goto out;
1565     }
1566 
1567     /* 4. succeed, mark a snapshot is created */
1568     state->created = true;
1569 
1570 out:
1571     aio_context_release(aio_context);
1572 }
1573 
1574 static void internal_snapshot_abort(BlkActionState *common)
1575 {
1576     InternalSnapshotState *state =
1577                              DO_UPCAST(InternalSnapshotState, common, common);
1578     BlockDriverState *bs = state->bs;
1579     QEMUSnapshotInfo *sn = &state->sn;
1580     AioContext *aio_context;
1581     Error *local_error = NULL;
1582 
1583     if (!state->created) {
1584         return;
1585     }
1586 
1587     aio_context = bdrv_get_aio_context(state->bs);
1588     aio_context_acquire(aio_context);
1589 
1590     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1591         error_reportf_err(local_error,
1592                           "Failed to delete snapshot with id '%s' and "
1593                           "name '%s' on device '%s' in abort: ",
1594                           sn->id_str, sn->name,
1595                           bdrv_get_device_name(bs));
1596     }
1597 
1598     aio_context_release(aio_context);
1599 }
1600 
1601 static void internal_snapshot_clean(BlkActionState *common)
1602 {
1603     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1604                                              common, common);
1605     AioContext *aio_context;
1606 
1607     if (!state->bs) {
1608         return;
1609     }
1610 
1611     aio_context = bdrv_get_aio_context(state->bs);
1612     aio_context_acquire(aio_context);
1613 
1614     bdrv_drained_end(state->bs);
1615 
1616     aio_context_release(aio_context);
1617 }
1618 
1619 /* external snapshot private data */
1620 typedef struct ExternalSnapshotState {
1621     BlkActionState common;
1622     BlockDriverState *old_bs;
1623     BlockDriverState *new_bs;
1624     bool overlay_appended;
1625 } ExternalSnapshotState;
1626 
1627 static void external_snapshot_prepare(BlkActionState *common,
1628                                       Error **errp)
1629 {
1630     int flags = 0;
1631     QDict *options = NULL;
1632     Error *local_err = NULL;
1633     /* Device and node name of the image to generate the snapshot from */
1634     const char *device;
1635     const char *node_name;
1636     /* Reference to the new image (for 'blockdev-snapshot') */
1637     const char *snapshot_ref;
1638     /* File name of the new image (for 'blockdev-snapshot-sync') */
1639     const char *new_image_file;
1640     ExternalSnapshotState *state =
1641                              DO_UPCAST(ExternalSnapshotState, common, common);
1642     TransactionAction *action = common->action;
1643     AioContext *aio_context;
1644 
1645     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1646      * purpose but a different set of parameters */
1647     switch (action->type) {
1648     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1649         {
1650             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1651             device = s->node;
1652             node_name = s->node;
1653             new_image_file = NULL;
1654             snapshot_ref = s->overlay;
1655         }
1656         break;
1657     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1658         {
1659             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1660             device = s->has_device ? s->device : NULL;
1661             node_name = s->has_node_name ? s->node_name : NULL;
1662             new_image_file = s->snapshot_file;
1663             snapshot_ref = NULL;
1664         }
1665         break;
1666     default:
1667         g_assert_not_reached();
1668     }
1669 
1670     /* start processing */
1671     if (action_check_completion_mode(common, errp) < 0) {
1672         return;
1673     }
1674 
1675     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1676     if (!state->old_bs) {
1677         return;
1678     }
1679 
1680     aio_context = bdrv_get_aio_context(state->old_bs);
1681     aio_context_acquire(aio_context);
1682 
1683     /* Paired with .clean() */
1684     bdrv_drained_begin(state->old_bs);
1685 
1686     if (!bdrv_is_inserted(state->old_bs)) {
1687         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1688         goto out;
1689     }
1690 
1691     if (bdrv_op_is_blocked(state->old_bs,
1692                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1693         goto out;
1694     }
1695 
1696     if (!bdrv_is_read_only(state->old_bs)) {
1697         if (bdrv_flush(state->old_bs)) {
1698             error_setg(errp, QERR_IO_ERROR);
1699             goto out;
1700         }
1701     }
1702 
1703     if (!bdrv_is_first_non_filter(state->old_bs)) {
1704         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1705         goto out;
1706     }
1707 
1708     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1709         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1710         const char *format = s->has_format ? s->format : "qcow2";
1711         enum NewImageMode mode;
1712         const char *snapshot_node_name =
1713             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1714 
1715         if (node_name && !snapshot_node_name) {
1716             error_setg(errp, "New snapshot node name missing");
1717             goto out;
1718         }
1719 
1720         if (snapshot_node_name &&
1721             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1722             error_setg(errp, "New snapshot node name already in use");
1723             goto out;
1724         }
1725 
1726         flags = state->old_bs->open_flags;
1727         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1728         flags |= BDRV_O_NO_BACKING;
1729 
1730         /* create new image w/backing file */
1731         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1732         if (mode != NEW_IMAGE_MODE_EXISTING) {
1733             int64_t size = bdrv_getlength(state->old_bs);
1734             if (size < 0) {
1735                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1736                 goto out;
1737             }
1738             bdrv_img_create(new_image_file, format,
1739                             state->old_bs->filename,
1740                             state->old_bs->drv->format_name,
1741                             NULL, size, flags, false, &local_err);
1742             if (local_err) {
1743                 error_propagate(errp, local_err);
1744                 goto out;
1745             }
1746         }
1747 
1748         options = qdict_new();
1749         if (s->has_snapshot_node_name) {
1750             qdict_put_str(options, "node-name", snapshot_node_name);
1751         }
1752         qdict_put_str(options, "driver", format);
1753     }
1754 
1755     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1756                               errp);
1757     /* We will manually add the backing_hd field to the bs later */
1758     if (!state->new_bs) {
1759         goto out;
1760     }
1761 
1762     if (bdrv_has_blk(state->new_bs)) {
1763         error_setg(errp, "The snapshot is already in use");
1764         goto out;
1765     }
1766 
1767     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1768                            errp)) {
1769         goto out;
1770     }
1771 
1772     if (state->new_bs->backing != NULL) {
1773         error_setg(errp, "The snapshot already has a backing image");
1774         goto out;
1775     }
1776 
1777     if (!state->new_bs->drv->supports_backing) {
1778         error_setg(errp, "The snapshot does not support backing images");
1779         goto out;
1780     }
1781 
1782     bdrv_set_aio_context(state->new_bs, aio_context);
1783 
1784     /* This removes our old bs and adds the new bs. This is an operation that
1785      * can fail, so we need to do it in .prepare; undoing it for abort is
1786      * always possible. */
1787     bdrv_ref(state->new_bs);
1788     bdrv_append(state->new_bs, state->old_bs, &local_err);
1789     if (local_err) {
1790         error_propagate(errp, local_err);
1791         goto out;
1792     }
1793     state->overlay_appended = true;
1794 
1795 out:
1796     aio_context_release(aio_context);
1797 }
1798 
1799 static void external_snapshot_commit(BlkActionState *common)
1800 {
1801     ExternalSnapshotState *state =
1802                              DO_UPCAST(ExternalSnapshotState, common, common);
1803     AioContext *aio_context;
1804 
1805     aio_context = bdrv_get_aio_context(state->old_bs);
1806     aio_context_acquire(aio_context);
1807 
1808     /* We don't need (or want) to use the transactional
1809      * bdrv_reopen_multiple() across all the entries at once, because we
1810      * don't want to abort all of them if one of them fails the reopen */
1811     if (!atomic_read(&state->old_bs->copy_on_read)) {
1812         bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1813                     NULL);
1814     }
1815 
1816     aio_context_release(aio_context);
1817 }
1818 
1819 static void external_snapshot_abort(BlkActionState *common)
1820 {
1821     ExternalSnapshotState *state =
1822                              DO_UPCAST(ExternalSnapshotState, common, common);
1823     if (state->new_bs) {
1824         if (state->overlay_appended) {
1825             AioContext *aio_context;
1826 
1827             aio_context = bdrv_get_aio_context(state->old_bs);
1828             aio_context_acquire(aio_context);
1829 
1830             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1831                                           close state->old_bs; we need it */
1832             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1833             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1834             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1835 
1836             aio_context_release(aio_context);
1837         }
1838     }
1839 }
1840 
1841 static void external_snapshot_clean(BlkActionState *common)
1842 {
1843     ExternalSnapshotState *state =
1844                              DO_UPCAST(ExternalSnapshotState, common, common);
1845     AioContext *aio_context;
1846 
1847     if (!state->old_bs) {
1848         return;
1849     }
1850 
1851     aio_context = bdrv_get_aio_context(state->old_bs);
1852     aio_context_acquire(aio_context);
1853 
1854     bdrv_drained_end(state->old_bs);
1855     bdrv_unref(state->new_bs);
1856 
1857     aio_context_release(aio_context);
1858 }
1859 
1860 typedef struct DriveBackupState {
1861     BlkActionState common;
1862     BlockDriverState *bs;
1863     BlockJob *job;
1864 } DriveBackupState;
1865 
1866 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
1867                             Error **errp);
1868 
1869 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1870 {
1871     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1872     BlockDriverState *bs;
1873     DriveBackup *backup;
1874     AioContext *aio_context;
1875     Error *local_err = NULL;
1876 
1877     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1878     backup = common->action->u.drive_backup.data;
1879 
1880     bs = qmp_get_root_bs(backup->device, errp);
1881     if (!bs) {
1882         return;
1883     }
1884 
1885     aio_context = bdrv_get_aio_context(bs);
1886     aio_context_acquire(aio_context);
1887 
1888     /* Paired with .clean() */
1889     bdrv_drained_begin(bs);
1890 
1891     state->bs = bs;
1892 
1893     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1894     if (local_err) {
1895         error_propagate(errp, local_err);
1896         goto out;
1897     }
1898 
1899 out:
1900     aio_context_release(aio_context);
1901 }
1902 
1903 static void drive_backup_commit(BlkActionState *common)
1904 {
1905     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1906     AioContext *aio_context;
1907 
1908     aio_context = bdrv_get_aio_context(state->bs);
1909     aio_context_acquire(aio_context);
1910 
1911     assert(state->job);
1912     block_job_start(state->job);
1913 
1914     aio_context_release(aio_context);
1915 }
1916 
1917 static void drive_backup_abort(BlkActionState *common)
1918 {
1919     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1920 
1921     if (state->job) {
1922         AioContext *aio_context;
1923 
1924         aio_context = bdrv_get_aio_context(state->bs);
1925         aio_context_acquire(aio_context);
1926 
1927         block_job_cancel_sync(state->job);
1928 
1929         aio_context_release(aio_context);
1930     }
1931 }
1932 
1933 static void drive_backup_clean(BlkActionState *common)
1934 {
1935     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1936     AioContext *aio_context;
1937 
1938     if (!state->bs) {
1939         return;
1940     }
1941 
1942     aio_context = bdrv_get_aio_context(state->bs);
1943     aio_context_acquire(aio_context);
1944 
1945     bdrv_drained_end(state->bs);
1946 
1947     aio_context_release(aio_context);
1948 }
1949 
1950 typedef struct BlockdevBackupState {
1951     BlkActionState common;
1952     BlockDriverState *bs;
1953     BlockJob *job;
1954 } BlockdevBackupState;
1955 
1956 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
1957                                     Error **errp);
1958 
1959 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1960 {
1961     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1962     BlockdevBackup *backup;
1963     BlockDriverState *bs, *target;
1964     AioContext *aio_context;
1965     Error *local_err = NULL;
1966 
1967     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1968     backup = common->action->u.blockdev_backup.data;
1969 
1970     bs = qmp_get_root_bs(backup->device, errp);
1971     if (!bs) {
1972         return;
1973     }
1974 
1975     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1976     if (!target) {
1977         return;
1978     }
1979 
1980     aio_context = bdrv_get_aio_context(bs);
1981     if (aio_context != bdrv_get_aio_context(target)) {
1982         error_setg(errp, "Backup between two IO threads is not implemented");
1983         return;
1984     }
1985     aio_context_acquire(aio_context);
1986     state->bs = bs;
1987 
1988     /* Paired with .clean() */
1989     bdrv_drained_begin(state->bs);
1990 
1991     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1992     if (local_err) {
1993         error_propagate(errp, local_err);
1994         goto out;
1995     }
1996 
1997 out:
1998     aio_context_release(aio_context);
1999 }
2000 
2001 static void blockdev_backup_commit(BlkActionState *common)
2002 {
2003     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2004     AioContext *aio_context;
2005 
2006     aio_context = bdrv_get_aio_context(state->bs);
2007     aio_context_acquire(aio_context);
2008 
2009     assert(state->job);
2010     block_job_start(state->job);
2011 
2012     aio_context_release(aio_context);
2013 }
2014 
2015 static void blockdev_backup_abort(BlkActionState *common)
2016 {
2017     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2018 
2019     if (state->job) {
2020         AioContext *aio_context;
2021 
2022         aio_context = bdrv_get_aio_context(state->bs);
2023         aio_context_acquire(aio_context);
2024 
2025         block_job_cancel_sync(state->job);
2026 
2027         aio_context_release(aio_context);
2028     }
2029 }
2030 
2031 static void blockdev_backup_clean(BlkActionState *common)
2032 {
2033     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2034     AioContext *aio_context;
2035 
2036     if (!state->bs) {
2037         return;
2038     }
2039 
2040     aio_context = bdrv_get_aio_context(state->bs);
2041     aio_context_acquire(aio_context);
2042 
2043     bdrv_drained_end(state->bs);
2044 
2045     aio_context_release(aio_context);
2046 }
2047 
2048 typedef struct BlockDirtyBitmapState {
2049     BlkActionState common;
2050     BdrvDirtyBitmap *bitmap;
2051     BlockDriverState *bs;
2052     AioContext *aio_context;
2053     HBitmap *backup;
2054     bool prepared;
2055 } BlockDirtyBitmapState;
2056 
2057 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
2058                                            Error **errp)
2059 {
2060     Error *local_err = NULL;
2061     BlockDirtyBitmapAdd *action;
2062     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2063                                              common, common);
2064 
2065     if (action_check_completion_mode(common, errp) < 0) {
2066         return;
2067     }
2068 
2069     action = common->action->u.block_dirty_bitmap_add.data;
2070     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
2071     qmp_block_dirty_bitmap_add(action->node, action->name,
2072                                action->has_granularity, action->granularity,
2073                                action->has_persistent, action->persistent,
2074                                action->has_autoload, action->autoload,
2075                                &local_err);
2076 
2077     if (!local_err) {
2078         state->prepared = true;
2079     } else {
2080         error_propagate(errp, local_err);
2081     }
2082 }
2083 
2084 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2085 {
2086     BlockDirtyBitmapAdd *action;
2087     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2088                                              common, common);
2089 
2090     action = common->action->u.block_dirty_bitmap_add.data;
2091     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2092      * then the node reference and bitmap name must have been valid.
2093      */
2094     if (state->prepared) {
2095         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2096     }
2097 }
2098 
2099 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2100                                              Error **errp)
2101 {
2102     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2103                                              common, common);
2104     BlockDirtyBitmap *action;
2105 
2106     if (action_check_completion_mode(common, errp) < 0) {
2107         return;
2108     }
2109 
2110     action = common->action->u.block_dirty_bitmap_clear.data;
2111     state->bitmap = block_dirty_bitmap_lookup(action->node,
2112                                               action->name,
2113                                               &state->bs,
2114                                               errp);
2115     if (!state->bitmap) {
2116         return;
2117     }
2118 
2119     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2120         error_setg(errp, "Cannot modify a frozen bitmap");
2121         return;
2122     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2123         error_setg(errp, "Cannot clear a disabled bitmap");
2124         return;
2125     } else if (bdrv_dirty_bitmap_readonly(state->bitmap)) {
2126         error_setg(errp, "Cannot clear a readonly bitmap");
2127         return;
2128     }
2129 
2130     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2131     /* AioContext is released in .clean() */
2132 }
2133 
2134 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2135 {
2136     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2137                                              common, common);
2138 
2139     if (state->backup) {
2140         bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2141     }
2142 }
2143 
2144 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2145 {
2146     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2147                                              common, common);
2148 
2149     hbitmap_free(state->backup);
2150 }
2151 
2152 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2153 {
2154     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2155                                              common, common);
2156 
2157     if (state->aio_context) {
2158         aio_context_release(state->aio_context);
2159     }
2160 }
2161 
2162 static void abort_prepare(BlkActionState *common, Error **errp)
2163 {
2164     error_setg(errp, "Transaction aborted using Abort action");
2165 }
2166 
2167 static void abort_commit(BlkActionState *common)
2168 {
2169     g_assert_not_reached(); /* this action never succeeds */
2170 }
2171 
2172 static const BlkActionOps actions[] = {
2173     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2174         .instance_size = sizeof(ExternalSnapshotState),
2175         .prepare  = external_snapshot_prepare,
2176         .commit   = external_snapshot_commit,
2177         .abort = external_snapshot_abort,
2178         .clean = external_snapshot_clean,
2179     },
2180     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2181         .instance_size = sizeof(ExternalSnapshotState),
2182         .prepare  = external_snapshot_prepare,
2183         .commit   = external_snapshot_commit,
2184         .abort = external_snapshot_abort,
2185         .clean = external_snapshot_clean,
2186     },
2187     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2188         .instance_size = sizeof(DriveBackupState),
2189         .prepare = drive_backup_prepare,
2190         .commit = drive_backup_commit,
2191         .abort = drive_backup_abort,
2192         .clean = drive_backup_clean,
2193     },
2194     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2195         .instance_size = sizeof(BlockdevBackupState),
2196         .prepare = blockdev_backup_prepare,
2197         .commit = blockdev_backup_commit,
2198         .abort = blockdev_backup_abort,
2199         .clean = blockdev_backup_clean,
2200     },
2201     [TRANSACTION_ACTION_KIND_ABORT] = {
2202         .instance_size = sizeof(BlkActionState),
2203         .prepare = abort_prepare,
2204         .commit = abort_commit,
2205     },
2206     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2207         .instance_size = sizeof(InternalSnapshotState),
2208         .prepare  = internal_snapshot_prepare,
2209         .abort = internal_snapshot_abort,
2210         .clean = internal_snapshot_clean,
2211     },
2212     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2213         .instance_size = sizeof(BlockDirtyBitmapState),
2214         .prepare = block_dirty_bitmap_add_prepare,
2215         .abort = block_dirty_bitmap_add_abort,
2216     },
2217     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2218         .instance_size = sizeof(BlockDirtyBitmapState),
2219         .prepare = block_dirty_bitmap_clear_prepare,
2220         .commit = block_dirty_bitmap_clear_commit,
2221         .abort = block_dirty_bitmap_clear_abort,
2222         .clean = block_dirty_bitmap_clear_clean,
2223     }
2224 };
2225 
2226 /**
2227  * Allocate a TransactionProperties structure if necessary, and fill
2228  * that structure with desired defaults if they are unset.
2229  */
2230 static TransactionProperties *get_transaction_properties(
2231     TransactionProperties *props)
2232 {
2233     if (!props) {
2234         props = g_new0(TransactionProperties, 1);
2235     }
2236 
2237     if (!props->has_completion_mode) {
2238         props->has_completion_mode = true;
2239         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2240     }
2241 
2242     return props;
2243 }
2244 
2245 /*
2246  * 'Atomic' group operations.  The operations are performed as a set, and if
2247  * any fail then we roll back all operations in the group.
2248  */
2249 void qmp_transaction(TransactionActionList *dev_list,
2250                      bool has_props,
2251                      struct TransactionProperties *props,
2252                      Error **errp)
2253 {
2254     TransactionActionList *dev_entry = dev_list;
2255     BlockJobTxn *block_job_txn = NULL;
2256     BlkActionState *state, *next;
2257     Error *local_err = NULL;
2258 
2259     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2260     QSIMPLEQ_INIT(&snap_bdrv_states);
2261 
2262     /* Does this transaction get canceled as a group on failure?
2263      * If not, we don't really need to make a BlockJobTxn.
2264      */
2265     props = get_transaction_properties(props);
2266     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2267         block_job_txn = block_job_txn_new();
2268     }
2269 
2270     /* drain all i/o before any operations */
2271     bdrv_drain_all();
2272 
2273     /* We don't do anything in this loop that commits us to the operations */
2274     while (NULL != dev_entry) {
2275         TransactionAction *dev_info = NULL;
2276         const BlkActionOps *ops;
2277 
2278         dev_info = dev_entry->value;
2279         dev_entry = dev_entry->next;
2280 
2281         assert(dev_info->type < ARRAY_SIZE(actions));
2282 
2283         ops = &actions[dev_info->type];
2284         assert(ops->instance_size > 0);
2285 
2286         state = g_malloc0(ops->instance_size);
2287         state->ops = ops;
2288         state->action = dev_info;
2289         state->block_job_txn = block_job_txn;
2290         state->txn_props = props;
2291         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2292 
2293         state->ops->prepare(state, &local_err);
2294         if (local_err) {
2295             error_propagate(errp, local_err);
2296             goto delete_and_fail;
2297         }
2298     }
2299 
2300     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2301         if (state->ops->commit) {
2302             state->ops->commit(state);
2303         }
2304     }
2305 
2306     /* success */
2307     goto exit;
2308 
2309 delete_and_fail:
2310     /* failure, and it is all-or-none; roll back all operations */
2311     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2312         if (state->ops->abort) {
2313             state->ops->abort(state);
2314         }
2315     }
2316 exit:
2317     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2318         if (state->ops->clean) {
2319             state->ops->clean(state);
2320         }
2321         g_free(state);
2322     }
2323     if (!has_props) {
2324         qapi_free_TransactionProperties(props);
2325     }
2326     block_job_txn_unref(block_job_txn);
2327 }
2328 
2329 void qmp_eject(bool has_device, const char *device,
2330                bool has_id, const char *id,
2331                bool has_force, bool force, Error **errp)
2332 {
2333     Error *local_err = NULL;
2334     int rc;
2335 
2336     if (!has_force) {
2337         force = false;
2338     }
2339 
2340     rc = do_open_tray(has_device ? device : NULL,
2341                       has_id ? id : NULL,
2342                       force, &local_err);
2343     if (rc && rc != -ENOSYS) {
2344         error_propagate(errp, local_err);
2345         return;
2346     }
2347     error_free(local_err);
2348 
2349     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, errp);
2350 }
2351 
2352 void qmp_block_passwd(bool has_device, const char *device,
2353                       bool has_node_name, const char *node_name,
2354                       const char *password, Error **errp)
2355 {
2356     error_setg(errp,
2357                "Setting block passwords directly is no longer supported");
2358 }
2359 
2360 /*
2361  * Attempt to open the tray of @device.
2362  * If @force, ignore its tray lock.
2363  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2364  * On error, store an error through @errp and return -errno.
2365  * If @device does not exist, return -ENODEV.
2366  * If it has no removable media, return -ENOTSUP.
2367  * If it has no tray, return -ENOSYS.
2368  * If the guest was asked to open the tray, return -EINPROGRESS.
2369  * Else, return 0.
2370  */
2371 static int do_open_tray(const char *blk_name, const char *qdev_id,
2372                         bool force, Error **errp)
2373 {
2374     BlockBackend *blk;
2375     const char *device = qdev_id ?: blk_name;
2376     bool locked;
2377 
2378     blk = qmp_get_blk(blk_name, qdev_id, errp);
2379     if (!blk) {
2380         return -ENODEV;
2381     }
2382 
2383     if (!blk_dev_has_removable_media(blk)) {
2384         error_setg(errp, "Device '%s' is not removable", device);
2385         return -ENOTSUP;
2386     }
2387 
2388     if (!blk_dev_has_tray(blk)) {
2389         error_setg(errp, "Device '%s' does not have a tray", device);
2390         return -ENOSYS;
2391     }
2392 
2393     if (blk_dev_is_tray_open(blk)) {
2394         return 0;
2395     }
2396 
2397     locked = blk_dev_is_medium_locked(blk);
2398     if (locked) {
2399         blk_dev_eject_request(blk, force);
2400     }
2401 
2402     if (!locked || force) {
2403         blk_dev_change_media_cb(blk, false, &error_abort);
2404     }
2405 
2406     if (locked && !force) {
2407         error_setg(errp, "Device '%s' is locked and force was not specified, "
2408                    "wait for tray to open and try again", device);
2409         return -EINPROGRESS;
2410     }
2411 
2412     return 0;
2413 }
2414 
2415 void qmp_blockdev_open_tray(bool has_device, const char *device,
2416                             bool has_id, const char *id,
2417                             bool has_force, bool force,
2418                             Error **errp)
2419 {
2420     Error *local_err = NULL;
2421     int rc;
2422 
2423     if (!has_force) {
2424         force = false;
2425     }
2426     rc = do_open_tray(has_device ? device : NULL,
2427                       has_id ? id : NULL,
2428                       force, &local_err);
2429     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2430         error_propagate(errp, local_err);
2431         return;
2432     }
2433     error_free(local_err);
2434 }
2435 
2436 void qmp_blockdev_close_tray(bool has_device, const char *device,
2437                              bool has_id, const char *id,
2438                              Error **errp)
2439 {
2440     BlockBackend *blk;
2441     Error *local_err = NULL;
2442 
2443     device = has_device ? device : NULL;
2444     id = has_id ? id : NULL;
2445 
2446     blk = qmp_get_blk(device, id, errp);
2447     if (!blk) {
2448         return;
2449     }
2450 
2451     if (!blk_dev_has_removable_media(blk)) {
2452         error_setg(errp, "Device '%s' is not removable", device ?: id);
2453         return;
2454     }
2455 
2456     if (!blk_dev_has_tray(blk)) {
2457         /* Ignore this command on tray-less devices */
2458         return;
2459     }
2460 
2461     if (!blk_dev_is_tray_open(blk)) {
2462         return;
2463     }
2464 
2465     blk_dev_change_media_cb(blk, true, &local_err);
2466     if (local_err) {
2467         error_propagate(errp, local_err);
2468         return;
2469     }
2470 }
2471 
2472 void qmp_x_blockdev_remove_medium(bool has_device, const char *device,
2473                                   bool has_id, const char *id, Error **errp)
2474 {
2475     BlockBackend *blk;
2476     BlockDriverState *bs;
2477     AioContext *aio_context;
2478     bool has_attached_device;
2479 
2480     device = has_device ? device : NULL;
2481     id = has_id ? id : NULL;
2482 
2483     blk = qmp_get_blk(device, id, errp);
2484     if (!blk) {
2485         return;
2486     }
2487 
2488     /* For BBs without a device, we can exchange the BDS tree at will */
2489     has_attached_device = blk_get_attached_dev(blk);
2490 
2491     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2492         error_setg(errp, "Device '%s' is not removable", device ?: id);
2493         return;
2494     }
2495 
2496     if (has_attached_device && blk_dev_has_tray(blk) &&
2497         !blk_dev_is_tray_open(blk))
2498     {
2499         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2500         return;
2501     }
2502 
2503     bs = blk_bs(blk);
2504     if (!bs) {
2505         return;
2506     }
2507 
2508     aio_context = bdrv_get_aio_context(bs);
2509     aio_context_acquire(aio_context);
2510 
2511     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2512         goto out;
2513     }
2514 
2515     blk_remove_bs(blk);
2516 
2517     if (!blk_dev_has_tray(blk)) {
2518         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2519          * called at all); therefore, the medium needs to be ejected here.
2520          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2521          * value passed here (i.e. false). */
2522         blk_dev_change_media_cb(blk, false, &error_abort);
2523     }
2524 
2525 out:
2526     aio_context_release(aio_context);
2527 }
2528 
2529 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2530                                             BlockDriverState *bs, Error **errp)
2531 {
2532     Error *local_err = NULL;
2533     bool has_device;
2534     int ret;
2535 
2536     /* For BBs without a device, we can exchange the BDS tree at will */
2537     has_device = blk_get_attached_dev(blk);
2538 
2539     if (has_device && !blk_dev_has_removable_media(blk)) {
2540         error_setg(errp, "Device is not removable");
2541         return;
2542     }
2543 
2544     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2545         error_setg(errp, "Tray of the device is not open");
2546         return;
2547     }
2548 
2549     if (blk_bs(blk)) {
2550         error_setg(errp, "There already is a medium in the device");
2551         return;
2552     }
2553 
2554     ret = blk_insert_bs(blk, bs, errp);
2555     if (ret < 0) {
2556         return;
2557     }
2558 
2559     if (!blk_dev_has_tray(blk)) {
2560         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2561          * called at all); therefore, the medium needs to be pushed into the
2562          * slot here.
2563          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2564          * value passed here (i.e. true). */
2565         blk_dev_change_media_cb(blk, true, &local_err);
2566         if (local_err) {
2567             error_propagate(errp, local_err);
2568             blk_remove_bs(blk);
2569             return;
2570         }
2571     }
2572 }
2573 
2574 void qmp_x_blockdev_insert_medium(bool has_device, const char *device,
2575                                   bool has_id, const char *id,
2576                                   const char *node_name, Error **errp)
2577 {
2578     BlockBackend *blk;
2579     BlockDriverState *bs;
2580 
2581     blk = qmp_get_blk(has_device ? device : NULL,
2582                       has_id ? id : NULL,
2583                       errp);
2584     if (!blk) {
2585         return;
2586     }
2587 
2588     bs = bdrv_find_node(node_name);
2589     if (!bs) {
2590         error_setg(errp, "Node '%s' not found", node_name);
2591         return;
2592     }
2593 
2594     if (bdrv_has_blk(bs)) {
2595         error_setg(errp, "Node '%s' is already in use", node_name);
2596         return;
2597     }
2598 
2599     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2600 }
2601 
2602 void qmp_blockdev_change_medium(bool has_device, const char *device,
2603                                 bool has_id, const char *id,
2604                                 const char *filename,
2605                                 bool has_format, const char *format,
2606                                 bool has_read_only,
2607                                 BlockdevChangeReadOnlyMode read_only,
2608                                 Error **errp)
2609 {
2610     BlockBackend *blk;
2611     BlockDriverState *medium_bs = NULL;
2612     int bdrv_flags;
2613     bool detect_zeroes;
2614     int rc;
2615     QDict *options = NULL;
2616     Error *err = NULL;
2617 
2618     blk = qmp_get_blk(has_device ? device : NULL,
2619                       has_id ? id : NULL,
2620                       errp);
2621     if (!blk) {
2622         goto fail;
2623     }
2624 
2625     if (blk_bs(blk)) {
2626         blk_update_root_state(blk);
2627     }
2628 
2629     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2630     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2631         BDRV_O_PROTOCOL);
2632 
2633     if (!has_read_only) {
2634         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2635     }
2636 
2637     switch (read_only) {
2638     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2639         break;
2640 
2641     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2642         bdrv_flags &= ~BDRV_O_RDWR;
2643         break;
2644 
2645     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2646         bdrv_flags |= BDRV_O_RDWR;
2647         break;
2648 
2649     default:
2650         abort();
2651     }
2652 
2653     options = qdict_new();
2654     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2655     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2656 
2657     if (has_format) {
2658         qdict_put_str(options, "driver", format);
2659     }
2660 
2661     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2662     if (!medium_bs) {
2663         goto fail;
2664     }
2665 
2666     rc = do_open_tray(has_device ? device : NULL,
2667                       has_id ? id : NULL,
2668                       false, &err);
2669     if (rc && rc != -ENOSYS) {
2670         error_propagate(errp, err);
2671         goto fail;
2672     }
2673     error_free(err);
2674     err = NULL;
2675 
2676     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, &err);
2677     if (err) {
2678         error_propagate(errp, err);
2679         goto fail;
2680     }
2681 
2682     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2683     if (err) {
2684         error_propagate(errp, err);
2685         goto fail;
2686     }
2687 
2688     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2689 
2690 fail:
2691     /* If the medium has been inserted, the device has its own reference, so
2692      * ours must be relinquished; and if it has not been inserted successfully,
2693      * the reference must be relinquished anyway */
2694     bdrv_unref(medium_bs);
2695 }
2696 
2697 /* throttling disk I/O limits */
2698 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2699 {
2700     ThrottleConfig cfg;
2701     BlockDriverState *bs;
2702     BlockBackend *blk;
2703     AioContext *aio_context;
2704 
2705     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2706                       arg->has_id ? arg->id : NULL,
2707                       errp);
2708     if (!blk) {
2709         return;
2710     }
2711 
2712     aio_context = blk_get_aio_context(blk);
2713     aio_context_acquire(aio_context);
2714 
2715     bs = blk_bs(blk);
2716     if (!bs) {
2717         error_setg(errp, "Device has no medium");
2718         goto out;
2719     }
2720 
2721     throttle_config_init(&cfg);
2722     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2723     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2724     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2725 
2726     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2727     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2728     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2729 
2730     if (arg->has_bps_max) {
2731         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2732     }
2733     if (arg->has_bps_rd_max) {
2734         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2735     }
2736     if (arg->has_bps_wr_max) {
2737         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2738     }
2739     if (arg->has_iops_max) {
2740         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2741     }
2742     if (arg->has_iops_rd_max) {
2743         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2744     }
2745     if (arg->has_iops_wr_max) {
2746         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2747     }
2748 
2749     if (arg->has_bps_max_length) {
2750         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2751     }
2752     if (arg->has_bps_rd_max_length) {
2753         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2754     }
2755     if (arg->has_bps_wr_max_length) {
2756         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2757     }
2758     if (arg->has_iops_max_length) {
2759         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2760     }
2761     if (arg->has_iops_rd_max_length) {
2762         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2763     }
2764     if (arg->has_iops_wr_max_length) {
2765         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2766     }
2767 
2768     if (arg->has_iops_size) {
2769         cfg.op_size = arg->iops_size;
2770     }
2771 
2772     if (!throttle_is_valid(&cfg, errp)) {
2773         goto out;
2774     }
2775 
2776     if (throttle_enabled(&cfg)) {
2777         /* Enable I/O limits if they're not enabled yet, otherwise
2778          * just update the throttling group. */
2779         if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2780             blk_io_limits_enable(blk,
2781                                  arg->has_group ? arg->group :
2782                                  arg->has_device ? arg->device :
2783                                  arg->id);
2784         } else if (arg->has_group) {
2785             blk_io_limits_update_group(blk, arg->group);
2786         }
2787         /* Set the new throttling configuration */
2788         blk_set_io_limits(blk, &cfg);
2789     } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2790         /* If all throttling settings are set to 0, disable I/O limits */
2791         blk_io_limits_disable(blk);
2792     }
2793 
2794 out:
2795     aio_context_release(aio_context);
2796 }
2797 
2798 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2799                                 bool has_granularity, uint32_t granularity,
2800                                 bool has_persistent, bool persistent,
2801                                 bool has_autoload, bool autoload,
2802                                 Error **errp)
2803 {
2804     BlockDriverState *bs;
2805     BdrvDirtyBitmap *bitmap;
2806 
2807     if (!name || name[0] == '\0') {
2808         error_setg(errp, "Bitmap name cannot be empty");
2809         return;
2810     }
2811 
2812     bs = bdrv_lookup_bs(node, node, errp);
2813     if (!bs) {
2814         return;
2815     }
2816 
2817     if (has_granularity) {
2818         if (granularity < 512 || !is_power_of_2(granularity)) {
2819             error_setg(errp, "Granularity must be power of 2 "
2820                              "and at least 512");
2821             return;
2822         }
2823     } else {
2824         /* Default to cluster size, if available: */
2825         granularity = bdrv_get_default_bitmap_granularity(bs);
2826     }
2827 
2828     if (!has_persistent) {
2829         persistent = false;
2830     }
2831     if (!has_autoload) {
2832         autoload = false;
2833     }
2834 
2835     if (has_autoload && !persistent) {
2836         error_setg(errp, "Autoload flag must be used only for persistent "
2837                          "bitmaps");
2838         return;
2839     }
2840 
2841     if (persistent &&
2842         !bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp))
2843     {
2844         return;
2845     }
2846 
2847     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2848     if (bitmap == NULL) {
2849         return;
2850     }
2851 
2852     bdrv_dirty_bitmap_set_persistance(bitmap, persistent);
2853     bdrv_dirty_bitmap_set_autoload(bitmap, autoload);
2854 }
2855 
2856 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2857                                    Error **errp)
2858 {
2859     BlockDriverState *bs;
2860     BdrvDirtyBitmap *bitmap;
2861     Error *local_err = NULL;
2862 
2863     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2864     if (!bitmap || !bs) {
2865         return;
2866     }
2867 
2868     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2869         error_setg(errp,
2870                    "Bitmap '%s' is currently frozen and cannot be removed",
2871                    name);
2872         return;
2873     }
2874 
2875     if (bdrv_dirty_bitmap_get_persistance(bitmap)) {
2876         bdrv_remove_persistent_dirty_bitmap(bs, name, &local_err);
2877         if (local_err != NULL) {
2878             error_propagate(errp, local_err);
2879             return;
2880         }
2881     }
2882 
2883     bdrv_dirty_bitmap_make_anon(bitmap);
2884     bdrv_release_dirty_bitmap(bs, bitmap);
2885 }
2886 
2887 /**
2888  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2889  * immediately after a full backup operation.
2890  */
2891 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2892                                   Error **errp)
2893 {
2894     BdrvDirtyBitmap *bitmap;
2895     BlockDriverState *bs;
2896 
2897     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2898     if (!bitmap || !bs) {
2899         return;
2900     }
2901 
2902     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2903         error_setg(errp,
2904                    "Bitmap '%s' is currently frozen and cannot be modified",
2905                    name);
2906         return;
2907     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2908         error_setg(errp,
2909                    "Bitmap '%s' is currently disabled and cannot be cleared",
2910                    name);
2911         return;
2912     } else if (bdrv_dirty_bitmap_readonly(bitmap)) {
2913         error_setg(errp, "Bitmap '%s' is readonly and cannot be cleared", name);
2914         return;
2915     }
2916 
2917     bdrv_clear_dirty_bitmap(bitmap, NULL);
2918 }
2919 
2920 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2921                                                               const char *name,
2922                                                               Error **errp)
2923 {
2924     BdrvDirtyBitmap *bitmap;
2925     BlockDriverState *bs;
2926     BlockDirtyBitmapSha256 *ret = NULL;
2927     char *sha256;
2928 
2929     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2930     if (!bitmap || !bs) {
2931         return NULL;
2932     }
2933 
2934     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2935     if (sha256 == NULL) {
2936         return NULL;
2937     }
2938 
2939     ret = g_new(BlockDirtyBitmapSha256, 1);
2940     ret->sha256 = sha256;
2941 
2942     return ret;
2943 }
2944 
2945 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2946 {
2947     const char *id = qdict_get_str(qdict, "id");
2948     BlockBackend *blk;
2949     BlockDriverState *bs;
2950     AioContext *aio_context;
2951     Error *local_err = NULL;
2952 
2953     bs = bdrv_find_node(id);
2954     if (bs) {
2955         qmp_blockdev_del(id, &local_err);
2956         if (local_err) {
2957             error_report_err(local_err);
2958         }
2959         return;
2960     }
2961 
2962     blk = blk_by_name(id);
2963     if (!blk) {
2964         error_report("Device '%s' not found", id);
2965         return;
2966     }
2967 
2968     if (!blk_legacy_dinfo(blk)) {
2969         error_report("Deleting device added with blockdev-add"
2970                      " is not supported");
2971         return;
2972     }
2973 
2974     aio_context = blk_get_aio_context(blk);
2975     aio_context_acquire(aio_context);
2976 
2977     bs = blk_bs(blk);
2978     if (bs) {
2979         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2980             error_report_err(local_err);
2981             aio_context_release(aio_context);
2982             return;
2983         }
2984 
2985         blk_remove_bs(blk);
2986     }
2987 
2988     /* Make the BlockBackend and the attached BlockDriverState anonymous */
2989     monitor_remove_blk(blk);
2990 
2991     /* If this BlockBackend has a device attached to it, its refcount will be
2992      * decremented when the device is removed; otherwise we have to do so here.
2993      */
2994     if (blk_get_attached_dev(blk)) {
2995         /* Further I/O must not pause the guest */
2996         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2997                          BLOCKDEV_ON_ERROR_REPORT);
2998     } else {
2999         blk_unref(blk);
3000     }
3001 
3002     aio_context_release(aio_context);
3003 }
3004 
3005 void qmp_block_resize(bool has_device, const char *device,
3006                       bool has_node_name, const char *node_name,
3007                       int64_t size, Error **errp)
3008 {
3009     Error *local_err = NULL;
3010     BlockBackend *blk = NULL;
3011     BlockDriverState *bs;
3012     AioContext *aio_context;
3013     int ret;
3014 
3015     bs = bdrv_lookup_bs(has_device ? device : NULL,
3016                         has_node_name ? node_name : NULL,
3017                         &local_err);
3018     if (local_err) {
3019         error_propagate(errp, local_err);
3020         return;
3021     }
3022 
3023     aio_context = bdrv_get_aio_context(bs);
3024     aio_context_acquire(aio_context);
3025 
3026     if (!bdrv_is_first_non_filter(bs)) {
3027         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
3028         goto out;
3029     }
3030 
3031     if (size < 0) {
3032         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
3033         goto out;
3034     }
3035 
3036     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3037         error_setg(errp, QERR_DEVICE_IN_USE, device);
3038         goto out;
3039     }
3040 
3041     blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
3042     ret = blk_insert_bs(blk, bs, errp);
3043     if (ret < 0) {
3044         goto out;
3045     }
3046 
3047     bdrv_drained_begin(bs);
3048     ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
3049     bdrv_drained_end(bs);
3050 
3051 out:
3052     blk_unref(blk);
3053     aio_context_release(aio_context);
3054 }
3055 
3056 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
3057                       bool has_base, const char *base,
3058                       bool has_base_node, const char *base_node,
3059                       bool has_backing_file, const char *backing_file,
3060                       bool has_speed, int64_t speed,
3061                       bool has_on_error, BlockdevOnError on_error,
3062                       Error **errp)
3063 {
3064     BlockDriverState *bs, *iter;
3065     BlockDriverState *base_bs = NULL;
3066     AioContext *aio_context;
3067     Error *local_err = NULL;
3068     const char *base_name = NULL;
3069 
3070     if (!has_on_error) {
3071         on_error = BLOCKDEV_ON_ERROR_REPORT;
3072     }
3073 
3074     bs = bdrv_lookup_bs(device, device, errp);
3075     if (!bs) {
3076         return;
3077     }
3078 
3079     aio_context = bdrv_get_aio_context(bs);
3080     aio_context_acquire(aio_context);
3081 
3082     if (has_base && has_base_node) {
3083         error_setg(errp, "'base' and 'base-node' cannot be specified "
3084                    "at the same time");
3085         goto out;
3086     }
3087 
3088     if (has_base) {
3089         base_bs = bdrv_find_backing_image(bs, base);
3090         if (base_bs == NULL) {
3091             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3092             goto out;
3093         }
3094         assert(bdrv_get_aio_context(base_bs) == aio_context);
3095         base_name = base;
3096     }
3097 
3098     if (has_base_node) {
3099         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3100         if (!base_bs) {
3101             goto out;
3102         }
3103         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3104             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3105                        base_node, device);
3106             goto out;
3107         }
3108         assert(bdrv_get_aio_context(base_bs) == aio_context);
3109         base_name = base_bs->filename;
3110     }
3111 
3112     /* Check for op blockers in the whole chain between bs and base */
3113     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3114         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3115             goto out;
3116         }
3117     }
3118 
3119     /* if we are streaming the entire chain, the result will have no backing
3120      * file, and specifying one is therefore an error */
3121     if (base_bs == NULL && has_backing_file) {
3122         error_setg(errp, "backing file specified, but streaming the "
3123                          "entire chain");
3124         goto out;
3125     }
3126 
3127     /* backing_file string overrides base bs filename */
3128     base_name = has_backing_file ? backing_file : base_name;
3129 
3130     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3131                  has_speed ? speed : 0, on_error, &local_err);
3132     if (local_err) {
3133         error_propagate(errp, local_err);
3134         goto out;
3135     }
3136 
3137     trace_qmp_block_stream(bs, bs->job);
3138 
3139 out:
3140     aio_context_release(aio_context);
3141 }
3142 
3143 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3144                       bool has_base, const char *base,
3145                       bool has_top, const char *top,
3146                       bool has_backing_file, const char *backing_file,
3147                       bool has_speed, int64_t speed,
3148                       bool has_filter_node_name, const char *filter_node_name,
3149                       Error **errp)
3150 {
3151     BlockDriverState *bs;
3152     BlockDriverState *iter;
3153     BlockDriverState *base_bs, *top_bs;
3154     AioContext *aio_context;
3155     Error *local_err = NULL;
3156     /* This will be part of the QMP command, if/when the
3157      * BlockdevOnError change for blkmirror makes it in
3158      */
3159     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3160 
3161     if (!has_speed) {
3162         speed = 0;
3163     }
3164     if (!has_filter_node_name) {
3165         filter_node_name = NULL;
3166     }
3167 
3168     /* Important Note:
3169      *  libvirt relies on the DeviceNotFound error class in order to probe for
3170      *  live commit feature versions; for this to work, we must make sure to
3171      *  perform the device lookup before any generic errors that may occur in a
3172      *  scenario in which all optional arguments are omitted. */
3173     bs = qmp_get_root_bs(device, &local_err);
3174     if (!bs) {
3175         bs = bdrv_lookup_bs(device, device, NULL);
3176         if (!bs) {
3177             error_free(local_err);
3178             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3179                       "Device '%s' not found", device);
3180         } else {
3181             error_propagate(errp, local_err);
3182         }
3183         return;
3184     }
3185 
3186     aio_context = bdrv_get_aio_context(bs);
3187     aio_context_acquire(aio_context);
3188 
3189     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3190         goto out;
3191     }
3192 
3193     /* default top_bs is the active layer */
3194     top_bs = bs;
3195 
3196     if (has_top && top) {
3197         if (strcmp(bs->filename, top) != 0) {
3198             top_bs = bdrv_find_backing_image(bs, top);
3199         }
3200     }
3201 
3202     if (top_bs == NULL) {
3203         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3204         goto out;
3205     }
3206 
3207     assert(bdrv_get_aio_context(top_bs) == aio_context);
3208 
3209     if (has_base && base) {
3210         base_bs = bdrv_find_backing_image(top_bs, base);
3211     } else {
3212         base_bs = bdrv_find_base(top_bs);
3213     }
3214 
3215     if (base_bs == NULL) {
3216         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3217         goto out;
3218     }
3219 
3220     assert(bdrv_get_aio_context(base_bs) == aio_context);
3221 
3222     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3223         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3224             goto out;
3225         }
3226     }
3227 
3228     /* Do not allow attempts to commit an image into itself */
3229     if (top_bs == base_bs) {
3230         error_setg(errp, "cannot commit an image into itself");
3231         goto out;
3232     }
3233 
3234     if (top_bs == bs) {
3235         if (has_backing_file) {
3236             error_setg(errp, "'backing-file' specified,"
3237                              " but 'top' is the active layer");
3238             goto out;
3239         }
3240         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3241                             BLOCK_JOB_DEFAULT, speed, on_error,
3242                             filter_node_name, NULL, NULL, false, &local_err);
3243     } else {
3244         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3245         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3246             goto out;
3247         }
3248         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, speed,
3249                      on_error, has_backing_file ? backing_file : NULL,
3250                      filter_node_name, &local_err);
3251     }
3252     if (local_err != NULL) {
3253         error_propagate(errp, local_err);
3254         goto out;
3255     }
3256 
3257 out:
3258     aio_context_release(aio_context);
3259 }
3260 
3261 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
3262                                  Error **errp)
3263 {
3264     BlockDriverState *bs;
3265     BlockDriverState *target_bs;
3266     BlockDriverState *source = NULL;
3267     BlockJob *job = NULL;
3268     BdrvDirtyBitmap *bmap = NULL;
3269     AioContext *aio_context;
3270     QDict *options = NULL;
3271     Error *local_err = NULL;
3272     int flags;
3273     int64_t size;
3274     bool set_backing_hd = false;
3275 
3276     if (!backup->has_speed) {
3277         backup->speed = 0;
3278     }
3279     if (!backup->has_on_source_error) {
3280         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3281     }
3282     if (!backup->has_on_target_error) {
3283         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3284     }
3285     if (!backup->has_mode) {
3286         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3287     }
3288     if (!backup->has_job_id) {
3289         backup->job_id = NULL;
3290     }
3291     if (!backup->has_compress) {
3292         backup->compress = false;
3293     }
3294 
3295     bs = qmp_get_root_bs(backup->device, errp);
3296     if (!bs) {
3297         return NULL;
3298     }
3299 
3300     aio_context = bdrv_get_aio_context(bs);
3301     aio_context_acquire(aio_context);
3302 
3303     if (!backup->has_format) {
3304         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3305                          NULL : (char*) bs->drv->format_name;
3306     }
3307 
3308     /* Early check to avoid creating target */
3309     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3310         goto out;
3311     }
3312 
3313     flags = bs->open_flags | BDRV_O_RDWR;
3314 
3315     /* See if we have a backing HD we can use to create our new image
3316      * on top of. */
3317     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3318         source = backing_bs(bs);
3319         if (!source) {
3320             backup->sync = MIRROR_SYNC_MODE_FULL;
3321         }
3322     }
3323     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3324         source = bs;
3325         flags |= BDRV_O_NO_BACKING;
3326         set_backing_hd = true;
3327     }
3328 
3329     size = bdrv_getlength(bs);
3330     if (size < 0) {
3331         error_setg_errno(errp, -size, "bdrv_getlength failed");
3332         goto out;
3333     }
3334 
3335     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3336         assert(backup->format);
3337         if (source) {
3338             bdrv_img_create(backup->target, backup->format, source->filename,
3339                             source->drv->format_name, NULL,
3340                             size, flags, false, &local_err);
3341         } else {
3342             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3343                             size, flags, false, &local_err);
3344         }
3345     }
3346 
3347     if (local_err) {
3348         error_propagate(errp, local_err);
3349         goto out;
3350     }
3351 
3352     if (backup->format) {
3353         if (!options) {
3354             options = qdict_new();
3355         }
3356         qdict_put_str(options, "driver", backup->format);
3357     }
3358 
3359     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3360     if (!target_bs) {
3361         goto out;
3362     }
3363 
3364     bdrv_set_aio_context(target_bs, aio_context);
3365 
3366     if (set_backing_hd) {
3367         bdrv_set_backing_hd(target_bs, source, &local_err);
3368         if (local_err) {
3369             bdrv_unref(target_bs);
3370             goto out;
3371         }
3372     }
3373 
3374     if (backup->has_bitmap) {
3375         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3376         if (!bmap) {
3377             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3378             bdrv_unref(target_bs);
3379             goto out;
3380         }
3381     }
3382 
3383     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3384                             backup->sync, bmap, backup->compress,
3385                             backup->on_source_error, backup->on_target_error,
3386                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3387     bdrv_unref(target_bs);
3388     if (local_err != NULL) {
3389         error_propagate(errp, local_err);
3390         goto out;
3391     }
3392 
3393 out:
3394     aio_context_release(aio_context);
3395     return job;
3396 }
3397 
3398 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3399 {
3400 
3401     BlockJob *job;
3402     job = do_drive_backup(arg, NULL, errp);
3403     if (job) {
3404         block_job_start(job);
3405     }
3406 }
3407 
3408 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3409 {
3410     return bdrv_named_nodes_list(errp);
3411 }
3412 
3413 BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
3414                              Error **errp)
3415 {
3416     BlockDriverState *bs;
3417     BlockDriverState *target_bs;
3418     Error *local_err = NULL;
3419     AioContext *aio_context;
3420     BlockJob *job = NULL;
3421 
3422     if (!backup->has_speed) {
3423         backup->speed = 0;
3424     }
3425     if (!backup->has_on_source_error) {
3426         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3427     }
3428     if (!backup->has_on_target_error) {
3429         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3430     }
3431     if (!backup->has_job_id) {
3432         backup->job_id = NULL;
3433     }
3434     if (!backup->has_compress) {
3435         backup->compress = false;
3436     }
3437 
3438     bs = qmp_get_root_bs(backup->device, errp);
3439     if (!bs) {
3440         return NULL;
3441     }
3442 
3443     aio_context = bdrv_get_aio_context(bs);
3444     aio_context_acquire(aio_context);
3445 
3446     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3447     if (!target_bs) {
3448         goto out;
3449     }
3450 
3451     if (bdrv_get_aio_context(target_bs) != aio_context) {
3452         if (!bdrv_has_blk(target_bs)) {
3453             /* The target BDS is not attached, we can safely move it to another
3454              * AioContext. */
3455             bdrv_set_aio_context(target_bs, aio_context);
3456         } else {
3457             error_setg(errp, "Target is attached to a different thread from "
3458                              "source.");
3459             goto out;
3460         }
3461     }
3462     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3463                             backup->sync, NULL, backup->compress,
3464                             backup->on_source_error, backup->on_target_error,
3465                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3466     if (local_err != NULL) {
3467         error_propagate(errp, local_err);
3468     }
3469 out:
3470     aio_context_release(aio_context);
3471     return job;
3472 }
3473 
3474 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3475 {
3476     BlockJob *job;
3477     job = do_blockdev_backup(arg, NULL, errp);
3478     if (job) {
3479         block_job_start(job);
3480     }
3481 }
3482 
3483 /* Parameter check and block job starting for drive mirroring.
3484  * Caller should hold @device and @target's aio context (must be the same).
3485  **/
3486 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3487                                    BlockDriverState *target,
3488                                    bool has_replaces, const char *replaces,
3489                                    enum MirrorSyncMode sync,
3490                                    BlockMirrorBackingMode backing_mode,
3491                                    bool has_speed, int64_t speed,
3492                                    bool has_granularity, uint32_t granularity,
3493                                    bool has_buf_size, int64_t buf_size,
3494                                    bool has_on_source_error,
3495                                    BlockdevOnError on_source_error,
3496                                    bool has_on_target_error,
3497                                    BlockdevOnError on_target_error,
3498                                    bool has_unmap, bool unmap,
3499                                    bool has_filter_node_name,
3500                                    const char *filter_node_name,
3501                                    Error **errp)
3502 {
3503 
3504     if (!has_speed) {
3505         speed = 0;
3506     }
3507     if (!has_on_source_error) {
3508         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3509     }
3510     if (!has_on_target_error) {
3511         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3512     }
3513     if (!has_granularity) {
3514         granularity = 0;
3515     }
3516     if (!has_buf_size) {
3517         buf_size = 0;
3518     }
3519     if (!has_unmap) {
3520         unmap = true;
3521     }
3522     if (!has_filter_node_name) {
3523         filter_node_name = NULL;
3524     }
3525 
3526     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3527         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3528                    "a value in range [512B, 64MB]");
3529         return;
3530     }
3531     if (granularity & (granularity - 1)) {
3532         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3533                    "power of 2");
3534         return;
3535     }
3536 
3537     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3538         return;
3539     }
3540     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3541         return;
3542     }
3543 
3544     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3545         sync = MIRROR_SYNC_MODE_FULL;
3546     }
3547 
3548     /* pass the node name to replace to mirror start since it's loose coupling
3549      * and will allow to check whether the node still exist at mirror completion
3550      */
3551     mirror_start(job_id, bs, target,
3552                  has_replaces ? replaces : NULL,
3553                  speed, granularity, buf_size, sync, backing_mode,
3554                  on_source_error, on_target_error, unmap, filter_node_name,
3555                  errp);
3556 }
3557 
3558 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3559 {
3560     BlockDriverState *bs;
3561     BlockDriverState *source, *target_bs;
3562     AioContext *aio_context;
3563     BlockMirrorBackingMode backing_mode;
3564     Error *local_err = NULL;
3565     QDict *options = NULL;
3566     int flags;
3567     int64_t size;
3568     const char *format = arg->format;
3569 
3570     bs = qmp_get_root_bs(arg->device, errp);
3571     if (!bs) {
3572         return;
3573     }
3574 
3575     aio_context = bdrv_get_aio_context(bs);
3576     aio_context_acquire(aio_context);
3577 
3578     if (!arg->has_mode) {
3579         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3580     }
3581 
3582     if (!arg->has_format) {
3583         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3584                   ? NULL : bs->drv->format_name);
3585     }
3586 
3587     flags = bs->open_flags | BDRV_O_RDWR;
3588     source = backing_bs(bs);
3589     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3590         arg->sync = MIRROR_SYNC_MODE_FULL;
3591     }
3592     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3593         source = bs;
3594     }
3595 
3596     size = bdrv_getlength(bs);
3597     if (size < 0) {
3598         error_setg_errno(errp, -size, "bdrv_getlength failed");
3599         goto out;
3600     }
3601 
3602     if (arg->has_replaces) {
3603         BlockDriverState *to_replace_bs;
3604         AioContext *replace_aio_context;
3605         int64_t replace_size;
3606 
3607         if (!arg->has_node_name) {
3608             error_setg(errp, "a node-name must be provided when replacing a"
3609                              " named node of the graph");
3610             goto out;
3611         }
3612 
3613         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3614 
3615         if (!to_replace_bs) {
3616             error_propagate(errp, local_err);
3617             goto out;
3618         }
3619 
3620         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3621         aio_context_acquire(replace_aio_context);
3622         replace_size = bdrv_getlength(to_replace_bs);
3623         aio_context_release(replace_aio_context);
3624 
3625         if (size != replace_size) {
3626             error_setg(errp, "cannot replace image with a mirror image of "
3627                              "different size");
3628             goto out;
3629         }
3630     }
3631 
3632     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3633         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3634     } else {
3635         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3636     }
3637 
3638     /* Don't open backing image in create() */
3639     flags |= BDRV_O_NO_BACKING;
3640 
3641     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3642         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3643     {
3644         /* create new image w/o backing file */
3645         assert(format);
3646         bdrv_img_create(arg->target, format,
3647                         NULL, NULL, NULL, size, flags, false, &local_err);
3648     } else {
3649         switch (arg->mode) {
3650         case NEW_IMAGE_MODE_EXISTING:
3651             break;
3652         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3653             /* create new image with backing file */
3654             bdrv_img_create(arg->target, format,
3655                             source->filename,
3656                             source->drv->format_name,
3657                             NULL, size, flags, false, &local_err);
3658             break;
3659         default:
3660             abort();
3661         }
3662     }
3663 
3664     if (local_err) {
3665         error_propagate(errp, local_err);
3666         goto out;
3667     }
3668 
3669     options = qdict_new();
3670     if (arg->has_node_name) {
3671         qdict_put_str(options, "node-name", arg->node_name);
3672     }
3673     if (format) {
3674         qdict_put_str(options, "driver", format);
3675     }
3676 
3677     /* Mirroring takes care of copy-on-write using the source's backing
3678      * file.
3679      */
3680     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3681     if (!target_bs) {
3682         goto out;
3683     }
3684 
3685     bdrv_set_aio_context(target_bs, aio_context);
3686 
3687     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3688                            arg->has_replaces, arg->replaces, arg->sync,
3689                            backing_mode, arg->has_speed, arg->speed,
3690                            arg->has_granularity, arg->granularity,
3691                            arg->has_buf_size, arg->buf_size,
3692                            arg->has_on_source_error, arg->on_source_error,
3693                            arg->has_on_target_error, arg->on_target_error,
3694                            arg->has_unmap, arg->unmap,
3695                            false, NULL,
3696                            &local_err);
3697     bdrv_unref(target_bs);
3698     error_propagate(errp, local_err);
3699 out:
3700     aio_context_release(aio_context);
3701 }
3702 
3703 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3704                          const char *device, const char *target,
3705                          bool has_replaces, const char *replaces,
3706                          MirrorSyncMode sync,
3707                          bool has_speed, int64_t speed,
3708                          bool has_granularity, uint32_t granularity,
3709                          bool has_buf_size, int64_t buf_size,
3710                          bool has_on_source_error,
3711                          BlockdevOnError on_source_error,
3712                          bool has_on_target_error,
3713                          BlockdevOnError on_target_error,
3714                          bool has_filter_node_name,
3715                          const char *filter_node_name,
3716                          Error **errp)
3717 {
3718     BlockDriverState *bs;
3719     BlockDriverState *target_bs;
3720     AioContext *aio_context;
3721     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3722     Error *local_err = NULL;
3723 
3724     bs = qmp_get_root_bs(device, errp);
3725     if (!bs) {
3726         return;
3727     }
3728 
3729     target_bs = bdrv_lookup_bs(target, target, errp);
3730     if (!target_bs) {
3731         return;
3732     }
3733 
3734     aio_context = bdrv_get_aio_context(bs);
3735     aio_context_acquire(aio_context);
3736 
3737     bdrv_set_aio_context(target_bs, aio_context);
3738 
3739     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3740                            has_replaces, replaces, sync, backing_mode,
3741                            has_speed, speed,
3742                            has_granularity, granularity,
3743                            has_buf_size, buf_size,
3744                            has_on_source_error, on_source_error,
3745                            has_on_target_error, on_target_error,
3746                            true, true,
3747                            has_filter_node_name, filter_node_name,
3748                            &local_err);
3749     error_propagate(errp, local_err);
3750 
3751     aio_context_release(aio_context);
3752 }
3753 
3754 /* Get a block job using its ID and acquire its AioContext */
3755 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3756                                 Error **errp)
3757 {
3758     BlockJob *job;
3759 
3760     assert(id != NULL);
3761 
3762     *aio_context = NULL;
3763 
3764     job = block_job_get(id);
3765 
3766     if (!job) {
3767         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3768                   "Block job '%s' not found", id);
3769         return NULL;
3770     }
3771 
3772     *aio_context = blk_get_aio_context(job->blk);
3773     aio_context_acquire(*aio_context);
3774 
3775     return job;
3776 }
3777 
3778 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3779 {
3780     AioContext *aio_context;
3781     BlockJob *job = find_block_job(device, &aio_context, errp);
3782 
3783     if (!job) {
3784         return;
3785     }
3786 
3787     block_job_set_speed(job, speed, errp);
3788     aio_context_release(aio_context);
3789 }
3790 
3791 void qmp_block_job_cancel(const char *device,
3792                           bool has_force, bool force, Error **errp)
3793 {
3794     AioContext *aio_context;
3795     BlockJob *job = find_block_job(device, &aio_context, errp);
3796 
3797     if (!job) {
3798         return;
3799     }
3800 
3801     if (!has_force) {
3802         force = false;
3803     }
3804 
3805     if (block_job_user_paused(job) && !force) {
3806         error_setg(errp, "The block job for device '%s' is currently paused",
3807                    device);
3808         goto out;
3809     }
3810 
3811     trace_qmp_block_job_cancel(job);
3812     block_job_cancel(job);
3813 out:
3814     aio_context_release(aio_context);
3815 }
3816 
3817 void qmp_block_job_pause(const char *device, Error **errp)
3818 {
3819     AioContext *aio_context;
3820     BlockJob *job = find_block_job(device, &aio_context, errp);
3821 
3822     if (!job || block_job_user_paused(job)) {
3823         return;
3824     }
3825 
3826     trace_qmp_block_job_pause(job);
3827     block_job_user_pause(job);
3828     aio_context_release(aio_context);
3829 }
3830 
3831 void qmp_block_job_resume(const char *device, Error **errp)
3832 {
3833     AioContext *aio_context;
3834     BlockJob *job = find_block_job(device, &aio_context, errp);
3835 
3836     if (!job || !block_job_user_paused(job)) {
3837         return;
3838     }
3839 
3840     trace_qmp_block_job_resume(job);
3841     block_job_user_resume(job);
3842     aio_context_release(aio_context);
3843 }
3844 
3845 void qmp_block_job_complete(const char *device, Error **errp)
3846 {
3847     AioContext *aio_context;
3848     BlockJob *job = find_block_job(device, &aio_context, errp);
3849 
3850     if (!job) {
3851         return;
3852     }
3853 
3854     trace_qmp_block_job_complete(job);
3855     block_job_complete(job, errp);
3856     aio_context_release(aio_context);
3857 }
3858 
3859 void qmp_change_backing_file(const char *device,
3860                              const char *image_node_name,
3861                              const char *backing_file,
3862                              Error **errp)
3863 {
3864     BlockDriverState *bs = NULL;
3865     AioContext *aio_context;
3866     BlockDriverState *image_bs = NULL;
3867     Error *local_err = NULL;
3868     bool ro;
3869     int open_flags;
3870     int ret;
3871 
3872     bs = qmp_get_root_bs(device, errp);
3873     if (!bs) {
3874         return;
3875     }
3876 
3877     aio_context = bdrv_get_aio_context(bs);
3878     aio_context_acquire(aio_context);
3879 
3880     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3881     if (local_err) {
3882         error_propagate(errp, local_err);
3883         goto out;
3884     }
3885 
3886     if (!image_bs) {
3887         error_setg(errp, "image file not found");
3888         goto out;
3889     }
3890 
3891     if (bdrv_find_base(image_bs) == image_bs) {
3892         error_setg(errp, "not allowing backing file change on an image "
3893                          "without a backing file");
3894         goto out;
3895     }
3896 
3897     /* even though we are not necessarily operating on bs, we need it to
3898      * determine if block ops are currently prohibited on the chain */
3899     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3900         goto out;
3901     }
3902 
3903     /* final sanity check */
3904     if (!bdrv_chain_contains(bs, image_bs)) {
3905         error_setg(errp, "'%s' and image file are not in the same chain",
3906                    device);
3907         goto out;
3908     }
3909 
3910     /* if not r/w, reopen to make r/w */
3911     open_flags = image_bs->open_flags;
3912     ro = bdrv_is_read_only(image_bs);
3913 
3914     if (ro) {
3915         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3916         if (local_err) {
3917             error_propagate(errp, local_err);
3918             goto out;
3919         }
3920     }
3921 
3922     ret = bdrv_change_backing_file(image_bs, backing_file,
3923                                image_bs->drv ? image_bs->drv->format_name : "");
3924 
3925     if (ret < 0) {
3926         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3927                          backing_file);
3928         /* don't exit here, so we can try to restore open flags if
3929          * appropriate */
3930     }
3931 
3932     if (ro) {
3933         bdrv_reopen(image_bs, open_flags, &local_err);
3934         error_propagate(errp, local_err);
3935     }
3936 
3937 out:
3938     aio_context_release(aio_context);
3939 }
3940 
3941 void hmp_drive_add_node(Monitor *mon, const char *optstr)
3942 {
3943     QemuOpts *opts;
3944     QDict *qdict;
3945     Error *local_err = NULL;
3946 
3947     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
3948     if (!opts) {
3949         return;
3950     }
3951 
3952     qdict = qemu_opts_to_qdict(opts, NULL);
3953 
3954     if (!qdict_get_try_str(qdict, "node-name")) {
3955         QDECREF(qdict);
3956         error_report("'node-name' needs to be specified");
3957         goto out;
3958     }
3959 
3960     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
3961     if (!bs) {
3962         error_report_err(local_err);
3963         goto out;
3964     }
3965 
3966     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3967 
3968 out:
3969     qemu_opts_del(opts);
3970 }
3971 
3972 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3973 {
3974     BlockDriverState *bs;
3975     QObject *obj;
3976     Visitor *v = qobject_output_visitor_new(&obj);
3977     QDict *qdict;
3978     const QDictEntry *ent;
3979     Error *local_err = NULL;
3980 
3981     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
3982     if (local_err) {
3983         error_propagate(errp, local_err);
3984         goto fail;
3985     }
3986 
3987     visit_complete(v, &obj);
3988     qdict = qobject_to_qdict(obj);
3989 
3990     qdict_flatten(qdict);
3991 
3992     /*
3993      * Rewrite "backing": null to "backing": ""
3994      * TODO Rewrite "" to null instead, and perhaps not even here
3995      */
3996     for (ent = qdict_first(qdict); ent; ent = qdict_next(qdict, ent)) {
3997         char *dot = strrchr(ent->key, '.');
3998 
3999         if (!strcmp(dot ? dot + 1 : ent->key, "backing")
4000             && qobject_type(ent->value) == QTYPE_QNULL) {
4001             qdict_put(qdict, ent->key, qstring_new());
4002         }
4003     }
4004 
4005     if (!qdict_get_try_str(qdict, "node-name")) {
4006         error_setg(errp, "'node-name' must be specified for the root node");
4007         goto fail;
4008     }
4009 
4010     bs = bds_tree_init(qdict, errp);
4011     if (!bs) {
4012         goto fail;
4013     }
4014 
4015     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4016 
4017 fail:
4018     visit_free(v);
4019 }
4020 
4021 void qmp_blockdev_del(const char *node_name, Error **errp)
4022 {
4023     AioContext *aio_context;
4024     BlockDriverState *bs;
4025 
4026     bs = bdrv_find_node(node_name);
4027     if (!bs) {
4028         error_setg(errp, "Cannot find node %s", node_name);
4029         return;
4030     }
4031     if (bdrv_has_blk(bs)) {
4032         error_setg(errp, "Node %s is in use", node_name);
4033         return;
4034     }
4035     aio_context = bdrv_get_aio_context(bs);
4036     aio_context_acquire(aio_context);
4037 
4038     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4039         goto out;
4040     }
4041 
4042     if (!bs->monitor_list.tqe_prev) {
4043         error_setg(errp, "Node %s is not owned by the monitor",
4044                    bs->node_name);
4045         goto out;
4046     }
4047 
4048     if (bs->refcnt > 1) {
4049         error_setg(errp, "Block device %s is in use",
4050                    bdrv_get_device_or_node_name(bs));
4051         goto out;
4052     }
4053 
4054     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4055     bdrv_unref(bs);
4056 
4057 out:
4058     aio_context_release(aio_context);
4059 }
4060 
4061 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
4062                                   const char *child_name)
4063 {
4064     BdrvChild *child;
4065 
4066     QLIST_FOREACH(child, &parent_bs->children, next) {
4067         if (strcmp(child->name, child_name) == 0) {
4068             return child;
4069         }
4070     }
4071 
4072     return NULL;
4073 }
4074 
4075 void qmp_x_blockdev_change(const char *parent, bool has_child,
4076                            const char *child, bool has_node,
4077                            const char *node, Error **errp)
4078 {
4079     BlockDriverState *parent_bs, *new_bs = NULL;
4080     BdrvChild *p_child;
4081 
4082     parent_bs = bdrv_lookup_bs(parent, parent, errp);
4083     if (!parent_bs) {
4084         return;
4085     }
4086 
4087     if (has_child == has_node) {
4088         if (has_child) {
4089             error_setg(errp, "The parameters child and node are in conflict");
4090         } else {
4091             error_setg(errp, "Either child or node must be specified");
4092         }
4093         return;
4094     }
4095 
4096     if (has_child) {
4097         p_child = bdrv_find_child(parent_bs, child);
4098         if (!p_child) {
4099             error_setg(errp, "Node '%s' does not have child '%s'",
4100                        parent, child);
4101             return;
4102         }
4103         bdrv_del_child(parent_bs, p_child, errp);
4104     }
4105 
4106     if (has_node) {
4107         new_bs = bdrv_find_node(node);
4108         if (!new_bs) {
4109             error_setg(errp, "Node '%s' not found", node);
4110             return;
4111         }
4112         bdrv_add_child(parent_bs, new_bs, errp);
4113     }
4114 }
4115 
4116 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4117 {
4118     BlockJobInfoList *head = NULL, **p_next = &head;
4119     BlockJob *job;
4120 
4121     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4122         BlockJobInfoList *elem;
4123         AioContext *aio_context;
4124 
4125         if (block_job_is_internal(job)) {
4126             continue;
4127         }
4128         elem = g_new0(BlockJobInfoList, 1);
4129         aio_context = blk_get_aio_context(job->blk);
4130         aio_context_acquire(aio_context);
4131         elem->value = block_job_query(job, errp);
4132         aio_context_release(aio_context);
4133         if (!elem->value) {
4134             g_free(elem);
4135             qapi_free_BlockJobInfoList(head);
4136             return NULL;
4137         }
4138         *p_next = elem;
4139         p_next = &elem->next;
4140     }
4141 
4142     return head;
4143 }
4144 
4145 QemuOptsList qemu_common_drive_opts = {
4146     .name = "drive",
4147     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4148     .desc = {
4149         {
4150             .name = "snapshot",
4151             .type = QEMU_OPT_BOOL,
4152             .help = "enable/disable snapshot mode",
4153         },{
4154             .name = "aio",
4155             .type = QEMU_OPT_STRING,
4156             .help = "host AIO implementation (threads, native)",
4157         },{
4158             .name = BDRV_OPT_CACHE_WB,
4159             .type = QEMU_OPT_BOOL,
4160             .help = "Enable writeback mode",
4161         },{
4162             .name = "format",
4163             .type = QEMU_OPT_STRING,
4164             .help = "disk format (raw, qcow2, ...)",
4165         },{
4166             .name = "rerror",
4167             .type = QEMU_OPT_STRING,
4168             .help = "read error action",
4169         },{
4170             .name = "werror",
4171             .type = QEMU_OPT_STRING,
4172             .help = "write error action",
4173         },{
4174             .name = BDRV_OPT_READ_ONLY,
4175             .type = QEMU_OPT_BOOL,
4176             .help = "open drive file as read-only",
4177         },
4178 
4179         THROTTLE_OPTS,
4180 
4181         {
4182             .name = "throttling.group",
4183             .type = QEMU_OPT_STRING,
4184             .help = "name of the block throttling group",
4185         },{
4186             .name = "copy-on-read",
4187             .type = QEMU_OPT_BOOL,
4188             .help = "copy read data from backing file into image file",
4189         },{
4190             .name = "detect-zeroes",
4191             .type = QEMU_OPT_STRING,
4192             .help = "try to optimize zero writes (off, on, unmap)",
4193         },{
4194             .name = "stats-account-invalid",
4195             .type = QEMU_OPT_BOOL,
4196             .help = "whether to account for invalid I/O operations "
4197                     "in the statistics",
4198         },{
4199             .name = "stats-account-failed",
4200             .type = QEMU_OPT_BOOL,
4201             .help = "whether to account for failed I/O operations "
4202                     "in the statistics",
4203         },
4204         { /* end of list */ }
4205     },
4206 };
4207 
4208 QemuOptsList qemu_drive_opts = {
4209     .name = "drive",
4210     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4211     .desc = {
4212         /*
4213          * no elements => accept any params
4214          * validation will happen later
4215          */
4216         { /* end of list */ }
4217     },
4218 };
4219