xref: /openbmc/qemu/blockdev.c (revision a8583393)
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     AioContext *aio_context;
1458     QEMUSnapshotInfo sn;
1459     bool created;
1460 } InternalSnapshotState;
1461 
1462 
1463 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1464 {
1465     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1466         error_setg(errp,
1467                    "Action '%s' does not support Transaction property "
1468                    "completion-mode = %s",
1469                    TransactionActionKind_str(s->action->type),
1470                    ActionCompletionMode_str(s->txn_props->completion_mode));
1471         return -1;
1472     }
1473     return 0;
1474 }
1475 
1476 static void internal_snapshot_prepare(BlkActionState *common,
1477                                       Error **errp)
1478 {
1479     Error *local_err = NULL;
1480     const char *device;
1481     const char *name;
1482     BlockDriverState *bs;
1483     QEMUSnapshotInfo old_sn, *sn;
1484     bool ret;
1485     qemu_timeval tv;
1486     BlockdevSnapshotInternal *internal;
1487     InternalSnapshotState *state;
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     /* AioContext is released in .clean() */
1510     state->aio_context = bdrv_get_aio_context(bs);
1511     aio_context_acquire(state->aio_context);
1512 
1513     state->bs = bs;
1514     bdrv_drained_begin(bs);
1515 
1516     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1517         return;
1518     }
1519 
1520     if (bdrv_is_read_only(bs)) {
1521         error_setg(errp, "Device '%s' is read only", device);
1522         return;
1523     }
1524 
1525     if (!bdrv_can_snapshot(bs)) {
1526         error_setg(errp, "Block format '%s' used by device '%s' "
1527                    "does not support internal snapshots",
1528                    bs->drv->format_name, device);
1529         return;
1530     }
1531 
1532     if (!strlen(name)) {
1533         error_setg(errp, "Name is empty");
1534         return;
1535     }
1536 
1537     /* check whether a snapshot with name exist */
1538     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1539                                             &local_err);
1540     if (local_err) {
1541         error_propagate(errp, local_err);
1542         return;
1543     } else if (ret) {
1544         error_setg(errp,
1545                    "Snapshot with name '%s' already exists on device '%s'",
1546                    name, device);
1547         return;
1548     }
1549 
1550     /* 3. take the snapshot */
1551     sn = &state->sn;
1552     pstrcpy(sn->name, sizeof(sn->name), name);
1553     qemu_gettimeofday(&tv);
1554     sn->date_sec = tv.tv_sec;
1555     sn->date_nsec = tv.tv_usec * 1000;
1556     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1557 
1558     ret1 = bdrv_snapshot_create(bs, sn);
1559     if (ret1 < 0) {
1560         error_setg_errno(errp, -ret1,
1561                          "Failed to create snapshot '%s' on device '%s'",
1562                          name, device);
1563         return;
1564     }
1565 
1566     /* 4. succeed, mark a snapshot is created */
1567     state->created = true;
1568 }
1569 
1570 static void internal_snapshot_abort(BlkActionState *common)
1571 {
1572     InternalSnapshotState *state =
1573                              DO_UPCAST(InternalSnapshotState, common, common);
1574     BlockDriverState *bs = state->bs;
1575     QEMUSnapshotInfo *sn = &state->sn;
1576     Error *local_error = NULL;
1577 
1578     if (!state->created) {
1579         return;
1580     }
1581 
1582     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1583         error_reportf_err(local_error,
1584                           "Failed to delete snapshot with id '%s' and "
1585                           "name '%s' on device '%s' in abort: ",
1586                           sn->id_str, sn->name,
1587                           bdrv_get_device_name(bs));
1588     }
1589 }
1590 
1591 static void internal_snapshot_clean(BlkActionState *common)
1592 {
1593     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1594                                              common, common);
1595 
1596     if (state->aio_context) {
1597         if (state->bs) {
1598             bdrv_drained_end(state->bs);
1599         }
1600         aio_context_release(state->aio_context);
1601     }
1602 }
1603 
1604 /* external snapshot private data */
1605 typedef struct ExternalSnapshotState {
1606     BlkActionState common;
1607     BlockDriverState *old_bs;
1608     BlockDriverState *new_bs;
1609     AioContext *aio_context;
1610     bool overlay_appended;
1611 } ExternalSnapshotState;
1612 
1613 static void external_snapshot_prepare(BlkActionState *common,
1614                                       Error **errp)
1615 {
1616     int flags = 0;
1617     QDict *options = NULL;
1618     Error *local_err = NULL;
1619     /* Device and node name of the image to generate the snapshot from */
1620     const char *device;
1621     const char *node_name;
1622     /* Reference to the new image (for 'blockdev-snapshot') */
1623     const char *snapshot_ref;
1624     /* File name of the new image (for 'blockdev-snapshot-sync') */
1625     const char *new_image_file;
1626     ExternalSnapshotState *state =
1627                              DO_UPCAST(ExternalSnapshotState, common, common);
1628     TransactionAction *action = common->action;
1629 
1630     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1631      * purpose but a different set of parameters */
1632     switch (action->type) {
1633     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1634         {
1635             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1636             device = s->node;
1637             node_name = s->node;
1638             new_image_file = NULL;
1639             snapshot_ref = s->overlay;
1640         }
1641         break;
1642     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1643         {
1644             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1645             device = s->has_device ? s->device : NULL;
1646             node_name = s->has_node_name ? s->node_name : NULL;
1647             new_image_file = s->snapshot_file;
1648             snapshot_ref = NULL;
1649         }
1650         break;
1651     default:
1652         g_assert_not_reached();
1653     }
1654 
1655     /* start processing */
1656     if (action_check_completion_mode(common, errp) < 0) {
1657         return;
1658     }
1659 
1660     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1661     if (!state->old_bs) {
1662         return;
1663     }
1664 
1665     /* Acquire AioContext now so any threads operating on old_bs stop */
1666     state->aio_context = bdrv_get_aio_context(state->old_bs);
1667     aio_context_acquire(state->aio_context);
1668     bdrv_drained_begin(state->old_bs);
1669 
1670     if (!bdrv_is_inserted(state->old_bs)) {
1671         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1672         return;
1673     }
1674 
1675     if (bdrv_op_is_blocked(state->old_bs,
1676                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1677         return;
1678     }
1679 
1680     if (!bdrv_is_read_only(state->old_bs)) {
1681         if (bdrv_flush(state->old_bs)) {
1682             error_setg(errp, QERR_IO_ERROR);
1683             return;
1684         }
1685     }
1686 
1687     if (!bdrv_is_first_non_filter(state->old_bs)) {
1688         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1689         return;
1690     }
1691 
1692     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1693         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1694         const char *format = s->has_format ? s->format : "qcow2";
1695         enum NewImageMode mode;
1696         const char *snapshot_node_name =
1697             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1698 
1699         if (node_name && !snapshot_node_name) {
1700             error_setg(errp, "New snapshot node name missing");
1701             return;
1702         }
1703 
1704         if (snapshot_node_name &&
1705             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1706             error_setg(errp, "New snapshot node name already in use");
1707             return;
1708         }
1709 
1710         flags = state->old_bs->open_flags;
1711         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1712         flags |= BDRV_O_NO_BACKING;
1713 
1714         /* create new image w/backing file */
1715         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1716         if (mode != NEW_IMAGE_MODE_EXISTING) {
1717             int64_t size = bdrv_getlength(state->old_bs);
1718             if (size < 0) {
1719                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1720                 return;
1721             }
1722             bdrv_img_create(new_image_file, format,
1723                             state->old_bs->filename,
1724                             state->old_bs->drv->format_name,
1725                             NULL, size, flags, false, &local_err);
1726             if (local_err) {
1727                 error_propagate(errp, local_err);
1728                 return;
1729             }
1730         }
1731 
1732         options = qdict_new();
1733         if (s->has_snapshot_node_name) {
1734             qdict_put_str(options, "node-name", snapshot_node_name);
1735         }
1736         qdict_put_str(options, "driver", format);
1737     }
1738 
1739     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1740                               errp);
1741     /* We will manually add the backing_hd field to the bs later */
1742     if (!state->new_bs) {
1743         return;
1744     }
1745 
1746     if (bdrv_has_blk(state->new_bs)) {
1747         error_setg(errp, "The snapshot is already in use");
1748         return;
1749     }
1750 
1751     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1752                            errp)) {
1753         return;
1754     }
1755 
1756     if (state->new_bs->backing != NULL) {
1757         error_setg(errp, "The snapshot already has a backing image");
1758         return;
1759     }
1760 
1761     if (!state->new_bs->drv->supports_backing) {
1762         error_setg(errp, "The snapshot does not support backing images");
1763         return;
1764     }
1765 
1766     bdrv_set_aio_context(state->new_bs, state->aio_context);
1767 
1768     /* This removes our old bs and adds the new bs. This is an operation that
1769      * can fail, so we need to do it in .prepare; undoing it for abort is
1770      * always possible. */
1771     bdrv_ref(state->new_bs);
1772     bdrv_append(state->new_bs, state->old_bs, &local_err);
1773     if (local_err) {
1774         error_propagate(errp, local_err);
1775         return;
1776     }
1777     state->overlay_appended = true;
1778 }
1779 
1780 static void external_snapshot_commit(BlkActionState *common)
1781 {
1782     ExternalSnapshotState *state =
1783                              DO_UPCAST(ExternalSnapshotState, common, common);
1784 
1785     /* We don't need (or want) to use the transactional
1786      * bdrv_reopen_multiple() across all the entries at once, because we
1787      * don't want to abort all of them if one of them fails the reopen */
1788     if (!atomic_read(&state->old_bs->copy_on_read)) {
1789         bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1790                     NULL);
1791     }
1792 }
1793 
1794 static void external_snapshot_abort(BlkActionState *common)
1795 {
1796     ExternalSnapshotState *state =
1797                              DO_UPCAST(ExternalSnapshotState, common, common);
1798     if (state->new_bs) {
1799         if (state->overlay_appended) {
1800             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1801                                           close state->old_bs; we need it */
1802             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1803             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1804             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1805         }
1806     }
1807 }
1808 
1809 static void external_snapshot_clean(BlkActionState *common)
1810 {
1811     ExternalSnapshotState *state =
1812                              DO_UPCAST(ExternalSnapshotState, common, common);
1813     if (state->aio_context) {
1814         bdrv_drained_end(state->old_bs);
1815         aio_context_release(state->aio_context);
1816         bdrv_unref(state->new_bs);
1817     }
1818 }
1819 
1820 typedef struct DriveBackupState {
1821     BlkActionState common;
1822     BlockDriverState *bs;
1823     AioContext *aio_context;
1824     BlockJob *job;
1825 } DriveBackupState;
1826 
1827 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
1828                             Error **errp);
1829 
1830 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1831 {
1832     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1833     BlockDriverState *bs;
1834     DriveBackup *backup;
1835     Error *local_err = NULL;
1836 
1837     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1838     backup = common->action->u.drive_backup.data;
1839 
1840     bs = qmp_get_root_bs(backup->device, errp);
1841     if (!bs) {
1842         return;
1843     }
1844 
1845     /* AioContext is released in .clean() */
1846     state->aio_context = bdrv_get_aio_context(bs);
1847     aio_context_acquire(state->aio_context);
1848     bdrv_drained_begin(bs);
1849     state->bs = bs;
1850 
1851     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1852     if (local_err) {
1853         error_propagate(errp, local_err);
1854         return;
1855     }
1856 }
1857 
1858 static void drive_backup_commit(BlkActionState *common)
1859 {
1860     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1861     assert(state->job);
1862     block_job_start(state->job);
1863 }
1864 
1865 static void drive_backup_abort(BlkActionState *common)
1866 {
1867     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1868 
1869     if (state->job) {
1870         block_job_cancel_sync(state->job);
1871     }
1872 }
1873 
1874 static void drive_backup_clean(BlkActionState *common)
1875 {
1876     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1877 
1878     if (state->aio_context) {
1879         bdrv_drained_end(state->bs);
1880         aio_context_release(state->aio_context);
1881     }
1882 }
1883 
1884 typedef struct BlockdevBackupState {
1885     BlkActionState common;
1886     BlockDriverState *bs;
1887     BlockJob *job;
1888     AioContext *aio_context;
1889 } BlockdevBackupState;
1890 
1891 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
1892                                     Error **errp);
1893 
1894 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1895 {
1896     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1897     BlockdevBackup *backup;
1898     BlockDriverState *bs, *target;
1899     Error *local_err = NULL;
1900 
1901     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1902     backup = common->action->u.blockdev_backup.data;
1903 
1904     bs = qmp_get_root_bs(backup->device, errp);
1905     if (!bs) {
1906         return;
1907     }
1908 
1909     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1910     if (!target) {
1911         return;
1912     }
1913 
1914     /* AioContext is released in .clean() */
1915     state->aio_context = bdrv_get_aio_context(bs);
1916     if (state->aio_context != bdrv_get_aio_context(target)) {
1917         state->aio_context = NULL;
1918         error_setg(errp, "Backup between two IO threads is not implemented");
1919         return;
1920     }
1921     aio_context_acquire(state->aio_context);
1922     state->bs = bs;
1923     bdrv_drained_begin(state->bs);
1924 
1925     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1926     if (local_err) {
1927         error_propagate(errp, local_err);
1928         return;
1929     }
1930 }
1931 
1932 static void blockdev_backup_commit(BlkActionState *common)
1933 {
1934     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1935     assert(state->job);
1936     block_job_start(state->job);
1937 }
1938 
1939 static void blockdev_backup_abort(BlkActionState *common)
1940 {
1941     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1942 
1943     if (state->job) {
1944         block_job_cancel_sync(state->job);
1945     }
1946 }
1947 
1948 static void blockdev_backup_clean(BlkActionState *common)
1949 {
1950     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1951 
1952     if (state->aio_context) {
1953         bdrv_drained_end(state->bs);
1954         aio_context_release(state->aio_context);
1955     }
1956 }
1957 
1958 typedef struct BlockDirtyBitmapState {
1959     BlkActionState common;
1960     BdrvDirtyBitmap *bitmap;
1961     BlockDriverState *bs;
1962     AioContext *aio_context;
1963     HBitmap *backup;
1964     bool prepared;
1965 } BlockDirtyBitmapState;
1966 
1967 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1968                                            Error **errp)
1969 {
1970     Error *local_err = NULL;
1971     BlockDirtyBitmapAdd *action;
1972     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1973                                              common, common);
1974 
1975     if (action_check_completion_mode(common, errp) < 0) {
1976         return;
1977     }
1978 
1979     action = common->action->u.block_dirty_bitmap_add.data;
1980     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1981     qmp_block_dirty_bitmap_add(action->node, action->name,
1982                                action->has_granularity, action->granularity,
1983                                action->has_persistent, action->persistent,
1984                                action->has_autoload, action->autoload,
1985                                &local_err);
1986 
1987     if (!local_err) {
1988         state->prepared = true;
1989     } else {
1990         error_propagate(errp, local_err);
1991     }
1992 }
1993 
1994 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1995 {
1996     BlockDirtyBitmapAdd *action;
1997     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1998                                              common, common);
1999 
2000     action = common->action->u.block_dirty_bitmap_add.data;
2001     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2002      * then the node reference and bitmap name must have been valid.
2003      */
2004     if (state->prepared) {
2005         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2006     }
2007 }
2008 
2009 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2010                                              Error **errp)
2011 {
2012     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2013                                              common, common);
2014     BlockDirtyBitmap *action;
2015 
2016     if (action_check_completion_mode(common, errp) < 0) {
2017         return;
2018     }
2019 
2020     action = common->action->u.block_dirty_bitmap_clear.data;
2021     state->bitmap = block_dirty_bitmap_lookup(action->node,
2022                                               action->name,
2023                                               &state->bs,
2024                                               errp);
2025     if (!state->bitmap) {
2026         return;
2027     }
2028 
2029     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2030         error_setg(errp, "Cannot modify a frozen bitmap");
2031         return;
2032     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2033         error_setg(errp, "Cannot clear a disabled bitmap");
2034         return;
2035     } else if (bdrv_dirty_bitmap_readonly(state->bitmap)) {
2036         error_setg(errp, "Cannot clear a readonly bitmap");
2037         return;
2038     }
2039 
2040     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2041     /* AioContext is released in .clean() */
2042 }
2043 
2044 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2045 {
2046     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2047                                              common, common);
2048 
2049     if (state->backup) {
2050         bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2051     }
2052 }
2053 
2054 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2055 {
2056     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2057                                              common, common);
2058 
2059     hbitmap_free(state->backup);
2060 }
2061 
2062 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2063 {
2064     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2065                                              common, common);
2066 
2067     if (state->aio_context) {
2068         aio_context_release(state->aio_context);
2069     }
2070 }
2071 
2072 static void abort_prepare(BlkActionState *common, Error **errp)
2073 {
2074     error_setg(errp, "Transaction aborted using Abort action");
2075 }
2076 
2077 static void abort_commit(BlkActionState *common)
2078 {
2079     g_assert_not_reached(); /* this action never succeeds */
2080 }
2081 
2082 static const BlkActionOps actions[] = {
2083     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2084         .instance_size = sizeof(ExternalSnapshotState),
2085         .prepare  = external_snapshot_prepare,
2086         .commit   = external_snapshot_commit,
2087         .abort = external_snapshot_abort,
2088         .clean = external_snapshot_clean,
2089     },
2090     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2091         .instance_size = sizeof(ExternalSnapshotState),
2092         .prepare  = external_snapshot_prepare,
2093         .commit   = external_snapshot_commit,
2094         .abort = external_snapshot_abort,
2095         .clean = external_snapshot_clean,
2096     },
2097     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2098         .instance_size = sizeof(DriveBackupState),
2099         .prepare = drive_backup_prepare,
2100         .commit = drive_backup_commit,
2101         .abort = drive_backup_abort,
2102         .clean = drive_backup_clean,
2103     },
2104     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2105         .instance_size = sizeof(BlockdevBackupState),
2106         .prepare = blockdev_backup_prepare,
2107         .commit = blockdev_backup_commit,
2108         .abort = blockdev_backup_abort,
2109         .clean = blockdev_backup_clean,
2110     },
2111     [TRANSACTION_ACTION_KIND_ABORT] = {
2112         .instance_size = sizeof(BlkActionState),
2113         .prepare = abort_prepare,
2114         .commit = abort_commit,
2115     },
2116     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2117         .instance_size = sizeof(InternalSnapshotState),
2118         .prepare  = internal_snapshot_prepare,
2119         .abort = internal_snapshot_abort,
2120         .clean = internal_snapshot_clean,
2121     },
2122     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2123         .instance_size = sizeof(BlockDirtyBitmapState),
2124         .prepare = block_dirty_bitmap_add_prepare,
2125         .abort = block_dirty_bitmap_add_abort,
2126     },
2127     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2128         .instance_size = sizeof(BlockDirtyBitmapState),
2129         .prepare = block_dirty_bitmap_clear_prepare,
2130         .commit = block_dirty_bitmap_clear_commit,
2131         .abort = block_dirty_bitmap_clear_abort,
2132         .clean = block_dirty_bitmap_clear_clean,
2133     }
2134 };
2135 
2136 /**
2137  * Allocate a TransactionProperties structure if necessary, and fill
2138  * that structure with desired defaults if they are unset.
2139  */
2140 static TransactionProperties *get_transaction_properties(
2141     TransactionProperties *props)
2142 {
2143     if (!props) {
2144         props = g_new0(TransactionProperties, 1);
2145     }
2146 
2147     if (!props->has_completion_mode) {
2148         props->has_completion_mode = true;
2149         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2150     }
2151 
2152     return props;
2153 }
2154 
2155 /*
2156  * 'Atomic' group operations.  The operations are performed as a set, and if
2157  * any fail then we roll back all operations in the group.
2158  */
2159 void qmp_transaction(TransactionActionList *dev_list,
2160                      bool has_props,
2161                      struct TransactionProperties *props,
2162                      Error **errp)
2163 {
2164     TransactionActionList *dev_entry = dev_list;
2165     BlockJobTxn *block_job_txn = NULL;
2166     BlkActionState *state, *next;
2167     Error *local_err = NULL;
2168 
2169     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2170     QSIMPLEQ_INIT(&snap_bdrv_states);
2171 
2172     /* Does this transaction get canceled as a group on failure?
2173      * If not, we don't really need to make a BlockJobTxn.
2174      */
2175     props = get_transaction_properties(props);
2176     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2177         block_job_txn = block_job_txn_new();
2178     }
2179 
2180     /* drain all i/o before any operations */
2181     bdrv_drain_all();
2182 
2183     /* We don't do anything in this loop that commits us to the operations */
2184     while (NULL != dev_entry) {
2185         TransactionAction *dev_info = NULL;
2186         const BlkActionOps *ops;
2187 
2188         dev_info = dev_entry->value;
2189         dev_entry = dev_entry->next;
2190 
2191         assert(dev_info->type < ARRAY_SIZE(actions));
2192 
2193         ops = &actions[dev_info->type];
2194         assert(ops->instance_size > 0);
2195 
2196         state = g_malloc0(ops->instance_size);
2197         state->ops = ops;
2198         state->action = dev_info;
2199         state->block_job_txn = block_job_txn;
2200         state->txn_props = props;
2201         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2202 
2203         state->ops->prepare(state, &local_err);
2204         if (local_err) {
2205             error_propagate(errp, local_err);
2206             goto delete_and_fail;
2207         }
2208     }
2209 
2210     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2211         if (state->ops->commit) {
2212             state->ops->commit(state);
2213         }
2214     }
2215 
2216     /* success */
2217     goto exit;
2218 
2219 delete_and_fail:
2220     /* failure, and it is all-or-none; roll back all operations */
2221     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2222         if (state->ops->abort) {
2223             state->ops->abort(state);
2224         }
2225     }
2226 exit:
2227     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2228         if (state->ops->clean) {
2229             state->ops->clean(state);
2230         }
2231         g_free(state);
2232     }
2233     if (!has_props) {
2234         qapi_free_TransactionProperties(props);
2235     }
2236     block_job_txn_unref(block_job_txn);
2237 }
2238 
2239 void qmp_eject(bool has_device, const char *device,
2240                bool has_id, const char *id,
2241                bool has_force, bool force, Error **errp)
2242 {
2243     Error *local_err = NULL;
2244     int rc;
2245 
2246     if (!has_force) {
2247         force = false;
2248     }
2249 
2250     rc = do_open_tray(has_device ? device : NULL,
2251                       has_id ? id : NULL,
2252                       force, &local_err);
2253     if (rc && rc != -ENOSYS) {
2254         error_propagate(errp, local_err);
2255         return;
2256     }
2257     error_free(local_err);
2258 
2259     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, errp);
2260 }
2261 
2262 void qmp_block_passwd(bool has_device, const char *device,
2263                       bool has_node_name, const char *node_name,
2264                       const char *password, Error **errp)
2265 {
2266     error_setg(errp,
2267                "Setting block passwords directly is no longer supported");
2268 }
2269 
2270 /*
2271  * Attempt to open the tray of @device.
2272  * If @force, ignore its tray lock.
2273  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2274  * On error, store an error through @errp and return -errno.
2275  * If @device does not exist, return -ENODEV.
2276  * If it has no removable media, return -ENOTSUP.
2277  * If it has no tray, return -ENOSYS.
2278  * If the guest was asked to open the tray, return -EINPROGRESS.
2279  * Else, return 0.
2280  */
2281 static int do_open_tray(const char *blk_name, const char *qdev_id,
2282                         bool force, Error **errp)
2283 {
2284     BlockBackend *blk;
2285     const char *device = qdev_id ?: blk_name;
2286     bool locked;
2287 
2288     blk = qmp_get_blk(blk_name, qdev_id, errp);
2289     if (!blk) {
2290         return -ENODEV;
2291     }
2292 
2293     if (!blk_dev_has_removable_media(blk)) {
2294         error_setg(errp, "Device '%s' is not removable", device);
2295         return -ENOTSUP;
2296     }
2297 
2298     if (!blk_dev_has_tray(blk)) {
2299         error_setg(errp, "Device '%s' does not have a tray", device);
2300         return -ENOSYS;
2301     }
2302 
2303     if (blk_dev_is_tray_open(blk)) {
2304         return 0;
2305     }
2306 
2307     locked = blk_dev_is_medium_locked(blk);
2308     if (locked) {
2309         blk_dev_eject_request(blk, force);
2310     }
2311 
2312     if (!locked || force) {
2313         blk_dev_change_media_cb(blk, false, &error_abort);
2314     }
2315 
2316     if (locked && !force) {
2317         error_setg(errp, "Device '%s' is locked and force was not specified, "
2318                    "wait for tray to open and try again", device);
2319         return -EINPROGRESS;
2320     }
2321 
2322     return 0;
2323 }
2324 
2325 void qmp_blockdev_open_tray(bool has_device, const char *device,
2326                             bool has_id, const char *id,
2327                             bool has_force, bool force,
2328                             Error **errp)
2329 {
2330     Error *local_err = NULL;
2331     int rc;
2332 
2333     if (!has_force) {
2334         force = false;
2335     }
2336     rc = do_open_tray(has_device ? device : NULL,
2337                       has_id ? id : NULL,
2338                       force, &local_err);
2339     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2340         error_propagate(errp, local_err);
2341         return;
2342     }
2343     error_free(local_err);
2344 }
2345 
2346 void qmp_blockdev_close_tray(bool has_device, const char *device,
2347                              bool has_id, const char *id,
2348                              Error **errp)
2349 {
2350     BlockBackend *blk;
2351     Error *local_err = NULL;
2352 
2353     device = has_device ? device : NULL;
2354     id = has_id ? id : NULL;
2355 
2356     blk = qmp_get_blk(device, id, errp);
2357     if (!blk) {
2358         return;
2359     }
2360 
2361     if (!blk_dev_has_removable_media(blk)) {
2362         error_setg(errp, "Device '%s' is not removable", device ?: id);
2363         return;
2364     }
2365 
2366     if (!blk_dev_has_tray(blk)) {
2367         /* Ignore this command on tray-less devices */
2368         return;
2369     }
2370 
2371     if (!blk_dev_is_tray_open(blk)) {
2372         return;
2373     }
2374 
2375     blk_dev_change_media_cb(blk, true, &local_err);
2376     if (local_err) {
2377         error_propagate(errp, local_err);
2378         return;
2379     }
2380 }
2381 
2382 void qmp_x_blockdev_remove_medium(bool has_device, const char *device,
2383                                   bool has_id, const char *id, Error **errp)
2384 {
2385     BlockBackend *blk;
2386     BlockDriverState *bs;
2387     AioContext *aio_context;
2388     bool has_attached_device;
2389 
2390     device = has_device ? device : NULL;
2391     id = has_id ? id : NULL;
2392 
2393     blk = qmp_get_blk(device, id, errp);
2394     if (!blk) {
2395         return;
2396     }
2397 
2398     /* For BBs without a device, we can exchange the BDS tree at will */
2399     has_attached_device = blk_get_attached_dev(blk);
2400 
2401     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2402         error_setg(errp, "Device '%s' is not removable", device ?: id);
2403         return;
2404     }
2405 
2406     if (has_attached_device && blk_dev_has_tray(blk) &&
2407         !blk_dev_is_tray_open(blk))
2408     {
2409         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2410         return;
2411     }
2412 
2413     bs = blk_bs(blk);
2414     if (!bs) {
2415         return;
2416     }
2417 
2418     aio_context = bdrv_get_aio_context(bs);
2419     aio_context_acquire(aio_context);
2420 
2421     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2422         goto out;
2423     }
2424 
2425     blk_remove_bs(blk);
2426 
2427     if (!blk_dev_has_tray(blk)) {
2428         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2429          * called at all); therefore, the medium needs to be ejected here.
2430          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2431          * value passed here (i.e. false). */
2432         blk_dev_change_media_cb(blk, false, &error_abort);
2433     }
2434 
2435 out:
2436     aio_context_release(aio_context);
2437 }
2438 
2439 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2440                                             BlockDriverState *bs, Error **errp)
2441 {
2442     Error *local_err = NULL;
2443     bool has_device;
2444     int ret;
2445 
2446     /* For BBs without a device, we can exchange the BDS tree at will */
2447     has_device = blk_get_attached_dev(blk);
2448 
2449     if (has_device && !blk_dev_has_removable_media(blk)) {
2450         error_setg(errp, "Device is not removable");
2451         return;
2452     }
2453 
2454     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2455         error_setg(errp, "Tray of the device is not open");
2456         return;
2457     }
2458 
2459     if (blk_bs(blk)) {
2460         error_setg(errp, "There already is a medium in the device");
2461         return;
2462     }
2463 
2464     ret = blk_insert_bs(blk, bs, errp);
2465     if (ret < 0) {
2466         return;
2467     }
2468 
2469     if (!blk_dev_has_tray(blk)) {
2470         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2471          * called at all); therefore, the medium needs to be pushed into the
2472          * slot here.
2473          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2474          * value passed here (i.e. true). */
2475         blk_dev_change_media_cb(blk, true, &local_err);
2476         if (local_err) {
2477             error_propagate(errp, local_err);
2478             blk_remove_bs(blk);
2479             return;
2480         }
2481     }
2482 }
2483 
2484 void qmp_x_blockdev_insert_medium(bool has_device, const char *device,
2485                                   bool has_id, const char *id,
2486                                   const char *node_name, Error **errp)
2487 {
2488     BlockBackend *blk;
2489     BlockDriverState *bs;
2490 
2491     blk = qmp_get_blk(has_device ? device : NULL,
2492                       has_id ? id : NULL,
2493                       errp);
2494     if (!blk) {
2495         return;
2496     }
2497 
2498     bs = bdrv_find_node(node_name);
2499     if (!bs) {
2500         error_setg(errp, "Node '%s' not found", node_name);
2501         return;
2502     }
2503 
2504     if (bdrv_has_blk(bs)) {
2505         error_setg(errp, "Node '%s' is already in use", node_name);
2506         return;
2507     }
2508 
2509     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2510 }
2511 
2512 void qmp_blockdev_change_medium(bool has_device, const char *device,
2513                                 bool has_id, const char *id,
2514                                 const char *filename,
2515                                 bool has_format, const char *format,
2516                                 bool has_read_only,
2517                                 BlockdevChangeReadOnlyMode read_only,
2518                                 Error **errp)
2519 {
2520     BlockBackend *blk;
2521     BlockDriverState *medium_bs = NULL;
2522     int bdrv_flags;
2523     bool detect_zeroes;
2524     int rc;
2525     QDict *options = NULL;
2526     Error *err = NULL;
2527 
2528     blk = qmp_get_blk(has_device ? device : NULL,
2529                       has_id ? id : NULL,
2530                       errp);
2531     if (!blk) {
2532         goto fail;
2533     }
2534 
2535     if (blk_bs(blk)) {
2536         blk_update_root_state(blk);
2537     }
2538 
2539     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2540     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2541         BDRV_O_PROTOCOL);
2542 
2543     if (!has_read_only) {
2544         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2545     }
2546 
2547     switch (read_only) {
2548     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2549         break;
2550 
2551     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2552         bdrv_flags &= ~BDRV_O_RDWR;
2553         break;
2554 
2555     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2556         bdrv_flags |= BDRV_O_RDWR;
2557         break;
2558 
2559     default:
2560         abort();
2561     }
2562 
2563     options = qdict_new();
2564     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2565     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2566 
2567     if (has_format) {
2568         qdict_put_str(options, "driver", format);
2569     }
2570 
2571     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2572     if (!medium_bs) {
2573         goto fail;
2574     }
2575 
2576     rc = do_open_tray(has_device ? device : NULL,
2577                       has_id ? id : NULL,
2578                       false, &err);
2579     if (rc && rc != -ENOSYS) {
2580         error_propagate(errp, err);
2581         goto fail;
2582     }
2583     error_free(err);
2584     err = NULL;
2585 
2586     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, &err);
2587     if (err) {
2588         error_propagate(errp, err);
2589         goto fail;
2590     }
2591 
2592     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2593     if (err) {
2594         error_propagate(errp, err);
2595         goto fail;
2596     }
2597 
2598     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2599 
2600 fail:
2601     /* If the medium has been inserted, the device has its own reference, so
2602      * ours must be relinquished; and if it has not been inserted successfully,
2603      * the reference must be relinquished anyway */
2604     bdrv_unref(medium_bs);
2605 }
2606 
2607 /* throttling disk I/O limits */
2608 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2609 {
2610     ThrottleConfig cfg;
2611     BlockDriverState *bs;
2612     BlockBackend *blk;
2613     AioContext *aio_context;
2614 
2615     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2616                       arg->has_id ? arg->id : NULL,
2617                       errp);
2618     if (!blk) {
2619         return;
2620     }
2621 
2622     aio_context = blk_get_aio_context(blk);
2623     aio_context_acquire(aio_context);
2624 
2625     bs = blk_bs(blk);
2626     if (!bs) {
2627         error_setg(errp, "Device has no medium");
2628         goto out;
2629     }
2630 
2631     throttle_config_init(&cfg);
2632     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2633     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2634     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2635 
2636     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2637     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2638     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2639 
2640     if (arg->has_bps_max) {
2641         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2642     }
2643     if (arg->has_bps_rd_max) {
2644         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2645     }
2646     if (arg->has_bps_wr_max) {
2647         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2648     }
2649     if (arg->has_iops_max) {
2650         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2651     }
2652     if (arg->has_iops_rd_max) {
2653         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2654     }
2655     if (arg->has_iops_wr_max) {
2656         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2657     }
2658 
2659     if (arg->has_bps_max_length) {
2660         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2661     }
2662     if (arg->has_bps_rd_max_length) {
2663         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2664     }
2665     if (arg->has_bps_wr_max_length) {
2666         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2667     }
2668     if (arg->has_iops_max_length) {
2669         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2670     }
2671     if (arg->has_iops_rd_max_length) {
2672         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2673     }
2674     if (arg->has_iops_wr_max_length) {
2675         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2676     }
2677 
2678     if (arg->has_iops_size) {
2679         cfg.op_size = arg->iops_size;
2680     }
2681 
2682     if (!throttle_is_valid(&cfg, errp)) {
2683         goto out;
2684     }
2685 
2686     if (throttle_enabled(&cfg)) {
2687         /* Enable I/O limits if they're not enabled yet, otherwise
2688          * just update the throttling group. */
2689         if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2690             blk_io_limits_enable(blk,
2691                                  arg->has_group ? arg->group :
2692                                  arg->has_device ? arg->device :
2693                                  arg->id);
2694         } else if (arg->has_group) {
2695             blk_io_limits_update_group(blk, arg->group);
2696         }
2697         /* Set the new throttling configuration */
2698         blk_set_io_limits(blk, &cfg);
2699     } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2700         /* If all throttling settings are set to 0, disable I/O limits */
2701         blk_io_limits_disable(blk);
2702     }
2703 
2704 out:
2705     aio_context_release(aio_context);
2706 }
2707 
2708 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2709                                 bool has_granularity, uint32_t granularity,
2710                                 bool has_persistent, bool persistent,
2711                                 bool has_autoload, bool autoload,
2712                                 Error **errp)
2713 {
2714     BlockDriverState *bs;
2715     BdrvDirtyBitmap *bitmap;
2716 
2717     if (!name || name[0] == '\0') {
2718         error_setg(errp, "Bitmap name cannot be empty");
2719         return;
2720     }
2721 
2722     bs = bdrv_lookup_bs(node, node, errp);
2723     if (!bs) {
2724         return;
2725     }
2726 
2727     if (has_granularity) {
2728         if (granularity < 512 || !is_power_of_2(granularity)) {
2729             error_setg(errp, "Granularity must be power of 2 "
2730                              "and at least 512");
2731             return;
2732         }
2733     } else {
2734         /* Default to cluster size, if available: */
2735         granularity = bdrv_get_default_bitmap_granularity(bs);
2736     }
2737 
2738     if (!has_persistent) {
2739         persistent = false;
2740     }
2741     if (!has_autoload) {
2742         autoload = false;
2743     }
2744 
2745     if (has_autoload && !persistent) {
2746         error_setg(errp, "Autoload flag must be used only for persistent "
2747                          "bitmaps");
2748         return;
2749     }
2750 
2751     if (persistent &&
2752         !bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp))
2753     {
2754         return;
2755     }
2756 
2757     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2758     if (bitmap == NULL) {
2759         return;
2760     }
2761 
2762     bdrv_dirty_bitmap_set_persistance(bitmap, persistent);
2763     bdrv_dirty_bitmap_set_autoload(bitmap, autoload);
2764 }
2765 
2766 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2767                                    Error **errp)
2768 {
2769     BlockDriverState *bs;
2770     BdrvDirtyBitmap *bitmap;
2771     Error *local_err = NULL;
2772 
2773     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2774     if (!bitmap || !bs) {
2775         return;
2776     }
2777 
2778     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2779         error_setg(errp,
2780                    "Bitmap '%s' is currently frozen and cannot be removed",
2781                    name);
2782         return;
2783     }
2784 
2785     if (bdrv_dirty_bitmap_get_persistance(bitmap)) {
2786         bdrv_remove_persistent_dirty_bitmap(bs, name, &local_err);
2787         if (local_err != NULL) {
2788             error_propagate(errp, local_err);
2789             return;
2790         }
2791     }
2792 
2793     bdrv_dirty_bitmap_make_anon(bitmap);
2794     bdrv_release_dirty_bitmap(bs, bitmap);
2795 }
2796 
2797 /**
2798  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2799  * immediately after a full backup operation.
2800  */
2801 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2802                                   Error **errp)
2803 {
2804     BdrvDirtyBitmap *bitmap;
2805     BlockDriverState *bs;
2806 
2807     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2808     if (!bitmap || !bs) {
2809         return;
2810     }
2811 
2812     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2813         error_setg(errp,
2814                    "Bitmap '%s' is currently frozen and cannot be modified",
2815                    name);
2816         return;
2817     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2818         error_setg(errp,
2819                    "Bitmap '%s' is currently disabled and cannot be cleared",
2820                    name);
2821         return;
2822     } else if (bdrv_dirty_bitmap_readonly(bitmap)) {
2823         error_setg(errp, "Bitmap '%s' is readonly and cannot be cleared", name);
2824         return;
2825     }
2826 
2827     bdrv_clear_dirty_bitmap(bitmap, NULL);
2828 }
2829 
2830 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2831                                                               const char *name,
2832                                                               Error **errp)
2833 {
2834     BdrvDirtyBitmap *bitmap;
2835     BlockDriverState *bs;
2836     BlockDirtyBitmapSha256 *ret = NULL;
2837     char *sha256;
2838 
2839     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2840     if (!bitmap || !bs) {
2841         return NULL;
2842     }
2843 
2844     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2845     if (sha256 == NULL) {
2846         return NULL;
2847     }
2848 
2849     ret = g_new(BlockDirtyBitmapSha256, 1);
2850     ret->sha256 = sha256;
2851 
2852     return ret;
2853 }
2854 
2855 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2856 {
2857     const char *id = qdict_get_str(qdict, "id");
2858     BlockBackend *blk;
2859     BlockDriverState *bs;
2860     AioContext *aio_context;
2861     Error *local_err = NULL;
2862 
2863     bs = bdrv_find_node(id);
2864     if (bs) {
2865         qmp_blockdev_del(id, &local_err);
2866         if (local_err) {
2867             error_report_err(local_err);
2868         }
2869         return;
2870     }
2871 
2872     blk = blk_by_name(id);
2873     if (!blk) {
2874         error_report("Device '%s' not found", id);
2875         return;
2876     }
2877 
2878     if (!blk_legacy_dinfo(blk)) {
2879         error_report("Deleting device added with blockdev-add"
2880                      " is not supported");
2881         return;
2882     }
2883 
2884     aio_context = blk_get_aio_context(blk);
2885     aio_context_acquire(aio_context);
2886 
2887     bs = blk_bs(blk);
2888     if (bs) {
2889         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2890             error_report_err(local_err);
2891             aio_context_release(aio_context);
2892             return;
2893         }
2894 
2895         blk_remove_bs(blk);
2896     }
2897 
2898     /* Make the BlockBackend and the attached BlockDriverState anonymous */
2899     monitor_remove_blk(blk);
2900 
2901     /* If this BlockBackend has a device attached to it, its refcount will be
2902      * decremented when the device is removed; otherwise we have to do so here.
2903      */
2904     if (blk_get_attached_dev(blk)) {
2905         /* Further I/O must not pause the guest */
2906         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2907                          BLOCKDEV_ON_ERROR_REPORT);
2908     } else {
2909         blk_unref(blk);
2910     }
2911 
2912     aio_context_release(aio_context);
2913 }
2914 
2915 void qmp_block_resize(bool has_device, const char *device,
2916                       bool has_node_name, const char *node_name,
2917                       int64_t size, Error **errp)
2918 {
2919     Error *local_err = NULL;
2920     BlockBackend *blk = NULL;
2921     BlockDriverState *bs;
2922     AioContext *aio_context;
2923     int ret;
2924 
2925     bs = bdrv_lookup_bs(has_device ? device : NULL,
2926                         has_node_name ? node_name : NULL,
2927                         &local_err);
2928     if (local_err) {
2929         error_propagate(errp, local_err);
2930         return;
2931     }
2932 
2933     aio_context = bdrv_get_aio_context(bs);
2934     aio_context_acquire(aio_context);
2935 
2936     if (!bdrv_is_first_non_filter(bs)) {
2937         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2938         goto out;
2939     }
2940 
2941     if (size < 0) {
2942         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2943         goto out;
2944     }
2945 
2946     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2947         error_setg(errp, QERR_DEVICE_IN_USE, device);
2948         goto out;
2949     }
2950 
2951     blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
2952     ret = blk_insert_bs(blk, bs, errp);
2953     if (ret < 0) {
2954         goto out;
2955     }
2956 
2957     bdrv_drained_begin(bs);
2958     ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
2959     bdrv_drained_end(bs);
2960 
2961 out:
2962     blk_unref(blk);
2963     aio_context_release(aio_context);
2964 }
2965 
2966 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2967                       bool has_base, const char *base,
2968                       bool has_base_node, const char *base_node,
2969                       bool has_backing_file, const char *backing_file,
2970                       bool has_speed, int64_t speed,
2971                       bool has_on_error, BlockdevOnError on_error,
2972                       Error **errp)
2973 {
2974     BlockDriverState *bs, *iter;
2975     BlockDriverState *base_bs = NULL;
2976     AioContext *aio_context;
2977     Error *local_err = NULL;
2978     const char *base_name = NULL;
2979 
2980     if (!has_on_error) {
2981         on_error = BLOCKDEV_ON_ERROR_REPORT;
2982     }
2983 
2984     bs = bdrv_lookup_bs(device, device, errp);
2985     if (!bs) {
2986         return;
2987     }
2988 
2989     aio_context = bdrv_get_aio_context(bs);
2990     aio_context_acquire(aio_context);
2991 
2992     if (has_base && has_base_node) {
2993         error_setg(errp, "'base' and 'base-node' cannot be specified "
2994                    "at the same time");
2995         goto out;
2996     }
2997 
2998     if (has_base) {
2999         base_bs = bdrv_find_backing_image(bs, base);
3000         if (base_bs == NULL) {
3001             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3002             goto out;
3003         }
3004         assert(bdrv_get_aio_context(base_bs) == aio_context);
3005         base_name = base;
3006     }
3007 
3008     if (has_base_node) {
3009         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3010         if (!base_bs) {
3011             goto out;
3012         }
3013         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3014             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3015                        base_node, device);
3016             goto out;
3017         }
3018         assert(bdrv_get_aio_context(base_bs) == aio_context);
3019         base_name = base_bs->filename;
3020     }
3021 
3022     /* Check for op blockers in the whole chain between bs and base */
3023     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3024         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3025             goto out;
3026         }
3027     }
3028 
3029     /* if we are streaming the entire chain, the result will have no backing
3030      * file, and specifying one is therefore an error */
3031     if (base_bs == NULL && has_backing_file) {
3032         error_setg(errp, "backing file specified, but streaming the "
3033                          "entire chain");
3034         goto out;
3035     }
3036 
3037     /* backing_file string overrides base bs filename */
3038     base_name = has_backing_file ? backing_file : base_name;
3039 
3040     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3041                  has_speed ? speed : 0, on_error, &local_err);
3042     if (local_err) {
3043         error_propagate(errp, local_err);
3044         goto out;
3045     }
3046 
3047     trace_qmp_block_stream(bs, bs->job);
3048 
3049 out:
3050     aio_context_release(aio_context);
3051 }
3052 
3053 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3054                       bool has_base, const char *base,
3055                       bool has_top, const char *top,
3056                       bool has_backing_file, const char *backing_file,
3057                       bool has_speed, int64_t speed,
3058                       bool has_filter_node_name, const char *filter_node_name,
3059                       Error **errp)
3060 {
3061     BlockDriverState *bs;
3062     BlockDriverState *iter;
3063     BlockDriverState *base_bs, *top_bs;
3064     AioContext *aio_context;
3065     Error *local_err = NULL;
3066     /* This will be part of the QMP command, if/when the
3067      * BlockdevOnError change for blkmirror makes it in
3068      */
3069     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3070 
3071     if (!has_speed) {
3072         speed = 0;
3073     }
3074     if (!has_filter_node_name) {
3075         filter_node_name = NULL;
3076     }
3077 
3078     /* Important Note:
3079      *  libvirt relies on the DeviceNotFound error class in order to probe for
3080      *  live commit feature versions; for this to work, we must make sure to
3081      *  perform the device lookup before any generic errors that may occur in a
3082      *  scenario in which all optional arguments are omitted. */
3083     bs = qmp_get_root_bs(device, &local_err);
3084     if (!bs) {
3085         bs = bdrv_lookup_bs(device, device, NULL);
3086         if (!bs) {
3087             error_free(local_err);
3088             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3089                       "Device '%s' not found", device);
3090         } else {
3091             error_propagate(errp, local_err);
3092         }
3093         return;
3094     }
3095 
3096     aio_context = bdrv_get_aio_context(bs);
3097     aio_context_acquire(aio_context);
3098 
3099     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3100         goto out;
3101     }
3102 
3103     /* default top_bs is the active layer */
3104     top_bs = bs;
3105 
3106     if (has_top && top) {
3107         if (strcmp(bs->filename, top) != 0) {
3108             top_bs = bdrv_find_backing_image(bs, top);
3109         }
3110     }
3111 
3112     if (top_bs == NULL) {
3113         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3114         goto out;
3115     }
3116 
3117     assert(bdrv_get_aio_context(top_bs) == aio_context);
3118 
3119     if (has_base && base) {
3120         base_bs = bdrv_find_backing_image(top_bs, base);
3121     } else {
3122         base_bs = bdrv_find_base(top_bs);
3123     }
3124 
3125     if (base_bs == NULL) {
3126         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3127         goto out;
3128     }
3129 
3130     assert(bdrv_get_aio_context(base_bs) == aio_context);
3131 
3132     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3133         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3134             goto out;
3135         }
3136     }
3137 
3138     /* Do not allow attempts to commit an image into itself */
3139     if (top_bs == base_bs) {
3140         error_setg(errp, "cannot commit an image into itself");
3141         goto out;
3142     }
3143 
3144     if (top_bs == bs) {
3145         if (has_backing_file) {
3146             error_setg(errp, "'backing-file' specified,"
3147                              " but 'top' is the active layer");
3148             goto out;
3149         }
3150         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3151                             BLOCK_JOB_DEFAULT, speed, on_error,
3152                             filter_node_name, NULL, NULL, false, &local_err);
3153     } else {
3154         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3155         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3156             goto out;
3157         }
3158         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, speed,
3159                      on_error, has_backing_file ? backing_file : NULL,
3160                      filter_node_name, &local_err);
3161     }
3162     if (local_err != NULL) {
3163         error_propagate(errp, local_err);
3164         goto out;
3165     }
3166 
3167 out:
3168     aio_context_release(aio_context);
3169 }
3170 
3171 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
3172                                  Error **errp)
3173 {
3174     BlockDriverState *bs;
3175     BlockDriverState *target_bs;
3176     BlockDriverState *source = NULL;
3177     BlockJob *job = NULL;
3178     BdrvDirtyBitmap *bmap = NULL;
3179     AioContext *aio_context;
3180     QDict *options = NULL;
3181     Error *local_err = NULL;
3182     int flags;
3183     int64_t size;
3184     bool set_backing_hd = false;
3185 
3186     if (!backup->has_speed) {
3187         backup->speed = 0;
3188     }
3189     if (!backup->has_on_source_error) {
3190         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3191     }
3192     if (!backup->has_on_target_error) {
3193         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3194     }
3195     if (!backup->has_mode) {
3196         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3197     }
3198     if (!backup->has_job_id) {
3199         backup->job_id = NULL;
3200     }
3201     if (!backup->has_compress) {
3202         backup->compress = false;
3203     }
3204 
3205     bs = qmp_get_root_bs(backup->device, errp);
3206     if (!bs) {
3207         return NULL;
3208     }
3209 
3210     aio_context = bdrv_get_aio_context(bs);
3211     aio_context_acquire(aio_context);
3212 
3213     if (!backup->has_format) {
3214         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3215                          NULL : (char*) bs->drv->format_name;
3216     }
3217 
3218     /* Early check to avoid creating target */
3219     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3220         goto out;
3221     }
3222 
3223     flags = bs->open_flags | BDRV_O_RDWR;
3224 
3225     /* See if we have a backing HD we can use to create our new image
3226      * on top of. */
3227     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3228         source = backing_bs(bs);
3229         if (!source) {
3230             backup->sync = MIRROR_SYNC_MODE_FULL;
3231         }
3232     }
3233     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3234         source = bs;
3235         flags |= BDRV_O_NO_BACKING;
3236         set_backing_hd = true;
3237     }
3238 
3239     size = bdrv_getlength(bs);
3240     if (size < 0) {
3241         error_setg_errno(errp, -size, "bdrv_getlength failed");
3242         goto out;
3243     }
3244 
3245     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3246         assert(backup->format);
3247         if (source) {
3248             bdrv_img_create(backup->target, backup->format, source->filename,
3249                             source->drv->format_name, NULL,
3250                             size, flags, false, &local_err);
3251         } else {
3252             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3253                             size, flags, false, &local_err);
3254         }
3255     }
3256 
3257     if (local_err) {
3258         error_propagate(errp, local_err);
3259         goto out;
3260     }
3261 
3262     if (backup->format) {
3263         if (!options) {
3264             options = qdict_new();
3265         }
3266         qdict_put_str(options, "driver", backup->format);
3267     }
3268 
3269     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3270     if (!target_bs) {
3271         goto out;
3272     }
3273 
3274     bdrv_set_aio_context(target_bs, aio_context);
3275 
3276     if (set_backing_hd) {
3277         bdrv_set_backing_hd(target_bs, source, &local_err);
3278         if (local_err) {
3279             bdrv_unref(target_bs);
3280             goto out;
3281         }
3282     }
3283 
3284     if (backup->has_bitmap) {
3285         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3286         if (!bmap) {
3287             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3288             bdrv_unref(target_bs);
3289             goto out;
3290         }
3291     }
3292 
3293     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3294                             backup->sync, bmap, backup->compress,
3295                             backup->on_source_error, backup->on_target_error,
3296                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3297     bdrv_unref(target_bs);
3298     if (local_err != NULL) {
3299         error_propagate(errp, local_err);
3300         goto out;
3301     }
3302 
3303 out:
3304     aio_context_release(aio_context);
3305     return job;
3306 }
3307 
3308 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3309 {
3310 
3311     BlockJob *job;
3312     job = do_drive_backup(arg, NULL, errp);
3313     if (job) {
3314         block_job_start(job);
3315     }
3316 }
3317 
3318 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3319 {
3320     return bdrv_named_nodes_list(errp);
3321 }
3322 
3323 BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
3324                              Error **errp)
3325 {
3326     BlockDriverState *bs;
3327     BlockDriverState *target_bs;
3328     Error *local_err = NULL;
3329     AioContext *aio_context;
3330     BlockJob *job = NULL;
3331 
3332     if (!backup->has_speed) {
3333         backup->speed = 0;
3334     }
3335     if (!backup->has_on_source_error) {
3336         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3337     }
3338     if (!backup->has_on_target_error) {
3339         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3340     }
3341     if (!backup->has_job_id) {
3342         backup->job_id = NULL;
3343     }
3344     if (!backup->has_compress) {
3345         backup->compress = false;
3346     }
3347 
3348     bs = qmp_get_root_bs(backup->device, errp);
3349     if (!bs) {
3350         return NULL;
3351     }
3352 
3353     aio_context = bdrv_get_aio_context(bs);
3354     aio_context_acquire(aio_context);
3355 
3356     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3357     if (!target_bs) {
3358         goto out;
3359     }
3360 
3361     if (bdrv_get_aio_context(target_bs) != aio_context) {
3362         if (!bdrv_has_blk(target_bs)) {
3363             /* The target BDS is not attached, we can safely move it to another
3364              * AioContext. */
3365             bdrv_set_aio_context(target_bs, aio_context);
3366         } else {
3367             error_setg(errp, "Target is attached to a different thread from "
3368                              "source.");
3369             goto out;
3370         }
3371     }
3372     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3373                             backup->sync, NULL, backup->compress,
3374                             backup->on_source_error, backup->on_target_error,
3375                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3376     if (local_err != NULL) {
3377         error_propagate(errp, local_err);
3378     }
3379 out:
3380     aio_context_release(aio_context);
3381     return job;
3382 }
3383 
3384 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3385 {
3386     BlockJob *job;
3387     job = do_blockdev_backup(arg, NULL, errp);
3388     if (job) {
3389         block_job_start(job);
3390     }
3391 }
3392 
3393 /* Parameter check and block job starting for drive mirroring.
3394  * Caller should hold @device and @target's aio context (must be the same).
3395  **/
3396 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3397                                    BlockDriverState *target,
3398                                    bool has_replaces, const char *replaces,
3399                                    enum MirrorSyncMode sync,
3400                                    BlockMirrorBackingMode backing_mode,
3401                                    bool has_speed, int64_t speed,
3402                                    bool has_granularity, uint32_t granularity,
3403                                    bool has_buf_size, int64_t buf_size,
3404                                    bool has_on_source_error,
3405                                    BlockdevOnError on_source_error,
3406                                    bool has_on_target_error,
3407                                    BlockdevOnError on_target_error,
3408                                    bool has_unmap, bool unmap,
3409                                    bool has_filter_node_name,
3410                                    const char *filter_node_name,
3411                                    Error **errp)
3412 {
3413 
3414     if (!has_speed) {
3415         speed = 0;
3416     }
3417     if (!has_on_source_error) {
3418         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3419     }
3420     if (!has_on_target_error) {
3421         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3422     }
3423     if (!has_granularity) {
3424         granularity = 0;
3425     }
3426     if (!has_buf_size) {
3427         buf_size = 0;
3428     }
3429     if (!has_unmap) {
3430         unmap = true;
3431     }
3432     if (!has_filter_node_name) {
3433         filter_node_name = NULL;
3434     }
3435 
3436     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3437         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3438                    "a value in range [512B, 64MB]");
3439         return;
3440     }
3441     if (granularity & (granularity - 1)) {
3442         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3443                    "power of 2");
3444         return;
3445     }
3446 
3447     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3448         return;
3449     }
3450     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3451         return;
3452     }
3453 
3454     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3455         sync = MIRROR_SYNC_MODE_FULL;
3456     }
3457 
3458     /* pass the node name to replace to mirror start since it's loose coupling
3459      * and will allow to check whether the node still exist at mirror completion
3460      */
3461     mirror_start(job_id, bs, target,
3462                  has_replaces ? replaces : NULL,
3463                  speed, granularity, buf_size, sync, backing_mode,
3464                  on_source_error, on_target_error, unmap, filter_node_name,
3465                  errp);
3466 }
3467 
3468 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3469 {
3470     BlockDriverState *bs;
3471     BlockDriverState *source, *target_bs;
3472     AioContext *aio_context;
3473     BlockMirrorBackingMode backing_mode;
3474     Error *local_err = NULL;
3475     QDict *options = NULL;
3476     int flags;
3477     int64_t size;
3478     const char *format = arg->format;
3479 
3480     bs = qmp_get_root_bs(arg->device, errp);
3481     if (!bs) {
3482         return;
3483     }
3484 
3485     aio_context = bdrv_get_aio_context(bs);
3486     aio_context_acquire(aio_context);
3487 
3488     if (!arg->has_mode) {
3489         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3490     }
3491 
3492     if (!arg->has_format) {
3493         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3494                   ? NULL : bs->drv->format_name);
3495     }
3496 
3497     flags = bs->open_flags | BDRV_O_RDWR;
3498     source = backing_bs(bs);
3499     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3500         arg->sync = MIRROR_SYNC_MODE_FULL;
3501     }
3502     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3503         source = bs;
3504     }
3505 
3506     size = bdrv_getlength(bs);
3507     if (size < 0) {
3508         error_setg_errno(errp, -size, "bdrv_getlength failed");
3509         goto out;
3510     }
3511 
3512     if (arg->has_replaces) {
3513         BlockDriverState *to_replace_bs;
3514         AioContext *replace_aio_context;
3515         int64_t replace_size;
3516 
3517         if (!arg->has_node_name) {
3518             error_setg(errp, "a node-name must be provided when replacing a"
3519                              " named node of the graph");
3520             goto out;
3521         }
3522 
3523         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3524 
3525         if (!to_replace_bs) {
3526             error_propagate(errp, local_err);
3527             goto out;
3528         }
3529 
3530         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3531         aio_context_acquire(replace_aio_context);
3532         replace_size = bdrv_getlength(to_replace_bs);
3533         aio_context_release(replace_aio_context);
3534 
3535         if (size != replace_size) {
3536             error_setg(errp, "cannot replace image with a mirror image of "
3537                              "different size");
3538             goto out;
3539         }
3540     }
3541 
3542     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3543         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3544     } else {
3545         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3546     }
3547 
3548     /* Don't open backing image in create() */
3549     flags |= BDRV_O_NO_BACKING;
3550 
3551     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3552         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3553     {
3554         /* create new image w/o backing file */
3555         assert(format);
3556         bdrv_img_create(arg->target, format,
3557                         NULL, NULL, NULL, size, flags, false, &local_err);
3558     } else {
3559         switch (arg->mode) {
3560         case NEW_IMAGE_MODE_EXISTING:
3561             break;
3562         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3563             /* create new image with backing file */
3564             bdrv_img_create(arg->target, format,
3565                             source->filename,
3566                             source->drv->format_name,
3567                             NULL, size, flags, false, &local_err);
3568             break;
3569         default:
3570             abort();
3571         }
3572     }
3573 
3574     if (local_err) {
3575         error_propagate(errp, local_err);
3576         goto out;
3577     }
3578 
3579     options = qdict_new();
3580     if (arg->has_node_name) {
3581         qdict_put_str(options, "node-name", arg->node_name);
3582     }
3583     if (format) {
3584         qdict_put_str(options, "driver", format);
3585     }
3586 
3587     /* Mirroring takes care of copy-on-write using the source's backing
3588      * file.
3589      */
3590     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3591     if (!target_bs) {
3592         goto out;
3593     }
3594 
3595     bdrv_set_aio_context(target_bs, aio_context);
3596 
3597     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3598                            arg->has_replaces, arg->replaces, arg->sync,
3599                            backing_mode, arg->has_speed, arg->speed,
3600                            arg->has_granularity, arg->granularity,
3601                            arg->has_buf_size, arg->buf_size,
3602                            arg->has_on_source_error, arg->on_source_error,
3603                            arg->has_on_target_error, arg->on_target_error,
3604                            arg->has_unmap, arg->unmap,
3605                            false, NULL,
3606                            &local_err);
3607     bdrv_unref(target_bs);
3608     error_propagate(errp, local_err);
3609 out:
3610     aio_context_release(aio_context);
3611 }
3612 
3613 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3614                          const char *device, const char *target,
3615                          bool has_replaces, const char *replaces,
3616                          MirrorSyncMode sync,
3617                          bool has_speed, int64_t speed,
3618                          bool has_granularity, uint32_t granularity,
3619                          bool has_buf_size, int64_t buf_size,
3620                          bool has_on_source_error,
3621                          BlockdevOnError on_source_error,
3622                          bool has_on_target_error,
3623                          BlockdevOnError on_target_error,
3624                          bool has_filter_node_name,
3625                          const char *filter_node_name,
3626                          Error **errp)
3627 {
3628     BlockDriverState *bs;
3629     BlockDriverState *target_bs;
3630     AioContext *aio_context;
3631     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3632     Error *local_err = NULL;
3633 
3634     bs = qmp_get_root_bs(device, errp);
3635     if (!bs) {
3636         return;
3637     }
3638 
3639     target_bs = bdrv_lookup_bs(target, target, errp);
3640     if (!target_bs) {
3641         return;
3642     }
3643 
3644     aio_context = bdrv_get_aio_context(bs);
3645     aio_context_acquire(aio_context);
3646 
3647     bdrv_set_aio_context(target_bs, aio_context);
3648 
3649     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3650                            has_replaces, replaces, sync, backing_mode,
3651                            has_speed, speed,
3652                            has_granularity, granularity,
3653                            has_buf_size, buf_size,
3654                            has_on_source_error, on_source_error,
3655                            has_on_target_error, on_target_error,
3656                            true, true,
3657                            has_filter_node_name, filter_node_name,
3658                            &local_err);
3659     error_propagate(errp, local_err);
3660 
3661     aio_context_release(aio_context);
3662 }
3663 
3664 /* Get a block job using its ID and acquire its AioContext */
3665 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3666                                 Error **errp)
3667 {
3668     BlockJob *job;
3669 
3670     assert(id != NULL);
3671 
3672     *aio_context = NULL;
3673 
3674     job = block_job_get(id);
3675 
3676     if (!job) {
3677         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3678                   "Block job '%s' not found", id);
3679         return NULL;
3680     }
3681 
3682     *aio_context = blk_get_aio_context(job->blk);
3683     aio_context_acquire(*aio_context);
3684 
3685     return job;
3686 }
3687 
3688 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3689 {
3690     AioContext *aio_context;
3691     BlockJob *job = find_block_job(device, &aio_context, errp);
3692 
3693     if (!job) {
3694         return;
3695     }
3696 
3697     block_job_set_speed(job, speed, errp);
3698     aio_context_release(aio_context);
3699 }
3700 
3701 void qmp_block_job_cancel(const char *device,
3702                           bool has_force, bool force, Error **errp)
3703 {
3704     AioContext *aio_context;
3705     BlockJob *job = find_block_job(device, &aio_context, errp);
3706 
3707     if (!job) {
3708         return;
3709     }
3710 
3711     if (!has_force) {
3712         force = false;
3713     }
3714 
3715     if (block_job_user_paused(job) && !force) {
3716         error_setg(errp, "The block job for device '%s' is currently paused",
3717                    device);
3718         goto out;
3719     }
3720 
3721     trace_qmp_block_job_cancel(job);
3722     block_job_cancel(job);
3723 out:
3724     aio_context_release(aio_context);
3725 }
3726 
3727 void qmp_block_job_pause(const char *device, Error **errp)
3728 {
3729     AioContext *aio_context;
3730     BlockJob *job = find_block_job(device, &aio_context, errp);
3731 
3732     if (!job || block_job_user_paused(job)) {
3733         return;
3734     }
3735 
3736     trace_qmp_block_job_pause(job);
3737     block_job_user_pause(job);
3738     aio_context_release(aio_context);
3739 }
3740 
3741 void qmp_block_job_resume(const char *device, Error **errp)
3742 {
3743     AioContext *aio_context;
3744     BlockJob *job = find_block_job(device, &aio_context, errp);
3745 
3746     if (!job || !block_job_user_paused(job)) {
3747         return;
3748     }
3749 
3750     trace_qmp_block_job_resume(job);
3751     block_job_user_resume(job);
3752     aio_context_release(aio_context);
3753 }
3754 
3755 void qmp_block_job_complete(const char *device, Error **errp)
3756 {
3757     AioContext *aio_context;
3758     BlockJob *job = find_block_job(device, &aio_context, errp);
3759 
3760     if (!job) {
3761         return;
3762     }
3763 
3764     trace_qmp_block_job_complete(job);
3765     block_job_complete(job, errp);
3766     aio_context_release(aio_context);
3767 }
3768 
3769 void qmp_change_backing_file(const char *device,
3770                              const char *image_node_name,
3771                              const char *backing_file,
3772                              Error **errp)
3773 {
3774     BlockDriverState *bs = NULL;
3775     AioContext *aio_context;
3776     BlockDriverState *image_bs = NULL;
3777     Error *local_err = NULL;
3778     bool ro;
3779     int open_flags;
3780     int ret;
3781 
3782     bs = qmp_get_root_bs(device, errp);
3783     if (!bs) {
3784         return;
3785     }
3786 
3787     aio_context = bdrv_get_aio_context(bs);
3788     aio_context_acquire(aio_context);
3789 
3790     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3791     if (local_err) {
3792         error_propagate(errp, local_err);
3793         goto out;
3794     }
3795 
3796     if (!image_bs) {
3797         error_setg(errp, "image file not found");
3798         goto out;
3799     }
3800 
3801     if (bdrv_find_base(image_bs) == image_bs) {
3802         error_setg(errp, "not allowing backing file change on an image "
3803                          "without a backing file");
3804         goto out;
3805     }
3806 
3807     /* even though we are not necessarily operating on bs, we need it to
3808      * determine if block ops are currently prohibited on the chain */
3809     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3810         goto out;
3811     }
3812 
3813     /* final sanity check */
3814     if (!bdrv_chain_contains(bs, image_bs)) {
3815         error_setg(errp, "'%s' and image file are not in the same chain",
3816                    device);
3817         goto out;
3818     }
3819 
3820     /* if not r/w, reopen to make r/w */
3821     open_flags = image_bs->open_flags;
3822     ro = bdrv_is_read_only(image_bs);
3823 
3824     if (ro) {
3825         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3826         if (local_err) {
3827             error_propagate(errp, local_err);
3828             goto out;
3829         }
3830     }
3831 
3832     ret = bdrv_change_backing_file(image_bs, backing_file,
3833                                image_bs->drv ? image_bs->drv->format_name : "");
3834 
3835     if (ret < 0) {
3836         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3837                          backing_file);
3838         /* don't exit here, so we can try to restore open flags if
3839          * appropriate */
3840     }
3841 
3842     if (ro) {
3843         bdrv_reopen(image_bs, open_flags, &local_err);
3844         error_propagate(errp, local_err);
3845     }
3846 
3847 out:
3848     aio_context_release(aio_context);
3849 }
3850 
3851 void hmp_drive_add_node(Monitor *mon, const char *optstr)
3852 {
3853     QemuOpts *opts;
3854     QDict *qdict;
3855     Error *local_err = NULL;
3856 
3857     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
3858     if (!opts) {
3859         return;
3860     }
3861 
3862     qdict = qemu_opts_to_qdict(opts, NULL);
3863 
3864     if (!qdict_get_try_str(qdict, "node-name")) {
3865         QDECREF(qdict);
3866         error_report("'node-name' needs to be specified");
3867         goto out;
3868     }
3869 
3870     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
3871     if (!bs) {
3872         error_report_err(local_err);
3873         goto out;
3874     }
3875 
3876     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3877 
3878 out:
3879     qemu_opts_del(opts);
3880 }
3881 
3882 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3883 {
3884     BlockDriverState *bs;
3885     QObject *obj;
3886     Visitor *v = qobject_output_visitor_new(&obj);
3887     QDict *qdict;
3888     const QDictEntry *ent;
3889     Error *local_err = NULL;
3890 
3891     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
3892     if (local_err) {
3893         error_propagate(errp, local_err);
3894         goto fail;
3895     }
3896 
3897     visit_complete(v, &obj);
3898     qdict = qobject_to_qdict(obj);
3899 
3900     qdict_flatten(qdict);
3901 
3902     /*
3903      * Rewrite "backing": null to "backing": ""
3904      * TODO Rewrite "" to null instead, and perhaps not even here
3905      */
3906     for (ent = qdict_first(qdict); ent; ent = qdict_next(qdict, ent)) {
3907         char *dot = strrchr(ent->key, '.');
3908 
3909         if (!strcmp(dot ? dot + 1 : ent->key, "backing")
3910             && qobject_type(ent->value) == QTYPE_QNULL) {
3911             qdict_put(qdict, ent->key, qstring_new());
3912         }
3913     }
3914 
3915     if (!qdict_get_try_str(qdict, "node-name")) {
3916         error_setg(errp, "'node-name' must be specified for the root node");
3917         goto fail;
3918     }
3919 
3920     bs = bds_tree_init(qdict, errp);
3921     if (!bs) {
3922         goto fail;
3923     }
3924 
3925     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3926 
3927 fail:
3928     visit_free(v);
3929 }
3930 
3931 void qmp_blockdev_del(const char *node_name, Error **errp)
3932 {
3933     AioContext *aio_context;
3934     BlockDriverState *bs;
3935 
3936     bs = bdrv_find_node(node_name);
3937     if (!bs) {
3938         error_setg(errp, "Cannot find node %s", node_name);
3939         return;
3940     }
3941     if (bdrv_has_blk(bs)) {
3942         error_setg(errp, "Node %s is in use", node_name);
3943         return;
3944     }
3945     aio_context = bdrv_get_aio_context(bs);
3946     aio_context_acquire(aio_context);
3947 
3948     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3949         goto out;
3950     }
3951 
3952     if (!bs->monitor_list.tqe_prev) {
3953         error_setg(errp, "Node %s is not owned by the monitor",
3954                    bs->node_name);
3955         goto out;
3956     }
3957 
3958     if (bs->refcnt > 1) {
3959         error_setg(errp, "Block device %s is in use",
3960                    bdrv_get_device_or_node_name(bs));
3961         goto out;
3962     }
3963 
3964     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3965     bdrv_unref(bs);
3966 
3967 out:
3968     aio_context_release(aio_context);
3969 }
3970 
3971 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3972                                   const char *child_name)
3973 {
3974     BdrvChild *child;
3975 
3976     QLIST_FOREACH(child, &parent_bs->children, next) {
3977         if (strcmp(child->name, child_name) == 0) {
3978             return child;
3979         }
3980     }
3981 
3982     return NULL;
3983 }
3984 
3985 void qmp_x_blockdev_change(const char *parent, bool has_child,
3986                            const char *child, bool has_node,
3987                            const char *node, Error **errp)
3988 {
3989     BlockDriverState *parent_bs, *new_bs = NULL;
3990     BdrvChild *p_child;
3991 
3992     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3993     if (!parent_bs) {
3994         return;
3995     }
3996 
3997     if (has_child == has_node) {
3998         if (has_child) {
3999             error_setg(errp, "The parameters child and node are in conflict");
4000         } else {
4001             error_setg(errp, "Either child or node must be specified");
4002         }
4003         return;
4004     }
4005 
4006     if (has_child) {
4007         p_child = bdrv_find_child(parent_bs, child);
4008         if (!p_child) {
4009             error_setg(errp, "Node '%s' does not have child '%s'",
4010                        parent, child);
4011             return;
4012         }
4013         bdrv_del_child(parent_bs, p_child, errp);
4014     }
4015 
4016     if (has_node) {
4017         new_bs = bdrv_find_node(node);
4018         if (!new_bs) {
4019             error_setg(errp, "Node '%s' not found", node);
4020             return;
4021         }
4022         bdrv_add_child(parent_bs, new_bs, errp);
4023     }
4024 }
4025 
4026 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4027 {
4028     BlockJobInfoList *head = NULL, **p_next = &head;
4029     BlockJob *job;
4030 
4031     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4032         BlockJobInfoList *elem;
4033         AioContext *aio_context;
4034 
4035         if (block_job_is_internal(job)) {
4036             continue;
4037         }
4038         elem = g_new0(BlockJobInfoList, 1);
4039         aio_context = blk_get_aio_context(job->blk);
4040         aio_context_acquire(aio_context);
4041         elem->value = block_job_query(job, errp);
4042         aio_context_release(aio_context);
4043         if (!elem->value) {
4044             g_free(elem);
4045             qapi_free_BlockJobInfoList(head);
4046             return NULL;
4047         }
4048         *p_next = elem;
4049         p_next = &elem->next;
4050     }
4051 
4052     return head;
4053 }
4054 
4055 QemuOptsList qemu_common_drive_opts = {
4056     .name = "drive",
4057     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4058     .desc = {
4059         {
4060             .name = "snapshot",
4061             .type = QEMU_OPT_BOOL,
4062             .help = "enable/disable snapshot mode",
4063         },{
4064             .name = "aio",
4065             .type = QEMU_OPT_STRING,
4066             .help = "host AIO implementation (threads, native)",
4067         },{
4068             .name = BDRV_OPT_CACHE_WB,
4069             .type = QEMU_OPT_BOOL,
4070             .help = "Enable writeback mode",
4071         },{
4072             .name = "format",
4073             .type = QEMU_OPT_STRING,
4074             .help = "disk format (raw, qcow2, ...)",
4075         },{
4076             .name = "rerror",
4077             .type = QEMU_OPT_STRING,
4078             .help = "read error action",
4079         },{
4080             .name = "werror",
4081             .type = QEMU_OPT_STRING,
4082             .help = "write error action",
4083         },{
4084             .name = BDRV_OPT_READ_ONLY,
4085             .type = QEMU_OPT_BOOL,
4086             .help = "open drive file as read-only",
4087         },
4088 
4089         THROTTLE_OPTS,
4090 
4091         {
4092             .name = "throttling.group",
4093             .type = QEMU_OPT_STRING,
4094             .help = "name of the block throttling group",
4095         },{
4096             .name = "copy-on-read",
4097             .type = QEMU_OPT_BOOL,
4098             .help = "copy read data from backing file into image file",
4099         },{
4100             .name = "detect-zeroes",
4101             .type = QEMU_OPT_STRING,
4102             .help = "try to optimize zero writes (off, on, unmap)",
4103         },{
4104             .name = "stats-account-invalid",
4105             .type = QEMU_OPT_BOOL,
4106             .help = "whether to account for invalid I/O operations "
4107                     "in the statistics",
4108         },{
4109             .name = "stats-account-failed",
4110             .type = QEMU_OPT_BOOL,
4111             .help = "whether to account for failed I/O operations "
4112                     "in the statistics",
4113         },
4114         { /* end of list */ }
4115     },
4116 };
4117 
4118 QemuOptsList qemu_drive_opts = {
4119     .name = "drive",
4120     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4121     .desc = {
4122         /*
4123          * no elements => accept any params
4124          * validation will happen later
4125          */
4126         { /* end of list */ }
4127     },
4128 };
4129