xref: /openbmc/qemu/blockdev.c (revision 39164c13)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/throttle-groups.h"
39 #include "monitor/monitor.h"
40 #include "qemu/error-report.h"
41 #include "qemu/option.h"
42 #include "qemu/config-file.h"
43 #include "qapi/qmp/types.h"
44 #include "qapi-visit.h"
45 #include "qapi/qmp/qerror.h"
46 #include "qapi/qobject-output-visitor.h"
47 #include "qapi/util.h"
48 #include "sysemu/sysemu.h"
49 #include "block/block_int.h"
50 #include "qmp-commands.h"
51 #include "block/trace.h"
52 #include "sysemu/arch_init.h"
53 #include "qemu/cutils.h"
54 #include "qemu/help_option.h"
55 #include "qemu/throttle-options.h"
56 
57 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
58     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
59 
60 static int do_open_tray(const char *blk_name, const char *qdev_id,
61                         bool force, Error **errp);
62 
63 static const char *const if_name[IF_COUNT] = {
64     [IF_NONE] = "none",
65     [IF_IDE] = "ide",
66     [IF_SCSI] = "scsi",
67     [IF_FLOPPY] = "floppy",
68     [IF_PFLASH] = "pflash",
69     [IF_MTD] = "mtd",
70     [IF_SD] = "sd",
71     [IF_VIRTIO] = "virtio",
72     [IF_XEN] = "xen",
73 };
74 
75 static int if_max_devs[IF_COUNT] = {
76     /*
77      * Do not change these numbers!  They govern how drive option
78      * index maps to unit and bus.  That mapping is ABI.
79      *
80      * All controllers used to implement if=T drives need to support
81      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
82      * Otherwise, some index values map to "impossible" bus, unit
83      * values.
84      *
85      * For instance, if you change [IF_SCSI] to 255, -drive
86      * if=scsi,index=12 no longer means bus=1,unit=5, but
87      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
88      * the drive can't be set up.  Regression.
89      */
90     [IF_IDE] = 2,
91     [IF_SCSI] = 7,
92 };
93 
94 /**
95  * Boards may call this to offer board-by-board overrides
96  * of the default, global values.
97  */
98 void override_max_devs(BlockInterfaceType type, int max_devs)
99 {
100     BlockBackend *blk;
101     DriveInfo *dinfo;
102 
103     if (max_devs <= 0) {
104         return;
105     }
106 
107     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
108         dinfo = blk_legacy_dinfo(blk);
109         if (dinfo->type == type) {
110             fprintf(stderr, "Cannot override units-per-bus property of"
111                     " the %s interface, because a drive of that type has"
112                     " already been added.\n", if_name[type]);
113             g_assert_not_reached();
114         }
115     }
116 
117     if_max_devs[type] = max_devs;
118 }
119 
120 /*
121  * We automatically delete the drive when a device using it gets
122  * unplugged.  Questionable feature, but we can't just drop it.
123  * Device models call blockdev_mark_auto_del() to schedule the
124  * automatic deletion, and generic qdev code calls blockdev_auto_del()
125  * when deletion is actually safe.
126  */
127 void blockdev_mark_auto_del(BlockBackend *blk)
128 {
129     DriveInfo *dinfo = blk_legacy_dinfo(blk);
130     BlockDriverState *bs = blk_bs(blk);
131     AioContext *aio_context;
132 
133     if (!dinfo) {
134         return;
135     }
136 
137     if (bs) {
138         aio_context = bdrv_get_aio_context(bs);
139         aio_context_acquire(aio_context);
140 
141         if (bs->job) {
142             block_job_cancel(bs->job);
143         }
144 
145         aio_context_release(aio_context);
146     }
147 
148     dinfo->auto_del = 1;
149 }
150 
151 void blockdev_auto_del(BlockBackend *blk)
152 {
153     DriveInfo *dinfo = blk_legacy_dinfo(blk);
154 
155     if (dinfo && dinfo->auto_del) {
156         monitor_remove_blk(blk);
157         blk_unref(blk);
158     }
159 }
160 
161 /**
162  * Returns the current mapping of how many units per bus
163  * a particular interface can support.
164  *
165  *  A positive integer indicates n units per bus.
166  *  0 implies the mapping has not been established.
167  * -1 indicates an invalid BlockInterfaceType was given.
168  */
169 int drive_get_max_devs(BlockInterfaceType type)
170 {
171     if (type >= IF_IDE && type < IF_COUNT) {
172         return if_max_devs[type];
173     }
174 
175     return -1;
176 }
177 
178 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
179 {
180     int max_devs = if_max_devs[type];
181     return max_devs ? index / max_devs : 0;
182 }
183 
184 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
185 {
186     int max_devs = if_max_devs[type];
187     return max_devs ? index % max_devs : index;
188 }
189 
190 QemuOpts *drive_def(const char *optstr)
191 {
192     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
193 }
194 
195 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
196                     const char *optstr)
197 {
198     QemuOpts *opts;
199 
200     opts = drive_def(optstr);
201     if (!opts) {
202         return NULL;
203     }
204     if (type != IF_DEFAULT) {
205         qemu_opt_set(opts, "if", if_name[type], &error_abort);
206     }
207     if (index >= 0) {
208         qemu_opt_set_number(opts, "index", index, &error_abort);
209     }
210     if (file)
211         qemu_opt_set(opts, "file", file, &error_abort);
212     return opts;
213 }
214 
215 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
216 {
217     BlockBackend *blk;
218     DriveInfo *dinfo;
219 
220     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
221         dinfo = blk_legacy_dinfo(blk);
222         if (dinfo && dinfo->type == type
223             && dinfo->bus == bus && dinfo->unit == unit) {
224             return dinfo;
225         }
226     }
227 
228     return NULL;
229 }
230 
231 void drive_check_orphaned(void)
232 {
233     BlockBackend *blk;
234     DriveInfo *dinfo;
235     Location loc;
236     bool orphans = false;
237 
238     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
239         dinfo = blk_legacy_dinfo(blk);
240         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
241             dinfo->type != IF_NONE) {
242             loc_push_none(&loc);
243             qemu_opts_loc_restore(dinfo->opts);
244             error_report("machine type does not support"
245                          " if=%s,bus=%d,unit=%d",
246                          if_name[dinfo->type], dinfo->bus, dinfo->unit);
247             loc_pop(&loc);
248             orphans = true;
249         }
250     }
251 
252     if (orphans) {
253         exit(1);
254     }
255 }
256 
257 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
258 {
259     return drive_get(type,
260                      drive_index_to_bus_id(type, index),
261                      drive_index_to_unit_id(type, index));
262 }
263 
264 int drive_get_max_bus(BlockInterfaceType type)
265 {
266     int max_bus;
267     BlockBackend *blk;
268     DriveInfo *dinfo;
269 
270     max_bus = -1;
271     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
272         dinfo = blk_legacy_dinfo(blk);
273         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
274             max_bus = dinfo->bus;
275         }
276     }
277     return max_bus;
278 }
279 
280 /* Get a block device.  This should only be used for single-drive devices
281    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
282    appropriate bus.  */
283 DriveInfo *drive_get_next(BlockInterfaceType type)
284 {
285     static int next_block_unit[IF_COUNT];
286 
287     return drive_get(type, 0, next_block_unit[type]++);
288 }
289 
290 static void bdrv_format_print(void *opaque, const char *name)
291 {
292     error_printf(" %s", name);
293 }
294 
295 typedef struct {
296     QEMUBH *bh;
297     BlockDriverState *bs;
298 } BDRVPutRefBH;
299 
300 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
301 {
302     if (!strcmp(buf, "ignore")) {
303         return BLOCKDEV_ON_ERROR_IGNORE;
304     } else if (!is_read && !strcmp(buf, "enospc")) {
305         return BLOCKDEV_ON_ERROR_ENOSPC;
306     } else if (!strcmp(buf, "stop")) {
307         return BLOCKDEV_ON_ERROR_STOP;
308     } else if (!strcmp(buf, "report")) {
309         return BLOCKDEV_ON_ERROR_REPORT;
310     } else {
311         error_setg(errp, "'%s' invalid %s error action",
312                    buf, is_read ? "read" : "write");
313         return -1;
314     }
315 }
316 
317 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
318                                   Error **errp)
319 {
320     const QListEntry *entry;
321     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
322         switch (qobject_type(entry->value)) {
323 
324         case QTYPE_QSTRING: {
325             unsigned long long length;
326             const char *str = qstring_get_str(qobject_to_qstring(entry->value));
327             if (parse_uint_full(str, &length, 10) == 0 &&
328                 length > 0 && length <= UINT_MAX) {
329                 block_acct_add_interval(stats, (unsigned) length);
330             } else {
331                 error_setg(errp, "Invalid interval length: %s", str);
332                 return false;
333             }
334             break;
335         }
336 
337         case QTYPE_QINT: {
338             int64_t length = qint_get_int(qobject_to_qint(entry->value));
339             if (length > 0 && length <= UINT_MAX) {
340                 block_acct_add_interval(stats, (unsigned) length);
341             } else {
342                 error_setg(errp, "Invalid interval length: %" PRId64, length);
343                 return false;
344             }
345             break;
346         }
347 
348         default:
349             error_setg(errp, "The specification of stats-intervals is invalid");
350             return false;
351         }
352     }
353     return true;
354 }
355 
356 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
357 
358 /* All parameters but @opts are optional and may be set to NULL. */
359 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
360     const char **throttling_group, ThrottleConfig *throttle_cfg,
361     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
362 {
363     Error *local_error = NULL;
364     const char *aio;
365 
366     if (bdrv_flags) {
367         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
368             *bdrv_flags |= BDRV_O_COPY_ON_READ;
369         }
370 
371         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
372             if (!strcmp(aio, "native")) {
373                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
374             } else if (!strcmp(aio, "threads")) {
375                 /* this is the default */
376             } else {
377                error_setg(errp, "invalid aio option");
378                return;
379             }
380         }
381     }
382 
383     /* disk I/O throttling */
384     if (throttling_group) {
385         *throttling_group = qemu_opt_get(opts, "throttling.group");
386     }
387 
388     if (throttle_cfg) {
389         throttle_config_init(throttle_cfg);
390         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
391             qemu_opt_get_number(opts, "throttling.bps-total", 0);
392         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
393             qemu_opt_get_number(opts, "throttling.bps-read", 0);
394         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
395             qemu_opt_get_number(opts, "throttling.bps-write", 0);
396         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
397             qemu_opt_get_number(opts, "throttling.iops-total", 0);
398         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
399             qemu_opt_get_number(opts, "throttling.iops-read", 0);
400         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
401             qemu_opt_get_number(opts, "throttling.iops-write", 0);
402 
403         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
404             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
405         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
406             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
407         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
408             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
409         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
410             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
411         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
412             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
413         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
414             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
415 
416         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
417             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
418         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
419             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
420         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
421             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
422         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
423             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
424         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
425             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
426         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
427             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
428 
429         throttle_cfg->op_size =
430             qemu_opt_get_number(opts, "throttling.iops-size", 0);
431 
432         if (!throttle_is_valid(throttle_cfg, errp)) {
433             return;
434         }
435     }
436 
437     if (detect_zeroes) {
438         *detect_zeroes =
439             qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
440                             qemu_opt_get(opts, "detect-zeroes"),
441                             BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
442                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
443                             &local_error);
444         if (local_error) {
445             error_propagate(errp, local_error);
446             return;
447         }
448     }
449 }
450 
451 /* Takes the ownership of bs_opts */
452 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
453                                    Error **errp)
454 {
455     const char *buf;
456     int bdrv_flags = 0;
457     int on_read_error, on_write_error;
458     bool account_invalid, account_failed;
459     bool writethrough, read_only;
460     BlockBackend *blk;
461     BlockDriverState *bs;
462     ThrottleConfig cfg;
463     int snapshot = 0;
464     Error *error = NULL;
465     QemuOpts *opts;
466     QDict *interval_dict = NULL;
467     QList *interval_list = NULL;
468     const char *id;
469     BlockdevDetectZeroesOptions detect_zeroes =
470         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
471     const char *throttling_group = NULL;
472 
473     /* Check common options by copying from bs_opts to opts, all other options
474      * stay in bs_opts for processing by bdrv_open(). */
475     id = qdict_get_try_str(bs_opts, "id");
476     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
477     if (error) {
478         error_propagate(errp, error);
479         goto err_no_opts;
480     }
481 
482     qemu_opts_absorb_qdict(opts, bs_opts, &error);
483     if (error) {
484         error_propagate(errp, error);
485         goto early_err;
486     }
487 
488     if (id) {
489         qdict_del(bs_opts, "id");
490     }
491 
492     /* extract parameters */
493     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
494 
495     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
496     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
497 
498     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
499 
500     id = qemu_opts_id(opts);
501 
502     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
503     qdict_array_split(interval_dict, &interval_list);
504 
505     if (qdict_size(interval_dict) != 0) {
506         error_setg(errp, "Invalid option stats-intervals.%s",
507                    qdict_first(interval_dict)->key);
508         goto early_err;
509     }
510 
511     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
512                                     &detect_zeroes, &error);
513     if (error) {
514         error_propagate(errp, error);
515         goto early_err;
516     }
517 
518     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
519         if (is_help_option(buf)) {
520             error_printf("Supported formats:");
521             bdrv_iterate_format(bdrv_format_print, NULL);
522             error_printf("\n");
523             goto early_err;
524         }
525 
526         if (qdict_haskey(bs_opts, "driver")) {
527             error_setg(errp, "Cannot specify both 'driver' and 'format'");
528             goto early_err;
529         }
530         qdict_put(bs_opts, "driver", qstring_from_str(buf));
531     }
532 
533     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
534     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
535         on_write_error = parse_block_error_action(buf, 0, &error);
536         if (error) {
537             error_propagate(errp, error);
538             goto early_err;
539         }
540     }
541 
542     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
543     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
544         on_read_error = parse_block_error_action(buf, 1, &error);
545         if (error) {
546             error_propagate(errp, error);
547             goto early_err;
548         }
549     }
550 
551     if (snapshot) {
552         bdrv_flags |= BDRV_O_SNAPSHOT;
553     }
554 
555     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
556 
557     /* init */
558     if ((!file || !*file) && !qdict_size(bs_opts)) {
559         BlockBackendRootState *blk_rs;
560 
561         blk = blk_new();
562         blk_rs = blk_get_root_state(blk);
563         blk_rs->open_flags    = bdrv_flags;
564         blk_rs->read_only     = read_only;
565         blk_rs->detect_zeroes = detect_zeroes;
566 
567         QDECREF(bs_opts);
568     } else {
569         if (file && !*file) {
570             file = NULL;
571         }
572 
573         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
574          * with other callers) rather than what we want as the real defaults.
575          * Apply the defaults here instead. */
576         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
577         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
578         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
579                               read_only ? "on" : "off");
580         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
581 
582         if (runstate_check(RUN_STATE_INMIGRATE)) {
583             bdrv_flags |= BDRV_O_INACTIVE;
584         }
585 
586         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
587         if (!blk) {
588             goto err_no_bs_opts;
589         }
590         bs = blk_bs(blk);
591 
592         bs->detect_zeroes = detect_zeroes;
593 
594         if (bdrv_key_required(bs)) {
595             autostart = 0;
596         }
597 
598         block_acct_init(blk_get_stats(blk), account_invalid, account_failed);
599 
600         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
601             blk_unref(blk);
602             blk = NULL;
603             goto err_no_bs_opts;
604         }
605     }
606 
607     /* disk I/O throttling */
608     if (throttle_enabled(&cfg)) {
609         if (!throttling_group) {
610             throttling_group = id;
611         }
612         blk_io_limits_enable(blk, throttling_group);
613         blk_set_io_limits(blk, &cfg);
614     }
615 
616     blk_set_enable_write_cache(blk, !writethrough);
617     blk_set_on_error(blk, on_read_error, on_write_error);
618 
619     if (!monitor_add_blk(blk, id, errp)) {
620         blk_unref(blk);
621         blk = NULL;
622         goto err_no_bs_opts;
623     }
624 
625 err_no_bs_opts:
626     qemu_opts_del(opts);
627     QDECREF(interval_dict);
628     QDECREF(interval_list);
629     return blk;
630 
631 early_err:
632     qemu_opts_del(opts);
633     QDECREF(interval_dict);
634     QDECREF(interval_list);
635 err_no_opts:
636     QDECREF(bs_opts);
637     return NULL;
638 }
639 
640 /* Takes the ownership of bs_opts */
641 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
642 {
643     int bdrv_flags = 0;
644 
645     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
646      * with other callers) rather than what we want as the real defaults.
647      * Apply the defaults here instead. */
648     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
649     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
650     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
651 
652     if (runstate_check(RUN_STATE_INMIGRATE)) {
653         bdrv_flags |= BDRV_O_INACTIVE;
654     }
655 
656     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
657 }
658 
659 void blockdev_close_all_bdrv_states(void)
660 {
661     BlockDriverState *bs, *next_bs;
662 
663     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
664         AioContext *ctx = bdrv_get_aio_context(bs);
665 
666         aio_context_acquire(ctx);
667         bdrv_unref(bs);
668         aio_context_release(ctx);
669     }
670 }
671 
672 /* Iterates over the list of monitor-owned BlockDriverStates */
673 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
674 {
675     return bs ? QTAILQ_NEXT(bs, monitor_list)
676               : QTAILQ_FIRST(&monitor_bdrv_states);
677 }
678 
679 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
680                             Error **errp)
681 {
682     const char *value;
683 
684     value = qemu_opt_get(opts, from);
685     if (value) {
686         if (qemu_opt_find(opts, to)) {
687             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
688                        "same time", to, from);
689             return;
690         }
691     }
692 
693     /* rename all items in opts */
694     while ((value = qemu_opt_get(opts, from))) {
695         qemu_opt_set(opts, to, value, &error_abort);
696         qemu_opt_unset(opts, from);
697     }
698 }
699 
700 QemuOptsList qemu_legacy_drive_opts = {
701     .name = "drive",
702     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
703     .desc = {
704         {
705             .name = "bus",
706             .type = QEMU_OPT_NUMBER,
707             .help = "bus number",
708         },{
709             .name = "unit",
710             .type = QEMU_OPT_NUMBER,
711             .help = "unit number (i.e. lun for scsi)",
712         },{
713             .name = "index",
714             .type = QEMU_OPT_NUMBER,
715             .help = "index number",
716         },{
717             .name = "media",
718             .type = QEMU_OPT_STRING,
719             .help = "media type (disk, cdrom)",
720         },{
721             .name = "if",
722             .type = QEMU_OPT_STRING,
723             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
724         },{
725             .name = "cyls",
726             .type = QEMU_OPT_NUMBER,
727             .help = "number of cylinders (ide disk geometry)",
728         },{
729             .name = "heads",
730             .type = QEMU_OPT_NUMBER,
731             .help = "number of heads (ide disk geometry)",
732         },{
733             .name = "secs",
734             .type = QEMU_OPT_NUMBER,
735             .help = "number of sectors (ide disk geometry)",
736         },{
737             .name = "trans",
738             .type = QEMU_OPT_STRING,
739             .help = "chs translation (auto, lba, none)",
740         },{
741             .name = "boot",
742             .type = QEMU_OPT_BOOL,
743             .help = "(deprecated, ignored)",
744         },{
745             .name = "addr",
746             .type = QEMU_OPT_STRING,
747             .help = "pci address (virtio only)",
748         },{
749             .name = "serial",
750             .type = QEMU_OPT_STRING,
751             .help = "disk serial number",
752         },{
753             .name = "file",
754             .type = QEMU_OPT_STRING,
755             .help = "file name",
756         },
757 
758         /* Options that are passed on, but have special semantics with -drive */
759         {
760             .name = BDRV_OPT_READ_ONLY,
761             .type = QEMU_OPT_BOOL,
762             .help = "open drive file as read-only",
763         },{
764             .name = "rerror",
765             .type = QEMU_OPT_STRING,
766             .help = "read error action",
767         },{
768             .name = "werror",
769             .type = QEMU_OPT_STRING,
770             .help = "write error action",
771         },{
772             .name = "copy-on-read",
773             .type = QEMU_OPT_BOOL,
774             .help = "copy read data from backing file into image file",
775         },
776 
777         { /* end of list */ }
778     },
779 };
780 
781 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
782 {
783     const char *value;
784     BlockBackend *blk;
785     DriveInfo *dinfo = NULL;
786     QDict *bs_opts;
787     QemuOpts *legacy_opts;
788     DriveMediaType media = MEDIA_DISK;
789     BlockInterfaceType type;
790     int cyls, heads, secs, translation;
791     int max_devs, bus_id, unit_id, index;
792     const char *devaddr;
793     const char *werror, *rerror;
794     bool read_only = false;
795     bool copy_on_read;
796     const char *serial;
797     const char *filename;
798     Error *local_err = NULL;
799     int i;
800 
801     /* Change legacy command line options into QMP ones */
802     static const struct {
803         const char *from;
804         const char *to;
805     } opt_renames[] = {
806         { "iops",           "throttling.iops-total" },
807         { "iops_rd",        "throttling.iops-read" },
808         { "iops_wr",        "throttling.iops-write" },
809 
810         { "bps",            "throttling.bps-total" },
811         { "bps_rd",         "throttling.bps-read" },
812         { "bps_wr",         "throttling.bps-write" },
813 
814         { "iops_max",       "throttling.iops-total-max" },
815         { "iops_rd_max",    "throttling.iops-read-max" },
816         { "iops_wr_max",    "throttling.iops-write-max" },
817 
818         { "bps_max",        "throttling.bps-total-max" },
819         { "bps_rd_max",     "throttling.bps-read-max" },
820         { "bps_wr_max",     "throttling.bps-write-max" },
821 
822         { "iops_size",      "throttling.iops-size" },
823 
824         { "group",          "throttling.group" },
825 
826         { "readonly",       BDRV_OPT_READ_ONLY },
827     };
828 
829     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
830         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
831                         &local_err);
832         if (local_err) {
833             error_report_err(local_err);
834             return NULL;
835         }
836     }
837 
838     value = qemu_opt_get(all_opts, "cache");
839     if (value) {
840         int flags = 0;
841         bool writethrough;
842 
843         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
844             error_report("invalid cache option");
845             return NULL;
846         }
847 
848         /* Specific options take precedence */
849         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
850             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
851                               !writethrough, &error_abort);
852         }
853         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
854             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
855                               !!(flags & BDRV_O_NOCACHE), &error_abort);
856         }
857         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
858             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
859                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
860         }
861         qemu_opt_unset(all_opts, "cache");
862     }
863 
864     /* Get a QDict for processing the options */
865     bs_opts = qdict_new();
866     qemu_opts_to_qdict(all_opts, bs_opts);
867 
868     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
869                                    &error_abort);
870     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
871     if (local_err) {
872         error_report_err(local_err);
873         goto fail;
874     }
875 
876     /* Deprecated option boot=[on|off] */
877     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
878         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
879                 "ignored. Future versions will reject this parameter. Please "
880                 "update your scripts.\n");
881     }
882 
883     /* Media type */
884     value = qemu_opt_get(legacy_opts, "media");
885     if (value) {
886         if (!strcmp(value, "disk")) {
887             media = MEDIA_DISK;
888         } else if (!strcmp(value, "cdrom")) {
889             media = MEDIA_CDROM;
890             read_only = true;
891         } else {
892             error_report("'%s' invalid media", value);
893             goto fail;
894         }
895     }
896 
897     /* copy-on-read is disabled with a warning for read-only devices */
898     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
899     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
900 
901     if (read_only && copy_on_read) {
902         error_report("warning: disabling copy-on-read on read-only drive");
903         copy_on_read = false;
904     }
905 
906     qdict_put(bs_opts, BDRV_OPT_READ_ONLY,
907               qstring_from_str(read_only ? "on" : "off"));
908     qdict_put(bs_opts, "copy-on-read",
909               qstring_from_str(copy_on_read ? "on" :"off"));
910 
911     /* Controller type */
912     value = qemu_opt_get(legacy_opts, "if");
913     if (value) {
914         for (type = 0;
915              type < IF_COUNT && strcmp(value, if_name[type]);
916              type++) {
917         }
918         if (type == IF_COUNT) {
919             error_report("unsupported bus type '%s'", value);
920             goto fail;
921         }
922     } else {
923         type = block_default_type;
924     }
925 
926     /* Geometry */
927     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
928     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
929     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
930 
931     if (cyls || heads || secs) {
932         if (cyls < 1) {
933             error_report("invalid physical cyls number");
934             goto fail;
935         }
936         if (heads < 1) {
937             error_report("invalid physical heads number");
938             goto fail;
939         }
940         if (secs < 1) {
941             error_report("invalid physical secs number");
942             goto fail;
943         }
944     }
945 
946     translation = BIOS_ATA_TRANSLATION_AUTO;
947     value = qemu_opt_get(legacy_opts, "trans");
948     if (value != NULL) {
949         if (!cyls) {
950             error_report("'%s' trans must be used with cyls, heads and secs",
951                          value);
952             goto fail;
953         }
954         if (!strcmp(value, "none")) {
955             translation = BIOS_ATA_TRANSLATION_NONE;
956         } else if (!strcmp(value, "lba")) {
957             translation = BIOS_ATA_TRANSLATION_LBA;
958         } else if (!strcmp(value, "large")) {
959             translation = BIOS_ATA_TRANSLATION_LARGE;
960         } else if (!strcmp(value, "rechs")) {
961             translation = BIOS_ATA_TRANSLATION_RECHS;
962         } else if (!strcmp(value, "auto")) {
963             translation = BIOS_ATA_TRANSLATION_AUTO;
964         } else {
965             error_report("'%s' invalid translation type", value);
966             goto fail;
967         }
968     }
969 
970     if (media == MEDIA_CDROM) {
971         if (cyls || secs || heads) {
972             error_report("CHS can't be set with media=cdrom");
973             goto fail;
974         }
975     }
976 
977     /* Device address specified by bus/unit or index.
978      * If none was specified, try to find the first free one. */
979     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
980     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
981     index   = qemu_opt_get_number(legacy_opts, "index", -1);
982 
983     max_devs = if_max_devs[type];
984 
985     if (index != -1) {
986         if (bus_id != 0 || unit_id != -1) {
987             error_report("index cannot be used with bus and unit");
988             goto fail;
989         }
990         bus_id = drive_index_to_bus_id(type, index);
991         unit_id = drive_index_to_unit_id(type, index);
992     }
993 
994     if (unit_id == -1) {
995        unit_id = 0;
996        while (drive_get(type, bus_id, unit_id) != NULL) {
997            unit_id++;
998            if (max_devs && unit_id >= max_devs) {
999                unit_id -= max_devs;
1000                bus_id++;
1001            }
1002        }
1003     }
1004 
1005     if (max_devs && unit_id >= max_devs) {
1006         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1007         goto fail;
1008     }
1009 
1010     if (drive_get(type, bus_id, unit_id) != NULL) {
1011         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1012                      bus_id, unit_id, index);
1013         goto fail;
1014     }
1015 
1016     /* Serial number */
1017     serial = qemu_opt_get(legacy_opts, "serial");
1018 
1019     /* no id supplied -> create one */
1020     if (qemu_opts_id(all_opts) == NULL) {
1021         char *new_id;
1022         const char *mediastr = "";
1023         if (type == IF_IDE || type == IF_SCSI) {
1024             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1025         }
1026         if (max_devs) {
1027             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1028                                      mediastr, unit_id);
1029         } else {
1030             new_id = g_strdup_printf("%s%s%i", if_name[type],
1031                                      mediastr, unit_id);
1032         }
1033         qdict_put(bs_opts, "id", qstring_from_str(new_id));
1034         g_free(new_id);
1035     }
1036 
1037     /* Add virtio block device */
1038     devaddr = qemu_opt_get(legacy_opts, "addr");
1039     if (devaddr && type != IF_VIRTIO) {
1040         error_report("addr is not supported by this bus type");
1041         goto fail;
1042     }
1043 
1044     if (type == IF_VIRTIO) {
1045         QemuOpts *devopts;
1046         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1047                                    &error_abort);
1048         if (arch_type == QEMU_ARCH_S390X) {
1049             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1050         } else {
1051             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1052         }
1053         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1054                      &error_abort);
1055         if (devaddr) {
1056             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1057         }
1058     }
1059 
1060     filename = qemu_opt_get(legacy_opts, "file");
1061 
1062     /* Check werror/rerror compatibility with if=... */
1063     werror = qemu_opt_get(legacy_opts, "werror");
1064     if (werror != NULL) {
1065         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1066             type != IF_NONE) {
1067             error_report("werror is not supported by this bus type");
1068             goto fail;
1069         }
1070         qdict_put(bs_opts, "werror", qstring_from_str(werror));
1071     }
1072 
1073     rerror = qemu_opt_get(legacy_opts, "rerror");
1074     if (rerror != NULL) {
1075         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1076             type != IF_NONE) {
1077             error_report("rerror is not supported by this bus type");
1078             goto fail;
1079         }
1080         qdict_put(bs_opts, "rerror", qstring_from_str(rerror));
1081     }
1082 
1083     /* Actual block device init: Functionality shared with blockdev-add */
1084     blk = blockdev_init(filename, bs_opts, &local_err);
1085     bs_opts = NULL;
1086     if (!blk) {
1087         if (local_err) {
1088             error_report_err(local_err);
1089         }
1090         goto fail;
1091     } else {
1092         assert(!local_err);
1093     }
1094 
1095     /* Create legacy DriveInfo */
1096     dinfo = g_malloc0(sizeof(*dinfo));
1097     dinfo->opts = all_opts;
1098 
1099     dinfo->cyls = cyls;
1100     dinfo->heads = heads;
1101     dinfo->secs = secs;
1102     dinfo->trans = translation;
1103 
1104     dinfo->type = type;
1105     dinfo->bus = bus_id;
1106     dinfo->unit = unit_id;
1107     dinfo->devaddr = devaddr;
1108     dinfo->serial = g_strdup(serial);
1109 
1110     blk_set_legacy_dinfo(blk, dinfo);
1111 
1112     switch(type) {
1113     case IF_IDE:
1114     case IF_SCSI:
1115     case IF_XEN:
1116     case IF_NONE:
1117         dinfo->media_cd = media == MEDIA_CDROM;
1118         break;
1119     default:
1120         break;
1121     }
1122 
1123 fail:
1124     qemu_opts_del(legacy_opts);
1125     QDECREF(bs_opts);
1126     return dinfo;
1127 }
1128 
1129 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1130 {
1131     BlockDriverState *bs;
1132 
1133     bs = bdrv_lookup_bs(name, name, errp);
1134     if (bs == NULL) {
1135         return NULL;
1136     }
1137 
1138     if (!bdrv_is_root_node(bs)) {
1139         error_setg(errp, "Need a root block node");
1140         return NULL;
1141     }
1142 
1143     if (!bdrv_is_inserted(bs)) {
1144         error_setg(errp, "Device has no medium");
1145         return NULL;
1146     }
1147 
1148     return bs;
1149 }
1150 
1151 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1152                                  Error **errp)
1153 {
1154     BlockBackend *blk;
1155 
1156     if (!blk_name == !qdev_id) {
1157         error_setg(errp, "Need exactly one of 'device' and 'id'");
1158         return NULL;
1159     }
1160 
1161     if (qdev_id) {
1162         blk = blk_by_qdev_id(qdev_id, errp);
1163     } else {
1164         blk = blk_by_name(blk_name);
1165         if (blk == NULL) {
1166             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1167                       "Device '%s' not found", blk_name);
1168         }
1169     }
1170 
1171     return blk;
1172 }
1173 
1174 void hmp_commit(Monitor *mon, const QDict *qdict)
1175 {
1176     const char *device = qdict_get_str(qdict, "device");
1177     BlockBackend *blk;
1178     int ret;
1179 
1180     if (!strcmp(device, "all")) {
1181         ret = blk_commit_all();
1182     } else {
1183         BlockDriverState *bs;
1184         AioContext *aio_context;
1185 
1186         blk = blk_by_name(device);
1187         if (!blk) {
1188             monitor_printf(mon, "Device '%s' not found\n", device);
1189             return;
1190         }
1191         if (!blk_is_available(blk)) {
1192             monitor_printf(mon, "Device '%s' has no medium\n", device);
1193             return;
1194         }
1195 
1196         bs = blk_bs(blk);
1197         aio_context = bdrv_get_aio_context(bs);
1198         aio_context_acquire(aio_context);
1199 
1200         ret = bdrv_commit(bs);
1201 
1202         aio_context_release(aio_context);
1203     }
1204     if (ret < 0) {
1205         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1206                        strerror(-ret));
1207     }
1208 }
1209 
1210 static void blockdev_do_action(TransactionAction *action, Error **errp)
1211 {
1212     TransactionActionList list;
1213 
1214     list.value = action;
1215     list.next = NULL;
1216     qmp_transaction(&list, false, NULL, errp);
1217 }
1218 
1219 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1220                                 bool has_node_name, const char *node_name,
1221                                 const char *snapshot_file,
1222                                 bool has_snapshot_node_name,
1223                                 const char *snapshot_node_name,
1224                                 bool has_format, const char *format,
1225                                 bool has_mode, NewImageMode mode, Error **errp)
1226 {
1227     BlockdevSnapshotSync snapshot = {
1228         .has_device = has_device,
1229         .device = (char *) device,
1230         .has_node_name = has_node_name,
1231         .node_name = (char *) node_name,
1232         .snapshot_file = (char *) snapshot_file,
1233         .has_snapshot_node_name = has_snapshot_node_name,
1234         .snapshot_node_name = (char *) snapshot_node_name,
1235         .has_format = has_format,
1236         .format = (char *) format,
1237         .has_mode = has_mode,
1238         .mode = mode,
1239     };
1240     TransactionAction action = {
1241         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1242         .u.blockdev_snapshot_sync.data = &snapshot,
1243     };
1244     blockdev_do_action(&action, errp);
1245 }
1246 
1247 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1248                            Error **errp)
1249 {
1250     BlockdevSnapshot snapshot_data = {
1251         .node = (char *) node,
1252         .overlay = (char *) overlay
1253     };
1254     TransactionAction action = {
1255         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1256         .u.blockdev_snapshot.data = &snapshot_data,
1257     };
1258     blockdev_do_action(&action, errp);
1259 }
1260 
1261 void qmp_blockdev_snapshot_internal_sync(const char *device,
1262                                          const char *name,
1263                                          Error **errp)
1264 {
1265     BlockdevSnapshotInternal snapshot = {
1266         .device = (char *) device,
1267         .name = (char *) name
1268     };
1269     TransactionAction action = {
1270         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1271         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1272     };
1273     blockdev_do_action(&action, errp);
1274 }
1275 
1276 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1277                                                          bool has_id,
1278                                                          const char *id,
1279                                                          bool has_name,
1280                                                          const char *name,
1281                                                          Error **errp)
1282 {
1283     BlockDriverState *bs;
1284     AioContext *aio_context;
1285     QEMUSnapshotInfo sn;
1286     Error *local_err = NULL;
1287     SnapshotInfo *info = NULL;
1288     int ret;
1289 
1290     bs = qmp_get_root_bs(device, errp);
1291     if (!bs) {
1292         return NULL;
1293     }
1294     aio_context = bdrv_get_aio_context(bs);
1295     aio_context_acquire(aio_context);
1296 
1297     if (!has_id) {
1298         id = NULL;
1299     }
1300 
1301     if (!has_name) {
1302         name = NULL;
1303     }
1304 
1305     if (!id && !name) {
1306         error_setg(errp, "Name or id must be provided");
1307         goto out_aio_context;
1308     }
1309 
1310     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1311         goto out_aio_context;
1312     }
1313 
1314     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1315     if (local_err) {
1316         error_propagate(errp, local_err);
1317         goto out_aio_context;
1318     }
1319     if (!ret) {
1320         error_setg(errp,
1321                    "Snapshot with id '%s' and name '%s' does not exist on "
1322                    "device '%s'",
1323                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1324         goto out_aio_context;
1325     }
1326 
1327     bdrv_snapshot_delete(bs, id, name, &local_err);
1328     if (local_err) {
1329         error_propagate(errp, local_err);
1330         goto out_aio_context;
1331     }
1332 
1333     aio_context_release(aio_context);
1334 
1335     info = g_new0(SnapshotInfo, 1);
1336     info->id = g_strdup(sn.id_str);
1337     info->name = g_strdup(sn.name);
1338     info->date_nsec = sn.date_nsec;
1339     info->date_sec = sn.date_sec;
1340     info->vm_state_size = sn.vm_state_size;
1341     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1342     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1343 
1344     return info;
1345 
1346 out_aio_context:
1347     aio_context_release(aio_context);
1348     return NULL;
1349 }
1350 
1351 /**
1352  * block_dirty_bitmap_lookup:
1353  * Return a dirty bitmap (if present), after validating
1354  * the node reference and bitmap names.
1355  *
1356  * @node: The name of the BDS node to search for bitmaps
1357  * @name: The name of the bitmap to search for
1358  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1359  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1360  * @errp: Output pointer for error information. Can be NULL.
1361  *
1362  * @return: A bitmap object on success, or NULL on failure.
1363  */
1364 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1365                                                   const char *name,
1366                                                   BlockDriverState **pbs,
1367                                                   AioContext **paio,
1368                                                   Error **errp)
1369 {
1370     BlockDriverState *bs;
1371     BdrvDirtyBitmap *bitmap;
1372     AioContext *aio_context;
1373 
1374     if (!node) {
1375         error_setg(errp, "Node cannot be NULL");
1376         return NULL;
1377     }
1378     if (!name) {
1379         error_setg(errp, "Bitmap name cannot be NULL");
1380         return NULL;
1381     }
1382     bs = bdrv_lookup_bs(node, node, NULL);
1383     if (!bs) {
1384         error_setg(errp, "Node '%s' not found", node);
1385         return NULL;
1386     }
1387 
1388     aio_context = bdrv_get_aio_context(bs);
1389     aio_context_acquire(aio_context);
1390 
1391     bitmap = bdrv_find_dirty_bitmap(bs, name);
1392     if (!bitmap) {
1393         error_setg(errp, "Dirty bitmap '%s' not found", name);
1394         goto fail;
1395     }
1396 
1397     if (pbs) {
1398         *pbs = bs;
1399     }
1400     if (paio) {
1401         *paio = aio_context;
1402     } else {
1403         aio_context_release(aio_context);
1404     }
1405 
1406     return bitmap;
1407 
1408  fail:
1409     aio_context_release(aio_context);
1410     return NULL;
1411 }
1412 
1413 /* New and old BlockDriverState structs for atomic group operations */
1414 
1415 typedef struct BlkActionState BlkActionState;
1416 
1417 /**
1418  * BlkActionOps:
1419  * Table of operations that define an Action.
1420  *
1421  * @instance_size: Size of state struct, in bytes.
1422  * @prepare: Prepare the work, must NOT be NULL.
1423  * @commit: Commit the changes, can be NULL.
1424  * @abort: Abort the changes on fail, can be NULL.
1425  * @clean: Clean up resources after all transaction actions have called
1426  *         commit() or abort(). Can be NULL.
1427  *
1428  * Only prepare() may fail. In a single transaction, only one of commit() or
1429  * abort() will be called. clean() will always be called if it is present.
1430  */
1431 typedef struct BlkActionOps {
1432     size_t instance_size;
1433     void (*prepare)(BlkActionState *common, Error **errp);
1434     void (*commit)(BlkActionState *common);
1435     void (*abort)(BlkActionState *common);
1436     void (*clean)(BlkActionState *common);
1437 } BlkActionOps;
1438 
1439 /**
1440  * BlkActionState:
1441  * Describes one Action's state within a Transaction.
1442  *
1443  * @action: QAPI-defined enum identifying which Action to perform.
1444  * @ops: Table of ActionOps this Action can perform.
1445  * @block_job_txn: Transaction which this action belongs to.
1446  * @entry: List membership for all Actions in this Transaction.
1447  *
1448  * This structure must be arranged as first member in a subclassed type,
1449  * assuming that the compiler will also arrange it to the same offsets as the
1450  * base class.
1451  */
1452 struct BlkActionState {
1453     TransactionAction *action;
1454     const BlkActionOps *ops;
1455     BlockJobTxn *block_job_txn;
1456     TransactionProperties *txn_props;
1457     QSIMPLEQ_ENTRY(BlkActionState) entry;
1458 };
1459 
1460 /* internal snapshot private data */
1461 typedef struct InternalSnapshotState {
1462     BlkActionState common;
1463     BlockDriverState *bs;
1464     AioContext *aio_context;
1465     QEMUSnapshotInfo sn;
1466     bool created;
1467 } InternalSnapshotState;
1468 
1469 
1470 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1471 {
1472     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1473         error_setg(errp,
1474                    "Action '%s' does not support Transaction property "
1475                    "completion-mode = %s",
1476                    TransactionActionKind_lookup[s->action->type],
1477                    ActionCompletionMode_lookup[s->txn_props->completion_mode]);
1478         return -1;
1479     }
1480     return 0;
1481 }
1482 
1483 static void internal_snapshot_prepare(BlkActionState *common,
1484                                       Error **errp)
1485 {
1486     Error *local_err = NULL;
1487     const char *device;
1488     const char *name;
1489     BlockDriverState *bs;
1490     QEMUSnapshotInfo old_sn, *sn;
1491     bool ret;
1492     qemu_timeval tv;
1493     BlockdevSnapshotInternal *internal;
1494     InternalSnapshotState *state;
1495     int ret1;
1496 
1497     g_assert(common->action->type ==
1498              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1499     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1500     state = DO_UPCAST(InternalSnapshotState, common, common);
1501 
1502     /* 1. parse input */
1503     device = internal->device;
1504     name = internal->name;
1505 
1506     /* 2. check for validation */
1507     if (action_check_completion_mode(common, errp) < 0) {
1508         return;
1509     }
1510 
1511     bs = qmp_get_root_bs(device, errp);
1512     if (!bs) {
1513         return;
1514     }
1515 
1516     /* AioContext is released in .clean() */
1517     state->aio_context = bdrv_get_aio_context(bs);
1518     aio_context_acquire(state->aio_context);
1519 
1520     state->bs = bs;
1521     bdrv_drained_begin(bs);
1522 
1523     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1524         return;
1525     }
1526 
1527     if (bdrv_is_read_only(bs)) {
1528         error_setg(errp, "Device '%s' is read only", device);
1529         return;
1530     }
1531 
1532     if (!bdrv_can_snapshot(bs)) {
1533         error_setg(errp, "Block format '%s' used by device '%s' "
1534                    "does not support internal snapshots",
1535                    bs->drv->format_name, device);
1536         return;
1537     }
1538 
1539     if (!strlen(name)) {
1540         error_setg(errp, "Name is empty");
1541         return;
1542     }
1543 
1544     /* check whether a snapshot with name exist */
1545     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1546                                             &local_err);
1547     if (local_err) {
1548         error_propagate(errp, local_err);
1549         return;
1550     } else if (ret) {
1551         error_setg(errp,
1552                    "Snapshot with name '%s' already exists on device '%s'",
1553                    name, device);
1554         return;
1555     }
1556 
1557     /* 3. take the snapshot */
1558     sn = &state->sn;
1559     pstrcpy(sn->name, sizeof(sn->name), name);
1560     qemu_gettimeofday(&tv);
1561     sn->date_sec = tv.tv_sec;
1562     sn->date_nsec = tv.tv_usec * 1000;
1563     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1564 
1565     ret1 = bdrv_snapshot_create(bs, sn);
1566     if (ret1 < 0) {
1567         error_setg_errno(errp, -ret1,
1568                          "Failed to create snapshot '%s' on device '%s'",
1569                          name, device);
1570         return;
1571     }
1572 
1573     /* 4. succeed, mark a snapshot is created */
1574     state->created = true;
1575 }
1576 
1577 static void internal_snapshot_abort(BlkActionState *common)
1578 {
1579     InternalSnapshotState *state =
1580                              DO_UPCAST(InternalSnapshotState, common, common);
1581     BlockDriverState *bs = state->bs;
1582     QEMUSnapshotInfo *sn = &state->sn;
1583     Error *local_error = NULL;
1584 
1585     if (!state->created) {
1586         return;
1587     }
1588 
1589     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1590         error_reportf_err(local_error,
1591                           "Failed to delete snapshot with id '%s' and "
1592                           "name '%s' on device '%s' in abort: ",
1593                           sn->id_str, sn->name,
1594                           bdrv_get_device_name(bs));
1595     }
1596 }
1597 
1598 static void internal_snapshot_clean(BlkActionState *common)
1599 {
1600     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1601                                              common, common);
1602 
1603     if (state->aio_context) {
1604         if (state->bs) {
1605             bdrv_drained_end(state->bs);
1606         }
1607         aio_context_release(state->aio_context);
1608     }
1609 }
1610 
1611 /* external snapshot private data */
1612 typedef struct ExternalSnapshotState {
1613     BlkActionState common;
1614     BlockDriverState *old_bs;
1615     BlockDriverState *new_bs;
1616     AioContext *aio_context;
1617 } ExternalSnapshotState;
1618 
1619 static void external_snapshot_prepare(BlkActionState *common,
1620                                       Error **errp)
1621 {
1622     int flags = 0;
1623     QDict *options = NULL;
1624     Error *local_err = NULL;
1625     /* Device and node name of the image to generate the snapshot from */
1626     const char *device;
1627     const char *node_name;
1628     /* Reference to the new image (for 'blockdev-snapshot') */
1629     const char *snapshot_ref;
1630     /* File name of the new image (for 'blockdev-snapshot-sync') */
1631     const char *new_image_file;
1632     ExternalSnapshotState *state =
1633                              DO_UPCAST(ExternalSnapshotState, common, common);
1634     TransactionAction *action = common->action;
1635 
1636     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1637      * purpose but a different set of parameters */
1638     switch (action->type) {
1639     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1640         {
1641             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1642             device = s->node;
1643             node_name = s->node;
1644             new_image_file = NULL;
1645             snapshot_ref = s->overlay;
1646         }
1647         break;
1648     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1649         {
1650             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1651             device = s->has_device ? s->device : NULL;
1652             node_name = s->has_node_name ? s->node_name : NULL;
1653             new_image_file = s->snapshot_file;
1654             snapshot_ref = NULL;
1655         }
1656         break;
1657     default:
1658         g_assert_not_reached();
1659     }
1660 
1661     /* start processing */
1662     if (action_check_completion_mode(common, errp) < 0) {
1663         return;
1664     }
1665 
1666     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1667     if (!state->old_bs) {
1668         return;
1669     }
1670 
1671     /* Acquire AioContext now so any threads operating on old_bs stop */
1672     state->aio_context = bdrv_get_aio_context(state->old_bs);
1673     aio_context_acquire(state->aio_context);
1674     bdrv_drained_begin(state->old_bs);
1675 
1676     if (!bdrv_is_inserted(state->old_bs)) {
1677         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1678         return;
1679     }
1680 
1681     if (bdrv_op_is_blocked(state->old_bs,
1682                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1683         return;
1684     }
1685 
1686     if (!bdrv_is_read_only(state->old_bs)) {
1687         if (bdrv_flush(state->old_bs)) {
1688             error_setg(errp, QERR_IO_ERROR);
1689             return;
1690         }
1691     }
1692 
1693     if (!bdrv_is_first_non_filter(state->old_bs)) {
1694         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1695         return;
1696     }
1697 
1698     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1699         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1700         const char *format = s->has_format ? s->format : "qcow2";
1701         enum NewImageMode mode;
1702         const char *snapshot_node_name =
1703             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1704 
1705         if (node_name && !snapshot_node_name) {
1706             error_setg(errp, "New snapshot node name missing");
1707             return;
1708         }
1709 
1710         if (snapshot_node_name &&
1711             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1712             error_setg(errp, "New snapshot node name already in use");
1713             return;
1714         }
1715 
1716         flags = state->old_bs->open_flags;
1717         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1718 
1719         /* create new image w/backing file */
1720         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1721         if (mode != NEW_IMAGE_MODE_EXISTING) {
1722             int64_t size = bdrv_getlength(state->old_bs);
1723             if (size < 0) {
1724                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1725                 return;
1726             }
1727             bdrv_img_create(new_image_file, format,
1728                             state->old_bs->filename,
1729                             state->old_bs->drv->format_name,
1730                             NULL, size, flags, &local_err, false);
1731             if (local_err) {
1732                 error_propagate(errp, local_err);
1733                 return;
1734             }
1735         }
1736 
1737         options = qdict_new();
1738         if (s->has_snapshot_node_name) {
1739             qdict_put(options, "node-name",
1740                       qstring_from_str(snapshot_node_name));
1741         }
1742         qdict_put(options, "driver", qstring_from_str(format));
1743 
1744         flags |= BDRV_O_NO_BACKING;
1745     }
1746 
1747     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1748                               errp);
1749     /* We will manually add the backing_hd field to the bs later */
1750     if (!state->new_bs) {
1751         return;
1752     }
1753 
1754     if (bdrv_has_blk(state->new_bs)) {
1755         error_setg(errp, "The snapshot is already in use");
1756         return;
1757     }
1758 
1759     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1760                            errp)) {
1761         return;
1762     }
1763 
1764     if (state->new_bs->backing != NULL) {
1765         error_setg(errp, "The snapshot already has a backing image");
1766         return;
1767     }
1768 
1769     if (!state->new_bs->drv->supports_backing) {
1770         error_setg(errp, "The snapshot does not support backing images");
1771     }
1772 }
1773 
1774 static void external_snapshot_commit(BlkActionState *common)
1775 {
1776     ExternalSnapshotState *state =
1777                              DO_UPCAST(ExternalSnapshotState, common, common);
1778 
1779     bdrv_set_aio_context(state->new_bs, state->aio_context);
1780 
1781     /* This removes our old bs and adds the new bs */
1782     bdrv_append(state->new_bs, state->old_bs);
1783     /* We don't need (or want) to use the transactional
1784      * bdrv_reopen_multiple() across all the entries at once, because we
1785      * don't want to abort all of them if one of them fails the reopen */
1786     if (!state->old_bs->copy_on_read) {
1787         bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1788                     NULL);
1789     }
1790 }
1791 
1792 static void external_snapshot_abort(BlkActionState *common)
1793 {
1794     ExternalSnapshotState *state =
1795                              DO_UPCAST(ExternalSnapshotState, common, common);
1796     if (state->new_bs) {
1797         bdrv_unref(state->new_bs);
1798     }
1799 }
1800 
1801 static void external_snapshot_clean(BlkActionState *common)
1802 {
1803     ExternalSnapshotState *state =
1804                              DO_UPCAST(ExternalSnapshotState, common, common);
1805     if (state->aio_context) {
1806         bdrv_drained_end(state->old_bs);
1807         aio_context_release(state->aio_context);
1808     }
1809 }
1810 
1811 typedef struct DriveBackupState {
1812     BlkActionState common;
1813     BlockDriverState *bs;
1814     AioContext *aio_context;
1815     BlockJob *job;
1816 } DriveBackupState;
1817 
1818 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
1819                             Error **errp);
1820 
1821 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1822 {
1823     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1824     BlockDriverState *bs;
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.data;
1830 
1831     bs = qmp_get_root_bs(backup->device, errp);
1832     if (!bs) {
1833         return;
1834     }
1835 
1836     /* AioContext is released in .clean() */
1837     state->aio_context = bdrv_get_aio_context(bs);
1838     aio_context_acquire(state->aio_context);
1839     bdrv_drained_begin(bs);
1840     state->bs = bs;
1841 
1842     state->job = do_drive_backup(backup, common->block_job_txn, &local_err);
1843     if (local_err) {
1844         error_propagate(errp, local_err);
1845         return;
1846     }
1847 }
1848 
1849 static void drive_backup_commit(BlkActionState *common)
1850 {
1851     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1852     assert(state->job);
1853     block_job_start(state->job);
1854 }
1855 
1856 static void drive_backup_abort(BlkActionState *common)
1857 {
1858     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1859 
1860     if (state->job) {
1861         block_job_cancel_sync(state->job);
1862     }
1863 }
1864 
1865 static void drive_backup_clean(BlkActionState *common)
1866 {
1867     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1868 
1869     if (state->aio_context) {
1870         bdrv_drained_end(state->bs);
1871         aio_context_release(state->aio_context);
1872     }
1873 }
1874 
1875 typedef struct BlockdevBackupState {
1876     BlkActionState common;
1877     BlockDriverState *bs;
1878     BlockJob *job;
1879     AioContext *aio_context;
1880 } BlockdevBackupState;
1881 
1882 static BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
1883                                     Error **errp);
1884 
1885 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1886 {
1887     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1888     BlockdevBackup *backup;
1889     BlockDriverState *bs, *target;
1890     Error *local_err = NULL;
1891 
1892     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1893     backup = common->action->u.blockdev_backup.data;
1894 
1895     bs = qmp_get_root_bs(backup->device, errp);
1896     if (!bs) {
1897         return;
1898     }
1899 
1900     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1901     if (!target) {
1902         return;
1903     }
1904 
1905     /* AioContext is released in .clean() */
1906     state->aio_context = bdrv_get_aio_context(bs);
1907     if (state->aio_context != bdrv_get_aio_context(target)) {
1908         state->aio_context = NULL;
1909         error_setg(errp, "Backup between two IO threads is not implemented");
1910         return;
1911     }
1912     aio_context_acquire(state->aio_context);
1913     state->bs = bs;
1914     bdrv_drained_begin(state->bs);
1915 
1916     state->job = do_blockdev_backup(backup, common->block_job_txn, &local_err);
1917     if (local_err) {
1918         error_propagate(errp, local_err);
1919         return;
1920     }
1921 }
1922 
1923 static void blockdev_backup_commit(BlkActionState *common)
1924 {
1925     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1926     assert(state->job);
1927     block_job_start(state->job);
1928 }
1929 
1930 static void blockdev_backup_abort(BlkActionState *common)
1931 {
1932     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1933 
1934     if (state->job) {
1935         block_job_cancel_sync(state->job);
1936     }
1937 }
1938 
1939 static void blockdev_backup_clean(BlkActionState *common)
1940 {
1941     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1942 
1943     if (state->aio_context) {
1944         bdrv_drained_end(state->bs);
1945         aio_context_release(state->aio_context);
1946     }
1947 }
1948 
1949 typedef struct BlockDirtyBitmapState {
1950     BlkActionState common;
1951     BdrvDirtyBitmap *bitmap;
1952     BlockDriverState *bs;
1953     AioContext *aio_context;
1954     HBitmap *backup;
1955     bool prepared;
1956 } BlockDirtyBitmapState;
1957 
1958 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1959                                            Error **errp)
1960 {
1961     Error *local_err = NULL;
1962     BlockDirtyBitmapAdd *action;
1963     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1964                                              common, common);
1965 
1966     if (action_check_completion_mode(common, errp) < 0) {
1967         return;
1968     }
1969 
1970     action = common->action->u.block_dirty_bitmap_add.data;
1971     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1972     qmp_block_dirty_bitmap_add(action->node, action->name,
1973                                action->has_granularity, action->granularity,
1974                                &local_err);
1975 
1976     if (!local_err) {
1977         state->prepared = true;
1978     } else {
1979         error_propagate(errp, local_err);
1980     }
1981 }
1982 
1983 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1984 {
1985     BlockDirtyBitmapAdd *action;
1986     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1987                                              common, common);
1988 
1989     action = common->action->u.block_dirty_bitmap_add.data;
1990     /* Should not be able to fail: IF the bitmap was added via .prepare(),
1991      * then the node reference and bitmap name must have been valid.
1992      */
1993     if (state->prepared) {
1994         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
1995     }
1996 }
1997 
1998 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
1999                                              Error **errp)
2000 {
2001     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2002                                              common, common);
2003     BlockDirtyBitmap *action;
2004 
2005     if (action_check_completion_mode(common, errp) < 0) {
2006         return;
2007     }
2008 
2009     action = common->action->u.block_dirty_bitmap_clear.data;
2010     state->bitmap = block_dirty_bitmap_lookup(action->node,
2011                                               action->name,
2012                                               &state->bs,
2013                                               &state->aio_context,
2014                                               errp);
2015     if (!state->bitmap) {
2016         return;
2017     }
2018 
2019     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2020         error_setg(errp, "Cannot modify a frozen bitmap");
2021         return;
2022     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2023         error_setg(errp, "Cannot clear a disabled bitmap");
2024         return;
2025     }
2026 
2027     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2028     /* AioContext is released in .clean() */
2029 }
2030 
2031 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2032 {
2033     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2034                                              common, common);
2035 
2036     bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2037 }
2038 
2039 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2040 {
2041     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2042                                              common, common);
2043 
2044     hbitmap_free(state->backup);
2045 }
2046 
2047 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2048 {
2049     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2050                                              common, common);
2051 
2052     if (state->aio_context) {
2053         aio_context_release(state->aio_context);
2054     }
2055 }
2056 
2057 static void abort_prepare(BlkActionState *common, Error **errp)
2058 {
2059     error_setg(errp, "Transaction aborted using Abort action");
2060 }
2061 
2062 static void abort_commit(BlkActionState *common)
2063 {
2064     g_assert_not_reached(); /* this action never succeeds */
2065 }
2066 
2067 static const BlkActionOps actions[] = {
2068     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2069         .instance_size = sizeof(ExternalSnapshotState),
2070         .prepare  = external_snapshot_prepare,
2071         .commit   = external_snapshot_commit,
2072         .abort = external_snapshot_abort,
2073         .clean = external_snapshot_clean,
2074     },
2075     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2076         .instance_size = sizeof(ExternalSnapshotState),
2077         .prepare  = external_snapshot_prepare,
2078         .commit   = external_snapshot_commit,
2079         .abort = external_snapshot_abort,
2080         .clean = external_snapshot_clean,
2081     },
2082     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2083         .instance_size = sizeof(DriveBackupState),
2084         .prepare = drive_backup_prepare,
2085         .commit = drive_backup_commit,
2086         .abort = drive_backup_abort,
2087         .clean = drive_backup_clean,
2088     },
2089     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2090         .instance_size = sizeof(BlockdevBackupState),
2091         .prepare = blockdev_backup_prepare,
2092         .commit = blockdev_backup_commit,
2093         .abort = blockdev_backup_abort,
2094         .clean = blockdev_backup_clean,
2095     },
2096     [TRANSACTION_ACTION_KIND_ABORT] = {
2097         .instance_size = sizeof(BlkActionState),
2098         .prepare = abort_prepare,
2099         .commit = abort_commit,
2100     },
2101     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2102         .instance_size = sizeof(InternalSnapshotState),
2103         .prepare  = internal_snapshot_prepare,
2104         .abort = internal_snapshot_abort,
2105         .clean = internal_snapshot_clean,
2106     },
2107     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2108         .instance_size = sizeof(BlockDirtyBitmapState),
2109         .prepare = block_dirty_bitmap_add_prepare,
2110         .abort = block_dirty_bitmap_add_abort,
2111     },
2112     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2113         .instance_size = sizeof(BlockDirtyBitmapState),
2114         .prepare = block_dirty_bitmap_clear_prepare,
2115         .commit = block_dirty_bitmap_clear_commit,
2116         .abort = block_dirty_bitmap_clear_abort,
2117         .clean = block_dirty_bitmap_clear_clean,
2118     }
2119 };
2120 
2121 /**
2122  * Allocate a TransactionProperties structure if necessary, and fill
2123  * that structure with desired defaults if they are unset.
2124  */
2125 static TransactionProperties *get_transaction_properties(
2126     TransactionProperties *props)
2127 {
2128     if (!props) {
2129         props = g_new0(TransactionProperties, 1);
2130     }
2131 
2132     if (!props->has_completion_mode) {
2133         props->has_completion_mode = true;
2134         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2135     }
2136 
2137     return props;
2138 }
2139 
2140 /*
2141  * 'Atomic' group operations.  The operations are performed as a set, and if
2142  * any fail then we roll back all operations in the group.
2143  */
2144 void qmp_transaction(TransactionActionList *dev_list,
2145                      bool has_props,
2146                      struct TransactionProperties *props,
2147                      Error **errp)
2148 {
2149     TransactionActionList *dev_entry = dev_list;
2150     BlockJobTxn *block_job_txn = NULL;
2151     BlkActionState *state, *next;
2152     Error *local_err = NULL;
2153 
2154     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2155     QSIMPLEQ_INIT(&snap_bdrv_states);
2156 
2157     /* Does this transaction get canceled as a group on failure?
2158      * If not, we don't really need to make a BlockJobTxn.
2159      */
2160     props = get_transaction_properties(props);
2161     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2162         block_job_txn = block_job_txn_new();
2163     }
2164 
2165     /* drain all i/o before any operations */
2166     bdrv_drain_all();
2167 
2168     /* We don't do anything in this loop that commits us to the operations */
2169     while (NULL != dev_entry) {
2170         TransactionAction *dev_info = NULL;
2171         const BlkActionOps *ops;
2172 
2173         dev_info = dev_entry->value;
2174         dev_entry = dev_entry->next;
2175 
2176         assert(dev_info->type < ARRAY_SIZE(actions));
2177 
2178         ops = &actions[dev_info->type];
2179         assert(ops->instance_size > 0);
2180 
2181         state = g_malloc0(ops->instance_size);
2182         state->ops = ops;
2183         state->action = dev_info;
2184         state->block_job_txn = block_job_txn;
2185         state->txn_props = props;
2186         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2187 
2188         state->ops->prepare(state, &local_err);
2189         if (local_err) {
2190             error_propagate(errp, local_err);
2191             goto delete_and_fail;
2192         }
2193     }
2194 
2195     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2196         if (state->ops->commit) {
2197             state->ops->commit(state);
2198         }
2199     }
2200 
2201     /* success */
2202     goto exit;
2203 
2204 delete_and_fail:
2205     /* failure, and it is all-or-none; roll back all operations */
2206     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2207         if (state->ops->abort) {
2208             state->ops->abort(state);
2209         }
2210     }
2211 exit:
2212     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2213         if (state->ops->clean) {
2214             state->ops->clean(state);
2215         }
2216         g_free(state);
2217     }
2218     if (!has_props) {
2219         qapi_free_TransactionProperties(props);
2220     }
2221     block_job_txn_unref(block_job_txn);
2222 }
2223 
2224 void qmp_eject(bool has_device, const char *device,
2225                bool has_id, const char *id,
2226                bool has_force, bool force, Error **errp)
2227 {
2228     Error *local_err = NULL;
2229     int rc;
2230 
2231     if (!has_force) {
2232         force = false;
2233     }
2234 
2235     rc = do_open_tray(has_device ? device : NULL,
2236                       has_id ? id : NULL,
2237                       force, &local_err);
2238     if (rc && rc != -ENOSYS) {
2239         error_propagate(errp, local_err);
2240         return;
2241     }
2242     error_free(local_err);
2243 
2244     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, errp);
2245 }
2246 
2247 void qmp_block_passwd(bool has_device, const char *device,
2248                       bool has_node_name, const char *node_name,
2249                       const char *password, Error **errp)
2250 {
2251     Error *local_err = NULL;
2252     BlockDriverState *bs;
2253     AioContext *aio_context;
2254 
2255     bs = bdrv_lookup_bs(has_device ? device : NULL,
2256                         has_node_name ? node_name : NULL,
2257                         &local_err);
2258     if (local_err) {
2259         error_propagate(errp, local_err);
2260         return;
2261     }
2262 
2263     aio_context = bdrv_get_aio_context(bs);
2264     aio_context_acquire(aio_context);
2265 
2266     bdrv_add_key(bs, password, errp);
2267 
2268     aio_context_release(aio_context);
2269 }
2270 
2271 /*
2272  * Attempt to open the tray of @device.
2273  * If @force, ignore its tray lock.
2274  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2275  * On error, store an error through @errp and return -errno.
2276  * If @device does not exist, return -ENODEV.
2277  * If it has no removable media, return -ENOTSUP.
2278  * If it has no tray, return -ENOSYS.
2279  * If the guest was asked to open the tray, return -EINPROGRESS.
2280  * Else, return 0.
2281  */
2282 static int do_open_tray(const char *blk_name, const char *qdev_id,
2283                         bool force, Error **errp)
2284 {
2285     BlockBackend *blk;
2286     const char *device = qdev_id ?: blk_name;
2287     bool locked;
2288 
2289     blk = qmp_get_blk(blk_name, qdev_id, errp);
2290     if (!blk) {
2291         return -ENODEV;
2292     }
2293 
2294     if (!blk_dev_has_removable_media(blk)) {
2295         error_setg(errp, "Device '%s' is not removable", device);
2296         return -ENOTSUP;
2297     }
2298 
2299     if (!blk_dev_has_tray(blk)) {
2300         error_setg(errp, "Device '%s' does not have a tray", device);
2301         return -ENOSYS;
2302     }
2303 
2304     if (blk_dev_is_tray_open(blk)) {
2305         return 0;
2306     }
2307 
2308     locked = blk_dev_is_medium_locked(blk);
2309     if (locked) {
2310         blk_dev_eject_request(blk, force);
2311     }
2312 
2313     if (!locked || force) {
2314         blk_dev_change_media_cb(blk, false);
2315     }
2316 
2317     if (locked && !force) {
2318         error_setg(errp, "Device '%s' is locked and force was not specified, "
2319                    "wait for tray to open and try again", device);
2320         return -EINPROGRESS;
2321     }
2322 
2323     return 0;
2324 }
2325 
2326 void qmp_blockdev_open_tray(bool has_device, const char *device,
2327                             bool has_id, const char *id,
2328                             bool has_force, bool force,
2329                             Error **errp)
2330 {
2331     Error *local_err = NULL;
2332     int rc;
2333 
2334     if (!has_force) {
2335         force = false;
2336     }
2337     rc = do_open_tray(has_device ? device : NULL,
2338                       has_id ? id : NULL,
2339                       force, &local_err);
2340     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2341         error_propagate(errp, local_err);
2342         return;
2343     }
2344     error_free(local_err);
2345 }
2346 
2347 void qmp_blockdev_close_tray(bool has_device, const char *device,
2348                              bool has_id, const char *id,
2349                              Error **errp)
2350 {
2351     BlockBackend *blk;
2352 
2353     device = has_device ? device : NULL;
2354     id = has_id ? id : NULL;
2355 
2356     blk = qmp_get_blk(device, id, errp);
2357     if (!blk) {
2358         return;
2359     }
2360 
2361     if (!blk_dev_has_removable_media(blk)) {
2362         error_setg(errp, "Device '%s' is not removable", device ?: id);
2363         return;
2364     }
2365 
2366     if (!blk_dev_has_tray(blk)) {
2367         /* Ignore this command on tray-less devices */
2368         return;
2369     }
2370 
2371     if (!blk_dev_is_tray_open(blk)) {
2372         return;
2373     }
2374 
2375     blk_dev_change_media_cb(blk, true);
2376 }
2377 
2378 void qmp_x_blockdev_remove_medium(bool has_device, const char *device,
2379                                   bool has_id, const char *id, Error **errp)
2380 {
2381     BlockBackend *blk;
2382     BlockDriverState *bs;
2383     AioContext *aio_context;
2384     bool has_attached_device;
2385 
2386     device = has_device ? device : NULL;
2387     id = has_id ? id : NULL;
2388 
2389     blk = qmp_get_blk(device, id, errp);
2390     if (!blk) {
2391         return;
2392     }
2393 
2394     /* For BBs without a device, we can exchange the BDS tree at will */
2395     has_attached_device = blk_get_attached_dev(blk);
2396 
2397     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2398         error_setg(errp, "Device '%s' is not removable", device ?: id);
2399         return;
2400     }
2401 
2402     if (has_attached_device && blk_dev_has_tray(blk) &&
2403         !blk_dev_is_tray_open(blk))
2404     {
2405         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2406         return;
2407     }
2408 
2409     bs = blk_bs(blk);
2410     if (!bs) {
2411         return;
2412     }
2413 
2414     aio_context = bdrv_get_aio_context(bs);
2415     aio_context_acquire(aio_context);
2416 
2417     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2418         goto out;
2419     }
2420 
2421     blk_remove_bs(blk);
2422 
2423     if (!blk_dev_has_tray(blk)) {
2424         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2425          * called at all); therefore, the medium needs to be ejected here.
2426          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2427          * value passed here (i.e. false). */
2428         blk_dev_change_media_cb(blk, false);
2429     }
2430 
2431 out:
2432     aio_context_release(aio_context);
2433 }
2434 
2435 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2436                                             BlockDriverState *bs, Error **errp)
2437 {
2438     bool has_device;
2439 
2440     /* For BBs without a device, we can exchange the BDS tree at will */
2441     has_device = blk_get_attached_dev(blk);
2442 
2443     if (has_device && !blk_dev_has_removable_media(blk)) {
2444         error_setg(errp, "Device is not removable");
2445         return;
2446     }
2447 
2448     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2449         error_setg(errp, "Tray of the device is not open");
2450         return;
2451     }
2452 
2453     if (blk_bs(blk)) {
2454         error_setg(errp, "There already is a medium in the device");
2455         return;
2456     }
2457 
2458     blk_insert_bs(blk, bs);
2459 
2460     if (!blk_dev_has_tray(blk)) {
2461         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2462          * called at all); therefore, the medium needs to be pushed into the
2463          * slot here.
2464          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2465          * value passed here (i.e. true). */
2466         blk_dev_change_media_cb(blk, true);
2467     }
2468 }
2469 
2470 void qmp_x_blockdev_insert_medium(bool has_device, const char *device,
2471                                   bool has_id, const char *id,
2472                                   const char *node_name, Error **errp)
2473 {
2474     BlockBackend *blk;
2475     BlockDriverState *bs;
2476 
2477     blk = qmp_get_blk(has_device ? device : NULL,
2478                       has_id ? id : NULL,
2479                       errp);
2480     if (!blk) {
2481         return;
2482     }
2483 
2484     bs = bdrv_find_node(node_name);
2485     if (!bs) {
2486         error_setg(errp, "Node '%s' not found", node_name);
2487         return;
2488     }
2489 
2490     if (bdrv_has_blk(bs)) {
2491         error_setg(errp, "Node '%s' is already in use", node_name);
2492         return;
2493     }
2494 
2495     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2496 }
2497 
2498 void qmp_blockdev_change_medium(bool has_device, const char *device,
2499                                 bool has_id, const char *id,
2500                                 const char *filename,
2501                                 bool has_format, const char *format,
2502                                 bool has_read_only,
2503                                 BlockdevChangeReadOnlyMode read_only,
2504                                 Error **errp)
2505 {
2506     BlockBackend *blk;
2507     BlockDriverState *medium_bs = NULL;
2508     int bdrv_flags;
2509     bool detect_zeroes;
2510     int rc;
2511     QDict *options = NULL;
2512     Error *err = NULL;
2513 
2514     blk = qmp_get_blk(has_device ? device : NULL,
2515                       has_id ? id : NULL,
2516                       errp);
2517     if (!blk) {
2518         goto fail;
2519     }
2520 
2521     if (blk_bs(blk)) {
2522         blk_update_root_state(blk);
2523     }
2524 
2525     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2526     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2527         BDRV_O_PROTOCOL);
2528 
2529     if (!has_read_only) {
2530         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2531     }
2532 
2533     switch (read_only) {
2534     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2535         break;
2536 
2537     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2538         bdrv_flags &= ~BDRV_O_RDWR;
2539         break;
2540 
2541     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2542         bdrv_flags |= BDRV_O_RDWR;
2543         break;
2544 
2545     default:
2546         abort();
2547     }
2548 
2549     options = qdict_new();
2550     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2551     qdict_put(options, "detect-zeroes",
2552               qstring_from_str(detect_zeroes ? "on" : "off"));
2553 
2554     if (has_format) {
2555         qdict_put(options, "driver", qstring_from_str(format));
2556     }
2557 
2558     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2559     if (!medium_bs) {
2560         goto fail;
2561     }
2562 
2563     bdrv_add_key(medium_bs, NULL, &err);
2564     if (err) {
2565         error_propagate(errp, err);
2566         goto fail;
2567     }
2568 
2569     rc = do_open_tray(has_device ? device : NULL,
2570                       has_id ? id : NULL,
2571                       false, &err);
2572     if (rc && rc != -ENOSYS) {
2573         error_propagate(errp, err);
2574         goto fail;
2575     }
2576     error_free(err);
2577     err = NULL;
2578 
2579     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, &err);
2580     if (err) {
2581         error_propagate(errp, err);
2582         goto fail;
2583     }
2584 
2585     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2586     if (err) {
2587         error_propagate(errp, err);
2588         goto fail;
2589     }
2590 
2591     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2592 
2593 fail:
2594     /* If the medium has been inserted, the device has its own reference, so
2595      * ours must be relinquished; and if it has not been inserted successfully,
2596      * the reference must be relinquished anyway */
2597     bdrv_unref(medium_bs);
2598 }
2599 
2600 /* throttling disk I/O limits */
2601 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2602 {
2603     ThrottleConfig cfg;
2604     BlockDriverState *bs;
2605     BlockBackend *blk;
2606     AioContext *aio_context;
2607 
2608     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2609                       arg->has_id ? arg->id : NULL,
2610                       errp);
2611     if (!blk) {
2612         return;
2613     }
2614 
2615     aio_context = blk_get_aio_context(blk);
2616     aio_context_acquire(aio_context);
2617 
2618     bs = blk_bs(blk);
2619     if (!bs) {
2620         error_setg(errp, "Device has no medium");
2621         goto out;
2622     }
2623 
2624     throttle_config_init(&cfg);
2625     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2626     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2627     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2628 
2629     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2630     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2631     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2632 
2633     if (arg->has_bps_max) {
2634         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2635     }
2636     if (arg->has_bps_rd_max) {
2637         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2638     }
2639     if (arg->has_bps_wr_max) {
2640         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2641     }
2642     if (arg->has_iops_max) {
2643         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2644     }
2645     if (arg->has_iops_rd_max) {
2646         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2647     }
2648     if (arg->has_iops_wr_max) {
2649         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2650     }
2651 
2652     if (arg->has_bps_max_length) {
2653         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2654     }
2655     if (arg->has_bps_rd_max_length) {
2656         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2657     }
2658     if (arg->has_bps_wr_max_length) {
2659         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2660     }
2661     if (arg->has_iops_max_length) {
2662         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2663     }
2664     if (arg->has_iops_rd_max_length) {
2665         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2666     }
2667     if (arg->has_iops_wr_max_length) {
2668         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2669     }
2670 
2671     if (arg->has_iops_size) {
2672         cfg.op_size = arg->iops_size;
2673     }
2674 
2675     if (!throttle_is_valid(&cfg, errp)) {
2676         goto out;
2677     }
2678 
2679     if (throttle_enabled(&cfg)) {
2680         /* Enable I/O limits if they're not enabled yet, otherwise
2681          * just update the throttling group. */
2682         if (!blk_get_public(blk)->throttle_state) {
2683             blk_io_limits_enable(blk,
2684                                  arg->has_group ? arg->group :
2685                                  arg->has_device ? arg->device :
2686                                  arg->id);
2687         } else if (arg->has_group) {
2688             blk_io_limits_update_group(blk, arg->group);
2689         }
2690         /* Set the new throttling configuration */
2691         blk_set_io_limits(blk, &cfg);
2692     } else if (blk_get_public(blk)->throttle_state) {
2693         /* If all throttling settings are set to 0, disable I/O limits */
2694         blk_io_limits_disable(blk);
2695     }
2696 
2697 out:
2698     aio_context_release(aio_context);
2699 }
2700 
2701 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2702                                 bool has_granularity, uint32_t granularity,
2703                                 Error **errp)
2704 {
2705     AioContext *aio_context;
2706     BlockDriverState *bs;
2707 
2708     if (!name || name[0] == '\0') {
2709         error_setg(errp, "Bitmap name cannot be empty");
2710         return;
2711     }
2712 
2713     bs = bdrv_lookup_bs(node, node, errp);
2714     if (!bs) {
2715         return;
2716     }
2717 
2718     aio_context = bdrv_get_aio_context(bs);
2719     aio_context_acquire(aio_context);
2720 
2721     if (has_granularity) {
2722         if (granularity < 512 || !is_power_of_2(granularity)) {
2723             error_setg(errp, "Granularity must be power of 2 "
2724                              "and at least 512");
2725             goto out;
2726         }
2727     } else {
2728         /* Default to cluster size, if available: */
2729         granularity = bdrv_get_default_bitmap_granularity(bs);
2730     }
2731 
2732     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2733 
2734  out:
2735     aio_context_release(aio_context);
2736 }
2737 
2738 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2739                                    Error **errp)
2740 {
2741     AioContext *aio_context;
2742     BlockDriverState *bs;
2743     BdrvDirtyBitmap *bitmap;
2744 
2745     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2746     if (!bitmap || !bs) {
2747         return;
2748     }
2749 
2750     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2751         error_setg(errp,
2752                    "Bitmap '%s' is currently frozen and cannot be removed",
2753                    name);
2754         goto out;
2755     }
2756     bdrv_dirty_bitmap_make_anon(bitmap);
2757     bdrv_release_dirty_bitmap(bs, bitmap);
2758 
2759  out:
2760     aio_context_release(aio_context);
2761 }
2762 
2763 /**
2764  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2765  * immediately after a full backup operation.
2766  */
2767 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2768                                   Error **errp)
2769 {
2770     AioContext *aio_context;
2771     BdrvDirtyBitmap *bitmap;
2772     BlockDriverState *bs;
2773 
2774     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2775     if (!bitmap || !bs) {
2776         return;
2777     }
2778 
2779     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2780         error_setg(errp,
2781                    "Bitmap '%s' is currently frozen and cannot be modified",
2782                    name);
2783         goto out;
2784     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2785         error_setg(errp,
2786                    "Bitmap '%s' is currently disabled and cannot be cleared",
2787                    name);
2788         goto out;
2789     }
2790 
2791     bdrv_clear_dirty_bitmap(bitmap, NULL);
2792 
2793  out:
2794     aio_context_release(aio_context);
2795 }
2796 
2797 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2798 {
2799     const char *id = qdict_get_str(qdict, "id");
2800     BlockBackend *blk;
2801     BlockDriverState *bs;
2802     AioContext *aio_context;
2803     Error *local_err = NULL;
2804 
2805     bs = bdrv_find_node(id);
2806     if (bs) {
2807         qmp_x_blockdev_del(id, &local_err);
2808         if (local_err) {
2809             error_report_err(local_err);
2810         }
2811         return;
2812     }
2813 
2814     blk = blk_by_name(id);
2815     if (!blk) {
2816         error_report("Device '%s' not found", id);
2817         return;
2818     }
2819 
2820     if (!blk_legacy_dinfo(blk)) {
2821         error_report("Deleting device added with blockdev-add"
2822                      " is not supported");
2823         return;
2824     }
2825 
2826     aio_context = blk_get_aio_context(blk);
2827     aio_context_acquire(aio_context);
2828 
2829     bs = blk_bs(blk);
2830     if (bs) {
2831         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2832             error_report_err(local_err);
2833             aio_context_release(aio_context);
2834             return;
2835         }
2836 
2837         blk_remove_bs(blk);
2838     }
2839 
2840     /* Make the BlockBackend and the attached BlockDriverState anonymous */
2841     monitor_remove_blk(blk);
2842 
2843     /* If this BlockBackend has a device attached to it, its refcount will be
2844      * decremented when the device is removed; otherwise we have to do so here.
2845      */
2846     if (blk_get_attached_dev(blk)) {
2847         /* Further I/O must not pause the guest */
2848         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2849                          BLOCKDEV_ON_ERROR_REPORT);
2850     } else {
2851         blk_unref(blk);
2852     }
2853 
2854     aio_context_release(aio_context);
2855 }
2856 
2857 void qmp_block_resize(bool has_device, const char *device,
2858                       bool has_node_name, const char *node_name,
2859                       int64_t size, Error **errp)
2860 {
2861     Error *local_err = NULL;
2862     BlockBackend *blk = NULL;
2863     BlockDriverState *bs;
2864     AioContext *aio_context;
2865     int ret;
2866 
2867     bs = bdrv_lookup_bs(has_device ? device : NULL,
2868                         has_node_name ? node_name : NULL,
2869                         &local_err);
2870     if (local_err) {
2871         error_propagate(errp, local_err);
2872         return;
2873     }
2874 
2875     aio_context = bdrv_get_aio_context(bs);
2876     aio_context_acquire(aio_context);
2877 
2878     if (!bdrv_is_first_non_filter(bs)) {
2879         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2880         goto out;
2881     }
2882 
2883     if (size < 0) {
2884         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2885         goto out;
2886     }
2887 
2888     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2889         error_setg(errp, QERR_DEVICE_IN_USE, device);
2890         goto out;
2891     }
2892 
2893     blk = blk_new();
2894     blk_insert_bs(blk, bs);
2895 
2896     /* complete all in-flight operations before resizing the device */
2897     bdrv_drain_all();
2898 
2899     ret = blk_truncate(blk, size);
2900     switch (ret) {
2901     case 0:
2902         break;
2903     case -ENOMEDIUM:
2904         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2905         break;
2906     case -ENOTSUP:
2907         error_setg(errp, QERR_UNSUPPORTED);
2908         break;
2909     case -EACCES:
2910         error_setg(errp, "Device '%s' is read only", device);
2911         break;
2912     case -EBUSY:
2913         error_setg(errp, QERR_DEVICE_IN_USE, device);
2914         break;
2915     default:
2916         error_setg_errno(errp, -ret, "Could not resize");
2917         break;
2918     }
2919 
2920 out:
2921     blk_unref(blk);
2922     aio_context_release(aio_context);
2923 }
2924 
2925 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2926                       bool has_base, const char *base,
2927                       bool has_base_node, const char *base_node,
2928                       bool has_backing_file, const char *backing_file,
2929                       bool has_speed, int64_t speed,
2930                       bool has_on_error, BlockdevOnError on_error,
2931                       Error **errp)
2932 {
2933     BlockDriverState *bs, *iter;
2934     BlockDriverState *base_bs = NULL;
2935     AioContext *aio_context;
2936     Error *local_err = NULL;
2937     const char *base_name = NULL;
2938 
2939     if (!has_on_error) {
2940         on_error = BLOCKDEV_ON_ERROR_REPORT;
2941     }
2942 
2943     bs = bdrv_lookup_bs(device, device, errp);
2944     if (!bs) {
2945         return;
2946     }
2947 
2948     aio_context = bdrv_get_aio_context(bs);
2949     aio_context_acquire(aio_context);
2950 
2951     if (has_base && has_base_node) {
2952         error_setg(errp, "'base' and 'base-node' cannot be specified "
2953                    "at the same time");
2954         goto out;
2955     }
2956 
2957     if (has_base) {
2958         base_bs = bdrv_find_backing_image(bs, base);
2959         if (base_bs == NULL) {
2960             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2961             goto out;
2962         }
2963         assert(bdrv_get_aio_context(base_bs) == aio_context);
2964         base_name = base;
2965     }
2966 
2967     if (has_base_node) {
2968         base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2969         if (!base_bs) {
2970             goto out;
2971         }
2972         if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2973             error_setg(errp, "Node '%s' is not a backing image of '%s'",
2974                        base_node, device);
2975             goto out;
2976         }
2977         assert(bdrv_get_aio_context(base_bs) == aio_context);
2978         base_name = base_bs->filename;
2979     }
2980 
2981     /* Check for op blockers in the whole chain between bs and base */
2982     for (iter = bs; iter && iter != base_bs; iter = backing_bs(iter)) {
2983         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2984             goto out;
2985         }
2986     }
2987 
2988     /* if we are streaming the entire chain, the result will have no backing
2989      * file, and specifying one is therefore an error */
2990     if (base_bs == NULL && has_backing_file) {
2991         error_setg(errp, "backing file specified, but streaming the "
2992                          "entire chain");
2993         goto out;
2994     }
2995 
2996     /* backing_file string overrides base bs filename */
2997     base_name = has_backing_file ? backing_file : base_name;
2998 
2999     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
3000                  has_speed ? speed : 0, on_error, &local_err);
3001     if (local_err) {
3002         error_propagate(errp, local_err);
3003         goto out;
3004     }
3005 
3006     trace_qmp_block_stream(bs, bs->job);
3007 
3008 out:
3009     aio_context_release(aio_context);
3010 }
3011 
3012 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
3013                       bool has_base, const char *base,
3014                       bool has_top, const char *top,
3015                       bool has_backing_file, const char *backing_file,
3016                       bool has_speed, int64_t speed,
3017                       Error **errp)
3018 {
3019     BlockDriverState *bs;
3020     BlockDriverState *iter;
3021     BlockDriverState *base_bs, *top_bs;
3022     AioContext *aio_context;
3023     Error *local_err = NULL;
3024     /* This will be part of the QMP command, if/when the
3025      * BlockdevOnError change for blkmirror makes it in
3026      */
3027     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3028 
3029     if (!has_speed) {
3030         speed = 0;
3031     }
3032 
3033     /* Important Note:
3034      *  libvirt relies on the DeviceNotFound error class in order to probe for
3035      *  live commit feature versions; for this to work, we must make sure to
3036      *  perform the device lookup before any generic errors that may occur in a
3037      *  scenario in which all optional arguments are omitted. */
3038     bs = qmp_get_root_bs(device, &local_err);
3039     if (!bs) {
3040         bs = bdrv_lookup_bs(device, device, NULL);
3041         if (!bs) {
3042             error_free(local_err);
3043             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3044                       "Device '%s' not found", device);
3045         } else {
3046             error_propagate(errp, local_err);
3047         }
3048         return;
3049     }
3050 
3051     aio_context = bdrv_get_aio_context(bs);
3052     aio_context_acquire(aio_context);
3053 
3054     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3055         goto out;
3056     }
3057 
3058     /* default top_bs is the active layer */
3059     top_bs = bs;
3060 
3061     if (has_top && top) {
3062         if (strcmp(bs->filename, top) != 0) {
3063             top_bs = bdrv_find_backing_image(bs, top);
3064         }
3065     }
3066 
3067     if (top_bs == NULL) {
3068         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3069         goto out;
3070     }
3071 
3072     assert(bdrv_get_aio_context(top_bs) == aio_context);
3073 
3074     if (has_base && base) {
3075         base_bs = bdrv_find_backing_image(top_bs, base);
3076     } else {
3077         base_bs = bdrv_find_base(top_bs);
3078     }
3079 
3080     if (base_bs == NULL) {
3081         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3082         goto out;
3083     }
3084 
3085     assert(bdrv_get_aio_context(base_bs) == aio_context);
3086 
3087     for (iter = top_bs; iter != backing_bs(base_bs); iter = backing_bs(iter)) {
3088         if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3089             goto out;
3090         }
3091     }
3092 
3093     /* Do not allow attempts to commit an image into itself */
3094     if (top_bs == base_bs) {
3095         error_setg(errp, "cannot commit an image into itself");
3096         goto out;
3097     }
3098 
3099     if (top_bs == bs) {
3100         if (has_backing_file) {
3101             error_setg(errp, "'backing-file' specified,"
3102                              " but 'top' is the active layer");
3103             goto out;
3104         }
3105         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs,
3106                             BLOCK_JOB_DEFAULT, speed, on_error, NULL, NULL,
3107                             &local_err, false);
3108     } else {
3109         BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
3110         if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3111             goto out;
3112         }
3113         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, speed,
3114                      on_error, has_backing_file ? backing_file : NULL,
3115                      &local_err);
3116     }
3117     if (local_err != NULL) {
3118         error_propagate(errp, local_err);
3119         goto out;
3120     }
3121 
3122 out:
3123     aio_context_release(aio_context);
3124 }
3125 
3126 static BlockJob *do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
3127                                  Error **errp)
3128 {
3129     BlockDriverState *bs;
3130     BlockDriverState *target_bs;
3131     BlockDriverState *source = NULL;
3132     BlockJob *job = NULL;
3133     BdrvDirtyBitmap *bmap = NULL;
3134     AioContext *aio_context;
3135     QDict *options = NULL;
3136     Error *local_err = NULL;
3137     int flags;
3138     int64_t size;
3139 
3140     if (!backup->has_speed) {
3141         backup->speed = 0;
3142     }
3143     if (!backup->has_on_source_error) {
3144         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3145     }
3146     if (!backup->has_on_target_error) {
3147         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3148     }
3149     if (!backup->has_mode) {
3150         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3151     }
3152     if (!backup->has_job_id) {
3153         backup->job_id = NULL;
3154     }
3155     if (!backup->has_compress) {
3156         backup->compress = false;
3157     }
3158 
3159     bs = qmp_get_root_bs(backup->device, errp);
3160     if (!bs) {
3161         return NULL;
3162     }
3163 
3164     aio_context = bdrv_get_aio_context(bs);
3165     aio_context_acquire(aio_context);
3166 
3167     if (!backup->has_format) {
3168         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3169                          NULL : (char*) bs->drv->format_name;
3170     }
3171 
3172     /* Early check to avoid creating target */
3173     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3174         goto out;
3175     }
3176 
3177     flags = bs->open_flags | BDRV_O_RDWR;
3178 
3179     /* See if we have a backing HD we can use to create our new image
3180      * on top of. */
3181     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3182         source = backing_bs(bs);
3183         if (!source) {
3184             backup->sync = MIRROR_SYNC_MODE_FULL;
3185         }
3186     }
3187     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3188         source = bs;
3189     }
3190 
3191     size = bdrv_getlength(bs);
3192     if (size < 0) {
3193         error_setg_errno(errp, -size, "bdrv_getlength failed");
3194         goto out;
3195     }
3196 
3197     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3198         assert(backup->format);
3199         if (source) {
3200             bdrv_img_create(backup->target, backup->format, source->filename,
3201                             source->drv->format_name, NULL,
3202                             size, flags, &local_err, false);
3203         } else {
3204             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3205                             size, flags, &local_err, false);
3206         }
3207     }
3208 
3209     if (local_err) {
3210         error_propagate(errp, local_err);
3211         goto out;
3212     }
3213 
3214     if (backup->format) {
3215         options = qdict_new();
3216         qdict_put(options, "driver", qstring_from_str(backup->format));
3217     }
3218 
3219     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3220     if (!target_bs) {
3221         goto out;
3222     }
3223 
3224     bdrv_set_aio_context(target_bs, aio_context);
3225 
3226     if (backup->has_bitmap) {
3227         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3228         if (!bmap) {
3229             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3230             bdrv_unref(target_bs);
3231             goto out;
3232         }
3233     }
3234 
3235     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3236                             backup->sync, bmap, backup->compress,
3237                             backup->on_source_error, backup->on_target_error,
3238                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3239     bdrv_unref(target_bs);
3240     if (local_err != NULL) {
3241         error_propagate(errp, local_err);
3242         goto out;
3243     }
3244 
3245 out:
3246     aio_context_release(aio_context);
3247     return job;
3248 }
3249 
3250 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3251 {
3252 
3253     BlockJob *job;
3254     job = do_drive_backup(arg, NULL, errp);
3255     if (job) {
3256         block_job_start(job);
3257     }
3258 }
3259 
3260 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3261 {
3262     return bdrv_named_nodes_list(errp);
3263 }
3264 
3265 BlockJob *do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
3266                              Error **errp)
3267 {
3268     BlockDriverState *bs;
3269     BlockDriverState *target_bs;
3270     Error *local_err = NULL;
3271     AioContext *aio_context;
3272     BlockJob *job = NULL;
3273 
3274     if (!backup->has_speed) {
3275         backup->speed = 0;
3276     }
3277     if (!backup->has_on_source_error) {
3278         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3279     }
3280     if (!backup->has_on_target_error) {
3281         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3282     }
3283     if (!backup->has_job_id) {
3284         backup->job_id = NULL;
3285     }
3286     if (!backup->has_compress) {
3287         backup->compress = false;
3288     }
3289 
3290     bs = qmp_get_root_bs(backup->device, errp);
3291     if (!bs) {
3292         return NULL;
3293     }
3294 
3295     aio_context = bdrv_get_aio_context(bs);
3296     aio_context_acquire(aio_context);
3297 
3298     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3299     if (!target_bs) {
3300         goto out;
3301     }
3302 
3303     if (bdrv_get_aio_context(target_bs) != aio_context) {
3304         if (!bdrv_has_blk(target_bs)) {
3305             /* The target BDS is not attached, we can safely move it to another
3306              * AioContext. */
3307             bdrv_set_aio_context(target_bs, aio_context);
3308         } else {
3309             error_setg(errp, "Target is attached to a different thread from "
3310                              "source.");
3311             goto out;
3312         }
3313     }
3314     job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
3315                             backup->sync, NULL, backup->compress,
3316                             backup->on_source_error, backup->on_target_error,
3317                             BLOCK_JOB_DEFAULT, NULL, NULL, txn, &local_err);
3318     if (local_err != NULL) {
3319         error_propagate(errp, local_err);
3320     }
3321 out:
3322     aio_context_release(aio_context);
3323     return job;
3324 }
3325 
3326 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3327 {
3328     BlockJob *job;
3329     job = do_blockdev_backup(arg, NULL, errp);
3330     if (job) {
3331         block_job_start(job);
3332     }
3333 }
3334 
3335 /* Parameter check and block job starting for drive mirroring.
3336  * Caller should hold @device and @target's aio context (must be the same).
3337  **/
3338 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3339                                    BlockDriverState *target,
3340                                    bool has_replaces, const char *replaces,
3341                                    enum MirrorSyncMode sync,
3342                                    BlockMirrorBackingMode backing_mode,
3343                                    bool has_speed, int64_t speed,
3344                                    bool has_granularity, uint32_t granularity,
3345                                    bool has_buf_size, int64_t buf_size,
3346                                    bool has_on_source_error,
3347                                    BlockdevOnError on_source_error,
3348                                    bool has_on_target_error,
3349                                    BlockdevOnError on_target_error,
3350                                    bool has_unmap, bool unmap,
3351                                    Error **errp)
3352 {
3353 
3354     if (!has_speed) {
3355         speed = 0;
3356     }
3357     if (!has_on_source_error) {
3358         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3359     }
3360     if (!has_on_target_error) {
3361         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3362     }
3363     if (!has_granularity) {
3364         granularity = 0;
3365     }
3366     if (!has_buf_size) {
3367         buf_size = 0;
3368     }
3369     if (!has_unmap) {
3370         unmap = true;
3371     }
3372 
3373     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3374         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3375                    "a value in range [512B, 64MB]");
3376         return;
3377     }
3378     if (granularity & (granularity - 1)) {
3379         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3380                    "power of 2");
3381         return;
3382     }
3383 
3384     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3385         return;
3386     }
3387     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3388         return;
3389     }
3390 
3391     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3392         sync = MIRROR_SYNC_MODE_FULL;
3393     }
3394 
3395     /* pass the node name to replace to mirror start since it's loose coupling
3396      * and will allow to check whether the node still exist at mirror completion
3397      */
3398     mirror_start(job_id, bs, target,
3399                  has_replaces ? replaces : NULL,
3400                  speed, granularity, buf_size, sync, backing_mode,
3401                  on_source_error, on_target_error, unmap, errp);
3402 }
3403 
3404 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3405 {
3406     BlockDriverState *bs;
3407     BlockDriverState *source, *target_bs;
3408     AioContext *aio_context;
3409     BlockMirrorBackingMode backing_mode;
3410     Error *local_err = NULL;
3411     QDict *options = NULL;
3412     int flags;
3413     int64_t size;
3414     const char *format = arg->format;
3415 
3416     bs = qmp_get_root_bs(arg->device, errp);
3417     if (!bs) {
3418         return;
3419     }
3420 
3421     aio_context = bdrv_get_aio_context(bs);
3422     aio_context_acquire(aio_context);
3423 
3424     if (!arg->has_mode) {
3425         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3426     }
3427 
3428     if (!arg->has_format) {
3429         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3430                   ? NULL : bs->drv->format_name);
3431     }
3432 
3433     flags = bs->open_flags | BDRV_O_RDWR;
3434     source = backing_bs(bs);
3435     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3436         arg->sync = MIRROR_SYNC_MODE_FULL;
3437     }
3438     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3439         source = bs;
3440     }
3441 
3442     size = bdrv_getlength(bs);
3443     if (size < 0) {
3444         error_setg_errno(errp, -size, "bdrv_getlength failed");
3445         goto out;
3446     }
3447 
3448     if (arg->has_replaces) {
3449         BlockDriverState *to_replace_bs;
3450         AioContext *replace_aio_context;
3451         int64_t replace_size;
3452 
3453         if (!arg->has_node_name) {
3454             error_setg(errp, "a node-name must be provided when replacing a"
3455                              " named node of the graph");
3456             goto out;
3457         }
3458 
3459         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3460 
3461         if (!to_replace_bs) {
3462             error_propagate(errp, local_err);
3463             goto out;
3464         }
3465 
3466         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3467         aio_context_acquire(replace_aio_context);
3468         replace_size = bdrv_getlength(to_replace_bs);
3469         aio_context_release(replace_aio_context);
3470 
3471         if (size != replace_size) {
3472             error_setg(errp, "cannot replace image with a mirror image of "
3473                              "different size");
3474             goto out;
3475         }
3476     }
3477 
3478     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3479         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3480     } else {
3481         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3482     }
3483 
3484     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3485         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3486     {
3487         /* create new image w/o backing file */
3488         assert(format);
3489         bdrv_img_create(arg->target, format,
3490                         NULL, NULL, NULL, size, flags, &local_err, false);
3491     } else {
3492         switch (arg->mode) {
3493         case NEW_IMAGE_MODE_EXISTING:
3494             break;
3495         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3496             /* create new image with backing file */
3497             bdrv_img_create(arg->target, format,
3498                             source->filename,
3499                             source->drv->format_name,
3500                             NULL, size, flags, &local_err, false);
3501             break;
3502         default:
3503             abort();
3504         }
3505     }
3506 
3507     if (local_err) {
3508         error_propagate(errp, local_err);
3509         goto out;
3510     }
3511 
3512     options = qdict_new();
3513     if (arg->has_node_name) {
3514         qdict_put(options, "node-name", qstring_from_str(arg->node_name));
3515     }
3516     if (format) {
3517         qdict_put(options, "driver", qstring_from_str(format));
3518     }
3519 
3520     /* Mirroring takes care of copy-on-write using the source's backing
3521      * file.
3522      */
3523     target_bs = bdrv_open(arg->target, NULL, options,
3524                           flags | BDRV_O_NO_BACKING, errp);
3525     if (!target_bs) {
3526         goto out;
3527     }
3528 
3529     bdrv_set_aio_context(target_bs, aio_context);
3530 
3531     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3532                            arg->has_replaces, arg->replaces, arg->sync,
3533                            backing_mode, arg->has_speed, arg->speed,
3534                            arg->has_granularity, arg->granularity,
3535                            arg->has_buf_size, arg->buf_size,
3536                            arg->has_on_source_error, arg->on_source_error,
3537                            arg->has_on_target_error, arg->on_target_error,
3538                            arg->has_unmap, arg->unmap,
3539                            &local_err);
3540     bdrv_unref(target_bs);
3541     error_propagate(errp, local_err);
3542 out:
3543     aio_context_release(aio_context);
3544 }
3545 
3546 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3547                          const char *device, const char *target,
3548                          bool has_replaces, const char *replaces,
3549                          MirrorSyncMode sync,
3550                          bool has_speed, int64_t speed,
3551                          bool has_granularity, uint32_t granularity,
3552                          bool has_buf_size, int64_t buf_size,
3553                          bool has_on_source_error,
3554                          BlockdevOnError on_source_error,
3555                          bool has_on_target_error,
3556                          BlockdevOnError on_target_error,
3557                          Error **errp)
3558 {
3559     BlockDriverState *bs;
3560     BlockDriverState *target_bs;
3561     AioContext *aio_context;
3562     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3563     Error *local_err = NULL;
3564 
3565     bs = qmp_get_root_bs(device, errp);
3566     if (!bs) {
3567         return;
3568     }
3569 
3570     target_bs = bdrv_lookup_bs(target, target, errp);
3571     if (!target_bs) {
3572         return;
3573     }
3574 
3575     aio_context = bdrv_get_aio_context(bs);
3576     aio_context_acquire(aio_context);
3577 
3578     bdrv_set_aio_context(target_bs, aio_context);
3579 
3580     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3581                            has_replaces, replaces, sync, backing_mode,
3582                            has_speed, speed,
3583                            has_granularity, granularity,
3584                            has_buf_size, buf_size,
3585                            has_on_source_error, on_source_error,
3586                            has_on_target_error, on_target_error,
3587                            true, true,
3588                            &local_err);
3589     error_propagate(errp, local_err);
3590 
3591     aio_context_release(aio_context);
3592 }
3593 
3594 /* Get a block job using its ID and acquire its AioContext */
3595 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3596                                 Error **errp)
3597 {
3598     BlockJob *job;
3599 
3600     assert(id != NULL);
3601 
3602     *aio_context = NULL;
3603 
3604     job = block_job_get(id);
3605 
3606     if (!job) {
3607         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3608                   "Block job '%s' not found", id);
3609         return NULL;
3610     }
3611 
3612     *aio_context = blk_get_aio_context(job->blk);
3613     aio_context_acquire(*aio_context);
3614 
3615     return job;
3616 }
3617 
3618 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3619 {
3620     AioContext *aio_context;
3621     BlockJob *job = find_block_job(device, &aio_context, errp);
3622 
3623     if (!job) {
3624         return;
3625     }
3626 
3627     block_job_set_speed(job, speed, errp);
3628     aio_context_release(aio_context);
3629 }
3630 
3631 void qmp_block_job_cancel(const char *device,
3632                           bool has_force, bool force, Error **errp)
3633 {
3634     AioContext *aio_context;
3635     BlockJob *job = find_block_job(device, &aio_context, errp);
3636 
3637     if (!job) {
3638         return;
3639     }
3640 
3641     if (!has_force) {
3642         force = false;
3643     }
3644 
3645     if (block_job_user_paused(job) && !force) {
3646         error_setg(errp, "The block job for device '%s' is currently paused",
3647                    device);
3648         goto out;
3649     }
3650 
3651     trace_qmp_block_job_cancel(job);
3652     block_job_cancel(job);
3653 out:
3654     aio_context_release(aio_context);
3655 }
3656 
3657 void qmp_block_job_pause(const char *device, Error **errp)
3658 {
3659     AioContext *aio_context;
3660     BlockJob *job = find_block_job(device, &aio_context, errp);
3661 
3662     if (!job || block_job_user_paused(job)) {
3663         return;
3664     }
3665 
3666     trace_qmp_block_job_pause(job);
3667     block_job_user_pause(job);
3668     aio_context_release(aio_context);
3669 }
3670 
3671 void qmp_block_job_resume(const char *device, Error **errp)
3672 {
3673     AioContext *aio_context;
3674     BlockJob *job = find_block_job(device, &aio_context, errp);
3675 
3676     if (!job || !block_job_user_paused(job)) {
3677         return;
3678     }
3679 
3680     trace_qmp_block_job_resume(job);
3681     block_job_iostatus_reset(job);
3682     block_job_user_resume(job);
3683     aio_context_release(aio_context);
3684 }
3685 
3686 void qmp_block_job_complete(const char *device, Error **errp)
3687 {
3688     AioContext *aio_context;
3689     BlockJob *job = find_block_job(device, &aio_context, errp);
3690 
3691     if (!job) {
3692         return;
3693     }
3694 
3695     trace_qmp_block_job_complete(job);
3696     block_job_complete(job, errp);
3697     aio_context_release(aio_context);
3698 }
3699 
3700 void qmp_change_backing_file(const char *device,
3701                              const char *image_node_name,
3702                              const char *backing_file,
3703                              Error **errp)
3704 {
3705     BlockDriverState *bs = NULL;
3706     AioContext *aio_context;
3707     BlockDriverState *image_bs = NULL;
3708     Error *local_err = NULL;
3709     bool ro;
3710     int open_flags;
3711     int ret;
3712 
3713     bs = qmp_get_root_bs(device, errp);
3714     if (!bs) {
3715         return;
3716     }
3717 
3718     aio_context = bdrv_get_aio_context(bs);
3719     aio_context_acquire(aio_context);
3720 
3721     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3722     if (local_err) {
3723         error_propagate(errp, local_err);
3724         goto out;
3725     }
3726 
3727     if (!image_bs) {
3728         error_setg(errp, "image file not found");
3729         goto out;
3730     }
3731 
3732     if (bdrv_find_base(image_bs) == image_bs) {
3733         error_setg(errp, "not allowing backing file change on an image "
3734                          "without a backing file");
3735         goto out;
3736     }
3737 
3738     /* even though we are not necessarily operating on bs, we need it to
3739      * determine if block ops are currently prohibited on the chain */
3740     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3741         goto out;
3742     }
3743 
3744     /* final sanity check */
3745     if (!bdrv_chain_contains(bs, image_bs)) {
3746         error_setg(errp, "'%s' and image file are not in the same chain",
3747                    device);
3748         goto out;
3749     }
3750 
3751     /* if not r/w, reopen to make r/w */
3752     open_flags = image_bs->open_flags;
3753     ro = bdrv_is_read_only(image_bs);
3754 
3755     if (ro) {
3756         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3757         if (local_err) {
3758             error_propagate(errp, local_err);
3759             goto out;
3760         }
3761     }
3762 
3763     ret = bdrv_change_backing_file(image_bs, backing_file,
3764                                image_bs->drv ? image_bs->drv->format_name : "");
3765 
3766     if (ret < 0) {
3767         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3768                          backing_file);
3769         /* don't exit here, so we can try to restore open flags if
3770          * appropriate */
3771     }
3772 
3773     if (ro) {
3774         bdrv_reopen(image_bs, open_flags, &local_err);
3775         error_propagate(errp, local_err);
3776     }
3777 
3778 out:
3779     aio_context_release(aio_context);
3780 }
3781 
3782 void hmp_drive_add_node(Monitor *mon, const char *optstr)
3783 {
3784     QemuOpts *opts;
3785     QDict *qdict;
3786     Error *local_err = NULL;
3787 
3788     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
3789     if (!opts) {
3790         return;
3791     }
3792 
3793     qdict = qemu_opts_to_qdict(opts, NULL);
3794 
3795     if (!qdict_get_try_str(qdict, "node-name")) {
3796         QDECREF(qdict);
3797         error_report("'node-name' needs to be specified");
3798         goto out;
3799     }
3800 
3801     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
3802     if (!bs) {
3803         error_report_err(local_err);
3804         goto out;
3805     }
3806 
3807     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3808 
3809 out:
3810     qemu_opts_del(opts);
3811 }
3812 
3813 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3814 {
3815     BlockDriverState *bs;
3816     QObject *obj;
3817     Visitor *v = qobject_output_visitor_new(&obj);
3818     QDict *qdict;
3819     Error *local_err = NULL;
3820 
3821     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
3822     if (local_err) {
3823         error_propagate(errp, local_err);
3824         goto fail;
3825     }
3826 
3827     visit_complete(v, &obj);
3828     qdict = qobject_to_qdict(obj);
3829 
3830     qdict_flatten(qdict);
3831 
3832     if (!qdict_get_try_str(qdict, "node-name")) {
3833         error_setg(errp, "'node-name' must be specified for the root node");
3834         goto fail;
3835     }
3836 
3837     bs = bds_tree_init(qdict, errp);
3838     if (!bs) {
3839         goto fail;
3840     }
3841 
3842     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3843 
3844     if (bs && bdrv_key_required(bs)) {
3845         QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3846         bdrv_unref(bs);
3847         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3848         goto fail;
3849     }
3850 
3851 fail:
3852     visit_free(v);
3853 }
3854 
3855 void qmp_x_blockdev_del(const char *node_name, Error **errp)
3856 {
3857     AioContext *aio_context;
3858     BlockDriverState *bs;
3859 
3860     bs = bdrv_find_node(node_name);
3861     if (!bs) {
3862         error_setg(errp, "Cannot find node %s", node_name);
3863         return;
3864     }
3865     if (bdrv_has_blk(bs)) {
3866         error_setg(errp, "Node %s is in use", node_name);
3867         return;
3868     }
3869     aio_context = bdrv_get_aio_context(bs);
3870     aio_context_acquire(aio_context);
3871 
3872     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3873         goto out;
3874     }
3875 
3876     if (!bs->monitor_list.tqe_prev) {
3877         error_setg(errp, "Node %s is not owned by the monitor",
3878                    bs->node_name);
3879         goto out;
3880     }
3881 
3882     if (bs->refcnt > 1) {
3883         error_setg(errp, "Block device %s is in use",
3884                    bdrv_get_device_or_node_name(bs));
3885         goto out;
3886     }
3887 
3888     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3889     bdrv_unref(bs);
3890 
3891 out:
3892     aio_context_release(aio_context);
3893 }
3894 
3895 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3896                                   const char *child_name)
3897 {
3898     BdrvChild *child;
3899 
3900     QLIST_FOREACH(child, &parent_bs->children, next) {
3901         if (strcmp(child->name, child_name) == 0) {
3902             return child;
3903         }
3904     }
3905 
3906     return NULL;
3907 }
3908 
3909 void qmp_x_blockdev_change(const char *parent, bool has_child,
3910                            const char *child, bool has_node,
3911                            const char *node, Error **errp)
3912 {
3913     BlockDriverState *parent_bs, *new_bs = NULL;
3914     BdrvChild *p_child;
3915 
3916     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3917     if (!parent_bs) {
3918         return;
3919     }
3920 
3921     if (has_child == has_node) {
3922         if (has_child) {
3923             error_setg(errp, "The parameters child and node are in conflict");
3924         } else {
3925             error_setg(errp, "Either child or node must be specified");
3926         }
3927         return;
3928     }
3929 
3930     if (has_child) {
3931         p_child = bdrv_find_child(parent_bs, child);
3932         if (!p_child) {
3933             error_setg(errp, "Node '%s' does not have child '%s'",
3934                        parent, child);
3935             return;
3936         }
3937         bdrv_del_child(parent_bs, p_child, errp);
3938     }
3939 
3940     if (has_node) {
3941         new_bs = bdrv_find_node(node);
3942         if (!new_bs) {
3943             error_setg(errp, "Node '%s' not found", node);
3944             return;
3945         }
3946         bdrv_add_child(parent_bs, new_bs, errp);
3947     }
3948 }
3949 
3950 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3951 {
3952     BlockJobInfoList *head = NULL, **p_next = &head;
3953     BlockJob *job;
3954 
3955     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
3956         BlockJobInfoList *elem;
3957         AioContext *aio_context;
3958 
3959         if (block_job_is_internal(job)) {
3960             continue;
3961         }
3962         elem = g_new0(BlockJobInfoList, 1);
3963         aio_context = blk_get_aio_context(job->blk);
3964         aio_context_acquire(aio_context);
3965         elem->value = block_job_query(job, errp);
3966         aio_context_release(aio_context);
3967         if (!elem->value) {
3968             g_free(elem);
3969             qapi_free_BlockJobInfoList(head);
3970             return NULL;
3971         }
3972         *p_next = elem;
3973         p_next = &elem->next;
3974     }
3975 
3976     return head;
3977 }
3978 
3979 QemuOptsList qemu_common_drive_opts = {
3980     .name = "drive",
3981     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3982     .desc = {
3983         {
3984             .name = "snapshot",
3985             .type = QEMU_OPT_BOOL,
3986             .help = "enable/disable snapshot mode",
3987         },{
3988             .name = "aio",
3989             .type = QEMU_OPT_STRING,
3990             .help = "host AIO implementation (threads, native)",
3991         },{
3992             .name = BDRV_OPT_CACHE_WB,
3993             .type = QEMU_OPT_BOOL,
3994             .help = "Enable writeback mode",
3995         },{
3996             .name = "format",
3997             .type = QEMU_OPT_STRING,
3998             .help = "disk format (raw, qcow2, ...)",
3999         },{
4000             .name = "rerror",
4001             .type = QEMU_OPT_STRING,
4002             .help = "read error action",
4003         },{
4004             .name = "werror",
4005             .type = QEMU_OPT_STRING,
4006             .help = "write error action",
4007         },{
4008             .name = BDRV_OPT_READ_ONLY,
4009             .type = QEMU_OPT_BOOL,
4010             .help = "open drive file as read-only",
4011         },
4012 
4013         THROTTLE_OPTS,
4014 
4015         {
4016             .name = "throttling.group",
4017             .type = QEMU_OPT_STRING,
4018             .help = "name of the block throttling group",
4019         },{
4020             .name = "copy-on-read",
4021             .type = QEMU_OPT_BOOL,
4022             .help = "copy read data from backing file into image file",
4023         },{
4024             .name = "detect-zeroes",
4025             .type = QEMU_OPT_STRING,
4026             .help = "try to optimize zero writes (off, on, unmap)",
4027         },{
4028             .name = "stats-account-invalid",
4029             .type = QEMU_OPT_BOOL,
4030             .help = "whether to account for invalid I/O operations "
4031                     "in the statistics",
4032         },{
4033             .name = "stats-account-failed",
4034             .type = QEMU_OPT_BOOL,
4035             .help = "whether to account for failed I/O operations "
4036                     "in the statistics",
4037         },
4038         { /* end of list */ }
4039     },
4040 };
4041 
4042 QemuOptsList qemu_drive_opts = {
4043     .name = "drive",
4044     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4045     .desc = {
4046         /*
4047          * no elements => accept any params
4048          * validation will happen later
4049          */
4050         { /* end of list */ }
4051     },
4052 };
4053