xref: /qemu/blockdev.c (revision 795c40b8)
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 "qapi/util.h"
48 #include "sysemu/sysemu.h"
49 #include "block/block_int.h"
50 #include "qmp-commands.h"
51 #include "block/trace.h"
52 #include "sysemu/arch_init.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_QINT: {
338             int64_t length = qint_get_int(qobject_to_qint(entry->value));
339             if (length > 0 && length <= UINT_MAX) {
340                 block_acct_add_interval(stats, (unsigned) length);
341             } else {
342                 error_setg(errp, "Invalid interval length: %" PRId64, length);
343                 return false;
344             }
345             break;
346         }
347 
348         default:
349             error_setg(errp, "The specification of stats-intervals is invalid");
350             return false;
351         }
352     }
353     return true;
354 }
355 
356 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
357 
358 /* All parameters but @opts are optional and may be set to NULL. */
359 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
360     const char **throttling_group, ThrottleConfig *throttle_cfg,
361     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
362 {
363     Error *local_error = NULL;
364     const char *aio;
365 
366     if (bdrv_flags) {
367         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
368             *bdrv_flags |= BDRV_O_COPY_ON_READ;
369         }
370 
371         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
372             if (!strcmp(aio, "native")) {
373                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
374             } else if (!strcmp(aio, "threads")) {
375                 /* this is the default */
376             } else {
377                error_setg(errp, "invalid aio option");
378                return;
379             }
380         }
381     }
382 
383     /* disk I/O throttling */
384     if (throttling_group) {
385         *throttling_group = qemu_opt_get(opts, "throttling.group");
386     }
387 
388     if (throttle_cfg) {
389         throttle_config_init(throttle_cfg);
390         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
391             qemu_opt_get_number(opts, "throttling.bps-total", 0);
392         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
393             qemu_opt_get_number(opts, "throttling.bps-read", 0);
394         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
395             qemu_opt_get_number(opts, "throttling.bps-write", 0);
396         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
397             qemu_opt_get_number(opts, "throttling.iops-total", 0);
398         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
399             qemu_opt_get_number(opts, "throttling.iops-read", 0);
400         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
401             qemu_opt_get_number(opts, "throttling.iops-write", 0);
402 
403         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
404             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
405         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
406             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
407         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
408             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
409         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
410             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
411         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
412             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
413         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
414             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
415 
416         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
417             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
418         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
419             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
420         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
421             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
422         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
423             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
424         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
425             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
426         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
427             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
428 
429         throttle_cfg->op_size =
430             qemu_opt_get_number(opts, "throttling.iops-size", 0);
431 
432         if (!throttle_is_valid(throttle_cfg, errp)) {
433             return;
434         }
435     }
436 
437     if (detect_zeroes) {
438         *detect_zeroes =
439             qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
440                             qemu_opt_get(opts, "detect-zeroes"),
441                             BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
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         if (bdrv_key_required(bs)) {
595             autostart = 0;
596         }
597 
598         block_acct_init(blk_get_stats(blk), account_invalid, account_failed);
599 
600         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
601             blk_unref(blk);
602             blk = NULL;
603             goto err_no_bs_opts;
604         }
605     }
606 
607     /* disk I/O throttling */
608     if (throttle_enabled(&cfg)) {
609         if (!throttling_group) {
610             throttling_group = id;
611         }
612         blk_io_limits_enable(blk, throttling_group);
613         blk_set_io_limits(blk, &cfg);
614     }
615 
616     blk_set_enable_write_cache(blk, !writethrough);
617     blk_set_on_error(blk, on_read_error, on_write_error);
618 
619     if (!monitor_add_blk(blk, id, errp)) {
620         blk_unref(blk);
621         blk = NULL;
622         goto err_no_bs_opts;
623     }
624 
625 err_no_bs_opts:
626     qemu_opts_del(opts);
627     QDECREF(interval_dict);
628     QDECREF(interval_list);
629     return blk;
630 
631 early_err:
632     qemu_opts_del(opts);
633     QDECREF(interval_dict);
634     QDECREF(interval_list);
635 err_no_opts:
636     QDECREF(bs_opts);
637     return NULL;
638 }
639 
640 /* Takes the ownership of bs_opts */
641 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
642 {
643     int bdrv_flags = 0;
644 
645     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
646      * with other callers) rather than what we want as the real defaults.
647      * Apply the defaults here instead. */
648     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
649     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
650     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
651 
652     if (runstate_check(RUN_STATE_INMIGRATE)) {
653         bdrv_flags |= BDRV_O_INACTIVE;
654     }
655 
656     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
657 }
658 
659 void blockdev_close_all_bdrv_states(void)
660 {
661     BlockDriverState *bs, *next_bs;
662 
663     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
664         AioContext *ctx = bdrv_get_aio_context(bs);
665 
666         aio_context_acquire(ctx);
667         bdrv_unref(bs);
668         aio_context_release(ctx);
669     }
670 }
671 
672 /* Iterates over the list of monitor-owned BlockDriverStates */
673 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
674 {
675     return bs ? QTAILQ_NEXT(bs, monitor_list)
676               : QTAILQ_FIRST(&monitor_bdrv_states);
677 }
678 
679 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
680                             Error **errp)
681 {
682     const char *value;
683 
684     value = qemu_opt_get(opts, from);
685     if (value) {
686         if (qemu_opt_find(opts, to)) {
687             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
688                        "same time", to, from);
689             return;
690         }
691     }
692 
693     /* rename all items in opts */
694     while ((value = qemu_opt_get(opts, from))) {
695         qemu_opt_set(opts, to, value, &error_abort);
696         qemu_opt_unset(opts, from);
697     }
698 }
699 
700 QemuOptsList qemu_legacy_drive_opts = {
701     .name = "drive",
702     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
703     .desc = {
704         {
705             .name = "bus",
706             .type = QEMU_OPT_NUMBER,
707             .help = "bus number",
708         },{
709             .name = "unit",
710             .type = QEMU_OPT_NUMBER,
711             .help = "unit number (i.e. lun for scsi)",
712         },{
713             .name = "index",
714             .type = QEMU_OPT_NUMBER,
715             .help = "index number",
716         },{
717             .name = "media",
718             .type = QEMU_OPT_STRING,
719             .help = "media type (disk, cdrom)",
720         },{
721             .name = "if",
722             .type = QEMU_OPT_STRING,
723             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
724         },{
725             .name = "cyls",
726             .type = QEMU_OPT_NUMBER,
727             .help = "number of cylinders (ide disk geometry)",
728         },{
729             .name = "heads",
730             .type = QEMU_OPT_NUMBER,
731             .help = "number of heads (ide disk geometry)",
732         },{
733             .name = "secs",
734             .type = QEMU_OPT_NUMBER,
735             .help = "number of sectors (ide disk geometry)",
736         },{
737             .name = "trans",
738             .type = QEMU_OPT_STRING,
739             .help = "chs translation (auto, lba, none)",
740         },{
741             .name = "boot",
742             .type = QEMU_OPT_BOOL,
743             .help = "(deprecated, ignored)",
744         },{
745             .name = "addr",
746             .type = QEMU_OPT_STRING,
747             .help = "pci address (virtio only)",
748         },{
749             .name = "serial",
750             .type = QEMU_OPT_STRING,
751             .help = "disk serial number",
752         },{
753             .name = "file",
754             .type = QEMU_OPT_STRING,
755             .help = "file name",
756         },
757 
758         /* Options that are passed on, but have special semantics with -drive */
759         {
760             .name = BDRV_OPT_READ_ONLY,
761             .type = QEMU_OPT_BOOL,
762             .help = "open drive file as read-only",
763         },{
764             .name = "rerror",
765             .type = QEMU_OPT_STRING,
766             .help = "read error action",
767         },{
768             .name = "werror",
769             .type = QEMU_OPT_STRING,
770             .help = "write error action",
771         },{
772             .name = "copy-on-read",
773             .type = QEMU_OPT_BOOL,
774             .help = "copy read data from backing file into image file",
775         },
776 
777         { /* end of list */ }
778     },
779 };
780 
781 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
782 {
783     const char *value;
784     BlockBackend *blk;
785     DriveInfo *dinfo = NULL;
786     QDict *bs_opts;
787     QemuOpts *legacy_opts;
788     DriveMediaType media = MEDIA_DISK;
789     BlockInterfaceType type;
790     int cyls, heads, secs, translation;
791     int max_devs, bus_id, unit_id, index;
792     const char *devaddr;
793     const char *werror, *rerror;
794     bool read_only = false;
795     bool copy_on_read;
796     const char *serial;
797     const char *filename;
798     Error *local_err = NULL;
799     int i;
800 
801     /* Change legacy command line options into QMP ones */
802     static const struct {
803         const char *from;
804         const char *to;
805     } opt_renames[] = {
806         { "iops",           "throttling.iops-total" },
807         { "iops_rd",        "throttling.iops-read" },
808         { "iops_wr",        "throttling.iops-write" },
809 
810         { "bps",            "throttling.bps-total" },
811         { "bps_rd",         "throttling.bps-read" },
812         { "bps_wr",         "throttling.bps-write" },
813 
814         { "iops_max",       "throttling.iops-total-max" },
815         { "iops_rd_max",    "throttling.iops-read-max" },
816         { "iops_wr_max",    "throttling.iops-write-max" },
817 
818         { "bps_max",        "throttling.bps-total-max" },
819         { "bps_rd_max",     "throttling.bps-read-max" },
820         { "bps_wr_max",     "throttling.bps-write-max" },
821 
822         { "iops_size",      "throttling.iops-size" },
823 
824         { "group",          "throttling.group" },
825 
826         { "readonly",       BDRV_OPT_READ_ONLY },
827     };
828 
829     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
830         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
831                         &local_err);
832         if (local_err) {
833             error_report_err(local_err);
834             return NULL;
835         }
836     }
837 
838     value = qemu_opt_get(all_opts, "cache");
839     if (value) {
840         int flags = 0;
841         bool writethrough;
842 
843         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
844             error_report("invalid cache option");
845             return NULL;
846         }
847 
848         /* Specific options take precedence */
849         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
850             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
851                               !writethrough, &error_abort);
852         }
853         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
854             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
855                               !!(flags & BDRV_O_NOCACHE), &error_abort);
856         }
857         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
858             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
859                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
860         }
861         qemu_opt_unset(all_opts, "cache");
862     }
863 
864     /* Get a QDict for processing the options */
865     bs_opts = qdict_new();
866     qemu_opts_to_qdict(all_opts, bs_opts);
867 
868     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
869                                    &error_abort);
870     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
871     if (local_err) {
872         error_report_err(local_err);
873         goto fail;
874     }
875 
876     /* Deprecated option boot=[on|off] */
877     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
878         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
879                 "ignored. Future versions will reject this parameter. Please "
880                 "update your scripts.\n");
881     }
882 
883     /* Media type */
884     value = qemu_opt_get(legacy_opts, "media");
885     if (value) {
886         if (!strcmp(value, "disk")) {
887             media = MEDIA_DISK;
888         } else if (!strcmp(value, "cdrom")) {
889             media = MEDIA_CDROM;
890             read_only = true;
891         } else {
892             error_report("'%s' invalid media", value);
893             goto fail;
894         }
895     }
896 
897     /* copy-on-read is disabled with a warning for read-only devices */
898     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
899     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
900 
901     if (read_only && copy_on_read) {
902         error_report("warning: disabling copy-on-read on read-only drive");
903         copy_on_read = false;
904     }
905 
906     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
907     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
908 
909     /* Controller type */
910     value = qemu_opt_get(legacy_opts, "if");
911     if (value) {
912         for (type = 0;
913              type < IF_COUNT && strcmp(value, if_name[type]);
914              type++) {
915         }
916         if (type == IF_COUNT) {
917             error_report("unsupported bus type '%s'", value);
918             goto fail;
919         }
920     } else {
921         type = block_default_type;
922     }
923 
924     /* Geometry */
925     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
926     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
927     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
928 
929     if (cyls || heads || secs) {
930         if (cyls < 1) {
931             error_report("invalid physical cyls number");
932             goto fail;
933         }
934         if (heads < 1) {
935             error_report("invalid physical heads number");
936             goto fail;
937         }
938         if (secs < 1) {
939             error_report("invalid physical secs number");
940             goto fail;
941         }
942     }
943 
944     translation = BIOS_ATA_TRANSLATION_AUTO;
945     value = qemu_opt_get(legacy_opts, "trans");
946     if (value != NULL) {
947         if (!cyls) {
948             error_report("'%s' trans must be used with cyls, heads and secs",
949                          value);
950             goto fail;
951         }
952         if (!strcmp(value, "none")) {
953             translation = BIOS_ATA_TRANSLATION_NONE;
954         } else if (!strcmp(value, "lba")) {
955             translation = BIOS_ATA_TRANSLATION_LBA;
956         } else if (!strcmp(value, "large")) {
957             translation = BIOS_ATA_TRANSLATION_LARGE;
958         } else if (!strcmp(value, "rechs")) {
959             translation = BIOS_ATA_TRANSLATION_RECHS;
960         } else if (!strcmp(value, "auto")) {
961             translation = BIOS_ATA_TRANSLATION_AUTO;
962         } else {
963             error_report("'%s' invalid translation type", value);
964             goto fail;
965         }
966     }
967 
968     if (media == MEDIA_CDROM) {
969         if (cyls || secs || heads) {
970             error_report("CHS can't be set with media=cdrom");
971             goto fail;
972         }
973     }
974 
975     /* Device address specified by bus/unit or index.
976      * If none was specified, try to find the first free one. */
977     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
978     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
979     index   = qemu_opt_get_number(legacy_opts, "index", -1);
980 
981     max_devs = if_max_devs[type];
982 
983     if (index != -1) {
984         if (bus_id != 0 || unit_id != -1) {
985             error_report("index cannot be used with bus and unit");
986             goto fail;
987         }
988         bus_id = drive_index_to_bus_id(type, index);
989         unit_id = drive_index_to_unit_id(type, index);
990     }
991 
992     if (unit_id == -1) {
993        unit_id = 0;
994        while (drive_get(type, bus_id, unit_id) != NULL) {
995            unit_id++;
996            if (max_devs && unit_id >= max_devs) {
997                unit_id -= max_devs;
998                bus_id++;
999            }
1000        }
1001     }
1002 
1003     if (max_devs && unit_id >= max_devs) {
1004         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1005         goto fail;
1006     }
1007 
1008     if (drive_get(type, bus_id, unit_id) != NULL) {
1009         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1010                      bus_id, unit_id, index);
1011         goto fail;
1012     }
1013 
1014     /* Serial number */
1015     serial = qemu_opt_get(legacy_opts, "serial");
1016 
1017     /* no id supplied -> create one */
1018     if (qemu_opts_id(all_opts) == NULL) {
1019         char *new_id;
1020         const char *mediastr = "";
1021         if (type == IF_IDE || type == IF_SCSI) {
1022             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1023         }
1024         if (max_devs) {
1025             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1026                                      mediastr, unit_id);
1027         } else {
1028             new_id = g_strdup_printf("%s%s%i", if_name[type],
1029                                      mediastr, unit_id);
1030         }
1031         qdict_put_str(bs_opts, "id", new_id);
1032         g_free(new_id);
1033     }
1034 
1035     /* Add virtio block device */
1036     devaddr = qemu_opt_get(legacy_opts, "addr");
1037     if (devaddr && type != IF_VIRTIO) {
1038         error_report("addr is not supported by this bus type");
1039         goto fail;
1040     }
1041 
1042     if (type == IF_VIRTIO) {
1043         QemuOpts *devopts;
1044         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1045                                    &error_abort);
1046         if (arch_type == QEMU_ARCH_S390X) {
1047             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1048         } else {
1049             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1050         }
1051         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1052                      &error_abort);
1053         if (devaddr) {
1054             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1055         }
1056     }
1057 
1058     filename = qemu_opt_get(legacy_opts, "file");
1059 
1060     /* Check werror/rerror compatibility with if=... */
1061     werror = qemu_opt_get(legacy_opts, "werror");
1062     if (werror != NULL) {
1063         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1064             type != IF_NONE) {
1065             error_report("werror is not supported by this bus type");
1066             goto fail;
1067         }
1068         qdict_put_str(bs_opts, "werror", werror);
1069     }
1070 
1071     rerror = qemu_opt_get(legacy_opts, "rerror");
1072     if (rerror != NULL) {
1073         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1074             type != IF_NONE) {
1075             error_report("rerror is not supported by this bus type");
1076             goto fail;
1077         }
1078         qdict_put_str(bs_opts, "rerror", rerror);
1079     }
1080 
1081     /* Actual block device init: Functionality shared with blockdev-add */
1082     blk = blockdev_init(filename, bs_opts, &local_err);
1083     bs_opts = NULL;
1084     if (!blk) {
1085         if (local_err) {
1086             error_report_err(local_err);
1087         }
1088         goto fail;
1089     } else {
1090         assert(!local_err);
1091     }
1092 
1093     /* Create legacy DriveInfo */
1094     dinfo = g_malloc0(sizeof(*dinfo));
1095     dinfo->opts = all_opts;
1096 
1097     dinfo->cyls = cyls;
1098     dinfo->heads = heads;
1099     dinfo->secs = secs;
1100     dinfo->trans = translation;
1101 
1102     dinfo->type = type;
1103     dinfo->bus = bus_id;
1104     dinfo->unit = unit_id;
1105     dinfo->devaddr = devaddr;
1106     dinfo->serial = g_strdup(serial);
1107 
1108     blk_set_legacy_dinfo(blk, dinfo);
1109 
1110     switch(type) {
1111     case IF_IDE:
1112     case IF_SCSI:
1113     case IF_XEN:
1114     case IF_NONE:
1115         dinfo->media_cd = media == MEDIA_CDROM;
1116         break;
1117     default:
1118         break;
1119     }
1120 
1121 fail:
1122     qemu_opts_del(legacy_opts);
1123     QDECREF(bs_opts);
1124     return dinfo;
1125 }
1126 
1127 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1128 {
1129     BlockDriverState *bs;
1130 
1131     bs = bdrv_lookup_bs(name, name, errp);
1132     if (bs == NULL) {
1133         return NULL;
1134     }
1135 
1136     if (!bdrv_is_root_node(bs)) {
1137         error_setg(errp, "Need a root block node");
1138         return NULL;
1139     }
1140 
1141     if (!bdrv_is_inserted(bs)) {
1142         error_setg(errp, "Device has no medium");
1143         return NULL;
1144     }
1145 
1146     return bs;
1147 }
1148 
1149 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1150                                  Error **errp)
1151 {
1152     BlockBackend *blk;
1153 
1154     if (!blk_name == !qdev_id) {
1155         error_setg(errp, "Need exactly one of 'device' and 'id'");
1156         return NULL;
1157     }
1158 
1159     if (qdev_id) {
1160         blk = blk_by_qdev_id(qdev_id, errp);
1161     } else {
1162         blk = blk_by_name(blk_name);
1163         if (blk == NULL) {
1164             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1165                       "Device '%s' not found", blk_name);
1166         }
1167     }
1168 
1169     return blk;
1170 }
1171 
1172 void hmp_commit(Monitor *mon, const QDict *qdict)
1173 {
1174     const char *device = qdict_get_str(qdict, "device");
1175     BlockBackend *blk;
1176     int ret;
1177 
1178     if (!strcmp(device, "all")) {
1179         ret = blk_commit_all();
1180     } else {
1181         BlockDriverState *bs;
1182         AioContext *aio_context;
1183 
1184         blk = blk_by_name(device);
1185         if (!blk) {
1186             monitor_printf(mon, "Device '%s' not found\n", device);
1187             return;
1188         }
1189         if (!blk_is_available(blk)) {
1190             monitor_printf(mon, "Device '%s' has no medium\n", device);
1191             return;
1192         }
1193 
1194         bs = blk_bs(blk);
1195         aio_context = bdrv_get_aio_context(bs);
1196         aio_context_acquire(aio_context);
1197 
1198         ret = bdrv_commit(bs);
1199 
1200         aio_context_release(aio_context);
1201     }
1202     if (ret < 0) {
1203         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1204                        strerror(-ret));
1205     }
1206 }
1207 
1208 static void blockdev_do_action(TransactionAction *action, Error **errp)
1209 {
1210     TransactionActionList list;
1211 
1212     list.value = action;
1213     list.next = NULL;
1214     qmp_transaction(&list, false, NULL, errp);
1215 }
1216 
1217 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1218                                 bool has_node_name, const char *node_name,
1219                                 const char *snapshot_file,
1220                                 bool has_snapshot_node_name,
1221                                 const char *snapshot_node_name,
1222                                 bool has_format, const char *format,
1223                                 bool has_mode, NewImageMode mode, Error **errp)
1224 {
1225     BlockdevSnapshotSync snapshot = {
1226         .has_device = has_device,
1227         .device = (char *) device,
1228         .has_node_name = has_node_name,
1229         .node_name = (char *) node_name,
1230         .snapshot_file = (char *) snapshot_file,
1231         .has_snapshot_node_name = has_snapshot_node_name,
1232         .snapshot_node_name = (char *) snapshot_node_name,
1233         .has_format = has_format,
1234         .format = (char *) format,
1235         .has_mode = has_mode,
1236         .mode = mode,
1237     };
1238     TransactionAction action = {
1239         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1240         .u.blockdev_snapshot_sync.data = &snapshot,
1241     };
1242     blockdev_do_action(&action, errp);
1243 }
1244 
1245 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1246                            Error **errp)
1247 {
1248     BlockdevSnapshot snapshot_data = {
1249         .node = (char *) node,
1250         .overlay = (char *) overlay
1251     };
1252     TransactionAction action = {
1253         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1254         .u.blockdev_snapshot.data = &snapshot_data,
1255     };
1256     blockdev_do_action(&action, errp);
1257 }
1258 
1259 void qmp_blockdev_snapshot_internal_sync(const char *device,
1260                                          const char *name,
1261                                          Error **errp)
1262 {
1263     BlockdevSnapshotInternal snapshot = {
1264         .device = (char *) device,
1265         .name = (char *) name
1266     };
1267     TransactionAction action = {
1268         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1269         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1270     };
1271     blockdev_do_action(&action, errp);
1272 }
1273 
1274 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1275                                                          bool has_id,
1276                                                          const char *id,
1277                                                          bool has_name,
1278                                                          const char *name,
1279                                                          Error **errp)
1280 {
1281     BlockDriverState *bs;
1282     AioContext *aio_context;
1283     QEMUSnapshotInfo sn;
1284     Error *local_err = NULL;
1285     SnapshotInfo *info = NULL;
1286     int ret;
1287 
1288     bs = qmp_get_root_bs(device, errp);
1289     if (!bs) {
1290         return NULL;
1291     }
1292     aio_context = bdrv_get_aio_context(bs);
1293     aio_context_acquire(aio_context);
1294 
1295     if (!has_id) {
1296         id = NULL;
1297     }
1298 
1299     if (!has_name) {
1300         name = NULL;
1301     }
1302 
1303     if (!id && !name) {
1304         error_setg(errp, "Name or id must be provided");
1305         goto out_aio_context;
1306     }
1307 
1308     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1309         goto out_aio_context;
1310     }
1311 
1312     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1313     if (local_err) {
1314         error_propagate(errp, local_err);
1315         goto out_aio_context;
1316     }
1317     if (!ret) {
1318         error_setg(errp,
1319                    "Snapshot with id '%s' and name '%s' does not exist on "
1320                    "device '%s'",
1321                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1322         goto out_aio_context;
1323     }
1324 
1325     bdrv_snapshot_delete(bs, id, name, &local_err);
1326     if (local_err) {
1327         error_propagate(errp, local_err);
1328         goto out_aio_context;
1329     }
1330 
1331     aio_context_release(aio_context);
1332 
1333     info = g_new0(SnapshotInfo, 1);
1334     info->id = g_strdup(sn.id_str);
1335     info->name = g_strdup(sn.name);
1336     info->date_nsec = sn.date_nsec;
1337     info->date_sec = sn.date_sec;
1338     info->vm_state_size = sn.vm_state_size;
1339     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1340     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1341 
1342     return info;
1343 
1344 out_aio_context:
1345     aio_context_release(aio_context);
1346     return NULL;
1347 }
1348 
1349 /**
1350  * block_dirty_bitmap_lookup:
1351  * Return a dirty bitmap (if present), after validating
1352  * the node reference and bitmap names.
1353  *
1354  * @node: The name of the BDS node to search for bitmaps
1355  * @name: The name of the bitmap to search for
1356  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1357  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1358  * @errp: Output pointer for error information. Can be NULL.
1359  *
1360  * @return: A bitmap object on success, or NULL on failure.
1361  */
1362 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1363                                                   const char *name,
1364                                                   BlockDriverState **pbs,
1365                                                   AioContext **paio,
1366                                                   Error **errp)
1367 {
1368     BlockDriverState *bs;
1369     BdrvDirtyBitmap *bitmap;
1370     AioContext *aio_context;
1371 
1372     if (!node) {
1373         error_setg(errp, "Node cannot be NULL");
1374         return NULL;
1375     }
1376     if (!name) {
1377         error_setg(errp, "Bitmap name cannot be NULL");
1378         return NULL;
1379     }
1380     bs = bdrv_lookup_bs(node, node, NULL);
1381     if (!bs) {
1382         error_setg(errp, "Node '%s' not found", node);
1383         return NULL;
1384     }
1385 
1386     aio_context = bdrv_get_aio_context(bs);
1387     aio_context_acquire(aio_context);
1388 
1389     bitmap = bdrv_find_dirty_bitmap(bs, name);
1390     if (!bitmap) {
1391         error_setg(errp, "Dirty bitmap '%s' not found", name);
1392         goto fail;
1393     }
1394 
1395     if (pbs) {
1396         *pbs = bs;
1397     }
1398     if (paio) {
1399         *paio = aio_context;
1400     } else {
1401         aio_context_release(aio_context);
1402     }
1403 
1404     return bitmap;
1405 
1406  fail:
1407     aio_context_release(aio_context);
1408     return NULL;
1409 }
1410 
1411 /* New and old BlockDriverState structs for atomic group operations */
1412 
1413 typedef struct BlkActionState BlkActionState;
1414 
1415 /**
1416  * BlkActionOps:
1417  * Table of operations that define an Action.
1418  *
1419  * @instance_size: Size of state struct, in bytes.
1420  * @prepare: Prepare the work, must NOT be NULL.
1421  * @commit: Commit the changes, can be NULL.
1422  * @abort: Abort the changes on fail, can be NULL.
1423  * @clean: Clean up resources after all transaction actions have called
1424  *         commit() or abort(). Can be NULL.
1425  *
1426  * Only prepare() may fail. In a single transaction, only one of commit() or
1427  * abort() will be called. clean() will always be called if it is present.
1428  */
1429 typedef struct BlkActionOps {
1430     size_t instance_size;
1431     void (*prepare)(BlkActionState *common, Error **errp);
1432     void (*commit)(BlkActionState *common);
1433     void (*abort)(BlkActionState *common);
1434     void (*clean)(BlkActionState *common);
1435 } BlkActionOps;
1436 
1437 /**
1438  * BlkActionState:
1439  * Describes one Action's state within a Transaction.
1440  *
1441  * @action: QAPI-defined enum identifying which Action to perform.
1442  * @ops: Table of ActionOps this Action can perform.
1443  * @block_job_txn: Transaction which this action belongs to.
1444  * @entry: List membership for all Actions in this Transaction.
1445  *
1446  * This structure must be arranged as first member in a subclassed type,
1447  * assuming that the compiler will also arrange it to the same offsets as the
1448  * base class.
1449  */
1450 struct BlkActionState {
1451     TransactionAction *action;
1452     const BlkActionOps *ops;
1453     BlockJobTxn *block_job_txn;
1454     TransactionProperties *txn_props;
1455     QSIMPLEQ_ENTRY(BlkActionState) entry;
1456 };
1457 
1458 /* internal snapshot private data */
1459 typedef struct InternalSnapshotState {
1460     BlkActionState common;
1461     BlockDriverState *bs;
1462     AioContext *aio_context;
1463     QEMUSnapshotInfo sn;
1464     bool created;
1465 } InternalSnapshotState;
1466 
1467 
1468 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1469 {
1470     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1471         error_setg(errp,
1472                    "Action '%s' does not support Transaction property "
1473                    "completion-mode = %s",
1474                    TransactionActionKind_lookup[s->action->type],
1475                    ActionCompletionMode_lookup[s->txn_props->completion_mode]);
1476         return -1;
1477     }
1478     return 0;
1479 }
1480 
1481 static void internal_snapshot_prepare(BlkActionState *common,
1482                                       Error **errp)
1483 {
1484     Error *local_err = NULL;
1485     const char *device;
1486     const char *name;
1487     BlockDriverState *bs;
1488     QEMUSnapshotInfo old_sn, *sn;
1489     bool ret;
1490     qemu_timeval tv;
1491     BlockdevSnapshotInternal *internal;
1492     InternalSnapshotState *state;
1493     int ret1;
1494 
1495     g_assert(common->action->type ==
1496              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1497     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1498     state = DO_UPCAST(InternalSnapshotState, common, common);
1499 
1500     /* 1. parse input */
1501     device = internal->device;
1502     name = internal->name;
1503 
1504     /* 2. check for validation */
1505     if (action_check_completion_mode(common, errp) < 0) {
1506         return;
1507     }
1508 
1509     bs = qmp_get_root_bs(device, errp);
1510     if (!bs) {
1511         return;
1512     }
1513 
1514     /* AioContext is released in .clean() */
1515     state->aio_context = bdrv_get_aio_context(bs);
1516     aio_context_acquire(state->aio_context);
1517 
1518     state->bs = bs;
1519     bdrv_drained_begin(bs);
1520 
1521     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1522         return;
1523     }
1524 
1525     if (bdrv_is_read_only(bs)) {
1526         error_setg(errp, "Device '%s' is read only", device);
1527         return;
1528     }
1529 
1530     if (!bdrv_can_snapshot(bs)) {
1531         error_setg(errp, "Block format '%s' used by device '%s' "
1532                    "does not support internal snapshots",
1533                    bs->drv->format_name, device);
1534         return;
1535     }
1536 
1537     if (!strlen(name)) {
1538         error_setg(errp, "Name is empty");
1539         return;
1540     }
1541 
1542     /* check whether a snapshot with name exist */
1543     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1544                                             &local_err);
1545     if (local_err) {
1546         error_propagate(errp, local_err);
1547         return;
1548     } else if (ret) {
1549         error_setg(errp,
1550                    "Snapshot with name '%s' already exists on device '%s'",
1551                    name, device);
1552         return;
1553     }
1554 
1555     /* 3. take the snapshot */
1556     sn = &state->sn;
1557     pstrcpy(sn->name, sizeof(sn->name), name);
1558     qemu_gettimeofday(&tv);
1559     sn->date_sec = tv.tv_sec;
1560     sn->date_nsec = tv.tv_usec * 1000;
1561     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1562 
1563     ret1 = bdrv_snapshot_create(bs, sn);
1564     if (ret1 < 0) {
1565         error_setg_errno(errp, -ret1,
1566                          "Failed to create snapshot '%s' on device '%s'",
1567                          name, device);
1568         return;
1569     }
1570 
1571     /* 4. succeed, mark a snapshot is created */
1572     state->created = true;
1573 }
1574 
1575 static void internal_snapshot_abort(BlkActionState *common)
1576 {
1577     InternalSnapshotState *state =
1578                              DO_UPCAST(InternalSnapshotState, common, common);
1579     BlockDriverState *bs = state->bs;
1580     QEMUSnapshotInfo *sn = &state->sn;
1581     Error *local_error = NULL;
1582 
1583     if (!state->created) {
1584         return;
1585     }
1586 
1587     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1588         error_reportf_err(local_error,
1589                           "Failed to delete snapshot with id '%s' and "
1590                           "name '%s' on device '%s' in abort: ",
1591                           sn->id_str, sn->name,
1592                           bdrv_get_device_name(bs));
1593     }
1594 }
1595 
1596 static void internal_snapshot_clean(BlkActionState *common)
1597 {
1598     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1599                                              common, common);
1600 
1601     if (state->aio_context) {
1602         if (state->bs) {
1603             bdrv_drained_end(state->bs);
1604         }
1605         aio_context_release(state->aio_context);
1606     }
1607 }
1608 
1609 /* external snapshot private data */
1610 typedef struct ExternalSnapshotState {
1611     BlkActionState common;
1612     BlockDriverState *old_bs;
1613     BlockDriverState *new_bs;
1614     AioContext *aio_context;
1615     bool overlay_appended;
1616 } ExternalSnapshotState;
1617 
1618 static void external_snapshot_prepare(BlkActionState *common,
1619                                       Error **errp)
1620 {
1621     int flags = 0;
1622     QDict *options = NULL;
1623     Error *local_err = NULL;
1624     /* Device and node name of the image to generate the snapshot from */
1625     const char *device;
1626     const char *node_name;
1627     /* Reference to the new image (for 'blockdev-snapshot') */
1628     const char *snapshot_ref;
1629     /* File name of the new image (for 'blockdev-snapshot-sync') */
1630     const char *new_image_file;
1631     ExternalSnapshotState *state =
1632                              DO_UPCAST(ExternalSnapshotState, common, common);
1633     TransactionAction *action = common->action;
1634 
1635     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1636      * purpose but a different set of parameters */
1637     switch (action->type) {
1638     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1639         {
1640             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1641             device = s->node;
1642             node_name = s->node;
1643             new_image_file = NULL;
1644             snapshot_ref = s->overlay;
1645         }
1646         break;
1647     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1648         {
1649             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1650             device = s->has_device ? s->device : NULL;
1651             node_name = s->has_node_name ? s->node_name : NULL;
1652             new_image_file = s->snapshot_file;
1653             snapshot_ref = NULL;
1654         }
1655         break;
1656     default:
1657         g_assert_not_reached();
1658     }
1659 
1660     /* start processing */
1661     if (action_check_completion_mode(common, errp) < 0) {
1662         return;
1663     }
1664 
1665     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1666     if (!state->old_bs) {
1667         return;
1668     }
1669 
1670     /* Acquire AioContext now so any threads operating on old_bs stop */
1671     state->aio_context = bdrv_get_aio_context(state->old_bs);
1672     aio_context_acquire(state->aio_context);
1673     bdrv_drained_begin(state->old_bs);
1674 
1675     if (!bdrv_is_inserted(state->old_bs)) {
1676         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1677         return;
1678     }
1679 
1680     if (bdrv_op_is_blocked(state->old_bs,
1681                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1682         return;
1683     }
1684 
1685     if (!bdrv_is_read_only(state->old_bs)) {
1686         if (bdrv_flush(state->old_bs)) {
1687             error_setg(errp, QERR_IO_ERROR);
1688             return;
1689         }
1690     }
1691 
1692     if (!bdrv_is_first_non_filter(state->old_bs)) {
1693         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1694         return;
1695     }
1696 
1697     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1698         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1699         const char *format = s->has_format ? s->format : "qcow2";
1700         enum NewImageMode mode;
1701         const char *snapshot_node_name =
1702             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1703 
1704         if (node_name && !snapshot_node_name) {
1705             error_setg(errp, "New snapshot node name missing");
1706             return;
1707         }
1708 
1709         if (snapshot_node_name &&
1710             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1711             error_setg(errp, "New snapshot node name already in use");
1712             return;
1713         }
1714 
1715         flags = state->old_bs->open_flags;
1716         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1717 
1718         /* create new image w/backing file */
1719         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1720         if (mode != NEW_IMAGE_MODE_EXISTING) {
1721             int64_t size = bdrv_getlength(state->old_bs);
1722             if (size < 0) {
1723                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1724                 return;
1725             }
1726             bdrv_img_create(new_image_file, format,
1727                             state->old_bs->filename,
1728                             state->old_bs->drv->format_name,
1729                             NULL, size, flags, false, &local_err);
1730             if (local_err) {
1731                 error_propagate(errp, local_err);
1732                 return;
1733             }
1734         }
1735 
1736         options = qdict_new();
1737         if (s->has_snapshot_node_name) {
1738             qdict_put_str(options, "node-name", snapshot_node_name);
1739         }
1740         qdict_put_str(options, "driver", format);
1741 
1742         flags |= BDRV_O_NO_BACKING;
1743     }
1744 
1745     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1746                               errp);
1747     /* We will manually add the backing_hd field to the bs later */
1748     if (!state->new_bs) {
1749         return;
1750     }
1751 
1752     if (bdrv_has_blk(state->new_bs)) {
1753         error_setg(errp, "The snapshot is already in use");
1754         return;
1755     }
1756 
1757     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1758                            errp)) {
1759         return;
1760     }
1761 
1762     if (state->new_bs->backing != NULL) {
1763         error_setg(errp, "The snapshot already has a backing image");
1764         return;
1765     }
1766 
1767     if (!state->new_bs->drv->supports_backing) {
1768         error_setg(errp, "The snapshot does not support backing images");
1769         return;
1770     }
1771 
1772     bdrv_set_aio_context(state->new_bs, state->aio_context);
1773 
1774     /* This removes our old bs and adds the new bs. This is an operation that
1775      * can fail, so we need to do it in .prepare; undoing it for abort is
1776      * always possible. */
1777     bdrv_ref(state->new_bs);
1778     bdrv_append(state->new_bs, state->old_bs, &local_err);
1779     if (local_err) {
1780         error_propagate(errp, local_err);
1781         return;
1782     }
1783     state->overlay_appended = true;
1784 }
1785 
1786 static void external_snapshot_commit(BlkActionState *common)
1787 {
1788     ExternalSnapshotState *state =
1789                              DO_UPCAST(ExternalSnapshotState, common, common);
1790 
1791     /* We don't need (or want) to use the transactional
1792      * bdrv_reopen_multiple() across all the entries at once, because we
1793      * don't want to abort all of them if one of them fails the reopen */
1794     if (!state->old_bs->copy_on_read) {
1795         bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1796                     NULL);
1797     }
1798 }
1799 
1800 static void external_snapshot_abort(BlkActionState *common)
1801 {
1802     ExternalSnapshotState *state =
1803                              DO_UPCAST(ExternalSnapshotState, common, common);
1804     if (state->new_bs) {
1805         if (state->overlay_appended) {
1806             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1807         }
1808     }
1809 }
1810 
1811 static void external_snapshot_clean(BlkActionState *common)
1812 {
1813     ExternalSnapshotState *state =
1814                              DO_UPCAST(ExternalSnapshotState, common, common);
1815     if (state->aio_context) {
1816         bdrv_drained_end(state->old_bs);
1817         aio_context_release(state->aio_context);
1818         bdrv_unref(state->new_bs);
1819     }
1820 }
1821 
1822 typedef struct DriveBackupState {
1823     BlkActionState common;
1824     BlockDriverState *bs;
1825     AioContext *aio_context;
1826     BlockJob *job;
1827 } DriveBackupState;
1828 
1829 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
1830                             Error **errp);
1831 
1832 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1833 {
1834     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1835     BlockDriverState *bs;
1836     DriveBackup *backup;
1837     Error *local_err = NULL;
1838 
1839     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1840     backup = common->action->u.drive_backup.data;
1841 
1842     bs = qmp_get_root_bs(backup->device, errp);
1843     if (!bs) {
1844         return;
1845     }
1846 
1847     /* AioContext is released in .clean() */
1848     state->aio_context = bdrv_get_aio_context(bs);
1849     aio_context_acquire(state->aio_context);
1850     bdrv_drained_begin(bs);
1851     state->bs = bs;
1852 
1853     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1854     if (local_err) {
1855         error_propagate(errp, local_err);
1856         return;
1857     }
1858 }
1859 
1860 static void drive_backup_commit(BlkActionState *common)
1861 {
1862     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1863     assert(state->job);
1864     block_job_start(state->job);
1865 }
1866 
1867 static void drive_backup_abort(BlkActionState *common)
1868 {
1869     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1870 
1871     if (state->job) {
1872         block_job_cancel_sync(state->job);
1873     }
1874 }
1875 
1876 static void drive_backup_clean(BlkActionState *common)
1877 {
1878     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1879 
1880     if (state->aio_context) {
1881         bdrv_drained_end(state->bs);
1882         aio_context_release(state->aio_context);
1883     }
1884 }
1885 
1886 typedef struct BlockdevBackupState {
1887     BlkActionState common;
1888     BlockDriverState *bs;
1889     BlockJob *job;
1890     AioContext *aio_context;
1891 } BlockdevBackupState;
1892 
1893 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
1894                                     Error **errp);
1895 
1896 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1897 {
1898     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1899     BlockdevBackup *backup;
1900     BlockDriverState *bs, *target;
1901     Error *local_err = NULL;
1902 
1903     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1904     backup = common->action->u.blockdev_backup.data;
1905 
1906     bs = qmp_get_root_bs(backup->device, errp);
1907     if (!bs) {
1908         return;
1909     }
1910 
1911     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1912     if (!target) {
1913         return;
1914     }
1915 
1916     /* AioContext is released in .clean() */
1917     state->aio_context = bdrv_get_aio_context(bs);
1918     if (state->aio_context != bdrv_get_aio_context(target)) {
1919         state->aio_context = NULL;
1920         error_setg(errp, "Backup between two IO threads is not implemented");
1921         return;
1922     }
1923     aio_context_acquire(state->aio_context);
1924     state->bs = bs;
1925     bdrv_drained_begin(state->bs);
1926 
1927     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1928     if (local_err) {
1929         error_propagate(errp, local_err);
1930         return;
1931     }
1932 }
1933 
1934 static void blockdev_backup_commit(BlkActionState *common)
1935 {
1936     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1937     assert(state->job);
1938     block_job_start(state->job);
1939 }
1940 
1941 static void blockdev_backup_abort(BlkActionState *common)
1942 {
1943     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1944 
1945     if (state->job) {
1946         block_job_cancel_sync(state->job);
1947     }
1948 }
1949 
1950 static void blockdev_backup_clean(BlkActionState *common)
1951 {
1952     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1953 
1954     if (state->aio_context) {
1955         bdrv_drained_end(state->bs);
1956         aio_context_release(state->aio_context);
1957     }
1958 }
1959 
1960 typedef struct BlockDirtyBitmapState {
1961     BlkActionState common;
1962     BdrvDirtyBitmap *bitmap;
1963     BlockDriverState *bs;
1964     AioContext *aio_context;
1965     HBitmap *backup;
1966     bool prepared;
1967 } BlockDirtyBitmapState;
1968 
1969 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1970                                            Error **errp)
1971 {
1972     Error *local_err = NULL;
1973     BlockDirtyBitmapAdd *action;
1974     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1975                                              common, common);
1976 
1977     if (action_check_completion_mode(common, errp) < 0) {
1978         return;
1979     }
1980 
1981     action = common->action->u.block_dirty_bitmap_add.data;
1982     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1983     qmp_block_dirty_bitmap_add(action->node, action->name,
1984                                action->has_granularity, action->granularity,
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                                               &state->aio_context,
2025                                               errp);
2026     if (!state->bitmap) {
2027         return;
2028     }
2029 
2030     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2031         error_setg(errp, "Cannot modify a frozen bitmap");
2032         return;
2033     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2034         error_setg(errp, "Cannot clear a disabled bitmap");
2035         return;
2036     }
2037 
2038     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2039     /* AioContext is released in .clean() */
2040 }
2041 
2042 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2043 {
2044     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2045                                              common, common);
2046 
2047     if (state->backup) {
2048         bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2049     }
2050 }
2051 
2052 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2053 {
2054     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2055                                              common, common);
2056 
2057     hbitmap_free(state->backup);
2058 }
2059 
2060 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2061 {
2062     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2063                                              common, common);
2064 
2065     if (state->aio_context) {
2066         aio_context_release(state->aio_context);
2067     }
2068 }
2069 
2070 static void abort_prepare(BlkActionState *common, Error **errp)
2071 {
2072     error_setg(errp, "Transaction aborted using Abort action");
2073 }
2074 
2075 static void abort_commit(BlkActionState *common)
2076 {
2077     g_assert_not_reached(); /* this action never succeeds */
2078 }
2079 
2080 static const BlkActionOps actions[] = {
2081     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2082         .instance_size = sizeof(ExternalSnapshotState),
2083         .prepare  = external_snapshot_prepare,
2084         .commit   = external_snapshot_commit,
2085         .abort = external_snapshot_abort,
2086         .clean = external_snapshot_clean,
2087     },
2088     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2089         .instance_size = sizeof(ExternalSnapshotState),
2090         .prepare  = external_snapshot_prepare,
2091         .commit   = external_snapshot_commit,
2092         .abort = external_snapshot_abort,
2093         .clean = external_snapshot_clean,
2094     },
2095     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2096         .instance_size = sizeof(DriveBackupState),
2097         .prepare = drive_backup_prepare,
2098         .commit = drive_backup_commit,
2099         .abort = drive_backup_abort,
2100         .clean = drive_backup_clean,
2101     },
2102     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2103         .instance_size = sizeof(BlockdevBackupState),
2104         .prepare = blockdev_backup_prepare,
2105         .commit = blockdev_backup_commit,
2106         .abort = blockdev_backup_abort,
2107         .clean = blockdev_backup_clean,
2108     },
2109     [TRANSACTION_ACTION_KIND_ABORT] = {
2110         .instance_size = sizeof(BlkActionState),
2111         .prepare = abort_prepare,
2112         .commit = abort_commit,
2113     },
2114     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2115         .instance_size = sizeof(InternalSnapshotState),
2116         .prepare  = internal_snapshot_prepare,
2117         .abort = internal_snapshot_abort,
2118         .clean = internal_snapshot_clean,
2119     },
2120     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2121         .instance_size = sizeof(BlockDirtyBitmapState),
2122         .prepare = block_dirty_bitmap_add_prepare,
2123         .abort = block_dirty_bitmap_add_abort,
2124     },
2125     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2126         .instance_size = sizeof(BlockDirtyBitmapState),
2127         .prepare = block_dirty_bitmap_clear_prepare,
2128         .commit = block_dirty_bitmap_clear_commit,
2129         .abort = block_dirty_bitmap_clear_abort,
2130         .clean = block_dirty_bitmap_clear_clean,
2131     }
2132 };
2133 
2134 /**
2135  * Allocate a TransactionProperties structure if necessary, and fill
2136  * that structure with desired defaults if they are unset.
2137  */
2138 static TransactionProperties *get_transaction_properties(
2139     TransactionProperties *props)
2140 {
2141     if (!props) {
2142         props = g_new0(TransactionProperties, 1);
2143     }
2144 
2145     if (!props->has_completion_mode) {
2146         props->has_completion_mode = true;
2147         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2148     }
2149 
2150     return props;
2151 }
2152 
2153 /*
2154  * 'Atomic' group operations.  The operations are performed as a set, and if
2155  * any fail then we roll back all operations in the group.
2156  */
2157 void qmp_transaction(TransactionActionList *dev_list,
2158                      bool has_props,
2159                      struct TransactionProperties *props,
2160                      Error **errp)
2161 {
2162     TransactionActionList *dev_entry = dev_list;
2163     BlockJobTxn *block_job_txn = NULL;
2164     BlkActionState *state, *next;
2165     Error *local_err = NULL;
2166 
2167     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2168     QSIMPLEQ_INIT(&snap_bdrv_states);
2169 
2170     /* Does this transaction get canceled as a group on failure?
2171      * If not, we don't really need to make a BlockJobTxn.
2172      */
2173     props = get_transaction_properties(props);
2174     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2175         block_job_txn = block_job_txn_new();
2176     }
2177 
2178     /* drain all i/o before any operations */
2179     bdrv_drain_all();
2180 
2181     /* We don't do anything in this loop that commits us to the operations */
2182     while (NULL != dev_entry) {
2183         TransactionAction *dev_info = NULL;
2184         const BlkActionOps *ops;
2185 
2186         dev_info = dev_entry->value;
2187         dev_entry = dev_entry->next;
2188 
2189         assert(dev_info->type < ARRAY_SIZE(actions));
2190 
2191         ops = &actions[dev_info->type];
2192         assert(ops->instance_size > 0);
2193 
2194         state = g_malloc0(ops->instance_size);
2195         state->ops = ops;
2196         state->action = dev_info;
2197         state->block_job_txn = block_job_txn;
2198         state->txn_props = props;
2199         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2200 
2201         state->ops->prepare(state, &local_err);
2202         if (local_err) {
2203             error_propagate(errp, local_err);
2204             goto delete_and_fail;
2205         }
2206     }
2207 
2208     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2209         if (state->ops->commit) {
2210             state->ops->commit(state);
2211         }
2212     }
2213 
2214     /* success */
2215     goto exit;
2216 
2217 delete_and_fail:
2218     /* failure, and it is all-or-none; roll back all operations */
2219     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2220         if (state->ops->abort) {
2221             state->ops->abort(state);
2222         }
2223     }
2224 exit:
2225     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2226         if (state->ops->clean) {
2227             state->ops->clean(state);
2228         }
2229         g_free(state);
2230     }
2231     if (!has_props) {
2232         qapi_free_TransactionProperties(props);
2233     }
2234     block_job_txn_unref(block_job_txn);
2235 }
2236 
2237 void qmp_eject(bool has_device, const char *device,
2238                bool has_id, const char *id,
2239                bool has_force, bool force, Error **errp)
2240 {
2241     Error *local_err = NULL;
2242     int rc;
2243 
2244     if (!has_force) {
2245         force = false;
2246     }
2247 
2248     rc = do_open_tray(has_device ? device : NULL,
2249                       has_id ? id : NULL,
2250                       force, &local_err);
2251     if (rc && rc != -ENOSYS) {
2252         error_propagate(errp, local_err);
2253         return;
2254     }
2255     error_free(local_err);
2256 
2257     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, errp);
2258 }
2259 
2260 void qmp_block_passwd(bool has_device, const char *device,
2261                       bool has_node_name, const char *node_name,
2262                       const char *password, Error **errp)
2263 {
2264     Error *local_err = NULL;
2265     BlockDriverState *bs;
2266     AioContext *aio_context;
2267 
2268     bs = bdrv_lookup_bs(has_device ? device : NULL,
2269                         has_node_name ? node_name : NULL,
2270                         &local_err);
2271     if (local_err) {
2272         error_propagate(errp, local_err);
2273         return;
2274     }
2275 
2276     aio_context = bdrv_get_aio_context(bs);
2277     aio_context_acquire(aio_context);
2278 
2279     bdrv_add_key(bs, password, errp);
2280 
2281     aio_context_release(aio_context);
2282 }
2283 
2284 /*
2285  * Attempt to open the tray of @device.
2286  * If @force, ignore its tray lock.
2287  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2288  * On error, store an error through @errp and return -errno.
2289  * If @device does not exist, return -ENODEV.
2290  * If it has no removable media, return -ENOTSUP.
2291  * If it has no tray, return -ENOSYS.
2292  * If the guest was asked to open the tray, return -EINPROGRESS.
2293  * Else, return 0.
2294  */
2295 static int do_open_tray(const char *blk_name, const char *qdev_id,
2296                         bool force, Error **errp)
2297 {
2298     BlockBackend *blk;
2299     const char *device = qdev_id ?: blk_name;
2300     bool locked;
2301 
2302     blk = qmp_get_blk(blk_name, qdev_id, errp);
2303     if (!blk) {
2304         return -ENODEV;
2305     }
2306 
2307     if (!blk_dev_has_removable_media(blk)) {
2308         error_setg(errp, "Device '%s' is not removable", device);
2309         return -ENOTSUP;
2310     }
2311 
2312     if (!blk_dev_has_tray(blk)) {
2313         error_setg(errp, "Device '%s' does not have a tray", device);
2314         return -ENOSYS;
2315     }
2316 
2317     if (blk_dev_is_tray_open(blk)) {
2318         return 0;
2319     }
2320 
2321     locked = blk_dev_is_medium_locked(blk);
2322     if (locked) {
2323         blk_dev_eject_request(blk, force);
2324     }
2325 
2326     if (!locked || force) {
2327         blk_dev_change_media_cb(blk, false, &error_abort);
2328     }
2329 
2330     if (locked && !force) {
2331         error_setg(errp, "Device '%s' is locked and force was not specified, "
2332                    "wait for tray to open and try again", device);
2333         return -EINPROGRESS;
2334     }
2335 
2336     return 0;
2337 }
2338 
2339 void qmp_blockdev_open_tray(bool has_device, const char *device,
2340                             bool has_id, const char *id,
2341                             bool has_force, bool force,
2342                             Error **errp)
2343 {
2344     Error *local_err = NULL;
2345     int rc;
2346 
2347     if (!has_force) {
2348         force = false;
2349     }
2350     rc = do_open_tray(has_device ? device : NULL,
2351                       has_id ? id : NULL,
2352                       force, &local_err);
2353     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2354         error_propagate(errp, local_err);
2355         return;
2356     }
2357     error_free(local_err);
2358 }
2359 
2360 void qmp_blockdev_close_tray(bool has_device, const char *device,
2361                              bool has_id, const char *id,
2362                              Error **errp)
2363 {
2364     BlockBackend *blk;
2365     Error *local_err = NULL;
2366 
2367     device = has_device ? device : NULL;
2368     id = has_id ? id : NULL;
2369 
2370     blk = qmp_get_blk(device, id, errp);
2371     if (!blk) {
2372         return;
2373     }
2374 
2375     if (!blk_dev_has_removable_media(blk)) {
2376         error_setg(errp, "Device '%s' is not removable", device ?: id);
2377         return;
2378     }
2379 
2380     if (!blk_dev_has_tray(blk)) {
2381         /* Ignore this command on tray-less devices */
2382         return;
2383     }
2384 
2385     if (!blk_dev_is_tray_open(blk)) {
2386         return;
2387     }
2388 
2389     blk_dev_change_media_cb(blk, true, &local_err);
2390     if (local_err) {
2391         error_propagate(errp, local_err);
2392         return;
2393     }
2394 }
2395 
2396 void qmp_x_blockdev_remove_medium(bool has_device, const char *device,
2397                                   bool has_id, const char *id, Error **errp)
2398 {
2399     BlockBackend *blk;
2400     BlockDriverState *bs;
2401     AioContext *aio_context;
2402     bool has_attached_device;
2403 
2404     device = has_device ? device : NULL;
2405     id = has_id ? id : NULL;
2406 
2407     blk = qmp_get_blk(device, id, errp);
2408     if (!blk) {
2409         return;
2410     }
2411 
2412     /* For BBs without a device, we can exchange the BDS tree at will */
2413     has_attached_device = blk_get_attached_dev(blk);
2414 
2415     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2416         error_setg(errp, "Device '%s' is not removable", device ?: id);
2417         return;
2418     }
2419 
2420     if (has_attached_device && blk_dev_has_tray(blk) &&
2421         !blk_dev_is_tray_open(blk))
2422     {
2423         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2424         return;
2425     }
2426 
2427     bs = blk_bs(blk);
2428     if (!bs) {
2429         return;
2430     }
2431 
2432     aio_context = bdrv_get_aio_context(bs);
2433     aio_context_acquire(aio_context);
2434 
2435     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2436         goto out;
2437     }
2438 
2439     blk_remove_bs(blk);
2440 
2441     if (!blk_dev_has_tray(blk)) {
2442         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2443          * called at all); therefore, the medium needs to be ejected here.
2444          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2445          * value passed here (i.e. false). */
2446         blk_dev_change_media_cb(blk, false, &error_abort);
2447     }
2448 
2449 out:
2450     aio_context_release(aio_context);
2451 }
2452 
2453 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2454                                             BlockDriverState *bs, Error **errp)
2455 {
2456     Error *local_err = NULL;
2457     bool has_device;
2458     int ret;
2459 
2460     /* For BBs without a device, we can exchange the BDS tree at will */
2461     has_device = blk_get_attached_dev(blk);
2462 
2463     if (has_device && !blk_dev_has_removable_media(blk)) {
2464         error_setg(errp, "Device is not removable");
2465         return;
2466     }
2467 
2468     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2469         error_setg(errp, "Tray of the device is not open");
2470         return;
2471     }
2472 
2473     if (blk_bs(blk)) {
2474         error_setg(errp, "There already is a medium in the device");
2475         return;
2476     }
2477 
2478     ret = blk_insert_bs(blk, bs, errp);
2479     if (ret < 0) {
2480         return;
2481     }
2482 
2483     if (!blk_dev_has_tray(blk)) {
2484         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2485          * called at all); therefore, the medium needs to be pushed into the
2486          * slot here.
2487          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2488          * value passed here (i.e. true). */
2489         blk_dev_change_media_cb(blk, true, &local_err);
2490         if (local_err) {
2491             error_propagate(errp, local_err);
2492             blk_remove_bs(blk);
2493             return;
2494         }
2495     }
2496 }
2497 
2498 void qmp_x_blockdev_insert_medium(bool has_device, const char *device,
2499                                   bool has_id, const char *id,
2500                                   const char *node_name, Error **errp)
2501 {
2502     BlockBackend *blk;
2503     BlockDriverState *bs;
2504 
2505     blk = qmp_get_blk(has_device ? device : NULL,
2506                       has_id ? id : NULL,
2507                       errp);
2508     if (!blk) {
2509         return;
2510     }
2511 
2512     bs = bdrv_find_node(node_name);
2513     if (!bs) {
2514         error_setg(errp, "Node '%s' not found", node_name);
2515         return;
2516     }
2517 
2518     if (bdrv_has_blk(bs)) {
2519         error_setg(errp, "Node '%s' is already in use", node_name);
2520         return;
2521     }
2522 
2523     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2524 }
2525 
2526 void qmp_blockdev_change_medium(bool has_device, const char *device,
2527                                 bool has_id, const char *id,
2528                                 const char *filename,
2529                                 bool has_format, const char *format,
2530                                 bool has_read_only,
2531                                 BlockdevChangeReadOnlyMode read_only,
2532                                 Error **errp)
2533 {
2534     BlockBackend *blk;
2535     BlockDriverState *medium_bs = NULL;
2536     int bdrv_flags;
2537     bool detect_zeroes;
2538     int rc;
2539     QDict *options = NULL;
2540     Error *err = NULL;
2541 
2542     blk = qmp_get_blk(has_device ? device : NULL,
2543                       has_id ? id : NULL,
2544                       errp);
2545     if (!blk) {
2546         goto fail;
2547     }
2548 
2549     if (blk_bs(blk)) {
2550         blk_update_root_state(blk);
2551     }
2552 
2553     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2554     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2555         BDRV_O_PROTOCOL);
2556 
2557     if (!has_read_only) {
2558         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2559     }
2560 
2561     switch (read_only) {
2562     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2563         break;
2564 
2565     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2566         bdrv_flags &= ~BDRV_O_RDWR;
2567         break;
2568 
2569     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2570         bdrv_flags |= BDRV_O_RDWR;
2571         break;
2572 
2573     default:
2574         abort();
2575     }
2576 
2577     options = qdict_new();
2578     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2579     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2580 
2581     if (has_format) {
2582         qdict_put_str(options, "driver", format);
2583     }
2584 
2585     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2586     if (!medium_bs) {
2587         goto fail;
2588     }
2589 
2590     bdrv_add_key(medium_bs, NULL, &err);
2591     if (err) {
2592         error_propagate(errp, err);
2593         goto fail;
2594     }
2595 
2596     rc = do_open_tray(has_device ? device : NULL,
2597                       has_id ? id : NULL,
2598                       false, &err);
2599     if (rc && rc != -ENOSYS) {
2600         error_propagate(errp, err);
2601         goto fail;
2602     }
2603     error_free(err);
2604     err = NULL;
2605 
2606     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, &err);
2607     if (err) {
2608         error_propagate(errp, err);
2609         goto fail;
2610     }
2611 
2612     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2613     if (err) {
2614         error_propagate(errp, err);
2615         goto fail;
2616     }
2617 
2618     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2619 
2620 fail:
2621     /* If the medium has been inserted, the device has its own reference, so
2622      * ours must be relinquished; and if it has not been inserted successfully,
2623      * the reference must be relinquished anyway */
2624     bdrv_unref(medium_bs);
2625 }
2626 
2627 /* throttling disk I/O limits */
2628 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2629 {
2630     ThrottleConfig cfg;
2631     BlockDriverState *bs;
2632     BlockBackend *blk;
2633     AioContext *aio_context;
2634 
2635     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2636                       arg->has_id ? arg->id : NULL,
2637                       errp);
2638     if (!blk) {
2639         return;
2640     }
2641 
2642     aio_context = blk_get_aio_context(blk);
2643     aio_context_acquire(aio_context);
2644 
2645     bs = blk_bs(blk);
2646     if (!bs) {
2647         error_setg(errp, "Device has no medium");
2648         goto out;
2649     }
2650 
2651     throttle_config_init(&cfg);
2652     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2653     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2654     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2655 
2656     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2657     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2658     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2659 
2660     if (arg->has_bps_max) {
2661         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2662     }
2663     if (arg->has_bps_rd_max) {
2664         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2665     }
2666     if (arg->has_bps_wr_max) {
2667         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2668     }
2669     if (arg->has_iops_max) {
2670         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2671     }
2672     if (arg->has_iops_rd_max) {
2673         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2674     }
2675     if (arg->has_iops_wr_max) {
2676         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2677     }
2678 
2679     if (arg->has_bps_max_length) {
2680         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2681     }
2682     if (arg->has_bps_rd_max_length) {
2683         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2684     }
2685     if (arg->has_bps_wr_max_length) {
2686         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2687     }
2688     if (arg->has_iops_max_length) {
2689         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2690     }
2691     if (arg->has_iops_rd_max_length) {
2692         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2693     }
2694     if (arg->has_iops_wr_max_length) {
2695         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2696     }
2697 
2698     if (arg->has_iops_size) {
2699         cfg.op_size = arg->iops_size;
2700     }
2701 
2702     if (!throttle_is_valid(&cfg, errp)) {
2703         goto out;
2704     }
2705 
2706     if (throttle_enabled(&cfg)) {
2707         /* Enable I/O limits if they're not enabled yet, otherwise
2708          * just update the throttling group. */
2709         if (!blk_get_public(blk)->throttle_state) {
2710             blk_io_limits_enable(blk,
2711                                  arg->has_group ? arg->group :
2712                                  arg->has_device ? arg->device :
2713                                  arg->id);
2714         } else if (arg->has_group) {
2715             blk_io_limits_update_group(blk, arg->group);
2716         }
2717         /* Set the new throttling configuration */
2718         blk_set_io_limits(blk, &cfg);
2719     } else if (blk_get_public(blk)->throttle_state) {
2720         /* If all throttling settings are set to 0, disable I/O limits */
2721         blk_io_limits_disable(blk);
2722     }
2723 
2724 out:
2725     aio_context_release(aio_context);
2726 }
2727 
2728 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2729                                 bool has_granularity, uint32_t granularity,
2730                                 Error **errp)
2731 {
2732     AioContext *aio_context;
2733     BlockDriverState *bs;
2734 
2735     if (!name || name[0] == '\0') {
2736         error_setg(errp, "Bitmap name cannot be empty");
2737         return;
2738     }
2739 
2740     bs = bdrv_lookup_bs(node, node, errp);
2741     if (!bs) {
2742         return;
2743     }
2744 
2745     aio_context = bdrv_get_aio_context(bs);
2746     aio_context_acquire(aio_context);
2747 
2748     if (has_granularity) {
2749         if (granularity < 512 || !is_power_of_2(granularity)) {
2750             error_setg(errp, "Granularity must be power of 2 "
2751                              "and at least 512");
2752             goto out;
2753         }
2754     } else {
2755         /* Default to cluster size, if available: */
2756         granularity = bdrv_get_default_bitmap_granularity(bs);
2757     }
2758 
2759     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2760 
2761  out:
2762     aio_context_release(aio_context);
2763 }
2764 
2765 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2766                                    Error **errp)
2767 {
2768     AioContext *aio_context;
2769     BlockDriverState *bs;
2770     BdrvDirtyBitmap *bitmap;
2771 
2772     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2773     if (!bitmap || !bs) {
2774         return;
2775     }
2776 
2777     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2778         error_setg(errp,
2779                    "Bitmap '%s' is currently frozen and cannot be removed",
2780                    name);
2781         goto out;
2782     }
2783     bdrv_dirty_bitmap_make_anon(bitmap);
2784     bdrv_release_dirty_bitmap(bs, bitmap);
2785 
2786  out:
2787     aio_context_release(aio_context);
2788 }
2789 
2790 /**
2791  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2792  * immediately after a full backup operation.
2793  */
2794 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2795                                   Error **errp)
2796 {
2797     AioContext *aio_context;
2798     BdrvDirtyBitmap *bitmap;
2799     BlockDriverState *bs;
2800 
2801     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2802     if (!bitmap || !bs) {
2803         return;
2804     }
2805 
2806     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2807         error_setg(errp,
2808                    "Bitmap '%s' is currently frozen and cannot be modified",
2809                    name);
2810         goto out;
2811     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2812         error_setg(errp,
2813                    "Bitmap '%s' is currently disabled and cannot be cleared",
2814                    name);
2815         goto out;
2816     }
2817 
2818     bdrv_clear_dirty_bitmap(bitmap, NULL);
2819 
2820  out:
2821     aio_context_release(aio_context);
2822 }
2823 
2824 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2825 {
2826     const char *id = qdict_get_str(qdict, "id");
2827     BlockBackend *blk;
2828     BlockDriverState *bs;
2829     AioContext *aio_context;
2830     Error *local_err = NULL;
2831 
2832     bs = bdrv_find_node(id);
2833     if (bs) {
2834         qmp_blockdev_del(id, &local_err);
2835         if (local_err) {
2836             error_report_err(local_err);
2837         }
2838         return;
2839     }
2840 
2841     blk = blk_by_name(id);
2842     if (!blk) {
2843         error_report("Device '%s' not found", id);
2844         return;
2845     }
2846 
2847     if (!blk_legacy_dinfo(blk)) {
2848         error_report("Deleting device added with blockdev-add"
2849                      " is not supported");
2850         return;
2851     }
2852 
2853     aio_context = blk_get_aio_context(blk);
2854     aio_context_acquire(aio_context);
2855 
2856     bs = blk_bs(blk);
2857     if (bs) {
2858         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2859             error_report_err(local_err);
2860             aio_context_release(aio_context);
2861             return;
2862         }
2863 
2864         blk_remove_bs(blk);
2865     }
2866 
2867     /* Make the BlockBackend and the attached BlockDriverState anonymous */
2868     monitor_remove_blk(blk);
2869 
2870     /* If this BlockBackend has a device attached to it, its refcount will be
2871      * decremented when the device is removed; otherwise we have to do so here.
2872      */
2873     if (blk_get_attached_dev(blk)) {
2874         /* Further I/O must not pause the guest */
2875         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2876                          BLOCKDEV_ON_ERROR_REPORT);
2877     } else {
2878         blk_unref(blk);
2879     }
2880 
2881     aio_context_release(aio_context);
2882 }
2883 
2884 void qmp_block_resize(bool has_device, const char *device,
2885                       bool has_node_name, const char *node_name,
2886                       int64_t size, Error **errp)
2887 {
2888     Error *local_err = NULL;
2889     BlockBackend *blk = NULL;
2890     BlockDriverState *bs;
2891     AioContext *aio_context;
2892     int ret;
2893 
2894     bs = bdrv_lookup_bs(has_device ? device : NULL,
2895                         has_node_name ? node_name : NULL,
2896                         &local_err);
2897     if (local_err) {
2898         error_propagate(errp, local_err);
2899         return;
2900     }
2901 
2902     aio_context = bdrv_get_aio_context(bs);
2903     aio_context_acquire(aio_context);
2904 
2905     if (!bdrv_is_first_non_filter(bs)) {
2906         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2907         goto out;
2908     }
2909 
2910     if (size < 0) {
2911         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2912         goto out;
2913     }
2914 
2915     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2916         error_setg(errp, QERR_DEVICE_IN_USE, device);
2917         goto out;
2918     }
2919 
2920     blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
2921     ret = blk_insert_bs(blk, bs, errp);
2922     if (ret < 0) {
2923         goto out;
2924     }
2925 
2926     bdrv_drained_begin(bs);
2927     ret = blk_truncate(blk, size, errp);
2928     bdrv_drained_end(bs);
2929 
2930 out:
2931     blk_unref(blk);
2932     aio_context_release(aio_context);
2933 }
2934 
2935 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2936                       bool has_base, const char *base,
2937                       bool has_base_node, const char *base_node,
2938                       bool has_backing_file, const char *backing_file,
2939                       bool has_speed, int64_t speed,
2940                       bool has_on_error, BlockdevOnError on_error,
2941                       Error **errp)
2942 {
2943     BlockDriverState *bs, *iter;
2944     BlockDriverState *base_bs = NULL;
2945     AioContext *aio_context;
2946     Error *local_err = NULL;
2947     const char *base_name = NULL;
2948 
2949     if (!has_on_error) {
2950         on_error = BLOCKDEV_ON_ERROR_REPORT;
2951     }
2952 
2953     bs = bdrv_lookup_bs(device, device, errp);
2954     if (!bs) {
2955         return;
2956     }
2957 
2958     aio_context = bdrv_get_aio_context(bs);
2959     aio_context_acquire(aio_context);
2960 
2961     if (has_base && has_base_node) {
2962         error_setg(errp, "'base' and 'base-node' cannot be specified "
2963                    "at the same time");
2964         goto out;
2965     }
2966 
2967     if (has_base) {
2968         base_bs = bdrv_find_backing_image(bs, base);
2969         if (base_bs == NULL) {
2970             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2971             goto out;
2972         }
2973         assert(bdrv_get_aio_context(base_bs) == aio_context);
2974         base_name = base;
2975     }
2976 
2977     if (has_base_node) {
2978         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2979         if (!base_bs) {
2980             goto out;
2981         }
2982         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2983             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2984                        base_node, device);
2985             goto out;
2986         }
2987         assert(bdrv_get_aio_context(base_bs) == aio_context);
2988         base_name = base_bs->filename;
2989     }
2990 
2991     /* Check for op blockers in the whole chain between bs and base */
2992     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
2993         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2994             goto out;
2995         }
2996     }
2997 
2998     /* if we are streaming the entire chain, the result will have no backing
2999      * file, and specifying one is therefore an error */
3000     if (base_bs == NULL && has_backing_file) {
3001         error_setg(errp, "backing file specified, but streaming the "
3002                          "entire chain");
3003         goto out;
3004     }
3005 
3006     /* backing_file string overrides base bs filename */
3007     base_name = has_backing_file ? backing_file : base_name;
3008 
3009     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3010                  has_speed ? speed : 0, on_error, &local_err);
3011     if (local_err) {
3012         error_propagate(errp, local_err);
3013         goto out;
3014     }
3015 
3016     trace_qmp_block_stream(bs, bs->job);
3017 
3018 out:
3019     aio_context_release(aio_context);
3020 }
3021 
3022 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3023                       bool has_base, const char *base,
3024                       bool has_top, const char *top,
3025                       bool has_backing_file, const char *backing_file,
3026                       bool has_speed, int64_t speed,
3027                       bool has_filter_node_name, const char *filter_node_name,
3028                       Error **errp)
3029 {
3030     BlockDriverState *bs;
3031     BlockDriverState *iter;
3032     BlockDriverState *base_bs, *top_bs;
3033     AioContext *aio_context;
3034     Error *local_err = NULL;
3035     /* This will be part of the QMP command, if/when the
3036      * BlockdevOnError change for blkmirror makes it in
3037      */
3038     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3039 
3040     if (!has_speed) {
3041         speed = 0;
3042     }
3043     if (!has_filter_node_name) {
3044         filter_node_name = NULL;
3045     }
3046 
3047     /* Important Note:
3048      *  libvirt relies on the DeviceNotFound error class in order to probe for
3049      *  live commit feature versions; for this to work, we must make sure to
3050      *  perform the device lookup before any generic errors that may occur in a
3051      *  scenario in which all optional arguments are omitted. */
3052     bs = qmp_get_root_bs(device, &local_err);
3053     if (!bs) {
3054         bs = bdrv_lookup_bs(device, device, NULL);
3055         if (!bs) {
3056             error_free(local_err);
3057             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3058                       "Device '%s' not found", device);
3059         } else {
3060             error_propagate(errp, local_err);
3061         }
3062         return;
3063     }
3064 
3065     aio_context = bdrv_get_aio_context(bs);
3066     aio_context_acquire(aio_context);
3067 
3068     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3069         goto out;
3070     }
3071 
3072     /* default top_bs is the active layer */
3073     top_bs = bs;
3074 
3075     if (has_top && top) {
3076         if (strcmp(bs->filename, top) != 0) {
3077             top_bs = bdrv_find_backing_image(bs, top);
3078         }
3079     }
3080 
3081     if (top_bs == NULL) {
3082         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3083         goto out;
3084     }
3085 
3086     assert(bdrv_get_aio_context(top_bs) == aio_context);
3087 
3088     if (has_base && base) {
3089         base_bs = bdrv_find_backing_image(top_bs, base);
3090     } else {
3091         base_bs = bdrv_find_base(top_bs);
3092     }
3093 
3094     if (base_bs == NULL) {
3095         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3096         goto out;
3097     }
3098 
3099     assert(bdrv_get_aio_context(base_bs) == aio_context);
3100 
3101     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3102         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3103             goto out;
3104         }
3105     }
3106 
3107     /* Do not allow attempts to commit an image into itself */
3108     if (top_bs == base_bs) {
3109         error_setg(errp, "cannot commit an image into itself");
3110         goto out;
3111     }
3112 
3113     if (top_bs == bs) {
3114         if (has_backing_file) {
3115             error_setg(errp, "'backing-file' specified,"
3116                              " but 'top' is the active layer");
3117             goto out;
3118         }
3119         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3120                             BLOCK_JOB_DEFAULT, speed, on_error,
3121                             filter_node_name, NULL, NULL, false, &local_err);
3122     } else {
3123         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3124         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3125             goto out;
3126         }
3127         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, speed,
3128                      on_error, has_backing_file ? backing_file : NULL,
3129                      filter_node_name, &local_err);
3130     }
3131     if (local_err != NULL) {
3132         error_propagate(errp, local_err);
3133         goto out;
3134     }
3135 
3136 out:
3137     aio_context_release(aio_context);
3138 }
3139 
3140 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
3141                                  Error **errp)
3142 {
3143     BlockDriverState *bs;
3144     BlockDriverState *target_bs;
3145     BlockDriverState *source = NULL;
3146     BlockJob *job = NULL;
3147     BdrvDirtyBitmap *bmap = NULL;
3148     AioContext *aio_context;
3149     QDict *options = NULL;
3150     Error *local_err = NULL;
3151     int flags;
3152     int64_t size;
3153     bool set_backing_hd = false;
3154 
3155     if (!backup->has_speed) {
3156         backup->speed = 0;
3157     }
3158     if (!backup->has_on_source_error) {
3159         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3160     }
3161     if (!backup->has_on_target_error) {
3162         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3163     }
3164     if (!backup->has_mode) {
3165         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3166     }
3167     if (!backup->has_job_id) {
3168         backup->job_id = NULL;
3169     }
3170     if (!backup->has_compress) {
3171         backup->compress = false;
3172     }
3173 
3174     bs = qmp_get_root_bs(backup->device, errp);
3175     if (!bs) {
3176         return NULL;
3177     }
3178 
3179     aio_context = bdrv_get_aio_context(bs);
3180     aio_context_acquire(aio_context);
3181 
3182     if (!backup->has_format) {
3183         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3184                          NULL : (char*) bs->drv->format_name;
3185     }
3186 
3187     /* Early check to avoid creating target */
3188     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3189         goto out;
3190     }
3191 
3192     flags = bs->open_flags | BDRV_O_RDWR;
3193 
3194     /* See if we have a backing HD we can use to create our new image
3195      * on top of. */
3196     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3197         source = backing_bs(bs);
3198         if (!source) {
3199             backup->sync = MIRROR_SYNC_MODE_FULL;
3200         }
3201     }
3202     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3203         source = bs;
3204         flags |= BDRV_O_NO_BACKING;
3205         set_backing_hd = true;
3206     }
3207 
3208     size = bdrv_getlength(bs);
3209     if (size < 0) {
3210         error_setg_errno(errp, -size, "bdrv_getlength failed");
3211         goto out;
3212     }
3213 
3214     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3215         assert(backup->format);
3216         if (source) {
3217             bdrv_img_create(backup->target, backup->format, source->filename,
3218                             source->drv->format_name, NULL,
3219                             size, flags, false, &local_err);
3220         } else {
3221             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3222                             size, flags, false, &local_err);
3223         }
3224     }
3225 
3226     if (local_err) {
3227         error_propagate(errp, local_err);
3228         goto out;
3229     }
3230 
3231     if (backup->format) {
3232         if (!options) {
3233             options = qdict_new();
3234         }
3235         qdict_put_str(options, "driver", backup->format);
3236     }
3237 
3238     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3239     if (!target_bs) {
3240         goto out;
3241     }
3242 
3243     bdrv_set_aio_context(target_bs, aio_context);
3244 
3245     if (set_backing_hd) {
3246         bdrv_set_backing_hd(target_bs, source, &local_err);
3247         if (local_err) {
3248             bdrv_unref(target_bs);
3249             goto out;
3250         }
3251     }
3252 
3253     if (backup->has_bitmap) {
3254         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3255         if (!bmap) {
3256             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3257             bdrv_unref(target_bs);
3258             goto out;
3259         }
3260     }
3261 
3262     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3263                             backup->sync, bmap, backup->compress,
3264                             backup->on_source_error, backup->on_target_error,
3265                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3266     bdrv_unref(target_bs);
3267     if (local_err != NULL) {
3268         error_propagate(errp, local_err);
3269         goto out;
3270     }
3271 
3272 out:
3273     aio_context_release(aio_context);
3274     return job;
3275 }
3276 
3277 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3278 {
3279 
3280     BlockJob *job;
3281     job = do_drive_backup(arg, NULL, errp);
3282     if (job) {
3283         block_job_start(job);
3284     }
3285 }
3286 
3287 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3288 {
3289     return bdrv_named_nodes_list(errp);
3290 }
3291 
3292 BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
3293                              Error **errp)
3294 {
3295     BlockDriverState *bs;
3296     BlockDriverState *target_bs;
3297     Error *local_err = NULL;
3298     AioContext *aio_context;
3299     BlockJob *job = NULL;
3300 
3301     if (!backup->has_speed) {
3302         backup->speed = 0;
3303     }
3304     if (!backup->has_on_source_error) {
3305         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3306     }
3307     if (!backup->has_on_target_error) {
3308         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3309     }
3310     if (!backup->has_job_id) {
3311         backup->job_id = NULL;
3312     }
3313     if (!backup->has_compress) {
3314         backup->compress = false;
3315     }
3316 
3317     bs = qmp_get_root_bs(backup->device, errp);
3318     if (!bs) {
3319         return NULL;
3320     }
3321 
3322     aio_context = bdrv_get_aio_context(bs);
3323     aio_context_acquire(aio_context);
3324 
3325     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3326     if (!target_bs) {
3327         goto out;
3328     }
3329 
3330     if (bdrv_get_aio_context(target_bs) != aio_context) {
3331         if (!bdrv_has_blk(target_bs)) {
3332             /* The target BDS is not attached, we can safely move it to another
3333              * AioContext. */
3334             bdrv_set_aio_context(target_bs, aio_context);
3335         } else {
3336             error_setg(errp, "Target is attached to a different thread from "
3337                              "source.");
3338             goto out;
3339         }
3340     }
3341     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3342                             backup->sync, NULL, backup->compress,
3343                             backup->on_source_error, backup->on_target_error,
3344                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3345     if (local_err != NULL) {
3346         error_propagate(errp, local_err);
3347     }
3348 out:
3349     aio_context_release(aio_context);
3350     return job;
3351 }
3352 
3353 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3354 {
3355     BlockJob *job;
3356     job = do_blockdev_backup(arg, NULL, errp);
3357     if (job) {
3358         block_job_start(job);
3359     }
3360 }
3361 
3362 /* Parameter check and block job starting for drive mirroring.
3363  * Caller should hold @device and @target's aio context (must be the same).
3364  **/
3365 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3366                                    BlockDriverState *target,
3367                                    bool has_replaces, const char *replaces,
3368                                    enum MirrorSyncMode sync,
3369                                    BlockMirrorBackingMode backing_mode,
3370                                    bool has_speed, int64_t speed,
3371                                    bool has_granularity, uint32_t granularity,
3372                                    bool has_buf_size, int64_t buf_size,
3373                                    bool has_on_source_error,
3374                                    BlockdevOnError on_source_error,
3375                                    bool has_on_target_error,
3376                                    BlockdevOnError on_target_error,
3377                                    bool has_unmap, bool unmap,
3378                                    bool has_filter_node_name,
3379                                    const char *filter_node_name,
3380                                    Error **errp)
3381 {
3382 
3383     if (!has_speed) {
3384         speed = 0;
3385     }
3386     if (!has_on_source_error) {
3387         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3388     }
3389     if (!has_on_target_error) {
3390         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3391     }
3392     if (!has_granularity) {
3393         granularity = 0;
3394     }
3395     if (!has_buf_size) {
3396         buf_size = 0;
3397     }
3398     if (!has_unmap) {
3399         unmap = true;
3400     }
3401     if (!has_filter_node_name) {
3402         filter_node_name = NULL;
3403     }
3404 
3405     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3406         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3407                    "a value in range [512B, 64MB]");
3408         return;
3409     }
3410     if (granularity & (granularity - 1)) {
3411         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3412                    "power of 2");
3413         return;
3414     }
3415 
3416     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3417         return;
3418     }
3419     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3420         return;
3421     }
3422 
3423     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3424         sync = MIRROR_SYNC_MODE_FULL;
3425     }
3426 
3427     /* pass the node name to replace to mirror start since it's loose coupling
3428      * and will allow to check whether the node still exist at mirror completion
3429      */
3430     mirror_start(job_id, bs, target,
3431                  has_replaces ? replaces : NULL,
3432                  speed, granularity, buf_size, sync, backing_mode,
3433                  on_source_error, on_target_error, unmap, filter_node_name,
3434                  errp);
3435 }
3436 
3437 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3438 {
3439     BlockDriverState *bs;
3440     BlockDriverState *source, *target_bs;
3441     AioContext *aio_context;
3442     BlockMirrorBackingMode backing_mode;
3443     Error *local_err = NULL;
3444     QDict *options = NULL;
3445     int flags;
3446     int64_t size;
3447     const char *format = arg->format;
3448 
3449     bs = qmp_get_root_bs(arg->device, errp);
3450     if (!bs) {
3451         return;
3452     }
3453 
3454     aio_context = bdrv_get_aio_context(bs);
3455     aio_context_acquire(aio_context);
3456 
3457     if (!arg->has_mode) {
3458         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3459     }
3460 
3461     if (!arg->has_format) {
3462         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3463                   ? NULL : bs->drv->format_name);
3464     }
3465 
3466     flags = bs->open_flags | BDRV_O_RDWR;
3467     source = backing_bs(bs);
3468     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3469         arg->sync = MIRROR_SYNC_MODE_FULL;
3470     }
3471     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3472         source = bs;
3473     }
3474 
3475     size = bdrv_getlength(bs);
3476     if (size < 0) {
3477         error_setg_errno(errp, -size, "bdrv_getlength failed");
3478         goto out;
3479     }
3480 
3481     if (arg->has_replaces) {
3482         BlockDriverState *to_replace_bs;
3483         AioContext *replace_aio_context;
3484         int64_t replace_size;
3485 
3486         if (!arg->has_node_name) {
3487             error_setg(errp, "a node-name must be provided when replacing a"
3488                              " named node of the graph");
3489             goto out;
3490         }
3491 
3492         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3493 
3494         if (!to_replace_bs) {
3495             error_propagate(errp, local_err);
3496             goto out;
3497         }
3498 
3499         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3500         aio_context_acquire(replace_aio_context);
3501         replace_size = bdrv_getlength(to_replace_bs);
3502         aio_context_release(replace_aio_context);
3503 
3504         if (size != replace_size) {
3505             error_setg(errp, "cannot replace image with a mirror image of "
3506                              "different size");
3507             goto out;
3508         }
3509     }
3510 
3511     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3512         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3513     } else {
3514         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3515     }
3516 
3517     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3518         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3519     {
3520         /* create new image w/o backing file */
3521         assert(format);
3522         bdrv_img_create(arg->target, format,
3523                         NULL, NULL, NULL, size, flags, false, &local_err);
3524     } else {
3525         switch (arg->mode) {
3526         case NEW_IMAGE_MODE_EXISTING:
3527             break;
3528         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3529             /* create new image with backing file */
3530             bdrv_img_create(arg->target, format,
3531                             source->filename,
3532                             source->drv->format_name,
3533                             NULL, size, flags, false, &local_err);
3534             break;
3535         default:
3536             abort();
3537         }
3538     }
3539 
3540     if (local_err) {
3541         error_propagate(errp, local_err);
3542         goto out;
3543     }
3544 
3545     options = qdict_new();
3546     if (arg->has_node_name) {
3547         qdict_put_str(options, "node-name", arg->node_name);
3548     }
3549     if (format) {
3550         qdict_put_str(options, "driver", format);
3551     }
3552 
3553     /* Mirroring takes care of copy-on-write using the source's backing
3554      * file.
3555      */
3556     target_bs = bdrv_open(arg->target, NULL, options,
3557                           flags | BDRV_O_NO_BACKING, errp);
3558     if (!target_bs) {
3559         goto out;
3560     }
3561 
3562     bdrv_set_aio_context(target_bs, aio_context);
3563 
3564     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3565                            arg->has_replaces, arg->replaces, arg->sync,
3566                            backing_mode, arg->has_speed, arg->speed,
3567                            arg->has_granularity, arg->granularity,
3568                            arg->has_buf_size, arg->buf_size,
3569                            arg->has_on_source_error, arg->on_source_error,
3570                            arg->has_on_target_error, arg->on_target_error,
3571                            arg->has_unmap, arg->unmap,
3572                            false, NULL,
3573                            &local_err);
3574     bdrv_unref(target_bs);
3575     error_propagate(errp, local_err);
3576 out:
3577     aio_context_release(aio_context);
3578 }
3579 
3580 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3581                          const char *device, const char *target,
3582                          bool has_replaces, const char *replaces,
3583                          MirrorSyncMode sync,
3584                          bool has_speed, int64_t speed,
3585                          bool has_granularity, uint32_t granularity,
3586                          bool has_buf_size, int64_t buf_size,
3587                          bool has_on_source_error,
3588                          BlockdevOnError on_source_error,
3589                          bool has_on_target_error,
3590                          BlockdevOnError on_target_error,
3591                          bool has_filter_node_name,
3592                          const char *filter_node_name,
3593                          Error **errp)
3594 {
3595     BlockDriverState *bs;
3596     BlockDriverState *target_bs;
3597     AioContext *aio_context;
3598     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3599     Error *local_err = NULL;
3600 
3601     bs = qmp_get_root_bs(device, errp);
3602     if (!bs) {
3603         return;
3604     }
3605 
3606     target_bs = bdrv_lookup_bs(target, target, errp);
3607     if (!target_bs) {
3608         return;
3609     }
3610 
3611     aio_context = bdrv_get_aio_context(bs);
3612     aio_context_acquire(aio_context);
3613 
3614     bdrv_set_aio_context(target_bs, aio_context);
3615 
3616     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3617                            has_replaces, replaces, sync, backing_mode,
3618                            has_speed, speed,
3619                            has_granularity, granularity,
3620                            has_buf_size, buf_size,
3621                            has_on_source_error, on_source_error,
3622                            has_on_target_error, on_target_error,
3623                            true, true,
3624                            has_filter_node_name, filter_node_name,
3625                            &local_err);
3626     error_propagate(errp, local_err);
3627 
3628     aio_context_release(aio_context);
3629 }
3630 
3631 /* Get a block job using its ID and acquire its AioContext */
3632 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3633                                 Error **errp)
3634 {
3635     BlockJob *job;
3636 
3637     assert(id != NULL);
3638 
3639     *aio_context = NULL;
3640 
3641     job = block_job_get(id);
3642 
3643     if (!job) {
3644         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3645                   "Block job '%s' not found", id);
3646         return NULL;
3647     }
3648 
3649     *aio_context = blk_get_aio_context(job->blk);
3650     aio_context_acquire(*aio_context);
3651 
3652     return job;
3653 }
3654 
3655 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3656 {
3657     AioContext *aio_context;
3658     BlockJob *job = find_block_job(device, &aio_context, errp);
3659 
3660     if (!job) {
3661         return;
3662     }
3663 
3664     block_job_set_speed(job, speed, errp);
3665     aio_context_release(aio_context);
3666 }
3667 
3668 void qmp_block_job_cancel(const char *device,
3669                           bool has_force, bool force, Error **errp)
3670 {
3671     AioContext *aio_context;
3672     BlockJob *job = find_block_job(device, &aio_context, errp);
3673 
3674     if (!job) {
3675         return;
3676     }
3677 
3678     if (!has_force) {
3679         force = false;
3680     }
3681 
3682     if (block_job_user_paused(job) && !force) {
3683         error_setg(errp, "The block job for device '%s' is currently paused",
3684                    device);
3685         goto out;
3686     }
3687 
3688     trace_qmp_block_job_cancel(job);
3689     block_job_cancel(job);
3690 out:
3691     aio_context_release(aio_context);
3692 }
3693 
3694 void qmp_block_job_pause(const char *device, Error **errp)
3695 {
3696     AioContext *aio_context;
3697     BlockJob *job = find_block_job(device, &aio_context, errp);
3698 
3699     if (!job || block_job_user_paused(job)) {
3700         return;
3701     }
3702 
3703     trace_qmp_block_job_pause(job);
3704     block_job_user_pause(job);
3705     aio_context_release(aio_context);
3706 }
3707 
3708 void qmp_block_job_resume(const char *device, Error **errp)
3709 {
3710     AioContext *aio_context;
3711     BlockJob *job = find_block_job(device, &aio_context, errp);
3712 
3713     if (!job || !block_job_user_paused(job)) {
3714         return;
3715     }
3716 
3717     trace_qmp_block_job_resume(job);
3718     block_job_iostatus_reset(job);
3719     block_job_user_resume(job);
3720     aio_context_release(aio_context);
3721 }
3722 
3723 void qmp_block_job_complete(const char *device, Error **errp)
3724 {
3725     AioContext *aio_context;
3726     BlockJob *job = find_block_job(device, &aio_context, errp);
3727 
3728     if (!job) {
3729         return;
3730     }
3731 
3732     trace_qmp_block_job_complete(job);
3733     block_job_complete(job, errp);
3734     aio_context_release(aio_context);
3735 }
3736 
3737 void qmp_change_backing_file(const char *device,
3738                              const char *image_node_name,
3739                              const char *backing_file,
3740                              Error **errp)
3741 {
3742     BlockDriverState *bs = NULL;
3743     AioContext *aio_context;
3744     BlockDriverState *image_bs = NULL;
3745     Error *local_err = NULL;
3746     bool ro;
3747     int open_flags;
3748     int ret;
3749 
3750     bs = qmp_get_root_bs(device, errp);
3751     if (!bs) {
3752         return;
3753     }
3754 
3755     aio_context = bdrv_get_aio_context(bs);
3756     aio_context_acquire(aio_context);
3757 
3758     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3759     if (local_err) {
3760         error_propagate(errp, local_err);
3761         goto out;
3762     }
3763 
3764     if (!image_bs) {
3765         error_setg(errp, "image file not found");
3766         goto out;
3767     }
3768 
3769     if (bdrv_find_base(image_bs) == image_bs) {
3770         error_setg(errp, "not allowing backing file change on an image "
3771                          "without a backing file");
3772         goto out;
3773     }
3774 
3775     /* even though we are not necessarily operating on bs, we need it to
3776      * determine if block ops are currently prohibited on the chain */
3777     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3778         goto out;
3779     }
3780 
3781     /* final sanity check */
3782     if (!bdrv_chain_contains(bs, image_bs)) {
3783         error_setg(errp, "'%s' and image file are not in the same chain",
3784                    device);
3785         goto out;
3786     }
3787 
3788     /* if not r/w, reopen to make r/w */
3789     open_flags = image_bs->open_flags;
3790     ro = bdrv_is_read_only(image_bs);
3791 
3792     if (ro) {
3793         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3794         if (local_err) {
3795             error_propagate(errp, local_err);
3796             goto out;
3797         }
3798     }
3799 
3800     ret = bdrv_change_backing_file(image_bs, backing_file,
3801                                image_bs->drv ? image_bs->drv->format_name : "");
3802 
3803     if (ret < 0) {
3804         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3805                          backing_file);
3806         /* don't exit here, so we can try to restore open flags if
3807          * appropriate */
3808     }
3809 
3810     if (ro) {
3811         bdrv_reopen(image_bs, open_flags, &local_err);
3812         error_propagate(errp, local_err);
3813     }
3814 
3815 out:
3816     aio_context_release(aio_context);
3817 }
3818 
3819 void hmp_drive_add_node(Monitor *mon, const char *optstr)
3820 {
3821     QemuOpts *opts;
3822     QDict *qdict;
3823     Error *local_err = NULL;
3824 
3825     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
3826     if (!opts) {
3827         return;
3828     }
3829 
3830     qdict = qemu_opts_to_qdict(opts, NULL);
3831 
3832     if (!qdict_get_try_str(qdict, "node-name")) {
3833         QDECREF(qdict);
3834         error_report("'node-name' needs to be specified");
3835         goto out;
3836     }
3837 
3838     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
3839     if (!bs) {
3840         error_report_err(local_err);
3841         goto out;
3842     }
3843 
3844     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3845 
3846 out:
3847     qemu_opts_del(opts);
3848 }
3849 
3850 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3851 {
3852     BlockDriverState *bs;
3853     QObject *obj;
3854     Visitor *v = qobject_output_visitor_new(&obj);
3855     QDict *qdict;
3856     Error *local_err = NULL;
3857 
3858     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
3859     if (local_err) {
3860         error_propagate(errp, local_err);
3861         goto fail;
3862     }
3863 
3864     visit_complete(v, &obj);
3865     qdict = qobject_to_qdict(obj);
3866 
3867     qdict_flatten(qdict);
3868 
3869     if (!qdict_get_try_str(qdict, "node-name")) {
3870         error_setg(errp, "'node-name' must be specified for the root node");
3871         goto fail;
3872     }
3873 
3874     bs = bds_tree_init(qdict, errp);
3875     if (!bs) {
3876         goto fail;
3877     }
3878 
3879     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3880 
3881     if (bs && bdrv_key_required(bs)) {
3882         QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3883         bdrv_unref(bs);
3884         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3885         goto fail;
3886     }
3887 
3888 fail:
3889     visit_free(v);
3890 }
3891 
3892 void qmp_blockdev_del(const char *node_name, Error **errp)
3893 {
3894     AioContext *aio_context;
3895     BlockDriverState *bs;
3896 
3897     bs = bdrv_find_node(node_name);
3898     if (!bs) {
3899         error_setg(errp, "Cannot find node %s", node_name);
3900         return;
3901     }
3902     if (bdrv_has_blk(bs)) {
3903         error_setg(errp, "Node %s is in use", node_name);
3904         return;
3905     }
3906     aio_context = bdrv_get_aio_context(bs);
3907     aio_context_acquire(aio_context);
3908 
3909     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3910         goto out;
3911     }
3912 
3913     if (!bs->monitor_list.tqe_prev) {
3914         error_setg(errp, "Node %s is not owned by the monitor",
3915                    bs->node_name);
3916         goto out;
3917     }
3918 
3919     if (bs->refcnt > 1) {
3920         error_setg(errp, "Block device %s is in use",
3921                    bdrv_get_device_or_node_name(bs));
3922         goto out;
3923     }
3924 
3925     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3926     bdrv_unref(bs);
3927 
3928 out:
3929     aio_context_release(aio_context);
3930 }
3931 
3932 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3933                                   const char *child_name)
3934 {
3935     BdrvChild *child;
3936 
3937     QLIST_FOREACH(child, &parent_bs->children, next) {
3938         if (strcmp(child->name, child_name) == 0) {
3939             return child;
3940         }
3941     }
3942 
3943     return NULL;
3944 }
3945 
3946 void qmp_x_blockdev_change(const char *parent, bool has_child,
3947                            const char *child, bool has_node,
3948                            const char *node, Error **errp)
3949 {
3950     BlockDriverState *parent_bs, *new_bs = NULL;
3951     BdrvChild *p_child;
3952 
3953     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3954     if (!parent_bs) {
3955         return;
3956     }
3957 
3958     if (has_child == has_node) {
3959         if (has_child) {
3960             error_setg(errp, "The parameters child and node are in conflict");
3961         } else {
3962             error_setg(errp, "Either child or node must be specified");
3963         }
3964         return;
3965     }
3966 
3967     if (has_child) {
3968         p_child = bdrv_find_child(parent_bs, child);
3969         if (!p_child) {
3970             error_setg(errp, "Node '%s' does not have child '%s'",
3971                        parent, child);
3972             return;
3973         }
3974         bdrv_del_child(parent_bs, p_child, errp);
3975     }
3976 
3977     if (has_node) {
3978         new_bs = bdrv_find_node(node);
3979         if (!new_bs) {
3980             error_setg(errp, "Node '%s' not found", node);
3981             return;
3982         }
3983         bdrv_add_child(parent_bs, new_bs, errp);
3984     }
3985 }
3986 
3987 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3988 {
3989     BlockJobInfoList *head = NULL, **p_next = &head;
3990     BlockJob *job;
3991 
3992     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
3993         BlockJobInfoList *elem;
3994         AioContext *aio_context;
3995 
3996         if (block_job_is_internal(job)) {
3997             continue;
3998         }
3999         elem = g_new0(BlockJobInfoList, 1);
4000         aio_context = blk_get_aio_context(job->blk);
4001         aio_context_acquire(aio_context);
4002         elem->value = block_job_query(job, errp);
4003         aio_context_release(aio_context);
4004         if (!elem->value) {
4005             g_free(elem);
4006             qapi_free_BlockJobInfoList(head);
4007             return NULL;
4008         }
4009         *p_next = elem;
4010         p_next = &elem->next;
4011     }
4012 
4013     return head;
4014 }
4015 
4016 QemuOptsList qemu_common_drive_opts = {
4017     .name = "drive",
4018     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4019     .desc = {
4020         {
4021             .name = "snapshot",
4022             .type = QEMU_OPT_BOOL,
4023             .help = "enable/disable snapshot mode",
4024         },{
4025             .name = "aio",
4026             .type = QEMU_OPT_STRING,
4027             .help = "host AIO implementation (threads, native)",
4028         },{
4029             .name = BDRV_OPT_CACHE_WB,
4030             .type = QEMU_OPT_BOOL,
4031             .help = "Enable writeback mode",
4032         },{
4033             .name = "format",
4034             .type = QEMU_OPT_STRING,
4035             .help = "disk format (raw, qcow2, ...)",
4036         },{
4037             .name = "rerror",
4038             .type = QEMU_OPT_STRING,
4039             .help = "read error action",
4040         },{
4041             .name = "werror",
4042             .type = QEMU_OPT_STRING,
4043             .help = "write error action",
4044         },{
4045             .name = BDRV_OPT_READ_ONLY,
4046             .type = QEMU_OPT_BOOL,
4047             .help = "open drive file as read-only",
4048         },
4049 
4050         THROTTLE_OPTS,
4051 
4052         {
4053             .name = "throttling.group",
4054             .type = QEMU_OPT_STRING,
4055             .help = "name of the block throttling group",
4056         },{
4057             .name = "copy-on-read",
4058             .type = QEMU_OPT_BOOL,
4059             .help = "copy read data from backing file into image file",
4060         },{
4061             .name = "detect-zeroes",
4062             .type = QEMU_OPT_STRING,
4063             .help = "try to optimize zero writes (off, on, unmap)",
4064         },{
4065             .name = "stats-account-invalid",
4066             .type = QEMU_OPT_BOOL,
4067             .help = "whether to account for invalid I/O operations "
4068                     "in the statistics",
4069         },{
4070             .name = "stats-account-failed",
4071             .type = QEMU_OPT_BOOL,
4072             .help = "whether to account for failed I/O operations "
4073                     "in the statistics",
4074         },
4075         { /* end of list */ }
4076     },
4077 };
4078 
4079 QemuOptsList qemu_drive_opts = {
4080     .name = "drive",
4081     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4082     .desc = {
4083         /*
4084          * no elements => accept any params
4085          * validation will happen later
4086          */
4087         { /* end of list */ }
4088     },
4089 };
4090