xref: /openbmc/qemu/blockdev.c (revision 14b6d44d4720681a57b5d2c58cabdfc6364f8263)
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/qmp-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 "trace.h"
52 #include "sysemu/arch_init.h"
53 
54 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
55     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
56 
57 static const char *const if_name[IF_COUNT] = {
58     [IF_NONE] = "none",
59     [IF_IDE] = "ide",
60     [IF_SCSI] = "scsi",
61     [IF_FLOPPY] = "floppy",
62     [IF_PFLASH] = "pflash",
63     [IF_MTD] = "mtd",
64     [IF_SD] = "sd",
65     [IF_VIRTIO] = "virtio",
66     [IF_XEN] = "xen",
67 };
68 
69 static int if_max_devs[IF_COUNT] = {
70     /*
71      * Do not change these numbers!  They govern how drive option
72      * index maps to unit and bus.  That mapping is ABI.
73      *
74      * All controllers used to imlement if=T drives need to support
75      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
76      * Otherwise, some index values map to "impossible" bus, unit
77      * values.
78      *
79      * For instance, if you change [IF_SCSI] to 255, -drive
80      * if=scsi,index=12 no longer means bus=1,unit=5, but
81      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
82      * the drive can't be set up.  Regression.
83      */
84     [IF_IDE] = 2,
85     [IF_SCSI] = 7,
86 };
87 
88 /**
89  * Boards may call this to offer board-by-board overrides
90  * of the default, global values.
91  */
92 void override_max_devs(BlockInterfaceType type, int max_devs)
93 {
94     BlockBackend *blk;
95     DriveInfo *dinfo;
96 
97     if (max_devs <= 0) {
98         return;
99     }
100 
101     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
102         dinfo = blk_legacy_dinfo(blk);
103         if (dinfo->type == type) {
104             fprintf(stderr, "Cannot override units-per-bus property of"
105                     " the %s interface, because a drive of that type has"
106                     " already been added.\n", if_name[type]);
107             g_assert_not_reached();
108         }
109     }
110 
111     if_max_devs[type] = max_devs;
112 }
113 
114 /*
115  * We automatically delete the drive when a device using it gets
116  * unplugged.  Questionable feature, but we can't just drop it.
117  * Device models call blockdev_mark_auto_del() to schedule the
118  * automatic deletion, and generic qdev code calls blockdev_auto_del()
119  * when deletion is actually safe.
120  */
121 void blockdev_mark_auto_del(BlockBackend *blk)
122 {
123     DriveInfo *dinfo = blk_legacy_dinfo(blk);
124     BlockDriverState *bs = blk_bs(blk);
125     AioContext *aio_context;
126 
127     if (!dinfo) {
128         return;
129     }
130 
131     if (bs) {
132         aio_context = bdrv_get_aio_context(bs);
133         aio_context_acquire(aio_context);
134 
135         if (bs->job) {
136             block_job_cancel(bs->job);
137         }
138 
139         aio_context_release(aio_context);
140     }
141 
142     dinfo->auto_del = 1;
143 }
144 
145 void blockdev_auto_del(BlockBackend *blk)
146 {
147     DriveInfo *dinfo = blk_legacy_dinfo(blk);
148 
149     if (dinfo && dinfo->auto_del) {
150         monitor_remove_blk(blk);
151         blk_unref(blk);
152     }
153 }
154 
155 /**
156  * Returns the current mapping of how many units per bus
157  * a particular interface can support.
158  *
159  *  A positive integer indicates n units per bus.
160  *  0 implies the mapping has not been established.
161  * -1 indicates an invalid BlockInterfaceType was given.
162  */
163 int drive_get_max_devs(BlockInterfaceType type)
164 {
165     if (type >= IF_IDE && type < IF_COUNT) {
166         return if_max_devs[type];
167     }
168 
169     return -1;
170 }
171 
172 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
173 {
174     int max_devs = if_max_devs[type];
175     return max_devs ? index / max_devs : 0;
176 }
177 
178 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
179 {
180     int max_devs = if_max_devs[type];
181     return max_devs ? index % max_devs : index;
182 }
183 
184 QemuOpts *drive_def(const char *optstr)
185 {
186     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
187 }
188 
189 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
190                     const char *optstr)
191 {
192     QemuOpts *opts;
193 
194     opts = drive_def(optstr);
195     if (!opts) {
196         return NULL;
197     }
198     if (type != IF_DEFAULT) {
199         qemu_opt_set(opts, "if", if_name[type], &error_abort);
200     }
201     if (index >= 0) {
202         qemu_opt_set_number(opts, "index", index, &error_abort);
203     }
204     if (file)
205         qemu_opt_set(opts, "file", file, &error_abort);
206     return opts;
207 }
208 
209 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
210 {
211     BlockBackend *blk;
212     DriveInfo *dinfo;
213 
214     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
215         dinfo = blk_legacy_dinfo(blk);
216         if (dinfo && dinfo->type == type
217             && dinfo->bus == bus && dinfo->unit == unit) {
218             return dinfo;
219         }
220     }
221 
222     return NULL;
223 }
224 
225 bool drive_check_orphaned(void)
226 {
227     BlockBackend *blk;
228     DriveInfo *dinfo;
229     bool rs = false;
230 
231     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
232         dinfo = blk_legacy_dinfo(blk);
233         /* If dinfo->bdrv->dev is NULL, it has no device attached. */
234         /* Unless this is a default drive, this may be an oversight. */
235         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
236             dinfo->type != IF_NONE) {
237             fprintf(stderr, "Warning: Orphaned drive without device: "
238                     "id=%s,file=%s,if=%s,bus=%d,unit=%d\n",
239                     blk_name(blk), blk_bs(blk) ? blk_bs(blk)->filename : "",
240                     if_name[dinfo->type], dinfo->bus, dinfo->unit);
241             rs = true;
242         }
243     }
244 
245     return rs;
246 }
247 
248 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
249 {
250     return drive_get(type,
251                      drive_index_to_bus_id(type, index),
252                      drive_index_to_unit_id(type, index));
253 }
254 
255 int drive_get_max_bus(BlockInterfaceType type)
256 {
257     int max_bus;
258     BlockBackend *blk;
259     DriveInfo *dinfo;
260 
261     max_bus = -1;
262     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
263         dinfo = blk_legacy_dinfo(blk);
264         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
265             max_bus = dinfo->bus;
266         }
267     }
268     return max_bus;
269 }
270 
271 /* Get a block device.  This should only be used for single-drive devices
272    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
273    appropriate bus.  */
274 DriveInfo *drive_get_next(BlockInterfaceType type)
275 {
276     static int next_block_unit[IF_COUNT];
277 
278     return drive_get(type, 0, next_block_unit[type]++);
279 }
280 
281 static void bdrv_format_print(void *opaque, const char *name)
282 {
283     error_printf(" %s", name);
284 }
285 
286 typedef struct {
287     QEMUBH *bh;
288     BlockDriverState *bs;
289 } BDRVPutRefBH;
290 
291 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
292 {
293     if (!strcmp(buf, "ignore")) {
294         return BLOCKDEV_ON_ERROR_IGNORE;
295     } else if (!is_read && !strcmp(buf, "enospc")) {
296         return BLOCKDEV_ON_ERROR_ENOSPC;
297     } else if (!strcmp(buf, "stop")) {
298         return BLOCKDEV_ON_ERROR_STOP;
299     } else if (!strcmp(buf, "report")) {
300         return BLOCKDEV_ON_ERROR_REPORT;
301     } else {
302         error_setg(errp, "'%s' invalid %s error action",
303                    buf, is_read ? "read" : "write");
304         return -1;
305     }
306 }
307 
308 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
309                                   Error **errp)
310 {
311     const QListEntry *entry;
312     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
313         switch (qobject_type(entry->value)) {
314 
315         case QTYPE_QSTRING: {
316             unsigned long long length;
317             const char *str = qstring_get_str(qobject_to_qstring(entry->value));
318             if (parse_uint_full(str, &length, 10) == 0 &&
319                 length > 0 && length <= UINT_MAX) {
320                 block_acct_add_interval(stats, (unsigned) length);
321             } else {
322                 error_setg(errp, "Invalid interval length: %s", str);
323                 return false;
324             }
325             break;
326         }
327 
328         case QTYPE_QINT: {
329             int64_t length = qint_get_int(qobject_to_qint(entry->value));
330             if (length > 0 && length <= UINT_MAX) {
331                 block_acct_add_interval(stats, (unsigned) length);
332             } else {
333                 error_setg(errp, "Invalid interval length: %" PRId64, length);
334                 return false;
335             }
336             break;
337         }
338 
339         default:
340             error_setg(errp, "The specification of stats-intervals is invalid");
341             return false;
342         }
343     }
344     return true;
345 }
346 
347 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
348 
349 /* All parameters but @opts are optional and may be set to NULL. */
350 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
351     const char **throttling_group, ThrottleConfig *throttle_cfg,
352     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
353 {
354     const char *discard;
355     Error *local_error = NULL;
356     const char *aio;
357 
358     if (bdrv_flags) {
359         if (!qemu_opt_get_bool(opts, "read-only", false)) {
360             *bdrv_flags |= BDRV_O_RDWR;
361         }
362         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
363             *bdrv_flags |= BDRV_O_COPY_ON_READ;
364         }
365 
366         if ((discard = qemu_opt_get(opts, "discard")) != NULL) {
367             if (bdrv_parse_discard_flags(discard, bdrv_flags) != 0) {
368                 error_setg(errp, "Invalid discard option");
369                 return;
370             }
371         }
372 
373         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
374             if (!strcmp(aio, "native")) {
375                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
376             } else if (!strcmp(aio, "threads")) {
377                 /* this is the default */
378             } else {
379                error_setg(errp, "invalid aio option");
380                return;
381             }
382         }
383     }
384 
385     /* disk I/O throttling */
386     if (throttling_group) {
387         *throttling_group = qemu_opt_get(opts, "throttling.group");
388     }
389 
390     if (throttle_cfg) {
391         throttle_config_init(throttle_cfg);
392         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
393             qemu_opt_get_number(opts, "throttling.bps-total", 0);
394         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
395             qemu_opt_get_number(opts, "throttling.bps-read", 0);
396         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
397             qemu_opt_get_number(opts, "throttling.bps-write", 0);
398         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
399             qemu_opt_get_number(opts, "throttling.iops-total", 0);
400         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
401             qemu_opt_get_number(opts, "throttling.iops-read", 0);
402         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
403             qemu_opt_get_number(opts, "throttling.iops-write", 0);
404 
405         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
406             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
407         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
408             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
409         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
410             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
411         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
412             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
413         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
414             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
415         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
416             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
417 
418         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
419             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
420         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
421             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
422         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
423             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
424         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
425             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
426         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
427             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
428         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
429             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
430 
431         throttle_cfg->op_size =
432             qemu_opt_get_number(opts, "throttling.iops-size", 0);
433 
434         if (!throttle_is_valid(throttle_cfg, errp)) {
435             return;
436         }
437     }
438 
439     if (detect_zeroes) {
440         *detect_zeroes =
441             qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
442                             qemu_opt_get(opts, "detect-zeroes"),
443                             BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
444                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
445                             &local_error);
446         if (local_error) {
447             error_propagate(errp, local_error);
448             return;
449         }
450 
451         if (bdrv_flags &&
452             *detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
453             !(*bdrv_flags & BDRV_O_UNMAP))
454         {
455             error_setg(errp, "setting detect-zeroes to unmap is not allowed "
456                              "without setting discard operation to unmap");
457             return;
458         }
459     }
460 }
461 
462 /* Takes the ownership of bs_opts */
463 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
464                                    Error **errp)
465 {
466     const char *buf;
467     int bdrv_flags = 0;
468     int on_read_error, on_write_error;
469     bool account_invalid, account_failed;
470     BlockBackend *blk;
471     BlockDriverState *bs;
472     ThrottleConfig cfg;
473     int snapshot = 0;
474     Error *error = NULL;
475     QemuOpts *opts;
476     QDict *interval_dict = NULL;
477     QList *interval_list = NULL;
478     const char *id;
479     BlockdevDetectZeroesOptions detect_zeroes =
480         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
481     const char *throttling_group = NULL;
482 
483     /* Check common options by copying from bs_opts to opts, all other options
484      * stay in bs_opts for processing by bdrv_open(). */
485     id = qdict_get_try_str(bs_opts, "id");
486     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
487     if (error) {
488         error_propagate(errp, error);
489         goto err_no_opts;
490     }
491 
492     qemu_opts_absorb_qdict(opts, bs_opts, &error);
493     if (error) {
494         error_propagate(errp, error);
495         goto early_err;
496     }
497 
498     if (id) {
499         qdict_del(bs_opts, "id");
500     }
501 
502     /* extract parameters */
503     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
504 
505     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
506     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
507 
508     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
509     qdict_array_split(interval_dict, &interval_list);
510 
511     if (qdict_size(interval_dict) != 0) {
512         error_setg(errp, "Invalid option stats-intervals.%s",
513                    qdict_first(interval_dict)->key);
514         goto early_err;
515     }
516 
517     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
518                                     &detect_zeroes, &error);
519     if (error) {
520         error_propagate(errp, error);
521         goto early_err;
522     }
523 
524     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
525         if (is_help_option(buf)) {
526             error_printf("Supported formats:");
527             bdrv_iterate_format(bdrv_format_print, NULL);
528             error_printf("\n");
529             goto early_err;
530         }
531 
532         if (qdict_haskey(bs_opts, "driver")) {
533             error_setg(errp, "Cannot specify both 'driver' and 'format'");
534             goto early_err;
535         }
536         qdict_put(bs_opts, "driver", qstring_from_str(buf));
537     }
538 
539     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
540     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
541         on_write_error = parse_block_error_action(buf, 0, &error);
542         if (error) {
543             error_propagate(errp, error);
544             goto early_err;
545         }
546     }
547 
548     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
549     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
550         on_read_error = parse_block_error_action(buf, 1, &error);
551         if (error) {
552             error_propagate(errp, error);
553             goto early_err;
554         }
555     }
556 
557     if (snapshot) {
558         bdrv_flags |= BDRV_O_SNAPSHOT;
559     }
560 
561     /* init */
562     if ((!file || !*file) && !qdict_size(bs_opts)) {
563         BlockBackendRootState *blk_rs;
564 
565         blk = blk_new(errp);
566         if (!blk) {
567             goto early_err;
568         }
569 
570         blk_rs = blk_get_root_state(blk);
571         blk_rs->open_flags    = bdrv_flags;
572         blk_rs->read_only     = !(bdrv_flags & BDRV_O_RDWR);
573         blk_rs->detect_zeroes = detect_zeroes;
574 
575         if (throttle_enabled(&cfg)) {
576             if (!throttling_group) {
577                 throttling_group = blk_name(blk);
578             }
579             blk_rs->throttle_group = g_strdup(throttling_group);
580             blk_rs->throttle_state = throttle_group_incref(throttling_group);
581             blk_rs->throttle_state->cfg = cfg;
582         }
583 
584         QDECREF(bs_opts);
585     } else {
586         if (file && !*file) {
587             file = NULL;
588         }
589 
590         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
591          * with other callers) rather than what we want as the real defaults.
592          * Apply the defaults here instead. */
593         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_WB, "on");
594         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
595         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
596 
597         if (runstate_check(RUN_STATE_INMIGRATE)) {
598             bdrv_flags |= BDRV_O_INACTIVE;
599         }
600 
601         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
602         if (!blk) {
603             goto err_no_bs_opts;
604         }
605         bs = blk_bs(blk);
606 
607         bs->detect_zeroes = detect_zeroes;
608 
609         /* disk I/O throttling */
610         if (throttle_enabled(&cfg)) {
611             if (!throttling_group) {
612                 throttling_group = blk_name(blk);
613             }
614             bdrv_io_limits_enable(bs, throttling_group);
615             bdrv_set_io_limits(bs, &cfg);
616         }
617 
618         if (bdrv_key_required(bs)) {
619             autostart = 0;
620         }
621 
622         block_acct_init(blk_get_stats(blk), account_invalid, account_failed);
623 
624         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
625             blk_unref(blk);
626             blk = NULL;
627             goto err_no_bs_opts;
628         }
629     }
630 
631     blk_set_on_error(blk, on_read_error, on_write_error);
632 
633     if (!monitor_add_blk(blk, qemu_opts_id(opts), errp)) {
634         blk_unref(blk);
635         blk = NULL;
636         goto err_no_bs_opts;
637     }
638 
639 err_no_bs_opts:
640     qemu_opts_del(opts);
641     QDECREF(interval_dict);
642     QDECREF(interval_list);
643     return blk;
644 
645 early_err:
646     qemu_opts_del(opts);
647     QDECREF(interval_dict);
648     QDECREF(interval_list);
649 err_no_opts:
650     QDECREF(bs_opts);
651     return NULL;
652 }
653 
654 static QemuOptsList qemu_root_bds_opts;
655 
656 /* Takes the ownership of bs_opts */
657 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
658 {
659     BlockDriverState *bs;
660     QemuOpts *opts;
661     Error *local_error = NULL;
662     BlockdevDetectZeroesOptions detect_zeroes;
663     int ret;
664     int bdrv_flags = 0;
665 
666     opts = qemu_opts_create(&qemu_root_bds_opts, NULL, 1, errp);
667     if (!opts) {
668         goto fail;
669     }
670 
671     qemu_opts_absorb_qdict(opts, bs_opts, &local_error);
672     if (local_error) {
673         error_propagate(errp, local_error);
674         goto fail;
675     }
676 
677     extract_common_blockdev_options(opts, &bdrv_flags, NULL, NULL,
678                                     &detect_zeroes, &local_error);
679     if (local_error) {
680         error_propagate(errp, local_error);
681         goto fail;
682     }
683 
684     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
685      * with other callers) rather than what we want as the real defaults.
686      * Apply the defaults here instead. */
687     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_WB, "on");
688     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
689     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
690 
691     if (runstate_check(RUN_STATE_INMIGRATE)) {
692         bdrv_flags |= BDRV_O_INACTIVE;
693     }
694 
695     bs = NULL;
696     ret = bdrv_open(&bs, NULL, NULL, bs_opts, bdrv_flags, errp);
697     if (ret < 0) {
698         goto fail_no_bs_opts;
699     }
700 
701     bs->detect_zeroes = detect_zeroes;
702 
703 fail_no_bs_opts:
704     qemu_opts_del(opts);
705     return bs;
706 
707 fail:
708     qemu_opts_del(opts);
709     QDECREF(bs_opts);
710     return NULL;
711 }
712 
713 void blockdev_close_all_bdrv_states(void)
714 {
715     BlockDriverState *bs, *next_bs;
716 
717     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
718         AioContext *ctx = bdrv_get_aio_context(bs);
719 
720         aio_context_acquire(ctx);
721         bdrv_unref(bs);
722         aio_context_release(ctx);
723     }
724 }
725 
726 /* Iterates over the list of monitor-owned BlockDriverStates */
727 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
728 {
729     return bs ? QTAILQ_NEXT(bs, monitor_list)
730               : QTAILQ_FIRST(&monitor_bdrv_states);
731 }
732 
733 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
734                             Error **errp)
735 {
736     const char *value;
737 
738     value = qemu_opt_get(opts, from);
739     if (value) {
740         if (qemu_opt_find(opts, to)) {
741             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
742                        "same time", to, from);
743             return;
744         }
745     }
746 
747     /* rename all items in opts */
748     while ((value = qemu_opt_get(opts, from))) {
749         qemu_opt_set(opts, to, value, &error_abort);
750         qemu_opt_unset(opts, from);
751     }
752 }
753 
754 QemuOptsList qemu_legacy_drive_opts = {
755     .name = "drive",
756     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
757     .desc = {
758         {
759             .name = "bus",
760             .type = QEMU_OPT_NUMBER,
761             .help = "bus number",
762         },{
763             .name = "unit",
764             .type = QEMU_OPT_NUMBER,
765             .help = "unit number (i.e. lun for scsi)",
766         },{
767             .name = "index",
768             .type = QEMU_OPT_NUMBER,
769             .help = "index number",
770         },{
771             .name = "media",
772             .type = QEMU_OPT_STRING,
773             .help = "media type (disk, cdrom)",
774         },{
775             .name = "if",
776             .type = QEMU_OPT_STRING,
777             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
778         },{
779             .name = "cyls",
780             .type = QEMU_OPT_NUMBER,
781             .help = "number of cylinders (ide disk geometry)",
782         },{
783             .name = "heads",
784             .type = QEMU_OPT_NUMBER,
785             .help = "number of heads (ide disk geometry)",
786         },{
787             .name = "secs",
788             .type = QEMU_OPT_NUMBER,
789             .help = "number of sectors (ide disk geometry)",
790         },{
791             .name = "trans",
792             .type = QEMU_OPT_STRING,
793             .help = "chs translation (auto, lba, none)",
794         },{
795             .name = "boot",
796             .type = QEMU_OPT_BOOL,
797             .help = "(deprecated, ignored)",
798         },{
799             .name = "addr",
800             .type = QEMU_OPT_STRING,
801             .help = "pci address (virtio only)",
802         },{
803             .name = "serial",
804             .type = QEMU_OPT_STRING,
805             .help = "disk serial number",
806         },{
807             .name = "file",
808             .type = QEMU_OPT_STRING,
809             .help = "file name",
810         },
811 
812         /* Options that are passed on, but have special semantics with -drive */
813         {
814             .name = "read-only",
815             .type = QEMU_OPT_BOOL,
816             .help = "open drive file as read-only",
817         },{
818             .name = "rerror",
819             .type = QEMU_OPT_STRING,
820             .help = "read error action",
821         },{
822             .name = "werror",
823             .type = QEMU_OPT_STRING,
824             .help = "write error action",
825         },{
826             .name = "copy-on-read",
827             .type = QEMU_OPT_BOOL,
828             .help = "copy read data from backing file into image file",
829         },
830 
831         { /* end of list */ }
832     },
833 };
834 
835 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
836 {
837     const char *value;
838     BlockBackend *blk;
839     DriveInfo *dinfo = NULL;
840     QDict *bs_opts;
841     QemuOpts *legacy_opts;
842     DriveMediaType media = MEDIA_DISK;
843     BlockInterfaceType type;
844     int cyls, heads, secs, translation;
845     int max_devs, bus_id, unit_id, index;
846     const char *devaddr;
847     const char *werror, *rerror;
848     bool read_only = false;
849     bool copy_on_read;
850     const char *serial;
851     const char *filename;
852     Error *local_err = NULL;
853     int i;
854 
855     /* Change legacy command line options into QMP ones */
856     static const struct {
857         const char *from;
858         const char *to;
859     } opt_renames[] = {
860         { "iops",           "throttling.iops-total" },
861         { "iops_rd",        "throttling.iops-read" },
862         { "iops_wr",        "throttling.iops-write" },
863 
864         { "bps",            "throttling.bps-total" },
865         { "bps_rd",         "throttling.bps-read" },
866         { "bps_wr",         "throttling.bps-write" },
867 
868         { "iops_max",       "throttling.iops-total-max" },
869         { "iops_rd_max",    "throttling.iops-read-max" },
870         { "iops_wr_max",    "throttling.iops-write-max" },
871 
872         { "bps_max",        "throttling.bps-total-max" },
873         { "bps_rd_max",     "throttling.bps-read-max" },
874         { "bps_wr_max",     "throttling.bps-write-max" },
875 
876         { "iops_size",      "throttling.iops-size" },
877 
878         { "group",          "throttling.group" },
879 
880         { "readonly",       "read-only" },
881     };
882 
883     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
884         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
885                         &local_err);
886         if (local_err) {
887             error_report_err(local_err);
888             return NULL;
889         }
890     }
891 
892     value = qemu_opt_get(all_opts, "cache");
893     if (value) {
894         int flags = 0;
895 
896         if (bdrv_parse_cache_flags(value, &flags) != 0) {
897             error_report("invalid cache option");
898             return NULL;
899         }
900 
901         /* Specific options take precedence */
902         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
903             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
904                               !!(flags & BDRV_O_CACHE_WB), &error_abort);
905         }
906         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
907             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
908                               !!(flags & BDRV_O_NOCACHE), &error_abort);
909         }
910         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
911             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
912                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
913         }
914         qemu_opt_unset(all_opts, "cache");
915     }
916 
917     /* Get a QDict for processing the options */
918     bs_opts = qdict_new();
919     qemu_opts_to_qdict(all_opts, bs_opts);
920 
921     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
922                                    &error_abort);
923     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
924     if (local_err) {
925         error_report_err(local_err);
926         goto fail;
927     }
928 
929     /* Deprecated option boot=[on|off] */
930     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
931         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
932                 "ignored. Future versions will reject this parameter. Please "
933                 "update your scripts.\n");
934     }
935 
936     /* Media type */
937     value = qemu_opt_get(legacy_opts, "media");
938     if (value) {
939         if (!strcmp(value, "disk")) {
940             media = MEDIA_DISK;
941         } else if (!strcmp(value, "cdrom")) {
942             media = MEDIA_CDROM;
943             read_only = true;
944         } else {
945             error_report("'%s' invalid media", value);
946             goto fail;
947         }
948     }
949 
950     /* copy-on-read is disabled with a warning for read-only devices */
951     read_only |= qemu_opt_get_bool(legacy_opts, "read-only", false);
952     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
953 
954     if (read_only && copy_on_read) {
955         error_report("warning: disabling copy-on-read on read-only drive");
956         copy_on_read = false;
957     }
958 
959     qdict_put(bs_opts, "read-only",
960               qstring_from_str(read_only ? "on" : "off"));
961     qdict_put(bs_opts, "copy-on-read",
962               qstring_from_str(copy_on_read ? "on" :"off"));
963 
964     /* Controller type */
965     value = qemu_opt_get(legacy_opts, "if");
966     if (value) {
967         for (type = 0;
968              type < IF_COUNT && strcmp(value, if_name[type]);
969              type++) {
970         }
971         if (type == IF_COUNT) {
972             error_report("unsupported bus type '%s'", value);
973             goto fail;
974         }
975     } else {
976         type = block_default_type;
977     }
978 
979     /* Geometry */
980     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
981     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
982     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
983 
984     if (cyls || heads || secs) {
985         if (cyls < 1) {
986             error_report("invalid physical cyls number");
987             goto fail;
988         }
989         if (heads < 1) {
990             error_report("invalid physical heads number");
991             goto fail;
992         }
993         if (secs < 1) {
994             error_report("invalid physical secs number");
995             goto fail;
996         }
997     }
998 
999     translation = BIOS_ATA_TRANSLATION_AUTO;
1000     value = qemu_opt_get(legacy_opts, "trans");
1001     if (value != NULL) {
1002         if (!cyls) {
1003             error_report("'%s' trans must be used with cyls, heads and secs",
1004                          value);
1005             goto fail;
1006         }
1007         if (!strcmp(value, "none")) {
1008             translation = BIOS_ATA_TRANSLATION_NONE;
1009         } else if (!strcmp(value, "lba")) {
1010             translation = BIOS_ATA_TRANSLATION_LBA;
1011         } else if (!strcmp(value, "large")) {
1012             translation = BIOS_ATA_TRANSLATION_LARGE;
1013         } else if (!strcmp(value, "rechs")) {
1014             translation = BIOS_ATA_TRANSLATION_RECHS;
1015         } else if (!strcmp(value, "auto")) {
1016             translation = BIOS_ATA_TRANSLATION_AUTO;
1017         } else {
1018             error_report("'%s' invalid translation type", value);
1019             goto fail;
1020         }
1021     }
1022 
1023     if (media == MEDIA_CDROM) {
1024         if (cyls || secs || heads) {
1025             error_report("CHS can't be set with media=cdrom");
1026             goto fail;
1027         }
1028     }
1029 
1030     /* Device address specified by bus/unit or index.
1031      * If none was specified, try to find the first free one. */
1032     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
1033     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
1034     index   = qemu_opt_get_number(legacy_opts, "index", -1);
1035 
1036     max_devs = if_max_devs[type];
1037 
1038     if (index != -1) {
1039         if (bus_id != 0 || unit_id != -1) {
1040             error_report("index cannot be used with bus and unit");
1041             goto fail;
1042         }
1043         bus_id = drive_index_to_bus_id(type, index);
1044         unit_id = drive_index_to_unit_id(type, index);
1045     }
1046 
1047     if (unit_id == -1) {
1048        unit_id = 0;
1049        while (drive_get(type, bus_id, unit_id) != NULL) {
1050            unit_id++;
1051            if (max_devs && unit_id >= max_devs) {
1052                unit_id -= max_devs;
1053                bus_id++;
1054            }
1055        }
1056     }
1057 
1058     if (max_devs && unit_id >= max_devs) {
1059         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1060         goto fail;
1061     }
1062 
1063     if (drive_get(type, bus_id, unit_id) != NULL) {
1064         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1065                      bus_id, unit_id, index);
1066         goto fail;
1067     }
1068 
1069     /* Serial number */
1070     serial = qemu_opt_get(legacy_opts, "serial");
1071 
1072     /* no id supplied -> create one */
1073     if (qemu_opts_id(all_opts) == NULL) {
1074         char *new_id;
1075         const char *mediastr = "";
1076         if (type == IF_IDE || type == IF_SCSI) {
1077             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1078         }
1079         if (max_devs) {
1080             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1081                                      mediastr, unit_id);
1082         } else {
1083             new_id = g_strdup_printf("%s%s%i", if_name[type],
1084                                      mediastr, unit_id);
1085         }
1086         qdict_put(bs_opts, "id", qstring_from_str(new_id));
1087         g_free(new_id);
1088     }
1089 
1090     /* Add virtio block device */
1091     devaddr = qemu_opt_get(legacy_opts, "addr");
1092     if (devaddr && type != IF_VIRTIO) {
1093         error_report("addr is not supported by this bus type");
1094         goto fail;
1095     }
1096 
1097     if (type == IF_VIRTIO) {
1098         QemuOpts *devopts;
1099         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1100                                    &error_abort);
1101         if (arch_type == QEMU_ARCH_S390X) {
1102             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1103         } else {
1104             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1105         }
1106         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1107                      &error_abort);
1108         if (devaddr) {
1109             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1110         }
1111     }
1112 
1113     filename = qemu_opt_get(legacy_opts, "file");
1114 
1115     /* Check werror/rerror compatibility with if=... */
1116     werror = qemu_opt_get(legacy_opts, "werror");
1117     if (werror != NULL) {
1118         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1119             type != IF_NONE) {
1120             error_report("werror is not supported by this bus type");
1121             goto fail;
1122         }
1123         qdict_put(bs_opts, "werror", qstring_from_str(werror));
1124     }
1125 
1126     rerror = qemu_opt_get(legacy_opts, "rerror");
1127     if (rerror != NULL) {
1128         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1129             type != IF_NONE) {
1130             error_report("rerror is not supported by this bus type");
1131             goto fail;
1132         }
1133         qdict_put(bs_opts, "rerror", qstring_from_str(rerror));
1134     }
1135 
1136     /* Actual block device init: Functionality shared with blockdev-add */
1137     blk = blockdev_init(filename, bs_opts, &local_err);
1138     bs_opts = NULL;
1139     if (!blk) {
1140         if (local_err) {
1141             error_report_err(local_err);
1142         }
1143         goto fail;
1144     } else {
1145         assert(!local_err);
1146     }
1147 
1148     /* Create legacy DriveInfo */
1149     dinfo = g_malloc0(sizeof(*dinfo));
1150     dinfo->opts = all_opts;
1151 
1152     dinfo->cyls = cyls;
1153     dinfo->heads = heads;
1154     dinfo->secs = secs;
1155     dinfo->trans = translation;
1156 
1157     dinfo->type = type;
1158     dinfo->bus = bus_id;
1159     dinfo->unit = unit_id;
1160     dinfo->devaddr = devaddr;
1161     dinfo->serial = g_strdup(serial);
1162 
1163     blk_set_legacy_dinfo(blk, dinfo);
1164 
1165     switch(type) {
1166     case IF_IDE:
1167     case IF_SCSI:
1168     case IF_XEN:
1169     case IF_NONE:
1170         dinfo->media_cd = media == MEDIA_CDROM;
1171         break;
1172     default:
1173         break;
1174     }
1175 
1176 fail:
1177     qemu_opts_del(legacy_opts);
1178     QDECREF(bs_opts);
1179     return dinfo;
1180 }
1181 
1182 void hmp_commit(Monitor *mon, const QDict *qdict)
1183 {
1184     const char *device = qdict_get_str(qdict, "device");
1185     BlockBackend *blk;
1186     int ret;
1187 
1188     if (!strcmp(device, "all")) {
1189         ret = blk_commit_all();
1190     } else {
1191         BlockDriverState *bs;
1192         AioContext *aio_context;
1193 
1194         blk = blk_by_name(device);
1195         if (!blk) {
1196             monitor_printf(mon, "Device '%s' not found\n", device);
1197             return;
1198         }
1199         if (!blk_is_available(blk)) {
1200             monitor_printf(mon, "Device '%s' has no medium\n", device);
1201             return;
1202         }
1203 
1204         bs = blk_bs(blk);
1205         aio_context = bdrv_get_aio_context(bs);
1206         aio_context_acquire(aio_context);
1207 
1208         ret = bdrv_commit(bs);
1209 
1210         aio_context_release(aio_context);
1211     }
1212     if (ret < 0) {
1213         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1214                        strerror(-ret));
1215     }
1216 }
1217 
1218 static void blockdev_do_action(TransactionAction *action, Error **errp)
1219 {
1220     TransactionActionList list;
1221 
1222     list.value = action;
1223     list.next = NULL;
1224     qmp_transaction(&list, false, NULL, errp);
1225 }
1226 
1227 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1228                                 bool has_node_name, const char *node_name,
1229                                 const char *snapshot_file,
1230                                 bool has_snapshot_node_name,
1231                                 const char *snapshot_node_name,
1232                                 bool has_format, const char *format,
1233                                 bool has_mode, NewImageMode mode, Error **errp)
1234 {
1235     BlockdevSnapshotSync snapshot = {
1236         .has_device = has_device,
1237         .device = (char *) device,
1238         .has_node_name = has_node_name,
1239         .node_name = (char *) node_name,
1240         .snapshot_file = (char *) snapshot_file,
1241         .has_snapshot_node_name = has_snapshot_node_name,
1242         .snapshot_node_name = (char *) snapshot_node_name,
1243         .has_format = has_format,
1244         .format = (char *) format,
1245         .has_mode = has_mode,
1246         .mode = mode,
1247     };
1248     TransactionAction action = {
1249         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1250         .u.blockdev_snapshot_sync.data = &snapshot,
1251     };
1252     blockdev_do_action(&action, errp);
1253 }
1254 
1255 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1256                            Error **errp)
1257 {
1258     BlockdevSnapshot snapshot_data = {
1259         .node = (char *) node,
1260         .overlay = (char *) overlay
1261     };
1262     TransactionAction action = {
1263         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1264         .u.blockdev_snapshot.data = &snapshot_data,
1265     };
1266     blockdev_do_action(&action, errp);
1267 }
1268 
1269 void qmp_blockdev_snapshot_internal_sync(const char *device,
1270                                          const char *name,
1271                                          Error **errp)
1272 {
1273     BlockdevSnapshotInternal snapshot = {
1274         .device = (char *) device,
1275         .name = (char *) name
1276     };
1277     TransactionAction action = {
1278         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1279         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1280     };
1281     blockdev_do_action(&action, errp);
1282 }
1283 
1284 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1285                                                          bool has_id,
1286                                                          const char *id,
1287                                                          bool has_name,
1288                                                          const char *name,
1289                                                          Error **errp)
1290 {
1291     BlockDriverState *bs;
1292     BlockBackend *blk;
1293     AioContext *aio_context;
1294     QEMUSnapshotInfo sn;
1295     Error *local_err = NULL;
1296     SnapshotInfo *info = NULL;
1297     int ret;
1298 
1299     blk = blk_by_name(device);
1300     if (!blk) {
1301         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1302                   "Device '%s' not found", device);
1303         return NULL;
1304     }
1305 
1306     aio_context = blk_get_aio_context(blk);
1307     aio_context_acquire(aio_context);
1308 
1309     if (!has_id) {
1310         id = NULL;
1311     }
1312 
1313     if (!has_name) {
1314         name = NULL;
1315     }
1316 
1317     if (!id && !name) {
1318         error_setg(errp, "Name or id must be provided");
1319         goto out_aio_context;
1320     }
1321 
1322     if (!blk_is_available(blk)) {
1323         error_setg(errp, "Device '%s' has no medium", device);
1324         goto out_aio_context;
1325     }
1326     bs = blk_bs(blk);
1327 
1328     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1329         goto out_aio_context;
1330     }
1331 
1332     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1333     if (local_err) {
1334         error_propagate(errp, local_err);
1335         goto out_aio_context;
1336     }
1337     if (!ret) {
1338         error_setg(errp,
1339                    "Snapshot with id '%s' and name '%s' does not exist on "
1340                    "device '%s'",
1341                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1342         goto out_aio_context;
1343     }
1344 
1345     bdrv_snapshot_delete(bs, id, name, &local_err);
1346     if (local_err) {
1347         error_propagate(errp, local_err);
1348         goto out_aio_context;
1349     }
1350 
1351     aio_context_release(aio_context);
1352 
1353     info = g_new0(SnapshotInfo, 1);
1354     info->id = g_strdup(sn.id_str);
1355     info->name = g_strdup(sn.name);
1356     info->date_nsec = sn.date_nsec;
1357     info->date_sec = sn.date_sec;
1358     info->vm_state_size = sn.vm_state_size;
1359     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1360     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1361 
1362     return info;
1363 
1364 out_aio_context:
1365     aio_context_release(aio_context);
1366     return NULL;
1367 }
1368 
1369 /**
1370  * block_dirty_bitmap_lookup:
1371  * Return a dirty bitmap (if present), after validating
1372  * the node reference and bitmap names.
1373  *
1374  * @node: The name of the BDS node to search for bitmaps
1375  * @name: The name of the bitmap to search for
1376  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1377  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1378  * @errp: Output pointer for error information. Can be NULL.
1379  *
1380  * @return: A bitmap object on success, or NULL on failure.
1381  */
1382 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1383                                                   const char *name,
1384                                                   BlockDriverState **pbs,
1385                                                   AioContext **paio,
1386                                                   Error **errp)
1387 {
1388     BlockDriverState *bs;
1389     BdrvDirtyBitmap *bitmap;
1390     AioContext *aio_context;
1391 
1392     if (!node) {
1393         error_setg(errp, "Node cannot be NULL");
1394         return NULL;
1395     }
1396     if (!name) {
1397         error_setg(errp, "Bitmap name cannot be NULL");
1398         return NULL;
1399     }
1400     bs = bdrv_lookup_bs(node, node, NULL);
1401     if (!bs) {
1402         error_setg(errp, "Node '%s' not found", node);
1403         return NULL;
1404     }
1405 
1406     aio_context = bdrv_get_aio_context(bs);
1407     aio_context_acquire(aio_context);
1408 
1409     bitmap = bdrv_find_dirty_bitmap(bs, name);
1410     if (!bitmap) {
1411         error_setg(errp, "Dirty bitmap '%s' not found", name);
1412         goto fail;
1413     }
1414 
1415     if (pbs) {
1416         *pbs = bs;
1417     }
1418     if (paio) {
1419         *paio = aio_context;
1420     } else {
1421         aio_context_release(aio_context);
1422     }
1423 
1424     return bitmap;
1425 
1426  fail:
1427     aio_context_release(aio_context);
1428     return NULL;
1429 }
1430 
1431 /* New and old BlockDriverState structs for atomic group operations */
1432 
1433 typedef struct BlkActionState BlkActionState;
1434 
1435 /**
1436  * BlkActionOps:
1437  * Table of operations that define an Action.
1438  *
1439  * @instance_size: Size of state struct, in bytes.
1440  * @prepare: Prepare the work, must NOT be NULL.
1441  * @commit: Commit the changes, can be NULL.
1442  * @abort: Abort the changes on fail, can be NULL.
1443  * @clean: Clean up resources after all transaction actions have called
1444  *         commit() or abort(). Can be NULL.
1445  *
1446  * Only prepare() may fail. In a single transaction, only one of commit() or
1447  * abort() will be called. clean() will always be called if it is present.
1448  */
1449 typedef struct BlkActionOps {
1450     size_t instance_size;
1451     void (*prepare)(BlkActionState *common, Error **errp);
1452     void (*commit)(BlkActionState *common);
1453     void (*abort)(BlkActionState *common);
1454     void (*clean)(BlkActionState *common);
1455 } BlkActionOps;
1456 
1457 /**
1458  * BlkActionState:
1459  * Describes one Action's state within a Transaction.
1460  *
1461  * @action: QAPI-defined enum identifying which Action to perform.
1462  * @ops: Table of ActionOps this Action can perform.
1463  * @block_job_txn: Transaction which this action belongs to.
1464  * @entry: List membership for all Actions in this Transaction.
1465  *
1466  * This structure must be arranged as first member in a subclassed type,
1467  * assuming that the compiler will also arrange it to the same offsets as the
1468  * base class.
1469  */
1470 struct BlkActionState {
1471     TransactionAction *action;
1472     const BlkActionOps *ops;
1473     BlockJobTxn *block_job_txn;
1474     TransactionProperties *txn_props;
1475     QSIMPLEQ_ENTRY(BlkActionState) entry;
1476 };
1477 
1478 /* internal snapshot private data */
1479 typedef struct InternalSnapshotState {
1480     BlkActionState common;
1481     BlockDriverState *bs;
1482     AioContext *aio_context;
1483     QEMUSnapshotInfo sn;
1484     bool created;
1485 } InternalSnapshotState;
1486 
1487 
1488 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1489 {
1490     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1491         error_setg(errp,
1492                    "Action '%s' does not support Transaction property "
1493                    "completion-mode = %s",
1494                    TransactionActionKind_lookup[s->action->type],
1495                    ActionCompletionMode_lookup[s->txn_props->completion_mode]);
1496         return -1;
1497     }
1498     return 0;
1499 }
1500 
1501 static void internal_snapshot_prepare(BlkActionState *common,
1502                                       Error **errp)
1503 {
1504     Error *local_err = NULL;
1505     const char *device;
1506     const char *name;
1507     BlockBackend *blk;
1508     BlockDriverState *bs;
1509     QEMUSnapshotInfo old_sn, *sn;
1510     bool ret;
1511     qemu_timeval tv;
1512     BlockdevSnapshotInternal *internal;
1513     InternalSnapshotState *state;
1514     int ret1;
1515 
1516     g_assert(common->action->type ==
1517              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1518     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1519     state = DO_UPCAST(InternalSnapshotState, common, common);
1520 
1521     /* 1. parse input */
1522     device = internal->device;
1523     name = internal->name;
1524 
1525     /* 2. check for validation */
1526     if (action_check_completion_mode(common, errp) < 0) {
1527         return;
1528     }
1529 
1530     blk = blk_by_name(device);
1531     if (!blk) {
1532         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1533                   "Device '%s' not found", device);
1534         return;
1535     }
1536 
1537     /* AioContext is released in .clean() */
1538     state->aio_context = blk_get_aio_context(blk);
1539     aio_context_acquire(state->aio_context);
1540 
1541     if (!blk_is_available(blk)) {
1542         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1543         return;
1544     }
1545     bs = blk_bs(blk);
1546 
1547     state->bs = bs;
1548     bdrv_drained_begin(bs);
1549 
1550     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1551         return;
1552     }
1553 
1554     if (bdrv_is_read_only(bs)) {
1555         error_setg(errp, "Device '%s' is read only", device);
1556         return;
1557     }
1558 
1559     if (!bdrv_can_snapshot(bs)) {
1560         error_setg(errp, "Block format '%s' used by device '%s' "
1561                    "does not support internal snapshots",
1562                    bs->drv->format_name, device);
1563         return;
1564     }
1565 
1566     if (!strlen(name)) {
1567         error_setg(errp, "Name is empty");
1568         return;
1569     }
1570 
1571     /* check whether a snapshot with name exist */
1572     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1573                                             &local_err);
1574     if (local_err) {
1575         error_propagate(errp, local_err);
1576         return;
1577     } else if (ret) {
1578         error_setg(errp,
1579                    "Snapshot with name '%s' already exists on device '%s'",
1580                    name, device);
1581         return;
1582     }
1583 
1584     /* 3. take the snapshot */
1585     sn = &state->sn;
1586     pstrcpy(sn->name, sizeof(sn->name), name);
1587     qemu_gettimeofday(&tv);
1588     sn->date_sec = tv.tv_sec;
1589     sn->date_nsec = tv.tv_usec * 1000;
1590     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1591 
1592     ret1 = bdrv_snapshot_create(bs, sn);
1593     if (ret1 < 0) {
1594         error_setg_errno(errp, -ret1,
1595                          "Failed to create snapshot '%s' on device '%s'",
1596                          name, device);
1597         return;
1598     }
1599 
1600     /* 4. succeed, mark a snapshot is created */
1601     state->created = true;
1602 }
1603 
1604 static void internal_snapshot_abort(BlkActionState *common)
1605 {
1606     InternalSnapshotState *state =
1607                              DO_UPCAST(InternalSnapshotState, common, common);
1608     BlockDriverState *bs = state->bs;
1609     QEMUSnapshotInfo *sn = &state->sn;
1610     Error *local_error = NULL;
1611 
1612     if (!state->created) {
1613         return;
1614     }
1615 
1616     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1617         error_reportf_err(local_error,
1618                           "Failed to delete snapshot with id '%s' and "
1619                           "name '%s' on device '%s' in abort: ",
1620                           sn->id_str, sn->name,
1621                           bdrv_get_device_name(bs));
1622     }
1623 }
1624 
1625 static void internal_snapshot_clean(BlkActionState *common)
1626 {
1627     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1628                                              common, common);
1629 
1630     if (state->aio_context) {
1631         if (state->bs) {
1632             bdrv_drained_end(state->bs);
1633         }
1634         aio_context_release(state->aio_context);
1635     }
1636 }
1637 
1638 /* external snapshot private data */
1639 typedef struct ExternalSnapshotState {
1640     BlkActionState common;
1641     BlockDriverState *old_bs;
1642     BlockDriverState *new_bs;
1643     AioContext *aio_context;
1644 } ExternalSnapshotState;
1645 
1646 static void external_snapshot_prepare(BlkActionState *common,
1647                                       Error **errp)
1648 {
1649     int flags = 0, ret;
1650     QDict *options = NULL;
1651     Error *local_err = NULL;
1652     /* Device and node name of the image to generate the snapshot from */
1653     const char *device;
1654     const char *node_name;
1655     /* Reference to the new image (for 'blockdev-snapshot') */
1656     const char *snapshot_ref;
1657     /* File name of the new image (for 'blockdev-snapshot-sync') */
1658     const char *new_image_file;
1659     ExternalSnapshotState *state =
1660                              DO_UPCAST(ExternalSnapshotState, common, common);
1661     TransactionAction *action = common->action;
1662 
1663     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1664      * purpose but a different set of parameters */
1665     switch (action->type) {
1666     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1667         {
1668             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1669             device = s->node;
1670             node_name = s->node;
1671             new_image_file = NULL;
1672             snapshot_ref = s->overlay;
1673         }
1674         break;
1675     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1676         {
1677             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1678             device = s->has_device ? s->device : NULL;
1679             node_name = s->has_node_name ? s->node_name : NULL;
1680             new_image_file = s->snapshot_file;
1681             snapshot_ref = NULL;
1682         }
1683         break;
1684     default:
1685         g_assert_not_reached();
1686     }
1687 
1688     /* start processing */
1689     if (action_check_completion_mode(common, errp) < 0) {
1690         return;
1691     }
1692 
1693     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1694     if (!state->old_bs) {
1695         return;
1696     }
1697 
1698     /* Acquire AioContext now so any threads operating on old_bs stop */
1699     state->aio_context = bdrv_get_aio_context(state->old_bs);
1700     aio_context_acquire(state->aio_context);
1701     bdrv_drained_begin(state->old_bs);
1702 
1703     if (!bdrv_is_inserted(state->old_bs)) {
1704         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1705         return;
1706     }
1707 
1708     if (bdrv_op_is_blocked(state->old_bs,
1709                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1710         return;
1711     }
1712 
1713     if (!bdrv_is_read_only(state->old_bs)) {
1714         if (bdrv_flush(state->old_bs)) {
1715             error_setg(errp, QERR_IO_ERROR);
1716             return;
1717         }
1718     }
1719 
1720     if (!bdrv_is_first_non_filter(state->old_bs)) {
1721         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1722         return;
1723     }
1724 
1725     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1726         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1727         const char *format = s->has_format ? s->format : "qcow2";
1728         enum NewImageMode mode;
1729         const char *snapshot_node_name =
1730             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1731 
1732         if (node_name && !snapshot_node_name) {
1733             error_setg(errp, "New snapshot node name missing");
1734             return;
1735         }
1736 
1737         if (snapshot_node_name &&
1738             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1739             error_setg(errp, "New snapshot node name already in use");
1740             return;
1741         }
1742 
1743         flags = state->old_bs->open_flags;
1744 
1745         /* create new image w/backing file */
1746         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1747         if (mode != NEW_IMAGE_MODE_EXISTING) {
1748             int64_t size = bdrv_getlength(state->old_bs);
1749             if (size < 0) {
1750                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1751                 return;
1752             }
1753             bdrv_img_create(new_image_file, format,
1754                             state->old_bs->filename,
1755                             state->old_bs->drv->format_name,
1756                             NULL, size, flags, &local_err, false);
1757             if (local_err) {
1758                 error_propagate(errp, local_err);
1759                 return;
1760             }
1761         }
1762 
1763         options = qdict_new();
1764         if (s->has_snapshot_node_name) {
1765             qdict_put(options, "node-name",
1766                       qstring_from_str(snapshot_node_name));
1767         }
1768         qdict_put(options, "driver", qstring_from_str(format));
1769 
1770         flags |= BDRV_O_NO_BACKING;
1771     }
1772 
1773     assert(state->new_bs == NULL);
1774     ret = bdrv_open(&state->new_bs, new_image_file, snapshot_ref, options,
1775                     flags, errp);
1776     /* We will manually add the backing_hd field to the bs later */
1777     if (ret != 0) {
1778         return;
1779     }
1780 
1781     if (state->new_bs->blk != NULL) {
1782         error_setg(errp, "The snapshot is already in use by %s",
1783                    blk_name(state->new_bs->blk));
1784         return;
1785     }
1786 
1787     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1788                            errp)) {
1789         return;
1790     }
1791 
1792     if (state->new_bs->backing != NULL) {
1793         error_setg(errp, "The snapshot already has a backing image");
1794         return;
1795     }
1796 
1797     if (!state->new_bs->drv->supports_backing) {
1798         error_setg(errp, "The snapshot does not support backing images");
1799     }
1800 }
1801 
1802 static void external_snapshot_commit(BlkActionState *common)
1803 {
1804     ExternalSnapshotState *state =
1805                              DO_UPCAST(ExternalSnapshotState, common, common);
1806 
1807     bdrv_set_aio_context(state->new_bs, state->aio_context);
1808 
1809     /* This removes our old bs and adds the new bs */
1810     bdrv_append(state->new_bs, state->old_bs);
1811     /* We don't need (or want) to use the transactional
1812      * bdrv_reopen_multiple() across all the entries at once, because we
1813      * don't want to abort all of them if one of them fails the reopen */
1814     bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1815                 NULL);
1816 }
1817 
1818 static void external_snapshot_abort(BlkActionState *common)
1819 {
1820     ExternalSnapshotState *state =
1821                              DO_UPCAST(ExternalSnapshotState, common, common);
1822     if (state->new_bs) {
1823         bdrv_unref(state->new_bs);
1824     }
1825 }
1826 
1827 static void external_snapshot_clean(BlkActionState *common)
1828 {
1829     ExternalSnapshotState *state =
1830                              DO_UPCAST(ExternalSnapshotState, common, common);
1831     if (state->aio_context) {
1832         bdrv_drained_end(state->old_bs);
1833         aio_context_release(state->aio_context);
1834     }
1835 }
1836 
1837 typedef struct DriveBackupState {
1838     BlkActionState common;
1839     BlockDriverState *bs;
1840     AioContext *aio_context;
1841     BlockJob *job;
1842 } DriveBackupState;
1843 
1844 static void do_drive_backup(const char *device, const char *target,
1845                             bool has_format, const char *format,
1846                             enum MirrorSyncMode sync,
1847                             bool has_mode, enum NewImageMode mode,
1848                             bool has_speed, int64_t speed,
1849                             bool has_bitmap, const char *bitmap,
1850                             bool has_on_source_error,
1851                             BlockdevOnError on_source_error,
1852                             bool has_on_target_error,
1853                             BlockdevOnError on_target_error,
1854                             BlockJobTxn *txn, Error **errp);
1855 
1856 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1857 {
1858     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1859     BlockBackend *blk;
1860     DriveBackup *backup;
1861     Error *local_err = NULL;
1862 
1863     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1864     backup = common->action->u.drive_backup.data;
1865 
1866     blk = blk_by_name(backup->device);
1867     if (!blk) {
1868         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1869                   "Device '%s' not found", backup->device);
1870         return;
1871     }
1872 
1873     if (!blk_is_available(blk)) {
1874         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, backup->device);
1875         return;
1876     }
1877 
1878     /* AioContext is released in .clean() */
1879     state->aio_context = blk_get_aio_context(blk);
1880     aio_context_acquire(state->aio_context);
1881     bdrv_drained_begin(blk_bs(blk));
1882     state->bs = blk_bs(blk);
1883 
1884     do_drive_backup(backup->device, backup->target,
1885                     backup->has_format, backup->format,
1886                     backup->sync,
1887                     backup->has_mode, backup->mode,
1888                     backup->has_speed, backup->speed,
1889                     backup->has_bitmap, backup->bitmap,
1890                     backup->has_on_source_error, backup->on_source_error,
1891                     backup->has_on_target_error, backup->on_target_error,
1892                     common->block_job_txn, &local_err);
1893     if (local_err) {
1894         error_propagate(errp, local_err);
1895         return;
1896     }
1897 
1898     state->job = state->bs->job;
1899 }
1900 
1901 static void drive_backup_abort(BlkActionState *common)
1902 {
1903     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1904     BlockDriverState *bs = state->bs;
1905 
1906     /* Only cancel if it's the job we started */
1907     if (bs && bs->job && bs->job == state->job) {
1908         block_job_cancel_sync(bs->job);
1909     }
1910 }
1911 
1912 static void drive_backup_clean(BlkActionState *common)
1913 {
1914     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1915 
1916     if (state->aio_context) {
1917         bdrv_drained_end(state->bs);
1918         aio_context_release(state->aio_context);
1919     }
1920 }
1921 
1922 typedef struct BlockdevBackupState {
1923     BlkActionState common;
1924     BlockDriverState *bs;
1925     BlockJob *job;
1926     AioContext *aio_context;
1927 } BlockdevBackupState;
1928 
1929 static void do_blockdev_backup(const char *device, const char *target,
1930                                enum MirrorSyncMode sync,
1931                                bool has_speed, int64_t speed,
1932                                bool has_on_source_error,
1933                                BlockdevOnError on_source_error,
1934                                bool has_on_target_error,
1935                                BlockdevOnError on_target_error,
1936                                BlockJobTxn *txn, Error **errp);
1937 
1938 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1939 {
1940     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1941     BlockdevBackup *backup;
1942     BlockBackend *blk, *target;
1943     Error *local_err = NULL;
1944 
1945     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1946     backup = common->action->u.blockdev_backup.data;
1947 
1948     blk = blk_by_name(backup->device);
1949     if (!blk) {
1950         error_setg(errp, "Device '%s' not found", backup->device);
1951         return;
1952     }
1953 
1954     if (!blk_is_available(blk)) {
1955         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, backup->device);
1956         return;
1957     }
1958 
1959     target = blk_by_name(backup->target);
1960     if (!target) {
1961         error_setg(errp, "Device '%s' not found", backup->target);
1962         return;
1963     }
1964 
1965     /* AioContext is released in .clean() */
1966     state->aio_context = blk_get_aio_context(blk);
1967     if (state->aio_context != blk_get_aio_context(target)) {
1968         state->aio_context = NULL;
1969         error_setg(errp, "Backup between two IO threads is not implemented");
1970         return;
1971     }
1972     aio_context_acquire(state->aio_context);
1973     state->bs = blk_bs(blk);
1974     bdrv_drained_begin(state->bs);
1975 
1976     do_blockdev_backup(backup->device, backup->target,
1977                        backup->sync,
1978                        backup->has_speed, backup->speed,
1979                        backup->has_on_source_error, backup->on_source_error,
1980                        backup->has_on_target_error, backup->on_target_error,
1981                        common->block_job_txn, &local_err);
1982     if (local_err) {
1983         error_propagate(errp, local_err);
1984         return;
1985     }
1986 
1987     state->job = state->bs->job;
1988 }
1989 
1990 static void blockdev_backup_abort(BlkActionState *common)
1991 {
1992     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1993     BlockDriverState *bs = state->bs;
1994 
1995     /* Only cancel if it's the job we started */
1996     if (bs && bs->job && bs->job == state->job) {
1997         block_job_cancel_sync(bs->job);
1998     }
1999 }
2000 
2001 static void blockdev_backup_clean(BlkActionState *common)
2002 {
2003     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
2004 
2005     if (state->aio_context) {
2006         bdrv_drained_end(state->bs);
2007         aio_context_release(state->aio_context);
2008     }
2009 }
2010 
2011 typedef struct BlockDirtyBitmapState {
2012     BlkActionState common;
2013     BdrvDirtyBitmap *bitmap;
2014     BlockDriverState *bs;
2015     AioContext *aio_context;
2016     HBitmap *backup;
2017     bool prepared;
2018 } BlockDirtyBitmapState;
2019 
2020 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
2021                                            Error **errp)
2022 {
2023     Error *local_err = NULL;
2024     BlockDirtyBitmapAdd *action;
2025     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2026                                              common, common);
2027 
2028     if (action_check_completion_mode(common, errp) < 0) {
2029         return;
2030     }
2031 
2032     action = common->action->u.block_dirty_bitmap_add.data;
2033     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
2034     qmp_block_dirty_bitmap_add(action->node, action->name,
2035                                action->has_granularity, action->granularity,
2036                                &local_err);
2037 
2038     if (!local_err) {
2039         state->prepared = true;
2040     } else {
2041         error_propagate(errp, local_err);
2042     }
2043 }
2044 
2045 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2046 {
2047     BlockDirtyBitmapAdd *action;
2048     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2049                                              common, common);
2050 
2051     action = common->action->u.block_dirty_bitmap_add.data;
2052     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2053      * then the node reference and bitmap name must have been valid.
2054      */
2055     if (state->prepared) {
2056         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2057     }
2058 }
2059 
2060 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2061                                              Error **errp)
2062 {
2063     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2064                                              common, common);
2065     BlockDirtyBitmap *action;
2066 
2067     if (action_check_completion_mode(common, errp) < 0) {
2068         return;
2069     }
2070 
2071     action = common->action->u.block_dirty_bitmap_clear.data;
2072     state->bitmap = block_dirty_bitmap_lookup(action->node,
2073                                               action->name,
2074                                               &state->bs,
2075                                               &state->aio_context,
2076                                               errp);
2077     if (!state->bitmap) {
2078         return;
2079     }
2080 
2081     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2082         error_setg(errp, "Cannot modify a frozen bitmap");
2083         return;
2084     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2085         error_setg(errp, "Cannot clear a disabled bitmap");
2086         return;
2087     }
2088 
2089     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2090     /* AioContext is released in .clean() */
2091 }
2092 
2093 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2094 {
2095     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2096                                              common, common);
2097 
2098     bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2099 }
2100 
2101 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2102 {
2103     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2104                                              common, common);
2105 
2106     hbitmap_free(state->backup);
2107 }
2108 
2109 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2110 {
2111     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2112                                              common, common);
2113 
2114     if (state->aio_context) {
2115         aio_context_release(state->aio_context);
2116     }
2117 }
2118 
2119 static void abort_prepare(BlkActionState *common, Error **errp)
2120 {
2121     error_setg(errp, "Transaction aborted using Abort action");
2122 }
2123 
2124 static void abort_commit(BlkActionState *common)
2125 {
2126     g_assert_not_reached(); /* this action never succeeds */
2127 }
2128 
2129 static const BlkActionOps actions[] = {
2130     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2131         .instance_size = sizeof(ExternalSnapshotState),
2132         .prepare  = external_snapshot_prepare,
2133         .commit   = external_snapshot_commit,
2134         .abort = external_snapshot_abort,
2135         .clean = external_snapshot_clean,
2136     },
2137     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2138         .instance_size = sizeof(ExternalSnapshotState),
2139         .prepare  = external_snapshot_prepare,
2140         .commit   = external_snapshot_commit,
2141         .abort = external_snapshot_abort,
2142         .clean = external_snapshot_clean,
2143     },
2144     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2145         .instance_size = sizeof(DriveBackupState),
2146         .prepare = drive_backup_prepare,
2147         .abort = drive_backup_abort,
2148         .clean = drive_backup_clean,
2149     },
2150     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2151         .instance_size = sizeof(BlockdevBackupState),
2152         .prepare = blockdev_backup_prepare,
2153         .abort = blockdev_backup_abort,
2154         .clean = blockdev_backup_clean,
2155     },
2156     [TRANSACTION_ACTION_KIND_ABORT] = {
2157         .instance_size = sizeof(BlkActionState),
2158         .prepare = abort_prepare,
2159         .commit = abort_commit,
2160     },
2161     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2162         .instance_size = sizeof(InternalSnapshotState),
2163         .prepare  = internal_snapshot_prepare,
2164         .abort = internal_snapshot_abort,
2165         .clean = internal_snapshot_clean,
2166     },
2167     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2168         .instance_size = sizeof(BlockDirtyBitmapState),
2169         .prepare = block_dirty_bitmap_add_prepare,
2170         .abort = block_dirty_bitmap_add_abort,
2171     },
2172     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2173         .instance_size = sizeof(BlockDirtyBitmapState),
2174         .prepare = block_dirty_bitmap_clear_prepare,
2175         .commit = block_dirty_bitmap_clear_commit,
2176         .abort = block_dirty_bitmap_clear_abort,
2177         .clean = block_dirty_bitmap_clear_clean,
2178     }
2179 };
2180 
2181 /**
2182  * Allocate a TransactionProperties structure if necessary, and fill
2183  * that structure with desired defaults if they are unset.
2184  */
2185 static TransactionProperties *get_transaction_properties(
2186     TransactionProperties *props)
2187 {
2188     if (!props) {
2189         props = g_new0(TransactionProperties, 1);
2190     }
2191 
2192     if (!props->has_completion_mode) {
2193         props->has_completion_mode = true;
2194         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2195     }
2196 
2197     return props;
2198 }
2199 
2200 /*
2201  * 'Atomic' group operations.  The operations are performed as a set, and if
2202  * any fail then we roll back all operations in the group.
2203  */
2204 void qmp_transaction(TransactionActionList *dev_list,
2205                      bool has_props,
2206                      struct TransactionProperties *props,
2207                      Error **errp)
2208 {
2209     TransactionActionList *dev_entry = dev_list;
2210     BlockJobTxn *block_job_txn = NULL;
2211     BlkActionState *state, *next;
2212     Error *local_err = NULL;
2213 
2214     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2215     QSIMPLEQ_INIT(&snap_bdrv_states);
2216 
2217     /* Does this transaction get canceled as a group on failure?
2218      * If not, we don't really need to make a BlockJobTxn.
2219      */
2220     props = get_transaction_properties(props);
2221     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2222         block_job_txn = block_job_txn_new();
2223     }
2224 
2225     /* drain all i/o before any operations */
2226     bdrv_drain_all();
2227 
2228     /* We don't do anything in this loop that commits us to the operations */
2229     while (NULL != dev_entry) {
2230         TransactionAction *dev_info = NULL;
2231         const BlkActionOps *ops;
2232 
2233         dev_info = dev_entry->value;
2234         dev_entry = dev_entry->next;
2235 
2236         assert(dev_info->type < ARRAY_SIZE(actions));
2237 
2238         ops = &actions[dev_info->type];
2239         assert(ops->instance_size > 0);
2240 
2241         state = g_malloc0(ops->instance_size);
2242         state->ops = ops;
2243         state->action = dev_info;
2244         state->block_job_txn = block_job_txn;
2245         state->txn_props = props;
2246         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2247 
2248         state->ops->prepare(state, &local_err);
2249         if (local_err) {
2250             error_propagate(errp, local_err);
2251             goto delete_and_fail;
2252         }
2253     }
2254 
2255     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2256         if (state->ops->commit) {
2257             state->ops->commit(state);
2258         }
2259     }
2260 
2261     /* success */
2262     goto exit;
2263 
2264 delete_and_fail:
2265     /* failure, and it is all-or-none; roll back all operations */
2266     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2267         if (state->ops->abort) {
2268             state->ops->abort(state);
2269         }
2270     }
2271 exit:
2272     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2273         if (state->ops->clean) {
2274             state->ops->clean(state);
2275         }
2276         g_free(state);
2277     }
2278     if (!has_props) {
2279         qapi_free_TransactionProperties(props);
2280     }
2281     block_job_txn_unref(block_job_txn);
2282 }
2283 
2284 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
2285 {
2286     Error *local_err = NULL;
2287 
2288     qmp_blockdev_open_tray(device, has_force, force, &local_err);
2289     if (local_err) {
2290         error_propagate(errp, local_err);
2291         return;
2292     }
2293 
2294     qmp_x_blockdev_remove_medium(device, errp);
2295 }
2296 
2297 void qmp_block_passwd(bool has_device, const char *device,
2298                       bool has_node_name, const char *node_name,
2299                       const char *password, Error **errp)
2300 {
2301     Error *local_err = NULL;
2302     BlockDriverState *bs;
2303     AioContext *aio_context;
2304 
2305     bs = bdrv_lookup_bs(has_device ? device : NULL,
2306                         has_node_name ? node_name : NULL,
2307                         &local_err);
2308     if (local_err) {
2309         error_propagate(errp, local_err);
2310         return;
2311     }
2312 
2313     aio_context = bdrv_get_aio_context(bs);
2314     aio_context_acquire(aio_context);
2315 
2316     bdrv_add_key(bs, password, errp);
2317 
2318     aio_context_release(aio_context);
2319 }
2320 
2321 void qmp_blockdev_open_tray(const char *device, bool has_force, bool force,
2322                             Error **errp)
2323 {
2324     BlockBackend *blk;
2325     bool locked;
2326 
2327     if (!has_force) {
2328         force = false;
2329     }
2330 
2331     blk = blk_by_name(device);
2332     if (!blk) {
2333         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2334                   "Device '%s' not found", device);
2335         return;
2336     }
2337 
2338     if (!blk_dev_has_removable_media(blk)) {
2339         error_setg(errp, "Device '%s' is not removable", device);
2340         return;
2341     }
2342 
2343     if (!blk_dev_has_tray(blk)) {
2344         /* Ignore this command on tray-less devices */
2345         return;
2346     }
2347 
2348     if (blk_dev_is_tray_open(blk)) {
2349         return;
2350     }
2351 
2352     locked = blk_dev_is_medium_locked(blk);
2353     if (locked) {
2354         blk_dev_eject_request(blk, force);
2355     }
2356 
2357     if (!locked || force) {
2358         blk_dev_change_media_cb(blk, false);
2359     }
2360 }
2361 
2362 void qmp_blockdev_close_tray(const char *device, Error **errp)
2363 {
2364     BlockBackend *blk;
2365 
2366     blk = blk_by_name(device);
2367     if (!blk) {
2368         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2369                   "Device '%s' not found", device);
2370         return;
2371     }
2372 
2373     if (!blk_dev_has_removable_media(blk)) {
2374         error_setg(errp, "Device '%s' is not removable", device);
2375         return;
2376     }
2377 
2378     if (!blk_dev_has_tray(blk)) {
2379         /* Ignore this command on tray-less devices */
2380         return;
2381     }
2382 
2383     if (!blk_dev_is_tray_open(blk)) {
2384         return;
2385     }
2386 
2387     blk_dev_change_media_cb(blk, true);
2388 }
2389 
2390 void qmp_x_blockdev_remove_medium(const char *device, Error **errp)
2391 {
2392     BlockBackend *blk;
2393     BlockDriverState *bs;
2394     AioContext *aio_context;
2395     bool has_device;
2396 
2397     blk = blk_by_name(device);
2398     if (!blk) {
2399         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2400                   "Device '%s' not found", device);
2401         return;
2402     }
2403 
2404     /* For BBs without a device, we can exchange the BDS tree at will */
2405     has_device = blk_get_attached_dev(blk);
2406 
2407     if (has_device && !blk_dev_has_removable_media(blk)) {
2408         error_setg(errp, "Device '%s' is not removable", device);
2409         return;
2410     }
2411 
2412     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2413         error_setg(errp, "Tray of device '%s' is not open", device);
2414         return;
2415     }
2416 
2417     bs = blk_bs(blk);
2418     if (!bs) {
2419         return;
2420     }
2421 
2422     aio_context = bdrv_get_aio_context(bs);
2423     aio_context_acquire(aio_context);
2424 
2425     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2426         goto out;
2427     }
2428 
2429     blk_remove_bs(blk);
2430 
2431     if (!blk_dev_has_tray(blk)) {
2432         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2433          * called at all); therefore, the medium needs to be ejected here.
2434          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2435          * value passed here (i.e. false). */
2436         blk_dev_change_media_cb(blk, false);
2437     }
2438 
2439 out:
2440     aio_context_release(aio_context);
2441 }
2442 
2443 static void qmp_blockdev_insert_anon_medium(const char *device,
2444                                             BlockDriverState *bs, Error **errp)
2445 {
2446     BlockBackend *blk;
2447     bool has_device;
2448 
2449     blk = blk_by_name(device);
2450     if (!blk) {
2451         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2452                   "Device '%s' not found", device);
2453         return;
2454     }
2455 
2456     /* For BBs without a device, we can exchange the BDS tree at will */
2457     has_device = blk_get_attached_dev(blk);
2458 
2459     if (has_device && !blk_dev_has_removable_media(blk)) {
2460         error_setg(errp, "Device '%s' is not removable", device);
2461         return;
2462     }
2463 
2464     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2465         error_setg(errp, "Tray of device '%s' is not open", device);
2466         return;
2467     }
2468 
2469     if (blk_bs(blk)) {
2470         error_setg(errp, "There already is a medium in device '%s'", device);
2471         return;
2472     }
2473 
2474     blk_insert_bs(blk, bs);
2475 
2476     if (!blk_dev_has_tray(blk)) {
2477         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2478          * called at all); therefore, the medium needs to be pushed into the
2479          * slot here.
2480          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2481          * value passed here (i.e. true). */
2482         blk_dev_change_media_cb(blk, true);
2483     }
2484 }
2485 
2486 void qmp_x_blockdev_insert_medium(const char *device, const char *node_name,
2487                                   Error **errp)
2488 {
2489     BlockDriverState *bs;
2490 
2491     bs = bdrv_find_node(node_name);
2492     if (!bs) {
2493         error_setg(errp, "Node '%s' not found", node_name);
2494         return;
2495     }
2496 
2497     if (bs->blk) {
2498         error_setg(errp, "Node '%s' is already in use by '%s'", node_name,
2499                    blk_name(bs->blk));
2500         return;
2501     }
2502 
2503     qmp_blockdev_insert_anon_medium(device, bs, errp);
2504 }
2505 
2506 void qmp_blockdev_change_medium(const char *device, const char *filename,
2507                                 bool has_format, const char *format,
2508                                 bool has_read_only,
2509                                 BlockdevChangeReadOnlyMode read_only,
2510                                 Error **errp)
2511 {
2512     BlockBackend *blk;
2513     BlockDriverState *medium_bs = NULL;
2514     int bdrv_flags, ret;
2515     QDict *options = NULL;
2516     Error *err = NULL;
2517 
2518     blk = blk_by_name(device);
2519     if (!blk) {
2520         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2521                   "Device '%s' not found", device);
2522         goto fail;
2523     }
2524 
2525     if (blk_bs(blk)) {
2526         blk_update_root_state(blk);
2527     }
2528 
2529     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2530     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2531         BDRV_O_PROTOCOL);
2532 
2533     if (!has_read_only) {
2534         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2535     }
2536 
2537     switch (read_only) {
2538     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2539         break;
2540 
2541     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2542         bdrv_flags &= ~BDRV_O_RDWR;
2543         break;
2544 
2545     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2546         bdrv_flags |= BDRV_O_RDWR;
2547         break;
2548 
2549     default:
2550         abort();
2551     }
2552 
2553     if (has_format) {
2554         options = qdict_new();
2555         qdict_put(options, "driver", qstring_from_str(format));
2556     }
2557 
2558     assert(!medium_bs);
2559     ret = bdrv_open(&medium_bs, filename, NULL, options, bdrv_flags, errp);
2560     if (ret < 0) {
2561         goto fail;
2562     }
2563 
2564     blk_apply_root_state(blk, medium_bs);
2565 
2566     bdrv_add_key(medium_bs, NULL, &err);
2567     if (err) {
2568         error_propagate(errp, err);
2569         goto fail;
2570     }
2571 
2572     qmp_blockdev_open_tray(device, false, false, &err);
2573     if (err) {
2574         error_propagate(errp, err);
2575         goto fail;
2576     }
2577 
2578     qmp_x_blockdev_remove_medium(device, &err);
2579     if (err) {
2580         error_propagate(errp, err);
2581         goto fail;
2582     }
2583 
2584     qmp_blockdev_insert_anon_medium(device, medium_bs, &err);
2585     if (err) {
2586         error_propagate(errp, err);
2587         goto fail;
2588     }
2589 
2590     qmp_blockdev_close_tray(device, errp);
2591 
2592 fail:
2593     /* If the medium has been inserted, the device has its own reference, so
2594      * ours must be relinquished; and if it has not been inserted successfully,
2595      * the reference must be relinquished anyway */
2596     bdrv_unref(medium_bs);
2597 }
2598 
2599 /* throttling disk I/O limits */
2600 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
2601                                int64_t bps_wr,
2602                                int64_t iops,
2603                                int64_t iops_rd,
2604                                int64_t iops_wr,
2605                                bool has_bps_max,
2606                                int64_t bps_max,
2607                                bool has_bps_rd_max,
2608                                int64_t bps_rd_max,
2609                                bool has_bps_wr_max,
2610                                int64_t bps_wr_max,
2611                                bool has_iops_max,
2612                                int64_t iops_max,
2613                                bool has_iops_rd_max,
2614                                int64_t iops_rd_max,
2615                                bool has_iops_wr_max,
2616                                int64_t iops_wr_max,
2617                                bool has_bps_max_length,
2618                                int64_t bps_max_length,
2619                                bool has_bps_rd_max_length,
2620                                int64_t bps_rd_max_length,
2621                                bool has_bps_wr_max_length,
2622                                int64_t bps_wr_max_length,
2623                                bool has_iops_max_length,
2624                                int64_t iops_max_length,
2625                                bool has_iops_rd_max_length,
2626                                int64_t iops_rd_max_length,
2627                                bool has_iops_wr_max_length,
2628                                int64_t iops_wr_max_length,
2629                                bool has_iops_size,
2630                                int64_t iops_size,
2631                                bool has_group,
2632                                const char *group, Error **errp)
2633 {
2634     ThrottleConfig cfg;
2635     BlockDriverState *bs;
2636     BlockBackend *blk;
2637     AioContext *aio_context;
2638 
2639     blk = blk_by_name(device);
2640     if (!blk) {
2641         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2642                   "Device '%s' not found", device);
2643         return;
2644     }
2645 
2646     aio_context = blk_get_aio_context(blk);
2647     aio_context_acquire(aio_context);
2648 
2649     bs = blk_bs(blk);
2650     if (!bs) {
2651         error_setg(errp, "Device '%s' has no medium", device);
2652         goto out;
2653     }
2654 
2655     throttle_config_init(&cfg);
2656     cfg.buckets[THROTTLE_BPS_TOTAL].avg = bps;
2657     cfg.buckets[THROTTLE_BPS_READ].avg  = bps_rd;
2658     cfg.buckets[THROTTLE_BPS_WRITE].avg = bps_wr;
2659 
2660     cfg.buckets[THROTTLE_OPS_TOTAL].avg = iops;
2661     cfg.buckets[THROTTLE_OPS_READ].avg  = iops_rd;
2662     cfg.buckets[THROTTLE_OPS_WRITE].avg = iops_wr;
2663 
2664     if (has_bps_max) {
2665         cfg.buckets[THROTTLE_BPS_TOTAL].max = bps_max;
2666     }
2667     if (has_bps_rd_max) {
2668         cfg.buckets[THROTTLE_BPS_READ].max = bps_rd_max;
2669     }
2670     if (has_bps_wr_max) {
2671         cfg.buckets[THROTTLE_BPS_WRITE].max = bps_wr_max;
2672     }
2673     if (has_iops_max) {
2674         cfg.buckets[THROTTLE_OPS_TOTAL].max = iops_max;
2675     }
2676     if (has_iops_rd_max) {
2677         cfg.buckets[THROTTLE_OPS_READ].max = iops_rd_max;
2678     }
2679     if (has_iops_wr_max) {
2680         cfg.buckets[THROTTLE_OPS_WRITE].max = iops_wr_max;
2681     }
2682 
2683     if (has_bps_max_length) {
2684         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = bps_max_length;
2685     }
2686     if (has_bps_rd_max_length) {
2687         cfg.buckets[THROTTLE_BPS_READ].burst_length = bps_rd_max_length;
2688     }
2689     if (has_bps_wr_max_length) {
2690         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = bps_wr_max_length;
2691     }
2692     if (has_iops_max_length) {
2693         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = iops_max_length;
2694     }
2695     if (has_iops_rd_max_length) {
2696         cfg.buckets[THROTTLE_OPS_READ].burst_length = iops_rd_max_length;
2697     }
2698     if (has_iops_wr_max_length) {
2699         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = iops_wr_max_length;
2700     }
2701 
2702     if (has_iops_size) {
2703         cfg.op_size = iops_size;
2704     }
2705 
2706     if (!throttle_is_valid(&cfg, errp)) {
2707         goto out;
2708     }
2709 
2710     if (throttle_enabled(&cfg)) {
2711         /* Enable I/O limits if they're not enabled yet, otherwise
2712          * just update the throttling group. */
2713         if (!bs->throttle_state) {
2714             bdrv_io_limits_enable(bs, has_group ? group : device);
2715         } else if (has_group) {
2716             bdrv_io_limits_update_group(bs, group);
2717         }
2718         /* Set the new throttling configuration */
2719         bdrv_set_io_limits(bs, &cfg);
2720     } else if (bs->throttle_state) {
2721         /* If all throttling settings are set to 0, disable I/O limits */
2722         bdrv_io_limits_disable(bs);
2723     }
2724 
2725 out:
2726     aio_context_release(aio_context);
2727 }
2728 
2729 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2730                                 bool has_granularity, uint32_t granularity,
2731                                 Error **errp)
2732 {
2733     AioContext *aio_context;
2734     BlockDriverState *bs;
2735 
2736     if (!name || name[0] == '\0') {
2737         error_setg(errp, "Bitmap name cannot be empty");
2738         return;
2739     }
2740 
2741     bs = bdrv_lookup_bs(node, node, errp);
2742     if (!bs) {
2743         return;
2744     }
2745 
2746     aio_context = bdrv_get_aio_context(bs);
2747     aio_context_acquire(aio_context);
2748 
2749     if (has_granularity) {
2750         if (granularity < 512 || !is_power_of_2(granularity)) {
2751             error_setg(errp, "Granularity must be power of 2 "
2752                              "and at least 512");
2753             goto out;
2754         }
2755     } else {
2756         /* Default to cluster size, if available: */
2757         granularity = bdrv_get_default_bitmap_granularity(bs);
2758     }
2759 
2760     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2761 
2762  out:
2763     aio_context_release(aio_context);
2764 }
2765 
2766 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2767                                    Error **errp)
2768 {
2769     AioContext *aio_context;
2770     BlockDriverState *bs;
2771     BdrvDirtyBitmap *bitmap;
2772 
2773     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2774     if (!bitmap || !bs) {
2775         return;
2776     }
2777 
2778     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2779         error_setg(errp,
2780                    "Bitmap '%s' is currently frozen and cannot be removed",
2781                    name);
2782         goto out;
2783     }
2784     bdrv_dirty_bitmap_make_anon(bitmap);
2785     bdrv_release_dirty_bitmap(bs, bitmap);
2786 
2787  out:
2788     aio_context_release(aio_context);
2789 }
2790 
2791 /**
2792  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2793  * immediately after a full backup operation.
2794  */
2795 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2796                                   Error **errp)
2797 {
2798     AioContext *aio_context;
2799     BdrvDirtyBitmap *bitmap;
2800     BlockDriverState *bs;
2801 
2802     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2803     if (!bitmap || !bs) {
2804         return;
2805     }
2806 
2807     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2808         error_setg(errp,
2809                    "Bitmap '%s' is currently frozen and cannot be modified",
2810                    name);
2811         goto out;
2812     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2813         error_setg(errp,
2814                    "Bitmap '%s' is currently disabled and cannot be cleared",
2815                    name);
2816         goto out;
2817     }
2818 
2819     bdrv_clear_dirty_bitmap(bitmap, NULL);
2820 
2821  out:
2822     aio_context_release(aio_context);
2823 }
2824 
2825 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2826 {
2827     const char *id = qdict_get_str(qdict, "id");
2828     BlockBackend *blk;
2829     BlockDriverState *bs;
2830     AioContext *aio_context;
2831     Error *local_err = NULL;
2832 
2833     bs = bdrv_find_node(id);
2834     if (bs) {
2835         qmp_x_blockdev_del(false, NULL, true, id, &local_err);
2836         if (local_err) {
2837             error_report_err(local_err);
2838         }
2839         return;
2840     }
2841 
2842     blk = blk_by_name(id);
2843     if (!blk) {
2844         error_report("Device '%s' not found", id);
2845         return;
2846     }
2847 
2848     if (!blk_legacy_dinfo(blk)) {
2849         error_report("Deleting device added with blockdev-add"
2850                      " is not supported");
2851         return;
2852     }
2853 
2854     aio_context = blk_get_aio_context(blk);
2855     aio_context_acquire(aio_context);
2856 
2857     bs = blk_bs(blk);
2858     if (bs) {
2859         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2860             error_report_err(local_err);
2861             aio_context_release(aio_context);
2862             return;
2863         }
2864 
2865         blk_remove_bs(blk);
2866     }
2867 
2868     /* Make the BlockBackend and the attached BlockDriverState anonymous */
2869     monitor_remove_blk(blk);
2870     if (blk_bs(blk)) {
2871         bdrv_make_anon(blk_bs(blk));
2872     }
2873 
2874     /* If this BlockBackend has a device attached to it, its refcount will be
2875      * decremented when the device is removed; otherwise we have to do so here.
2876      */
2877     if (blk_get_attached_dev(blk)) {
2878         /* Further I/O must not pause the guest */
2879         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2880                          BLOCKDEV_ON_ERROR_REPORT);
2881     } else {
2882         blk_unref(blk);
2883     }
2884 
2885     aio_context_release(aio_context);
2886 }
2887 
2888 void qmp_block_resize(bool has_device, const char *device,
2889                       bool has_node_name, const char *node_name,
2890                       int64_t size, Error **errp)
2891 {
2892     Error *local_err = NULL;
2893     BlockDriverState *bs;
2894     AioContext *aio_context;
2895     int ret;
2896 
2897     bs = bdrv_lookup_bs(has_device ? device : NULL,
2898                         has_node_name ? node_name : NULL,
2899                         &local_err);
2900     if (local_err) {
2901         error_propagate(errp, local_err);
2902         return;
2903     }
2904 
2905     aio_context = bdrv_get_aio_context(bs);
2906     aio_context_acquire(aio_context);
2907 
2908     if (!bdrv_is_first_non_filter(bs)) {
2909         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2910         goto out;
2911     }
2912 
2913     if (size < 0) {
2914         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2915         goto out;
2916     }
2917 
2918     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2919         error_setg(errp, QERR_DEVICE_IN_USE, device);
2920         goto out;
2921     }
2922 
2923     /* complete all in-flight operations before resizing the device */
2924     bdrv_drain_all();
2925 
2926     ret = bdrv_truncate(bs, size);
2927     switch (ret) {
2928     case 0:
2929         break;
2930     case -ENOMEDIUM:
2931         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2932         break;
2933     case -ENOTSUP:
2934         error_setg(errp, QERR_UNSUPPORTED);
2935         break;
2936     case -EACCES:
2937         error_setg(errp, "Device '%s' is read only", device);
2938         break;
2939     case -EBUSY:
2940         error_setg(errp, QERR_DEVICE_IN_USE, device);
2941         break;
2942     default:
2943         error_setg_errno(errp, -ret, "Could not resize");
2944         break;
2945     }
2946 
2947 out:
2948     aio_context_release(aio_context);
2949 }
2950 
2951 static void block_job_cb(void *opaque, int ret)
2952 {
2953     /* Note that this function may be executed from another AioContext besides
2954      * the QEMU main loop.  If you need to access anything that assumes the
2955      * QEMU global mutex, use a BH or introduce a mutex.
2956      */
2957 
2958     BlockDriverState *bs = opaque;
2959     const char *msg = NULL;
2960 
2961     trace_block_job_cb(bs, bs->job, ret);
2962 
2963     assert(bs->job);
2964 
2965     if (ret < 0) {
2966         msg = strerror(-ret);
2967     }
2968 
2969     if (block_job_is_cancelled(bs->job)) {
2970         block_job_event_cancelled(bs->job);
2971     } else {
2972         block_job_event_completed(bs->job, msg);
2973     }
2974 }
2975 
2976 void qmp_block_stream(const char *device,
2977                       bool has_base, const char *base,
2978                       bool has_backing_file, const char *backing_file,
2979                       bool has_speed, int64_t speed,
2980                       bool has_on_error, BlockdevOnError on_error,
2981                       Error **errp)
2982 {
2983     BlockBackend *blk;
2984     BlockDriverState *bs;
2985     BlockDriverState *base_bs = NULL;
2986     AioContext *aio_context;
2987     Error *local_err = NULL;
2988     const char *base_name = NULL;
2989 
2990     if (!has_on_error) {
2991         on_error = BLOCKDEV_ON_ERROR_REPORT;
2992     }
2993 
2994     blk = blk_by_name(device);
2995     if (!blk) {
2996         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2997                   "Device '%s' not found", device);
2998         return;
2999     }
3000 
3001     aio_context = blk_get_aio_context(blk);
3002     aio_context_acquire(aio_context);
3003 
3004     if (!blk_is_available(blk)) {
3005         error_setg(errp, "Device '%s' has no medium", device);
3006         goto out;
3007     }
3008     bs = blk_bs(blk);
3009 
3010     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_STREAM, errp)) {
3011         goto out;
3012     }
3013 
3014     if (has_base) {
3015         base_bs = bdrv_find_backing_image(bs, base);
3016         if (base_bs == NULL) {
3017             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3018             goto out;
3019         }
3020         assert(bdrv_get_aio_context(base_bs) == aio_context);
3021         base_name = base;
3022     }
3023 
3024     /* if we are streaming the entire chain, the result will have no backing
3025      * file, and specifying one is therefore an error */
3026     if (base_bs == NULL && has_backing_file) {
3027         error_setg(errp, "backing file specified, but streaming the "
3028                          "entire chain");
3029         goto out;
3030     }
3031 
3032     /* backing_file string overrides base bs filename */
3033     base_name = has_backing_file ? backing_file : base_name;
3034 
3035     stream_start(bs, base_bs, base_name, has_speed ? speed : 0,
3036                  on_error, block_job_cb, bs, &local_err);
3037     if (local_err) {
3038         error_propagate(errp, local_err);
3039         goto out;
3040     }
3041 
3042     trace_qmp_block_stream(bs, bs->job);
3043 
3044 out:
3045     aio_context_release(aio_context);
3046 }
3047 
3048 void qmp_block_commit(const char *device,
3049                       bool has_base, const char *base,
3050                       bool has_top, const char *top,
3051                       bool has_backing_file, const char *backing_file,
3052                       bool has_speed, int64_t speed,
3053                       Error **errp)
3054 {
3055     BlockBackend *blk;
3056     BlockDriverState *bs;
3057     BlockDriverState *base_bs, *top_bs;
3058     AioContext *aio_context;
3059     Error *local_err = NULL;
3060     /* This will be part of the QMP command, if/when the
3061      * BlockdevOnError change for blkmirror makes it in
3062      */
3063     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3064 
3065     if (!has_speed) {
3066         speed = 0;
3067     }
3068 
3069     /* Important Note:
3070      *  libvirt relies on the DeviceNotFound error class in order to probe for
3071      *  live commit feature versions; for this to work, we must make sure to
3072      *  perform the device lookup before any generic errors that may occur in a
3073      *  scenario in which all optional arguments are omitted. */
3074     blk = blk_by_name(device);
3075     if (!blk) {
3076         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3077                   "Device '%s' not found", device);
3078         return;
3079     }
3080 
3081     aio_context = blk_get_aio_context(blk);
3082     aio_context_acquire(aio_context);
3083 
3084     if (!blk_is_available(blk)) {
3085         error_setg(errp, "Device '%s' has no medium", device);
3086         goto out;
3087     }
3088     bs = blk_bs(blk);
3089 
3090     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3091         goto out;
3092     }
3093 
3094     /* default top_bs is the active layer */
3095     top_bs = bs;
3096 
3097     if (has_top && top) {
3098         if (strcmp(bs->filename, top) != 0) {
3099             top_bs = bdrv_find_backing_image(bs, top);
3100         }
3101     }
3102 
3103     if (top_bs == NULL) {
3104         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3105         goto out;
3106     }
3107 
3108     assert(bdrv_get_aio_context(top_bs) == aio_context);
3109 
3110     if (has_base && base) {
3111         base_bs = bdrv_find_backing_image(top_bs, base);
3112     } else {
3113         base_bs = bdrv_find_base(top_bs);
3114     }
3115 
3116     if (base_bs == NULL) {
3117         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3118         goto out;
3119     }
3120 
3121     assert(bdrv_get_aio_context(base_bs) == aio_context);
3122 
3123     if (bdrv_op_is_blocked(base_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3124         goto out;
3125     }
3126 
3127     /* Do not allow attempts to commit an image into itself */
3128     if (top_bs == base_bs) {
3129         error_setg(errp, "cannot commit an image into itself");
3130         goto out;
3131     }
3132 
3133     if (top_bs == bs) {
3134         if (has_backing_file) {
3135             error_setg(errp, "'backing-file' specified,"
3136                              " but 'top' is the active layer");
3137             goto out;
3138         }
3139         commit_active_start(bs, base_bs, speed, on_error, block_job_cb,
3140                             bs, &local_err);
3141     } else {
3142         commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
3143                      has_backing_file ? backing_file : NULL, &local_err);
3144     }
3145     if (local_err != NULL) {
3146         error_propagate(errp, local_err);
3147         goto out;
3148     }
3149 
3150 out:
3151     aio_context_release(aio_context);
3152 }
3153 
3154 static void do_drive_backup(const char *device, const char *target,
3155                             bool has_format, const char *format,
3156                             enum MirrorSyncMode sync,
3157                             bool has_mode, enum NewImageMode mode,
3158                             bool has_speed, int64_t speed,
3159                             bool has_bitmap, const char *bitmap,
3160                             bool has_on_source_error,
3161                             BlockdevOnError on_source_error,
3162                             bool has_on_target_error,
3163                             BlockdevOnError on_target_error,
3164                             BlockJobTxn *txn, Error **errp)
3165 {
3166     BlockBackend *blk;
3167     BlockDriverState *bs;
3168     BlockDriverState *target_bs;
3169     BlockDriverState *source = NULL;
3170     BdrvDirtyBitmap *bmap = NULL;
3171     AioContext *aio_context;
3172     QDict *options = NULL;
3173     Error *local_err = NULL;
3174     int flags;
3175     int64_t size;
3176     int ret;
3177 
3178     if (!has_speed) {
3179         speed = 0;
3180     }
3181     if (!has_on_source_error) {
3182         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3183     }
3184     if (!has_on_target_error) {
3185         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3186     }
3187     if (!has_mode) {
3188         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3189     }
3190 
3191     blk = blk_by_name(device);
3192     if (!blk) {
3193         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3194                   "Device '%s' not found", device);
3195         return;
3196     }
3197 
3198     aio_context = blk_get_aio_context(blk);
3199     aio_context_acquire(aio_context);
3200 
3201     /* Although backup_run has this check too, we need to use bs->drv below, so
3202      * do an early check redundantly. */
3203     if (!blk_is_available(blk)) {
3204         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
3205         goto out;
3206     }
3207     bs = blk_bs(blk);
3208 
3209     if (!has_format) {
3210         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
3211     }
3212 
3213     /* Early check to avoid creating target */
3214     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3215         goto out;
3216     }
3217 
3218     flags = bs->open_flags | BDRV_O_RDWR;
3219 
3220     /* See if we have a backing HD we can use to create our new image
3221      * on top of. */
3222     if (sync == MIRROR_SYNC_MODE_TOP) {
3223         source = backing_bs(bs);
3224         if (!source) {
3225             sync = MIRROR_SYNC_MODE_FULL;
3226         }
3227     }
3228     if (sync == MIRROR_SYNC_MODE_NONE) {
3229         source = bs;
3230     }
3231 
3232     size = bdrv_getlength(bs);
3233     if (size < 0) {
3234         error_setg_errno(errp, -size, "bdrv_getlength failed");
3235         goto out;
3236     }
3237 
3238     if (mode != NEW_IMAGE_MODE_EXISTING) {
3239         assert(format);
3240         if (source) {
3241             bdrv_img_create(target, format, source->filename,
3242                             source->drv->format_name, NULL,
3243                             size, flags, &local_err, false);
3244         } else {
3245             bdrv_img_create(target, format, NULL, NULL, NULL,
3246                             size, flags, &local_err, false);
3247         }
3248     }
3249 
3250     if (local_err) {
3251         error_propagate(errp, local_err);
3252         goto out;
3253     }
3254 
3255     if (format) {
3256         options = qdict_new();
3257         qdict_put(options, "driver", qstring_from_str(format));
3258     }
3259 
3260     target_bs = NULL;
3261     ret = bdrv_open(&target_bs, target, NULL, options, flags, &local_err);
3262     if (ret < 0) {
3263         error_propagate(errp, local_err);
3264         goto out;
3265     }
3266 
3267     bdrv_set_aio_context(target_bs, aio_context);
3268 
3269     if (has_bitmap) {
3270         bmap = bdrv_find_dirty_bitmap(bs, bitmap);
3271         if (!bmap) {
3272             error_setg(errp, "Bitmap '%s' could not be found", bitmap);
3273             bdrv_unref(target_bs);
3274             goto out;
3275         }
3276     }
3277 
3278     backup_start(bs, target_bs, speed, sync, bmap,
3279                  on_source_error, on_target_error,
3280                  block_job_cb, bs, txn, &local_err);
3281     if (local_err != NULL) {
3282         bdrv_unref(target_bs);
3283         error_propagate(errp, local_err);
3284         goto out;
3285     }
3286 
3287 out:
3288     aio_context_release(aio_context);
3289 }
3290 
3291 void qmp_drive_backup(const char *device, const char *target,
3292                       bool has_format, const char *format,
3293                       enum MirrorSyncMode sync,
3294                       bool has_mode, enum NewImageMode mode,
3295                       bool has_speed, int64_t speed,
3296                       bool has_bitmap, const char *bitmap,
3297                       bool has_on_source_error, BlockdevOnError on_source_error,
3298                       bool has_on_target_error, BlockdevOnError on_target_error,
3299                       Error **errp)
3300 {
3301     return do_drive_backup(device, target, has_format, format, sync,
3302                            has_mode, mode, has_speed, speed,
3303                            has_bitmap, bitmap,
3304                            has_on_source_error, on_source_error,
3305                            has_on_target_error, on_target_error,
3306                            NULL, errp);
3307 }
3308 
3309 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3310 {
3311     return bdrv_named_nodes_list(errp);
3312 }
3313 
3314 void do_blockdev_backup(const char *device, const char *target,
3315                          enum MirrorSyncMode sync,
3316                          bool has_speed, int64_t speed,
3317                          bool has_on_source_error,
3318                          BlockdevOnError on_source_error,
3319                          bool has_on_target_error,
3320                          BlockdevOnError on_target_error,
3321                          BlockJobTxn *txn, Error **errp)
3322 {
3323     BlockBackend *blk, *target_blk;
3324     BlockDriverState *bs;
3325     BlockDriverState *target_bs;
3326     Error *local_err = NULL;
3327     AioContext *aio_context;
3328 
3329     if (!has_speed) {
3330         speed = 0;
3331     }
3332     if (!has_on_source_error) {
3333         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3334     }
3335     if (!has_on_target_error) {
3336         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3337     }
3338 
3339     blk = blk_by_name(device);
3340     if (!blk) {
3341         error_setg(errp, "Device '%s' not found", device);
3342         return;
3343     }
3344 
3345     aio_context = blk_get_aio_context(blk);
3346     aio_context_acquire(aio_context);
3347 
3348     if (!blk_is_available(blk)) {
3349         error_setg(errp, "Device '%s' has no medium", device);
3350         goto out;
3351     }
3352     bs = blk_bs(blk);
3353 
3354     target_blk = blk_by_name(target);
3355     if (!target_blk) {
3356         error_setg(errp, "Device '%s' not found", target);
3357         goto out;
3358     }
3359 
3360     if (!blk_is_available(target_blk)) {
3361         error_setg(errp, "Device '%s' has no medium", target);
3362         goto out;
3363     }
3364     target_bs = blk_bs(target_blk);
3365 
3366     bdrv_ref(target_bs);
3367     bdrv_set_aio_context(target_bs, aio_context);
3368     backup_start(bs, target_bs, speed, sync, NULL, on_source_error,
3369                  on_target_error, block_job_cb, bs, txn, &local_err);
3370     if (local_err != NULL) {
3371         bdrv_unref(target_bs);
3372         error_propagate(errp, local_err);
3373     }
3374 out:
3375     aio_context_release(aio_context);
3376 }
3377 
3378 void qmp_blockdev_backup(const char *device, const char *target,
3379                          enum MirrorSyncMode sync,
3380                          bool has_speed, int64_t speed,
3381                          bool has_on_source_error,
3382                          BlockdevOnError on_source_error,
3383                          bool has_on_target_error,
3384                          BlockdevOnError on_target_error,
3385                          Error **errp)
3386 {
3387     do_blockdev_backup(device, target, sync, has_speed, speed,
3388                        has_on_source_error, on_source_error,
3389                        has_on_target_error, on_target_error,
3390                        NULL, errp);
3391 }
3392 
3393 /* Parameter check and block job starting for drive mirroring.
3394  * Caller should hold @device and @target's aio context (must be the same).
3395  **/
3396 static void blockdev_mirror_common(BlockDriverState *bs,
3397                                    BlockDriverState *target,
3398                                    bool has_replaces, const char *replaces,
3399                                    enum MirrorSyncMode sync,
3400                                    bool has_speed, int64_t speed,
3401                                    bool has_granularity, uint32_t granularity,
3402                                    bool has_buf_size, int64_t buf_size,
3403                                    bool has_on_source_error,
3404                                    BlockdevOnError on_source_error,
3405                                    bool has_on_target_error,
3406                                    BlockdevOnError on_target_error,
3407                                    bool has_unmap, bool unmap,
3408                                    Error **errp)
3409 {
3410 
3411     if (!has_speed) {
3412         speed = 0;
3413     }
3414     if (!has_on_source_error) {
3415         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3416     }
3417     if (!has_on_target_error) {
3418         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3419     }
3420     if (!has_granularity) {
3421         granularity = 0;
3422     }
3423     if (!has_buf_size) {
3424         buf_size = 0;
3425     }
3426     if (!has_unmap) {
3427         unmap = true;
3428     }
3429 
3430     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3431         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3432                    "a value in range [512B, 64MB]");
3433         return;
3434     }
3435     if (granularity & (granularity - 1)) {
3436         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3437                    "power of 2");
3438         return;
3439     }
3440 
3441     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3442         return;
3443     }
3444     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3445         return;
3446     }
3447     if (target->blk) {
3448         error_setg(errp, "Cannot mirror to an attached block device");
3449         return;
3450     }
3451 
3452     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3453         sync = MIRROR_SYNC_MODE_FULL;
3454     }
3455 
3456     /* pass the node name to replace to mirror start since it's loose coupling
3457      * and will allow to check whether the node still exist at mirror completion
3458      */
3459     mirror_start(bs, target,
3460                  has_replaces ? replaces : NULL,
3461                  speed, granularity, buf_size, sync,
3462                  on_source_error, on_target_error, unmap,
3463                  block_job_cb, bs, errp);
3464 }
3465 
3466 void qmp_drive_mirror(const char *device, const char *target,
3467                       bool has_format, const char *format,
3468                       bool has_node_name, const char *node_name,
3469                       bool has_replaces, const char *replaces,
3470                       enum MirrorSyncMode sync,
3471                       bool has_mode, enum NewImageMode mode,
3472                       bool has_speed, int64_t speed,
3473                       bool has_granularity, uint32_t granularity,
3474                       bool has_buf_size, int64_t buf_size,
3475                       bool has_on_source_error, BlockdevOnError on_source_error,
3476                       bool has_on_target_error, BlockdevOnError on_target_error,
3477                       bool has_unmap, bool unmap,
3478                       Error **errp)
3479 {
3480     BlockDriverState *bs;
3481     BlockBackend *blk;
3482     BlockDriverState *source, *target_bs;
3483     AioContext *aio_context;
3484     Error *local_err = NULL;
3485     QDict *options = NULL;
3486     int flags;
3487     int64_t size;
3488     int ret;
3489 
3490     blk = blk_by_name(device);
3491     if (!blk) {
3492         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3493                   "Device '%s' not found", device);
3494         return;
3495     }
3496 
3497     aio_context = blk_get_aio_context(blk);
3498     aio_context_acquire(aio_context);
3499 
3500     if (!blk_is_available(blk)) {
3501         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
3502         goto out;
3503     }
3504     bs = blk_bs(blk);
3505     if (!has_mode) {
3506         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3507     }
3508 
3509     if (!has_format) {
3510         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
3511     }
3512 
3513     flags = bs->open_flags | BDRV_O_RDWR;
3514     source = backing_bs(bs);
3515     if (!source && sync == MIRROR_SYNC_MODE_TOP) {
3516         sync = MIRROR_SYNC_MODE_FULL;
3517     }
3518     if (sync == MIRROR_SYNC_MODE_NONE) {
3519         source = bs;
3520     }
3521 
3522     size = bdrv_getlength(bs);
3523     if (size < 0) {
3524         error_setg_errno(errp, -size, "bdrv_getlength failed");
3525         goto out;
3526     }
3527 
3528     if (has_replaces) {
3529         BlockDriverState *to_replace_bs;
3530         AioContext *replace_aio_context;
3531         int64_t replace_size;
3532 
3533         if (!has_node_name) {
3534             error_setg(errp, "a node-name must be provided when replacing a"
3535                              " named node of the graph");
3536             goto out;
3537         }
3538 
3539         to_replace_bs = check_to_replace_node(bs, replaces, &local_err);
3540 
3541         if (!to_replace_bs) {
3542             error_propagate(errp, local_err);
3543             goto out;
3544         }
3545 
3546         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3547         aio_context_acquire(replace_aio_context);
3548         replace_size = bdrv_getlength(to_replace_bs);
3549         aio_context_release(replace_aio_context);
3550 
3551         if (size != replace_size) {
3552             error_setg(errp, "cannot replace image with a mirror image of "
3553                              "different size");
3554             goto out;
3555         }
3556     }
3557 
3558     if ((sync == MIRROR_SYNC_MODE_FULL || !source)
3559         && mode != NEW_IMAGE_MODE_EXISTING)
3560     {
3561         /* create new image w/o backing file */
3562         assert(format);
3563         bdrv_img_create(target, format,
3564                         NULL, NULL, NULL, size, flags, &local_err, false);
3565     } else {
3566         switch (mode) {
3567         case NEW_IMAGE_MODE_EXISTING:
3568             break;
3569         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3570             /* create new image with backing file */
3571             bdrv_img_create(target, format,
3572                             source->filename,
3573                             source->drv->format_name,
3574                             NULL, size, flags, &local_err, false);
3575             break;
3576         default:
3577             abort();
3578         }
3579     }
3580 
3581     if (local_err) {
3582         error_propagate(errp, local_err);
3583         goto out;
3584     }
3585 
3586     options = qdict_new();
3587     if (has_node_name) {
3588         qdict_put(options, "node-name", qstring_from_str(node_name));
3589     }
3590     if (format) {
3591         qdict_put(options, "driver", qstring_from_str(format));
3592     }
3593 
3594     /* Mirroring takes care of copy-on-write using the source's backing
3595      * file.
3596      */
3597     target_bs = NULL;
3598     ret = bdrv_open(&target_bs, target, NULL, options,
3599                     flags | BDRV_O_NO_BACKING, &local_err);
3600     if (ret < 0) {
3601         error_propagate(errp, local_err);
3602         goto out;
3603     }
3604 
3605     bdrv_set_aio_context(target_bs, aio_context);
3606 
3607     blockdev_mirror_common(bs, target_bs,
3608                            has_replaces, replaces, sync,
3609                            has_speed, speed,
3610                            has_granularity, granularity,
3611                            has_buf_size, buf_size,
3612                            has_on_source_error, on_source_error,
3613                            has_on_target_error, on_target_error,
3614                            has_unmap, unmap,
3615                            &local_err);
3616     if (local_err) {
3617         error_propagate(errp, local_err);
3618         bdrv_unref(target_bs);
3619     }
3620 out:
3621     aio_context_release(aio_context);
3622 }
3623 
3624 void qmp_blockdev_mirror(const char *device, const char *target,
3625                          bool has_replaces, const char *replaces,
3626                          MirrorSyncMode sync,
3627                          bool has_speed, int64_t speed,
3628                          bool has_granularity, uint32_t granularity,
3629                          bool has_buf_size, int64_t buf_size,
3630                          bool has_on_source_error,
3631                          BlockdevOnError on_source_error,
3632                          bool has_on_target_error,
3633                          BlockdevOnError on_target_error,
3634                          Error **errp)
3635 {
3636     BlockDriverState *bs;
3637     BlockBackend *blk;
3638     BlockDriverState *target_bs;
3639     AioContext *aio_context;
3640     Error *local_err = NULL;
3641 
3642     blk = blk_by_name(device);
3643     if (!blk) {
3644         error_setg(errp, "Device '%s' not found", device);
3645         return;
3646     }
3647     bs = blk_bs(blk);
3648 
3649     if (!bs) {
3650         error_setg(errp, "Device '%s' has no media", device);
3651         return;
3652     }
3653 
3654     target_bs = bdrv_lookup_bs(target, target, errp);
3655     if (!target_bs) {
3656         return;
3657     }
3658 
3659     aio_context = bdrv_get_aio_context(bs);
3660     aio_context_acquire(aio_context);
3661 
3662     bdrv_ref(target_bs);
3663     bdrv_set_aio_context(target_bs, aio_context);
3664 
3665     blockdev_mirror_common(bs, target_bs,
3666                            has_replaces, replaces, sync,
3667                            has_speed, speed,
3668                            has_granularity, granularity,
3669                            has_buf_size, buf_size,
3670                            has_on_source_error, on_source_error,
3671                            has_on_target_error, on_target_error,
3672                            true, true,
3673                            &local_err);
3674     if (local_err) {
3675         error_propagate(errp, local_err);
3676         bdrv_unref(target_bs);
3677     }
3678 
3679     aio_context_release(aio_context);
3680 }
3681 
3682 /* Get the block job for a given device name and acquire its AioContext */
3683 static BlockJob *find_block_job(const char *device, AioContext **aio_context,
3684                                 Error **errp)
3685 {
3686     BlockBackend *blk;
3687     BlockDriverState *bs;
3688 
3689     *aio_context = NULL;
3690 
3691     blk = blk_by_name(device);
3692     if (!blk) {
3693         goto notfound;
3694     }
3695 
3696     *aio_context = blk_get_aio_context(blk);
3697     aio_context_acquire(*aio_context);
3698 
3699     if (!blk_is_available(blk)) {
3700         goto notfound;
3701     }
3702     bs = blk_bs(blk);
3703 
3704     if (!bs->job) {
3705         goto notfound;
3706     }
3707 
3708     return bs->job;
3709 
3710 notfound:
3711     error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3712               "No active block job on device '%s'", device);
3713     if (*aio_context) {
3714         aio_context_release(*aio_context);
3715         *aio_context = NULL;
3716     }
3717     return NULL;
3718 }
3719 
3720 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3721 {
3722     AioContext *aio_context;
3723     BlockJob *job = find_block_job(device, &aio_context, errp);
3724 
3725     if (!job) {
3726         return;
3727     }
3728 
3729     block_job_set_speed(job, speed, errp);
3730     aio_context_release(aio_context);
3731 }
3732 
3733 void qmp_block_job_cancel(const char *device,
3734                           bool has_force, bool force, Error **errp)
3735 {
3736     AioContext *aio_context;
3737     BlockJob *job = find_block_job(device, &aio_context, errp);
3738 
3739     if (!job) {
3740         return;
3741     }
3742 
3743     if (!has_force) {
3744         force = false;
3745     }
3746 
3747     if (job->user_paused && !force) {
3748         error_setg(errp, "The block job for device '%s' is currently paused",
3749                    device);
3750         goto out;
3751     }
3752 
3753     trace_qmp_block_job_cancel(job);
3754     block_job_cancel(job);
3755 out:
3756     aio_context_release(aio_context);
3757 }
3758 
3759 void qmp_block_job_pause(const char *device, Error **errp)
3760 {
3761     AioContext *aio_context;
3762     BlockJob *job = find_block_job(device, &aio_context, errp);
3763 
3764     if (!job || job->user_paused) {
3765         return;
3766     }
3767 
3768     job->user_paused = true;
3769     trace_qmp_block_job_pause(job);
3770     block_job_pause(job);
3771     aio_context_release(aio_context);
3772 }
3773 
3774 void qmp_block_job_resume(const char *device, Error **errp)
3775 {
3776     AioContext *aio_context;
3777     BlockJob *job = find_block_job(device, &aio_context, errp);
3778 
3779     if (!job || !job->user_paused) {
3780         return;
3781     }
3782 
3783     job->user_paused = false;
3784     trace_qmp_block_job_resume(job);
3785     block_job_resume(job);
3786     aio_context_release(aio_context);
3787 }
3788 
3789 void qmp_block_job_complete(const char *device, Error **errp)
3790 {
3791     AioContext *aio_context;
3792     BlockJob *job = find_block_job(device, &aio_context, errp);
3793 
3794     if (!job) {
3795         return;
3796     }
3797 
3798     trace_qmp_block_job_complete(job);
3799     block_job_complete(job, errp);
3800     aio_context_release(aio_context);
3801 }
3802 
3803 void qmp_change_backing_file(const char *device,
3804                              const char *image_node_name,
3805                              const char *backing_file,
3806                              Error **errp)
3807 {
3808     BlockBackend *blk;
3809     BlockDriverState *bs = NULL;
3810     AioContext *aio_context;
3811     BlockDriverState *image_bs = NULL;
3812     Error *local_err = NULL;
3813     bool ro;
3814     int open_flags;
3815     int ret;
3816 
3817     blk = blk_by_name(device);
3818     if (!blk) {
3819         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3820                   "Device '%s' not found", device);
3821         return;
3822     }
3823 
3824     aio_context = blk_get_aio_context(blk);
3825     aio_context_acquire(aio_context);
3826 
3827     if (!blk_is_available(blk)) {
3828         error_setg(errp, "Device '%s' has no medium", device);
3829         goto out;
3830     }
3831     bs = blk_bs(blk);
3832 
3833     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3834     if (local_err) {
3835         error_propagate(errp, local_err);
3836         goto out;
3837     }
3838 
3839     if (!image_bs) {
3840         error_setg(errp, "image file not found");
3841         goto out;
3842     }
3843 
3844     if (bdrv_find_base(image_bs) == image_bs) {
3845         error_setg(errp, "not allowing backing file change on an image "
3846                          "without a backing file");
3847         goto out;
3848     }
3849 
3850     /* even though we are not necessarily operating on bs, we need it to
3851      * determine if block ops are currently prohibited on the chain */
3852     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3853         goto out;
3854     }
3855 
3856     /* final sanity check */
3857     if (!bdrv_chain_contains(bs, image_bs)) {
3858         error_setg(errp, "'%s' and image file are not in the same chain",
3859                    device);
3860         goto out;
3861     }
3862 
3863     /* if not r/w, reopen to make r/w */
3864     open_flags = image_bs->open_flags;
3865     ro = bdrv_is_read_only(image_bs);
3866 
3867     if (ro) {
3868         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3869         if (local_err) {
3870             error_propagate(errp, local_err);
3871             goto out;
3872         }
3873     }
3874 
3875     ret = bdrv_change_backing_file(image_bs, backing_file,
3876                                image_bs->drv ? image_bs->drv->format_name : "");
3877 
3878     if (ret < 0) {
3879         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3880                          backing_file);
3881         /* don't exit here, so we can try to restore open flags if
3882          * appropriate */
3883     }
3884 
3885     if (ro) {
3886         bdrv_reopen(image_bs, open_flags, &local_err);
3887         if (local_err) {
3888             error_propagate(errp, local_err); /* will preserve prior errp */
3889         }
3890     }
3891 
3892 out:
3893     aio_context_release(aio_context);
3894 }
3895 
3896 void hmp_drive_add_node(Monitor *mon, const char *optstr)
3897 {
3898     QemuOpts *opts;
3899     QDict *qdict;
3900     Error *local_err = NULL;
3901 
3902     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
3903     if (!opts) {
3904         return;
3905     }
3906 
3907     qdict = qemu_opts_to_qdict(opts, NULL);
3908 
3909     if (!qdict_get_try_str(qdict, "node-name")) {
3910         QDECREF(qdict);
3911         error_report("'node-name' needs to be specified");
3912         goto out;
3913     }
3914 
3915     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
3916     if (!bs) {
3917         error_report_err(local_err);
3918         goto out;
3919     }
3920 
3921     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3922 
3923 out:
3924     qemu_opts_del(opts);
3925 }
3926 
3927 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3928 {
3929     QmpOutputVisitor *ov = qmp_output_visitor_new();
3930     BlockDriverState *bs;
3931     BlockBackend *blk = NULL;
3932     QObject *obj;
3933     QDict *qdict;
3934     Error *local_err = NULL;
3935 
3936     /* TODO Sort it out in raw-posix and drive_new(): Reject aio=native with
3937      * cache.direct=false instead of silently switching to aio=threads, except
3938      * when called from drive_new().
3939      *
3940      * For now, simply forbidding the combination for all drivers will do. */
3941     if (options->has_aio && options->aio == BLOCKDEV_AIO_OPTIONS_NATIVE) {
3942         bool direct = options->has_cache &&
3943                       options->cache->has_direct &&
3944                       options->cache->direct;
3945         if (!direct) {
3946             error_setg(errp, "aio=native requires cache.direct=true");
3947             goto fail;
3948         }
3949     }
3950 
3951     visit_type_BlockdevOptions(qmp_output_get_visitor(ov), NULL, &options,
3952                                &local_err);
3953     if (local_err) {
3954         error_propagate(errp, local_err);
3955         goto fail;
3956     }
3957 
3958     obj = qmp_output_get_qobject(ov);
3959     qdict = qobject_to_qdict(obj);
3960 
3961     qdict_flatten(qdict);
3962 
3963     if (options->has_id) {
3964         blk = blockdev_init(NULL, qdict, &local_err);
3965         if (local_err) {
3966             error_propagate(errp, local_err);
3967             goto fail;
3968         }
3969 
3970         bs = blk_bs(blk);
3971     } else {
3972         if (!qdict_get_try_str(qdict, "node-name")) {
3973             error_setg(errp, "'id' and/or 'node-name' need to be specified for "
3974                        "the root node");
3975             goto fail;
3976         }
3977 
3978         bs = bds_tree_init(qdict, errp);
3979         if (!bs) {
3980             goto fail;
3981         }
3982 
3983         QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3984     }
3985 
3986     if (bs && bdrv_key_required(bs)) {
3987         if (blk) {
3988             monitor_remove_blk(blk);
3989             blk_unref(blk);
3990         } else {
3991             QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3992             bdrv_unref(bs);
3993         }
3994         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3995         goto fail;
3996     }
3997 
3998 fail:
3999     qmp_output_visitor_cleanup(ov);
4000 }
4001 
4002 void qmp_x_blockdev_del(bool has_id, const char *id,
4003                         bool has_node_name, const char *node_name, Error **errp)
4004 {
4005     AioContext *aio_context;
4006     BlockBackend *blk;
4007     BlockDriverState *bs;
4008 
4009     if (has_id && has_node_name) {
4010         error_setg(errp, "Only one of id and node-name must be specified");
4011         return;
4012     } else if (!has_id && !has_node_name) {
4013         error_setg(errp, "No block device specified");
4014         return;
4015     }
4016 
4017     if (has_id) {
4018         /* blk_by_name() never returns a BB that is not owned by the monitor */
4019         blk = blk_by_name(id);
4020         if (!blk) {
4021             error_setg(errp, "Cannot find block backend %s", id);
4022             return;
4023         }
4024         if (blk_get_refcnt(blk) > 1) {
4025             error_setg(errp, "Block backend %s is in use", id);
4026             return;
4027         }
4028         bs = blk_bs(blk);
4029         aio_context = blk_get_aio_context(blk);
4030     } else {
4031         bs = bdrv_find_node(node_name);
4032         if (!bs) {
4033             error_setg(errp, "Cannot find node %s", node_name);
4034             return;
4035         }
4036         blk = bs->blk;
4037         if (blk) {
4038             error_setg(errp, "Node %s is in use by %s",
4039                        node_name, blk_name(blk));
4040             return;
4041         }
4042         aio_context = bdrv_get_aio_context(bs);
4043     }
4044 
4045     aio_context_acquire(aio_context);
4046 
4047     if (bs) {
4048         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4049             goto out;
4050         }
4051 
4052         if (!blk && !bs->monitor_list.tqe_prev) {
4053             error_setg(errp, "Node %s is not owned by the monitor",
4054                        bs->node_name);
4055             goto out;
4056         }
4057 
4058         if (bs->refcnt > 1) {
4059             error_setg(errp, "Block device %s is in use",
4060                        bdrv_get_device_or_node_name(bs));
4061             goto out;
4062         }
4063     }
4064 
4065     if (blk) {
4066         monitor_remove_blk(blk);
4067         blk_unref(blk);
4068     } else {
4069         QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4070         bdrv_unref(bs);
4071     }
4072 
4073 out:
4074     aio_context_release(aio_context);
4075 }
4076 
4077 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4078 {
4079     BlockJobInfoList *head = NULL, **p_next = &head;
4080     BlockDriverState *bs;
4081 
4082     for (bs = bdrv_next(NULL); bs; bs = bdrv_next(bs)) {
4083         AioContext *aio_context = bdrv_get_aio_context(bs);
4084 
4085         aio_context_acquire(aio_context);
4086 
4087         if (bs->job) {
4088             BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
4089             elem->value = block_job_query(bs->job);
4090             *p_next = elem;
4091             p_next = &elem->next;
4092         }
4093 
4094         aio_context_release(aio_context);
4095     }
4096 
4097     return head;
4098 }
4099 
4100 QemuOptsList qemu_common_drive_opts = {
4101     .name = "drive",
4102     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4103     .desc = {
4104         {
4105             .name = "snapshot",
4106             .type = QEMU_OPT_BOOL,
4107             .help = "enable/disable snapshot mode",
4108         },{
4109             .name = "discard",
4110             .type = QEMU_OPT_STRING,
4111             .help = "discard operation (ignore/off, unmap/on)",
4112         },{
4113             .name = "aio",
4114             .type = QEMU_OPT_STRING,
4115             .help = "host AIO implementation (threads, native)",
4116         },{
4117             .name = "format",
4118             .type = QEMU_OPT_STRING,
4119             .help = "disk format (raw, qcow2, ...)",
4120         },{
4121             .name = "rerror",
4122             .type = QEMU_OPT_STRING,
4123             .help = "read error action",
4124         },{
4125             .name = "werror",
4126             .type = QEMU_OPT_STRING,
4127             .help = "write error action",
4128         },{
4129             .name = "read-only",
4130             .type = QEMU_OPT_BOOL,
4131             .help = "open drive file as read-only",
4132         },{
4133             .name = "throttling.iops-total",
4134             .type = QEMU_OPT_NUMBER,
4135             .help = "limit total I/O operations per second",
4136         },{
4137             .name = "throttling.iops-read",
4138             .type = QEMU_OPT_NUMBER,
4139             .help = "limit read operations per second",
4140         },{
4141             .name = "throttling.iops-write",
4142             .type = QEMU_OPT_NUMBER,
4143             .help = "limit write operations per second",
4144         },{
4145             .name = "throttling.bps-total",
4146             .type = QEMU_OPT_NUMBER,
4147             .help = "limit total bytes per second",
4148         },{
4149             .name = "throttling.bps-read",
4150             .type = QEMU_OPT_NUMBER,
4151             .help = "limit read bytes per second",
4152         },{
4153             .name = "throttling.bps-write",
4154             .type = QEMU_OPT_NUMBER,
4155             .help = "limit write bytes per second",
4156         },{
4157             .name = "throttling.iops-total-max",
4158             .type = QEMU_OPT_NUMBER,
4159             .help = "I/O operations burst",
4160         },{
4161             .name = "throttling.iops-read-max",
4162             .type = QEMU_OPT_NUMBER,
4163             .help = "I/O operations read burst",
4164         },{
4165             .name = "throttling.iops-write-max",
4166             .type = QEMU_OPT_NUMBER,
4167             .help = "I/O operations write burst",
4168         },{
4169             .name = "throttling.bps-total-max",
4170             .type = QEMU_OPT_NUMBER,
4171             .help = "total bytes burst",
4172         },{
4173             .name = "throttling.bps-read-max",
4174             .type = QEMU_OPT_NUMBER,
4175             .help = "total bytes read burst",
4176         },{
4177             .name = "throttling.bps-write-max",
4178             .type = QEMU_OPT_NUMBER,
4179             .help = "total bytes write burst",
4180         },{
4181             .name = "throttling.iops-total-max-length",
4182             .type = QEMU_OPT_NUMBER,
4183             .help = "length of the iops-total-max burst period, in seconds",
4184         },{
4185             .name = "throttling.iops-read-max-length",
4186             .type = QEMU_OPT_NUMBER,
4187             .help = "length of the iops-read-max burst period, in seconds",
4188         },{
4189             .name = "throttling.iops-write-max-length",
4190             .type = QEMU_OPT_NUMBER,
4191             .help = "length of the iops-write-max burst period, in seconds",
4192         },{
4193             .name = "throttling.bps-total-max-length",
4194             .type = QEMU_OPT_NUMBER,
4195             .help = "length of the bps-total-max burst period, in seconds",
4196         },{
4197             .name = "throttling.bps-read-max-length",
4198             .type = QEMU_OPT_NUMBER,
4199             .help = "length of the bps-read-max burst period, in seconds",
4200         },{
4201             .name = "throttling.bps-write-max-length",
4202             .type = QEMU_OPT_NUMBER,
4203             .help = "length of the bps-write-max burst period, in seconds",
4204         },{
4205             .name = "throttling.iops-size",
4206             .type = QEMU_OPT_NUMBER,
4207             .help = "when limiting by iops max size of an I/O in bytes",
4208         },{
4209             .name = "throttling.group",
4210             .type = QEMU_OPT_STRING,
4211             .help = "name of the block throttling group",
4212         },{
4213             .name = "copy-on-read",
4214             .type = QEMU_OPT_BOOL,
4215             .help = "copy read data from backing file into image file",
4216         },{
4217             .name = "detect-zeroes",
4218             .type = QEMU_OPT_STRING,
4219             .help = "try to optimize zero writes (off, on, unmap)",
4220         },{
4221             .name = "stats-account-invalid",
4222             .type = QEMU_OPT_BOOL,
4223             .help = "whether to account for invalid I/O operations "
4224                     "in the statistics",
4225         },{
4226             .name = "stats-account-failed",
4227             .type = QEMU_OPT_BOOL,
4228             .help = "whether to account for failed I/O operations "
4229                     "in the statistics",
4230         },
4231         { /* end of list */ }
4232     },
4233 };
4234 
4235 static QemuOptsList qemu_root_bds_opts = {
4236     .name = "root-bds",
4237     .head = QTAILQ_HEAD_INITIALIZER(qemu_root_bds_opts.head),
4238     .desc = {
4239         {
4240             .name = "discard",
4241             .type = QEMU_OPT_STRING,
4242             .help = "discard operation (ignore/off, unmap/on)",
4243         },{
4244             .name = "aio",
4245             .type = QEMU_OPT_STRING,
4246             .help = "host AIO implementation (threads, native)",
4247         },{
4248             .name = "read-only",
4249             .type = QEMU_OPT_BOOL,
4250             .help = "open drive file as read-only",
4251         },{
4252             .name = "copy-on-read",
4253             .type = QEMU_OPT_BOOL,
4254             .help = "copy read data from backing file into image file",
4255         },{
4256             .name = "detect-zeroes",
4257             .type = QEMU_OPT_STRING,
4258             .help = "try to optimize zero writes (off, on, unmap)",
4259         },
4260         { /* end of list */ }
4261     },
4262 };
4263 
4264 QemuOptsList qemu_drive_opts = {
4265     .name = "drive",
4266     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4267     .desc = {
4268         /*
4269          * no elements => accept any params
4270          * validation will happen later
4271          */
4272         { /* end of list */ }
4273     },
4274 };
4275