xref: /openbmc/qemu/blockdev.c (revision fa59483d)
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/qdict.h"
39 #include "block/throttle-groups.h"
40 #include "monitor/monitor.h"
41 #include "qemu/error-report.h"
42 #include "qemu/option.h"
43 #include "qemu/config-file.h"
44 #include "qapi/qapi-commands-block.h"
45 #include "qapi/qapi-commands-transaction.h"
46 #include "qapi/qapi-visit-block-core.h"
47 #include "qapi/qmp/qdict.h"
48 #include "qapi/qmp/qnum.h"
49 #include "qapi/qmp/qstring.h"
50 #include "qapi/error.h"
51 #include "qapi/qmp/qerror.h"
52 #include "qapi/qmp/qlist.h"
53 #include "qapi/qobject-output-visitor.h"
54 #include "sysemu/sysemu.h"
55 #include "sysemu/iothread.h"
56 #include "block/block_int.h"
57 #include "block/trace.h"
58 #include "sysemu/arch_init.h"
59 #include "sysemu/qtest.h"
60 #include "qemu/cutils.h"
61 #include "qemu/help_option.h"
62 #include "qemu/throttle-options.h"
63 
64 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
65     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
66 
67 static int do_open_tray(const char *blk_name, const char *qdev_id,
68                         bool force, Error **errp);
69 static void blockdev_remove_medium(bool has_device, const char *device,
70                                    bool has_id, const char *id, Error **errp);
71 static void blockdev_insert_medium(bool has_device, const char *device,
72                                    bool has_id, const char *id,
73                                    const char *node_name, Error **errp);
74 
75 static const char *const if_name[IF_COUNT] = {
76     [IF_NONE] = "none",
77     [IF_IDE] = "ide",
78     [IF_SCSI] = "scsi",
79     [IF_FLOPPY] = "floppy",
80     [IF_PFLASH] = "pflash",
81     [IF_MTD] = "mtd",
82     [IF_SD] = "sd",
83     [IF_VIRTIO] = "virtio",
84     [IF_XEN] = "xen",
85 };
86 
87 static int if_max_devs[IF_COUNT] = {
88     /*
89      * Do not change these numbers!  They govern how drive option
90      * index maps to unit and bus.  That mapping is ABI.
91      *
92      * All controllers used to implement if=T drives need to support
93      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
94      * Otherwise, some index values map to "impossible" bus, unit
95      * values.
96      *
97      * For instance, if you change [IF_SCSI] to 255, -drive
98      * if=scsi,index=12 no longer means bus=1,unit=5, but
99      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
100      * the drive can't be set up.  Regression.
101      */
102     [IF_IDE] = 2,
103     [IF_SCSI] = 7,
104 };
105 
106 /**
107  * Boards may call this to offer board-by-board overrides
108  * of the default, global values.
109  */
110 void override_max_devs(BlockInterfaceType type, int max_devs)
111 {
112     BlockBackend *blk;
113     DriveInfo *dinfo;
114 
115     if (max_devs <= 0) {
116         return;
117     }
118 
119     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
120         dinfo = blk_legacy_dinfo(blk);
121         if (dinfo->type == type) {
122             fprintf(stderr, "Cannot override units-per-bus property of"
123                     " the %s interface, because a drive of that type has"
124                     " already been added.\n", if_name[type]);
125             g_assert_not_reached();
126         }
127     }
128 
129     if_max_devs[type] = max_devs;
130 }
131 
132 /*
133  * We automatically delete the drive when a device using it gets
134  * unplugged.  Questionable feature, but we can't just drop it.
135  * Device models call blockdev_mark_auto_del() to schedule the
136  * automatic deletion, and generic qdev code calls blockdev_auto_del()
137  * when deletion is actually safe.
138  */
139 void blockdev_mark_auto_del(BlockBackend *blk)
140 {
141     DriveInfo *dinfo = blk_legacy_dinfo(blk);
142     BlockDriverState *bs = blk_bs(blk);
143     AioContext *aio_context;
144 
145     if (!dinfo) {
146         return;
147     }
148 
149     if (bs) {
150         aio_context = bdrv_get_aio_context(bs);
151         aio_context_acquire(aio_context);
152 
153         if (bs->job) {
154             job_cancel(&bs->job->job, false);
155         }
156 
157         aio_context_release(aio_context);
158     }
159 
160     dinfo->auto_del = 1;
161 }
162 
163 void blockdev_auto_del(BlockBackend *blk)
164 {
165     DriveInfo *dinfo = blk_legacy_dinfo(blk);
166 
167     if (dinfo && dinfo->auto_del) {
168         monitor_remove_blk(blk);
169         blk_unref(blk);
170     }
171 }
172 
173 /**
174  * Returns the current mapping of how many units per bus
175  * a particular interface can support.
176  *
177  *  A positive integer indicates n units per bus.
178  *  0 implies the mapping has not been established.
179  * -1 indicates an invalid BlockInterfaceType was given.
180  */
181 int drive_get_max_devs(BlockInterfaceType type)
182 {
183     if (type >= IF_IDE && type < IF_COUNT) {
184         return if_max_devs[type];
185     }
186 
187     return -1;
188 }
189 
190 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
191 {
192     int max_devs = if_max_devs[type];
193     return max_devs ? index / max_devs : 0;
194 }
195 
196 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
197 {
198     int max_devs = if_max_devs[type];
199     return max_devs ? index % max_devs : index;
200 }
201 
202 QemuOpts *drive_def(const char *optstr)
203 {
204     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
205 }
206 
207 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
208                     const char *optstr)
209 {
210     QemuOpts *opts;
211 
212     opts = drive_def(optstr);
213     if (!opts) {
214         return NULL;
215     }
216     if (type != IF_DEFAULT) {
217         qemu_opt_set(opts, "if", if_name[type], &error_abort);
218     }
219     if (index >= 0) {
220         qemu_opt_set_number(opts, "index", index, &error_abort);
221     }
222     if (file)
223         qemu_opt_set(opts, "file", file, &error_abort);
224     return opts;
225 }
226 
227 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
228 {
229     BlockBackend *blk;
230     DriveInfo *dinfo;
231 
232     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
233         dinfo = blk_legacy_dinfo(blk);
234         if (dinfo && dinfo->type == type
235             && dinfo->bus == bus && dinfo->unit == unit) {
236             return dinfo;
237         }
238     }
239 
240     return NULL;
241 }
242 
243 void drive_check_orphaned(void)
244 {
245     BlockBackend *blk;
246     DriveInfo *dinfo;
247     Location loc;
248     bool orphans = false;
249 
250     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
251         dinfo = blk_legacy_dinfo(blk);
252         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
253             dinfo->type != IF_NONE) {
254             loc_push_none(&loc);
255             qemu_opts_loc_restore(dinfo->opts);
256             error_report("machine type does not support"
257                          " if=%s,bus=%d,unit=%d",
258                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
259             loc_pop(&loc);
260             orphans = true;
261         }
262     }
263 
264     if (orphans) {
265         exit(1);
266     }
267 }
268 
269 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
270 {
271     return drive_get(type,
272                      drive_index_to_bus_id(type, index),
273                      drive_index_to_unit_id(type, index));
274 }
275 
276 int drive_get_max_bus(BlockInterfaceType type)
277 {
278     int max_bus;
279     BlockBackend *blk;
280     DriveInfo *dinfo;
281 
282     max_bus = -1;
283     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
284         dinfo = blk_legacy_dinfo(blk);
285         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
286             max_bus = dinfo->bus;
287         }
288     }
289     return max_bus;
290 }
291 
292 /* Get a block device.  This should only be used for single-drive devices
293    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
294    appropriate bus.  */
295 DriveInfo *drive_get_next(BlockInterfaceType type)
296 {
297     static int next_block_unit[IF_COUNT];
298 
299     return drive_get(type, 0, next_block_unit[type]++);
300 }
301 
302 static void bdrv_format_print(void *opaque, const char *name)
303 {
304     error_printf(" %s", name);
305 }
306 
307 typedef struct {
308     QEMUBH *bh;
309     BlockDriverState *bs;
310 } BDRVPutRefBH;
311 
312 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
313 {
314     if (!strcmp(buf, "ignore")) {
315         return BLOCKDEV_ON_ERROR_IGNORE;
316     } else if (!is_read && !strcmp(buf, "enospc")) {
317         return BLOCKDEV_ON_ERROR_ENOSPC;
318     } else if (!strcmp(buf, "stop")) {
319         return BLOCKDEV_ON_ERROR_STOP;
320     } else if (!strcmp(buf, "report")) {
321         return BLOCKDEV_ON_ERROR_REPORT;
322     } else {
323         error_setg(errp, "'%s' invalid %s error action",
324                    buf, is_read ? "read" : "write");
325         return -1;
326     }
327 }
328 
329 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
330                                   Error **errp)
331 {
332     const QListEntry *entry;
333     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
334         switch (qobject_type(entry->value)) {
335 
336         case QTYPE_QSTRING: {
337             unsigned long long length;
338             const char *str = qstring_get_str(qobject_to(QString,
339                                                          entry->value));
340             if (parse_uint_full(str, &length, 10) == 0 &&
341                 length > 0 && length <= UINT_MAX) {
342                 block_acct_add_interval(stats, (unsigned) length);
343             } else {
344                 error_setg(errp, "Invalid interval length: %s", str);
345                 return false;
346             }
347             break;
348         }
349 
350         case QTYPE_QNUM: {
351             int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
352 
353             if (length > 0 && length <= UINT_MAX) {
354                 block_acct_add_interval(stats, (unsigned) length);
355             } else {
356                 error_setg(errp, "Invalid interval length: %" PRId64, length);
357                 return false;
358             }
359             break;
360         }
361 
362         default:
363             error_setg(errp, "The specification of stats-intervals is invalid");
364             return false;
365         }
366     }
367     return true;
368 }
369 
370 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
371 
372 /* All parameters but @opts are optional and may be set to NULL. */
373 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
374     const char **throttling_group, ThrottleConfig *throttle_cfg,
375     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
376 {
377     Error *local_error = NULL;
378     const char *aio;
379 
380     if (bdrv_flags) {
381         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
382             *bdrv_flags |= BDRV_O_COPY_ON_READ;
383         }
384 
385         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
386             if (!strcmp(aio, "native")) {
387                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
388             } else if (!strcmp(aio, "threads")) {
389                 /* this is the default */
390             } else {
391                error_setg(errp, "invalid aio option");
392                return;
393             }
394         }
395     }
396 
397     /* disk I/O throttling */
398     if (throttling_group) {
399         *throttling_group = qemu_opt_get(opts, "throttling.group");
400     }
401 
402     if (throttle_cfg) {
403         throttle_config_init(throttle_cfg);
404         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
405             qemu_opt_get_number(opts, "throttling.bps-total", 0);
406         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
407             qemu_opt_get_number(opts, "throttling.bps-read", 0);
408         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
409             qemu_opt_get_number(opts, "throttling.bps-write", 0);
410         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
411             qemu_opt_get_number(opts, "throttling.iops-total", 0);
412         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
413             qemu_opt_get_number(opts, "throttling.iops-read", 0);
414         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
415             qemu_opt_get_number(opts, "throttling.iops-write", 0);
416 
417         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
418             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
419         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
420             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
421         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
422             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
423         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
424             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
425         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
426             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
427         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
428             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
429 
430         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
431             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
432         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
433             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
434         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
435             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
436         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
437             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
438         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
439             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
440         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
441             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
442 
443         throttle_cfg->op_size =
444             qemu_opt_get_number(opts, "throttling.iops-size", 0);
445 
446         if (!throttle_is_valid(throttle_cfg, errp)) {
447             return;
448         }
449     }
450 
451     if (detect_zeroes) {
452         *detect_zeroes =
453             qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
454                             qemu_opt_get(opts, "detect-zeroes"),
455                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
456                             &local_error);
457         if (local_error) {
458             error_propagate(errp, local_error);
459             return;
460         }
461     }
462 }
463 
464 /* Takes the ownership of bs_opts */
465 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
466                                    Error **errp)
467 {
468     const char *buf;
469     int bdrv_flags = 0;
470     int on_read_error, on_write_error;
471     bool account_invalid, account_failed;
472     bool writethrough, read_only;
473     BlockBackend *blk;
474     BlockDriverState *bs;
475     ThrottleConfig cfg;
476     int snapshot = 0;
477     Error *error = NULL;
478     QemuOpts *opts;
479     QDict *interval_dict = NULL;
480     QList *interval_list = NULL;
481     const char *id;
482     BlockdevDetectZeroesOptions detect_zeroes =
483         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
484     const char *throttling_group = NULL;
485 
486     /* Check common options by copying from bs_opts to opts, all other options
487      * stay in bs_opts for processing by bdrv_open(). */
488     id = qdict_get_try_str(bs_opts, "id");
489     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
490     if (error) {
491         error_propagate(errp, error);
492         goto err_no_opts;
493     }
494 
495     qemu_opts_absorb_qdict(opts, bs_opts, &error);
496     if (error) {
497         error_propagate(errp, error);
498         goto early_err;
499     }
500 
501     if (id) {
502         qdict_del(bs_opts, "id");
503     }
504 
505     /* extract parameters */
506     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
507 
508     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
509     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
510 
511     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
512 
513     id = qemu_opts_id(opts);
514 
515     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
516     qdict_array_split(interval_dict, &interval_list);
517 
518     if (qdict_size(interval_dict) != 0) {
519         error_setg(errp, "Invalid option stats-intervals.%s",
520                    qdict_first(interval_dict)->key);
521         goto early_err;
522     }
523 
524     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
525                                     &detect_zeroes, &error);
526     if (error) {
527         error_propagate(errp, error);
528         goto early_err;
529     }
530 
531     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
532         if (is_help_option(buf)) {
533             error_printf("Supported formats:");
534             bdrv_iterate_format(bdrv_format_print, NULL);
535             error_printf("\n");
536             goto early_err;
537         }
538 
539         if (qdict_haskey(bs_opts, "driver")) {
540             error_setg(errp, "Cannot specify both 'driver' and 'format'");
541             goto early_err;
542         }
543         qdict_put_str(bs_opts, "driver", buf);
544     }
545 
546     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
547     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
548         on_write_error = parse_block_error_action(buf, 0, &error);
549         if (error) {
550             error_propagate(errp, error);
551             goto early_err;
552         }
553     }
554 
555     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
556     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
557         on_read_error = parse_block_error_action(buf, 1, &error);
558         if (error) {
559             error_propagate(errp, error);
560             goto early_err;
561         }
562     }
563 
564     if (snapshot) {
565         bdrv_flags |= BDRV_O_SNAPSHOT;
566     }
567 
568     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
569 
570     /* init */
571     if ((!file || !*file) && !qdict_size(bs_opts)) {
572         BlockBackendRootState *blk_rs;
573 
574         blk = blk_new(0, BLK_PERM_ALL);
575         blk_rs = blk_get_root_state(blk);
576         blk_rs->open_flags    = bdrv_flags;
577         blk_rs->read_only     = read_only;
578         blk_rs->detect_zeroes = detect_zeroes;
579 
580         qobject_unref(bs_opts);
581     } else {
582         if (file && !*file) {
583             file = NULL;
584         }
585 
586         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
587          * with other callers) rather than what we want as the real defaults.
588          * Apply the defaults here instead. */
589         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
590         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
591         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
592                               read_only ? "on" : "off");
593         qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
594         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
595 
596         if (runstate_check(RUN_STATE_INMIGRATE)) {
597             bdrv_flags |= BDRV_O_INACTIVE;
598         }
599 
600         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
601         if (!blk) {
602             goto err_no_bs_opts;
603         }
604         bs = blk_bs(blk);
605 
606         bs->detect_zeroes = detect_zeroes;
607 
608         block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
609 
610         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
611             blk_unref(blk);
612             blk = NULL;
613             goto err_no_bs_opts;
614         }
615     }
616 
617     /* disk I/O throttling */
618     if (throttle_enabled(&cfg)) {
619         if (!throttling_group) {
620             throttling_group = id;
621         }
622         blk_io_limits_enable(blk, throttling_group);
623         blk_set_io_limits(blk, &cfg);
624     }
625 
626     blk_set_enable_write_cache(blk, !writethrough);
627     blk_set_on_error(blk, on_read_error, on_write_error);
628 
629     if (!monitor_add_blk(blk, id, errp)) {
630         blk_unref(blk);
631         blk = NULL;
632         goto err_no_bs_opts;
633     }
634 
635 err_no_bs_opts:
636     qemu_opts_del(opts);
637     qobject_unref(interval_dict);
638     qobject_unref(interval_list);
639     return blk;
640 
641 early_err:
642     qemu_opts_del(opts);
643     qobject_unref(interval_dict);
644     qobject_unref(interval_list);
645 err_no_opts:
646     qobject_unref(bs_opts);
647     return NULL;
648 }
649 
650 /* Takes the ownership of bs_opts */
651 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
652 {
653     int bdrv_flags = 0;
654 
655     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
656      * with other callers) rather than what we want as the real defaults.
657      * Apply the defaults here instead. */
658     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
659     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
660     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
661 
662     if (runstate_check(RUN_STATE_INMIGRATE)) {
663         bdrv_flags |= BDRV_O_INACTIVE;
664     }
665 
666     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
667 }
668 
669 void blockdev_close_all_bdrv_states(void)
670 {
671     BlockDriverState *bs, *next_bs;
672 
673     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
674         AioContext *ctx = bdrv_get_aio_context(bs);
675 
676         aio_context_acquire(ctx);
677         bdrv_unref(bs);
678         aio_context_release(ctx);
679     }
680 }
681 
682 /* Iterates over the list of monitor-owned BlockDriverStates */
683 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
684 {
685     return bs ? QTAILQ_NEXT(bs, monitor_list)
686               : QTAILQ_FIRST(&monitor_bdrv_states);
687 }
688 
689 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
690                             Error **errp)
691 {
692     const char *value;
693 
694     value = qemu_opt_get(opts, from);
695     if (value) {
696         if (qemu_opt_find(opts, to)) {
697             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
698                        "same time", to, from);
699             return;
700         }
701     }
702 
703     /* rename all items in opts */
704     while ((value = qemu_opt_get(opts, from))) {
705         qemu_opt_set(opts, to, value, &error_abort);
706         qemu_opt_unset(opts, from);
707     }
708 }
709 
710 QemuOptsList qemu_legacy_drive_opts = {
711     .name = "drive",
712     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
713     .desc = {
714         {
715             .name = "bus",
716             .type = QEMU_OPT_NUMBER,
717             .help = "bus number",
718         },{
719             .name = "unit",
720             .type = QEMU_OPT_NUMBER,
721             .help = "unit number (i.e. lun for scsi)",
722         },{
723             .name = "index",
724             .type = QEMU_OPT_NUMBER,
725             .help = "index number",
726         },{
727             .name = "media",
728             .type = QEMU_OPT_STRING,
729             .help = "media type (disk, cdrom)",
730         },{
731             .name = "if",
732             .type = QEMU_OPT_STRING,
733             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
734         },{
735             .name = "file",
736             .type = QEMU_OPT_STRING,
737             .help = "file name",
738         },
739 
740         /* Options that are passed on, but have special semantics with -drive */
741         {
742             .name = BDRV_OPT_READ_ONLY,
743             .type = QEMU_OPT_BOOL,
744             .help = "open drive file as read-only",
745         },{
746             .name = "rerror",
747             .type = QEMU_OPT_STRING,
748             .help = "read error action",
749         },{
750             .name = "werror",
751             .type = QEMU_OPT_STRING,
752             .help = "write error action",
753         },{
754             .name = "copy-on-read",
755             .type = QEMU_OPT_BOOL,
756             .help = "copy read data from backing file into image file",
757         },
758 
759         { /* end of list */ }
760     },
761 };
762 
763 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
764                      Error **errp)
765 {
766     const char *value;
767     BlockBackend *blk;
768     DriveInfo *dinfo = NULL;
769     QDict *bs_opts;
770     QemuOpts *legacy_opts;
771     DriveMediaType media = MEDIA_DISK;
772     BlockInterfaceType type;
773     int max_devs, bus_id, unit_id, index;
774     const char *werror, *rerror;
775     bool read_only = false;
776     bool copy_on_read;
777     const char *filename;
778     Error *local_err = NULL;
779     int i;
780 
781     /* Change legacy command line options into QMP ones */
782     static const struct {
783         const char *from;
784         const char *to;
785     } opt_renames[] = {
786         { "iops",           "throttling.iops-total" },
787         { "iops_rd",        "throttling.iops-read" },
788         { "iops_wr",        "throttling.iops-write" },
789 
790         { "bps",            "throttling.bps-total" },
791         { "bps_rd",         "throttling.bps-read" },
792         { "bps_wr",         "throttling.bps-write" },
793 
794         { "iops_max",       "throttling.iops-total-max" },
795         { "iops_rd_max",    "throttling.iops-read-max" },
796         { "iops_wr_max",    "throttling.iops-write-max" },
797 
798         { "bps_max",        "throttling.bps-total-max" },
799         { "bps_rd_max",     "throttling.bps-read-max" },
800         { "bps_wr_max",     "throttling.bps-write-max" },
801 
802         { "iops_size",      "throttling.iops-size" },
803 
804         { "group",          "throttling.group" },
805 
806         { "readonly",       BDRV_OPT_READ_ONLY },
807     };
808 
809     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
810         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
811                         &local_err);
812         if (local_err) {
813             error_propagate(errp, local_err);
814             return NULL;
815         }
816     }
817 
818     value = qemu_opt_get(all_opts, "cache");
819     if (value) {
820         int flags = 0;
821         bool writethrough;
822 
823         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
824             error_setg(errp, "invalid cache option");
825             return NULL;
826         }
827 
828         /* Specific options take precedence */
829         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
830             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
831                               !writethrough, &error_abort);
832         }
833         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
834             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
835                               !!(flags & BDRV_O_NOCACHE), &error_abort);
836         }
837         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
838             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
839                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
840         }
841         qemu_opt_unset(all_opts, "cache");
842     }
843 
844     /* Get a QDict for processing the options */
845     bs_opts = qdict_new();
846     qemu_opts_to_qdict(all_opts, bs_opts);
847 
848     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
849                                    &error_abort);
850     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
851     if (local_err) {
852         error_propagate(errp, local_err);
853         goto fail;
854     }
855 
856     /* Media type */
857     value = qemu_opt_get(legacy_opts, "media");
858     if (value) {
859         if (!strcmp(value, "disk")) {
860             media = MEDIA_DISK;
861         } else if (!strcmp(value, "cdrom")) {
862             media = MEDIA_CDROM;
863             read_only = true;
864         } else {
865             error_setg(errp, "'%s' invalid media", value);
866             goto fail;
867         }
868     }
869 
870     /* copy-on-read is disabled with a warning for read-only devices */
871     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
872     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
873 
874     if (read_only && copy_on_read) {
875         warn_report("disabling copy-on-read on read-only drive");
876         copy_on_read = false;
877     }
878 
879     qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
880     qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
881 
882     /* Controller type */
883     value = qemu_opt_get(legacy_opts, "if");
884     if (value) {
885         for (type = 0;
886              type < IF_COUNT && strcmp(value, if_name[type]);
887              type++) {
888         }
889         if (type == IF_COUNT) {
890             error_setg(errp, "unsupported bus type '%s'", value);
891             goto fail;
892         }
893     } else {
894         type = block_default_type;
895     }
896 
897     /* Device address specified by bus/unit or index.
898      * If none was specified, try to find the first free one. */
899     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
900     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
901     index   = qemu_opt_get_number(legacy_opts, "index", -1);
902 
903     max_devs = if_max_devs[type];
904 
905     if (index != -1) {
906         if (bus_id != 0 || unit_id != -1) {
907             error_setg(errp, "index cannot be used with bus and unit");
908             goto fail;
909         }
910         bus_id = drive_index_to_bus_id(type, index);
911         unit_id = drive_index_to_unit_id(type, index);
912     }
913 
914     if (unit_id == -1) {
915        unit_id = 0;
916        while (drive_get(type, bus_id, unit_id) != NULL) {
917            unit_id++;
918            if (max_devs && unit_id >= max_devs) {
919                unit_id -= max_devs;
920                bus_id++;
921            }
922        }
923     }
924 
925     if (max_devs && unit_id >= max_devs) {
926         error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
927         goto fail;
928     }
929 
930     if (drive_get(type, bus_id, unit_id) != NULL) {
931         error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
932                    bus_id, unit_id, index);
933         goto fail;
934     }
935 
936     /* no id supplied -> create one */
937     if (qemu_opts_id(all_opts) == NULL) {
938         char *new_id;
939         const char *mediastr = "";
940         if (type == IF_IDE || type == IF_SCSI) {
941             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
942         }
943         if (max_devs) {
944             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
945                                      mediastr, unit_id);
946         } else {
947             new_id = g_strdup_printf("%s%s%i", if_name[type],
948                                      mediastr, unit_id);
949         }
950         qdict_put_str(bs_opts, "id", new_id);
951         g_free(new_id);
952     }
953 
954     /* Add virtio block device */
955     if (type == IF_VIRTIO) {
956         QemuOpts *devopts;
957         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
958                                    &error_abort);
959         if (arch_type == QEMU_ARCH_S390X) {
960             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
961         } else {
962             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
963         }
964         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
965                      &error_abort);
966     }
967 
968     filename = qemu_opt_get(legacy_opts, "file");
969 
970     /* Check werror/rerror compatibility with if=... */
971     werror = qemu_opt_get(legacy_opts, "werror");
972     if (werror != NULL) {
973         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
974             type != IF_NONE) {
975             error_setg(errp, "werror is not supported by this bus type");
976             goto fail;
977         }
978         qdict_put_str(bs_opts, "werror", werror);
979     }
980 
981     rerror = qemu_opt_get(legacy_opts, "rerror");
982     if (rerror != NULL) {
983         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
984             type != IF_NONE) {
985             error_setg(errp, "rerror is not supported by this bus type");
986             goto fail;
987         }
988         qdict_put_str(bs_opts, "rerror", rerror);
989     }
990 
991     /* Actual block device init: Functionality shared with blockdev-add */
992     blk = blockdev_init(filename, bs_opts, &local_err);
993     bs_opts = NULL;
994     if (!blk) {
995         error_propagate(errp, local_err);
996         goto fail;
997     } else {
998         assert(!local_err);
999     }
1000 
1001     /* Create legacy DriveInfo */
1002     dinfo = g_malloc0(sizeof(*dinfo));
1003     dinfo->opts = all_opts;
1004 
1005     dinfo->type = type;
1006     dinfo->bus = bus_id;
1007     dinfo->unit = unit_id;
1008 
1009     blk_set_legacy_dinfo(blk, dinfo);
1010 
1011     switch(type) {
1012     case IF_IDE:
1013     case IF_SCSI:
1014     case IF_XEN:
1015     case IF_NONE:
1016         dinfo->media_cd = media == MEDIA_CDROM;
1017         break;
1018     default:
1019         break;
1020     }
1021 
1022 fail:
1023     qemu_opts_del(legacy_opts);
1024     qobject_unref(bs_opts);
1025     return dinfo;
1026 }
1027 
1028 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1029 {
1030     BlockDriverState *bs;
1031 
1032     bs = bdrv_lookup_bs(name, name, errp);
1033     if (bs == NULL) {
1034         return NULL;
1035     }
1036 
1037     if (!bdrv_is_root_node(bs)) {
1038         error_setg(errp, "Need a root block node");
1039         return NULL;
1040     }
1041 
1042     if (!bdrv_is_inserted(bs)) {
1043         error_setg(errp, "Device has no medium");
1044         return NULL;
1045     }
1046 
1047     return bs;
1048 }
1049 
1050 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1051                                  Error **errp)
1052 {
1053     BlockBackend *blk;
1054 
1055     if (!blk_name == !qdev_id) {
1056         error_setg(errp, "Need exactly one of 'device' and 'id'");
1057         return NULL;
1058     }
1059 
1060     if (qdev_id) {
1061         blk = blk_by_qdev_id(qdev_id, errp);
1062     } else {
1063         blk = blk_by_name(blk_name);
1064         if (blk == NULL) {
1065             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1066                       "Device '%s' not found", blk_name);
1067         }
1068     }
1069 
1070     return blk;
1071 }
1072 
1073 void hmp_commit(Monitor *mon, const QDict *qdict)
1074 {
1075     const char *device = qdict_get_str(qdict, "device");
1076     BlockBackend *blk;
1077     int ret;
1078 
1079     if (!strcmp(device, "all")) {
1080         ret = blk_commit_all();
1081     } else {
1082         BlockDriverState *bs;
1083         AioContext *aio_context;
1084 
1085         blk = blk_by_name(device);
1086         if (!blk) {
1087             monitor_printf(mon, "Device '%s' not found\n", device);
1088             return;
1089         }
1090         if (!blk_is_available(blk)) {
1091             monitor_printf(mon, "Device '%s' has no medium\n", device);
1092             return;
1093         }
1094 
1095         bs = blk_bs(blk);
1096         aio_context = bdrv_get_aio_context(bs);
1097         aio_context_acquire(aio_context);
1098 
1099         ret = bdrv_commit(bs);
1100 
1101         aio_context_release(aio_context);
1102     }
1103     if (ret < 0) {
1104         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1105                        strerror(-ret));
1106     }
1107 }
1108 
1109 static void blockdev_do_action(TransactionAction *action, Error **errp)
1110 {
1111     TransactionActionList list;
1112 
1113     list.value = action;
1114     list.next = NULL;
1115     qmp_transaction(&list, false, NULL, errp);
1116 }
1117 
1118 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1119                                 bool has_node_name, const char *node_name,
1120                                 const char *snapshot_file,
1121                                 bool has_snapshot_node_name,
1122                                 const char *snapshot_node_name,
1123                                 bool has_format, const char *format,
1124                                 bool has_mode, NewImageMode mode, Error **errp)
1125 {
1126     BlockdevSnapshotSync snapshot = {
1127         .has_device = has_device,
1128         .device = (char *) device,
1129         .has_node_name = has_node_name,
1130         .node_name = (char *) node_name,
1131         .snapshot_file = (char *) snapshot_file,
1132         .has_snapshot_node_name = has_snapshot_node_name,
1133         .snapshot_node_name = (char *) snapshot_node_name,
1134         .has_format = has_format,
1135         .format = (char *) format,
1136         .has_mode = has_mode,
1137         .mode = mode,
1138     };
1139     TransactionAction action = {
1140         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1141         .u.blockdev_snapshot_sync.data = &snapshot,
1142     };
1143     blockdev_do_action(&action, errp);
1144 }
1145 
1146 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1147                            Error **errp)
1148 {
1149     BlockdevSnapshot snapshot_data = {
1150         .node = (char *) node,
1151         .overlay = (char *) overlay
1152     };
1153     TransactionAction action = {
1154         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1155         .u.blockdev_snapshot.data = &snapshot_data,
1156     };
1157     blockdev_do_action(&action, errp);
1158 }
1159 
1160 void qmp_blockdev_snapshot_internal_sync(const char *device,
1161                                          const char *name,
1162                                          Error **errp)
1163 {
1164     BlockdevSnapshotInternal snapshot = {
1165         .device = (char *) device,
1166         .name = (char *) name
1167     };
1168     TransactionAction action = {
1169         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1170         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1171     };
1172     blockdev_do_action(&action, errp);
1173 }
1174 
1175 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1176                                                          bool has_id,
1177                                                          const char *id,
1178                                                          bool has_name,
1179                                                          const char *name,
1180                                                          Error **errp)
1181 {
1182     BlockDriverState *bs;
1183     AioContext *aio_context;
1184     QEMUSnapshotInfo sn;
1185     Error *local_err = NULL;
1186     SnapshotInfo *info = NULL;
1187     int ret;
1188 
1189     bs = qmp_get_root_bs(device, errp);
1190     if (!bs) {
1191         return NULL;
1192     }
1193     aio_context = bdrv_get_aio_context(bs);
1194     aio_context_acquire(aio_context);
1195 
1196     if (!has_id) {
1197         id = NULL;
1198     }
1199 
1200     if (!has_name) {
1201         name = NULL;
1202     }
1203 
1204     if (!id && !name) {
1205         error_setg(errp, "Name or id must be provided");
1206         goto out_aio_context;
1207     }
1208 
1209     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1210         goto out_aio_context;
1211     }
1212 
1213     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1214     if (local_err) {
1215         error_propagate(errp, local_err);
1216         goto out_aio_context;
1217     }
1218     if (!ret) {
1219         error_setg(errp,
1220                    "Snapshot with id '%s' and name '%s' does not exist on "
1221                    "device '%s'",
1222                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1223         goto out_aio_context;
1224     }
1225 
1226     bdrv_snapshot_delete(bs, id, name, &local_err);
1227     if (local_err) {
1228         error_propagate(errp, local_err);
1229         goto out_aio_context;
1230     }
1231 
1232     aio_context_release(aio_context);
1233 
1234     info = g_new0(SnapshotInfo, 1);
1235     info->id = g_strdup(sn.id_str);
1236     info->name = g_strdup(sn.name);
1237     info->date_nsec = sn.date_nsec;
1238     info->date_sec = sn.date_sec;
1239     info->vm_state_size = sn.vm_state_size;
1240     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1241     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1242 
1243     return info;
1244 
1245 out_aio_context:
1246     aio_context_release(aio_context);
1247     return NULL;
1248 }
1249 
1250 /**
1251  * block_dirty_bitmap_lookup:
1252  * Return a dirty bitmap (if present), after validating
1253  * the node reference and bitmap names.
1254  *
1255  * @node: The name of the BDS node to search for bitmaps
1256  * @name: The name of the bitmap to search for
1257  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1258  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1259  * @errp: Output pointer for error information. Can be NULL.
1260  *
1261  * @return: A bitmap object on success, or NULL on failure.
1262  */
1263 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1264                                                   const char *name,
1265                                                   BlockDriverState **pbs,
1266                                                   Error **errp)
1267 {
1268     BlockDriverState *bs;
1269     BdrvDirtyBitmap *bitmap;
1270 
1271     if (!node) {
1272         error_setg(errp, "Node cannot be NULL");
1273         return NULL;
1274     }
1275     if (!name) {
1276         error_setg(errp, "Bitmap name cannot be NULL");
1277         return NULL;
1278     }
1279     bs = bdrv_lookup_bs(node, node, NULL);
1280     if (!bs) {
1281         error_setg(errp, "Node '%s' not found", node);
1282         return NULL;
1283     }
1284 
1285     bitmap = bdrv_find_dirty_bitmap(bs, name);
1286     if (!bitmap) {
1287         error_setg(errp, "Dirty bitmap '%s' not found", name);
1288         return NULL;
1289     }
1290 
1291     if (pbs) {
1292         *pbs = bs;
1293     }
1294 
1295     return bitmap;
1296 }
1297 
1298 /* New and old BlockDriverState structs for atomic group operations */
1299 
1300 typedef struct BlkActionState BlkActionState;
1301 
1302 /**
1303  * BlkActionOps:
1304  * Table of operations that define an Action.
1305  *
1306  * @instance_size: Size of state struct, in bytes.
1307  * @prepare: Prepare the work, must NOT be NULL.
1308  * @commit: Commit the changes, can be NULL.
1309  * @abort: Abort the changes on fail, can be NULL.
1310  * @clean: Clean up resources after all transaction actions have called
1311  *         commit() or abort(). Can be NULL.
1312  *
1313  * Only prepare() may fail. In a single transaction, only one of commit() or
1314  * abort() will be called. clean() will always be called if it is present.
1315  */
1316 typedef struct BlkActionOps {
1317     size_t instance_size;
1318     void (*prepare)(BlkActionState *common, Error **errp);
1319     void (*commit)(BlkActionState *common);
1320     void (*abort)(BlkActionState *common);
1321     void (*clean)(BlkActionState *common);
1322 } BlkActionOps;
1323 
1324 /**
1325  * BlkActionState:
1326  * Describes one Action's state within a Transaction.
1327  *
1328  * @action: QAPI-defined enum identifying which Action to perform.
1329  * @ops: Table of ActionOps this Action can perform.
1330  * @block_job_txn: Transaction which this action belongs to.
1331  * @entry: List membership for all Actions in this Transaction.
1332  *
1333  * This structure must be arranged as first member in a subclassed type,
1334  * assuming that the compiler will also arrange it to the same offsets as the
1335  * base class.
1336  */
1337 struct BlkActionState {
1338     TransactionAction *action;
1339     const BlkActionOps *ops;
1340     JobTxn *block_job_txn;
1341     TransactionProperties *txn_props;
1342     QTAILQ_ENTRY(BlkActionState) entry;
1343 };
1344 
1345 /* internal snapshot private data */
1346 typedef struct InternalSnapshotState {
1347     BlkActionState common;
1348     BlockDriverState *bs;
1349     QEMUSnapshotInfo sn;
1350     bool created;
1351 } InternalSnapshotState;
1352 
1353 
1354 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1355 {
1356     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1357         error_setg(errp,
1358                    "Action '%s' does not support Transaction property "
1359                    "completion-mode = %s",
1360                    TransactionActionKind_str(s->action->type),
1361                    ActionCompletionMode_str(s->txn_props->completion_mode));
1362         return -1;
1363     }
1364     return 0;
1365 }
1366 
1367 static void internal_snapshot_prepare(BlkActionState *common,
1368                                       Error **errp)
1369 {
1370     Error *local_err = NULL;
1371     const char *device;
1372     const char *name;
1373     BlockDriverState *bs;
1374     QEMUSnapshotInfo old_sn, *sn;
1375     bool ret;
1376     qemu_timeval tv;
1377     BlockdevSnapshotInternal *internal;
1378     InternalSnapshotState *state;
1379     AioContext *aio_context;
1380     int ret1;
1381 
1382     g_assert(common->action->type ==
1383              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1384     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1385     state = DO_UPCAST(InternalSnapshotState, common, common);
1386 
1387     /* 1. parse input */
1388     device = internal->device;
1389     name = internal->name;
1390 
1391     /* 2. check for validation */
1392     if (action_check_completion_mode(common, errp) < 0) {
1393         return;
1394     }
1395 
1396     bs = qmp_get_root_bs(device, errp);
1397     if (!bs) {
1398         return;
1399     }
1400 
1401     aio_context = bdrv_get_aio_context(bs);
1402     aio_context_acquire(aio_context);
1403 
1404     state->bs = bs;
1405 
1406     /* Paired with .clean() */
1407     bdrv_drained_begin(bs);
1408 
1409     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1410         goto out;
1411     }
1412 
1413     if (bdrv_is_read_only(bs)) {
1414         error_setg(errp, "Device '%s' is read only", device);
1415         goto out;
1416     }
1417 
1418     if (!bdrv_can_snapshot(bs)) {
1419         error_setg(errp, "Block format '%s' used by device '%s' "
1420                    "does not support internal snapshots",
1421                    bs->drv->format_name, device);
1422         goto out;
1423     }
1424 
1425     if (!strlen(name)) {
1426         error_setg(errp, "Name is empty");
1427         goto out;
1428     }
1429 
1430     /* check whether a snapshot with name exist */
1431     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1432                                             &local_err);
1433     if (local_err) {
1434         error_propagate(errp, local_err);
1435         goto out;
1436     } else if (ret) {
1437         error_setg(errp,
1438                    "Snapshot with name '%s' already exists on device '%s'",
1439                    name, device);
1440         goto out;
1441     }
1442 
1443     /* 3. take the snapshot */
1444     sn = &state->sn;
1445     pstrcpy(sn->name, sizeof(sn->name), name);
1446     qemu_gettimeofday(&tv);
1447     sn->date_sec = tv.tv_sec;
1448     sn->date_nsec = tv.tv_usec * 1000;
1449     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1450 
1451     ret1 = bdrv_snapshot_create(bs, sn);
1452     if (ret1 < 0) {
1453         error_setg_errno(errp, -ret1,
1454                          "Failed to create snapshot '%s' on device '%s'",
1455                          name, device);
1456         goto out;
1457     }
1458 
1459     /* 4. succeed, mark a snapshot is created */
1460     state->created = true;
1461 
1462 out:
1463     aio_context_release(aio_context);
1464 }
1465 
1466 static void internal_snapshot_abort(BlkActionState *common)
1467 {
1468     InternalSnapshotState *state =
1469                              DO_UPCAST(InternalSnapshotState, common, common);
1470     BlockDriverState *bs = state->bs;
1471     QEMUSnapshotInfo *sn = &state->sn;
1472     AioContext *aio_context;
1473     Error *local_error = NULL;
1474 
1475     if (!state->created) {
1476         return;
1477     }
1478 
1479     aio_context = bdrv_get_aio_context(state->bs);
1480     aio_context_acquire(aio_context);
1481 
1482     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1483         error_reportf_err(local_error,
1484                           "Failed to delete snapshot with id '%s' and "
1485                           "name '%s' on device '%s' in abort: ",
1486                           sn->id_str, sn->name,
1487                           bdrv_get_device_name(bs));
1488     }
1489 
1490     aio_context_release(aio_context);
1491 }
1492 
1493 static void internal_snapshot_clean(BlkActionState *common)
1494 {
1495     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1496                                              common, common);
1497     AioContext *aio_context;
1498 
1499     if (!state->bs) {
1500         return;
1501     }
1502 
1503     aio_context = bdrv_get_aio_context(state->bs);
1504     aio_context_acquire(aio_context);
1505 
1506     bdrv_drained_end(state->bs);
1507 
1508     aio_context_release(aio_context);
1509 }
1510 
1511 /* external snapshot private data */
1512 typedef struct ExternalSnapshotState {
1513     BlkActionState common;
1514     BlockDriverState *old_bs;
1515     BlockDriverState *new_bs;
1516     bool overlay_appended;
1517 } ExternalSnapshotState;
1518 
1519 static void external_snapshot_prepare(BlkActionState *common,
1520                                       Error **errp)
1521 {
1522     int flags = 0;
1523     QDict *options = NULL;
1524     Error *local_err = NULL;
1525     /* Device and node name of the image to generate the snapshot from */
1526     const char *device;
1527     const char *node_name;
1528     /* Reference to the new image (for 'blockdev-snapshot') */
1529     const char *snapshot_ref;
1530     /* File name of the new image (for 'blockdev-snapshot-sync') */
1531     const char *new_image_file;
1532     ExternalSnapshotState *state =
1533                              DO_UPCAST(ExternalSnapshotState, common, common);
1534     TransactionAction *action = common->action;
1535     AioContext *aio_context;
1536 
1537     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1538      * purpose but a different set of parameters */
1539     switch (action->type) {
1540     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1541         {
1542             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1543             device = s->node;
1544             node_name = s->node;
1545             new_image_file = NULL;
1546             snapshot_ref = s->overlay;
1547         }
1548         break;
1549     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1550         {
1551             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1552             device = s->has_device ? s->device : NULL;
1553             node_name = s->has_node_name ? s->node_name : NULL;
1554             new_image_file = s->snapshot_file;
1555             snapshot_ref = NULL;
1556         }
1557         break;
1558     default:
1559         g_assert_not_reached();
1560     }
1561 
1562     /* start processing */
1563     if (action_check_completion_mode(common, errp) < 0) {
1564         return;
1565     }
1566 
1567     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1568     if (!state->old_bs) {
1569         return;
1570     }
1571 
1572     aio_context = bdrv_get_aio_context(state->old_bs);
1573     aio_context_acquire(aio_context);
1574 
1575     /* Paired with .clean() */
1576     bdrv_drained_begin(state->old_bs);
1577 
1578     if (!bdrv_is_inserted(state->old_bs)) {
1579         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1580         goto out;
1581     }
1582 
1583     if (bdrv_op_is_blocked(state->old_bs,
1584                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1585         goto out;
1586     }
1587 
1588     if (!bdrv_is_read_only(state->old_bs)) {
1589         if (bdrv_flush(state->old_bs)) {
1590             error_setg(errp, QERR_IO_ERROR);
1591             goto out;
1592         }
1593     }
1594 
1595     if (!bdrv_is_first_non_filter(state->old_bs)) {
1596         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1597         goto out;
1598     }
1599 
1600     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1601         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1602         const char *format = s->has_format ? s->format : "qcow2";
1603         enum NewImageMode mode;
1604         const char *snapshot_node_name =
1605             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1606 
1607         if (node_name && !snapshot_node_name) {
1608             error_setg(errp, "New snapshot node name missing");
1609             goto out;
1610         }
1611 
1612         if (snapshot_node_name &&
1613             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1614             error_setg(errp, "New snapshot node name already in use");
1615             goto out;
1616         }
1617 
1618         flags = state->old_bs->open_flags;
1619         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1620         flags |= BDRV_O_NO_BACKING;
1621 
1622         /* create new image w/backing file */
1623         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1624         if (mode != NEW_IMAGE_MODE_EXISTING) {
1625             int64_t size = bdrv_getlength(state->old_bs);
1626             if (size < 0) {
1627                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1628                 goto out;
1629             }
1630             bdrv_refresh_filename(state->old_bs);
1631             bdrv_img_create(new_image_file, format,
1632                             state->old_bs->filename,
1633                             state->old_bs->drv->format_name,
1634                             NULL, size, flags, false, &local_err);
1635             if (local_err) {
1636                 error_propagate(errp, local_err);
1637                 goto out;
1638             }
1639         }
1640 
1641         options = qdict_new();
1642         if (snapshot_node_name) {
1643             qdict_put_str(options, "node-name", snapshot_node_name);
1644         }
1645         qdict_put_str(options, "driver", format);
1646     }
1647 
1648     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1649                               errp);
1650     /* We will manually add the backing_hd field to the bs later */
1651     if (!state->new_bs) {
1652         goto out;
1653     }
1654 
1655     if (bdrv_has_blk(state->new_bs)) {
1656         error_setg(errp, "The snapshot is already in use");
1657         goto out;
1658     }
1659 
1660     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1661                            errp)) {
1662         goto out;
1663     }
1664 
1665     if (state->new_bs->backing != NULL) {
1666         error_setg(errp, "The snapshot already has a backing image");
1667         goto out;
1668     }
1669 
1670     if (!state->new_bs->drv->supports_backing) {
1671         error_setg(errp, "The snapshot does not support backing images");
1672         goto out;
1673     }
1674 
1675     bdrv_set_aio_context(state->new_bs, aio_context);
1676 
1677     /* This removes our old bs and adds the new bs. This is an operation that
1678      * can fail, so we need to do it in .prepare; undoing it for abort is
1679      * always possible. */
1680     bdrv_ref(state->new_bs);
1681     bdrv_append(state->new_bs, state->old_bs, &local_err);
1682     if (local_err) {
1683         error_propagate(errp, local_err);
1684         goto out;
1685     }
1686     state->overlay_appended = true;
1687 
1688 out:
1689     aio_context_release(aio_context);
1690 }
1691 
1692 static void external_snapshot_commit(BlkActionState *common)
1693 {
1694     ExternalSnapshotState *state =
1695                              DO_UPCAST(ExternalSnapshotState, common, common);
1696     AioContext *aio_context;
1697 
1698     aio_context = bdrv_get_aio_context(state->old_bs);
1699     aio_context_acquire(aio_context);
1700 
1701     /* We don't need (or want) to use the transactional
1702      * bdrv_reopen_multiple() across all the entries at once, because we
1703      * don't want to abort all of them if one of them fails the reopen */
1704     if (!atomic_read(&state->old_bs->copy_on_read)) {
1705         bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1706     }
1707 
1708     aio_context_release(aio_context);
1709 }
1710 
1711 static void external_snapshot_abort(BlkActionState *common)
1712 {
1713     ExternalSnapshotState *state =
1714                              DO_UPCAST(ExternalSnapshotState, common, common);
1715     if (state->new_bs) {
1716         if (state->overlay_appended) {
1717             AioContext *aio_context;
1718 
1719             aio_context = bdrv_get_aio_context(state->old_bs);
1720             aio_context_acquire(aio_context);
1721 
1722             bdrv_ref(state->old_bs);   /* we can't let bdrv_set_backind_hd()
1723                                           close state->old_bs; we need it */
1724             bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1725             bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1726             bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1727 
1728             aio_context_release(aio_context);
1729         }
1730     }
1731 }
1732 
1733 static void external_snapshot_clean(BlkActionState *common)
1734 {
1735     ExternalSnapshotState *state =
1736                              DO_UPCAST(ExternalSnapshotState, common, common);
1737     AioContext *aio_context;
1738 
1739     if (!state->old_bs) {
1740         return;
1741     }
1742 
1743     aio_context = bdrv_get_aio_context(state->old_bs);
1744     aio_context_acquire(aio_context);
1745 
1746     bdrv_drained_end(state->old_bs);
1747     bdrv_unref(state->new_bs);
1748 
1749     aio_context_release(aio_context);
1750 }
1751 
1752 typedef struct DriveBackupState {
1753     BlkActionState common;
1754     BlockDriverState *bs;
1755     BlockJob *job;
1756 } DriveBackupState;
1757 
1758 static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
1759                             Error **errp);
1760 
1761 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1762 {
1763     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1764     BlockDriverState *bs;
1765     DriveBackup *backup;
1766     AioContext *aio_context;
1767     Error *local_err = NULL;
1768 
1769     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1770     backup = common->action->u.drive_backup.data;
1771 
1772     bs = qmp_get_root_bs(backup->device, errp);
1773     if (!bs) {
1774         return;
1775     }
1776 
1777     aio_context = bdrv_get_aio_context(bs);
1778     aio_context_acquire(aio_context);
1779 
1780     /* Paired with .clean() */
1781     bdrv_drained_begin(bs);
1782 
1783     state->bs = bs;
1784 
1785     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1786     if (local_err) {
1787         error_propagate(errp, local_err);
1788         goto out;
1789     }
1790 
1791 out:
1792     aio_context_release(aio_context);
1793 }
1794 
1795 static void drive_backup_commit(BlkActionState *common)
1796 {
1797     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1798     AioContext *aio_context;
1799 
1800     aio_context = bdrv_get_aio_context(state->bs);
1801     aio_context_acquire(aio_context);
1802 
1803     assert(state->job);
1804     job_start(&state->job->job);
1805 
1806     aio_context_release(aio_context);
1807 }
1808 
1809 static void drive_backup_abort(BlkActionState *common)
1810 {
1811     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1812 
1813     if (state->job) {
1814         AioContext *aio_context;
1815 
1816         aio_context = bdrv_get_aio_context(state->bs);
1817         aio_context_acquire(aio_context);
1818 
1819         job_cancel_sync(&state->job->job);
1820 
1821         aio_context_release(aio_context);
1822     }
1823 }
1824 
1825 static void drive_backup_clean(BlkActionState *common)
1826 {
1827     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1828     AioContext *aio_context;
1829 
1830     if (!state->bs) {
1831         return;
1832     }
1833 
1834     aio_context = bdrv_get_aio_context(state->bs);
1835     aio_context_acquire(aio_context);
1836 
1837     bdrv_drained_end(state->bs);
1838 
1839     aio_context_release(aio_context);
1840 }
1841 
1842 typedef struct BlockdevBackupState {
1843     BlkActionState common;
1844     BlockDriverState *bs;
1845     BlockJob *job;
1846 } BlockdevBackupState;
1847 
1848 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
1849                                     Error **errp);
1850 
1851 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1852 {
1853     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1854     BlockdevBackup *backup;
1855     BlockDriverState *bs, *target;
1856     AioContext *aio_context;
1857     Error *local_err = NULL;
1858 
1859     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1860     backup = common->action->u.blockdev_backup.data;
1861 
1862     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1863     if (!bs) {
1864         return;
1865     }
1866 
1867     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1868     if (!target) {
1869         return;
1870     }
1871 
1872     aio_context = bdrv_get_aio_context(bs);
1873     if (aio_context != bdrv_get_aio_context(target)) {
1874         error_setg(errp, "Backup between two IO threads is not implemented");
1875         return;
1876     }
1877     aio_context_acquire(aio_context);
1878     state->bs = bs;
1879 
1880     /* Paired with .clean() */
1881     bdrv_drained_begin(state->bs);
1882 
1883     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1884     if (local_err) {
1885         error_propagate(errp, local_err);
1886         goto out;
1887     }
1888 
1889 out:
1890     aio_context_release(aio_context);
1891 }
1892 
1893 static void blockdev_backup_commit(BlkActionState *common)
1894 {
1895     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1896     AioContext *aio_context;
1897 
1898     aio_context = bdrv_get_aio_context(state->bs);
1899     aio_context_acquire(aio_context);
1900 
1901     assert(state->job);
1902     job_start(&state->job->job);
1903 
1904     aio_context_release(aio_context);
1905 }
1906 
1907 static void blockdev_backup_abort(BlkActionState *common)
1908 {
1909     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1910 
1911     if (state->job) {
1912         AioContext *aio_context;
1913 
1914         aio_context = bdrv_get_aio_context(state->bs);
1915         aio_context_acquire(aio_context);
1916 
1917         job_cancel_sync(&state->job->job);
1918 
1919         aio_context_release(aio_context);
1920     }
1921 }
1922 
1923 static void blockdev_backup_clean(BlkActionState *common)
1924 {
1925     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1926     AioContext *aio_context;
1927 
1928     if (!state->bs) {
1929         return;
1930     }
1931 
1932     aio_context = bdrv_get_aio_context(state->bs);
1933     aio_context_acquire(aio_context);
1934 
1935     bdrv_drained_end(state->bs);
1936 
1937     aio_context_release(aio_context);
1938 }
1939 
1940 typedef struct BlockDirtyBitmapState {
1941     BlkActionState common;
1942     BdrvDirtyBitmap *bitmap;
1943     BlockDriverState *bs;
1944     HBitmap *backup;
1945     bool prepared;
1946     bool was_enabled;
1947 } BlockDirtyBitmapState;
1948 
1949 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1950                                            Error **errp)
1951 {
1952     Error *local_err = NULL;
1953     BlockDirtyBitmapAdd *action;
1954     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1955                                              common, common);
1956 
1957     if (action_check_completion_mode(common, errp) < 0) {
1958         return;
1959     }
1960 
1961     action = common->action->u.block_dirty_bitmap_add.data;
1962     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1963     qmp_block_dirty_bitmap_add(action->node, action->name,
1964                                action->has_granularity, action->granularity,
1965                                action->has_persistent, action->persistent,
1966                                action->has_autoload, action->autoload,
1967                                action->has_disabled, action->disabled,
1968                                &local_err);
1969 
1970     if (!local_err) {
1971         state->prepared = true;
1972     } else {
1973         error_propagate(errp, local_err);
1974     }
1975 }
1976 
1977 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1978 {
1979     BlockDirtyBitmapAdd *action;
1980     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1981                                              common, common);
1982 
1983     action = common->action->u.block_dirty_bitmap_add.data;
1984     /* Should not be able to fail: IF the bitmap was added via .prepare(),
1985      * then the node reference and bitmap name must have been valid.
1986      */
1987     if (state->prepared) {
1988         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
1989     }
1990 }
1991 
1992 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
1993                                              Error **errp)
1994 {
1995     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1996                                              common, common);
1997     BlockDirtyBitmap *action;
1998 
1999     if (action_check_completion_mode(common, errp) < 0) {
2000         return;
2001     }
2002 
2003     action = common->action->u.block_dirty_bitmap_clear.data;
2004     state->bitmap = block_dirty_bitmap_lookup(action->node,
2005                                               action->name,
2006                                               &state->bs,
2007                                               errp);
2008     if (!state->bitmap) {
2009         return;
2010     }
2011 
2012     if (bdrv_dirty_bitmap_user_locked(state->bitmap)) {
2013         error_setg(errp, "Cannot modify a bitmap in use by another operation");
2014         return;
2015     } else if (bdrv_dirty_bitmap_readonly(state->bitmap)) {
2016         error_setg(errp, "Cannot clear a readonly bitmap");
2017         return;
2018     }
2019 
2020     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2021 }
2022 
2023 static void block_dirty_bitmap_restore(BlkActionState *common)
2024 {
2025     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2026                                              common, common);
2027 
2028     if (state->backup) {
2029         bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2030     }
2031 }
2032 
2033 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2034 {
2035     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2036                                              common, common);
2037 
2038     hbitmap_free(state->backup);
2039 }
2040 
2041 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2042                                               Error **errp)
2043 {
2044     BlockDirtyBitmap *action;
2045     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2046                                              common, common);
2047 
2048     if (action_check_completion_mode(common, errp) < 0) {
2049         return;
2050     }
2051 
2052     action = common->action->u.block_dirty_bitmap_enable.data;
2053     state->bitmap = block_dirty_bitmap_lookup(action->node,
2054                                               action->name,
2055                                               NULL,
2056                                               errp);
2057     if (!state->bitmap) {
2058         return;
2059     }
2060 
2061     if (bdrv_dirty_bitmap_user_locked(state->bitmap)) {
2062         error_setg(errp,
2063                    "Bitmap '%s' is currently in use by another operation"
2064                    " and cannot be enabled", action->name);
2065         return;
2066     }
2067 
2068     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2069     bdrv_enable_dirty_bitmap(state->bitmap);
2070 }
2071 
2072 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2073 {
2074     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2075                                              common, common);
2076 
2077     if (!state->was_enabled) {
2078         bdrv_disable_dirty_bitmap(state->bitmap);
2079     }
2080 }
2081 
2082 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2083                                                Error **errp)
2084 {
2085     BlockDirtyBitmap *action;
2086     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2087                                              common, common);
2088 
2089     if (action_check_completion_mode(common, errp) < 0) {
2090         return;
2091     }
2092 
2093     action = common->action->u.block_dirty_bitmap_disable.data;
2094     state->bitmap = block_dirty_bitmap_lookup(action->node,
2095                                               action->name,
2096                                               NULL,
2097                                               errp);
2098     if (!state->bitmap) {
2099         return;
2100     }
2101 
2102     if (bdrv_dirty_bitmap_user_locked(state->bitmap)) {
2103         error_setg(errp,
2104                    "Bitmap '%s' is currently in use by another operation"
2105                    " and cannot be disabled", action->name);
2106         return;
2107     }
2108 
2109     state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2110     bdrv_disable_dirty_bitmap(state->bitmap);
2111 }
2112 
2113 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2114 {
2115     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2116                                              common, common);
2117 
2118     if (state->was_enabled) {
2119         bdrv_enable_dirty_bitmap(state->bitmap);
2120     }
2121 }
2122 
2123 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(const char *node,
2124                                                     const char *target,
2125                                                     strList *bitmaps,
2126                                                     HBitmap **backup,
2127                                                     Error **errp);
2128 
2129 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2130                                              Error **errp)
2131 {
2132     BlockDirtyBitmapMerge *action;
2133     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2134                                              common, common);
2135 
2136     if (action_check_completion_mode(common, errp) < 0) {
2137         return;
2138     }
2139 
2140     action = common->action->u.block_dirty_bitmap_merge.data;
2141 
2142     state->bitmap = do_block_dirty_bitmap_merge(action->node, action->target,
2143                                                 action->bitmaps, &state->backup,
2144                                                 errp);
2145 }
2146 
2147 static void abort_prepare(BlkActionState *common, Error **errp)
2148 {
2149     error_setg(errp, "Transaction aborted using Abort action");
2150 }
2151 
2152 static void abort_commit(BlkActionState *common)
2153 {
2154     g_assert_not_reached(); /* this action never succeeds */
2155 }
2156 
2157 static const BlkActionOps actions[] = {
2158     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2159         .instance_size = sizeof(ExternalSnapshotState),
2160         .prepare  = external_snapshot_prepare,
2161         .commit   = external_snapshot_commit,
2162         .abort = external_snapshot_abort,
2163         .clean = external_snapshot_clean,
2164     },
2165     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2166         .instance_size = sizeof(ExternalSnapshotState),
2167         .prepare  = external_snapshot_prepare,
2168         .commit   = external_snapshot_commit,
2169         .abort = external_snapshot_abort,
2170         .clean = external_snapshot_clean,
2171     },
2172     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2173         .instance_size = sizeof(DriveBackupState),
2174         .prepare = drive_backup_prepare,
2175         .commit = drive_backup_commit,
2176         .abort = drive_backup_abort,
2177         .clean = drive_backup_clean,
2178     },
2179     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2180         .instance_size = sizeof(BlockdevBackupState),
2181         .prepare = blockdev_backup_prepare,
2182         .commit = blockdev_backup_commit,
2183         .abort = blockdev_backup_abort,
2184         .clean = blockdev_backup_clean,
2185     },
2186     [TRANSACTION_ACTION_KIND_ABORT] = {
2187         .instance_size = sizeof(BlkActionState),
2188         .prepare = abort_prepare,
2189         .commit = abort_commit,
2190     },
2191     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2192         .instance_size = sizeof(InternalSnapshotState),
2193         .prepare  = internal_snapshot_prepare,
2194         .abort = internal_snapshot_abort,
2195         .clean = internal_snapshot_clean,
2196     },
2197     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2198         .instance_size = sizeof(BlockDirtyBitmapState),
2199         .prepare = block_dirty_bitmap_add_prepare,
2200         .abort = block_dirty_bitmap_add_abort,
2201     },
2202     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2203         .instance_size = sizeof(BlockDirtyBitmapState),
2204         .prepare = block_dirty_bitmap_clear_prepare,
2205         .commit = block_dirty_bitmap_free_backup,
2206         .abort = block_dirty_bitmap_restore,
2207     },
2208     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2209         .instance_size = sizeof(BlockDirtyBitmapState),
2210         .prepare = block_dirty_bitmap_enable_prepare,
2211         .abort = block_dirty_bitmap_enable_abort,
2212     },
2213     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2214         .instance_size = sizeof(BlockDirtyBitmapState),
2215         .prepare = block_dirty_bitmap_disable_prepare,
2216         .abort = block_dirty_bitmap_disable_abort,
2217     },
2218     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2219         .instance_size = sizeof(BlockDirtyBitmapState),
2220         .prepare = block_dirty_bitmap_merge_prepare,
2221         .commit = block_dirty_bitmap_free_backup,
2222         .abort = block_dirty_bitmap_restore,
2223     },
2224     /* Where are transactions for MIRROR, COMMIT and STREAM?
2225      * Although these blockjobs use transaction callbacks like the backup job,
2226      * these jobs do not necessarily adhere to transaction semantics.
2227      * These jobs may not fully undo all of their actions on abort, nor do they
2228      * necessarily work in transactions with more than one job in them.
2229      */
2230 };
2231 
2232 /**
2233  * Allocate a TransactionProperties structure if necessary, and fill
2234  * that structure with desired defaults if they are unset.
2235  */
2236 static TransactionProperties *get_transaction_properties(
2237     TransactionProperties *props)
2238 {
2239     if (!props) {
2240         props = g_new0(TransactionProperties, 1);
2241     }
2242 
2243     if (!props->has_completion_mode) {
2244         props->has_completion_mode = true;
2245         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2246     }
2247 
2248     return props;
2249 }
2250 
2251 /*
2252  * 'Atomic' group operations.  The operations are performed as a set, and if
2253  * any fail then we roll back all operations in the group.
2254  */
2255 void qmp_transaction(TransactionActionList *dev_list,
2256                      bool has_props,
2257                      struct TransactionProperties *props,
2258                      Error **errp)
2259 {
2260     TransactionActionList *dev_entry = dev_list;
2261     JobTxn *block_job_txn = NULL;
2262     BlkActionState *state, *next;
2263     Error *local_err = NULL;
2264 
2265     QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2266     QTAILQ_INIT(&snap_bdrv_states);
2267 
2268     /* Does this transaction get canceled as a group on failure?
2269      * If not, we don't really need to make a JobTxn.
2270      */
2271     props = get_transaction_properties(props);
2272     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2273         block_job_txn = job_txn_new();
2274     }
2275 
2276     /* drain all i/o before any operations */
2277     bdrv_drain_all();
2278 
2279     /* We don't do anything in this loop that commits us to the operations */
2280     while (NULL != dev_entry) {
2281         TransactionAction *dev_info = NULL;
2282         const BlkActionOps *ops;
2283 
2284         dev_info = dev_entry->value;
2285         dev_entry = dev_entry->next;
2286 
2287         assert(dev_info->type < ARRAY_SIZE(actions));
2288 
2289         ops = &actions[dev_info->type];
2290         assert(ops->instance_size > 0);
2291 
2292         state = g_malloc0(ops->instance_size);
2293         state->ops = ops;
2294         state->action = dev_info;
2295         state->block_job_txn = block_job_txn;
2296         state->txn_props = props;
2297         QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2298 
2299         state->ops->prepare(state, &local_err);
2300         if (local_err) {
2301             error_propagate(errp, local_err);
2302             goto delete_and_fail;
2303         }
2304     }
2305 
2306     QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2307         if (state->ops->commit) {
2308             state->ops->commit(state);
2309         }
2310     }
2311 
2312     /* success */
2313     goto exit;
2314 
2315 delete_and_fail:
2316     /* failure, and it is all-or-none; roll back all operations */
2317     QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2318         if (state->ops->abort) {
2319             state->ops->abort(state);
2320         }
2321     }
2322 exit:
2323     QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2324         if (state->ops->clean) {
2325             state->ops->clean(state);
2326         }
2327         g_free(state);
2328     }
2329     if (!has_props) {
2330         qapi_free_TransactionProperties(props);
2331     }
2332     job_txn_unref(block_job_txn);
2333 }
2334 
2335 void qmp_eject(bool has_device, const char *device,
2336                bool has_id, const char *id,
2337                bool has_force, bool force, Error **errp)
2338 {
2339     Error *local_err = NULL;
2340     int rc;
2341 
2342     if (!has_force) {
2343         force = false;
2344     }
2345 
2346     rc = do_open_tray(has_device ? device : NULL,
2347                       has_id ? id : NULL,
2348                       force, &local_err);
2349     if (rc && rc != -ENOSYS) {
2350         error_propagate(errp, local_err);
2351         return;
2352     }
2353     error_free(local_err);
2354 
2355     blockdev_remove_medium(has_device, device, has_id, id, errp);
2356 }
2357 
2358 void qmp_block_passwd(bool has_device, const char *device,
2359                       bool has_node_name, const char *node_name,
2360                       const char *password, Error **errp)
2361 {
2362     error_setg(errp,
2363                "Setting block passwords directly is no longer supported");
2364 }
2365 
2366 /*
2367  * Attempt to open the tray of @device.
2368  * If @force, ignore its tray lock.
2369  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2370  * On error, store an error through @errp and return -errno.
2371  * If @device does not exist, return -ENODEV.
2372  * If it has no removable media, return -ENOTSUP.
2373  * If it has no tray, return -ENOSYS.
2374  * If the guest was asked to open the tray, return -EINPROGRESS.
2375  * Else, return 0.
2376  */
2377 static int do_open_tray(const char *blk_name, const char *qdev_id,
2378                         bool force, Error **errp)
2379 {
2380     BlockBackend *blk;
2381     const char *device = qdev_id ?: blk_name;
2382     bool locked;
2383 
2384     blk = qmp_get_blk(blk_name, qdev_id, errp);
2385     if (!blk) {
2386         return -ENODEV;
2387     }
2388 
2389     if (!blk_dev_has_removable_media(blk)) {
2390         error_setg(errp, "Device '%s' is not removable", device);
2391         return -ENOTSUP;
2392     }
2393 
2394     if (!blk_dev_has_tray(blk)) {
2395         error_setg(errp, "Device '%s' does not have a tray", device);
2396         return -ENOSYS;
2397     }
2398 
2399     if (blk_dev_is_tray_open(blk)) {
2400         return 0;
2401     }
2402 
2403     locked = blk_dev_is_medium_locked(blk);
2404     if (locked) {
2405         blk_dev_eject_request(blk, force);
2406     }
2407 
2408     if (!locked || force) {
2409         blk_dev_change_media_cb(blk, false, &error_abort);
2410     }
2411 
2412     if (locked && !force) {
2413         error_setg(errp, "Device '%s' is locked and force was not specified, "
2414                    "wait for tray to open and try again", device);
2415         return -EINPROGRESS;
2416     }
2417 
2418     return 0;
2419 }
2420 
2421 void qmp_blockdev_open_tray(bool has_device, const char *device,
2422                             bool has_id, const char *id,
2423                             bool has_force, bool force,
2424                             Error **errp)
2425 {
2426     Error *local_err = NULL;
2427     int rc;
2428 
2429     if (!has_force) {
2430         force = false;
2431     }
2432     rc = do_open_tray(has_device ? device : NULL,
2433                       has_id ? id : NULL,
2434                       force, &local_err);
2435     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2436         error_propagate(errp, local_err);
2437         return;
2438     }
2439     error_free(local_err);
2440 }
2441 
2442 void qmp_blockdev_close_tray(bool has_device, const char *device,
2443                              bool has_id, const char *id,
2444                              Error **errp)
2445 {
2446     BlockBackend *blk;
2447     Error *local_err = NULL;
2448 
2449     device = has_device ? device : NULL;
2450     id = has_id ? id : NULL;
2451 
2452     blk = qmp_get_blk(device, id, errp);
2453     if (!blk) {
2454         return;
2455     }
2456 
2457     if (!blk_dev_has_removable_media(blk)) {
2458         error_setg(errp, "Device '%s' is not removable", device ?: id);
2459         return;
2460     }
2461 
2462     if (!blk_dev_has_tray(blk)) {
2463         /* Ignore this command on tray-less devices */
2464         return;
2465     }
2466 
2467     if (!blk_dev_is_tray_open(blk)) {
2468         return;
2469     }
2470 
2471     blk_dev_change_media_cb(blk, true, &local_err);
2472     if (local_err) {
2473         error_propagate(errp, local_err);
2474         return;
2475     }
2476 }
2477 
2478 static void blockdev_remove_medium(bool has_device, const char *device,
2479                                    bool has_id, const char *id, Error **errp)
2480 {
2481     BlockBackend *blk;
2482     BlockDriverState *bs;
2483     AioContext *aio_context;
2484     bool has_attached_device;
2485 
2486     device = has_device ? device : NULL;
2487     id = has_id ? id : NULL;
2488 
2489     blk = qmp_get_blk(device, id, errp);
2490     if (!blk) {
2491         return;
2492     }
2493 
2494     /* For BBs without a device, we can exchange the BDS tree at will */
2495     has_attached_device = blk_get_attached_dev(blk);
2496 
2497     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2498         error_setg(errp, "Device '%s' is not removable", device ?: id);
2499         return;
2500     }
2501 
2502     if (has_attached_device && blk_dev_has_tray(blk) &&
2503         !blk_dev_is_tray_open(blk))
2504     {
2505         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2506         return;
2507     }
2508 
2509     bs = blk_bs(blk);
2510     if (!bs) {
2511         return;
2512     }
2513 
2514     aio_context = bdrv_get_aio_context(bs);
2515     aio_context_acquire(aio_context);
2516 
2517     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2518         goto out;
2519     }
2520 
2521     blk_remove_bs(blk);
2522 
2523     if (!blk_dev_has_tray(blk)) {
2524         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2525          * called at all); therefore, the medium needs to be ejected here.
2526          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2527          * value passed here (i.e. false). */
2528         blk_dev_change_media_cb(blk, false, &error_abort);
2529     }
2530 
2531 out:
2532     aio_context_release(aio_context);
2533 }
2534 
2535 void qmp_blockdev_remove_medium(const char *id, Error **errp)
2536 {
2537     blockdev_remove_medium(false, NULL, true, id, errp);
2538 }
2539 
2540 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2541                                             BlockDriverState *bs, Error **errp)
2542 {
2543     Error *local_err = NULL;
2544     bool has_device;
2545     int ret;
2546 
2547     /* For BBs without a device, we can exchange the BDS tree at will */
2548     has_device = blk_get_attached_dev(blk);
2549 
2550     if (has_device && !blk_dev_has_removable_media(blk)) {
2551         error_setg(errp, "Device is not removable");
2552         return;
2553     }
2554 
2555     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2556         error_setg(errp, "Tray of the device is not open");
2557         return;
2558     }
2559 
2560     if (blk_bs(blk)) {
2561         error_setg(errp, "There already is a medium in the device");
2562         return;
2563     }
2564 
2565     ret = blk_insert_bs(blk, bs, errp);
2566     if (ret < 0) {
2567         return;
2568     }
2569 
2570     if (!blk_dev_has_tray(blk)) {
2571         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2572          * called at all); therefore, the medium needs to be pushed into the
2573          * slot here.
2574          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2575          * value passed here (i.e. true). */
2576         blk_dev_change_media_cb(blk, true, &local_err);
2577         if (local_err) {
2578             error_propagate(errp, local_err);
2579             blk_remove_bs(blk);
2580             return;
2581         }
2582     }
2583 }
2584 
2585 static void blockdev_insert_medium(bool has_device, const char *device,
2586                                    bool has_id, const char *id,
2587                                    const char *node_name, Error **errp)
2588 {
2589     BlockBackend *blk;
2590     BlockDriverState *bs;
2591 
2592     blk = qmp_get_blk(has_device ? device : NULL,
2593                       has_id ? id : NULL,
2594                       errp);
2595     if (!blk) {
2596         return;
2597     }
2598 
2599     bs = bdrv_find_node(node_name);
2600     if (!bs) {
2601         error_setg(errp, "Node '%s' not found", node_name);
2602         return;
2603     }
2604 
2605     if (bdrv_has_blk(bs)) {
2606         error_setg(errp, "Node '%s' is already in use", node_name);
2607         return;
2608     }
2609 
2610     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2611 }
2612 
2613 void qmp_blockdev_insert_medium(const char *id, const char *node_name,
2614                                 Error **errp)
2615 {
2616     blockdev_insert_medium(false, NULL, true, id, node_name, errp);
2617 }
2618 
2619 void qmp_blockdev_change_medium(bool has_device, const char *device,
2620                                 bool has_id, const char *id,
2621                                 const char *filename,
2622                                 bool has_format, const char *format,
2623                                 bool has_read_only,
2624                                 BlockdevChangeReadOnlyMode read_only,
2625                                 Error **errp)
2626 {
2627     BlockBackend *blk;
2628     BlockDriverState *medium_bs = NULL;
2629     int bdrv_flags;
2630     bool detect_zeroes;
2631     int rc;
2632     QDict *options = NULL;
2633     Error *err = NULL;
2634 
2635     blk = qmp_get_blk(has_device ? device : NULL,
2636                       has_id ? id : NULL,
2637                       errp);
2638     if (!blk) {
2639         goto fail;
2640     }
2641 
2642     if (blk_bs(blk)) {
2643         blk_update_root_state(blk);
2644     }
2645 
2646     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2647     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2648         BDRV_O_PROTOCOL | BDRV_O_AUTO_RDONLY);
2649 
2650     if (!has_read_only) {
2651         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2652     }
2653 
2654     switch (read_only) {
2655     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2656         break;
2657 
2658     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2659         bdrv_flags &= ~BDRV_O_RDWR;
2660         break;
2661 
2662     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2663         bdrv_flags |= BDRV_O_RDWR;
2664         break;
2665 
2666     default:
2667         abort();
2668     }
2669 
2670     options = qdict_new();
2671     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2672     qdict_put_str(options, "detect-zeroes", detect_zeroes ? "on" : "off");
2673 
2674     if (has_format) {
2675         qdict_put_str(options, "driver", format);
2676     }
2677 
2678     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2679     if (!medium_bs) {
2680         goto fail;
2681     }
2682 
2683     rc = do_open_tray(has_device ? device : NULL,
2684                       has_id ? id : NULL,
2685                       false, &err);
2686     if (rc && rc != -ENOSYS) {
2687         error_propagate(errp, err);
2688         goto fail;
2689     }
2690     error_free(err);
2691     err = NULL;
2692 
2693     blockdev_remove_medium(has_device, device, has_id, id, &err);
2694     if (err) {
2695         error_propagate(errp, err);
2696         goto fail;
2697     }
2698 
2699     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2700     if (err) {
2701         error_propagate(errp, err);
2702         goto fail;
2703     }
2704 
2705     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2706 
2707 fail:
2708     /* If the medium has been inserted, the device has its own reference, so
2709      * ours must be relinquished; and if it has not been inserted successfully,
2710      * the reference must be relinquished anyway */
2711     bdrv_unref(medium_bs);
2712 }
2713 
2714 /* throttling disk I/O limits */
2715 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2716 {
2717     ThrottleConfig cfg;
2718     BlockDriverState *bs;
2719     BlockBackend *blk;
2720     AioContext *aio_context;
2721 
2722     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2723                       arg->has_id ? arg->id : NULL,
2724                       errp);
2725     if (!blk) {
2726         return;
2727     }
2728 
2729     aio_context = blk_get_aio_context(blk);
2730     aio_context_acquire(aio_context);
2731 
2732     bs = blk_bs(blk);
2733     if (!bs) {
2734         error_setg(errp, "Device has no medium");
2735         goto out;
2736     }
2737 
2738     throttle_config_init(&cfg);
2739     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2740     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2741     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2742 
2743     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2744     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2745     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2746 
2747     if (arg->has_bps_max) {
2748         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2749     }
2750     if (arg->has_bps_rd_max) {
2751         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2752     }
2753     if (arg->has_bps_wr_max) {
2754         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2755     }
2756     if (arg->has_iops_max) {
2757         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2758     }
2759     if (arg->has_iops_rd_max) {
2760         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2761     }
2762     if (arg->has_iops_wr_max) {
2763         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2764     }
2765 
2766     if (arg->has_bps_max_length) {
2767         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2768     }
2769     if (arg->has_bps_rd_max_length) {
2770         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2771     }
2772     if (arg->has_bps_wr_max_length) {
2773         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2774     }
2775     if (arg->has_iops_max_length) {
2776         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2777     }
2778     if (arg->has_iops_rd_max_length) {
2779         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2780     }
2781     if (arg->has_iops_wr_max_length) {
2782         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2783     }
2784 
2785     if (arg->has_iops_size) {
2786         cfg.op_size = arg->iops_size;
2787     }
2788 
2789     if (!throttle_is_valid(&cfg, errp)) {
2790         goto out;
2791     }
2792 
2793     if (throttle_enabled(&cfg)) {
2794         /* Enable I/O limits if they're not enabled yet, otherwise
2795          * just update the throttling group. */
2796         if (!blk_get_public(blk)->throttle_group_member.throttle_state) {
2797             blk_io_limits_enable(blk,
2798                                  arg->has_group ? arg->group :
2799                                  arg->has_device ? arg->device :
2800                                  arg->id);
2801         } else if (arg->has_group) {
2802             blk_io_limits_update_group(blk, arg->group);
2803         }
2804         /* Set the new throttling configuration */
2805         blk_set_io_limits(blk, &cfg);
2806     } else if (blk_get_public(blk)->throttle_group_member.throttle_state) {
2807         /* If all throttling settings are set to 0, disable I/O limits */
2808         blk_io_limits_disable(blk);
2809     }
2810 
2811 out:
2812     aio_context_release(aio_context);
2813 }
2814 
2815 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2816                                 bool has_granularity, uint32_t granularity,
2817                                 bool has_persistent, bool persistent,
2818                                 bool has_autoload, bool autoload,
2819                                 bool has_disabled, bool disabled,
2820                                 Error **errp)
2821 {
2822     BlockDriverState *bs;
2823     BdrvDirtyBitmap *bitmap;
2824     AioContext *aio_context = NULL;
2825 
2826     if (!name || name[0] == '\0') {
2827         error_setg(errp, "Bitmap name cannot be empty");
2828         return;
2829     }
2830 
2831     bs = bdrv_lookup_bs(node, node, errp);
2832     if (!bs) {
2833         return;
2834     }
2835 
2836     if (has_granularity) {
2837         if (granularity < 512 || !is_power_of_2(granularity)) {
2838             error_setg(errp, "Granularity must be power of 2 "
2839                              "and at least 512");
2840             return;
2841         }
2842     } else {
2843         /* Default to cluster size, if available: */
2844         granularity = bdrv_get_default_bitmap_granularity(bs);
2845     }
2846 
2847     if (!has_persistent) {
2848         persistent = false;
2849     }
2850 
2851     if (has_autoload) {
2852         warn_report("Autoload option is deprecated and its value is ignored");
2853     }
2854 
2855     if (!has_disabled) {
2856         disabled = false;
2857     }
2858 
2859     if (persistent) {
2860         aio_context = bdrv_get_aio_context(bs);
2861         aio_context_acquire(aio_context);
2862         if (!bdrv_can_store_new_dirty_bitmap(bs, name, granularity, errp)) {
2863             goto out;
2864         }
2865     }
2866 
2867     bitmap = bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2868     if (bitmap == NULL) {
2869         goto out;
2870     }
2871 
2872     if (disabled) {
2873         bdrv_disable_dirty_bitmap(bitmap);
2874     }
2875 
2876     bdrv_dirty_bitmap_set_persistance(bitmap, persistent);
2877  out:
2878     if (aio_context) {
2879         aio_context_release(aio_context);
2880     }
2881 }
2882 
2883 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2884                                    Error **errp)
2885 {
2886     BlockDriverState *bs;
2887     BdrvDirtyBitmap *bitmap;
2888     Error *local_err = NULL;
2889     AioContext *aio_context = NULL;
2890 
2891     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2892     if (!bitmap || !bs) {
2893         return;
2894     }
2895 
2896     if (bdrv_dirty_bitmap_user_locked(bitmap)) {
2897         error_setg(errp,
2898                    "Bitmap '%s' is currently in use by another operation and"
2899                    " cannot be removed", name);
2900         return;
2901     }
2902 
2903     if (bdrv_dirty_bitmap_get_persistance(bitmap)) {
2904         aio_context = bdrv_get_aio_context(bs);
2905         aio_context_acquire(aio_context);
2906         bdrv_remove_persistent_dirty_bitmap(bs, name, &local_err);
2907         if (local_err != NULL) {
2908             error_propagate(errp, local_err);
2909             goto out;
2910         }
2911     }
2912 
2913     bdrv_release_dirty_bitmap(bs, bitmap);
2914  out:
2915     if (aio_context) {
2916         aio_context_release(aio_context);
2917     }
2918 }
2919 
2920 /**
2921  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2922  * immediately after a full backup operation.
2923  */
2924 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2925                                   Error **errp)
2926 {
2927     BdrvDirtyBitmap *bitmap;
2928     BlockDriverState *bs;
2929 
2930     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2931     if (!bitmap || !bs) {
2932         return;
2933     }
2934 
2935     if (bdrv_dirty_bitmap_user_locked(bitmap)) {
2936         error_setg(errp,
2937                    "Bitmap '%s' is currently in use by another operation"
2938                    " and cannot be cleared", name);
2939         return;
2940     } else if (bdrv_dirty_bitmap_readonly(bitmap)) {
2941         error_setg(errp, "Bitmap '%s' is readonly and cannot be cleared", name);
2942         return;
2943     }
2944 
2945     bdrv_clear_dirty_bitmap(bitmap, NULL);
2946 }
2947 
2948 void qmp_block_dirty_bitmap_enable(const char *node, const char *name,
2949                                    Error **errp)
2950 {
2951     BlockDriverState *bs;
2952     BdrvDirtyBitmap *bitmap;
2953 
2954     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2955     if (!bitmap) {
2956         return;
2957     }
2958 
2959     if (bdrv_dirty_bitmap_user_locked(bitmap)) {
2960         error_setg(errp,
2961                    "Bitmap '%s' is currently in use by another operation"
2962                    " and cannot be enabled", name);
2963         return;
2964     }
2965 
2966     bdrv_enable_dirty_bitmap(bitmap);
2967 }
2968 
2969 void qmp_block_dirty_bitmap_disable(const char *node, const char *name,
2970                                     Error **errp)
2971 {
2972     BlockDriverState *bs;
2973     BdrvDirtyBitmap *bitmap;
2974 
2975     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2976     if (!bitmap) {
2977         return;
2978     }
2979 
2980     if (bdrv_dirty_bitmap_user_locked(bitmap)) {
2981         error_setg(errp,
2982                    "Bitmap '%s' is currently in use by another operation"
2983                    " and cannot be disabled", name);
2984         return;
2985     }
2986 
2987     bdrv_disable_dirty_bitmap(bitmap);
2988 }
2989 
2990 static BdrvDirtyBitmap *do_block_dirty_bitmap_merge(const char *node,
2991                                                     const char *target,
2992                                                     strList *bitmaps,
2993                                                     HBitmap **backup,
2994                                                     Error **errp)
2995 {
2996     BlockDriverState *bs;
2997     BdrvDirtyBitmap *dst, *src, *anon;
2998     strList *lst;
2999     Error *local_err = NULL;
3000 
3001     dst = block_dirty_bitmap_lookup(node, target, &bs, errp);
3002     if (!dst) {
3003         return NULL;
3004     }
3005 
3006     anon = bdrv_create_dirty_bitmap(bs, bdrv_dirty_bitmap_granularity(dst),
3007                                     NULL, errp);
3008     if (!anon) {
3009         return NULL;
3010     }
3011 
3012     for (lst = bitmaps; lst; lst = lst->next) {
3013         src = bdrv_find_dirty_bitmap(bs, lst->value);
3014         if (!src) {
3015             error_setg(errp, "Dirty bitmap '%s' not found", lst->value);
3016             dst = NULL;
3017             goto out;
3018         }
3019 
3020         bdrv_merge_dirty_bitmap(anon, src, NULL, &local_err);
3021         if (local_err) {
3022             error_propagate(errp, local_err);
3023             dst = NULL;
3024             goto out;
3025         }
3026     }
3027 
3028     /* Merge into dst; dst is unchanged on failure. */
3029     bdrv_merge_dirty_bitmap(dst, anon, backup, errp);
3030 
3031  out:
3032     bdrv_release_dirty_bitmap(bs, anon);
3033     return dst;
3034 }
3035 
3036 void qmp_block_dirty_bitmap_merge(const char *node, const char *target,
3037                                   strList *bitmaps, Error **errp)
3038 {
3039     do_block_dirty_bitmap_merge(node, target, bitmaps, NULL, errp);
3040 }
3041 
3042 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
3043                                                               const char *name,
3044                                                               Error **errp)
3045 {
3046     BdrvDirtyBitmap *bitmap;
3047     BlockDriverState *bs;
3048     BlockDirtyBitmapSha256 *ret = NULL;
3049     char *sha256;
3050 
3051     bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
3052     if (!bitmap || !bs) {
3053         return NULL;
3054     }
3055 
3056     sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
3057     if (sha256 == NULL) {
3058         return NULL;
3059     }
3060 
3061     ret = g_new(BlockDirtyBitmapSha256, 1);
3062     ret->sha256 = sha256;
3063 
3064     return ret;
3065 }
3066 
3067 void hmp_drive_del(Monitor *mon, const QDict *qdict)
3068 {
3069     const char *id = qdict_get_str(qdict, "id");
3070     BlockBackend *blk;
3071     BlockDriverState *bs;
3072     AioContext *aio_context;
3073     Error *local_err = NULL;
3074 
3075     bs = bdrv_find_node(id);
3076     if (bs) {
3077         qmp_blockdev_del(id, &local_err);
3078         if (local_err) {
3079             error_report_err(local_err);
3080         }
3081         return;
3082     }
3083 
3084     blk = blk_by_name(id);
3085     if (!blk) {
3086         error_report("Device '%s' not found", id);
3087         return;
3088     }
3089 
3090     if (!blk_legacy_dinfo(blk)) {
3091         error_report("Deleting device added with blockdev-add"
3092                      " is not supported");
3093         return;
3094     }
3095 
3096     aio_context = blk_get_aio_context(blk);
3097     aio_context_acquire(aio_context);
3098 
3099     bs = blk_bs(blk);
3100     if (bs) {
3101         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
3102             error_report_err(local_err);
3103             aio_context_release(aio_context);
3104             return;
3105         }
3106 
3107         blk_remove_bs(blk);
3108     }
3109 
3110     /* Make the BlockBackend and the attached BlockDriverState anonymous */
3111     monitor_remove_blk(blk);
3112 
3113     /* If this BlockBackend has a device attached to it, its refcount will be
3114      * decremented when the device is removed; otherwise we have to do so here.
3115      */
3116     if (blk_get_attached_dev(blk)) {
3117         /* Further I/O must not pause the guest */
3118         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
3119                          BLOCKDEV_ON_ERROR_REPORT);
3120     } else {
3121         blk_unref(blk);
3122     }
3123 
3124     aio_context_release(aio_context);
3125 }
3126 
3127 void qmp_block_resize(bool has_device, const char *device,
3128                       bool has_node_name, const char *node_name,
3129                       int64_t size, Error **errp)
3130 {
3131     Error *local_err = NULL;
3132     BlockBackend *blk = NULL;
3133     BlockDriverState *bs;
3134     AioContext *aio_context;
3135     int ret;
3136 
3137     bs = bdrv_lookup_bs(has_device ? device : NULL,
3138                         has_node_name ? node_name : NULL,
3139                         &local_err);
3140     if (local_err) {
3141         error_propagate(errp, local_err);
3142         return;
3143     }
3144 
3145     aio_context = bdrv_get_aio_context(bs);
3146     aio_context_acquire(aio_context);
3147 
3148     if (!bdrv_is_first_non_filter(bs)) {
3149         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
3150         goto out;
3151     }
3152 
3153     if (size < 0) {
3154         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
3155         goto out;
3156     }
3157 
3158     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3159         error_setg(errp, QERR_DEVICE_IN_USE, device);
3160         goto out;
3161     }
3162 
3163     blk = blk_new(BLK_PERM_RESIZE, BLK_PERM_ALL);
3164     ret = blk_insert_bs(blk, bs, errp);
3165     if (ret < 0) {
3166         goto out;
3167     }
3168 
3169     bdrv_drained_begin(bs);
3170     ret = blk_truncate(blk, size, PREALLOC_MODE_OFF, errp);
3171     bdrv_drained_end(bs);
3172 
3173 out:
3174     blk_unref(blk);
3175     aio_context_release(aio_context);
3176 }
3177 
3178 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
3179                       bool has_base, const char *base,
3180                       bool has_base_node, const char *base_node,
3181                       bool has_backing_file, const char *backing_file,
3182                       bool has_speed, int64_t speed,
3183                       bool has_on_error, BlockdevOnError on_error,
3184                       bool has_auto_finalize, bool auto_finalize,
3185                       bool has_auto_dismiss, bool auto_dismiss,
3186                       Error **errp)
3187 {
3188     BlockDriverState *bs, *iter;
3189     BlockDriverState *base_bs = NULL;
3190     AioContext *aio_context;
3191     Error *local_err = NULL;
3192     const char *base_name = NULL;
3193     int job_flags = JOB_DEFAULT;
3194 
3195     if (!has_on_error) {
3196         on_error = BLOCKDEV_ON_ERROR_REPORT;
3197     }
3198 
3199     bs = bdrv_lookup_bs(device, device, errp);
3200     if (!bs) {
3201         return;
3202     }
3203 
3204     aio_context = bdrv_get_aio_context(bs);
3205     aio_context_acquire(aio_context);
3206 
3207     if (has_base && has_base_node) {
3208         error_setg(errp, "'base' and 'base-node' cannot be specified "
3209                    "at the same time");
3210         goto out;
3211     }
3212 
3213     if (has_base) {
3214         base_bs = bdrv_find_backing_image(bs, base);
3215         if (base_bs == NULL) {
3216             error_setg(errp, QERR_BASE_NOT_FOUND, base);
3217             goto out;
3218         }
3219         assert(bdrv_get_aio_context(base_bs) == aio_context);
3220         base_name = base;
3221     }
3222 
3223     if (has_base_node) {
3224         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3225         if (!base_bs) {
3226             goto out;
3227         }
3228         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
3229             error_setg(errp, "Node '%s' is not a backing image of '%s'",
3230                        base_node, device);
3231             goto out;
3232         }
3233         assert(bdrv_get_aio_context(base_bs) == aio_context);
3234         bdrv_refresh_filename(base_bs);
3235         base_name = base_bs->filename;
3236     }
3237 
3238     /* Check for op blockers in the whole chain between bs and base */
3239     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
3240         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
3241             goto out;
3242         }
3243     }
3244 
3245     /* if we are streaming the entire chain, the result will have no backing
3246      * file, and specifying one is therefore an error */
3247     if (base_bs == NULL && has_backing_file) {
3248         error_setg(errp, "backing file specified, but streaming the "
3249                          "entire chain");
3250         goto out;
3251     }
3252 
3253     /* backing_file string overrides base bs filename */
3254     base_name = has_backing_file ? backing_file : base_name;
3255 
3256     if (has_auto_finalize && !auto_finalize) {
3257         job_flags |= JOB_MANUAL_FINALIZE;
3258     }
3259     if (has_auto_dismiss && !auto_dismiss) {
3260         job_flags |= JOB_MANUAL_DISMISS;
3261     }
3262 
3263     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3264                  job_flags, has_speed ? speed : 0, on_error, &local_err);
3265     if (local_err) {
3266         error_propagate(errp, local_err);
3267         goto out;
3268     }
3269 
3270     trace_qmp_block_stream(bs, bs->job);
3271 
3272 out:
3273     aio_context_release(aio_context);
3274 }
3275 
3276 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3277                       bool has_base_node, const char *base_node,
3278                       bool has_base, const char *base,
3279                       bool has_top_node, const char *top_node,
3280                       bool has_top, const char *top,
3281                       bool has_backing_file, const char *backing_file,
3282                       bool has_speed, int64_t speed,
3283                       bool has_filter_node_name, const char *filter_node_name,
3284                       bool has_auto_finalize, bool auto_finalize,
3285                       bool has_auto_dismiss, bool auto_dismiss,
3286                       Error **errp)
3287 {
3288     BlockDriverState *bs;
3289     BlockDriverState *iter;
3290     BlockDriverState *base_bs, *top_bs;
3291     AioContext *aio_context;
3292     Error *local_err = NULL;
3293     /* This will be part of the QMP command, if/when the
3294      * BlockdevOnError change for blkmirror makes it in
3295      */
3296     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3297     int job_flags = JOB_DEFAULT;
3298 
3299     if (!has_speed) {
3300         speed = 0;
3301     }
3302     if (!has_filter_node_name) {
3303         filter_node_name = NULL;
3304     }
3305     if (has_auto_finalize && !auto_finalize) {
3306         job_flags |= JOB_MANUAL_FINALIZE;
3307     }
3308     if (has_auto_dismiss && !auto_dismiss) {
3309         job_flags |= JOB_MANUAL_DISMISS;
3310     }
3311 
3312     /* Important Note:
3313      *  libvirt relies on the DeviceNotFound error class in order to probe for
3314      *  live commit feature versions; for this to work, we must make sure to
3315      *  perform the device lookup before any generic errors that may occur in a
3316      *  scenario in which all optional arguments are omitted. */
3317     bs = qmp_get_root_bs(device, &local_err);
3318     if (!bs) {
3319         bs = bdrv_lookup_bs(device, device, NULL);
3320         if (!bs) {
3321             error_free(local_err);
3322             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3323                       "Device '%s' not found", device);
3324         } else {
3325             error_propagate(errp, local_err);
3326         }
3327         return;
3328     }
3329 
3330     aio_context = bdrv_get_aio_context(bs);
3331     aio_context_acquire(aio_context);
3332 
3333     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3334         goto out;
3335     }
3336 
3337     /* default top_bs is the active layer */
3338     top_bs = bs;
3339 
3340     if (has_top_node && has_top) {
3341         error_setg(errp, "'top-node' and 'top' are mutually exclusive");
3342         goto out;
3343     } else if (has_top_node) {
3344         top_bs = bdrv_lookup_bs(NULL, top_node, errp);
3345         if (top_bs == NULL) {
3346             goto out;
3347         }
3348         if (!bdrv_chain_contains(bs, top_bs)) {
3349             error_setg(errp, "'%s' is not in this backing file chain",
3350                        top_node);
3351             goto out;
3352         }
3353     } else if (has_top && top) {
3354         /* This strcmp() is just a shortcut, there is no need to
3355          * refresh @bs's filename.  If it mismatches,
3356          * bdrv_find_backing_image() will do the refresh and may still
3357          * return @bs. */
3358         if (strcmp(bs->filename, top) != 0) {
3359             top_bs = bdrv_find_backing_image(bs, top);
3360         }
3361     }
3362 
3363     if (top_bs == NULL) {
3364         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3365         goto out;
3366     }
3367 
3368     assert(bdrv_get_aio_context(top_bs) == aio_context);
3369 
3370     if (has_base_node && has_base) {
3371         error_setg(errp, "'base-node' and 'base' are mutually exclusive");
3372         goto out;
3373     } else if (has_base_node) {
3374         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
3375         if (base_bs == NULL) {
3376             goto out;
3377         }
3378         if (!bdrv_chain_contains(top_bs, base_bs)) {
3379             error_setg(errp, "'%s' is not in this backing file chain",
3380                        base_node);
3381             goto out;
3382         }
3383     } else if (has_base && base) {
3384         base_bs = bdrv_find_backing_image(top_bs, base);
3385     } else {
3386         base_bs = bdrv_find_base(top_bs);
3387     }
3388 
3389     if (base_bs == NULL) {
3390         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3391         goto out;
3392     }
3393 
3394     assert(bdrv_get_aio_context(base_bs) == aio_context);
3395 
3396     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3397         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3398             goto out;
3399         }
3400     }
3401 
3402     /* Do not allow attempts to commit an image into itself */
3403     if (top_bs == base_bs) {
3404         error_setg(errp, "cannot commit an image into itself");
3405         goto out;
3406     }
3407 
3408     if (top_bs == bs) {
3409         if (has_backing_file) {
3410             error_setg(errp, "'backing-file' specified,"
3411                              " but 'top' is the active layer");
3412             goto out;
3413         }
3414         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3415                             job_flags, speed, on_error,
3416                             filter_node_name, NULL, NULL, false, &local_err);
3417     } else {
3418         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3419         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3420             goto out;
3421         }
3422         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, job_flags,
3423                      speed, on_error, has_backing_file ? backing_file : NULL,
3424                      filter_node_name, &local_err);
3425     }
3426     if (local_err != NULL) {
3427         error_propagate(errp, local_err);
3428         goto out;
3429     }
3430 
3431 out:
3432     aio_context_release(aio_context);
3433 }
3434 
3435 static BlockJob *do_drive_backup(DriveBackup *backup, JobTxn *txn,
3436                                  Error **errp)
3437 {
3438     BlockDriverState *bs;
3439     BlockDriverState *target_bs;
3440     BlockDriverState *source = NULL;
3441     BlockJob *job = NULL;
3442     BdrvDirtyBitmap *bmap = NULL;
3443     AioContext *aio_context;
3444     QDict *options = NULL;
3445     Error *local_err = NULL;
3446     int flags, job_flags = JOB_DEFAULT;
3447     int64_t size;
3448     bool set_backing_hd = false;
3449 
3450     if (!backup->has_speed) {
3451         backup->speed = 0;
3452     }
3453     if (!backup->has_on_source_error) {
3454         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3455     }
3456     if (!backup->has_on_target_error) {
3457         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3458     }
3459     if (!backup->has_mode) {
3460         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3461     }
3462     if (!backup->has_job_id) {
3463         backup->job_id = NULL;
3464     }
3465     if (!backup->has_auto_finalize) {
3466         backup->auto_finalize = true;
3467     }
3468     if (!backup->has_auto_dismiss) {
3469         backup->auto_dismiss = true;
3470     }
3471     if (!backup->has_compress) {
3472         backup->compress = false;
3473     }
3474 
3475     bs = qmp_get_root_bs(backup->device, errp);
3476     if (!bs) {
3477         return NULL;
3478     }
3479 
3480     aio_context = bdrv_get_aio_context(bs);
3481     aio_context_acquire(aio_context);
3482 
3483     if (!backup->has_format) {
3484         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3485                          NULL : (char*) bs->drv->format_name;
3486     }
3487 
3488     /* Early check to avoid creating target */
3489     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3490         goto out;
3491     }
3492 
3493     flags = bs->open_flags | BDRV_O_RDWR;
3494 
3495     /* See if we have a backing HD we can use to create our new image
3496      * on top of. */
3497     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3498         source = backing_bs(bs);
3499         if (!source) {
3500             backup->sync = MIRROR_SYNC_MODE_FULL;
3501         }
3502     }
3503     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3504         source = bs;
3505         flags |= BDRV_O_NO_BACKING;
3506         set_backing_hd = true;
3507     }
3508 
3509     size = bdrv_getlength(bs);
3510     if (size < 0) {
3511         error_setg_errno(errp, -size, "bdrv_getlength failed");
3512         goto out;
3513     }
3514 
3515     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3516         assert(backup->format);
3517         if (source) {
3518             bdrv_refresh_filename(source);
3519             bdrv_img_create(backup->target, backup->format, source->filename,
3520                             source->drv->format_name, NULL,
3521                             size, flags, false, &local_err);
3522         } else {
3523             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3524                             size, flags, false, &local_err);
3525         }
3526     }
3527 
3528     if (local_err) {
3529         error_propagate(errp, local_err);
3530         goto out;
3531     }
3532 
3533     if (backup->format) {
3534         if (!options) {
3535             options = qdict_new();
3536         }
3537         qdict_put_str(options, "driver", backup->format);
3538     }
3539 
3540     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3541     if (!target_bs) {
3542         goto out;
3543     }
3544 
3545     bdrv_set_aio_context(target_bs, aio_context);
3546 
3547     if (set_backing_hd) {
3548         bdrv_set_backing_hd(target_bs, source, &local_err);
3549         if (local_err) {
3550             bdrv_unref(target_bs);
3551             goto out;
3552         }
3553     }
3554 
3555     if (backup->has_bitmap) {
3556         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3557         if (!bmap) {
3558             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3559             bdrv_unref(target_bs);
3560             goto out;
3561         }
3562         if (bdrv_dirty_bitmap_user_locked(bmap)) {
3563             error_setg(errp,
3564                        "Bitmap '%s' is currently in use by another operation"
3565                        " and cannot be used for backup", backup->bitmap);
3566             goto out;
3567         }
3568     }
3569     if (!backup->auto_finalize) {
3570         job_flags |= JOB_MANUAL_FINALIZE;
3571     }
3572     if (!backup->auto_dismiss) {
3573         job_flags |= JOB_MANUAL_DISMISS;
3574     }
3575 
3576     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3577                             backup->sync, bmap, backup->compress,
3578                             backup->on_source_error, backup->on_target_error,
3579                             job_flags, NULL, NULL, txn, &local_err);
3580     bdrv_unref(target_bs);
3581     if (local_err != NULL) {
3582         error_propagate(errp, local_err);
3583         goto out;
3584     }
3585 
3586 out:
3587     aio_context_release(aio_context);
3588     return job;
3589 }
3590 
3591 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3592 {
3593 
3594     BlockJob *job;
3595     job = do_drive_backup(arg, NULL, errp);
3596     if (job) {
3597         job_start(&job->job);
3598     }
3599 }
3600 
3601 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3602 {
3603     return bdrv_named_nodes_list(errp);
3604 }
3605 
3606 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
3607 {
3608     return bdrv_get_xdbg_block_graph(errp);
3609 }
3610 
3611 BlockJob *do_blockdev_backup(BlockdevBackup *backup, JobTxn *txn,
3612                              Error **errp)
3613 {
3614     BlockDriverState *bs;
3615     BlockDriverState *target_bs;
3616     Error *local_err = NULL;
3617     BdrvDirtyBitmap *bmap = NULL;
3618     AioContext *aio_context;
3619     BlockJob *job = NULL;
3620     int job_flags = JOB_DEFAULT;
3621 
3622     if (!backup->has_speed) {
3623         backup->speed = 0;
3624     }
3625     if (!backup->has_on_source_error) {
3626         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3627     }
3628     if (!backup->has_on_target_error) {
3629         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3630     }
3631     if (!backup->has_job_id) {
3632         backup->job_id = NULL;
3633     }
3634     if (!backup->has_auto_finalize) {
3635         backup->auto_finalize = true;
3636     }
3637     if (!backup->has_auto_dismiss) {
3638         backup->auto_dismiss = true;
3639     }
3640     if (!backup->has_compress) {
3641         backup->compress = false;
3642     }
3643 
3644     bs = bdrv_lookup_bs(backup->device, backup->device, errp);
3645     if (!bs) {
3646         return NULL;
3647     }
3648 
3649     aio_context = bdrv_get_aio_context(bs);
3650     aio_context_acquire(aio_context);
3651 
3652     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3653     if (!target_bs) {
3654         goto out;
3655     }
3656 
3657     if (bdrv_get_aio_context(target_bs) != aio_context) {
3658         if (!bdrv_has_blk(target_bs)) {
3659             /* The target BDS is not attached, we can safely move it to another
3660              * AioContext. */
3661             bdrv_set_aio_context(target_bs, aio_context);
3662         } else {
3663             error_setg(errp, "Target is attached to a different thread from "
3664                              "source.");
3665             goto out;
3666         }
3667     }
3668 
3669     if (backup->has_bitmap) {
3670         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3671         if (!bmap) {
3672             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3673             goto out;
3674         }
3675         if (bdrv_dirty_bitmap_user_locked(bmap)) {
3676             error_setg(errp,
3677                        "Bitmap '%s' is currently in use by another operation"
3678                        " and cannot be used for backup", backup->bitmap);
3679             goto out;
3680         }
3681     }
3682 
3683     if (!backup->auto_finalize) {
3684         job_flags |= JOB_MANUAL_FINALIZE;
3685     }
3686     if (!backup->auto_dismiss) {
3687         job_flags |= JOB_MANUAL_DISMISS;
3688     }
3689     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3690                             backup->sync, bmap, backup->compress,
3691                             backup->on_source_error, backup->on_target_error,
3692                             job_flags, NULL, NULL, txn, &local_err);
3693     if (local_err != NULL) {
3694         error_propagate(errp, local_err);
3695     }
3696 out:
3697     aio_context_release(aio_context);
3698     return job;
3699 }
3700 
3701 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3702 {
3703     BlockJob *job;
3704     job = do_blockdev_backup(arg, NULL, errp);
3705     if (job) {
3706         job_start(&job->job);
3707     }
3708 }
3709 
3710 /* Parameter check and block job starting for drive mirroring.
3711  * Caller should hold @device and @target's aio context (must be the same).
3712  **/
3713 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3714                                    BlockDriverState *target,
3715                                    bool has_replaces, const char *replaces,
3716                                    enum MirrorSyncMode sync,
3717                                    BlockMirrorBackingMode backing_mode,
3718                                    bool has_speed, int64_t speed,
3719                                    bool has_granularity, uint32_t granularity,
3720                                    bool has_buf_size, int64_t buf_size,
3721                                    bool has_on_source_error,
3722                                    BlockdevOnError on_source_error,
3723                                    bool has_on_target_error,
3724                                    BlockdevOnError on_target_error,
3725                                    bool has_unmap, bool unmap,
3726                                    bool has_filter_node_name,
3727                                    const char *filter_node_name,
3728                                    bool has_copy_mode, MirrorCopyMode copy_mode,
3729                                    bool has_auto_finalize, bool auto_finalize,
3730                                    bool has_auto_dismiss, bool auto_dismiss,
3731                                    Error **errp)
3732 {
3733     int job_flags = JOB_DEFAULT;
3734 
3735     if (!has_speed) {
3736         speed = 0;
3737     }
3738     if (!has_on_source_error) {
3739         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3740     }
3741     if (!has_on_target_error) {
3742         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3743     }
3744     if (!has_granularity) {
3745         granularity = 0;
3746     }
3747     if (!has_buf_size) {
3748         buf_size = 0;
3749     }
3750     if (!has_unmap) {
3751         unmap = true;
3752     }
3753     if (!has_filter_node_name) {
3754         filter_node_name = NULL;
3755     }
3756     if (!has_copy_mode) {
3757         copy_mode = MIRROR_COPY_MODE_BACKGROUND;
3758     }
3759     if (has_auto_finalize && !auto_finalize) {
3760         job_flags |= JOB_MANUAL_FINALIZE;
3761     }
3762     if (has_auto_dismiss && !auto_dismiss) {
3763         job_flags |= JOB_MANUAL_DISMISS;
3764     }
3765 
3766     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3767         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3768                    "a value in range [512B, 64MB]");
3769         return;
3770     }
3771     if (granularity & (granularity - 1)) {
3772         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3773                    "power of 2");
3774         return;
3775     }
3776 
3777     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3778         return;
3779     }
3780     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3781         return;
3782     }
3783 
3784     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3785         sync = MIRROR_SYNC_MODE_FULL;
3786     }
3787 
3788     /* pass the node name to replace to mirror start since it's loose coupling
3789      * and will allow to check whether the node still exist at mirror completion
3790      */
3791     mirror_start(job_id, bs, target,
3792                  has_replaces ? replaces : NULL, job_flags,
3793                  speed, granularity, buf_size, sync, backing_mode,
3794                  on_source_error, on_target_error, unmap, filter_node_name,
3795                  copy_mode, errp);
3796 }
3797 
3798 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3799 {
3800     BlockDriverState *bs;
3801     BlockDriverState *source, *target_bs;
3802     AioContext *aio_context;
3803     BlockMirrorBackingMode backing_mode;
3804     Error *local_err = NULL;
3805     QDict *options = NULL;
3806     int flags;
3807     int64_t size;
3808     const char *format = arg->format;
3809 
3810     bs = qmp_get_root_bs(arg->device, errp);
3811     if (!bs) {
3812         return;
3813     }
3814 
3815     /* Early check to avoid creating target */
3816     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3817         return;
3818     }
3819 
3820     aio_context = bdrv_get_aio_context(bs);
3821     aio_context_acquire(aio_context);
3822 
3823     if (!arg->has_mode) {
3824         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3825     }
3826 
3827     if (!arg->has_format) {
3828         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3829                   ? NULL : bs->drv->format_name);
3830     }
3831 
3832     flags = bs->open_flags | BDRV_O_RDWR;
3833     source = backing_bs(bs);
3834     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3835         arg->sync = MIRROR_SYNC_MODE_FULL;
3836     }
3837     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3838         source = bs;
3839     }
3840 
3841     size = bdrv_getlength(bs);
3842     if (size < 0) {
3843         error_setg_errno(errp, -size, "bdrv_getlength failed");
3844         goto out;
3845     }
3846 
3847     if (arg->has_replaces) {
3848         BlockDriverState *to_replace_bs;
3849         AioContext *replace_aio_context;
3850         int64_t replace_size;
3851 
3852         if (!arg->has_node_name) {
3853             error_setg(errp, "a node-name must be provided when replacing a"
3854                              " named node of the graph");
3855             goto out;
3856         }
3857 
3858         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3859 
3860         if (!to_replace_bs) {
3861             error_propagate(errp, local_err);
3862             goto out;
3863         }
3864 
3865         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3866         aio_context_acquire(replace_aio_context);
3867         replace_size = bdrv_getlength(to_replace_bs);
3868         aio_context_release(replace_aio_context);
3869 
3870         if (size != replace_size) {
3871             error_setg(errp, "cannot replace image with a mirror image of "
3872                              "different size");
3873             goto out;
3874         }
3875     }
3876 
3877     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3878         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3879     } else {
3880         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3881     }
3882 
3883     /* Don't open backing image in create() */
3884     flags |= BDRV_O_NO_BACKING;
3885 
3886     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3887         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3888     {
3889         /* create new image w/o backing file */
3890         assert(format);
3891         bdrv_img_create(arg->target, format,
3892                         NULL, NULL, NULL, size, flags, false, &local_err);
3893     } else {
3894         switch (arg->mode) {
3895         case NEW_IMAGE_MODE_EXISTING:
3896             break;
3897         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3898             /* create new image with backing file */
3899             bdrv_refresh_filename(source);
3900             bdrv_img_create(arg->target, format,
3901                             source->filename,
3902                             source->drv->format_name,
3903                             NULL, size, flags, false, &local_err);
3904             break;
3905         default:
3906             abort();
3907         }
3908     }
3909 
3910     if (local_err) {
3911         error_propagate(errp, local_err);
3912         goto out;
3913     }
3914 
3915     options = qdict_new();
3916     if (arg->has_node_name) {
3917         qdict_put_str(options, "node-name", arg->node_name);
3918     }
3919     if (format) {
3920         qdict_put_str(options, "driver", format);
3921     }
3922 
3923     /* Mirroring takes care of copy-on-write using the source's backing
3924      * file.
3925      */
3926     target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3927     if (!target_bs) {
3928         goto out;
3929     }
3930 
3931     bdrv_set_aio_context(target_bs, aio_context);
3932 
3933     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3934                            arg->has_replaces, arg->replaces, arg->sync,
3935                            backing_mode, arg->has_speed, arg->speed,
3936                            arg->has_granularity, arg->granularity,
3937                            arg->has_buf_size, arg->buf_size,
3938                            arg->has_on_source_error, arg->on_source_error,
3939                            arg->has_on_target_error, arg->on_target_error,
3940                            arg->has_unmap, arg->unmap,
3941                            false, NULL,
3942                            arg->has_copy_mode, arg->copy_mode,
3943                            arg->has_auto_finalize, arg->auto_finalize,
3944                            arg->has_auto_dismiss, arg->auto_dismiss,
3945                            &local_err);
3946     bdrv_unref(target_bs);
3947     error_propagate(errp, local_err);
3948 out:
3949     aio_context_release(aio_context);
3950 }
3951 
3952 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3953                          const char *device, const char *target,
3954                          bool has_replaces, const char *replaces,
3955                          MirrorSyncMode sync,
3956                          bool has_speed, int64_t speed,
3957                          bool has_granularity, uint32_t granularity,
3958                          bool has_buf_size, int64_t buf_size,
3959                          bool has_on_source_error,
3960                          BlockdevOnError on_source_error,
3961                          bool has_on_target_error,
3962                          BlockdevOnError on_target_error,
3963                          bool has_filter_node_name,
3964                          const char *filter_node_name,
3965                          bool has_copy_mode, MirrorCopyMode copy_mode,
3966                          bool has_auto_finalize, bool auto_finalize,
3967                          bool has_auto_dismiss, bool auto_dismiss,
3968                          Error **errp)
3969 {
3970     BlockDriverState *bs;
3971     BlockDriverState *target_bs;
3972     AioContext *aio_context;
3973     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3974     Error *local_err = NULL;
3975 
3976     bs = qmp_get_root_bs(device, errp);
3977     if (!bs) {
3978         return;
3979     }
3980 
3981     target_bs = bdrv_lookup_bs(target, target, errp);
3982     if (!target_bs) {
3983         return;
3984     }
3985 
3986     aio_context = bdrv_get_aio_context(bs);
3987     aio_context_acquire(aio_context);
3988 
3989     bdrv_set_aio_context(target_bs, aio_context);
3990 
3991     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3992                            has_replaces, replaces, sync, backing_mode,
3993                            has_speed, speed,
3994                            has_granularity, granularity,
3995                            has_buf_size, buf_size,
3996                            has_on_source_error, on_source_error,
3997                            has_on_target_error, on_target_error,
3998                            true, true,
3999                            has_filter_node_name, filter_node_name,
4000                            has_copy_mode, copy_mode,
4001                            has_auto_finalize, auto_finalize,
4002                            has_auto_dismiss, auto_dismiss,
4003                            &local_err);
4004     error_propagate(errp, local_err);
4005 
4006     aio_context_release(aio_context);
4007 }
4008 
4009 /* Get a block job using its ID and acquire its AioContext */
4010 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
4011                                 Error **errp)
4012 {
4013     BlockJob *job;
4014 
4015     assert(id != NULL);
4016 
4017     *aio_context = NULL;
4018 
4019     job = block_job_get(id);
4020 
4021     if (!job) {
4022         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
4023                   "Block job '%s' not found", id);
4024         return NULL;
4025     }
4026 
4027     *aio_context = blk_get_aio_context(job->blk);
4028     aio_context_acquire(*aio_context);
4029 
4030     return job;
4031 }
4032 
4033 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
4034 {
4035     AioContext *aio_context;
4036     BlockJob *job = find_block_job(device, &aio_context, errp);
4037 
4038     if (!job) {
4039         return;
4040     }
4041 
4042     block_job_set_speed(job, speed, errp);
4043     aio_context_release(aio_context);
4044 }
4045 
4046 void qmp_block_job_cancel(const char *device,
4047                           bool has_force, bool force, Error **errp)
4048 {
4049     AioContext *aio_context;
4050     BlockJob *job = find_block_job(device, &aio_context, errp);
4051 
4052     if (!job) {
4053         return;
4054     }
4055 
4056     if (!has_force) {
4057         force = false;
4058     }
4059 
4060     if (job_user_paused(&job->job) && !force) {
4061         error_setg(errp, "The block job for device '%s' is currently paused",
4062                    device);
4063         goto out;
4064     }
4065 
4066     trace_qmp_block_job_cancel(job);
4067     job_user_cancel(&job->job, force, errp);
4068 out:
4069     aio_context_release(aio_context);
4070 }
4071 
4072 void qmp_block_job_pause(const char *device, Error **errp)
4073 {
4074     AioContext *aio_context;
4075     BlockJob *job = find_block_job(device, &aio_context, errp);
4076 
4077     if (!job) {
4078         return;
4079     }
4080 
4081     trace_qmp_block_job_pause(job);
4082     job_user_pause(&job->job, errp);
4083     aio_context_release(aio_context);
4084 }
4085 
4086 void qmp_block_job_resume(const char *device, Error **errp)
4087 {
4088     AioContext *aio_context;
4089     BlockJob *job = find_block_job(device, &aio_context, errp);
4090 
4091     if (!job) {
4092         return;
4093     }
4094 
4095     trace_qmp_block_job_resume(job);
4096     job_user_resume(&job->job, errp);
4097     aio_context_release(aio_context);
4098 }
4099 
4100 void qmp_block_job_complete(const char *device, Error **errp)
4101 {
4102     AioContext *aio_context;
4103     BlockJob *job = find_block_job(device, &aio_context, errp);
4104 
4105     if (!job) {
4106         return;
4107     }
4108 
4109     trace_qmp_block_job_complete(job);
4110     job_complete(&job->job, errp);
4111     aio_context_release(aio_context);
4112 }
4113 
4114 void qmp_block_job_finalize(const char *id, Error **errp)
4115 {
4116     AioContext *aio_context;
4117     BlockJob *job = find_block_job(id, &aio_context, errp);
4118 
4119     if (!job) {
4120         return;
4121     }
4122 
4123     trace_qmp_block_job_finalize(job);
4124     job_finalize(&job->job, errp);
4125     aio_context_release(aio_context);
4126 }
4127 
4128 void qmp_block_job_dismiss(const char *id, Error **errp)
4129 {
4130     AioContext *aio_context;
4131     BlockJob *bjob = find_block_job(id, &aio_context, errp);
4132     Job *job;
4133 
4134     if (!bjob) {
4135         return;
4136     }
4137 
4138     trace_qmp_block_job_dismiss(bjob);
4139     job = &bjob->job;
4140     job_dismiss(&job, errp);
4141     aio_context_release(aio_context);
4142 }
4143 
4144 void qmp_change_backing_file(const char *device,
4145                              const char *image_node_name,
4146                              const char *backing_file,
4147                              Error **errp)
4148 {
4149     BlockDriverState *bs = NULL;
4150     AioContext *aio_context;
4151     BlockDriverState *image_bs = NULL;
4152     Error *local_err = NULL;
4153     bool ro;
4154     int ret;
4155 
4156     bs = qmp_get_root_bs(device, errp);
4157     if (!bs) {
4158         return;
4159     }
4160 
4161     aio_context = bdrv_get_aio_context(bs);
4162     aio_context_acquire(aio_context);
4163 
4164     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
4165     if (local_err) {
4166         error_propagate(errp, local_err);
4167         goto out;
4168     }
4169 
4170     if (!image_bs) {
4171         error_setg(errp, "image file not found");
4172         goto out;
4173     }
4174 
4175     if (bdrv_find_base(image_bs) == image_bs) {
4176         error_setg(errp, "not allowing backing file change on an image "
4177                          "without a backing file");
4178         goto out;
4179     }
4180 
4181     /* even though we are not necessarily operating on bs, we need it to
4182      * determine if block ops are currently prohibited on the chain */
4183     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
4184         goto out;
4185     }
4186 
4187     /* final sanity check */
4188     if (!bdrv_chain_contains(bs, image_bs)) {
4189         error_setg(errp, "'%s' and image file are not in the same chain",
4190                    device);
4191         goto out;
4192     }
4193 
4194     /* if not r/w, reopen to make r/w */
4195     ro = bdrv_is_read_only(image_bs);
4196 
4197     if (ro) {
4198         if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
4199             goto out;
4200         }
4201     }
4202 
4203     ret = bdrv_change_backing_file(image_bs, backing_file,
4204                                image_bs->drv ? image_bs->drv->format_name : "");
4205 
4206     if (ret < 0) {
4207         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
4208                          backing_file);
4209         /* don't exit here, so we can try to restore open flags if
4210          * appropriate */
4211     }
4212 
4213     if (ro) {
4214         bdrv_reopen_set_read_only(image_bs, true, &local_err);
4215         error_propagate(errp, local_err);
4216     }
4217 
4218 out:
4219     aio_context_release(aio_context);
4220 }
4221 
4222 void hmp_drive_add_node(Monitor *mon, const char *optstr)
4223 {
4224     QemuOpts *opts;
4225     QDict *qdict;
4226     Error *local_err = NULL;
4227 
4228     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
4229     if (!opts) {
4230         return;
4231     }
4232 
4233     qdict = qemu_opts_to_qdict(opts, NULL);
4234 
4235     if (!qdict_get_try_str(qdict, "node-name")) {
4236         qobject_unref(qdict);
4237         error_report("'node-name' needs to be specified");
4238         goto out;
4239     }
4240 
4241     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
4242     if (!bs) {
4243         error_report_err(local_err);
4244         goto out;
4245     }
4246 
4247     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4248 
4249 out:
4250     qemu_opts_del(opts);
4251 }
4252 
4253 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
4254 {
4255     BlockDriverState *bs;
4256     QObject *obj;
4257     Visitor *v = qobject_output_visitor_new(&obj);
4258     QDict *qdict;
4259     Error *local_err = NULL;
4260 
4261     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
4262     if (local_err) {
4263         error_propagate(errp, local_err);
4264         goto fail;
4265     }
4266 
4267     visit_complete(v, &obj);
4268     qdict = qobject_to(QDict, obj);
4269 
4270     qdict_flatten(qdict);
4271 
4272     if (!qdict_get_try_str(qdict, "node-name")) {
4273         error_setg(errp, "'node-name' must be specified for the root node");
4274         goto fail;
4275     }
4276 
4277     bs = bds_tree_init(qdict, errp);
4278     if (!bs) {
4279         goto fail;
4280     }
4281 
4282     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
4283 
4284 fail:
4285     visit_free(v);
4286 }
4287 
4288 void qmp_blockdev_del(const char *node_name, Error **errp)
4289 {
4290     AioContext *aio_context;
4291     BlockDriverState *bs;
4292 
4293     bs = bdrv_find_node(node_name);
4294     if (!bs) {
4295         error_setg(errp, "Cannot find node %s", node_name);
4296         return;
4297     }
4298     if (bdrv_has_blk(bs)) {
4299         error_setg(errp, "Node %s is in use", node_name);
4300         return;
4301     }
4302     aio_context = bdrv_get_aio_context(bs);
4303     aio_context_acquire(aio_context);
4304 
4305     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
4306         goto out;
4307     }
4308 
4309     if (!QTAILQ_IN_USE(bs, monitor_list)) {
4310         error_setg(errp, "Node %s is not owned by the monitor",
4311                    bs->node_name);
4312         goto out;
4313     }
4314 
4315     if (bs->refcnt > 1) {
4316         error_setg(errp, "Block device %s is in use",
4317                    bdrv_get_device_or_node_name(bs));
4318         goto out;
4319     }
4320 
4321     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
4322     bdrv_unref(bs);
4323 
4324 out:
4325     aio_context_release(aio_context);
4326 }
4327 
4328 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
4329                                   const char *child_name)
4330 {
4331     BdrvChild *child;
4332 
4333     QLIST_FOREACH(child, &parent_bs->children, next) {
4334         if (strcmp(child->name, child_name) == 0) {
4335             return child;
4336         }
4337     }
4338 
4339     return NULL;
4340 }
4341 
4342 void qmp_x_blockdev_change(const char *parent, bool has_child,
4343                            const char *child, bool has_node,
4344                            const char *node, Error **errp)
4345 {
4346     BlockDriverState *parent_bs, *new_bs = NULL;
4347     BdrvChild *p_child;
4348 
4349     parent_bs = bdrv_lookup_bs(parent, parent, errp);
4350     if (!parent_bs) {
4351         return;
4352     }
4353 
4354     if (has_child == has_node) {
4355         if (has_child) {
4356             error_setg(errp, "The parameters child and node are in conflict");
4357         } else {
4358             error_setg(errp, "Either child or node must be specified");
4359         }
4360         return;
4361     }
4362 
4363     if (has_child) {
4364         p_child = bdrv_find_child(parent_bs, child);
4365         if (!p_child) {
4366             error_setg(errp, "Node '%s' does not have child '%s'",
4367                        parent, child);
4368             return;
4369         }
4370         bdrv_del_child(parent_bs, p_child, errp);
4371     }
4372 
4373     if (has_node) {
4374         new_bs = bdrv_find_node(node);
4375         if (!new_bs) {
4376             error_setg(errp, "Node '%s' not found", node);
4377             return;
4378         }
4379         bdrv_add_child(parent_bs, new_bs, errp);
4380     }
4381 }
4382 
4383 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
4384 {
4385     BlockJobInfoList *head = NULL, **p_next = &head;
4386     BlockJob *job;
4387 
4388     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
4389         BlockJobInfoList *elem;
4390         AioContext *aio_context;
4391 
4392         if (block_job_is_internal(job)) {
4393             continue;
4394         }
4395         elem = g_new0(BlockJobInfoList, 1);
4396         aio_context = blk_get_aio_context(job->blk);
4397         aio_context_acquire(aio_context);
4398         elem->value = block_job_query(job, errp);
4399         aio_context_release(aio_context);
4400         if (!elem->value) {
4401             g_free(elem);
4402             qapi_free_BlockJobInfoList(head);
4403             return NULL;
4404         }
4405         *p_next = elem;
4406         p_next = &elem->next;
4407     }
4408 
4409     return head;
4410 }
4411 
4412 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
4413                                  bool has_force, bool force, Error **errp)
4414 {
4415     AioContext *old_context;
4416     AioContext *new_context;
4417     BlockDriverState *bs;
4418 
4419     bs = bdrv_find_node(node_name);
4420     if (!bs) {
4421         error_setg(errp, "Cannot find node %s", node_name);
4422         return;
4423     }
4424 
4425     /* Protects against accidents. */
4426     if (!(has_force && force) && bdrv_has_blk(bs)) {
4427         error_setg(errp, "Node %s is associated with a BlockBackend and could "
4428                          "be in use (use force=true to override this check)",
4429                          node_name);
4430         return;
4431     }
4432 
4433     if (iothread->type == QTYPE_QSTRING) {
4434         IOThread *obj = iothread_by_id(iothread->u.s);
4435         if (!obj) {
4436             error_setg(errp, "Cannot find iothread %s", iothread->u.s);
4437             return;
4438         }
4439 
4440         new_context = iothread_get_aio_context(obj);
4441     } else {
4442         new_context = qemu_get_aio_context();
4443     }
4444 
4445     old_context = bdrv_get_aio_context(bs);
4446     aio_context_acquire(old_context);
4447 
4448     bdrv_set_aio_context(bs, new_context);
4449 
4450     aio_context_release(old_context);
4451 }
4452 
4453 void qmp_x_block_latency_histogram_set(
4454     const char *device,
4455     bool has_boundaries, uint64List *boundaries,
4456     bool has_boundaries_read, uint64List *boundaries_read,
4457     bool has_boundaries_write, uint64List *boundaries_write,
4458     bool has_boundaries_flush, uint64List *boundaries_flush,
4459     Error **errp)
4460 {
4461     BlockBackend *blk = blk_by_name(device);
4462     BlockAcctStats *stats;
4463     int ret;
4464 
4465     if (!blk) {
4466         error_setg(errp, "Device '%s' not found", device);
4467         return;
4468     }
4469     stats = blk_get_stats(blk);
4470 
4471     if (!has_boundaries && !has_boundaries_read && !has_boundaries_write &&
4472         !has_boundaries_flush)
4473     {
4474         block_latency_histograms_clear(stats);
4475         return;
4476     }
4477 
4478     if (has_boundaries || has_boundaries_read) {
4479         ret = block_latency_histogram_set(
4480             stats, BLOCK_ACCT_READ,
4481             has_boundaries_read ? boundaries_read : boundaries);
4482         if (ret) {
4483             error_setg(errp, "Device '%s' set read boundaries fail", device);
4484             return;
4485         }
4486     }
4487 
4488     if (has_boundaries || has_boundaries_write) {
4489         ret = block_latency_histogram_set(
4490             stats, BLOCK_ACCT_WRITE,
4491             has_boundaries_write ? boundaries_write : boundaries);
4492         if (ret) {
4493             error_setg(errp, "Device '%s' set write boundaries fail", device);
4494             return;
4495         }
4496     }
4497 
4498     if (has_boundaries || has_boundaries_flush) {
4499         ret = block_latency_histogram_set(
4500             stats, BLOCK_ACCT_FLUSH,
4501             has_boundaries_flush ? boundaries_flush : boundaries);
4502         if (ret) {
4503             error_setg(errp, "Device '%s' set flush boundaries fail", device);
4504             return;
4505         }
4506     }
4507 }
4508 
4509 QemuOptsList qemu_common_drive_opts = {
4510     .name = "drive",
4511     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4512     .desc = {
4513         {
4514             .name = "snapshot",
4515             .type = QEMU_OPT_BOOL,
4516             .help = "enable/disable snapshot mode",
4517         },{
4518             .name = "aio",
4519             .type = QEMU_OPT_STRING,
4520             .help = "host AIO implementation (threads, native)",
4521         },{
4522             .name = BDRV_OPT_CACHE_WB,
4523             .type = QEMU_OPT_BOOL,
4524             .help = "Enable writeback mode",
4525         },{
4526             .name = "format",
4527             .type = QEMU_OPT_STRING,
4528             .help = "disk format (raw, qcow2, ...)",
4529         },{
4530             .name = "rerror",
4531             .type = QEMU_OPT_STRING,
4532             .help = "read error action",
4533         },{
4534             .name = "werror",
4535             .type = QEMU_OPT_STRING,
4536             .help = "write error action",
4537         },{
4538             .name = BDRV_OPT_READ_ONLY,
4539             .type = QEMU_OPT_BOOL,
4540             .help = "open drive file as read-only",
4541         },
4542 
4543         THROTTLE_OPTS,
4544 
4545         {
4546             .name = "throttling.group",
4547             .type = QEMU_OPT_STRING,
4548             .help = "name of the block throttling group",
4549         },{
4550             .name = "copy-on-read",
4551             .type = QEMU_OPT_BOOL,
4552             .help = "copy read data from backing file into image file",
4553         },{
4554             .name = "detect-zeroes",
4555             .type = QEMU_OPT_STRING,
4556             .help = "try to optimize zero writes (off, on, unmap)",
4557         },{
4558             .name = "stats-account-invalid",
4559             .type = QEMU_OPT_BOOL,
4560             .help = "whether to account for invalid I/O operations "
4561                     "in the statistics",
4562         },{
4563             .name = "stats-account-failed",
4564             .type = QEMU_OPT_BOOL,
4565             .help = "whether to account for failed I/O operations "
4566                     "in the statistics",
4567         },
4568         { /* end of list */ }
4569     },
4570 };
4571 
4572 QemuOptsList qemu_drive_opts = {
4573     .name = "drive",
4574     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4575     .desc = {
4576         /*
4577          * no elements => accept any params
4578          * validation will happen later
4579          */
4580         { /* end of list */ }
4581     },
4582 };
4583