xref: /openbmc/qemu/blockdev.c (revision 84a3a53c)
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_report("Failed to delete snapshot with id '%s' and name '%s' on "
1586                      "device '%s' in abort: %s",
1587                      sn->id_str,
1588                      sn->name,
1589                      bdrv_get_device_name(bs),
1590                      error_get_pretty(local_error));
1591         error_free(local_error);
1592     }
1593 }
1594 
1595 static void internal_snapshot_clean(BlkActionState *common)
1596 {
1597     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1598                                              common, common);
1599 
1600     if (state->aio_context) {
1601         if (state->bs) {
1602             bdrv_drained_end(state->bs);
1603         }
1604         aio_context_release(state->aio_context);
1605     }
1606 }
1607 
1608 /* external snapshot private data */
1609 typedef struct ExternalSnapshotState {
1610     BlkActionState common;
1611     BlockDriverState *old_bs;
1612     BlockDriverState *new_bs;
1613     AioContext *aio_context;
1614 } ExternalSnapshotState;
1615 
1616 static void external_snapshot_prepare(BlkActionState *common,
1617                                       Error **errp)
1618 {
1619     int flags = 0, ret;
1620     QDict *options = NULL;
1621     Error *local_err = NULL;
1622     /* Device and node name of the image to generate the snapshot from */
1623     const char *device;
1624     const char *node_name;
1625     /* Reference to the new image (for 'blockdev-snapshot') */
1626     const char *snapshot_ref;
1627     /* File name of the new image (for 'blockdev-snapshot-sync') */
1628     const char *new_image_file;
1629     ExternalSnapshotState *state =
1630                              DO_UPCAST(ExternalSnapshotState, common, common);
1631     TransactionAction *action = common->action;
1632 
1633     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1634      * purpose but a different set of parameters */
1635     switch (action->type) {
1636     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1637         {
1638             BlockdevSnapshot *s = action->u.blockdev_snapshot;
1639             device = s->node;
1640             node_name = s->node;
1641             new_image_file = NULL;
1642             snapshot_ref = s->overlay;
1643         }
1644         break;
1645     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1646         {
1647             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync;
1648             device = s->has_device ? s->device : NULL;
1649             node_name = s->has_node_name ? s->node_name : NULL;
1650             new_image_file = s->snapshot_file;
1651             snapshot_ref = NULL;
1652         }
1653         break;
1654     default:
1655         g_assert_not_reached();
1656     }
1657 
1658     /* start processing */
1659     if (action_check_completion_mode(common, errp) < 0) {
1660         return;
1661     }
1662 
1663     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1664     if (!state->old_bs) {
1665         return;
1666     }
1667 
1668     /* Acquire AioContext now so any threads operating on old_bs stop */
1669     state->aio_context = bdrv_get_aio_context(state->old_bs);
1670     aio_context_acquire(state->aio_context);
1671     bdrv_drained_begin(state->old_bs);
1672 
1673     if (!bdrv_is_inserted(state->old_bs)) {
1674         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1675         return;
1676     }
1677 
1678     if (bdrv_op_is_blocked(state->old_bs,
1679                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1680         return;
1681     }
1682 
1683     if (!bdrv_is_read_only(state->old_bs)) {
1684         if (bdrv_flush(state->old_bs)) {
1685             error_setg(errp, QERR_IO_ERROR);
1686             return;
1687         }
1688     }
1689 
1690     if (!bdrv_is_first_non_filter(state->old_bs)) {
1691         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1692         return;
1693     }
1694 
1695     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1696         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync;
1697         const char *format = s->has_format ? s->format : "qcow2";
1698         enum NewImageMode mode;
1699         const char *snapshot_node_name =
1700             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1701 
1702         if (node_name && !snapshot_node_name) {
1703             error_setg(errp, "New snapshot node name missing");
1704             return;
1705         }
1706 
1707         if (snapshot_node_name &&
1708             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1709             error_setg(errp, "New snapshot node name already in use");
1710             return;
1711         }
1712 
1713         flags = state->old_bs->open_flags;
1714 
1715         /* create new image w/backing file */
1716         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1717         if (mode != NEW_IMAGE_MODE_EXISTING) {
1718             bdrv_img_create(new_image_file, format,
1719                             state->old_bs->filename,
1720                             state->old_bs->drv->format_name,
1721                             NULL, -1, flags, &local_err, false);
1722             if (local_err) {
1723                 error_propagate(errp, local_err);
1724                 return;
1725             }
1726         }
1727 
1728         options = qdict_new();
1729         if (s->has_snapshot_node_name) {
1730             qdict_put(options, "node-name",
1731                       qstring_from_str(snapshot_node_name));
1732         }
1733         qdict_put(options, "driver", qstring_from_str(format));
1734 
1735         flags |= BDRV_O_NO_BACKING;
1736     }
1737 
1738     assert(state->new_bs == NULL);
1739     ret = bdrv_open(&state->new_bs, new_image_file, snapshot_ref, options,
1740                     flags, errp);
1741     /* We will manually add the backing_hd field to the bs later */
1742     if (ret != 0) {
1743         return;
1744     }
1745 
1746     if (state->new_bs->blk != NULL) {
1747         error_setg(errp, "The snapshot is already in use by %s",
1748                    blk_name(state->new_bs->blk));
1749         return;
1750     }
1751 
1752     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1753                            errp)) {
1754         return;
1755     }
1756 
1757     if (state->new_bs->backing != NULL) {
1758         error_setg(errp, "The snapshot already has a backing image");
1759         return;
1760     }
1761 
1762     if (!state->new_bs->drv->supports_backing) {
1763         error_setg(errp, "The snapshot does not support backing images");
1764     }
1765 }
1766 
1767 static void external_snapshot_commit(BlkActionState *common)
1768 {
1769     ExternalSnapshotState *state =
1770                              DO_UPCAST(ExternalSnapshotState, common, common);
1771 
1772     bdrv_set_aio_context(state->new_bs, state->aio_context);
1773 
1774     /* This removes our old bs and adds the new bs */
1775     bdrv_append(state->new_bs, state->old_bs);
1776     /* We don't need (or want) to use the transactional
1777      * bdrv_reopen_multiple() across all the entries at once, because we
1778      * don't want to abort all of them if one of them fails the reopen */
1779     bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1780                 NULL);
1781 }
1782 
1783 static void external_snapshot_abort(BlkActionState *common)
1784 {
1785     ExternalSnapshotState *state =
1786                              DO_UPCAST(ExternalSnapshotState, common, common);
1787     if (state->new_bs) {
1788         bdrv_unref(state->new_bs);
1789     }
1790 }
1791 
1792 static void external_snapshot_clean(BlkActionState *common)
1793 {
1794     ExternalSnapshotState *state =
1795                              DO_UPCAST(ExternalSnapshotState, common, common);
1796     if (state->aio_context) {
1797         bdrv_drained_end(state->old_bs);
1798         aio_context_release(state->aio_context);
1799     }
1800 }
1801 
1802 typedef struct DriveBackupState {
1803     BlkActionState common;
1804     BlockDriverState *bs;
1805     AioContext *aio_context;
1806     BlockJob *job;
1807 } DriveBackupState;
1808 
1809 static void do_drive_backup(const char *device, const char *target,
1810                             bool has_format, const char *format,
1811                             enum MirrorSyncMode sync,
1812                             bool has_mode, enum NewImageMode mode,
1813                             bool has_speed, int64_t speed,
1814                             bool has_bitmap, const char *bitmap,
1815                             bool has_on_source_error,
1816                             BlockdevOnError on_source_error,
1817                             bool has_on_target_error,
1818                             BlockdevOnError on_target_error,
1819                             BlockJobTxn *txn, Error **errp);
1820 
1821 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1822 {
1823     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1824     BlockBackend *blk;
1825     DriveBackup *backup;
1826     Error *local_err = NULL;
1827 
1828     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1829     backup = common->action->u.drive_backup;
1830 
1831     blk = blk_by_name(backup->device);
1832     if (!blk) {
1833         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1834                   "Device '%s' not found", backup->device);
1835         return;
1836     }
1837 
1838     if (!blk_is_available(blk)) {
1839         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, backup->device);
1840         return;
1841     }
1842 
1843     /* AioContext is released in .clean() */
1844     state->aio_context = blk_get_aio_context(blk);
1845     aio_context_acquire(state->aio_context);
1846     bdrv_drained_begin(blk_bs(blk));
1847     state->bs = blk_bs(blk);
1848 
1849     do_drive_backup(backup->device, backup->target,
1850                     backup->has_format, backup->format,
1851                     backup->sync,
1852                     backup->has_mode, backup->mode,
1853                     backup->has_speed, backup->speed,
1854                     backup->has_bitmap, backup->bitmap,
1855                     backup->has_on_source_error, backup->on_source_error,
1856                     backup->has_on_target_error, backup->on_target_error,
1857                     common->block_job_txn, &local_err);
1858     if (local_err) {
1859         error_propagate(errp, local_err);
1860         return;
1861     }
1862 
1863     state->job = state->bs->job;
1864 }
1865 
1866 static void drive_backup_abort(BlkActionState *common)
1867 {
1868     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1869     BlockDriverState *bs = state->bs;
1870 
1871     /* Only cancel if it's the job we started */
1872     if (bs && bs->job && bs->job == state->job) {
1873         block_job_cancel_sync(bs->job);
1874     }
1875 }
1876 
1877 static void drive_backup_clean(BlkActionState *common)
1878 {
1879     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1880 
1881     if (state->aio_context) {
1882         bdrv_drained_end(state->bs);
1883         aio_context_release(state->aio_context);
1884     }
1885 }
1886 
1887 typedef struct BlockdevBackupState {
1888     BlkActionState common;
1889     BlockDriverState *bs;
1890     BlockJob *job;
1891     AioContext *aio_context;
1892 } BlockdevBackupState;
1893 
1894 static void do_blockdev_backup(const char *device, const char *target,
1895                                enum MirrorSyncMode sync,
1896                                bool has_speed, int64_t speed,
1897                                bool has_on_source_error,
1898                                BlockdevOnError on_source_error,
1899                                bool has_on_target_error,
1900                                BlockdevOnError on_target_error,
1901                                BlockJobTxn *txn, Error **errp);
1902 
1903 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1904 {
1905     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1906     BlockdevBackup *backup;
1907     BlockBackend *blk, *target;
1908     Error *local_err = NULL;
1909 
1910     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1911     backup = common->action->u.blockdev_backup;
1912 
1913     blk = blk_by_name(backup->device);
1914     if (!blk) {
1915         error_setg(errp, "Device '%s' not found", backup->device);
1916         return;
1917     }
1918 
1919     if (!blk_is_available(blk)) {
1920         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, backup->device);
1921         return;
1922     }
1923 
1924     target = blk_by_name(backup->target);
1925     if (!target) {
1926         error_setg(errp, "Device '%s' not found", backup->target);
1927         return;
1928     }
1929 
1930     /* AioContext is released in .clean() */
1931     state->aio_context = blk_get_aio_context(blk);
1932     if (state->aio_context != blk_get_aio_context(target)) {
1933         state->aio_context = NULL;
1934         error_setg(errp, "Backup between two IO threads is not implemented");
1935         return;
1936     }
1937     aio_context_acquire(state->aio_context);
1938     state->bs = blk_bs(blk);
1939     bdrv_drained_begin(state->bs);
1940 
1941     do_blockdev_backup(backup->device, backup->target,
1942                        backup->sync,
1943                        backup->has_speed, backup->speed,
1944                        backup->has_on_source_error, backup->on_source_error,
1945                        backup->has_on_target_error, backup->on_target_error,
1946                        common->block_job_txn, &local_err);
1947     if (local_err) {
1948         error_propagate(errp, local_err);
1949         return;
1950     }
1951 
1952     state->job = state->bs->job;
1953 }
1954 
1955 static void blockdev_backup_abort(BlkActionState *common)
1956 {
1957     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1958     BlockDriverState *bs = state->bs;
1959 
1960     /* Only cancel if it's the job we started */
1961     if (bs && bs->job && bs->job == state->job) {
1962         block_job_cancel_sync(bs->job);
1963     }
1964 }
1965 
1966 static void blockdev_backup_clean(BlkActionState *common)
1967 {
1968     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1969 
1970     if (state->aio_context) {
1971         bdrv_drained_end(state->bs);
1972         aio_context_release(state->aio_context);
1973     }
1974 }
1975 
1976 typedef struct BlockDirtyBitmapState {
1977     BlkActionState common;
1978     BdrvDirtyBitmap *bitmap;
1979     BlockDriverState *bs;
1980     AioContext *aio_context;
1981     HBitmap *backup;
1982     bool prepared;
1983 } BlockDirtyBitmapState;
1984 
1985 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1986                                            Error **errp)
1987 {
1988     Error *local_err = NULL;
1989     BlockDirtyBitmapAdd *action;
1990     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1991                                              common, common);
1992 
1993     if (action_check_completion_mode(common, errp) < 0) {
1994         return;
1995     }
1996 
1997     action = common->action->u.block_dirty_bitmap_add;
1998     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1999     qmp_block_dirty_bitmap_add(action->node, action->name,
2000                                action->has_granularity, action->granularity,
2001                                &local_err);
2002 
2003     if (!local_err) {
2004         state->prepared = true;
2005     } else {
2006         error_propagate(errp, local_err);
2007     }
2008 }
2009 
2010 static void block_dirty_bitmap_add_abort(BlkActionState *common)
2011 {
2012     BlockDirtyBitmapAdd *action;
2013     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2014                                              common, common);
2015 
2016     action = common->action->u.block_dirty_bitmap_add;
2017     /* Should not be able to fail: IF the bitmap was added via .prepare(),
2018      * then the node reference and bitmap name must have been valid.
2019      */
2020     if (state->prepared) {
2021         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
2022     }
2023 }
2024 
2025 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
2026                                              Error **errp)
2027 {
2028     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2029                                              common, common);
2030     BlockDirtyBitmap *action;
2031 
2032     if (action_check_completion_mode(common, errp) < 0) {
2033         return;
2034     }
2035 
2036     action = common->action->u.block_dirty_bitmap_clear;
2037     state->bitmap = block_dirty_bitmap_lookup(action->node,
2038                                               action->name,
2039                                               &state->bs,
2040                                               &state->aio_context,
2041                                               errp);
2042     if (!state->bitmap) {
2043         return;
2044     }
2045 
2046     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2047         error_setg(errp, "Cannot modify a frozen bitmap");
2048         return;
2049     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2050         error_setg(errp, "Cannot clear a disabled bitmap");
2051         return;
2052     }
2053 
2054     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2055     /* AioContext is released in .clean() */
2056 }
2057 
2058 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2059 {
2060     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2061                                              common, common);
2062 
2063     bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2064 }
2065 
2066 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2067 {
2068     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2069                                              common, common);
2070 
2071     hbitmap_free(state->backup);
2072 }
2073 
2074 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2075 {
2076     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2077                                              common, common);
2078 
2079     if (state->aio_context) {
2080         aio_context_release(state->aio_context);
2081     }
2082 }
2083 
2084 static void abort_prepare(BlkActionState *common, Error **errp)
2085 {
2086     error_setg(errp, "Transaction aborted using Abort action");
2087 }
2088 
2089 static void abort_commit(BlkActionState *common)
2090 {
2091     g_assert_not_reached(); /* this action never succeeds */
2092 }
2093 
2094 static const BlkActionOps actions[] = {
2095     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2096         .instance_size = sizeof(ExternalSnapshotState),
2097         .prepare  = external_snapshot_prepare,
2098         .commit   = external_snapshot_commit,
2099         .abort = external_snapshot_abort,
2100         .clean = external_snapshot_clean,
2101     },
2102     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2103         .instance_size = sizeof(ExternalSnapshotState),
2104         .prepare  = external_snapshot_prepare,
2105         .commit   = external_snapshot_commit,
2106         .abort = external_snapshot_abort,
2107         .clean = external_snapshot_clean,
2108     },
2109     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2110         .instance_size = sizeof(DriveBackupState),
2111         .prepare = drive_backup_prepare,
2112         .abort = drive_backup_abort,
2113         .clean = drive_backup_clean,
2114     },
2115     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2116         .instance_size = sizeof(BlockdevBackupState),
2117         .prepare = blockdev_backup_prepare,
2118         .abort = blockdev_backup_abort,
2119         .clean = blockdev_backup_clean,
2120     },
2121     [TRANSACTION_ACTION_KIND_ABORT] = {
2122         .instance_size = sizeof(BlkActionState),
2123         .prepare = abort_prepare,
2124         .commit = abort_commit,
2125     },
2126     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2127         .instance_size = sizeof(InternalSnapshotState),
2128         .prepare  = internal_snapshot_prepare,
2129         .abort = internal_snapshot_abort,
2130         .clean = internal_snapshot_clean,
2131     },
2132     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2133         .instance_size = sizeof(BlockDirtyBitmapState),
2134         .prepare = block_dirty_bitmap_add_prepare,
2135         .abort = block_dirty_bitmap_add_abort,
2136     },
2137     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2138         .instance_size = sizeof(BlockDirtyBitmapState),
2139         .prepare = block_dirty_bitmap_clear_prepare,
2140         .commit = block_dirty_bitmap_clear_commit,
2141         .abort = block_dirty_bitmap_clear_abort,
2142         .clean = block_dirty_bitmap_clear_clean,
2143     }
2144 };
2145 
2146 /**
2147  * Allocate a TransactionProperties structure if necessary, and fill
2148  * that structure with desired defaults if they are unset.
2149  */
2150 static TransactionProperties *get_transaction_properties(
2151     TransactionProperties *props)
2152 {
2153     if (!props) {
2154         props = g_new0(TransactionProperties, 1);
2155     }
2156 
2157     if (!props->has_completion_mode) {
2158         props->has_completion_mode = true;
2159         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2160     }
2161 
2162     return props;
2163 }
2164 
2165 /*
2166  * 'Atomic' group operations.  The operations are performed as a set, and if
2167  * any fail then we roll back all operations in the group.
2168  */
2169 void qmp_transaction(TransactionActionList *dev_list,
2170                      bool has_props,
2171                      struct TransactionProperties *props,
2172                      Error **errp)
2173 {
2174     TransactionActionList *dev_entry = dev_list;
2175     BlockJobTxn *block_job_txn = NULL;
2176     BlkActionState *state, *next;
2177     Error *local_err = NULL;
2178 
2179     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2180     QSIMPLEQ_INIT(&snap_bdrv_states);
2181 
2182     /* Does this transaction get canceled as a group on failure?
2183      * If not, we don't really need to make a BlockJobTxn.
2184      */
2185     props = get_transaction_properties(props);
2186     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2187         block_job_txn = block_job_txn_new();
2188     }
2189 
2190     /* drain all i/o before any operations */
2191     bdrv_drain_all();
2192 
2193     /* We don't do anything in this loop that commits us to the operations */
2194     while (NULL != dev_entry) {
2195         TransactionAction *dev_info = NULL;
2196         const BlkActionOps *ops;
2197 
2198         dev_info = dev_entry->value;
2199         dev_entry = dev_entry->next;
2200 
2201         assert(dev_info->type < ARRAY_SIZE(actions));
2202 
2203         ops = &actions[dev_info->type];
2204         assert(ops->instance_size > 0);
2205 
2206         state = g_malloc0(ops->instance_size);
2207         state->ops = ops;
2208         state->action = dev_info;
2209         state->block_job_txn = block_job_txn;
2210         state->txn_props = props;
2211         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2212 
2213         state->ops->prepare(state, &local_err);
2214         if (local_err) {
2215             error_propagate(errp, local_err);
2216             goto delete_and_fail;
2217         }
2218     }
2219 
2220     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2221         if (state->ops->commit) {
2222             state->ops->commit(state);
2223         }
2224     }
2225 
2226     /* success */
2227     goto exit;
2228 
2229 delete_and_fail:
2230     /* failure, and it is all-or-none; roll back all operations */
2231     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2232         if (state->ops->abort) {
2233             state->ops->abort(state);
2234         }
2235     }
2236 exit:
2237     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2238         if (state->ops->clean) {
2239             state->ops->clean(state);
2240         }
2241         g_free(state);
2242     }
2243     if (!has_props) {
2244         qapi_free_TransactionProperties(props);
2245     }
2246     block_job_txn_unref(block_job_txn);
2247 }
2248 
2249 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
2250 {
2251     Error *local_err = NULL;
2252 
2253     qmp_blockdev_open_tray(device, has_force, force, &local_err);
2254     if (local_err) {
2255         error_propagate(errp, local_err);
2256         return;
2257     }
2258 
2259     qmp_x_blockdev_remove_medium(device, errp);
2260 }
2261 
2262 void qmp_block_passwd(bool has_device, const char *device,
2263                       bool has_node_name, const char *node_name,
2264                       const char *password, Error **errp)
2265 {
2266     Error *local_err = NULL;
2267     BlockDriverState *bs;
2268     AioContext *aio_context;
2269 
2270     bs = bdrv_lookup_bs(has_device ? device : NULL,
2271                         has_node_name ? node_name : NULL,
2272                         &local_err);
2273     if (local_err) {
2274         error_propagate(errp, local_err);
2275         return;
2276     }
2277 
2278     aio_context = bdrv_get_aio_context(bs);
2279     aio_context_acquire(aio_context);
2280 
2281     bdrv_add_key(bs, password, errp);
2282 
2283     aio_context_release(aio_context);
2284 }
2285 
2286 void qmp_blockdev_open_tray(const char *device, bool has_force, bool force,
2287                             Error **errp)
2288 {
2289     BlockBackend *blk;
2290     bool locked;
2291 
2292     if (!has_force) {
2293         force = false;
2294     }
2295 
2296     blk = blk_by_name(device);
2297     if (!blk) {
2298         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2299                   "Device '%s' not found", device);
2300         return;
2301     }
2302 
2303     if (!blk_dev_has_removable_media(blk)) {
2304         error_setg(errp, "Device '%s' is not removable", device);
2305         return;
2306     }
2307 
2308     if (blk_dev_is_tray_open(blk)) {
2309         return;
2310     }
2311 
2312     locked = blk_dev_is_medium_locked(blk);
2313     if (locked) {
2314         blk_dev_eject_request(blk, force);
2315     }
2316 
2317     if (!locked || force) {
2318         blk_dev_change_media_cb(blk, false);
2319     }
2320 }
2321 
2322 void qmp_blockdev_close_tray(const char *device, Error **errp)
2323 {
2324     BlockBackend *blk;
2325 
2326     blk = blk_by_name(device);
2327     if (!blk) {
2328         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2329                   "Device '%s' not found", device);
2330         return;
2331     }
2332 
2333     if (!blk_dev_has_removable_media(blk)) {
2334         error_setg(errp, "Device '%s' is not removable", device);
2335         return;
2336     }
2337 
2338     if (!blk_dev_is_tray_open(blk)) {
2339         return;
2340     }
2341 
2342     blk_dev_change_media_cb(blk, true);
2343 }
2344 
2345 void qmp_x_blockdev_remove_medium(const char *device, Error **errp)
2346 {
2347     BlockBackend *blk;
2348     BlockDriverState *bs;
2349     AioContext *aio_context;
2350     bool has_device;
2351 
2352     blk = blk_by_name(device);
2353     if (!blk) {
2354         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2355                   "Device '%s' not found", device);
2356         return;
2357     }
2358 
2359     /* For BBs without a device, we can exchange the BDS tree at will */
2360     has_device = blk_get_attached_dev(blk);
2361 
2362     if (has_device && !blk_dev_has_removable_media(blk)) {
2363         error_setg(errp, "Device '%s' is not removable", device);
2364         return;
2365     }
2366 
2367     if (has_device && !blk_dev_is_tray_open(blk)) {
2368         error_setg(errp, "Tray of device '%s' is not open", device);
2369         return;
2370     }
2371 
2372     bs = blk_bs(blk);
2373     if (!bs) {
2374         return;
2375     }
2376 
2377     aio_context = bdrv_get_aio_context(bs);
2378     aio_context_acquire(aio_context);
2379 
2380     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2381         goto out;
2382     }
2383 
2384     /* This follows the convention established by bdrv_make_anon() */
2385     if (bs->device_list.tqe_prev) {
2386         QTAILQ_REMOVE(&bdrv_states, bs, device_list);
2387         bs->device_list.tqe_prev = NULL;
2388     }
2389 
2390     blk_remove_bs(blk);
2391 
2392 out:
2393     aio_context_release(aio_context);
2394 }
2395 
2396 static void qmp_blockdev_insert_anon_medium(const char *device,
2397                                             BlockDriverState *bs, Error **errp)
2398 {
2399     BlockBackend *blk;
2400     bool has_device;
2401 
2402     blk = blk_by_name(device);
2403     if (!blk) {
2404         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2405                   "Device '%s' not found", device);
2406         return;
2407     }
2408 
2409     /* For BBs without a device, we can exchange the BDS tree at will */
2410     has_device = blk_get_attached_dev(blk);
2411 
2412     if (has_device && !blk_dev_has_removable_media(blk)) {
2413         error_setg(errp, "Device '%s' is not removable", device);
2414         return;
2415     }
2416 
2417     if (has_device && !blk_dev_is_tray_open(blk)) {
2418         error_setg(errp, "Tray of device '%s' is not open", device);
2419         return;
2420     }
2421 
2422     if (blk_bs(blk)) {
2423         error_setg(errp, "There already is a medium in device '%s'", device);
2424         return;
2425     }
2426 
2427     blk_insert_bs(blk, bs);
2428 
2429     QTAILQ_INSERT_TAIL(&bdrv_states, bs, device_list);
2430 }
2431 
2432 void qmp_x_blockdev_insert_medium(const char *device, const char *node_name,
2433                                   Error **errp)
2434 {
2435     BlockDriverState *bs;
2436 
2437     bs = bdrv_find_node(node_name);
2438     if (!bs) {
2439         error_setg(errp, "Node '%s' not found", node_name);
2440         return;
2441     }
2442 
2443     if (bs->blk) {
2444         error_setg(errp, "Node '%s' is already in use by '%s'", node_name,
2445                    blk_name(bs->blk));
2446         return;
2447     }
2448 
2449     qmp_blockdev_insert_anon_medium(device, bs, errp);
2450 }
2451 
2452 void qmp_blockdev_change_medium(const char *device, const char *filename,
2453                                 bool has_format, const char *format,
2454                                 bool has_read_only,
2455                                 BlockdevChangeReadOnlyMode read_only,
2456                                 Error **errp)
2457 {
2458     BlockBackend *blk;
2459     BlockDriverState *medium_bs = NULL;
2460     int bdrv_flags, ret;
2461     QDict *options = NULL;
2462     Error *err = NULL;
2463 
2464     blk = blk_by_name(device);
2465     if (!blk) {
2466         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2467                   "Device '%s' not found", device);
2468         goto fail;
2469     }
2470 
2471     if (blk_bs(blk)) {
2472         blk_update_root_state(blk);
2473     }
2474 
2475     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2476 
2477     if (!has_read_only) {
2478         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2479     }
2480 
2481     switch (read_only) {
2482     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2483         break;
2484 
2485     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2486         bdrv_flags &= ~BDRV_O_RDWR;
2487         break;
2488 
2489     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2490         bdrv_flags |= BDRV_O_RDWR;
2491         break;
2492 
2493     default:
2494         abort();
2495     }
2496 
2497     if (has_format) {
2498         options = qdict_new();
2499         qdict_put(options, "driver", qstring_from_str(format));
2500     }
2501 
2502     assert(!medium_bs);
2503     ret = bdrv_open(&medium_bs, filename, NULL, options, bdrv_flags, errp);
2504     if (ret < 0) {
2505         goto fail;
2506     }
2507 
2508     blk_apply_root_state(blk, medium_bs);
2509 
2510     bdrv_add_key(medium_bs, NULL, &err);
2511     if (err) {
2512         error_propagate(errp, err);
2513         goto fail;
2514     }
2515 
2516     qmp_blockdev_open_tray(device, false, false, &err);
2517     if (err) {
2518         error_propagate(errp, err);
2519         goto fail;
2520     }
2521 
2522     qmp_x_blockdev_remove_medium(device, &err);
2523     if (err) {
2524         error_propagate(errp, err);
2525         goto fail;
2526     }
2527 
2528     qmp_blockdev_insert_anon_medium(device, medium_bs, &err);
2529     if (err) {
2530         error_propagate(errp, err);
2531         goto fail;
2532     }
2533 
2534     qmp_blockdev_close_tray(device, errp);
2535 
2536 fail:
2537     /* If the medium has been inserted, the device has its own reference, so
2538      * ours must be relinquished; and if it has not been inserted successfully,
2539      * the reference must be relinquished anyway */
2540     bdrv_unref(medium_bs);
2541 }
2542 
2543 /* throttling disk I/O limits */
2544 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
2545                                int64_t bps_wr,
2546                                int64_t iops,
2547                                int64_t iops_rd,
2548                                int64_t iops_wr,
2549                                bool has_bps_max,
2550                                int64_t bps_max,
2551                                bool has_bps_rd_max,
2552                                int64_t bps_rd_max,
2553                                bool has_bps_wr_max,
2554                                int64_t bps_wr_max,
2555                                bool has_iops_max,
2556                                int64_t iops_max,
2557                                bool has_iops_rd_max,
2558                                int64_t iops_rd_max,
2559                                bool has_iops_wr_max,
2560                                int64_t iops_wr_max,
2561                                bool has_iops_size,
2562                                int64_t iops_size,
2563                                bool has_group,
2564                                const char *group, Error **errp)
2565 {
2566     ThrottleConfig cfg;
2567     BlockDriverState *bs;
2568     BlockBackend *blk;
2569     AioContext *aio_context;
2570 
2571     blk = blk_by_name(device);
2572     if (!blk) {
2573         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2574                   "Device '%s' not found", device);
2575         return;
2576     }
2577 
2578     aio_context = blk_get_aio_context(blk);
2579     aio_context_acquire(aio_context);
2580 
2581     bs = blk_bs(blk);
2582     if (!bs) {
2583         error_setg(errp, "Device '%s' has no medium", device);
2584         goto out;
2585     }
2586 
2587     memset(&cfg, 0, sizeof(cfg));
2588     cfg.buckets[THROTTLE_BPS_TOTAL].avg = bps;
2589     cfg.buckets[THROTTLE_BPS_READ].avg  = bps_rd;
2590     cfg.buckets[THROTTLE_BPS_WRITE].avg = bps_wr;
2591 
2592     cfg.buckets[THROTTLE_OPS_TOTAL].avg = iops;
2593     cfg.buckets[THROTTLE_OPS_READ].avg  = iops_rd;
2594     cfg.buckets[THROTTLE_OPS_WRITE].avg = iops_wr;
2595 
2596     if (has_bps_max) {
2597         cfg.buckets[THROTTLE_BPS_TOTAL].max = bps_max;
2598     }
2599     if (has_bps_rd_max) {
2600         cfg.buckets[THROTTLE_BPS_READ].max = bps_rd_max;
2601     }
2602     if (has_bps_wr_max) {
2603         cfg.buckets[THROTTLE_BPS_WRITE].max = bps_wr_max;
2604     }
2605     if (has_iops_max) {
2606         cfg.buckets[THROTTLE_OPS_TOTAL].max = iops_max;
2607     }
2608     if (has_iops_rd_max) {
2609         cfg.buckets[THROTTLE_OPS_READ].max = iops_rd_max;
2610     }
2611     if (has_iops_wr_max) {
2612         cfg.buckets[THROTTLE_OPS_WRITE].max = iops_wr_max;
2613     }
2614 
2615     if (has_iops_size) {
2616         cfg.op_size = iops_size;
2617     }
2618 
2619     if (!check_throttle_config(&cfg, errp)) {
2620         goto out;
2621     }
2622 
2623     if (throttle_enabled(&cfg)) {
2624         /* Enable I/O limits if they're not enabled yet, otherwise
2625          * just update the throttling group. */
2626         if (!bs->throttle_state) {
2627             bdrv_io_limits_enable(bs, has_group ? group : device);
2628         } else if (has_group) {
2629             bdrv_io_limits_update_group(bs, group);
2630         }
2631         /* Set the new throttling configuration */
2632         bdrv_set_io_limits(bs, &cfg);
2633     } else if (bs->throttle_state) {
2634         /* If all throttling settings are set to 0, disable I/O limits */
2635         bdrv_io_limits_disable(bs);
2636     }
2637 
2638 out:
2639     aio_context_release(aio_context);
2640 }
2641 
2642 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2643                                 bool has_granularity, uint32_t granularity,
2644                                 Error **errp)
2645 {
2646     AioContext *aio_context;
2647     BlockDriverState *bs;
2648 
2649     if (!name || name[0] == '\0') {
2650         error_setg(errp, "Bitmap name cannot be empty");
2651         return;
2652     }
2653 
2654     bs = bdrv_lookup_bs(node, node, errp);
2655     if (!bs) {
2656         return;
2657     }
2658 
2659     aio_context = bdrv_get_aio_context(bs);
2660     aio_context_acquire(aio_context);
2661 
2662     if (has_granularity) {
2663         if (granularity < 512 || !is_power_of_2(granularity)) {
2664             error_setg(errp, "Granularity must be power of 2 "
2665                              "and at least 512");
2666             goto out;
2667         }
2668     } else {
2669         /* Default to cluster size, if available: */
2670         granularity = bdrv_get_default_bitmap_granularity(bs);
2671     }
2672 
2673     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2674 
2675  out:
2676     aio_context_release(aio_context);
2677 }
2678 
2679 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2680                                    Error **errp)
2681 {
2682     AioContext *aio_context;
2683     BlockDriverState *bs;
2684     BdrvDirtyBitmap *bitmap;
2685 
2686     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2687     if (!bitmap || !bs) {
2688         return;
2689     }
2690 
2691     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2692         error_setg(errp,
2693                    "Bitmap '%s' is currently frozen and cannot be removed",
2694                    name);
2695         goto out;
2696     }
2697     bdrv_dirty_bitmap_make_anon(bitmap);
2698     bdrv_release_dirty_bitmap(bs, bitmap);
2699 
2700  out:
2701     aio_context_release(aio_context);
2702 }
2703 
2704 /**
2705  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2706  * immediately after a full backup operation.
2707  */
2708 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2709                                   Error **errp)
2710 {
2711     AioContext *aio_context;
2712     BdrvDirtyBitmap *bitmap;
2713     BlockDriverState *bs;
2714 
2715     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2716     if (!bitmap || !bs) {
2717         return;
2718     }
2719 
2720     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2721         error_setg(errp,
2722                    "Bitmap '%s' is currently frozen and cannot be modified",
2723                    name);
2724         goto out;
2725     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2726         error_setg(errp,
2727                    "Bitmap '%s' is currently disabled and cannot be cleared",
2728                    name);
2729         goto out;
2730     }
2731 
2732     bdrv_clear_dirty_bitmap(bitmap, NULL);
2733 
2734  out:
2735     aio_context_release(aio_context);
2736 }
2737 
2738 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2739 {
2740     const char *id = qdict_get_str(qdict, "id");
2741     BlockBackend *blk;
2742     BlockDriverState *bs;
2743     AioContext *aio_context;
2744     Error *local_err = NULL;
2745 
2746     blk = blk_by_name(id);
2747     if (!blk) {
2748         error_report("Device '%s' not found", id);
2749         return;
2750     }
2751 
2752     if (!blk_legacy_dinfo(blk)) {
2753         error_report("Deleting device added with blockdev-add"
2754                      " is not supported");
2755         return;
2756     }
2757 
2758     aio_context = blk_get_aio_context(blk);
2759     aio_context_acquire(aio_context);
2760 
2761     bs = blk_bs(blk);
2762     if (bs) {
2763         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2764             error_report_err(local_err);
2765             aio_context_release(aio_context);
2766             return;
2767         }
2768 
2769         bdrv_close(bs);
2770     }
2771 
2772     /* if we have a device attached to this BlockDriverState
2773      * then we need to make the drive anonymous until the device
2774      * can be removed.  If this is a drive with no device backing
2775      * then we can just get rid of the block driver state right here.
2776      */
2777     if (blk_get_attached_dev(blk)) {
2778         blk_hide_on_behalf_of_hmp_drive_del(blk);
2779         /* Further I/O must not pause the guest */
2780         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2781                          BLOCKDEV_ON_ERROR_REPORT);
2782     } else {
2783         blk_unref(blk);
2784     }
2785 
2786     aio_context_release(aio_context);
2787 }
2788 
2789 void qmp_block_resize(bool has_device, const char *device,
2790                       bool has_node_name, const char *node_name,
2791                       int64_t size, Error **errp)
2792 {
2793     Error *local_err = NULL;
2794     BlockDriverState *bs;
2795     AioContext *aio_context;
2796     int ret;
2797 
2798     bs = bdrv_lookup_bs(has_device ? device : NULL,
2799                         has_node_name ? node_name : NULL,
2800                         &local_err);
2801     if (local_err) {
2802         error_propagate(errp, local_err);
2803         return;
2804     }
2805 
2806     aio_context = bdrv_get_aio_context(bs);
2807     aio_context_acquire(aio_context);
2808 
2809     if (!bdrv_is_first_non_filter(bs)) {
2810         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2811         goto out;
2812     }
2813 
2814     if (size < 0) {
2815         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2816         goto out;
2817     }
2818 
2819     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2820         error_setg(errp, QERR_DEVICE_IN_USE, device);
2821         goto out;
2822     }
2823 
2824     /* complete all in-flight operations before resizing the device */
2825     bdrv_drain_all();
2826 
2827     ret = bdrv_truncate(bs, size);
2828     switch (ret) {
2829     case 0:
2830         break;
2831     case -ENOMEDIUM:
2832         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2833         break;
2834     case -ENOTSUP:
2835         error_setg(errp, QERR_UNSUPPORTED);
2836         break;
2837     case -EACCES:
2838         error_setg(errp, "Device '%s' is read only", device);
2839         break;
2840     case -EBUSY:
2841         error_setg(errp, QERR_DEVICE_IN_USE, device);
2842         break;
2843     default:
2844         error_setg_errno(errp, -ret, "Could not resize");
2845         break;
2846     }
2847 
2848 out:
2849     aio_context_release(aio_context);
2850 }
2851 
2852 static void block_job_cb(void *opaque, int ret)
2853 {
2854     /* Note that this function may be executed from another AioContext besides
2855      * the QEMU main loop.  If you need to access anything that assumes the
2856      * QEMU global mutex, use a BH or introduce a mutex.
2857      */
2858 
2859     BlockDriverState *bs = opaque;
2860     const char *msg = NULL;
2861 
2862     trace_block_job_cb(bs, bs->job, ret);
2863 
2864     assert(bs->job);
2865 
2866     if (ret < 0) {
2867         msg = strerror(-ret);
2868     }
2869 
2870     if (block_job_is_cancelled(bs->job)) {
2871         block_job_event_cancelled(bs->job);
2872     } else {
2873         block_job_event_completed(bs->job, msg);
2874     }
2875 }
2876 
2877 void qmp_block_stream(const char *device,
2878                       bool has_base, const char *base,
2879                       bool has_backing_file, const char *backing_file,
2880                       bool has_speed, int64_t speed,
2881                       bool has_on_error, BlockdevOnError on_error,
2882                       Error **errp)
2883 {
2884     BlockBackend *blk;
2885     BlockDriverState *bs;
2886     BlockDriverState *base_bs = NULL;
2887     AioContext *aio_context;
2888     Error *local_err = NULL;
2889     const char *base_name = NULL;
2890 
2891     if (!has_on_error) {
2892         on_error = BLOCKDEV_ON_ERROR_REPORT;
2893     }
2894 
2895     blk = blk_by_name(device);
2896     if (!blk) {
2897         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2898                   "Device '%s' not found", device);
2899         return;
2900     }
2901 
2902     aio_context = blk_get_aio_context(blk);
2903     aio_context_acquire(aio_context);
2904 
2905     if (!blk_is_available(blk)) {
2906         error_setg(errp, "Device '%s' has no medium", device);
2907         goto out;
2908     }
2909     bs = blk_bs(blk);
2910 
2911     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_STREAM, errp)) {
2912         goto out;
2913     }
2914 
2915     if (has_base) {
2916         base_bs = bdrv_find_backing_image(bs, base);
2917         if (base_bs == NULL) {
2918             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2919             goto out;
2920         }
2921         assert(bdrv_get_aio_context(base_bs) == aio_context);
2922         base_name = base;
2923     }
2924 
2925     /* if we are streaming the entire chain, the result will have no backing
2926      * file, and specifying one is therefore an error */
2927     if (base_bs == NULL && has_backing_file) {
2928         error_setg(errp, "backing file specified, but streaming the "
2929                          "entire chain");
2930         goto out;
2931     }
2932 
2933     /* backing_file string overrides base bs filename */
2934     base_name = has_backing_file ? backing_file : base_name;
2935 
2936     stream_start(bs, base_bs, base_name, has_speed ? speed : 0,
2937                  on_error, block_job_cb, bs, &local_err);
2938     if (local_err) {
2939         error_propagate(errp, local_err);
2940         goto out;
2941     }
2942 
2943     trace_qmp_block_stream(bs, bs->job);
2944 
2945 out:
2946     aio_context_release(aio_context);
2947 }
2948 
2949 void qmp_block_commit(const char *device,
2950                       bool has_base, const char *base,
2951                       bool has_top, const char *top,
2952                       bool has_backing_file, const char *backing_file,
2953                       bool has_speed, int64_t speed,
2954                       Error **errp)
2955 {
2956     BlockBackend *blk;
2957     BlockDriverState *bs;
2958     BlockDriverState *base_bs, *top_bs;
2959     AioContext *aio_context;
2960     Error *local_err = NULL;
2961     /* This will be part of the QMP command, if/when the
2962      * BlockdevOnError change for blkmirror makes it in
2963      */
2964     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
2965 
2966     if (!has_speed) {
2967         speed = 0;
2968     }
2969 
2970     /* Important Note:
2971      *  libvirt relies on the DeviceNotFound error class in order to probe for
2972      *  live commit feature versions; for this to work, we must make sure to
2973      *  perform the device lookup before any generic errors that may occur in a
2974      *  scenario in which all optional arguments are omitted. */
2975     blk = blk_by_name(device);
2976     if (!blk) {
2977         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2978                   "Device '%s' not found", device);
2979         return;
2980     }
2981 
2982     aio_context = blk_get_aio_context(blk);
2983     aio_context_acquire(aio_context);
2984 
2985     if (!blk_is_available(blk)) {
2986         error_setg(errp, "Device '%s' has no medium", device);
2987         goto out;
2988     }
2989     bs = blk_bs(blk);
2990 
2991     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2992         goto out;
2993     }
2994 
2995     /* default top_bs is the active layer */
2996     top_bs = bs;
2997 
2998     if (has_top && top) {
2999         if (strcmp(bs->filename, top) != 0) {
3000             top_bs = bdrv_find_backing_image(bs, top);
3001         }
3002     }
3003 
3004     if (top_bs == NULL) {
3005         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3006         goto out;
3007     }
3008 
3009     assert(bdrv_get_aio_context(top_bs) == aio_context);
3010 
3011     if (has_base && base) {
3012         base_bs = bdrv_find_backing_image(top_bs, base);
3013     } else {
3014         base_bs = bdrv_find_base(top_bs);
3015     }
3016 
3017     if (base_bs == NULL) {
3018         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3019         goto out;
3020     }
3021 
3022     assert(bdrv_get_aio_context(base_bs) == aio_context);
3023 
3024     if (bdrv_op_is_blocked(base_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3025         goto out;
3026     }
3027 
3028     /* Do not allow attempts to commit an image into itself */
3029     if (top_bs == base_bs) {
3030         error_setg(errp, "cannot commit an image into itself");
3031         goto out;
3032     }
3033 
3034     if (top_bs == bs) {
3035         if (has_backing_file) {
3036             error_setg(errp, "'backing-file' specified,"
3037                              " but 'top' is the active layer");
3038             goto out;
3039         }
3040         commit_active_start(bs, base_bs, speed, on_error, block_job_cb,
3041                             bs, &local_err);
3042     } else {
3043         commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
3044                      has_backing_file ? backing_file : NULL, &local_err);
3045     }
3046     if (local_err != NULL) {
3047         error_propagate(errp, local_err);
3048         goto out;
3049     }
3050 
3051 out:
3052     aio_context_release(aio_context);
3053 }
3054 
3055 static void do_drive_backup(const char *device, const char *target,
3056                             bool has_format, const char *format,
3057                             enum MirrorSyncMode sync,
3058                             bool has_mode, enum NewImageMode mode,
3059                             bool has_speed, int64_t speed,
3060                             bool has_bitmap, const char *bitmap,
3061                             bool has_on_source_error,
3062                             BlockdevOnError on_source_error,
3063                             bool has_on_target_error,
3064                             BlockdevOnError on_target_error,
3065                             BlockJobTxn *txn, Error **errp)
3066 {
3067     BlockBackend *blk;
3068     BlockDriverState *bs;
3069     BlockDriverState *target_bs;
3070     BlockDriverState *source = NULL;
3071     BdrvDirtyBitmap *bmap = NULL;
3072     AioContext *aio_context;
3073     QDict *options = NULL;
3074     Error *local_err = NULL;
3075     int flags;
3076     int64_t size;
3077     int ret;
3078 
3079     if (!has_speed) {
3080         speed = 0;
3081     }
3082     if (!has_on_source_error) {
3083         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3084     }
3085     if (!has_on_target_error) {
3086         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3087     }
3088     if (!has_mode) {
3089         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3090     }
3091 
3092     blk = blk_by_name(device);
3093     if (!blk) {
3094         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3095                   "Device '%s' not found", device);
3096         return;
3097     }
3098 
3099     aio_context = blk_get_aio_context(blk);
3100     aio_context_acquire(aio_context);
3101 
3102     /* Although backup_run has this check too, we need to use bs->drv below, so
3103      * do an early check redundantly. */
3104     if (!blk_is_available(blk)) {
3105         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
3106         goto out;
3107     }
3108     bs = blk_bs(blk);
3109 
3110     if (!has_format) {
3111         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
3112     }
3113 
3114     /* Early check to avoid creating target */
3115     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3116         goto out;
3117     }
3118 
3119     flags = bs->open_flags | BDRV_O_RDWR;
3120 
3121     /* See if we have a backing HD we can use to create our new image
3122      * on top of. */
3123     if (sync == MIRROR_SYNC_MODE_TOP) {
3124         source = backing_bs(bs);
3125         if (!source) {
3126             sync = MIRROR_SYNC_MODE_FULL;
3127         }
3128     }
3129     if (sync == MIRROR_SYNC_MODE_NONE) {
3130         source = bs;
3131     }
3132 
3133     size = bdrv_getlength(bs);
3134     if (size < 0) {
3135         error_setg_errno(errp, -size, "bdrv_getlength failed");
3136         goto out;
3137     }
3138 
3139     if (mode != NEW_IMAGE_MODE_EXISTING) {
3140         assert(format);
3141         if (source) {
3142             bdrv_img_create(target, format, source->filename,
3143                             source->drv->format_name, NULL,
3144                             size, flags, &local_err, false);
3145         } else {
3146             bdrv_img_create(target, format, NULL, NULL, NULL,
3147                             size, flags, &local_err, false);
3148         }
3149     }
3150 
3151     if (local_err) {
3152         error_propagate(errp, local_err);
3153         goto out;
3154     }
3155 
3156     if (format) {
3157         options = qdict_new();
3158         qdict_put(options, "driver", qstring_from_str(format));
3159     }
3160 
3161     target_bs = NULL;
3162     ret = bdrv_open(&target_bs, target, NULL, options, flags, &local_err);
3163     if (ret < 0) {
3164         error_propagate(errp, local_err);
3165         goto out;
3166     }
3167 
3168     bdrv_set_aio_context(target_bs, aio_context);
3169 
3170     if (has_bitmap) {
3171         bmap = bdrv_find_dirty_bitmap(bs, bitmap);
3172         if (!bmap) {
3173             error_setg(errp, "Bitmap '%s' could not be found", bitmap);
3174             bdrv_unref(target_bs);
3175             goto out;
3176         }
3177     }
3178 
3179     backup_start(bs, target_bs, speed, sync, bmap,
3180                  on_source_error, on_target_error,
3181                  block_job_cb, bs, txn, &local_err);
3182     if (local_err != NULL) {
3183         bdrv_unref(target_bs);
3184         error_propagate(errp, local_err);
3185         goto out;
3186     }
3187 
3188 out:
3189     aio_context_release(aio_context);
3190 }
3191 
3192 void qmp_drive_backup(const char *device, const char *target,
3193                       bool has_format, const char *format,
3194                       enum MirrorSyncMode sync,
3195                       bool has_mode, enum NewImageMode mode,
3196                       bool has_speed, int64_t speed,
3197                       bool has_bitmap, const char *bitmap,
3198                       bool has_on_source_error, BlockdevOnError on_source_error,
3199                       bool has_on_target_error, BlockdevOnError on_target_error,
3200                       Error **errp)
3201 {
3202     return do_drive_backup(device, target, has_format, format, sync,
3203                            has_mode, mode, has_speed, speed,
3204                            has_bitmap, bitmap,
3205                            has_on_source_error, on_source_error,
3206                            has_on_target_error, on_target_error,
3207                            NULL, errp);
3208 }
3209 
3210 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3211 {
3212     return bdrv_named_nodes_list(errp);
3213 }
3214 
3215 void do_blockdev_backup(const char *device, const char *target,
3216                          enum MirrorSyncMode sync,
3217                          bool has_speed, int64_t speed,
3218                          bool has_on_source_error,
3219                          BlockdevOnError on_source_error,
3220                          bool has_on_target_error,
3221                          BlockdevOnError on_target_error,
3222                          BlockJobTxn *txn, Error **errp)
3223 {
3224     BlockBackend *blk, *target_blk;
3225     BlockDriverState *bs;
3226     BlockDriverState *target_bs;
3227     Error *local_err = NULL;
3228     AioContext *aio_context;
3229 
3230     if (!has_speed) {
3231         speed = 0;
3232     }
3233     if (!has_on_source_error) {
3234         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3235     }
3236     if (!has_on_target_error) {
3237         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3238     }
3239 
3240     blk = blk_by_name(device);
3241     if (!blk) {
3242         error_setg(errp, "Device '%s' not found", device);
3243         return;
3244     }
3245 
3246     aio_context = blk_get_aio_context(blk);
3247     aio_context_acquire(aio_context);
3248 
3249     if (!blk_is_available(blk)) {
3250         error_setg(errp, "Device '%s' has no medium", device);
3251         goto out;
3252     }
3253     bs = blk_bs(blk);
3254 
3255     target_blk = blk_by_name(target);
3256     if (!target_blk) {
3257         error_setg(errp, "Device '%s' not found", target);
3258         goto out;
3259     }
3260 
3261     if (!blk_is_available(target_blk)) {
3262         error_setg(errp, "Device '%s' has no medium", target);
3263         goto out;
3264     }
3265     target_bs = blk_bs(target_blk);
3266 
3267     bdrv_ref(target_bs);
3268     bdrv_set_aio_context(target_bs, aio_context);
3269     backup_start(bs, target_bs, speed, sync, NULL, on_source_error,
3270                  on_target_error, block_job_cb, bs, txn, &local_err);
3271     if (local_err != NULL) {
3272         bdrv_unref(target_bs);
3273         error_propagate(errp, local_err);
3274     }
3275 out:
3276     aio_context_release(aio_context);
3277 }
3278 
3279 void qmp_blockdev_backup(const char *device, const char *target,
3280                          enum MirrorSyncMode sync,
3281                          bool has_speed, int64_t speed,
3282                          bool has_on_source_error,
3283                          BlockdevOnError on_source_error,
3284                          bool has_on_target_error,
3285                          BlockdevOnError on_target_error,
3286                          Error **errp)
3287 {
3288     do_blockdev_backup(device, target, sync, has_speed, speed,
3289                        has_on_source_error, on_source_error,
3290                        has_on_target_error, on_target_error,
3291                        NULL, errp);
3292 }
3293 
3294 /* Parameter check and block job starting for drive mirroring.
3295  * Caller should hold @device and @target's aio context (must be the same).
3296  **/
3297 static void blockdev_mirror_common(BlockDriverState *bs,
3298                                    BlockDriverState *target,
3299                                    bool has_replaces, const char *replaces,
3300                                    enum MirrorSyncMode sync,
3301                                    bool has_speed, int64_t speed,
3302                                    bool has_granularity, uint32_t granularity,
3303                                    bool has_buf_size, int64_t buf_size,
3304                                    bool has_on_source_error,
3305                                    BlockdevOnError on_source_error,
3306                                    bool has_on_target_error,
3307                                    BlockdevOnError on_target_error,
3308                                    bool has_unmap, bool unmap,
3309                                    Error **errp)
3310 {
3311 
3312     if (!has_speed) {
3313         speed = 0;
3314     }
3315     if (!has_on_source_error) {
3316         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3317     }
3318     if (!has_on_target_error) {
3319         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3320     }
3321     if (!has_granularity) {
3322         granularity = 0;
3323     }
3324     if (!has_buf_size) {
3325         buf_size = 0;
3326     }
3327     if (!has_unmap) {
3328         unmap = true;
3329     }
3330 
3331     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3332         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3333                    "a value in range [512B, 64MB]");
3334         return;
3335     }
3336     if (granularity & (granularity - 1)) {
3337         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3338                    "power of 2");
3339         return;
3340     }
3341 
3342     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3343         return;
3344     }
3345     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3346         return;
3347     }
3348     if (target->blk) {
3349         error_setg(errp, "Cannot mirror to an attached block device");
3350         return;
3351     }
3352 
3353     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3354         sync = MIRROR_SYNC_MODE_FULL;
3355     }
3356 
3357     /* pass the node name to replace to mirror start since it's loose coupling
3358      * and will allow to check whether the node still exist at mirror completion
3359      */
3360     mirror_start(bs, target,
3361                  has_replaces ? replaces : NULL,
3362                  speed, granularity, buf_size, sync,
3363                  on_source_error, on_target_error, unmap,
3364                  block_job_cb, bs, errp);
3365 }
3366 
3367 void qmp_drive_mirror(const char *device, const char *target,
3368                       bool has_format, const char *format,
3369                       bool has_node_name, const char *node_name,
3370                       bool has_replaces, const char *replaces,
3371                       enum MirrorSyncMode sync,
3372                       bool has_mode, enum NewImageMode mode,
3373                       bool has_speed, int64_t speed,
3374                       bool has_granularity, uint32_t granularity,
3375                       bool has_buf_size, int64_t buf_size,
3376                       bool has_on_source_error, BlockdevOnError on_source_error,
3377                       bool has_on_target_error, BlockdevOnError on_target_error,
3378                       bool has_unmap, bool unmap,
3379                       Error **errp)
3380 {
3381     BlockDriverState *bs;
3382     BlockBackend *blk;
3383     BlockDriverState *source, *target_bs;
3384     AioContext *aio_context;
3385     Error *local_err = NULL;
3386     QDict *options = NULL;
3387     int flags;
3388     int64_t size;
3389     int ret;
3390 
3391     blk = blk_by_name(device);
3392     if (!blk) {
3393         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3394                   "Device '%s' not found", device);
3395         return;
3396     }
3397 
3398     aio_context = blk_get_aio_context(blk);
3399     aio_context_acquire(aio_context);
3400 
3401     if (!blk_is_available(blk)) {
3402         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
3403         goto out;
3404     }
3405     bs = blk_bs(blk);
3406     if (!has_mode) {
3407         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3408     }
3409 
3410     if (!has_format) {
3411         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
3412     }
3413 
3414     flags = bs->open_flags | BDRV_O_RDWR;
3415     source = backing_bs(bs);
3416     if (!source && sync == MIRROR_SYNC_MODE_TOP) {
3417         sync = MIRROR_SYNC_MODE_FULL;
3418     }
3419     if (sync == MIRROR_SYNC_MODE_NONE) {
3420         source = bs;
3421     }
3422 
3423     size = bdrv_getlength(bs);
3424     if (size < 0) {
3425         error_setg_errno(errp, -size, "bdrv_getlength failed");
3426         goto out;
3427     }
3428 
3429     if (has_replaces) {
3430         BlockDriverState *to_replace_bs;
3431         AioContext *replace_aio_context;
3432         int64_t replace_size;
3433 
3434         if (!has_node_name) {
3435             error_setg(errp, "a node-name must be provided when replacing a"
3436                              " named node of the graph");
3437             goto out;
3438         }
3439 
3440         to_replace_bs = check_to_replace_node(bs, replaces, &local_err);
3441 
3442         if (!to_replace_bs) {
3443             error_propagate(errp, local_err);
3444             goto out;
3445         }
3446 
3447         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3448         aio_context_acquire(replace_aio_context);
3449         replace_size = bdrv_getlength(to_replace_bs);
3450         aio_context_release(replace_aio_context);
3451 
3452         if (size != replace_size) {
3453             error_setg(errp, "cannot replace image with a mirror image of "
3454                              "different size");
3455             goto out;
3456         }
3457     }
3458 
3459     if ((sync == MIRROR_SYNC_MODE_FULL || !source)
3460         && mode != NEW_IMAGE_MODE_EXISTING)
3461     {
3462         /* create new image w/o backing file */
3463         assert(format);
3464         bdrv_img_create(target, format,
3465                         NULL, NULL, NULL, size, flags, &local_err, false);
3466     } else {
3467         switch (mode) {
3468         case NEW_IMAGE_MODE_EXISTING:
3469             break;
3470         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3471             /* create new image with backing file */
3472             bdrv_img_create(target, format,
3473                             source->filename,
3474                             source->drv->format_name,
3475                             NULL, size, flags, &local_err, false);
3476             break;
3477         default:
3478             abort();
3479         }
3480     }
3481 
3482     if (local_err) {
3483         error_propagate(errp, local_err);
3484         goto out;
3485     }
3486 
3487     options = qdict_new();
3488     if (has_node_name) {
3489         qdict_put(options, "node-name", qstring_from_str(node_name));
3490     }
3491     if (format) {
3492         qdict_put(options, "driver", qstring_from_str(format));
3493     }
3494 
3495     /* Mirroring takes care of copy-on-write using the source's backing
3496      * file.
3497      */
3498     target_bs = NULL;
3499     ret = bdrv_open(&target_bs, target, NULL, options,
3500                     flags | BDRV_O_NO_BACKING, &local_err);
3501     if (ret < 0) {
3502         error_propagate(errp, local_err);
3503         goto out;
3504     }
3505 
3506     bdrv_set_aio_context(target_bs, aio_context);
3507 
3508     blockdev_mirror_common(bs, target_bs,
3509                            has_replaces, replaces, sync,
3510                            has_speed, speed,
3511                            has_granularity, granularity,
3512                            has_buf_size, buf_size,
3513                            has_on_source_error, on_source_error,
3514                            has_on_target_error, on_target_error,
3515                            has_unmap, unmap,
3516                            &local_err);
3517     if (local_err) {
3518         error_propagate(errp, local_err);
3519         bdrv_unref(target_bs);
3520     }
3521 out:
3522     aio_context_release(aio_context);
3523 }
3524 
3525 void qmp_blockdev_mirror(const char *device, const char *target,
3526                          bool has_replaces, const char *replaces,
3527                          MirrorSyncMode sync,
3528                          bool has_speed, int64_t speed,
3529                          bool has_granularity, uint32_t granularity,
3530                          bool has_buf_size, int64_t buf_size,
3531                          bool has_on_source_error,
3532                          BlockdevOnError on_source_error,
3533                          bool has_on_target_error,
3534                          BlockdevOnError on_target_error,
3535                          Error **errp)
3536 {
3537     BlockDriverState *bs;
3538     BlockBackend *blk;
3539     BlockDriverState *target_bs;
3540     AioContext *aio_context;
3541     Error *local_err = NULL;
3542 
3543     blk = blk_by_name(device);
3544     if (!blk) {
3545         error_setg(errp, "Device '%s' not found", device);
3546         return;
3547     }
3548     bs = blk_bs(blk);
3549 
3550     if (!bs) {
3551         error_setg(errp, "Device '%s' has no media", device);
3552         return;
3553     }
3554 
3555     target_bs = bdrv_lookup_bs(target, target, errp);
3556     if (!target_bs) {
3557         return;
3558     }
3559 
3560     aio_context = bdrv_get_aio_context(bs);
3561     aio_context_acquire(aio_context);
3562 
3563     bdrv_ref(target_bs);
3564     bdrv_set_aio_context(target_bs, aio_context);
3565 
3566     blockdev_mirror_common(bs, target_bs,
3567                            has_replaces, replaces, sync,
3568                            has_speed, speed,
3569                            has_granularity, granularity,
3570                            has_buf_size, buf_size,
3571                            has_on_source_error, on_source_error,
3572                            has_on_target_error, on_target_error,
3573                            true, true,
3574                            &local_err);
3575     if (local_err) {
3576         error_propagate(errp, local_err);
3577         bdrv_unref(target_bs);
3578     }
3579 
3580     aio_context_release(aio_context);
3581 }
3582 
3583 /* Get the block job for a given device name and acquire its AioContext */
3584 static BlockJob *find_block_job(const char *device, AioContext **aio_context,
3585                                 Error **errp)
3586 {
3587     BlockBackend *blk;
3588     BlockDriverState *bs;
3589 
3590     *aio_context = NULL;
3591 
3592     blk = blk_by_name(device);
3593     if (!blk) {
3594         goto notfound;
3595     }
3596 
3597     *aio_context = blk_get_aio_context(blk);
3598     aio_context_acquire(*aio_context);
3599 
3600     if (!blk_is_available(blk)) {
3601         goto notfound;
3602     }
3603     bs = blk_bs(blk);
3604 
3605     if (!bs->job) {
3606         goto notfound;
3607     }
3608 
3609     return bs->job;
3610 
3611 notfound:
3612     error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3613               "No active block job on device '%s'", device);
3614     if (*aio_context) {
3615         aio_context_release(*aio_context);
3616         *aio_context = NULL;
3617     }
3618     return NULL;
3619 }
3620 
3621 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3622 {
3623     AioContext *aio_context;
3624     BlockJob *job = find_block_job(device, &aio_context, errp);
3625 
3626     if (!job) {
3627         return;
3628     }
3629 
3630     block_job_set_speed(job, speed, errp);
3631     aio_context_release(aio_context);
3632 }
3633 
3634 void qmp_block_job_cancel(const char *device,
3635                           bool has_force, bool force, Error **errp)
3636 {
3637     AioContext *aio_context;
3638     BlockJob *job = find_block_job(device, &aio_context, errp);
3639 
3640     if (!job) {
3641         return;
3642     }
3643 
3644     if (!has_force) {
3645         force = false;
3646     }
3647 
3648     if (job->user_paused && !force) {
3649         error_setg(errp, "The block job for device '%s' is currently paused",
3650                    device);
3651         goto out;
3652     }
3653 
3654     trace_qmp_block_job_cancel(job);
3655     block_job_cancel(job);
3656 out:
3657     aio_context_release(aio_context);
3658 }
3659 
3660 void qmp_block_job_pause(const char *device, Error **errp)
3661 {
3662     AioContext *aio_context;
3663     BlockJob *job = find_block_job(device, &aio_context, errp);
3664 
3665     if (!job || job->user_paused) {
3666         return;
3667     }
3668 
3669     job->user_paused = true;
3670     trace_qmp_block_job_pause(job);
3671     block_job_pause(job);
3672     aio_context_release(aio_context);
3673 }
3674 
3675 void qmp_block_job_resume(const char *device, Error **errp)
3676 {
3677     AioContext *aio_context;
3678     BlockJob *job = find_block_job(device, &aio_context, errp);
3679 
3680     if (!job || !job->user_paused) {
3681         return;
3682     }
3683 
3684     job->user_paused = false;
3685     trace_qmp_block_job_resume(job);
3686     block_job_resume(job);
3687     aio_context_release(aio_context);
3688 }
3689 
3690 void qmp_block_job_complete(const char *device, Error **errp)
3691 {
3692     AioContext *aio_context;
3693     BlockJob *job = find_block_job(device, &aio_context, errp);
3694 
3695     if (!job) {
3696         return;
3697     }
3698 
3699     trace_qmp_block_job_complete(job);
3700     block_job_complete(job, errp);
3701     aio_context_release(aio_context);
3702 }
3703 
3704 void qmp_change_backing_file(const char *device,
3705                              const char *image_node_name,
3706                              const char *backing_file,
3707                              Error **errp)
3708 {
3709     BlockBackend *blk;
3710     BlockDriverState *bs = NULL;
3711     AioContext *aio_context;
3712     BlockDriverState *image_bs = NULL;
3713     Error *local_err = NULL;
3714     bool ro;
3715     int open_flags;
3716     int ret;
3717 
3718     blk = blk_by_name(device);
3719     if (!blk) {
3720         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3721                   "Device '%s' not found", device);
3722         return;
3723     }
3724 
3725     aio_context = blk_get_aio_context(blk);
3726     aio_context_acquire(aio_context);
3727 
3728     if (!blk_is_available(blk)) {
3729         error_setg(errp, "Device '%s' has no medium", device);
3730         goto out;
3731     }
3732     bs = blk_bs(blk);
3733 
3734     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3735     if (local_err) {
3736         error_propagate(errp, local_err);
3737         goto out;
3738     }
3739 
3740     if (!image_bs) {
3741         error_setg(errp, "image file not found");
3742         goto out;
3743     }
3744 
3745     if (bdrv_find_base(image_bs) == image_bs) {
3746         error_setg(errp, "not allowing backing file change on an image "
3747                          "without a backing file");
3748         goto out;
3749     }
3750 
3751     /* even though we are not necessarily operating on bs, we need it to
3752      * determine if block ops are currently prohibited on the chain */
3753     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3754         goto out;
3755     }
3756 
3757     /* final sanity check */
3758     if (!bdrv_chain_contains(bs, image_bs)) {
3759         error_setg(errp, "'%s' and image file are not in the same chain",
3760                    device);
3761         goto out;
3762     }
3763 
3764     /* if not r/w, reopen to make r/w */
3765     open_flags = image_bs->open_flags;
3766     ro = bdrv_is_read_only(image_bs);
3767 
3768     if (ro) {
3769         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3770         if (local_err) {
3771             error_propagate(errp, local_err);
3772             goto out;
3773         }
3774     }
3775 
3776     ret = bdrv_change_backing_file(image_bs, backing_file,
3777                                image_bs->drv ? image_bs->drv->format_name : "");
3778 
3779     if (ret < 0) {
3780         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3781                          backing_file);
3782         /* don't exit here, so we can try to restore open flags if
3783          * appropriate */
3784     }
3785 
3786     if (ro) {
3787         bdrv_reopen(image_bs, open_flags, &local_err);
3788         if (local_err) {
3789             error_propagate(errp, local_err); /* will preserve prior errp */
3790         }
3791     }
3792 
3793 out:
3794     aio_context_release(aio_context);
3795 }
3796 
3797 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3798 {
3799     QmpOutputVisitor *ov = qmp_output_visitor_new();
3800     BlockDriverState *bs;
3801     BlockBackend *blk = NULL;
3802     QObject *obj;
3803     QDict *qdict;
3804     Error *local_err = NULL;
3805 
3806     /* TODO Sort it out in raw-posix and drive_new(): Reject aio=native with
3807      * cache.direct=false instead of silently switching to aio=threads, except
3808      * when called from drive_new().
3809      *
3810      * For now, simply forbidding the combination for all drivers will do. */
3811     if (options->has_aio && options->aio == BLOCKDEV_AIO_OPTIONS_NATIVE) {
3812         bool direct = options->has_cache &&
3813                       options->cache->has_direct &&
3814                       options->cache->direct;
3815         if (!direct) {
3816             error_setg(errp, "aio=native requires cache.direct=true");
3817             goto fail;
3818         }
3819     }
3820 
3821     visit_type_BlockdevOptions(qmp_output_get_visitor(ov),
3822                                &options, NULL, &local_err);
3823     if (local_err) {
3824         error_propagate(errp, local_err);
3825         goto fail;
3826     }
3827 
3828     obj = qmp_output_get_qobject(ov);
3829     qdict = qobject_to_qdict(obj);
3830 
3831     qdict_flatten(qdict);
3832 
3833     if (options->has_id) {
3834         blk = blockdev_init(NULL, qdict, &local_err);
3835         if (local_err) {
3836             error_propagate(errp, local_err);
3837             goto fail;
3838         }
3839 
3840         bs = blk_bs(blk);
3841     } else {
3842         if (!qdict_get_try_str(qdict, "node-name")) {
3843             error_setg(errp, "'id' and/or 'node-name' need to be specified for "
3844                        "the root node");
3845             goto fail;
3846         }
3847 
3848         bs = bds_tree_init(qdict, errp);
3849         if (!bs) {
3850             goto fail;
3851         }
3852     }
3853 
3854     if (bs && bdrv_key_required(bs)) {
3855         if (blk) {
3856             blk_unref(blk);
3857         } else {
3858             bdrv_unref(bs);
3859         }
3860         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3861         goto fail;
3862     }
3863 
3864 fail:
3865     qmp_output_visitor_cleanup(ov);
3866 }
3867 
3868 void qmp_x_blockdev_del(bool has_id, const char *id,
3869                         bool has_node_name, const char *node_name, Error **errp)
3870 {
3871     AioContext *aio_context;
3872     BlockBackend *blk;
3873     BlockDriverState *bs;
3874 
3875     if (has_id && has_node_name) {
3876         error_setg(errp, "Only one of id and node-name must be specified");
3877         return;
3878     } else if (!has_id && !has_node_name) {
3879         error_setg(errp, "No block device specified");
3880         return;
3881     }
3882 
3883     if (has_id) {
3884         blk = blk_by_name(id);
3885         if (!blk) {
3886             error_setg(errp, "Cannot find block backend %s", id);
3887             return;
3888         }
3889         if (blk_get_refcnt(blk) > 1) {
3890             error_setg(errp, "Block backend %s is in use", id);
3891             return;
3892         }
3893         bs = blk_bs(blk);
3894         aio_context = blk_get_aio_context(blk);
3895     } else {
3896         bs = bdrv_find_node(node_name);
3897         if (!bs) {
3898             error_setg(errp, "Cannot find node %s", node_name);
3899             return;
3900         }
3901         blk = bs->blk;
3902         if (blk) {
3903             error_setg(errp, "Node %s is in use by %s",
3904                        node_name, blk_name(blk));
3905             return;
3906         }
3907         aio_context = bdrv_get_aio_context(bs);
3908     }
3909 
3910     aio_context_acquire(aio_context);
3911 
3912     if (bs) {
3913         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3914             goto out;
3915         }
3916 
3917         if (bs->refcnt > 1 || !QLIST_EMPTY(&bs->parents)) {
3918             error_setg(errp, "Block device %s is in use",
3919                        bdrv_get_device_or_node_name(bs));
3920             goto out;
3921         }
3922     }
3923 
3924     if (blk) {
3925         blk_unref(blk);
3926     } else {
3927         bdrv_unref(bs);
3928     }
3929 
3930 out:
3931     aio_context_release(aio_context);
3932 }
3933 
3934 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3935 {
3936     BlockJobInfoList *head = NULL, **p_next = &head;
3937     BlockDriverState *bs;
3938 
3939     for (bs = bdrv_next(NULL); bs; bs = bdrv_next(bs)) {
3940         AioContext *aio_context = bdrv_get_aio_context(bs);
3941 
3942         aio_context_acquire(aio_context);
3943 
3944         if (bs->job) {
3945             BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
3946             elem->value = block_job_query(bs->job);
3947             *p_next = elem;
3948             p_next = &elem->next;
3949         }
3950 
3951         aio_context_release(aio_context);
3952     }
3953 
3954     return head;
3955 }
3956 
3957 QemuOptsList qemu_common_drive_opts = {
3958     .name = "drive",
3959     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3960     .desc = {
3961         {
3962             .name = "snapshot",
3963             .type = QEMU_OPT_BOOL,
3964             .help = "enable/disable snapshot mode",
3965         },{
3966             .name = "discard",
3967             .type = QEMU_OPT_STRING,
3968             .help = "discard operation (ignore/off, unmap/on)",
3969         },{
3970             .name = "aio",
3971             .type = QEMU_OPT_STRING,
3972             .help = "host AIO implementation (threads, native)",
3973         },{
3974             .name = "format",
3975             .type = QEMU_OPT_STRING,
3976             .help = "disk format (raw, qcow2, ...)",
3977         },{
3978             .name = "rerror",
3979             .type = QEMU_OPT_STRING,
3980             .help = "read error action",
3981         },{
3982             .name = "werror",
3983             .type = QEMU_OPT_STRING,
3984             .help = "write error action",
3985         },{
3986             .name = "read-only",
3987             .type = QEMU_OPT_BOOL,
3988             .help = "open drive file as read-only",
3989         },{
3990             .name = "throttling.iops-total",
3991             .type = QEMU_OPT_NUMBER,
3992             .help = "limit total I/O operations per second",
3993         },{
3994             .name = "throttling.iops-read",
3995             .type = QEMU_OPT_NUMBER,
3996             .help = "limit read operations per second",
3997         },{
3998             .name = "throttling.iops-write",
3999             .type = QEMU_OPT_NUMBER,
4000             .help = "limit write operations per second",
4001         },{
4002             .name = "throttling.bps-total",
4003             .type = QEMU_OPT_NUMBER,
4004             .help = "limit total bytes per second",
4005         },{
4006             .name = "throttling.bps-read",
4007             .type = QEMU_OPT_NUMBER,
4008             .help = "limit read bytes per second",
4009         },{
4010             .name = "throttling.bps-write",
4011             .type = QEMU_OPT_NUMBER,
4012             .help = "limit write bytes per second",
4013         },{
4014             .name = "throttling.iops-total-max",
4015             .type = QEMU_OPT_NUMBER,
4016             .help = "I/O operations burst",
4017         },{
4018             .name = "throttling.iops-read-max",
4019             .type = QEMU_OPT_NUMBER,
4020             .help = "I/O operations read burst",
4021         },{
4022             .name = "throttling.iops-write-max",
4023             .type = QEMU_OPT_NUMBER,
4024             .help = "I/O operations write burst",
4025         },{
4026             .name = "throttling.bps-total-max",
4027             .type = QEMU_OPT_NUMBER,
4028             .help = "total bytes burst",
4029         },{
4030             .name = "throttling.bps-read-max",
4031             .type = QEMU_OPT_NUMBER,
4032             .help = "total bytes read burst",
4033         },{
4034             .name = "throttling.bps-write-max",
4035             .type = QEMU_OPT_NUMBER,
4036             .help = "total bytes write burst",
4037         },{
4038             .name = "throttling.iops-size",
4039             .type = QEMU_OPT_NUMBER,
4040             .help = "when limiting by iops max size of an I/O in bytes",
4041         },{
4042             .name = "throttling.group",
4043             .type = QEMU_OPT_STRING,
4044             .help = "name of the block throttling group",
4045         },{
4046             .name = "copy-on-read",
4047             .type = QEMU_OPT_BOOL,
4048             .help = "copy read data from backing file into image file",
4049         },{
4050             .name = "detect-zeroes",
4051             .type = QEMU_OPT_STRING,
4052             .help = "try to optimize zero writes (off, on, unmap)",
4053         },{
4054             .name = "stats-account-invalid",
4055             .type = QEMU_OPT_BOOL,
4056             .help = "whether to account for invalid I/O operations "
4057                     "in the statistics",
4058         },{
4059             .name = "stats-account-failed",
4060             .type = QEMU_OPT_BOOL,
4061             .help = "whether to account for failed I/O operations "
4062                     "in the statistics",
4063         },
4064         { /* end of list */ }
4065     },
4066 };
4067 
4068 static QemuOptsList qemu_root_bds_opts = {
4069     .name = "root-bds",
4070     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
4071     .desc = {
4072         {
4073             .name = "discard",
4074             .type = QEMU_OPT_STRING,
4075             .help = "discard operation (ignore/off, unmap/on)",
4076         },{
4077             .name = "aio",
4078             .type = QEMU_OPT_STRING,
4079             .help = "host AIO implementation (threads, native)",
4080         },{
4081             .name = "read-only",
4082             .type = QEMU_OPT_BOOL,
4083             .help = "open drive file as read-only",
4084         },{
4085             .name = "copy-on-read",
4086             .type = QEMU_OPT_BOOL,
4087             .help = "copy read data from backing file into image file",
4088         },{
4089             .name = "detect-zeroes",
4090             .type = QEMU_OPT_STRING,
4091             .help = "try to optimize zero writes (off, on, unmap)",
4092         },
4093         { /* end of list */ }
4094     },
4095 };
4096 
4097 QemuOptsList qemu_drive_opts = {
4098     .name = "drive",
4099     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4100     .desc = {
4101         /*
4102          * no elements => accept any params
4103          * validation will happen later
4104          */
4105         { /* end of list */ }
4106     },
4107 };
4108