xref: /qemu/blockdev.c (revision c81ab0ac90ce50d85a50934f7a519a8bc68cb155)
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 "sysemu/block-backend.h"
34 #include "sysemu/blockdev.h"
35 #include "hw/block/block.h"
36 #include "block/blockjob.h"
37 #include "block/throttle-groups.h"
38 #include "monitor/monitor.h"
39 #include "qemu/error-report.h"
40 #include "qemu/option.h"
41 #include "qemu/config-file.h"
42 #include "qapi/qmp/types.h"
43 #include "qapi-visit.h"
44 #include "qapi/qmp/qerror.h"
45 #include "qapi/qmp-output-visitor.h"
46 #include "qapi/util.h"
47 #include "sysemu/sysemu.h"
48 #include "block/block_int.h"
49 #include "qmp-commands.h"
50 #include "trace.h"
51 #include "sysemu/arch_init.h"
52 
53 static const char *const if_name[IF_COUNT] = {
54     [IF_NONE] = "none",
55     [IF_IDE] = "ide",
56     [IF_SCSI] = "scsi",
57     [IF_FLOPPY] = "floppy",
58     [IF_PFLASH] = "pflash",
59     [IF_MTD] = "mtd",
60     [IF_SD] = "sd",
61     [IF_VIRTIO] = "virtio",
62     [IF_XEN] = "xen",
63 };
64 
65 static int if_max_devs[IF_COUNT] = {
66     /*
67      * Do not change these numbers!  They govern how drive option
68      * index maps to unit and bus.  That mapping is ABI.
69      *
70      * All controllers used to imlement if=T drives need to support
71      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
72      * Otherwise, some index values map to "impossible" bus, unit
73      * values.
74      *
75      * For instance, if you change [IF_SCSI] to 255, -drive
76      * if=scsi,index=12 no longer means bus=1,unit=5, but
77      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
78      * the drive can't be set up.  Regression.
79      */
80     [IF_IDE] = 2,
81     [IF_SCSI] = 7,
82 };
83 
84 /**
85  * Boards may call this to offer board-by-board overrides
86  * of the default, global values.
87  */
88 void override_max_devs(BlockInterfaceType type, int max_devs)
89 {
90     BlockBackend *blk;
91     DriveInfo *dinfo;
92 
93     if (max_devs <= 0) {
94         return;
95     }
96 
97     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
98         dinfo = blk_legacy_dinfo(blk);
99         if (dinfo->type == type) {
100             fprintf(stderr, "Cannot override units-per-bus property of"
101                     " the %s interface, because a drive of that type has"
102                     " already been added.\n", if_name[type]);
103             g_assert_not_reached();
104         }
105     }
106 
107     if_max_devs[type] = max_devs;
108 }
109 
110 /*
111  * We automatically delete the drive when a device using it gets
112  * unplugged.  Questionable feature, but we can't just drop it.
113  * Device models call blockdev_mark_auto_del() to schedule the
114  * automatic deletion, and generic qdev code calls blockdev_auto_del()
115  * when deletion is actually safe.
116  */
117 void blockdev_mark_auto_del(BlockBackend *blk)
118 {
119     DriveInfo *dinfo = blk_legacy_dinfo(blk);
120     BlockDriverState *bs = blk_bs(blk);
121     AioContext *aio_context;
122 
123     if (!dinfo) {
124         return;
125     }
126 
127     if (bs) {
128         aio_context = bdrv_get_aio_context(bs);
129         aio_context_acquire(aio_context);
130 
131         if (bs->job) {
132             block_job_cancel(bs->job);
133         }
134 
135         aio_context_release(aio_context);
136     }
137 
138     dinfo->auto_del = 1;
139 }
140 
141 void blockdev_auto_del(BlockBackend *blk)
142 {
143     DriveInfo *dinfo = blk_legacy_dinfo(blk);
144 
145     if (dinfo && dinfo->auto_del) {
146         blk_unref(blk);
147     }
148 }
149 
150 /**
151  * Returns the current mapping of how many units per bus
152  * a particular interface can support.
153  *
154  *  A positive integer indicates n units per bus.
155  *  0 implies the mapping has not been established.
156  * -1 indicates an invalid BlockInterfaceType was given.
157  */
158 int drive_get_max_devs(BlockInterfaceType type)
159 {
160     if (type >= IF_IDE && type < IF_COUNT) {
161         return if_max_devs[type];
162     }
163 
164     return -1;
165 }
166 
167 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
168 {
169     int max_devs = if_max_devs[type];
170     return max_devs ? index / max_devs : 0;
171 }
172 
173 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
174 {
175     int max_devs = if_max_devs[type];
176     return max_devs ? index % max_devs : index;
177 }
178 
179 QemuOpts *drive_def(const char *optstr)
180 {
181     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
182 }
183 
184 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
185                     const char *optstr)
186 {
187     QemuOpts *opts;
188 
189     opts = drive_def(optstr);
190     if (!opts) {
191         return NULL;
192     }
193     if (type != IF_DEFAULT) {
194         qemu_opt_set(opts, "if", if_name[type], &error_abort);
195     }
196     if (index >= 0) {
197         qemu_opt_set_number(opts, "index", index, &error_abort);
198     }
199     if (file)
200         qemu_opt_set(opts, "file", file, &error_abort);
201     return opts;
202 }
203 
204 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
205 {
206     BlockBackend *blk;
207     DriveInfo *dinfo;
208 
209     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
210         dinfo = blk_legacy_dinfo(blk);
211         if (dinfo && dinfo->type == type
212             && dinfo->bus == bus && dinfo->unit == unit) {
213             return dinfo;
214         }
215     }
216 
217     return NULL;
218 }
219 
220 bool drive_check_orphaned(void)
221 {
222     BlockBackend *blk;
223     DriveInfo *dinfo;
224     bool rs = false;
225 
226     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
227         dinfo = blk_legacy_dinfo(blk);
228         /* If dinfo->bdrv->dev is NULL, it has no device attached. */
229         /* Unless this is a default drive, this may be an oversight. */
230         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
231             dinfo->type != IF_NONE) {
232             fprintf(stderr, "Warning: Orphaned drive without device: "
233                     "id=%s,file=%s,if=%s,bus=%d,unit=%d\n",
234                     blk_name(blk), blk_bs(blk) ? blk_bs(blk)->filename : "",
235                     if_name[dinfo->type], dinfo->bus, dinfo->unit);
236             rs = true;
237         }
238     }
239 
240     return rs;
241 }
242 
243 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
244 {
245     return drive_get(type,
246                      drive_index_to_bus_id(type, index),
247                      drive_index_to_unit_id(type, index));
248 }
249 
250 int drive_get_max_bus(BlockInterfaceType type)
251 {
252     int max_bus;
253     BlockBackend *blk;
254     DriveInfo *dinfo;
255 
256     max_bus = -1;
257     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
258         dinfo = blk_legacy_dinfo(blk);
259         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
260             max_bus = dinfo->bus;
261         }
262     }
263     return max_bus;
264 }
265 
266 /* Get a block device.  This should only be used for single-drive devices
267    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
268    appropriate bus.  */
269 DriveInfo *drive_get_next(BlockInterfaceType type)
270 {
271     static int next_block_unit[IF_COUNT];
272 
273     return drive_get(type, 0, next_block_unit[type]++);
274 }
275 
276 static void bdrv_format_print(void *opaque, const char *name)
277 {
278     error_printf(" %s", name);
279 }
280 
281 typedef struct {
282     QEMUBH *bh;
283     BlockDriverState *bs;
284 } BDRVPutRefBH;
285 
286 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
287 {
288     if (!strcmp(buf, "ignore")) {
289         return BLOCKDEV_ON_ERROR_IGNORE;
290     } else if (!is_read && !strcmp(buf, "enospc")) {
291         return BLOCKDEV_ON_ERROR_ENOSPC;
292     } else if (!strcmp(buf, "stop")) {
293         return BLOCKDEV_ON_ERROR_STOP;
294     } else if (!strcmp(buf, "report")) {
295         return BLOCKDEV_ON_ERROR_REPORT;
296     } else {
297         error_setg(errp, "'%s' invalid %s error action",
298                    buf, is_read ? "read" : "write");
299         return -1;
300     }
301 }
302 
303 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
304                                   Error **errp)
305 {
306     const QListEntry *entry;
307     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
308         switch (qobject_type(entry->value)) {
309 
310         case QTYPE_QSTRING: {
311             unsigned long long length;
312             const char *str = qstring_get_str(qobject_to_qstring(entry->value));
313             if (parse_uint_full(str, &length, 10) == 0 &&
314                 length > 0 && length <= UINT_MAX) {
315                 block_acct_add_interval(stats, (unsigned) length);
316             } else {
317                 error_setg(errp, "Invalid interval length: %s", str);
318                 return false;
319             }
320             break;
321         }
322 
323         case QTYPE_QINT: {
324             int64_t length = qint_get_int(qobject_to_qint(entry->value));
325             if (length > 0 && length <= UINT_MAX) {
326                 block_acct_add_interval(stats, (unsigned) length);
327             } else {
328                 error_setg(errp, "Invalid interval length: %" PRId64, length);
329                 return false;
330             }
331             break;
332         }
333 
334         default:
335             error_setg(errp, "The specification of stats-intervals is invalid");
336             return false;
337         }
338     }
339     return true;
340 }
341 
342 static bool check_throttle_config(ThrottleConfig *cfg, Error **errp)
343 {
344     if (throttle_conflicting(cfg)) {
345         error_setg(errp, "bps/iops/max total values and read/write values"
346                          " cannot be used at the same time");
347         return false;
348     }
349 
350     if (!throttle_is_valid(cfg)) {
351         error_setg(errp, "bps/iops/maxs values must be 0 or greater");
352         return false;
353     }
354 
355     if (throttle_max_is_missing_limit(cfg)) {
356         error_setg(errp, "bps_max/iops_max require corresponding"
357                          " bps/iops values");
358         return false;
359     }
360 
361     return true;
362 }
363 
364 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
365 
366 /* All parameters but @opts are optional and may be set to NULL. */
367 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
368     const char **throttling_group, ThrottleConfig *throttle_cfg,
369     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
370 {
371     const char *discard;
372     Error *local_error = NULL;
373     const char *aio;
374 
375     if (bdrv_flags) {
376         if (!qemu_opt_get_bool(opts, "read-only", false)) {
377             *bdrv_flags |= BDRV_O_RDWR;
378         }
379         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
380             *bdrv_flags |= BDRV_O_COPY_ON_READ;
381         }
382 
383         if ((discard = qemu_opt_get(opts, "discard")) != NULL) {
384             if (bdrv_parse_discard_flags(discard, bdrv_flags) != 0) {
385                 error_setg(errp, "Invalid discard option");
386                 return;
387             }
388         }
389 
390         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
391             if (!strcmp(aio, "native")) {
392                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
393             } else if (!strcmp(aio, "threads")) {
394                 /* this is the default */
395             } else {
396                error_setg(errp, "invalid aio option");
397                return;
398             }
399         }
400     }
401 
402     /* disk I/O throttling */
403     if (throttling_group) {
404         *throttling_group = qemu_opt_get(opts, "throttling.group");
405     }
406 
407     if (throttle_cfg) {
408         memset(throttle_cfg, 0, sizeof(*throttle_cfg));
409         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
410             qemu_opt_get_number(opts, "throttling.bps-total", 0);
411         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
412             qemu_opt_get_number(opts, "throttling.bps-read", 0);
413         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
414             qemu_opt_get_number(opts, "throttling.bps-write", 0);
415         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
416             qemu_opt_get_number(opts, "throttling.iops-total", 0);
417         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
418             qemu_opt_get_number(opts, "throttling.iops-read", 0);
419         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
420             qemu_opt_get_number(opts, "throttling.iops-write", 0);
421 
422         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
423             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
424         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
425             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
426         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
427             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
428         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
429             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
430         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
431             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
432         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
433             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
434 
435         throttle_cfg->op_size =
436             qemu_opt_get_number(opts, "throttling.iops-size", 0);
437 
438         if (!check_throttle_config(throttle_cfg, errp)) {
439             return;
440         }
441     }
442 
443     if (detect_zeroes) {
444         *detect_zeroes =
445             qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
446                             qemu_opt_get(opts, "detect-zeroes"),
447                             BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
448                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
449                             &local_error);
450         if (local_error) {
451             error_propagate(errp, local_error);
452             return;
453         }
454 
455         if (bdrv_flags &&
456             *detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
457             !(*bdrv_flags & BDRV_O_UNMAP))
458         {
459             error_setg(errp, "setting detect-zeroes to unmap is not allowed "
460                              "without setting discard operation to unmap");
461             return;
462         }
463     }
464 }
465 
466 /* Takes the ownership of bs_opts */
467 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
468                                    Error **errp)
469 {
470     const char *buf;
471     int bdrv_flags = 0;
472     int on_read_error, on_write_error;
473     bool account_invalid, account_failed;
474     BlockBackend *blk;
475     BlockDriverState *bs;
476     ThrottleConfig cfg;
477     int snapshot = 0;
478     Error *error = NULL;
479     QemuOpts *opts;
480     QDict *interval_dict = NULL;
481     QList *interval_list = NULL;
482     const char *id;
483     BlockdevDetectZeroesOptions detect_zeroes =
484         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
485     const char *throttling_group = NULL;
486 
487     /* Check common options by copying from bs_opts to opts, all other options
488      * stay in bs_opts for processing by bdrv_open(). */
489     id = qdict_get_try_str(bs_opts, "id");
490     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
491     if (error) {
492         error_propagate(errp, error);
493         goto err_no_opts;
494     }
495 
496     qemu_opts_absorb_qdict(opts, bs_opts, &error);
497     if (error) {
498         error_propagate(errp, error);
499         goto early_err;
500     }
501 
502     if (id) {
503         qdict_del(bs_opts, "id");
504     }
505 
506     /* extract parameters */
507     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
508 
509     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
510     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
511 
512     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
513     qdict_array_split(interval_dict, &interval_list);
514 
515     if (qdict_size(interval_dict) != 0) {
516         error_setg(errp, "Invalid option stats-intervals.%s",
517                    qdict_first(interval_dict)->key);
518         goto early_err;
519     }
520 
521     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
522                                     &detect_zeroes, &error);
523     if (error) {
524         error_propagate(errp, error);
525         goto early_err;
526     }
527 
528     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
529         if (is_help_option(buf)) {
530             error_printf("Supported formats:");
531             bdrv_iterate_format(bdrv_format_print, NULL);
532             error_printf("\n");
533             goto early_err;
534         }
535 
536         if (qdict_haskey(bs_opts, "driver")) {
537             error_setg(errp, "Cannot specify both 'driver' and 'format'");
538             goto early_err;
539         }
540         qdict_put(bs_opts, "driver", qstring_from_str(buf));
541     }
542 
543     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
544     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
545         on_write_error = parse_block_error_action(buf, 0, &error);
546         if (error) {
547             error_propagate(errp, error);
548             goto early_err;
549         }
550     }
551 
552     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
553     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
554         on_read_error = parse_block_error_action(buf, 1, &error);
555         if (error) {
556             error_propagate(errp, error);
557             goto early_err;
558         }
559     }
560 
561     if (snapshot) {
562         bdrv_flags |= BDRV_O_SNAPSHOT;
563     }
564 
565     /* init */
566     if ((!file || !*file) && !qdict_size(bs_opts)) {
567         BlockBackendRootState *blk_rs;
568 
569         blk = blk_new(qemu_opts_id(opts), errp);
570         if (!blk) {
571             goto early_err;
572         }
573 
574         blk_rs = blk_get_root_state(blk);
575         blk_rs->open_flags    = bdrv_flags;
576         blk_rs->read_only     = !(bdrv_flags & BDRV_O_RDWR);
577         blk_rs->detect_zeroes = detect_zeroes;
578 
579         if (throttle_enabled(&cfg)) {
580             if (!throttling_group) {
581                 throttling_group = blk_name(blk);
582             }
583             blk_rs->throttle_group = g_strdup(throttling_group);
584             blk_rs->throttle_state = throttle_group_incref(throttling_group);
585             blk_rs->throttle_state->cfg = cfg;
586         }
587 
588         QDECREF(bs_opts);
589     } else {
590         if (file && !*file) {
591             file = NULL;
592         }
593 
594         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
595          * with other callers) rather than what we want as the real defaults.
596          * Apply the defaults here instead. */
597         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_WB, "on");
598         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
599         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
600 
601         if (snapshot) {
602             /* always use cache=unsafe with snapshot */
603             qdict_put(bs_opts, BDRV_OPT_CACHE_WB, qstring_from_str("on"));
604             qdict_put(bs_opts, BDRV_OPT_CACHE_DIRECT, qstring_from_str("off"));
605             qdict_put(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, qstring_from_str("on"));
606         }
607 
608         blk = blk_new_open(qemu_opts_id(opts), file, NULL, bs_opts, bdrv_flags,
609                            errp);
610         if (!blk) {
611             goto err_no_bs_opts;
612         }
613         bs = blk_bs(blk);
614 
615         bs->detect_zeroes = detect_zeroes;
616 
617         /* disk I/O throttling */
618         if (throttle_enabled(&cfg)) {
619             if (!throttling_group) {
620                 throttling_group = blk_name(blk);
621             }
622             bdrv_io_limits_enable(bs, throttling_group);
623             bdrv_set_io_limits(bs, &cfg);
624         }
625 
626         if (bdrv_key_required(bs)) {
627             autostart = 0;
628         }
629 
630         block_acct_init(blk_get_stats(blk), account_invalid, account_failed);
631 
632         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
633             blk_unref(blk);
634             blk = NULL;
635             goto err_no_bs_opts;
636         }
637     }
638 
639     blk_set_on_error(blk, on_read_error, on_write_error);
640 
641 err_no_bs_opts:
642     qemu_opts_del(opts);
643     QDECREF(interval_dict);
644     QDECREF(interval_list);
645     return blk;
646 
647 early_err:
648     qemu_opts_del(opts);
649     QDECREF(interval_dict);
650     QDECREF(interval_list);
651 err_no_opts:
652     QDECREF(bs_opts);
653     return NULL;
654 }
655 
656 static QemuOptsList qemu_root_bds_opts;
657 
658 /* Takes the ownership of bs_opts */
659 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
660 {
661     BlockDriverState *bs;
662     QemuOpts *opts;
663     Error *local_error = NULL;
664     BlockdevDetectZeroesOptions detect_zeroes;
665     int ret;
666     int bdrv_flags = 0;
667 
668     opts = qemu_opts_create(&qemu_root_bds_opts, NULL, 1, errp);
669     if (!opts) {
670         goto fail;
671     }
672 
673     qemu_opts_absorb_qdict(opts, bs_opts, &local_error);
674     if (local_error) {
675         error_propagate(errp, local_error);
676         goto fail;
677     }
678 
679     extract_common_blockdev_options(opts, &bdrv_flags, NULL, NULL,
680                                     &detect_zeroes, &local_error);
681     if (local_error) {
682         error_propagate(errp, local_error);
683         goto fail;
684     }
685 
686     bs = NULL;
687     ret = bdrv_open(&bs, NULL, NULL, bs_opts, bdrv_flags, errp);
688     if (ret < 0) {
689         goto fail_no_bs_opts;
690     }
691 
692     bs->detect_zeroes = detect_zeroes;
693 
694 fail_no_bs_opts:
695     qemu_opts_del(opts);
696     return bs;
697 
698 fail:
699     qemu_opts_del(opts);
700     QDECREF(bs_opts);
701     return NULL;
702 }
703 
704 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
705                             Error **errp)
706 {
707     const char *value;
708 
709     value = qemu_opt_get(opts, from);
710     if (value) {
711         if (qemu_opt_find(opts, to)) {
712             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
713                        "same time", to, from);
714             return;
715         }
716     }
717 
718     /* rename all items in opts */
719     while ((value = qemu_opt_get(opts, from))) {
720         qemu_opt_set(opts, to, value, &error_abort);
721         qemu_opt_unset(opts, from);
722     }
723 }
724 
725 QemuOptsList qemu_legacy_drive_opts = {
726     .name = "drive",
727     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
728     .desc = {
729         {
730             .name = "bus",
731             .type = QEMU_OPT_NUMBER,
732             .help = "bus number",
733         },{
734             .name = "unit",
735             .type = QEMU_OPT_NUMBER,
736             .help = "unit number (i.e. lun for scsi)",
737         },{
738             .name = "index",
739             .type = QEMU_OPT_NUMBER,
740             .help = "index number",
741         },{
742             .name = "media",
743             .type = QEMU_OPT_STRING,
744             .help = "media type (disk, cdrom)",
745         },{
746             .name = "if",
747             .type = QEMU_OPT_STRING,
748             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
749         },{
750             .name = "cyls",
751             .type = QEMU_OPT_NUMBER,
752             .help = "number of cylinders (ide disk geometry)",
753         },{
754             .name = "heads",
755             .type = QEMU_OPT_NUMBER,
756             .help = "number of heads (ide disk geometry)",
757         },{
758             .name = "secs",
759             .type = QEMU_OPT_NUMBER,
760             .help = "number of sectors (ide disk geometry)",
761         },{
762             .name = "trans",
763             .type = QEMU_OPT_STRING,
764             .help = "chs translation (auto, lba, none)",
765         },{
766             .name = "boot",
767             .type = QEMU_OPT_BOOL,
768             .help = "(deprecated, ignored)",
769         },{
770             .name = "addr",
771             .type = QEMU_OPT_STRING,
772             .help = "pci address (virtio only)",
773         },{
774             .name = "serial",
775             .type = QEMU_OPT_STRING,
776             .help = "disk serial number",
777         },{
778             .name = "file",
779             .type = QEMU_OPT_STRING,
780             .help = "file name",
781         },
782 
783         /* Options that are passed on, but have special semantics with -drive */
784         {
785             .name = "read-only",
786             .type = QEMU_OPT_BOOL,
787             .help = "open drive file as read-only",
788         },{
789             .name = "rerror",
790             .type = QEMU_OPT_STRING,
791             .help = "read error action",
792         },{
793             .name = "werror",
794             .type = QEMU_OPT_STRING,
795             .help = "write error action",
796         },{
797             .name = "copy-on-read",
798             .type = QEMU_OPT_BOOL,
799             .help = "copy read data from backing file into image file",
800         },
801 
802         { /* end of list */ }
803     },
804 };
805 
806 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
807 {
808     const char *value;
809     BlockBackend *blk;
810     DriveInfo *dinfo = NULL;
811     QDict *bs_opts;
812     QemuOpts *legacy_opts;
813     DriveMediaType media = MEDIA_DISK;
814     BlockInterfaceType type;
815     int cyls, heads, secs, translation;
816     int max_devs, bus_id, unit_id, index;
817     const char *devaddr;
818     const char *werror, *rerror;
819     bool read_only = false;
820     bool copy_on_read;
821     const char *serial;
822     const char *filename;
823     Error *local_err = NULL;
824     int i;
825 
826     /* Change legacy command line options into QMP ones */
827     static const struct {
828         const char *from;
829         const char *to;
830     } opt_renames[] = {
831         { "iops",           "throttling.iops-total" },
832         { "iops_rd",        "throttling.iops-read" },
833         { "iops_wr",        "throttling.iops-write" },
834 
835         { "bps",            "throttling.bps-total" },
836         { "bps_rd",         "throttling.bps-read" },
837         { "bps_wr",         "throttling.bps-write" },
838 
839         { "iops_max",       "throttling.iops-total-max" },
840         { "iops_rd_max",    "throttling.iops-read-max" },
841         { "iops_wr_max",    "throttling.iops-write-max" },
842 
843         { "bps_max",        "throttling.bps-total-max" },
844         { "bps_rd_max",     "throttling.bps-read-max" },
845         { "bps_wr_max",     "throttling.bps-write-max" },
846 
847         { "iops_size",      "throttling.iops-size" },
848 
849         { "group",          "throttling.group" },
850 
851         { "readonly",       "read-only" },
852     };
853 
854     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
855         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
856                         &local_err);
857         if (local_err) {
858             error_report_err(local_err);
859             return NULL;
860         }
861     }
862 
863     value = qemu_opt_get(all_opts, "cache");
864     if (value) {
865         int flags = 0;
866 
867         if (bdrv_parse_cache_flags(value, &flags) != 0) {
868             error_report("invalid cache option");
869             return NULL;
870         }
871 
872         /* Specific options take precedence */
873         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
874             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
875                               !!(flags & BDRV_O_CACHE_WB), &error_abort);
876         }
877         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
878             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
879                               !!(flags & BDRV_O_NOCACHE), &error_abort);
880         }
881         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
882             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
883                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
884         }
885         qemu_opt_unset(all_opts, "cache");
886     }
887 
888     /* Get a QDict for processing the options */
889     bs_opts = qdict_new();
890     qemu_opts_to_qdict(all_opts, bs_opts);
891 
892     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
893                                    &error_abort);
894     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
895     if (local_err) {
896         error_report_err(local_err);
897         goto fail;
898     }
899 
900     /* Deprecated option boot=[on|off] */
901     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
902         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
903                 "ignored. Future versions will reject this parameter. Please "
904                 "update your scripts.\n");
905     }
906 
907     /* Media type */
908     value = qemu_opt_get(legacy_opts, "media");
909     if (value) {
910         if (!strcmp(value, "disk")) {
911             media = MEDIA_DISK;
912         } else if (!strcmp(value, "cdrom")) {
913             media = MEDIA_CDROM;
914             read_only = true;
915         } else {
916             error_report("'%s' invalid media", value);
917             goto fail;
918         }
919     }
920 
921     /* copy-on-read is disabled with a warning for read-only devices */
922     read_only |= qemu_opt_get_bool(legacy_opts, "read-only", false);
923     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
924 
925     if (read_only && copy_on_read) {
926         error_report("warning: disabling copy-on-read on read-only drive");
927         copy_on_read = false;
928     }
929 
930     qdict_put(bs_opts, "read-only",
931               qstring_from_str(read_only ? "on" : "off"));
932     qdict_put(bs_opts, "copy-on-read",
933               qstring_from_str(copy_on_read ? "on" :"off"));
934 
935     /* Controller type */
936     value = qemu_opt_get(legacy_opts, "if");
937     if (value) {
938         for (type = 0;
939              type < IF_COUNT && strcmp(value, if_name[type]);
940              type++) {
941         }
942         if (type == IF_COUNT) {
943             error_report("unsupported bus type '%s'", value);
944             goto fail;
945         }
946     } else {
947         type = block_default_type;
948     }
949 
950     /* Geometry */
951     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
952     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
953     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
954 
955     if (cyls || heads || secs) {
956         if (cyls < 1) {
957             error_report("invalid physical cyls number");
958             goto fail;
959         }
960         if (heads < 1) {
961             error_report("invalid physical heads number");
962             goto fail;
963         }
964         if (secs < 1) {
965             error_report("invalid physical secs number");
966             goto fail;
967         }
968     }
969 
970     translation = BIOS_ATA_TRANSLATION_AUTO;
971     value = qemu_opt_get(legacy_opts, "trans");
972     if (value != NULL) {
973         if (!cyls) {
974             error_report("'%s' trans must be used with cyls, heads and secs",
975                          value);
976             goto fail;
977         }
978         if (!strcmp(value, "none")) {
979             translation = BIOS_ATA_TRANSLATION_NONE;
980         } else if (!strcmp(value, "lba")) {
981             translation = BIOS_ATA_TRANSLATION_LBA;
982         } else if (!strcmp(value, "large")) {
983             translation = BIOS_ATA_TRANSLATION_LARGE;
984         } else if (!strcmp(value, "rechs")) {
985             translation = BIOS_ATA_TRANSLATION_RECHS;
986         } else if (!strcmp(value, "auto")) {
987             translation = BIOS_ATA_TRANSLATION_AUTO;
988         } else {
989             error_report("'%s' invalid translation type", value);
990             goto fail;
991         }
992     }
993 
994     if (media == MEDIA_CDROM) {
995         if (cyls || secs || heads) {
996             error_report("CHS can't be set with media=cdrom");
997             goto fail;
998         }
999     }
1000 
1001     /* Device address specified by bus/unit or index.
1002      * If none was specified, try to find the first free one. */
1003     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
1004     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
1005     index   = qemu_opt_get_number(legacy_opts, "index", -1);
1006 
1007     max_devs = if_max_devs[type];
1008 
1009     if (index != -1) {
1010         if (bus_id != 0 || unit_id != -1) {
1011             error_report("index cannot be used with bus and unit");
1012             goto fail;
1013         }
1014         bus_id = drive_index_to_bus_id(type, index);
1015         unit_id = drive_index_to_unit_id(type, index);
1016     }
1017 
1018     if (unit_id == -1) {
1019        unit_id = 0;
1020        while (drive_get(type, bus_id, unit_id) != NULL) {
1021            unit_id++;
1022            if (max_devs && unit_id >= max_devs) {
1023                unit_id -= max_devs;
1024                bus_id++;
1025            }
1026        }
1027     }
1028 
1029     if (max_devs && unit_id >= max_devs) {
1030         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1031         goto fail;
1032     }
1033 
1034     if (drive_get(type, bus_id, unit_id) != NULL) {
1035         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1036                      bus_id, unit_id, index);
1037         goto fail;
1038     }
1039 
1040     /* Serial number */
1041     serial = qemu_opt_get(legacy_opts, "serial");
1042 
1043     /* no id supplied -> create one */
1044     if (qemu_opts_id(all_opts) == NULL) {
1045         char *new_id;
1046         const char *mediastr = "";
1047         if (type == IF_IDE || type == IF_SCSI) {
1048             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1049         }
1050         if (max_devs) {
1051             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1052                                      mediastr, unit_id);
1053         } else {
1054             new_id = g_strdup_printf("%s%s%i", if_name[type],
1055                                      mediastr, unit_id);
1056         }
1057         qdict_put(bs_opts, "id", qstring_from_str(new_id));
1058         g_free(new_id);
1059     }
1060 
1061     /* Add virtio block device */
1062     devaddr = qemu_opt_get(legacy_opts, "addr");
1063     if (devaddr && type != IF_VIRTIO) {
1064         error_report("addr is not supported by this bus type");
1065         goto fail;
1066     }
1067 
1068     if (type == IF_VIRTIO) {
1069         QemuOpts *devopts;
1070         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1071                                    &error_abort);
1072         if (arch_type == QEMU_ARCH_S390X) {
1073             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1074         } else {
1075             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1076         }
1077         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1078                      &error_abort);
1079         if (devaddr) {
1080             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1081         }
1082     }
1083 
1084     filename = qemu_opt_get(legacy_opts, "file");
1085 
1086     /* Check werror/rerror compatibility with if=... */
1087     werror = qemu_opt_get(legacy_opts, "werror");
1088     if (werror != NULL) {
1089         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1090             type != IF_NONE) {
1091             error_report("werror is not supported by this bus type");
1092             goto fail;
1093         }
1094         qdict_put(bs_opts, "werror", qstring_from_str(werror));
1095     }
1096 
1097     rerror = qemu_opt_get(legacy_opts, "rerror");
1098     if (rerror != NULL) {
1099         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1100             type != IF_NONE) {
1101             error_report("rerror is not supported by this bus type");
1102             goto fail;
1103         }
1104         qdict_put(bs_opts, "rerror", qstring_from_str(rerror));
1105     }
1106 
1107     /* Actual block device init: Functionality shared with blockdev-add */
1108     blk = blockdev_init(filename, bs_opts, &local_err);
1109     bs_opts = NULL;
1110     if (!blk) {
1111         if (local_err) {
1112             error_report_err(local_err);
1113         }
1114         goto fail;
1115     } else {
1116         assert(!local_err);
1117     }
1118 
1119     /* Create legacy DriveInfo */
1120     dinfo = g_malloc0(sizeof(*dinfo));
1121     dinfo->opts = all_opts;
1122 
1123     dinfo->cyls = cyls;
1124     dinfo->heads = heads;
1125     dinfo->secs = secs;
1126     dinfo->trans = translation;
1127 
1128     dinfo->type = type;
1129     dinfo->bus = bus_id;
1130     dinfo->unit = unit_id;
1131     dinfo->devaddr = devaddr;
1132     dinfo->serial = g_strdup(serial);
1133 
1134     blk_set_legacy_dinfo(blk, dinfo);
1135 
1136     switch(type) {
1137     case IF_IDE:
1138     case IF_SCSI:
1139     case IF_XEN:
1140     case IF_NONE:
1141         dinfo->media_cd = media == MEDIA_CDROM;
1142         break;
1143     default:
1144         break;
1145     }
1146 
1147 fail:
1148     qemu_opts_del(legacy_opts);
1149     QDECREF(bs_opts);
1150     return dinfo;
1151 }
1152 
1153 void hmp_commit(Monitor *mon, const QDict *qdict)
1154 {
1155     const char *device = qdict_get_str(qdict, "device");
1156     BlockBackend *blk;
1157     int ret;
1158 
1159     if (!strcmp(device, "all")) {
1160         ret = bdrv_commit_all();
1161     } else {
1162         BlockDriverState *bs;
1163         AioContext *aio_context;
1164 
1165         blk = blk_by_name(device);
1166         if (!blk) {
1167             monitor_printf(mon, "Device '%s' not found\n", device);
1168             return;
1169         }
1170         if (!blk_is_available(blk)) {
1171             monitor_printf(mon, "Device '%s' has no medium\n", device);
1172             return;
1173         }
1174 
1175         bs = blk_bs(blk);
1176         aio_context = bdrv_get_aio_context(bs);
1177         aio_context_acquire(aio_context);
1178 
1179         ret = bdrv_commit(bs);
1180 
1181         aio_context_release(aio_context);
1182     }
1183     if (ret < 0) {
1184         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1185                        strerror(-ret));
1186     }
1187 }
1188 
1189 static void blockdev_do_action(TransactionActionKind type, void *data,
1190                                Error **errp)
1191 {
1192     TransactionAction action;
1193     TransactionActionList list;
1194 
1195     action.type = type;
1196     action.u.data = data;
1197     list.value = &action;
1198     list.next = NULL;
1199     qmp_transaction(&list, false, NULL, errp);
1200 }
1201 
1202 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1203                                 bool has_node_name, const char *node_name,
1204                                 const char *snapshot_file,
1205                                 bool has_snapshot_node_name,
1206                                 const char *snapshot_node_name,
1207                                 bool has_format, const char *format,
1208                                 bool has_mode, NewImageMode mode, Error **errp)
1209 {
1210     BlockdevSnapshotSync snapshot = {
1211         .has_device = has_device,
1212         .device = (char *) device,
1213         .has_node_name = has_node_name,
1214         .node_name = (char *) node_name,
1215         .snapshot_file = (char *) snapshot_file,
1216         .has_snapshot_node_name = has_snapshot_node_name,
1217         .snapshot_node_name = (char *) snapshot_node_name,
1218         .has_format = has_format,
1219         .format = (char *) format,
1220         .has_mode = has_mode,
1221         .mode = mode,
1222     };
1223     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1224                        &snapshot, errp);
1225 }
1226 
1227 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1228                            Error **errp)
1229 {
1230     BlockdevSnapshot snapshot_data = {
1231         .node = (char *) node,
1232         .overlay = (char *) overlay
1233     };
1234 
1235     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1236                        &snapshot_data, errp);
1237 }
1238 
1239 void qmp_blockdev_snapshot_internal_sync(const char *device,
1240                                          const char *name,
1241                                          Error **errp)
1242 {
1243     BlockdevSnapshotInternal snapshot = {
1244         .device = (char *) device,
1245         .name = (char *) name
1246     };
1247 
1248     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1249                        &snapshot, errp);
1250 }
1251 
1252 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1253                                                          bool has_id,
1254                                                          const char *id,
1255                                                          bool has_name,
1256                                                          const char *name,
1257                                                          Error **errp)
1258 {
1259     BlockDriverState *bs;
1260     BlockBackend *blk;
1261     AioContext *aio_context;
1262     QEMUSnapshotInfo sn;
1263     Error *local_err = NULL;
1264     SnapshotInfo *info = NULL;
1265     int ret;
1266 
1267     blk = blk_by_name(device);
1268     if (!blk) {
1269         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1270                   "Device '%s' not found", device);
1271         return NULL;
1272     }
1273 
1274     aio_context = blk_get_aio_context(blk);
1275     aio_context_acquire(aio_context);
1276 
1277     if (!has_id) {
1278         id = NULL;
1279     }
1280 
1281     if (!has_name) {
1282         name = NULL;
1283     }
1284 
1285     if (!id && !name) {
1286         error_setg(errp, "Name or id must be provided");
1287         goto out_aio_context;
1288     }
1289 
1290     if (!blk_is_available(blk)) {
1291         error_setg(errp, "Device '%s' has no medium", device);
1292         goto out_aio_context;
1293     }
1294     bs = blk_bs(blk);
1295 
1296     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1297         goto out_aio_context;
1298     }
1299 
1300     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1301     if (local_err) {
1302         error_propagate(errp, local_err);
1303         goto out_aio_context;
1304     }
1305     if (!ret) {
1306         error_setg(errp,
1307                    "Snapshot with id '%s' and name '%s' does not exist on "
1308                    "device '%s'",
1309                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1310         goto out_aio_context;
1311     }
1312 
1313     bdrv_snapshot_delete(bs, id, name, &local_err);
1314     if (local_err) {
1315         error_propagate(errp, local_err);
1316         goto out_aio_context;
1317     }
1318 
1319     aio_context_release(aio_context);
1320 
1321     info = g_new0(SnapshotInfo, 1);
1322     info->id = g_strdup(sn.id_str);
1323     info->name = g_strdup(sn.name);
1324     info->date_nsec = sn.date_nsec;
1325     info->date_sec = sn.date_sec;
1326     info->vm_state_size = sn.vm_state_size;
1327     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1328     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1329 
1330     return info;
1331 
1332 out_aio_context:
1333     aio_context_release(aio_context);
1334     return NULL;
1335 }
1336 
1337 /**
1338  * block_dirty_bitmap_lookup:
1339  * Return a dirty bitmap (if present), after validating
1340  * the node reference and bitmap names.
1341  *
1342  * @node: The name of the BDS node to search for bitmaps
1343  * @name: The name of the bitmap to search for
1344  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1345  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1346  * @errp: Output pointer for error information. Can be NULL.
1347  *
1348  * @return: A bitmap object on success, or NULL on failure.
1349  */
1350 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1351                                                   const char *name,
1352                                                   BlockDriverState **pbs,
1353                                                   AioContext **paio,
1354                                                   Error **errp)
1355 {
1356     BlockDriverState *bs;
1357     BdrvDirtyBitmap *bitmap;
1358     AioContext *aio_context;
1359 
1360     if (!node) {
1361         error_setg(errp, "Node cannot be NULL");
1362         return NULL;
1363     }
1364     if (!name) {
1365         error_setg(errp, "Bitmap name cannot be NULL");
1366         return NULL;
1367     }
1368     bs = bdrv_lookup_bs(node, node, NULL);
1369     if (!bs) {
1370         error_setg(errp, "Node '%s' not found", node);
1371         return NULL;
1372     }
1373 
1374     aio_context = bdrv_get_aio_context(bs);
1375     aio_context_acquire(aio_context);
1376 
1377     bitmap = bdrv_find_dirty_bitmap(bs, name);
1378     if (!bitmap) {
1379         error_setg(errp, "Dirty bitmap '%s' not found", name);
1380         goto fail;
1381     }
1382 
1383     if (pbs) {
1384         *pbs = bs;
1385     }
1386     if (paio) {
1387         *paio = aio_context;
1388     } else {
1389         aio_context_release(aio_context);
1390     }
1391 
1392     return bitmap;
1393 
1394  fail:
1395     aio_context_release(aio_context);
1396     return NULL;
1397 }
1398 
1399 /* New and old BlockDriverState structs for atomic group operations */
1400 
1401 typedef struct BlkActionState BlkActionState;
1402 
1403 /**
1404  * BlkActionOps:
1405  * Table of operations that define an Action.
1406  *
1407  * @instance_size: Size of state struct, in bytes.
1408  * @prepare: Prepare the work, must NOT be NULL.
1409  * @commit: Commit the changes, can be NULL.
1410  * @abort: Abort the changes on fail, can be NULL.
1411  * @clean: Clean up resources after all transaction actions have called
1412  *         commit() or abort(). Can be NULL.
1413  *
1414  * Only prepare() may fail. In a single transaction, only one of commit() or
1415  * abort() will be called. clean() will always be called if it is present.
1416  */
1417 typedef struct BlkActionOps {
1418     size_t instance_size;
1419     void (*prepare)(BlkActionState *common, Error **errp);
1420     void (*commit)(BlkActionState *common);
1421     void (*abort)(BlkActionState *common);
1422     void (*clean)(BlkActionState *common);
1423 } BlkActionOps;
1424 
1425 /**
1426  * BlkActionState:
1427  * Describes one Action's state within a Transaction.
1428  *
1429  * @action: QAPI-defined enum identifying which Action to perform.
1430  * @ops: Table of ActionOps this Action can perform.
1431  * @block_job_txn: Transaction which this action belongs to.
1432  * @entry: List membership for all Actions in this Transaction.
1433  *
1434  * This structure must be arranged as first member in a subclassed type,
1435  * assuming that the compiler will also arrange it to the same offsets as the
1436  * base class.
1437  */
1438 struct BlkActionState {
1439     TransactionAction *action;
1440     const BlkActionOps *ops;
1441     BlockJobTxn *block_job_txn;
1442     TransactionProperties *txn_props;
1443     QSIMPLEQ_ENTRY(BlkActionState) entry;
1444 };
1445 
1446 /* internal snapshot private data */
1447 typedef struct InternalSnapshotState {
1448     BlkActionState common;
1449     BlockDriverState *bs;
1450     AioContext *aio_context;
1451     QEMUSnapshotInfo sn;
1452     bool created;
1453 } InternalSnapshotState;
1454 
1455 
1456 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1457 {
1458     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1459         error_setg(errp,
1460                    "Action '%s' does not support Transaction property "
1461                    "completion-mode = %s",
1462                    TransactionActionKind_lookup[s->action->type],
1463                    ActionCompletionMode_lookup[s->txn_props->completion_mode]);
1464         return -1;
1465     }
1466     return 0;
1467 }
1468 
1469 static void internal_snapshot_prepare(BlkActionState *common,
1470                                       Error **errp)
1471 {
1472     Error *local_err = NULL;
1473     const char *device;
1474     const char *name;
1475     BlockBackend *blk;
1476     BlockDriverState *bs;
1477     QEMUSnapshotInfo old_sn, *sn;
1478     bool ret;
1479     qemu_timeval tv;
1480     BlockdevSnapshotInternal *internal;
1481     InternalSnapshotState *state;
1482     int ret1;
1483 
1484     g_assert(common->action->type ==
1485              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1486     internal = common->action->u.blockdev_snapshot_internal_sync;
1487     state = DO_UPCAST(InternalSnapshotState, common, common);
1488 
1489     /* 1. parse input */
1490     device = internal->device;
1491     name = internal->name;
1492 
1493     /* 2. check for validation */
1494     if (action_check_completion_mode(common, errp) < 0) {
1495         return;
1496     }
1497 
1498     blk = blk_by_name(device);
1499     if (!blk) {
1500         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1501                   "Device '%s' not found", device);
1502         return;
1503     }
1504 
1505     /* AioContext is released in .clean() */
1506     state->aio_context = blk_get_aio_context(blk);
1507     aio_context_acquire(state->aio_context);
1508 
1509     if (!blk_is_available(blk)) {
1510         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1511         return;
1512     }
1513     bs = blk_bs(blk);
1514 
1515     state->bs = bs;
1516     bdrv_drained_begin(bs);
1517 
1518     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1519         return;
1520     }
1521 
1522     if (bdrv_is_read_only(bs)) {
1523         error_setg(errp, "Device '%s' is read only", device);
1524         return;
1525     }
1526 
1527     if (!bdrv_can_snapshot(bs)) {
1528         error_setg(errp, "Block format '%s' used by device '%s' "
1529                    "does not support internal snapshots",
1530                    bs->drv->format_name, device);
1531         return;
1532     }
1533 
1534     if (!strlen(name)) {
1535         error_setg(errp, "Name is empty");
1536         return;
1537     }
1538 
1539     /* check whether a snapshot with name exist */
1540     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1541                                             &local_err);
1542     if (local_err) {
1543         error_propagate(errp, local_err);
1544         return;
1545     } else if (ret) {
1546         error_setg(errp,
1547                    "Snapshot with name '%s' already exists on device '%s'",
1548                    name, device);
1549         return;
1550     }
1551 
1552     /* 3. take the snapshot */
1553     sn = &state->sn;
1554     pstrcpy(sn->name, sizeof(sn->name), name);
1555     qemu_gettimeofday(&tv);
1556     sn->date_sec = tv.tv_sec;
1557     sn->date_nsec = tv.tv_usec * 1000;
1558     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1559 
1560     ret1 = bdrv_snapshot_create(bs, sn);
1561     if (ret1 < 0) {
1562         error_setg_errno(errp, -ret1,
1563                          "Failed to create snapshot '%s' on device '%s'",
1564                          name, device);
1565         return;
1566     }
1567 
1568     /* 4. succeed, mark a snapshot is created */
1569     state->created = true;
1570 }
1571 
1572 static void internal_snapshot_abort(BlkActionState *common)
1573 {
1574     InternalSnapshotState *state =
1575                              DO_UPCAST(InternalSnapshotState, common, common);
1576     BlockDriverState *bs = state->bs;
1577     QEMUSnapshotInfo *sn = &state->sn;
1578     Error *local_error = NULL;
1579 
1580     if (!state->created) {
1581         return;
1582     }
1583 
1584     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1585         error_reportf_err(local_error,
1586                           "Failed to delete snapshot with id '%s' and "
1587                           "name '%s' on device '%s' in abort: ",
1588                           sn->id_str, sn->name,
1589                           bdrv_get_device_name(bs));
1590     }
1591 }
1592 
1593 static void internal_snapshot_clean(BlkActionState *common)
1594 {
1595     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1596                                              common, common);
1597 
1598     if (state->aio_context) {
1599         if (state->bs) {
1600             bdrv_drained_end(state->bs);
1601         }
1602         aio_context_release(state->aio_context);
1603     }
1604 }
1605 
1606 /* external snapshot private data */
1607 typedef struct ExternalSnapshotState {
1608     BlkActionState common;
1609     BlockDriverState *old_bs;
1610     BlockDriverState *new_bs;
1611     AioContext *aio_context;
1612 } ExternalSnapshotState;
1613 
1614 static void external_snapshot_prepare(BlkActionState *common,
1615                                       Error **errp)
1616 {
1617     int flags = 0, ret;
1618     QDict *options = NULL;
1619     Error *local_err = NULL;
1620     /* Device and node name of the image to generate the snapshot from */
1621     const char *device;
1622     const char *node_name;
1623     /* Reference to the new image (for 'blockdev-snapshot') */
1624     const char *snapshot_ref;
1625     /* File name of the new image (for 'blockdev-snapshot-sync') */
1626     const char *new_image_file;
1627     ExternalSnapshotState *state =
1628                              DO_UPCAST(ExternalSnapshotState, common, common);
1629     TransactionAction *action = common->action;
1630 
1631     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1632      * purpose but a different set of parameters */
1633     switch (action->type) {
1634     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1635         {
1636             BlockdevSnapshot *s = action->u.blockdev_snapshot;
1637             device = s->node;
1638             node_name = s->node;
1639             new_image_file = NULL;
1640             snapshot_ref = s->overlay;
1641         }
1642         break;
1643     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1644         {
1645             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync;
1646             device = s->has_device ? s->device : NULL;
1647             node_name = s->has_node_name ? s->node_name : NULL;
1648             new_image_file = s->snapshot_file;
1649             snapshot_ref = NULL;
1650         }
1651         break;
1652     default:
1653         g_assert_not_reached();
1654     }
1655 
1656     /* start processing */
1657     if (action_check_completion_mode(common, errp) < 0) {
1658         return;
1659     }
1660 
1661     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1662     if (!state->old_bs) {
1663         return;
1664     }
1665 
1666     /* Acquire AioContext now so any threads operating on old_bs stop */
1667     state->aio_context = bdrv_get_aio_context(state->old_bs);
1668     aio_context_acquire(state->aio_context);
1669     bdrv_drained_begin(state->old_bs);
1670 
1671     if (!bdrv_is_inserted(state->old_bs)) {
1672         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1673         return;
1674     }
1675 
1676     if (bdrv_op_is_blocked(state->old_bs,
1677                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1678         return;
1679     }
1680 
1681     if (!bdrv_is_read_only(state->old_bs)) {
1682         if (bdrv_flush(state->old_bs)) {
1683             error_setg(errp, QERR_IO_ERROR);
1684             return;
1685         }
1686     }
1687 
1688     if (!bdrv_is_first_non_filter(state->old_bs)) {
1689         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1690         return;
1691     }
1692 
1693     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1694         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync;
1695         const char *format = s->has_format ? s->format : "qcow2";
1696         enum NewImageMode mode;
1697         const char *snapshot_node_name =
1698             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1699 
1700         if (node_name && !snapshot_node_name) {
1701             error_setg(errp, "New snapshot node name missing");
1702             return;
1703         }
1704 
1705         if (snapshot_node_name &&
1706             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1707             error_setg(errp, "New snapshot node name already in use");
1708             return;
1709         }
1710 
1711         flags = state->old_bs->open_flags;
1712 
1713         /* create new image w/backing file */
1714         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1715         if (mode != NEW_IMAGE_MODE_EXISTING) {
1716             bdrv_img_create(new_image_file, format,
1717                             state->old_bs->filename,
1718                             state->old_bs->drv->format_name,
1719                             NULL, -1, flags, &local_err, false);
1720             if (local_err) {
1721                 error_propagate(errp, local_err);
1722                 return;
1723             }
1724         }
1725 
1726         options = qdict_new();
1727         if (s->has_snapshot_node_name) {
1728             qdict_put(options, "node-name",
1729                       qstring_from_str(snapshot_node_name));
1730         }
1731         qdict_put(options, "driver", qstring_from_str(format));
1732 
1733         flags |= BDRV_O_NO_BACKING;
1734     }
1735 
1736     assert(state->new_bs == NULL);
1737     ret = bdrv_open(&state->new_bs, new_image_file, snapshot_ref, options,
1738                     flags, errp);
1739     /* We will manually add the backing_hd field to the bs later */
1740     if (ret != 0) {
1741         return;
1742     }
1743 
1744     if (state->new_bs->blk != NULL) {
1745         error_setg(errp, "The snapshot is already in use by %s",
1746                    blk_name(state->new_bs->blk));
1747         return;
1748     }
1749 
1750     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1751                            errp)) {
1752         return;
1753     }
1754 
1755     if (state->new_bs->backing != NULL) {
1756         error_setg(errp, "The snapshot already has a backing image");
1757         return;
1758     }
1759 
1760     if (!state->new_bs->drv->supports_backing) {
1761         error_setg(errp, "The snapshot does not support backing images");
1762     }
1763 }
1764 
1765 static void external_snapshot_commit(BlkActionState *common)
1766 {
1767     ExternalSnapshotState *state =
1768                              DO_UPCAST(ExternalSnapshotState, common, common);
1769 
1770     bdrv_set_aio_context(state->new_bs, state->aio_context);
1771 
1772     /* This removes our old bs and adds the new bs */
1773     bdrv_append(state->new_bs, state->old_bs);
1774     /* We don't need (or want) to use the transactional
1775      * bdrv_reopen_multiple() across all the entries at once, because we
1776      * don't want to abort all of them if one of them fails the reopen */
1777     bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1778                 NULL);
1779 }
1780 
1781 static void external_snapshot_abort(BlkActionState *common)
1782 {
1783     ExternalSnapshotState *state =
1784                              DO_UPCAST(ExternalSnapshotState, common, common);
1785     if (state->new_bs) {
1786         bdrv_unref(state->new_bs);
1787     }
1788 }
1789 
1790 static void external_snapshot_clean(BlkActionState *common)
1791 {
1792     ExternalSnapshotState *state =
1793                              DO_UPCAST(ExternalSnapshotState, common, common);
1794     if (state->aio_context) {
1795         bdrv_drained_end(state->old_bs);
1796         aio_context_release(state->aio_context);
1797     }
1798 }
1799 
1800 typedef struct DriveBackupState {
1801     BlkActionState common;
1802     BlockDriverState *bs;
1803     AioContext *aio_context;
1804     BlockJob *job;
1805 } DriveBackupState;
1806 
1807 static void do_drive_backup(const char *device, const char *target,
1808                             bool has_format, const char *format,
1809                             enum MirrorSyncMode sync,
1810                             bool has_mode, enum NewImageMode mode,
1811                             bool has_speed, int64_t speed,
1812                             bool has_bitmap, const char *bitmap,
1813                             bool has_on_source_error,
1814                             BlockdevOnError on_source_error,
1815                             bool has_on_target_error,
1816                             BlockdevOnError on_target_error,
1817                             BlockJobTxn *txn, Error **errp);
1818 
1819 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1820 {
1821     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1822     BlockBackend *blk;
1823     DriveBackup *backup;
1824     Error *local_err = NULL;
1825 
1826     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1827     backup = common->action->u.drive_backup;
1828 
1829     blk = blk_by_name(backup->device);
1830     if (!blk) {
1831         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1832                   "Device '%s' not found", backup->device);
1833         return;
1834     }
1835 
1836     if (!blk_is_available(blk)) {
1837         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, backup->device);
1838         return;
1839     }
1840 
1841     /* AioContext is released in .clean() */
1842     state->aio_context = blk_get_aio_context(blk);
1843     aio_context_acquire(state->aio_context);
1844     bdrv_drained_begin(blk_bs(blk));
1845     state->bs = blk_bs(blk);
1846 
1847     do_drive_backup(backup->device, backup->target,
1848                     backup->has_format, backup->format,
1849                     backup->sync,
1850                     backup->has_mode, backup->mode,
1851                     backup->has_speed, backup->speed,
1852                     backup->has_bitmap, backup->bitmap,
1853                     backup->has_on_source_error, backup->on_source_error,
1854                     backup->has_on_target_error, backup->on_target_error,
1855                     common->block_job_txn, &local_err);
1856     if (local_err) {
1857         error_propagate(errp, local_err);
1858         return;
1859     }
1860 
1861     state->job = state->bs->job;
1862 }
1863 
1864 static void drive_backup_abort(BlkActionState *common)
1865 {
1866     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1867     BlockDriverState *bs = state->bs;
1868 
1869     /* Only cancel if it's the job we started */
1870     if (bs && bs->job && bs->job == state->job) {
1871         block_job_cancel_sync(bs->job);
1872     }
1873 }
1874 
1875 static void drive_backup_clean(BlkActionState *common)
1876 {
1877     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1878 
1879     if (state->aio_context) {
1880         bdrv_drained_end(state->bs);
1881         aio_context_release(state->aio_context);
1882     }
1883 }
1884 
1885 typedef struct BlockdevBackupState {
1886     BlkActionState common;
1887     BlockDriverState *bs;
1888     BlockJob *job;
1889     AioContext *aio_context;
1890 } BlockdevBackupState;
1891 
1892 static void do_blockdev_backup(const char *device, const char *target,
1893                                enum MirrorSyncMode sync,
1894                                bool has_speed, int64_t speed,
1895                                bool has_on_source_error,
1896                                BlockdevOnError on_source_error,
1897                                bool has_on_target_error,
1898                                BlockdevOnError on_target_error,
1899                                BlockJobTxn *txn, Error **errp);
1900 
1901 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1902 {
1903     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1904     BlockdevBackup *backup;
1905     BlockBackend *blk, *target;
1906     Error *local_err = NULL;
1907 
1908     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1909     backup = common->action->u.blockdev_backup;
1910 
1911     blk = blk_by_name(backup->device);
1912     if (!blk) {
1913         error_setg(errp, "Device '%s' not found", backup->device);
1914         return;
1915     }
1916 
1917     if (!blk_is_available(blk)) {
1918         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, backup->device);
1919         return;
1920     }
1921 
1922     target = blk_by_name(backup->target);
1923     if (!target) {
1924         error_setg(errp, "Device '%s' not found", backup->target);
1925         return;
1926     }
1927 
1928     /* AioContext is released in .clean() */
1929     state->aio_context = blk_get_aio_context(blk);
1930     if (state->aio_context != blk_get_aio_context(target)) {
1931         state->aio_context = NULL;
1932         error_setg(errp, "Backup between two IO threads is not implemented");
1933         return;
1934     }
1935     aio_context_acquire(state->aio_context);
1936     state->bs = blk_bs(blk);
1937     bdrv_drained_begin(state->bs);
1938 
1939     do_blockdev_backup(backup->device, backup->target,
1940                        backup->sync,
1941                        backup->has_speed, backup->speed,
1942                        backup->has_on_source_error, backup->on_source_error,
1943                        backup->has_on_target_error, backup->on_target_error,
1944                        common->block_job_txn, &local_err);
1945     if (local_err) {
1946         error_propagate(errp, local_err);
1947         return;
1948     }
1949 
1950     state->job = state->bs->job;
1951 }
1952 
1953 static void blockdev_backup_abort(BlkActionState *common)
1954 {
1955     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1956     BlockDriverState *bs = state->bs;
1957 
1958     /* Only cancel if it's the job we started */
1959     if (bs && bs->job && bs->job == state->job) {
1960         block_job_cancel_sync(bs->job);
1961     }
1962 }
1963 
1964 static void blockdev_backup_clean(BlkActionState *common)
1965 {
1966     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1967 
1968     if (state->aio_context) {
1969         bdrv_drained_end(state->bs);
1970         aio_context_release(state->aio_context);
1971     }
1972 }
1973 
1974 typedef struct BlockDirtyBitmapState {
1975     BlkActionState common;
1976     BdrvDirtyBitmap *bitmap;
1977     BlockDriverState *bs;
1978     AioContext *aio_context;
1979     HBitmap *backup;
1980     bool prepared;
1981 } BlockDirtyBitmapState;
1982 
1983 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1984                                            Error **errp)
1985 {
1986     Error *local_err = NULL;
1987     BlockDirtyBitmapAdd *action;
1988     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1989                                              common, common);
1990 
1991     if (action_check_completion_mode(common, errp) < 0) {
1992         return;
1993     }
1994 
1995     action = common->action->u.block_dirty_bitmap_add;
1996     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1997     qmp_block_dirty_bitmap_add(action->node, action->name,
1998                                action->has_granularity, action->granularity,
1999                                &local_err);
2000 
2001     if (!local_err) {
2002         state->prepared = true;
2003     } else {
2004         error_propagate(errp, local_err);
2005     }
2006 }
2007 
2008 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2009 {
2010     BlockDirtyBitmapAdd *action;
2011     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2012                                              common, common);
2013 
2014     action = common->action->u.block_dirty_bitmap_add;
2015     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2016      * then the node reference and bitmap name must have been valid.
2017      */
2018     if (state->prepared) {
2019         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2020     }
2021 }
2022 
2023 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2024                                              Error **errp)
2025 {
2026     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2027                                              common, common);
2028     BlockDirtyBitmap *action;
2029 
2030     if (action_check_completion_mode(common, errp) < 0) {
2031         return;
2032     }
2033 
2034     action = common->action->u.block_dirty_bitmap_clear;
2035     state->bitmap = block_dirty_bitmap_lookup(action->node,
2036                                               action->name,
2037                                               &state->bs,
2038                                               &state->aio_context,
2039                                               errp);
2040     if (!state->bitmap) {
2041         return;
2042     }
2043 
2044     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2045         error_setg(errp, "Cannot modify a frozen bitmap");
2046         return;
2047     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2048         error_setg(errp, "Cannot clear a disabled bitmap");
2049         return;
2050     }
2051 
2052     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2053     /* AioContext is released in .clean() */
2054 }
2055 
2056 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2057 {
2058     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2059                                              common, common);
2060 
2061     bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2062 }
2063 
2064 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2065 {
2066     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2067                                              common, common);
2068 
2069     hbitmap_free(state->backup);
2070 }
2071 
2072 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2073 {
2074     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2075                                              common, common);
2076 
2077     if (state->aio_context) {
2078         aio_context_release(state->aio_context);
2079     }
2080 }
2081 
2082 static void abort_prepare(BlkActionState *common, Error **errp)
2083 {
2084     error_setg(errp, "Transaction aborted using Abort action");
2085 }
2086 
2087 static void abort_commit(BlkActionState *common)
2088 {
2089     g_assert_not_reached(); /* this action never succeeds */
2090 }
2091 
2092 static const BlkActionOps actions[] = {
2093     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2094         .instance_size = sizeof(ExternalSnapshotState),
2095         .prepare  = external_snapshot_prepare,
2096         .commit   = external_snapshot_commit,
2097         .abort = external_snapshot_abort,
2098         .clean = external_snapshot_clean,
2099     },
2100     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2101         .instance_size = sizeof(ExternalSnapshotState),
2102         .prepare  = external_snapshot_prepare,
2103         .commit   = external_snapshot_commit,
2104         .abort = external_snapshot_abort,
2105         .clean = external_snapshot_clean,
2106     },
2107     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2108         .instance_size = sizeof(DriveBackupState),
2109         .prepare = drive_backup_prepare,
2110         .abort = drive_backup_abort,
2111         .clean = drive_backup_clean,
2112     },
2113     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2114         .instance_size = sizeof(BlockdevBackupState),
2115         .prepare = blockdev_backup_prepare,
2116         .abort = blockdev_backup_abort,
2117         .clean = blockdev_backup_clean,
2118     },
2119     [TRANSACTION_ACTION_KIND_ABORT] = {
2120         .instance_size = sizeof(BlkActionState),
2121         .prepare = abort_prepare,
2122         .commit = abort_commit,
2123     },
2124     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2125         .instance_size = sizeof(InternalSnapshotState),
2126         .prepare  = internal_snapshot_prepare,
2127         .abort = internal_snapshot_abort,
2128         .clean = internal_snapshot_clean,
2129     },
2130     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2131         .instance_size = sizeof(BlockDirtyBitmapState),
2132         .prepare = block_dirty_bitmap_add_prepare,
2133         .abort = block_dirty_bitmap_add_abort,
2134     },
2135     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2136         .instance_size = sizeof(BlockDirtyBitmapState),
2137         .prepare = block_dirty_bitmap_clear_prepare,
2138         .commit = block_dirty_bitmap_clear_commit,
2139         .abort = block_dirty_bitmap_clear_abort,
2140         .clean = block_dirty_bitmap_clear_clean,
2141     }
2142 };
2143 
2144 /**
2145  * Allocate a TransactionProperties structure if necessary, and fill
2146  * that structure with desired defaults if they are unset.
2147  */
2148 static TransactionProperties *get_transaction_properties(
2149     TransactionProperties *props)
2150 {
2151     if (!props) {
2152         props = g_new0(TransactionProperties, 1);
2153     }
2154 
2155     if (!props->has_completion_mode) {
2156         props->has_completion_mode = true;
2157         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2158     }
2159 
2160     return props;
2161 }
2162 
2163 /*
2164  * 'Atomic' group operations.  The operations are performed as a set, and if
2165  * any fail then we roll back all operations in the group.
2166  */
2167 void qmp_transaction(TransactionActionList *dev_list,
2168                      bool has_props,
2169                      struct TransactionProperties *props,
2170                      Error **errp)
2171 {
2172     TransactionActionList *dev_entry = dev_list;
2173     BlockJobTxn *block_job_txn = NULL;
2174     BlkActionState *state, *next;
2175     Error *local_err = NULL;
2176 
2177     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2178     QSIMPLEQ_INIT(&snap_bdrv_states);
2179 
2180     /* Does this transaction get canceled as a group on failure?
2181      * If not, we don't really need to make a BlockJobTxn.
2182      */
2183     props = get_transaction_properties(props);
2184     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2185         block_job_txn = block_job_txn_new();
2186     }
2187 
2188     /* drain all i/o before any operations */
2189     bdrv_drain_all();
2190 
2191     /* We don't do anything in this loop that commits us to the operations */
2192     while (NULL != dev_entry) {
2193         TransactionAction *dev_info = NULL;
2194         const BlkActionOps *ops;
2195 
2196         dev_info = dev_entry->value;
2197         dev_entry = dev_entry->next;
2198 
2199         assert(dev_info->type < ARRAY_SIZE(actions));
2200 
2201         ops = &actions[dev_info->type];
2202         assert(ops->instance_size > 0);
2203 
2204         state = g_malloc0(ops->instance_size);
2205         state->ops = ops;
2206         state->action = dev_info;
2207         state->block_job_txn = block_job_txn;
2208         state->txn_props = props;
2209         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2210 
2211         state->ops->prepare(state, &local_err);
2212         if (local_err) {
2213             error_propagate(errp, local_err);
2214             goto delete_and_fail;
2215         }
2216     }
2217 
2218     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2219         if (state->ops->commit) {
2220             state->ops->commit(state);
2221         }
2222     }
2223 
2224     /* success */
2225     goto exit;
2226 
2227 delete_and_fail:
2228     /* failure, and it is all-or-none; roll back all operations */
2229     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2230         if (state->ops->abort) {
2231             state->ops->abort(state);
2232         }
2233     }
2234 exit:
2235     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2236         if (state->ops->clean) {
2237             state->ops->clean(state);
2238         }
2239         g_free(state);
2240     }
2241     if (!has_props) {
2242         qapi_free_TransactionProperties(props);
2243     }
2244     block_job_txn_unref(block_job_txn);
2245 }
2246 
2247 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
2248 {
2249     Error *local_err = NULL;
2250 
2251     qmp_blockdev_open_tray(device, has_force, force, &local_err);
2252     if (local_err) {
2253         error_propagate(errp, local_err);
2254         return;
2255     }
2256 
2257     qmp_x_blockdev_remove_medium(device, errp);
2258 }
2259 
2260 void qmp_block_passwd(bool has_device, const char *device,
2261                       bool has_node_name, const char *node_name,
2262                       const char *password, Error **errp)
2263 {
2264     Error *local_err = NULL;
2265     BlockDriverState *bs;
2266     AioContext *aio_context;
2267 
2268     bs = bdrv_lookup_bs(has_device ? device : NULL,
2269                         has_node_name ? node_name : NULL,
2270                         &local_err);
2271     if (local_err) {
2272         error_propagate(errp, local_err);
2273         return;
2274     }
2275 
2276     aio_context = bdrv_get_aio_context(bs);
2277     aio_context_acquire(aio_context);
2278 
2279     bdrv_add_key(bs, password, errp);
2280 
2281     aio_context_release(aio_context);
2282 }
2283 
2284 void qmp_blockdev_open_tray(const char *device, bool has_force, bool force,
2285                             Error **errp)
2286 {
2287     BlockBackend *blk;
2288     bool locked;
2289 
2290     if (!has_force) {
2291         force = false;
2292     }
2293 
2294     blk = blk_by_name(device);
2295     if (!blk) {
2296         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2297                   "Device '%s' not found", device);
2298         return;
2299     }
2300 
2301     if (!blk_dev_has_removable_media(blk)) {
2302         error_setg(errp, "Device '%s' is not removable", device);
2303         return;
2304     }
2305 
2306     if (blk_dev_is_tray_open(blk)) {
2307         return;
2308     }
2309 
2310     locked = blk_dev_is_medium_locked(blk);
2311     if (locked) {
2312         blk_dev_eject_request(blk, force);
2313     }
2314 
2315     if (!locked || force) {
2316         blk_dev_change_media_cb(blk, false);
2317     }
2318 }
2319 
2320 void qmp_blockdev_close_tray(const char *device, Error **errp)
2321 {
2322     BlockBackend *blk;
2323 
2324     blk = blk_by_name(device);
2325     if (!blk) {
2326         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2327                   "Device '%s' not found", device);
2328         return;
2329     }
2330 
2331     if (!blk_dev_has_removable_media(blk)) {
2332         error_setg(errp, "Device '%s' is not removable", device);
2333         return;
2334     }
2335 
2336     if (!blk_dev_is_tray_open(blk)) {
2337         return;
2338     }
2339 
2340     blk_dev_change_media_cb(blk, true);
2341 }
2342 
2343 void qmp_x_blockdev_remove_medium(const char *device, Error **errp)
2344 {
2345     BlockBackend *blk;
2346     BlockDriverState *bs;
2347     AioContext *aio_context;
2348     bool has_device;
2349 
2350     blk = blk_by_name(device);
2351     if (!blk) {
2352         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2353                   "Device '%s' not found", device);
2354         return;
2355     }
2356 
2357     /* For BBs without a device, we can exchange the BDS tree at will */
2358     has_device = blk_get_attached_dev(blk);
2359 
2360     if (has_device && !blk_dev_has_removable_media(blk)) {
2361         error_setg(errp, "Device '%s' is not removable", device);
2362         return;
2363     }
2364 
2365     if (has_device && !blk_dev_is_tray_open(blk)) {
2366         error_setg(errp, "Tray of device '%s' is not open", device);
2367         return;
2368     }
2369 
2370     bs = blk_bs(blk);
2371     if (!bs) {
2372         return;
2373     }
2374 
2375     aio_context = bdrv_get_aio_context(bs);
2376     aio_context_acquire(aio_context);
2377 
2378     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2379         goto out;
2380     }
2381 
2382     /* This follows the convention established by bdrv_make_anon() */
2383     if (bs->device_list.tqe_prev) {
2384         QTAILQ_REMOVE(&bdrv_states, bs, device_list);
2385         bs->device_list.tqe_prev = NULL;
2386     }
2387 
2388     blk_remove_bs(blk);
2389 
2390 out:
2391     aio_context_release(aio_context);
2392 }
2393 
2394 static void qmp_blockdev_insert_anon_medium(const char *device,
2395                                             BlockDriverState *bs, Error **errp)
2396 {
2397     BlockBackend *blk;
2398     bool has_device;
2399 
2400     blk = blk_by_name(device);
2401     if (!blk) {
2402         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2403                   "Device '%s' not found", device);
2404         return;
2405     }
2406 
2407     /* For BBs without a device, we can exchange the BDS tree at will */
2408     has_device = blk_get_attached_dev(blk);
2409 
2410     if (has_device && !blk_dev_has_removable_media(blk)) {
2411         error_setg(errp, "Device '%s' is not removable", device);
2412         return;
2413     }
2414 
2415     if (has_device && !blk_dev_is_tray_open(blk)) {
2416         error_setg(errp, "Tray of device '%s' is not open", device);
2417         return;
2418     }
2419 
2420     if (blk_bs(blk)) {
2421         error_setg(errp, "There already is a medium in device '%s'", device);
2422         return;
2423     }
2424 
2425     blk_insert_bs(blk, bs);
2426 
2427     QTAILQ_INSERT_TAIL(&bdrv_states, bs, device_list);
2428 }
2429 
2430 void qmp_x_blockdev_insert_medium(const char *device, const char *node_name,
2431                                   Error **errp)
2432 {
2433     BlockDriverState *bs;
2434 
2435     bs = bdrv_find_node(node_name);
2436     if (!bs) {
2437         error_setg(errp, "Node '%s' not found", node_name);
2438         return;
2439     }
2440 
2441     if (bs->blk) {
2442         error_setg(errp, "Node '%s' is already in use by '%s'", node_name,
2443                    blk_name(bs->blk));
2444         return;
2445     }
2446 
2447     qmp_blockdev_insert_anon_medium(device, bs, errp);
2448 }
2449 
2450 void qmp_blockdev_change_medium(const char *device, const char *filename,
2451                                 bool has_format, const char *format,
2452                                 bool has_read_only,
2453                                 BlockdevChangeReadOnlyMode read_only,
2454                                 Error **errp)
2455 {
2456     BlockBackend *blk;
2457     BlockDriverState *medium_bs = NULL;
2458     int bdrv_flags, ret;
2459     QDict *options = NULL;
2460     Error *err = NULL;
2461 
2462     blk = blk_by_name(device);
2463     if (!blk) {
2464         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2465                   "Device '%s' not found", device);
2466         goto fail;
2467     }
2468 
2469     if (blk_bs(blk)) {
2470         blk_update_root_state(blk);
2471     }
2472 
2473     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2474 
2475     if (!has_read_only) {
2476         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2477     }
2478 
2479     switch (read_only) {
2480     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2481         break;
2482 
2483     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2484         bdrv_flags &= ~BDRV_O_RDWR;
2485         break;
2486 
2487     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2488         bdrv_flags |= BDRV_O_RDWR;
2489         break;
2490 
2491     default:
2492         abort();
2493     }
2494 
2495     if (has_format) {
2496         options = qdict_new();
2497         qdict_put(options, "driver", qstring_from_str(format));
2498     }
2499 
2500     assert(!medium_bs);
2501     ret = bdrv_open(&medium_bs, filename, NULL, options, bdrv_flags, errp);
2502     if (ret < 0) {
2503         goto fail;
2504     }
2505 
2506     blk_apply_root_state(blk, medium_bs);
2507 
2508     bdrv_add_key(medium_bs, NULL, &err);
2509     if (err) {
2510         error_propagate(errp, err);
2511         goto fail;
2512     }
2513 
2514     qmp_blockdev_open_tray(device, false, false, &err);
2515     if (err) {
2516         error_propagate(errp, err);
2517         goto fail;
2518     }
2519 
2520     qmp_x_blockdev_remove_medium(device, &err);
2521     if (err) {
2522         error_propagate(errp, err);
2523         goto fail;
2524     }
2525 
2526     qmp_blockdev_insert_anon_medium(device, medium_bs, &err);
2527     if (err) {
2528         error_propagate(errp, err);
2529         goto fail;
2530     }
2531 
2532     qmp_blockdev_close_tray(device, errp);
2533 
2534 fail:
2535     /* If the medium has been inserted, the device has its own reference, so
2536      * ours must be relinquished; and if it has not been inserted successfully,
2537      * the reference must be relinquished anyway */
2538     bdrv_unref(medium_bs);
2539 }
2540 
2541 /* throttling disk I/O limits */
2542 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
2543                                int64_t bps_wr,
2544                                int64_t iops,
2545                                int64_t iops_rd,
2546                                int64_t iops_wr,
2547                                bool has_bps_max,
2548                                int64_t bps_max,
2549                                bool has_bps_rd_max,
2550                                int64_t bps_rd_max,
2551                                bool has_bps_wr_max,
2552                                int64_t bps_wr_max,
2553                                bool has_iops_max,
2554                                int64_t iops_max,
2555                                bool has_iops_rd_max,
2556                                int64_t iops_rd_max,
2557                                bool has_iops_wr_max,
2558                                int64_t iops_wr_max,
2559                                bool has_iops_size,
2560                                int64_t iops_size,
2561                                bool has_group,
2562                                const char *group, Error **errp)
2563 {
2564     ThrottleConfig cfg;
2565     BlockDriverState *bs;
2566     BlockBackend *blk;
2567     AioContext *aio_context;
2568 
2569     blk = blk_by_name(device);
2570     if (!blk) {
2571         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2572                   "Device '%s' not found", device);
2573         return;
2574     }
2575 
2576     aio_context = blk_get_aio_context(blk);
2577     aio_context_acquire(aio_context);
2578 
2579     bs = blk_bs(blk);
2580     if (!bs) {
2581         error_setg(errp, "Device '%s' has no medium", device);
2582         goto out;
2583     }
2584 
2585     memset(&cfg, 0, sizeof(cfg));
2586     cfg.buckets[THROTTLE_BPS_TOTAL].avg = bps;
2587     cfg.buckets[THROTTLE_BPS_READ].avg  = bps_rd;
2588     cfg.buckets[THROTTLE_BPS_WRITE].avg = bps_wr;
2589 
2590     cfg.buckets[THROTTLE_OPS_TOTAL].avg = iops;
2591     cfg.buckets[THROTTLE_OPS_READ].avg  = iops_rd;
2592     cfg.buckets[THROTTLE_OPS_WRITE].avg = iops_wr;
2593 
2594     if (has_bps_max) {
2595         cfg.buckets[THROTTLE_BPS_TOTAL].max = bps_max;
2596     }
2597     if (has_bps_rd_max) {
2598         cfg.buckets[THROTTLE_BPS_READ].max = bps_rd_max;
2599     }
2600     if (has_bps_wr_max) {
2601         cfg.buckets[THROTTLE_BPS_WRITE].max = bps_wr_max;
2602     }
2603     if (has_iops_max) {
2604         cfg.buckets[THROTTLE_OPS_TOTAL].max = iops_max;
2605     }
2606     if (has_iops_rd_max) {
2607         cfg.buckets[THROTTLE_OPS_READ].max = iops_rd_max;
2608     }
2609     if (has_iops_wr_max) {
2610         cfg.buckets[THROTTLE_OPS_WRITE].max = iops_wr_max;
2611     }
2612 
2613     if (has_iops_size) {
2614         cfg.op_size = iops_size;
2615     }
2616 
2617     if (!check_throttle_config(&cfg, errp)) {
2618         goto out;
2619     }
2620 
2621     if (throttle_enabled(&cfg)) {
2622         /* Enable I/O limits if they're not enabled yet, otherwise
2623          * just update the throttling group. */
2624         if (!bs->throttle_state) {
2625             bdrv_io_limits_enable(bs, has_group ? group : device);
2626         } else if (has_group) {
2627             bdrv_io_limits_update_group(bs, group);
2628         }
2629         /* Set the new throttling configuration */
2630         bdrv_set_io_limits(bs, &cfg);
2631     } else if (bs->throttle_state) {
2632         /* If all throttling settings are set to 0, disable I/O limits */
2633         bdrv_io_limits_disable(bs);
2634     }
2635 
2636 out:
2637     aio_context_release(aio_context);
2638 }
2639 
2640 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2641                                 bool has_granularity, uint32_t granularity,
2642                                 Error **errp)
2643 {
2644     AioContext *aio_context;
2645     BlockDriverState *bs;
2646 
2647     if (!name || name[0] == '\0') {
2648         error_setg(errp, "Bitmap name cannot be empty");
2649         return;
2650     }
2651 
2652     bs = bdrv_lookup_bs(node, node, errp);
2653     if (!bs) {
2654         return;
2655     }
2656 
2657     aio_context = bdrv_get_aio_context(bs);
2658     aio_context_acquire(aio_context);
2659 
2660     if (has_granularity) {
2661         if (granularity < 512 || !is_power_of_2(granularity)) {
2662             error_setg(errp, "Granularity must be power of 2 "
2663                              "and at least 512");
2664             goto out;
2665         }
2666     } else {
2667         /* Default to cluster size, if available: */
2668         granularity = bdrv_get_default_bitmap_granularity(bs);
2669     }
2670 
2671     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2672 
2673  out:
2674     aio_context_release(aio_context);
2675 }
2676 
2677 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2678                                    Error **errp)
2679 {
2680     AioContext *aio_context;
2681     BlockDriverState *bs;
2682     BdrvDirtyBitmap *bitmap;
2683 
2684     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2685     if (!bitmap || !bs) {
2686         return;
2687     }
2688 
2689     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2690         error_setg(errp,
2691                    "Bitmap '%s' is currently frozen and cannot be removed",
2692                    name);
2693         goto out;
2694     }
2695     bdrv_dirty_bitmap_make_anon(bitmap);
2696     bdrv_release_dirty_bitmap(bs, bitmap);
2697 
2698  out:
2699     aio_context_release(aio_context);
2700 }
2701 
2702 /**
2703  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2704  * immediately after a full backup operation.
2705  */
2706 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2707                                   Error **errp)
2708 {
2709     AioContext *aio_context;
2710     BdrvDirtyBitmap *bitmap;
2711     BlockDriverState *bs;
2712 
2713     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2714     if (!bitmap || !bs) {
2715         return;
2716     }
2717 
2718     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2719         error_setg(errp,
2720                    "Bitmap '%s' is currently frozen and cannot be modified",
2721                    name);
2722         goto out;
2723     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2724         error_setg(errp,
2725                    "Bitmap '%s' is currently disabled and cannot be cleared",
2726                    name);
2727         goto out;
2728     }
2729 
2730     bdrv_clear_dirty_bitmap(bitmap, NULL);
2731 
2732  out:
2733     aio_context_release(aio_context);
2734 }
2735 
2736 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2737 {
2738     const char *id = qdict_get_str(qdict, "id");
2739     BlockBackend *blk;
2740     BlockDriverState *bs;
2741     AioContext *aio_context;
2742     Error *local_err = NULL;
2743 
2744     blk = blk_by_name(id);
2745     if (!blk) {
2746         error_report("Device '%s' not found", id);
2747         return;
2748     }
2749 
2750     if (!blk_legacy_dinfo(blk)) {
2751         error_report("Deleting device added with blockdev-add"
2752                      " is not supported");
2753         return;
2754     }
2755 
2756     aio_context = blk_get_aio_context(blk);
2757     aio_context_acquire(aio_context);
2758 
2759     bs = blk_bs(blk);
2760     if (bs) {
2761         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2762             error_report_err(local_err);
2763             aio_context_release(aio_context);
2764             return;
2765         }
2766 
2767         bdrv_close(bs);
2768     }
2769 
2770     /* if we have a device attached to this BlockDriverState
2771      * then we need to make the drive anonymous until the device
2772      * can be removed.  If this is a drive with no device backing
2773      * then we can just get rid of the block driver state right here.
2774      */
2775     if (blk_get_attached_dev(blk)) {
2776         blk_hide_on_behalf_of_hmp_drive_del(blk);
2777         /* Further I/O must not pause the guest */
2778         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2779                          BLOCKDEV_ON_ERROR_REPORT);
2780     } else {
2781         blk_unref(blk);
2782     }
2783 
2784     aio_context_release(aio_context);
2785 }
2786 
2787 void qmp_block_resize(bool has_device, const char *device,
2788                       bool has_node_name, const char *node_name,
2789                       int64_t size, Error **errp)
2790 {
2791     Error *local_err = NULL;
2792     BlockDriverState *bs;
2793     AioContext *aio_context;
2794     int ret;
2795 
2796     bs = bdrv_lookup_bs(has_device ? device : NULL,
2797                         has_node_name ? node_name : NULL,
2798                         &local_err);
2799     if (local_err) {
2800         error_propagate(errp, local_err);
2801         return;
2802     }
2803 
2804     aio_context = bdrv_get_aio_context(bs);
2805     aio_context_acquire(aio_context);
2806 
2807     if (!bdrv_is_first_non_filter(bs)) {
2808         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2809         goto out;
2810     }
2811 
2812     if (size < 0) {
2813         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2814         goto out;
2815     }
2816 
2817     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2818         error_setg(errp, QERR_DEVICE_IN_USE, device);
2819         goto out;
2820     }
2821 
2822     /* complete all in-flight operations before resizing the device */
2823     bdrv_drain_all();
2824 
2825     ret = bdrv_truncate(bs, size);
2826     switch (ret) {
2827     case 0:
2828         break;
2829     case -ENOMEDIUM:
2830         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2831         break;
2832     case -ENOTSUP:
2833         error_setg(errp, QERR_UNSUPPORTED);
2834         break;
2835     case -EACCES:
2836         error_setg(errp, "Device '%s' is read only", device);
2837         break;
2838     case -EBUSY:
2839         error_setg(errp, QERR_DEVICE_IN_USE, device);
2840         break;
2841     default:
2842         error_setg_errno(errp, -ret, "Could not resize");
2843         break;
2844     }
2845 
2846 out:
2847     aio_context_release(aio_context);
2848 }
2849 
2850 static void block_job_cb(void *opaque, int ret)
2851 {
2852     /* Note that this function may be executed from another AioContext besides
2853      * the QEMU main loop.  If you need to access anything that assumes the
2854      * QEMU global mutex, use a BH or introduce a mutex.
2855      */
2856 
2857     BlockDriverState *bs = opaque;
2858     const char *msg = NULL;
2859 
2860     trace_block_job_cb(bs, bs->job, ret);
2861 
2862     assert(bs->job);
2863 
2864     if (ret < 0) {
2865         msg = strerror(-ret);
2866     }
2867 
2868     if (block_job_is_cancelled(bs->job)) {
2869         block_job_event_cancelled(bs->job);
2870     } else {
2871         block_job_event_completed(bs->job, msg);
2872     }
2873 }
2874 
2875 void qmp_block_stream(const char *device,
2876                       bool has_base, const char *base,
2877                       bool has_backing_file, const char *backing_file,
2878                       bool has_speed, int64_t speed,
2879                       bool has_on_error, BlockdevOnError on_error,
2880                       Error **errp)
2881 {
2882     BlockBackend *blk;
2883     BlockDriverState *bs;
2884     BlockDriverState *base_bs = NULL;
2885     AioContext *aio_context;
2886     Error *local_err = NULL;
2887     const char *base_name = NULL;
2888 
2889     if (!has_on_error) {
2890         on_error = BLOCKDEV_ON_ERROR_REPORT;
2891     }
2892 
2893     blk = blk_by_name(device);
2894     if (!blk) {
2895         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2896                   "Device '%s' not found", device);
2897         return;
2898     }
2899 
2900     aio_context = blk_get_aio_context(blk);
2901     aio_context_acquire(aio_context);
2902 
2903     if (!blk_is_available(blk)) {
2904         error_setg(errp, "Device '%s' has no medium", device);
2905         goto out;
2906     }
2907     bs = blk_bs(blk);
2908 
2909     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_STREAM, errp)) {
2910         goto out;
2911     }
2912 
2913     if (has_base) {
2914         base_bs = bdrv_find_backing_image(bs, base);
2915         if (base_bs == NULL) {
2916             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2917             goto out;
2918         }
2919         assert(bdrv_get_aio_context(base_bs) == aio_context);
2920         base_name = base;
2921     }
2922 
2923     /* if we are streaming the entire chain, the result will have no backing
2924      * file, and specifying one is therefore an error */
2925     if (base_bs == NULL && has_backing_file) {
2926         error_setg(errp, "backing file specified, but streaming the "
2927                          "entire chain");
2928         goto out;
2929     }
2930 
2931     /* backing_file string overrides base bs filename */
2932     base_name = has_backing_file ? backing_file : base_name;
2933 
2934     stream_start(bs, base_bs, base_name, has_speed ? speed : 0,
2935                  on_error, block_job_cb, bs, &local_err);
2936     if (local_err) {
2937         error_propagate(errp, local_err);
2938         goto out;
2939     }
2940 
2941     trace_qmp_block_stream(bs, bs->job);
2942 
2943 out:
2944     aio_context_release(aio_context);
2945 }
2946 
2947 void qmp_block_commit(const char *device,
2948                       bool has_base, const char *base,
2949                       bool has_top, const char *top,
2950                       bool has_backing_file, const char *backing_file,
2951                       bool has_speed, int64_t speed,
2952                       Error **errp)
2953 {
2954     BlockBackend *blk;
2955     BlockDriverState *bs;
2956     BlockDriverState *base_bs, *top_bs;
2957     AioContext *aio_context;
2958     Error *local_err = NULL;
2959     /* This will be part of the QMP command, if/when the
2960      * BlockdevOnError change for blkmirror makes it in
2961      */
2962     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
2963 
2964     if (!has_speed) {
2965         speed = 0;
2966     }
2967 
2968     /* Important Note:
2969      *  libvirt relies on the DeviceNotFound error class in order to probe for
2970      *  live commit feature versions; for this to work, we must make sure to
2971      *  perform the device lookup before any generic errors that may occur in a
2972      *  scenario in which all optional arguments are omitted. */
2973     blk = blk_by_name(device);
2974     if (!blk) {
2975         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2976                   "Device '%s' not found", device);
2977         return;
2978     }
2979 
2980     aio_context = blk_get_aio_context(blk);
2981     aio_context_acquire(aio_context);
2982 
2983     if (!blk_is_available(blk)) {
2984         error_setg(errp, "Device '%s' has no medium", device);
2985         goto out;
2986     }
2987     bs = blk_bs(blk);
2988 
2989     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2990         goto out;
2991     }
2992 
2993     /* default top_bs is the active layer */
2994     top_bs = bs;
2995 
2996     if (has_top && top) {
2997         if (strcmp(bs->filename, top) != 0) {
2998             top_bs = bdrv_find_backing_image(bs, top);
2999         }
3000     }
3001 
3002     if (top_bs == NULL) {
3003         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3004         goto out;
3005     }
3006 
3007     assert(bdrv_get_aio_context(top_bs) == aio_context);
3008 
3009     if (has_base && base) {
3010         base_bs = bdrv_find_backing_image(top_bs, base);
3011     } else {
3012         base_bs = bdrv_find_base(top_bs);
3013     }
3014 
3015     if (base_bs == NULL) {
3016         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3017         goto out;
3018     }
3019 
3020     assert(bdrv_get_aio_context(base_bs) == aio_context);
3021 
3022     if (bdrv_op_is_blocked(base_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3023         goto out;
3024     }
3025 
3026     /* Do not allow attempts to commit an image into itself */
3027     if (top_bs == base_bs) {
3028         error_setg(errp, "cannot commit an image into itself");
3029         goto out;
3030     }
3031 
3032     if (top_bs == bs) {
3033         if (has_backing_file) {
3034             error_setg(errp, "'backing-file' specified,"
3035                              " but 'top' is the active layer");
3036             goto out;
3037         }
3038         commit_active_start(bs, base_bs, speed, on_error, block_job_cb,
3039                             bs, &local_err);
3040     } else {
3041         commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
3042                      has_backing_file ? backing_file : NULL, &local_err);
3043     }
3044     if (local_err != NULL) {
3045         error_propagate(errp, local_err);
3046         goto out;
3047     }
3048 
3049 out:
3050     aio_context_release(aio_context);
3051 }
3052 
3053 static void do_drive_backup(const char *device, const char *target,
3054                             bool has_format, const char *format,
3055                             enum MirrorSyncMode sync,
3056                             bool has_mode, enum NewImageMode mode,
3057                             bool has_speed, int64_t speed,
3058                             bool has_bitmap, const char *bitmap,
3059                             bool has_on_source_error,
3060                             BlockdevOnError on_source_error,
3061                             bool has_on_target_error,
3062                             BlockdevOnError on_target_error,
3063                             BlockJobTxn *txn, Error **errp)
3064 {
3065     BlockBackend *blk;
3066     BlockDriverState *bs;
3067     BlockDriverState *target_bs;
3068     BlockDriverState *source = NULL;
3069     BdrvDirtyBitmap *bmap = NULL;
3070     AioContext *aio_context;
3071     QDict *options = NULL;
3072     Error *local_err = NULL;
3073     int flags;
3074     int64_t size;
3075     int ret;
3076 
3077     if (!has_speed) {
3078         speed = 0;
3079     }
3080     if (!has_on_source_error) {
3081         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3082     }
3083     if (!has_on_target_error) {
3084         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3085     }
3086     if (!has_mode) {
3087         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3088     }
3089 
3090     blk = blk_by_name(device);
3091     if (!blk) {
3092         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3093                   "Device '%s' not found", device);
3094         return;
3095     }
3096 
3097     aio_context = blk_get_aio_context(blk);
3098     aio_context_acquire(aio_context);
3099 
3100     /* Although backup_run has this check too, we need to use bs->drv below, so
3101      * do an early check redundantly. */
3102     if (!blk_is_available(blk)) {
3103         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
3104         goto out;
3105     }
3106     bs = blk_bs(blk);
3107 
3108     if (!has_format) {
3109         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
3110     }
3111 
3112     /* Early check to avoid creating target */
3113     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3114         goto out;
3115     }
3116 
3117     flags = bs->open_flags | BDRV_O_RDWR;
3118 
3119     /* See if we have a backing HD we can use to create our new image
3120      * on top of. */
3121     if (sync == MIRROR_SYNC_MODE_TOP) {
3122         source = backing_bs(bs);
3123         if (!source) {
3124             sync = MIRROR_SYNC_MODE_FULL;
3125         }
3126     }
3127     if (sync == MIRROR_SYNC_MODE_NONE) {
3128         source = bs;
3129     }
3130 
3131     size = bdrv_getlength(bs);
3132     if (size < 0) {
3133         error_setg_errno(errp, -size, "bdrv_getlength failed");
3134         goto out;
3135     }
3136 
3137     if (mode != NEW_IMAGE_MODE_EXISTING) {
3138         assert(format);
3139         if (source) {
3140             bdrv_img_create(target, format, source->filename,
3141                             source->drv->format_name, NULL,
3142                             size, flags, &local_err, false);
3143         } else {
3144             bdrv_img_create(target, format, NULL, NULL, NULL,
3145                             size, flags, &local_err, false);
3146         }
3147     }
3148 
3149     if (local_err) {
3150         error_propagate(errp, local_err);
3151         goto out;
3152     }
3153 
3154     if (format) {
3155         options = qdict_new();
3156         qdict_put(options, "driver", qstring_from_str(format));
3157     }
3158 
3159     target_bs = NULL;
3160     ret = bdrv_open(&target_bs, target, NULL, options, flags, &local_err);
3161     if (ret < 0) {
3162         error_propagate(errp, local_err);
3163         goto out;
3164     }
3165 
3166     bdrv_set_aio_context(target_bs, aio_context);
3167 
3168     if (has_bitmap) {
3169         bmap = bdrv_find_dirty_bitmap(bs, bitmap);
3170         if (!bmap) {
3171             error_setg(errp, "Bitmap '%s' could not be found", bitmap);
3172             bdrv_unref(target_bs);
3173             goto out;
3174         }
3175     }
3176 
3177     backup_start(bs, target_bs, speed, sync, bmap,
3178                  on_source_error, on_target_error,
3179                  block_job_cb, bs, txn, &local_err);
3180     if (local_err != NULL) {
3181         bdrv_unref(target_bs);
3182         error_propagate(errp, local_err);
3183         goto out;
3184     }
3185 
3186 out:
3187     aio_context_release(aio_context);
3188 }
3189 
3190 void qmp_drive_backup(const char *device, const char *target,
3191                       bool has_format, const char *format,
3192                       enum MirrorSyncMode sync,
3193                       bool has_mode, enum NewImageMode mode,
3194                       bool has_speed, int64_t speed,
3195                       bool has_bitmap, const char *bitmap,
3196                       bool has_on_source_error, BlockdevOnError on_source_error,
3197                       bool has_on_target_error, BlockdevOnError on_target_error,
3198                       Error **errp)
3199 {
3200     return do_drive_backup(device, target, has_format, format, sync,
3201                            has_mode, mode, has_speed, speed,
3202                            has_bitmap, bitmap,
3203                            has_on_source_error, on_source_error,
3204                            has_on_target_error, on_target_error,
3205                            NULL, errp);
3206 }
3207 
3208 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3209 {
3210     return bdrv_named_nodes_list(errp);
3211 }
3212 
3213 void do_blockdev_backup(const char *device, const char *target,
3214                          enum MirrorSyncMode sync,
3215                          bool has_speed, int64_t speed,
3216                          bool has_on_source_error,
3217                          BlockdevOnError on_source_error,
3218                          bool has_on_target_error,
3219                          BlockdevOnError on_target_error,
3220                          BlockJobTxn *txn, Error **errp)
3221 {
3222     BlockBackend *blk, *target_blk;
3223     BlockDriverState *bs;
3224     BlockDriverState *target_bs;
3225     Error *local_err = NULL;
3226     AioContext *aio_context;
3227 
3228     if (!has_speed) {
3229         speed = 0;
3230     }
3231     if (!has_on_source_error) {
3232         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3233     }
3234     if (!has_on_target_error) {
3235         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3236     }
3237 
3238     blk = blk_by_name(device);
3239     if (!blk) {
3240         error_setg(errp, "Device '%s' not found", device);
3241         return;
3242     }
3243 
3244     aio_context = blk_get_aio_context(blk);
3245     aio_context_acquire(aio_context);
3246 
3247     if (!blk_is_available(blk)) {
3248         error_setg(errp, "Device '%s' has no medium", device);
3249         goto out;
3250     }
3251     bs = blk_bs(blk);
3252 
3253     target_blk = blk_by_name(target);
3254     if (!target_blk) {
3255         error_setg(errp, "Device '%s' not found", target);
3256         goto out;
3257     }
3258 
3259     if (!blk_is_available(target_blk)) {
3260         error_setg(errp, "Device '%s' has no medium", target);
3261         goto out;
3262     }
3263     target_bs = blk_bs(target_blk);
3264 
3265     bdrv_ref(target_bs);
3266     bdrv_set_aio_context(target_bs, aio_context);
3267     backup_start(bs, target_bs, speed, sync, NULL, on_source_error,
3268                  on_target_error, block_job_cb, bs, txn, &local_err);
3269     if (local_err != NULL) {
3270         bdrv_unref(target_bs);
3271         error_propagate(errp, local_err);
3272     }
3273 out:
3274     aio_context_release(aio_context);
3275 }
3276 
3277 void qmp_blockdev_backup(const char *device, const char *target,
3278                          enum MirrorSyncMode sync,
3279                          bool has_speed, int64_t speed,
3280                          bool has_on_source_error,
3281                          BlockdevOnError on_source_error,
3282                          bool has_on_target_error,
3283                          BlockdevOnError on_target_error,
3284                          Error **errp)
3285 {
3286     do_blockdev_backup(device, target, sync, has_speed, speed,
3287                        has_on_source_error, on_source_error,
3288                        has_on_target_error, on_target_error,
3289                        NULL, errp);
3290 }
3291 
3292 /* Parameter check and block job starting for drive mirroring.
3293  * Caller should hold @device and @target's aio context (must be the same).
3294  **/
3295 static void blockdev_mirror_common(BlockDriverState *bs,
3296                                    BlockDriverState *target,
3297                                    bool has_replaces, const char *replaces,
3298                                    enum MirrorSyncMode sync,
3299                                    bool has_speed, int64_t speed,
3300                                    bool has_granularity, uint32_t granularity,
3301                                    bool has_buf_size, int64_t buf_size,
3302                                    bool has_on_source_error,
3303                                    BlockdevOnError on_source_error,
3304                                    bool has_on_target_error,
3305                                    BlockdevOnError on_target_error,
3306                                    bool has_unmap, bool unmap,
3307                                    Error **errp)
3308 {
3309 
3310     if (!has_speed) {
3311         speed = 0;
3312     }
3313     if (!has_on_source_error) {
3314         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3315     }
3316     if (!has_on_target_error) {
3317         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3318     }
3319     if (!has_granularity) {
3320         granularity = 0;
3321     }
3322     if (!has_buf_size) {
3323         buf_size = 0;
3324     }
3325     if (!has_unmap) {
3326         unmap = true;
3327     }
3328 
3329     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3330         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3331                    "a value in range [512B, 64MB]");
3332         return;
3333     }
3334     if (granularity & (granularity - 1)) {
3335         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3336                    "power of 2");
3337         return;
3338     }
3339 
3340     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3341         return;
3342     }
3343     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3344         return;
3345     }
3346     if (target->blk) {
3347         error_setg(errp, "Cannot mirror to an attached block device");
3348         return;
3349     }
3350 
3351     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3352         sync = MIRROR_SYNC_MODE_FULL;
3353     }
3354 
3355     /* pass the node name to replace to mirror start since it's loose coupling
3356      * and will allow to check whether the node still exist at mirror completion
3357      */
3358     mirror_start(bs, target,
3359                  has_replaces ? replaces : NULL,
3360                  speed, granularity, buf_size, sync,
3361                  on_source_error, on_target_error, unmap,
3362                  block_job_cb, bs, errp);
3363 }
3364 
3365 void qmp_drive_mirror(const char *device, const char *target,
3366                       bool has_format, const char *format,
3367                       bool has_node_name, const char *node_name,
3368                       bool has_replaces, const char *replaces,
3369                       enum MirrorSyncMode sync,
3370                       bool has_mode, enum NewImageMode mode,
3371                       bool has_speed, int64_t speed,
3372                       bool has_granularity, uint32_t granularity,
3373                       bool has_buf_size, int64_t buf_size,
3374                       bool has_on_source_error, BlockdevOnError on_source_error,
3375                       bool has_on_target_error, BlockdevOnError on_target_error,
3376                       bool has_unmap, bool unmap,
3377                       Error **errp)
3378 {
3379     BlockDriverState *bs;
3380     BlockBackend *blk;
3381     BlockDriverState *source, *target_bs;
3382     AioContext *aio_context;
3383     Error *local_err = NULL;
3384     QDict *options = NULL;
3385     int flags;
3386     int64_t size;
3387     int ret;
3388 
3389     blk = blk_by_name(device);
3390     if (!blk) {
3391         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3392                   "Device '%s' not found", device);
3393         return;
3394     }
3395 
3396     aio_context = blk_get_aio_context(blk);
3397     aio_context_acquire(aio_context);
3398 
3399     if (!blk_is_available(blk)) {
3400         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
3401         goto out;
3402     }
3403     bs = blk_bs(blk);
3404     if (!has_mode) {
3405         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3406     }
3407 
3408     if (!has_format) {
3409         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
3410     }
3411 
3412     flags = bs->open_flags | BDRV_O_RDWR;
3413     source = backing_bs(bs);
3414     if (!source && sync == MIRROR_SYNC_MODE_TOP) {
3415         sync = MIRROR_SYNC_MODE_FULL;
3416     }
3417     if (sync == MIRROR_SYNC_MODE_NONE) {
3418         source = bs;
3419     }
3420 
3421     size = bdrv_getlength(bs);
3422     if (size < 0) {
3423         error_setg_errno(errp, -size, "bdrv_getlength failed");
3424         goto out;
3425     }
3426 
3427     if (has_replaces) {
3428         BlockDriverState *to_replace_bs;
3429         AioContext *replace_aio_context;
3430         int64_t replace_size;
3431 
3432         if (!has_node_name) {
3433             error_setg(errp, "a node-name must be provided when replacing a"
3434                              " named node of the graph");
3435             goto out;
3436         }
3437 
3438         to_replace_bs = check_to_replace_node(bs, replaces, &local_err);
3439 
3440         if (!to_replace_bs) {
3441             error_propagate(errp, local_err);
3442             goto out;
3443         }
3444 
3445         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3446         aio_context_acquire(replace_aio_context);
3447         replace_size = bdrv_getlength(to_replace_bs);
3448         aio_context_release(replace_aio_context);
3449 
3450         if (size != replace_size) {
3451             error_setg(errp, "cannot replace image with a mirror image of "
3452                              "different size");
3453             goto out;
3454         }
3455     }
3456 
3457     if ((sync == MIRROR_SYNC_MODE_FULL || !source)
3458         && mode != NEW_IMAGE_MODE_EXISTING)
3459     {
3460         /* create new image w/o backing file */
3461         assert(format);
3462         bdrv_img_create(target, format,
3463                         NULL, NULL, NULL, size, flags, &local_err, false);
3464     } else {
3465         switch (mode) {
3466         case NEW_IMAGE_MODE_EXISTING:
3467             break;
3468         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3469             /* create new image with backing file */
3470             bdrv_img_create(target, format,
3471                             source->filename,
3472                             source->drv->format_name,
3473                             NULL, size, flags, &local_err, false);
3474             break;
3475         default:
3476             abort();
3477         }
3478     }
3479 
3480     if (local_err) {
3481         error_propagate(errp, local_err);
3482         goto out;
3483     }
3484 
3485     options = qdict_new();
3486     if (has_node_name) {
3487         qdict_put(options, "node-name", qstring_from_str(node_name));
3488     }
3489     if (format) {
3490         qdict_put(options, "driver", qstring_from_str(format));
3491     }
3492 
3493     /* Mirroring takes care of copy-on-write using the source's backing
3494      * file.
3495      */
3496     target_bs = NULL;
3497     ret = bdrv_open(&target_bs, target, NULL, options,
3498                     flags | BDRV_O_NO_BACKING, &local_err);
3499     if (ret < 0) {
3500         error_propagate(errp, local_err);
3501         goto out;
3502     }
3503 
3504     bdrv_set_aio_context(target_bs, aio_context);
3505 
3506     blockdev_mirror_common(bs, target_bs,
3507                            has_replaces, replaces, sync,
3508                            has_speed, speed,
3509                            has_granularity, granularity,
3510                            has_buf_size, buf_size,
3511                            has_on_source_error, on_source_error,
3512                            has_on_target_error, on_target_error,
3513                            has_unmap, unmap,
3514                            &local_err);
3515     if (local_err) {
3516         error_propagate(errp, local_err);
3517         bdrv_unref(target_bs);
3518     }
3519 out:
3520     aio_context_release(aio_context);
3521 }
3522 
3523 void qmp_blockdev_mirror(const char *device, const char *target,
3524                          bool has_replaces, const char *replaces,
3525                          MirrorSyncMode sync,
3526                          bool has_speed, int64_t speed,
3527                          bool has_granularity, uint32_t granularity,
3528                          bool has_buf_size, int64_t buf_size,
3529                          bool has_on_source_error,
3530                          BlockdevOnError on_source_error,
3531                          bool has_on_target_error,
3532                          BlockdevOnError on_target_error,
3533                          Error **errp)
3534 {
3535     BlockDriverState *bs;
3536     BlockBackend *blk;
3537     BlockDriverState *target_bs;
3538     AioContext *aio_context;
3539     Error *local_err = NULL;
3540 
3541     blk = blk_by_name(device);
3542     if (!blk) {
3543         error_setg(errp, "Device '%s' not found", device);
3544         return;
3545     }
3546     bs = blk_bs(blk);
3547 
3548     if (!bs) {
3549         error_setg(errp, "Device '%s' has no media", device);
3550         return;
3551     }
3552 
3553     target_bs = bdrv_lookup_bs(target, target, errp);
3554     if (!target_bs) {
3555         return;
3556     }
3557 
3558     aio_context = bdrv_get_aio_context(bs);
3559     aio_context_acquire(aio_context);
3560 
3561     bdrv_ref(target_bs);
3562     bdrv_set_aio_context(target_bs, aio_context);
3563 
3564     blockdev_mirror_common(bs, target_bs,
3565                            has_replaces, replaces, sync,
3566                            has_speed, speed,
3567                            has_granularity, granularity,
3568                            has_buf_size, buf_size,
3569                            has_on_source_error, on_source_error,
3570                            has_on_target_error, on_target_error,
3571                            true, true,
3572                            &local_err);
3573     if (local_err) {
3574         error_propagate(errp, local_err);
3575         bdrv_unref(target_bs);
3576     }
3577 
3578     aio_context_release(aio_context);
3579 }
3580 
3581 /* Get the block job for a given device name and acquire its AioContext */
3582 static BlockJob *find_block_job(const char *device, AioContext **aio_context,
3583                                 Error **errp)
3584 {
3585     BlockBackend *blk;
3586     BlockDriverState *bs;
3587 
3588     *aio_context = NULL;
3589 
3590     blk = blk_by_name(device);
3591     if (!blk) {
3592         goto notfound;
3593     }
3594 
3595     *aio_context = blk_get_aio_context(blk);
3596     aio_context_acquire(*aio_context);
3597 
3598     if (!blk_is_available(blk)) {
3599         goto notfound;
3600     }
3601     bs = blk_bs(blk);
3602 
3603     if (!bs->job) {
3604         goto notfound;
3605     }
3606 
3607     return bs->job;
3608 
3609 notfound:
3610     error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3611               "No active block job on device '%s'", device);
3612     if (*aio_context) {
3613         aio_context_release(*aio_context);
3614         *aio_context = NULL;
3615     }
3616     return NULL;
3617 }
3618 
3619 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3620 {
3621     AioContext *aio_context;
3622     BlockJob *job = find_block_job(device, &aio_context, errp);
3623 
3624     if (!job) {
3625         return;
3626     }
3627 
3628     block_job_set_speed(job, speed, errp);
3629     aio_context_release(aio_context);
3630 }
3631 
3632 void qmp_block_job_cancel(const char *device,
3633                           bool has_force, bool force, Error **errp)
3634 {
3635     AioContext *aio_context;
3636     BlockJob *job = find_block_job(device, &aio_context, errp);
3637 
3638     if (!job) {
3639         return;
3640     }
3641 
3642     if (!has_force) {
3643         force = false;
3644     }
3645 
3646     if (job->user_paused && !force) {
3647         error_setg(errp, "The block job for device '%s' is currently paused",
3648                    device);
3649         goto out;
3650     }
3651 
3652     trace_qmp_block_job_cancel(job);
3653     block_job_cancel(job);
3654 out:
3655     aio_context_release(aio_context);
3656 }
3657 
3658 void qmp_block_job_pause(const char *device, Error **errp)
3659 {
3660     AioContext *aio_context;
3661     BlockJob *job = find_block_job(device, &aio_context, errp);
3662 
3663     if (!job || job->user_paused) {
3664         return;
3665     }
3666 
3667     job->user_paused = true;
3668     trace_qmp_block_job_pause(job);
3669     block_job_pause(job);
3670     aio_context_release(aio_context);
3671 }
3672 
3673 void qmp_block_job_resume(const char *device, Error **errp)
3674 {
3675     AioContext *aio_context;
3676     BlockJob *job = find_block_job(device, &aio_context, errp);
3677 
3678     if (!job || !job->user_paused) {
3679         return;
3680     }
3681 
3682     job->user_paused = false;
3683     trace_qmp_block_job_resume(job);
3684     block_job_resume(job);
3685     aio_context_release(aio_context);
3686 }
3687 
3688 void qmp_block_job_complete(const char *device, Error **errp)
3689 {
3690     AioContext *aio_context;
3691     BlockJob *job = find_block_job(device, &aio_context, errp);
3692 
3693     if (!job) {
3694         return;
3695     }
3696 
3697     trace_qmp_block_job_complete(job);
3698     block_job_complete(job, errp);
3699     aio_context_release(aio_context);
3700 }
3701 
3702 void qmp_change_backing_file(const char *device,
3703                              const char *image_node_name,
3704                              const char *backing_file,
3705                              Error **errp)
3706 {
3707     BlockBackend *blk;
3708     BlockDriverState *bs = NULL;
3709     AioContext *aio_context;
3710     BlockDriverState *image_bs = NULL;
3711     Error *local_err = NULL;
3712     bool ro;
3713     int open_flags;
3714     int ret;
3715 
3716     blk = blk_by_name(device);
3717     if (!blk) {
3718         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3719                   "Device '%s' not found", device);
3720         return;
3721     }
3722 
3723     aio_context = blk_get_aio_context(blk);
3724     aio_context_acquire(aio_context);
3725 
3726     if (!blk_is_available(blk)) {
3727         error_setg(errp, "Device '%s' has no medium", device);
3728         goto out;
3729     }
3730     bs = blk_bs(blk);
3731 
3732     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3733     if (local_err) {
3734         error_propagate(errp, local_err);
3735         goto out;
3736     }
3737 
3738     if (!image_bs) {
3739         error_setg(errp, "image file not found");
3740         goto out;
3741     }
3742 
3743     if (bdrv_find_base(image_bs) == image_bs) {
3744         error_setg(errp, "not allowing backing file change on an image "
3745                          "without a backing file");
3746         goto out;
3747     }
3748 
3749     /* even though we are not necessarily operating on bs, we need it to
3750      * determine if block ops are currently prohibited on the chain */
3751     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3752         goto out;
3753     }
3754 
3755     /* final sanity check */
3756     if (!bdrv_chain_contains(bs, image_bs)) {
3757         error_setg(errp, "'%s' and image file are not in the same chain",
3758                    device);
3759         goto out;
3760     }
3761 
3762     /* if not r/w, reopen to make r/w */
3763     open_flags = image_bs->open_flags;
3764     ro = bdrv_is_read_only(image_bs);
3765 
3766     if (ro) {
3767         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3768         if (local_err) {
3769             error_propagate(errp, local_err);
3770             goto out;
3771         }
3772     }
3773 
3774     ret = bdrv_change_backing_file(image_bs, backing_file,
3775                                image_bs->drv ? image_bs->drv->format_name : "");
3776 
3777     if (ret < 0) {
3778         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3779                          backing_file);
3780         /* don't exit here, so we can try to restore open flags if
3781          * appropriate */
3782     }
3783 
3784     if (ro) {
3785         bdrv_reopen(image_bs, open_flags, &local_err);
3786         if (local_err) {
3787             error_propagate(errp, local_err); /* will preserve prior errp */
3788         }
3789     }
3790 
3791 out:
3792     aio_context_release(aio_context);
3793 }
3794 
3795 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3796 {
3797     QmpOutputVisitor *ov = qmp_output_visitor_new();
3798     BlockDriverState *bs;
3799     BlockBackend *blk = NULL;
3800     QObject *obj;
3801     QDict *qdict;
3802     Error *local_err = NULL;
3803 
3804     /* TODO Sort it out in raw-posix and drive_new(): Reject aio=native with
3805      * cache.direct=false instead of silently switching to aio=threads, except
3806      * when called from drive_new().
3807      *
3808      * For now, simply forbidding the combination for all drivers will do. */
3809     if (options->has_aio && options->aio == BLOCKDEV_AIO_OPTIONS_NATIVE) {
3810         bool direct = options->has_cache &&
3811                       options->cache->has_direct &&
3812                       options->cache->direct;
3813         if (!direct) {
3814             error_setg(errp, "aio=native requires cache.direct=true");
3815             goto fail;
3816         }
3817     }
3818 
3819     visit_type_BlockdevOptions(qmp_output_get_visitor(ov),
3820                                &options, NULL, &local_err);
3821     if (local_err) {
3822         error_propagate(errp, local_err);
3823         goto fail;
3824     }
3825 
3826     obj = qmp_output_get_qobject(ov);
3827     qdict = qobject_to_qdict(obj);
3828 
3829     qdict_flatten(qdict);
3830 
3831     if (options->has_id) {
3832         blk = blockdev_init(NULL, qdict, &local_err);
3833         if (local_err) {
3834             error_propagate(errp, local_err);
3835             goto fail;
3836         }
3837 
3838         bs = blk_bs(blk);
3839     } else {
3840         if (!qdict_get_try_str(qdict, "node-name")) {
3841             error_setg(errp, "'id' and/or 'node-name' need to be specified for "
3842                        "the root node");
3843             goto fail;
3844         }
3845 
3846         bs = bds_tree_init(qdict, errp);
3847         if (!bs) {
3848             goto fail;
3849         }
3850     }
3851 
3852     if (bs && bdrv_key_required(bs)) {
3853         if (blk) {
3854             blk_unref(blk);
3855         } else {
3856             bdrv_unref(bs);
3857         }
3858         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3859         goto fail;
3860     }
3861 
3862 fail:
3863     qmp_output_visitor_cleanup(ov);
3864 }
3865 
3866 void qmp_x_blockdev_del(bool has_id, const char *id,
3867                         bool has_node_name, const char *node_name, Error **errp)
3868 {
3869     AioContext *aio_context;
3870     BlockBackend *blk;
3871     BlockDriverState *bs;
3872 
3873     if (has_id && has_node_name) {
3874         error_setg(errp, "Only one of id and node-name must be specified");
3875         return;
3876     } else if (!has_id && !has_node_name) {
3877         error_setg(errp, "No block device specified");
3878         return;
3879     }
3880 
3881     if (has_id) {
3882         blk = blk_by_name(id);
3883         if (!blk) {
3884             error_setg(errp, "Cannot find block backend %s", id);
3885             return;
3886         }
3887         if (blk_get_refcnt(blk) > 1) {
3888             error_setg(errp, "Block backend %s is in use", id);
3889             return;
3890         }
3891         bs = blk_bs(blk);
3892         aio_context = blk_get_aio_context(blk);
3893     } else {
3894         bs = bdrv_find_node(node_name);
3895         if (!bs) {
3896             error_setg(errp, "Cannot find node %s", node_name);
3897             return;
3898         }
3899         blk = bs->blk;
3900         if (blk) {
3901             error_setg(errp, "Node %s is in use by %s",
3902                        node_name, blk_name(blk));
3903             return;
3904         }
3905         aio_context = bdrv_get_aio_context(bs);
3906     }
3907 
3908     aio_context_acquire(aio_context);
3909 
3910     if (bs) {
3911         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3912             goto out;
3913         }
3914 
3915         if (bs->refcnt > 1 || !QLIST_EMPTY(&bs->parents)) {
3916             error_setg(errp, "Block device %s is in use",
3917                        bdrv_get_device_or_node_name(bs));
3918             goto out;
3919         }
3920     }
3921 
3922     if (blk) {
3923         blk_unref(blk);
3924     } else {
3925         bdrv_unref(bs);
3926     }
3927 
3928 out:
3929     aio_context_release(aio_context);
3930 }
3931 
3932 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3933 {
3934     BlockJobInfoList *head = NULL, **p_next = &head;
3935     BlockDriverState *bs;
3936 
3937     for (bs = bdrv_next(NULL); bs; bs = bdrv_next(bs)) {
3938         AioContext *aio_context = bdrv_get_aio_context(bs);
3939 
3940         aio_context_acquire(aio_context);
3941 
3942         if (bs->job) {
3943             BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
3944             elem->value = block_job_query(bs->job);
3945             *p_next = elem;
3946             p_next = &elem->next;
3947         }
3948 
3949         aio_context_release(aio_context);
3950     }
3951 
3952     return head;
3953 }
3954 
3955 QemuOptsList qemu_common_drive_opts = {
3956     .name = "drive",
3957     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3958     .desc = {
3959         {
3960             .name = "snapshot",
3961             .type = QEMU_OPT_BOOL,
3962             .help = "enable/disable snapshot mode",
3963         },{
3964             .name = "discard",
3965             .type = QEMU_OPT_STRING,
3966             .help = "discard operation (ignore/off, unmap/on)",
3967         },{
3968             .name = "aio",
3969             .type = QEMU_OPT_STRING,
3970             .help = "host AIO implementation (threads, native)",
3971         },{
3972             .name = "format",
3973             .type = QEMU_OPT_STRING,
3974             .help = "disk format (raw, qcow2, ...)",
3975         },{
3976             .name = "rerror",
3977             .type = QEMU_OPT_STRING,
3978             .help = "read error action",
3979         },{
3980             .name = "werror",
3981             .type = QEMU_OPT_STRING,
3982             .help = "write error action",
3983         },{
3984             .name = "read-only",
3985             .type = QEMU_OPT_BOOL,
3986             .help = "open drive file as read-only",
3987         },{
3988             .name = "throttling.iops-total",
3989             .type = QEMU_OPT_NUMBER,
3990             .help = "limit total I/O operations per second",
3991         },{
3992             .name = "throttling.iops-read",
3993             .type = QEMU_OPT_NUMBER,
3994             .help = "limit read operations per second",
3995         },{
3996             .name = "throttling.iops-write",
3997             .type = QEMU_OPT_NUMBER,
3998             .help = "limit write operations per second",
3999         },{
4000             .name = "throttling.bps-total",
4001             .type = QEMU_OPT_NUMBER,
4002             .help = "limit total bytes per second",
4003         },{
4004             .name = "throttling.bps-read",
4005             .type = QEMU_OPT_NUMBER,
4006             .help = "limit read bytes per second",
4007         },{
4008             .name = "throttling.bps-write",
4009             .type = QEMU_OPT_NUMBER,
4010             .help = "limit write bytes per second",
4011         },{
4012             .name = "throttling.iops-total-max",
4013             .type = QEMU_OPT_NUMBER,
4014             .help = "I/O operations burst",
4015         },{
4016             .name = "throttling.iops-read-max",
4017             .type = QEMU_OPT_NUMBER,
4018             .help = "I/O operations read burst",
4019         },{
4020             .name = "throttling.iops-write-max",
4021             .type = QEMU_OPT_NUMBER,
4022             .help = "I/O operations write burst",
4023         },{
4024             .name = "throttling.bps-total-max",
4025             .type = QEMU_OPT_NUMBER,
4026             .help = "total bytes burst",
4027         },{
4028             .name = "throttling.bps-read-max",
4029             .type = QEMU_OPT_NUMBER,
4030             .help = "total bytes read burst",
4031         },{
4032             .name = "throttling.bps-write-max",
4033             .type = QEMU_OPT_NUMBER,
4034             .help = "total bytes write burst",
4035         },{
4036             .name = "throttling.iops-size",
4037             .type = QEMU_OPT_NUMBER,
4038             .help = "when limiting by iops max size of an I/O in bytes",
4039         },{
4040             .name = "throttling.group",
4041             .type = QEMU_OPT_STRING,
4042             .help = "name of the block throttling group",
4043         },{
4044             .name = "copy-on-read",
4045             .type = QEMU_OPT_BOOL,
4046             .help = "copy read data from backing file into image file",
4047         },{
4048             .name = "detect-zeroes",
4049             .type = QEMU_OPT_STRING,
4050             .help = "try to optimize zero writes (off, on, unmap)",
4051         },{
4052             .name = "stats-account-invalid",
4053             .type = QEMU_OPT_BOOL,
4054             .help = "whether to account for invalid I/O operations "
4055                     "in the statistics",
4056         },{
4057             .name = "stats-account-failed",
4058             .type = QEMU_OPT_BOOL,
4059             .help = "whether to account for failed I/O operations "
4060                     "in the statistics",
4061         },
4062         { /* end of list */ }
4063     },
4064 };
4065 
4066 static QemuOptsList qemu_root_bds_opts = {
4067     .name = "root-bds",
4068     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4069     .desc = {
4070         {
4071             .name = "discard",
4072             .type = QEMU_OPT_STRING,
4073             .help = "discard operation (ignore/off, unmap/on)",
4074         },{
4075             .name = "aio",
4076             .type = QEMU_OPT_STRING,
4077             .help = "host AIO implementation (threads, native)",
4078         },{
4079             .name = "read-only",
4080             .type = QEMU_OPT_BOOL,
4081             .help = "open drive file as read-only",
4082         },{
4083             .name = "copy-on-read",
4084             .type = QEMU_OPT_BOOL,
4085             .help = "copy read data from backing file into image file",
4086         },{
4087             .name = "detect-zeroes",
4088             .type = QEMU_OPT_STRING,
4089             .help = "try to optimize zero writes (off, on, unmap)",
4090         },
4091         { /* end of list */ }
4092     },
4093 };
4094 
4095 QemuOptsList qemu_drive_opts = {
4096     .name = "drive",
4097     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4098     .desc = {
4099         /*
4100          * no elements => accept any params
4101          * validation will happen later
4102          */
4103         { /* end of list */ }
4104     },
4105 };
4106