xref: /qemu/blockdev.c (revision 814bb12a)
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 "trace.h"
52 #include "sysemu/arch_init.h"
53 #include "qemu/cutils.h"
54 #include "qemu/help_option.h"
55 
56 static QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
57     QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
58 
59 static int do_open_tray(const char *blk_name, const char *qdev_id,
60                         bool force, Error **errp);
61 
62 static const char *const if_name[IF_COUNT] = {
63     [IF_NONE] = "none",
64     [IF_IDE] = "ide",
65     [IF_SCSI] = "scsi",
66     [IF_FLOPPY] = "floppy",
67     [IF_PFLASH] = "pflash",
68     [IF_MTD] = "mtd",
69     [IF_SD] = "sd",
70     [IF_VIRTIO] = "virtio",
71     [IF_XEN] = "xen",
72 };
73 
74 static int if_max_devs[IF_COUNT] = {
75     /*
76      * Do not change these numbers!  They govern how drive option
77      * index maps to unit and bus.  That mapping is ABI.
78      *
79      * All controllers used to implement if=T drives need to support
80      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
81      * Otherwise, some index values map to "impossible" bus, unit
82      * values.
83      *
84      * For instance, if you change [IF_SCSI] to 255, -drive
85      * if=scsi,index=12 no longer means bus=1,unit=5, but
86      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
87      * the drive can't be set up.  Regression.
88      */
89     [IF_IDE] = 2,
90     [IF_SCSI] = 7,
91 };
92 
93 /**
94  * Boards may call this to offer board-by-board overrides
95  * of the default, global values.
96  */
97 void override_max_devs(BlockInterfaceType type, int max_devs)
98 {
99     BlockBackend *blk;
100     DriveInfo *dinfo;
101 
102     if (max_devs <= 0) {
103         return;
104     }
105 
106     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
107         dinfo = blk_legacy_dinfo(blk);
108         if (dinfo->type == type) {
109             fprintf(stderr, "Cannot override units-per-bus property of"
110                     " the %s interface, because a drive of that type has"
111                     " already been added.\n", if_name[type]);
112             g_assert_not_reached();
113         }
114     }
115 
116     if_max_devs[type] = max_devs;
117 }
118 
119 /*
120  * We automatically delete the drive when a device using it gets
121  * unplugged.  Questionable feature, but we can't just drop it.
122  * Device models call blockdev_mark_auto_del() to schedule the
123  * automatic deletion, and generic qdev code calls blockdev_auto_del()
124  * when deletion is actually safe.
125  */
126 void blockdev_mark_auto_del(BlockBackend *blk)
127 {
128     DriveInfo *dinfo = blk_legacy_dinfo(blk);
129     BlockDriverState *bs = blk_bs(blk);
130     AioContext *aio_context;
131 
132     if (!dinfo) {
133         return;
134     }
135 
136     if (bs) {
137         aio_context = bdrv_get_aio_context(bs);
138         aio_context_acquire(aio_context);
139 
140         if (bs->job) {
141             block_job_cancel(bs->job);
142         }
143 
144         aio_context_release(aio_context);
145     }
146 
147     dinfo->auto_del = 1;
148 }
149 
150 void blockdev_auto_del(BlockBackend *blk)
151 {
152     DriveInfo *dinfo = blk_legacy_dinfo(blk);
153 
154     if (dinfo && dinfo->auto_del) {
155         monitor_remove_blk(blk);
156         blk_unref(blk);
157     }
158 }
159 
160 /**
161  * Returns the current mapping of how many units per bus
162  * a particular interface can support.
163  *
164  *  A positive integer indicates n units per bus.
165  *  0 implies the mapping has not been established.
166  * -1 indicates an invalid BlockInterfaceType was given.
167  */
168 int drive_get_max_devs(BlockInterfaceType type)
169 {
170     if (type >= IF_IDE && type < IF_COUNT) {
171         return if_max_devs[type];
172     }
173 
174     return -1;
175 }
176 
177 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
178 {
179     int max_devs = if_max_devs[type];
180     return max_devs ? index / max_devs : 0;
181 }
182 
183 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
184 {
185     int max_devs = if_max_devs[type];
186     return max_devs ? index % max_devs : index;
187 }
188 
189 QemuOpts *drive_def(const char *optstr)
190 {
191     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
192 }
193 
194 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
195                     const char *optstr)
196 {
197     QemuOpts *opts;
198 
199     opts = drive_def(optstr);
200     if (!opts) {
201         return NULL;
202     }
203     if (type != IF_DEFAULT) {
204         qemu_opt_set(opts, "if", if_name[type], &error_abort);
205     }
206     if (index >= 0) {
207         qemu_opt_set_number(opts, "index", index, &error_abort);
208     }
209     if (file)
210         qemu_opt_set(opts, "file", file, &error_abort);
211     return opts;
212 }
213 
214 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
215 {
216     BlockBackend *blk;
217     DriveInfo *dinfo;
218 
219     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
220         dinfo = blk_legacy_dinfo(blk);
221         if (dinfo && dinfo->type == type
222             && dinfo->bus == bus && dinfo->unit == unit) {
223             return dinfo;
224         }
225     }
226 
227     return NULL;
228 }
229 
230 bool drive_check_orphaned(void)
231 {
232     BlockBackend *blk;
233     DriveInfo *dinfo;
234     bool rs = false;
235 
236     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
237         dinfo = blk_legacy_dinfo(blk);
238         /* If dinfo->bdrv->dev is NULL, it has no device attached. */
239         /* Unless this is a default drive, this may be an oversight. */
240         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
241             dinfo->type != IF_NONE) {
242             fprintf(stderr, "Warning: Orphaned drive without device: "
243                     "id=%s,file=%s,if=%s,bus=%d,unit=%d\n",
244                     blk_name(blk), blk_bs(blk) ? blk_bs(blk)->filename : "",
245                     if_name[dinfo->type], dinfo->bus, dinfo->unit);
246             rs = true;
247         }
248     }
249 
250     return rs;
251 }
252 
253 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
254 {
255     return drive_get(type,
256                      drive_index_to_bus_id(type, index),
257                      drive_index_to_unit_id(type, index));
258 }
259 
260 int drive_get_max_bus(BlockInterfaceType type)
261 {
262     int max_bus;
263     BlockBackend *blk;
264     DriveInfo *dinfo;
265 
266     max_bus = -1;
267     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
268         dinfo = blk_legacy_dinfo(blk);
269         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
270             max_bus = dinfo->bus;
271         }
272     }
273     return max_bus;
274 }
275 
276 /* Get a block device.  This should only be used for single-drive devices
277    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
278    appropriate bus.  */
279 DriveInfo *drive_get_next(BlockInterfaceType type)
280 {
281     static int next_block_unit[IF_COUNT];
282 
283     return drive_get(type, 0, next_block_unit[type]++);
284 }
285 
286 static void bdrv_format_print(void *opaque, const char *name)
287 {
288     error_printf(" %s", name);
289 }
290 
291 typedef struct {
292     QEMUBH *bh;
293     BlockDriverState *bs;
294 } BDRVPutRefBH;
295 
296 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
297 {
298     if (!strcmp(buf, "ignore")) {
299         return BLOCKDEV_ON_ERROR_IGNORE;
300     } else if (!is_read && !strcmp(buf, "enospc")) {
301         return BLOCKDEV_ON_ERROR_ENOSPC;
302     } else if (!strcmp(buf, "stop")) {
303         return BLOCKDEV_ON_ERROR_STOP;
304     } else if (!strcmp(buf, "report")) {
305         return BLOCKDEV_ON_ERROR_REPORT;
306     } else {
307         error_setg(errp, "'%s' invalid %s error action",
308                    buf, is_read ? "read" : "write");
309         return -1;
310     }
311 }
312 
313 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
314                                   Error **errp)
315 {
316     const QListEntry *entry;
317     for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
318         switch (qobject_type(entry->value)) {
319 
320         case QTYPE_QSTRING: {
321             unsigned long long length;
322             const char *str = qstring_get_str(qobject_to_qstring(entry->value));
323             if (parse_uint_full(str, &length, 10) == 0 &&
324                 length > 0 && length <= UINT_MAX) {
325                 block_acct_add_interval(stats, (unsigned) length);
326             } else {
327                 error_setg(errp, "Invalid interval length: %s", str);
328                 return false;
329             }
330             break;
331         }
332 
333         case QTYPE_QINT: {
334             int64_t length = qint_get_int(qobject_to_qint(entry->value));
335             if (length > 0 && length <= UINT_MAX) {
336                 block_acct_add_interval(stats, (unsigned) length);
337             } else {
338                 error_setg(errp, "Invalid interval length: %" PRId64, length);
339                 return false;
340             }
341             break;
342         }
343 
344         default:
345             error_setg(errp, "The specification of stats-intervals is invalid");
346             return false;
347         }
348     }
349     return true;
350 }
351 
352 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
353 
354 /* All parameters but @opts are optional and may be set to NULL. */
355 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
356     const char **throttling_group, ThrottleConfig *throttle_cfg,
357     BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
358 {
359     Error *local_error = NULL;
360     const char *aio;
361 
362     if (bdrv_flags) {
363         if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
364             *bdrv_flags |= BDRV_O_COPY_ON_READ;
365         }
366 
367         if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
368             if (!strcmp(aio, "native")) {
369                 *bdrv_flags |= BDRV_O_NATIVE_AIO;
370             } else if (!strcmp(aio, "threads")) {
371                 /* this is the default */
372             } else {
373                error_setg(errp, "invalid aio option");
374                return;
375             }
376         }
377     }
378 
379     /* disk I/O throttling */
380     if (throttling_group) {
381         *throttling_group = qemu_opt_get(opts, "throttling.group");
382     }
383 
384     if (throttle_cfg) {
385         throttle_config_init(throttle_cfg);
386         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
387             qemu_opt_get_number(opts, "throttling.bps-total", 0);
388         throttle_cfg->buckets[THROTTLE_BPS_READ].avg  =
389             qemu_opt_get_number(opts, "throttling.bps-read", 0);
390         throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
391             qemu_opt_get_number(opts, "throttling.bps-write", 0);
392         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
393             qemu_opt_get_number(opts, "throttling.iops-total", 0);
394         throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
395             qemu_opt_get_number(opts, "throttling.iops-read", 0);
396         throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
397             qemu_opt_get_number(opts, "throttling.iops-write", 0);
398 
399         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
400             qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
401         throttle_cfg->buckets[THROTTLE_BPS_READ].max  =
402             qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
403         throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
404             qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
405         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
406             qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
407         throttle_cfg->buckets[THROTTLE_OPS_READ].max =
408             qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
409         throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
410             qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
411 
412         throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
413             qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
414         throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length  =
415             qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
416         throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
417             qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
418         throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
419             qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
420         throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
421             qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
422         throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
423             qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
424 
425         throttle_cfg->op_size =
426             qemu_opt_get_number(opts, "throttling.iops-size", 0);
427 
428         if (!throttle_is_valid(throttle_cfg, errp)) {
429             return;
430         }
431     }
432 
433     if (detect_zeroes) {
434         *detect_zeroes =
435             qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
436                             qemu_opt_get(opts, "detect-zeroes"),
437                             BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX,
438                             BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
439                             &local_error);
440         if (local_error) {
441             error_propagate(errp, local_error);
442             return;
443         }
444     }
445 }
446 
447 /* Takes the ownership of bs_opts */
448 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
449                                    Error **errp)
450 {
451     const char *buf;
452     int bdrv_flags = 0;
453     int on_read_error, on_write_error;
454     bool account_invalid, account_failed;
455     bool writethrough, read_only;
456     BlockBackend *blk;
457     BlockDriverState *bs;
458     ThrottleConfig cfg;
459     int snapshot = 0;
460     Error *error = NULL;
461     QemuOpts *opts;
462     QDict *interval_dict = NULL;
463     QList *interval_list = NULL;
464     const char *id;
465     BlockdevDetectZeroesOptions detect_zeroes =
466         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
467     const char *throttling_group = NULL;
468 
469     /* Check common options by copying from bs_opts to opts, all other options
470      * stay in bs_opts for processing by bdrv_open(). */
471     id = qdict_get_try_str(bs_opts, "id");
472     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
473     if (error) {
474         error_propagate(errp, error);
475         goto err_no_opts;
476     }
477 
478     qemu_opts_absorb_qdict(opts, bs_opts, &error);
479     if (error) {
480         error_propagate(errp, error);
481         goto early_err;
482     }
483 
484     if (id) {
485         qdict_del(bs_opts, "id");
486     }
487 
488     /* extract parameters */
489     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
490 
491     account_invalid = qemu_opt_get_bool(opts, "stats-account-invalid", true);
492     account_failed = qemu_opt_get_bool(opts, "stats-account-failed", true);
493 
494     writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
495 
496     id = qemu_opts_id(opts);
497 
498     qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
499     qdict_array_split(interval_dict, &interval_list);
500 
501     if (qdict_size(interval_dict) != 0) {
502         error_setg(errp, "Invalid option stats-intervals.%s",
503                    qdict_first(interval_dict)->key);
504         goto early_err;
505     }
506 
507     extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
508                                     &detect_zeroes, &error);
509     if (error) {
510         error_propagate(errp, error);
511         goto early_err;
512     }
513 
514     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
515         if (is_help_option(buf)) {
516             error_printf("Supported formats:");
517             bdrv_iterate_format(bdrv_format_print, NULL);
518             error_printf("\n");
519             goto early_err;
520         }
521 
522         if (qdict_haskey(bs_opts, "driver")) {
523             error_setg(errp, "Cannot specify both 'driver' and 'format'");
524             goto early_err;
525         }
526         qdict_put(bs_opts, "driver", qstring_from_str(buf));
527     }
528 
529     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
530     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
531         on_write_error = parse_block_error_action(buf, 0, &error);
532         if (error) {
533             error_propagate(errp, error);
534             goto early_err;
535         }
536     }
537 
538     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
539     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
540         on_read_error = parse_block_error_action(buf, 1, &error);
541         if (error) {
542             error_propagate(errp, error);
543             goto early_err;
544         }
545     }
546 
547     if (snapshot) {
548         bdrv_flags |= BDRV_O_SNAPSHOT;
549     }
550 
551     read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
552 
553     /* init */
554     if ((!file || !*file) && !qdict_size(bs_opts)) {
555         BlockBackendRootState *blk_rs;
556 
557         blk = blk_new();
558         blk_rs = blk_get_root_state(blk);
559         blk_rs->open_flags    = bdrv_flags;
560         blk_rs->read_only     = read_only;
561         blk_rs->detect_zeroes = detect_zeroes;
562 
563         QDECREF(bs_opts);
564     } else {
565         if (file && !*file) {
566             file = NULL;
567         }
568 
569         /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
570          * with other callers) rather than what we want as the real defaults.
571          * Apply the defaults here instead. */
572         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
573         qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
574         qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
575                               read_only ? "on" : "off");
576         assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
577 
578         if (runstate_check(RUN_STATE_INMIGRATE)) {
579             bdrv_flags |= BDRV_O_INACTIVE;
580         }
581 
582         blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
583         if (!blk) {
584             goto err_no_bs_opts;
585         }
586         bs = blk_bs(blk);
587 
588         bs->detect_zeroes = detect_zeroes;
589 
590         if (bdrv_key_required(bs)) {
591             autostart = 0;
592         }
593 
594         block_acct_init(blk_get_stats(blk), account_invalid, account_failed);
595 
596         if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
597             blk_unref(blk);
598             blk = NULL;
599             goto err_no_bs_opts;
600         }
601     }
602 
603     /* disk I/O throttling */
604     if (throttle_enabled(&cfg)) {
605         if (!throttling_group) {
606             throttling_group = id;
607         }
608         blk_io_limits_enable(blk, throttling_group);
609         blk_set_io_limits(blk, &cfg);
610     }
611 
612     blk_set_enable_write_cache(blk, !writethrough);
613     blk_set_on_error(blk, on_read_error, on_write_error);
614 
615     if (!monitor_add_blk(blk, id, errp)) {
616         blk_unref(blk);
617         blk = NULL;
618         goto err_no_bs_opts;
619     }
620 
621 err_no_bs_opts:
622     qemu_opts_del(opts);
623     QDECREF(interval_dict);
624     QDECREF(interval_list);
625     return blk;
626 
627 early_err:
628     qemu_opts_del(opts);
629     QDECREF(interval_dict);
630     QDECREF(interval_list);
631 err_no_opts:
632     QDECREF(bs_opts);
633     return NULL;
634 }
635 
636 /* Takes the ownership of bs_opts */
637 static BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
638 {
639     int bdrv_flags = 0;
640 
641     /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
642      * with other callers) rather than what we want as the real defaults.
643      * Apply the defaults here instead. */
644     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
645     qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
646     qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
647 
648     if (runstate_check(RUN_STATE_INMIGRATE)) {
649         bdrv_flags |= BDRV_O_INACTIVE;
650     }
651 
652     return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
653 }
654 
655 void blockdev_close_all_bdrv_states(void)
656 {
657     BlockDriverState *bs, *next_bs;
658 
659     QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
660         AioContext *ctx = bdrv_get_aio_context(bs);
661 
662         aio_context_acquire(ctx);
663         bdrv_unref(bs);
664         aio_context_release(ctx);
665     }
666 }
667 
668 /* Iterates over the list of monitor-owned BlockDriverStates */
669 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
670 {
671     return bs ? QTAILQ_NEXT(bs, monitor_list)
672               : QTAILQ_FIRST(&monitor_bdrv_states);
673 }
674 
675 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
676                             Error **errp)
677 {
678     const char *value;
679 
680     value = qemu_opt_get(opts, from);
681     if (value) {
682         if (qemu_opt_find(opts, to)) {
683             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
684                        "same time", to, from);
685             return;
686         }
687     }
688 
689     /* rename all items in opts */
690     while ((value = qemu_opt_get(opts, from))) {
691         qemu_opt_set(opts, to, value, &error_abort);
692         qemu_opt_unset(opts, from);
693     }
694 }
695 
696 QemuOptsList qemu_legacy_drive_opts = {
697     .name = "drive",
698     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
699     .desc = {
700         {
701             .name = "bus",
702             .type = QEMU_OPT_NUMBER,
703             .help = "bus number",
704         },{
705             .name = "unit",
706             .type = QEMU_OPT_NUMBER,
707             .help = "unit number (i.e. lun for scsi)",
708         },{
709             .name = "index",
710             .type = QEMU_OPT_NUMBER,
711             .help = "index number",
712         },{
713             .name = "media",
714             .type = QEMU_OPT_STRING,
715             .help = "media type (disk, cdrom)",
716         },{
717             .name = "if",
718             .type = QEMU_OPT_STRING,
719             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
720         },{
721             .name = "cyls",
722             .type = QEMU_OPT_NUMBER,
723             .help = "number of cylinders (ide disk geometry)",
724         },{
725             .name = "heads",
726             .type = QEMU_OPT_NUMBER,
727             .help = "number of heads (ide disk geometry)",
728         },{
729             .name = "secs",
730             .type = QEMU_OPT_NUMBER,
731             .help = "number of sectors (ide disk geometry)",
732         },{
733             .name = "trans",
734             .type = QEMU_OPT_STRING,
735             .help = "chs translation (auto, lba, none)",
736         },{
737             .name = "boot",
738             .type = QEMU_OPT_BOOL,
739             .help = "(deprecated, ignored)",
740         },{
741             .name = "addr",
742             .type = QEMU_OPT_STRING,
743             .help = "pci address (virtio only)",
744         },{
745             .name = "serial",
746             .type = QEMU_OPT_STRING,
747             .help = "disk serial number",
748         },{
749             .name = "file",
750             .type = QEMU_OPT_STRING,
751             .help = "file name",
752         },
753 
754         /* Options that are passed on, but have special semantics with -drive */
755         {
756             .name = BDRV_OPT_READ_ONLY,
757             .type = QEMU_OPT_BOOL,
758             .help = "open drive file as read-only",
759         },{
760             .name = "rerror",
761             .type = QEMU_OPT_STRING,
762             .help = "read error action",
763         },{
764             .name = "werror",
765             .type = QEMU_OPT_STRING,
766             .help = "write error action",
767         },{
768             .name = "copy-on-read",
769             .type = QEMU_OPT_BOOL,
770             .help = "copy read data from backing file into image file",
771         },
772 
773         { /* end of list */ }
774     },
775 };
776 
777 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
778 {
779     const char *value;
780     BlockBackend *blk;
781     DriveInfo *dinfo = NULL;
782     QDict *bs_opts;
783     QemuOpts *legacy_opts;
784     DriveMediaType media = MEDIA_DISK;
785     BlockInterfaceType type;
786     int cyls, heads, secs, translation;
787     int max_devs, bus_id, unit_id, index;
788     const char *devaddr;
789     const char *werror, *rerror;
790     bool read_only = false;
791     bool copy_on_read;
792     const char *serial;
793     const char *filename;
794     Error *local_err = NULL;
795     int i;
796 
797     /* Change legacy command line options into QMP ones */
798     static const struct {
799         const char *from;
800         const char *to;
801     } opt_renames[] = {
802         { "iops",           "throttling.iops-total" },
803         { "iops_rd",        "throttling.iops-read" },
804         { "iops_wr",        "throttling.iops-write" },
805 
806         { "bps",            "throttling.bps-total" },
807         { "bps_rd",         "throttling.bps-read" },
808         { "bps_wr",         "throttling.bps-write" },
809 
810         { "iops_max",       "throttling.iops-total-max" },
811         { "iops_rd_max",    "throttling.iops-read-max" },
812         { "iops_wr_max",    "throttling.iops-write-max" },
813 
814         { "bps_max",        "throttling.bps-total-max" },
815         { "bps_rd_max",     "throttling.bps-read-max" },
816         { "bps_wr_max",     "throttling.bps-write-max" },
817 
818         { "iops_size",      "throttling.iops-size" },
819 
820         { "group",          "throttling.group" },
821 
822         { "readonly",       BDRV_OPT_READ_ONLY },
823     };
824 
825     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
826         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
827                         &local_err);
828         if (local_err) {
829             error_report_err(local_err);
830             return NULL;
831         }
832     }
833 
834     value = qemu_opt_get(all_opts, "cache");
835     if (value) {
836         int flags = 0;
837         bool writethrough;
838 
839         if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
840             error_report("invalid cache option");
841             return NULL;
842         }
843 
844         /* Specific options take precedence */
845         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
846             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
847                               !writethrough, &error_abort);
848         }
849         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
850             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
851                               !!(flags & BDRV_O_NOCACHE), &error_abort);
852         }
853         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
854             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
855                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
856         }
857         qemu_opt_unset(all_opts, "cache");
858     }
859 
860     /* Get a QDict for processing the options */
861     bs_opts = qdict_new();
862     qemu_opts_to_qdict(all_opts, bs_opts);
863 
864     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
865                                    &error_abort);
866     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
867     if (local_err) {
868         error_report_err(local_err);
869         goto fail;
870     }
871 
872     /* Deprecated option boot=[on|off] */
873     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
874         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
875                 "ignored. Future versions will reject this parameter. Please "
876                 "update your scripts.\n");
877     }
878 
879     /* Media type */
880     value = qemu_opt_get(legacy_opts, "media");
881     if (value) {
882         if (!strcmp(value, "disk")) {
883             media = MEDIA_DISK;
884         } else if (!strcmp(value, "cdrom")) {
885             media = MEDIA_CDROM;
886             read_only = true;
887         } else {
888             error_report("'%s' invalid media", value);
889             goto fail;
890         }
891     }
892 
893     /* copy-on-read is disabled with a warning for read-only devices */
894     read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
895     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
896 
897     if (read_only && copy_on_read) {
898         error_report("warning: disabling copy-on-read on read-only drive");
899         copy_on_read = false;
900     }
901 
902     qdict_put(bs_opts, BDRV_OPT_READ_ONLY,
903               qstring_from_str(read_only ? "on" : "off"));
904     qdict_put(bs_opts, "copy-on-read",
905               qstring_from_str(copy_on_read ? "on" :"off"));
906 
907     /* Controller type */
908     value = qemu_opt_get(legacy_opts, "if");
909     if (value) {
910         for (type = 0;
911              type < IF_COUNT && strcmp(value, if_name[type]);
912              type++) {
913         }
914         if (type == IF_COUNT) {
915             error_report("unsupported bus type '%s'", value);
916             goto fail;
917         }
918     } else {
919         type = block_default_type;
920     }
921 
922     /* Geometry */
923     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
924     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
925     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
926 
927     if (cyls || heads || secs) {
928         if (cyls < 1) {
929             error_report("invalid physical cyls number");
930             goto fail;
931         }
932         if (heads < 1) {
933             error_report("invalid physical heads number");
934             goto fail;
935         }
936         if (secs < 1) {
937             error_report("invalid physical secs number");
938             goto fail;
939         }
940     }
941 
942     translation = BIOS_ATA_TRANSLATION_AUTO;
943     value = qemu_opt_get(legacy_opts, "trans");
944     if (value != NULL) {
945         if (!cyls) {
946             error_report("'%s' trans must be used with cyls, heads and secs",
947                          value);
948             goto fail;
949         }
950         if (!strcmp(value, "none")) {
951             translation = BIOS_ATA_TRANSLATION_NONE;
952         } else if (!strcmp(value, "lba")) {
953             translation = BIOS_ATA_TRANSLATION_LBA;
954         } else if (!strcmp(value, "large")) {
955             translation = BIOS_ATA_TRANSLATION_LARGE;
956         } else if (!strcmp(value, "rechs")) {
957             translation = BIOS_ATA_TRANSLATION_RECHS;
958         } else if (!strcmp(value, "auto")) {
959             translation = BIOS_ATA_TRANSLATION_AUTO;
960         } else {
961             error_report("'%s' invalid translation type", value);
962             goto fail;
963         }
964     }
965 
966     if (media == MEDIA_CDROM) {
967         if (cyls || secs || heads) {
968             error_report("CHS can't be set with media=cdrom");
969             goto fail;
970         }
971     }
972 
973     /* Device address specified by bus/unit or index.
974      * If none was specified, try to find the first free one. */
975     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
976     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
977     index   = qemu_opt_get_number(legacy_opts, "index", -1);
978 
979     max_devs = if_max_devs[type];
980 
981     if (index != -1) {
982         if (bus_id != 0 || unit_id != -1) {
983             error_report("index cannot be used with bus and unit");
984             goto fail;
985         }
986         bus_id = drive_index_to_bus_id(type, index);
987         unit_id = drive_index_to_unit_id(type, index);
988     }
989 
990     if (unit_id == -1) {
991        unit_id = 0;
992        while (drive_get(type, bus_id, unit_id) != NULL) {
993            unit_id++;
994            if (max_devs && unit_id >= max_devs) {
995                unit_id -= max_devs;
996                bus_id++;
997            }
998        }
999     }
1000 
1001     if (max_devs && unit_id >= max_devs) {
1002         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
1003         goto fail;
1004     }
1005 
1006     if (drive_get(type, bus_id, unit_id) != NULL) {
1007         error_report("drive with bus=%d, unit=%d (index=%d) exists",
1008                      bus_id, unit_id, index);
1009         goto fail;
1010     }
1011 
1012     /* Serial number */
1013     serial = qemu_opt_get(legacy_opts, "serial");
1014 
1015     /* no id supplied -> create one */
1016     if (qemu_opts_id(all_opts) == NULL) {
1017         char *new_id;
1018         const char *mediastr = "";
1019         if (type == IF_IDE || type == IF_SCSI) {
1020             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
1021         }
1022         if (max_devs) {
1023             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
1024                                      mediastr, unit_id);
1025         } else {
1026             new_id = g_strdup_printf("%s%s%i", if_name[type],
1027                                      mediastr, unit_id);
1028         }
1029         qdict_put(bs_opts, "id", qstring_from_str(new_id));
1030         g_free(new_id);
1031     }
1032 
1033     /* Add virtio block device */
1034     devaddr = qemu_opt_get(legacy_opts, "addr");
1035     if (devaddr && type != IF_VIRTIO) {
1036         error_report("addr is not supported by this bus type");
1037         goto fail;
1038     }
1039 
1040     if (type == IF_VIRTIO) {
1041         QemuOpts *devopts;
1042         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
1043                                    &error_abort);
1044         if (arch_type == QEMU_ARCH_S390X) {
1045             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
1046         } else {
1047             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
1048         }
1049         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
1050                      &error_abort);
1051         if (devaddr) {
1052             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
1053         }
1054     }
1055 
1056     filename = qemu_opt_get(legacy_opts, "file");
1057 
1058     /* Check werror/rerror compatibility with if=... */
1059     werror = qemu_opt_get(legacy_opts, "werror");
1060     if (werror != NULL) {
1061         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
1062             type != IF_NONE) {
1063             error_report("werror is not supported by this bus type");
1064             goto fail;
1065         }
1066         qdict_put(bs_opts, "werror", qstring_from_str(werror));
1067     }
1068 
1069     rerror = qemu_opt_get(legacy_opts, "rerror");
1070     if (rerror != NULL) {
1071         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
1072             type != IF_NONE) {
1073             error_report("rerror is not supported by this bus type");
1074             goto fail;
1075         }
1076         qdict_put(bs_opts, "rerror", qstring_from_str(rerror));
1077     }
1078 
1079     /* Actual block device init: Functionality shared with blockdev-add */
1080     blk = blockdev_init(filename, bs_opts, &local_err);
1081     bs_opts = NULL;
1082     if (!blk) {
1083         if (local_err) {
1084             error_report_err(local_err);
1085         }
1086         goto fail;
1087     } else {
1088         assert(!local_err);
1089     }
1090 
1091     /* Create legacy DriveInfo */
1092     dinfo = g_malloc0(sizeof(*dinfo));
1093     dinfo->opts = all_opts;
1094 
1095     dinfo->cyls = cyls;
1096     dinfo->heads = heads;
1097     dinfo->secs = secs;
1098     dinfo->trans = translation;
1099 
1100     dinfo->type = type;
1101     dinfo->bus = bus_id;
1102     dinfo->unit = unit_id;
1103     dinfo->devaddr = devaddr;
1104     dinfo->serial = g_strdup(serial);
1105 
1106     blk_set_legacy_dinfo(blk, dinfo);
1107 
1108     switch(type) {
1109     case IF_IDE:
1110     case IF_SCSI:
1111     case IF_XEN:
1112     case IF_NONE:
1113         dinfo->media_cd = media == MEDIA_CDROM;
1114         break;
1115     default:
1116         break;
1117     }
1118 
1119 fail:
1120     qemu_opts_del(legacy_opts);
1121     QDECREF(bs_opts);
1122     return dinfo;
1123 }
1124 
1125 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1126 {
1127     BlockDriverState *bs;
1128 
1129     bs = bdrv_lookup_bs(name, name, errp);
1130     if (bs == NULL) {
1131         return NULL;
1132     }
1133 
1134     if (!bdrv_is_root_node(bs)) {
1135         error_setg(errp, "Need a root block node");
1136         return NULL;
1137     }
1138 
1139     if (!bdrv_is_inserted(bs)) {
1140         error_setg(errp, "Device has no medium");
1141         return NULL;
1142     }
1143 
1144     return bs;
1145 }
1146 
1147 static BlockBackend *qmp_get_blk(const char *blk_name, const char *qdev_id,
1148                                  Error **errp)
1149 {
1150     BlockBackend *blk;
1151 
1152     if (!blk_name == !qdev_id) {
1153         error_setg(errp, "Need exactly one of 'device' and 'id'");
1154         return NULL;
1155     }
1156 
1157     if (qdev_id) {
1158         blk = blk_by_qdev_id(qdev_id, errp);
1159     } else {
1160         blk = blk_by_name(blk_name);
1161         if (blk == NULL) {
1162             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1163                       "Device '%s' not found", blk_name);
1164         }
1165     }
1166 
1167     return blk;
1168 }
1169 
1170 void hmp_commit(Monitor *mon, const QDict *qdict)
1171 {
1172     const char *device = qdict_get_str(qdict, "device");
1173     BlockBackend *blk;
1174     int ret;
1175 
1176     if (!strcmp(device, "all")) {
1177         ret = blk_commit_all();
1178     } else {
1179         BlockDriverState *bs;
1180         AioContext *aio_context;
1181 
1182         blk = blk_by_name(device);
1183         if (!blk) {
1184             monitor_printf(mon, "Device '%s' not found\n", device);
1185             return;
1186         }
1187         if (!blk_is_available(blk)) {
1188             monitor_printf(mon, "Device '%s' has no medium\n", device);
1189             return;
1190         }
1191 
1192         bs = blk_bs(blk);
1193         aio_context = bdrv_get_aio_context(bs);
1194         aio_context_acquire(aio_context);
1195 
1196         ret = bdrv_commit(bs);
1197 
1198         aio_context_release(aio_context);
1199     }
1200     if (ret < 0) {
1201         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1202                        strerror(-ret));
1203     }
1204 }
1205 
1206 static void blockdev_do_action(TransactionAction *action, Error **errp)
1207 {
1208     TransactionActionList list;
1209 
1210     list.value = action;
1211     list.next = NULL;
1212     qmp_transaction(&list, false, NULL, errp);
1213 }
1214 
1215 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1216                                 bool has_node_name, const char *node_name,
1217                                 const char *snapshot_file,
1218                                 bool has_snapshot_node_name,
1219                                 const char *snapshot_node_name,
1220                                 bool has_format, const char *format,
1221                                 bool has_mode, NewImageMode mode, Error **errp)
1222 {
1223     BlockdevSnapshotSync snapshot = {
1224         .has_device = has_device,
1225         .device = (char *) device,
1226         .has_node_name = has_node_name,
1227         .node_name = (char *) node_name,
1228         .snapshot_file = (char *) snapshot_file,
1229         .has_snapshot_node_name = has_snapshot_node_name,
1230         .snapshot_node_name = (char *) snapshot_node_name,
1231         .has_format = has_format,
1232         .format = (char *) format,
1233         .has_mode = has_mode,
1234         .mode = mode,
1235     };
1236     TransactionAction action = {
1237         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1238         .u.blockdev_snapshot_sync.data = &snapshot,
1239     };
1240     blockdev_do_action(&action, errp);
1241 }
1242 
1243 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1244                            Error **errp)
1245 {
1246     BlockdevSnapshot snapshot_data = {
1247         .node = (char *) node,
1248         .overlay = (char *) overlay
1249     };
1250     TransactionAction action = {
1251         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1252         .u.blockdev_snapshot.data = &snapshot_data,
1253     };
1254     blockdev_do_action(&action, errp);
1255 }
1256 
1257 void qmp_blockdev_snapshot_internal_sync(const char *device,
1258                                          const char *name,
1259                                          Error **errp)
1260 {
1261     BlockdevSnapshotInternal snapshot = {
1262         .device = (char *) device,
1263         .name = (char *) name
1264     };
1265     TransactionAction action = {
1266         .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1267         .u.blockdev_snapshot_internal_sync.data = &snapshot,
1268     };
1269     blockdev_do_action(&action, errp);
1270 }
1271 
1272 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1273                                                          bool has_id,
1274                                                          const char *id,
1275                                                          bool has_name,
1276                                                          const char *name,
1277                                                          Error **errp)
1278 {
1279     BlockDriverState *bs;
1280     AioContext *aio_context;
1281     QEMUSnapshotInfo sn;
1282     Error *local_err = NULL;
1283     SnapshotInfo *info = NULL;
1284     int ret;
1285 
1286     bs = qmp_get_root_bs(device, errp);
1287     if (!bs) {
1288         return NULL;
1289     }
1290     aio_context = bdrv_get_aio_context(bs);
1291     aio_context_acquire(aio_context);
1292 
1293     if (!has_id) {
1294         id = NULL;
1295     }
1296 
1297     if (!has_name) {
1298         name = NULL;
1299     }
1300 
1301     if (!id && !name) {
1302         error_setg(errp, "Name or id must be provided");
1303         goto out_aio_context;
1304     }
1305 
1306     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1307         goto out_aio_context;
1308     }
1309 
1310     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1311     if (local_err) {
1312         error_propagate(errp, local_err);
1313         goto out_aio_context;
1314     }
1315     if (!ret) {
1316         error_setg(errp,
1317                    "Snapshot with id '%s' and name '%s' does not exist on "
1318                    "device '%s'",
1319                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1320         goto out_aio_context;
1321     }
1322 
1323     bdrv_snapshot_delete(bs, id, name, &local_err);
1324     if (local_err) {
1325         error_propagate(errp, local_err);
1326         goto out_aio_context;
1327     }
1328 
1329     aio_context_release(aio_context);
1330 
1331     info = g_new0(SnapshotInfo, 1);
1332     info->id = g_strdup(sn.id_str);
1333     info->name = g_strdup(sn.name);
1334     info->date_nsec = sn.date_nsec;
1335     info->date_sec = sn.date_sec;
1336     info->vm_state_size = sn.vm_state_size;
1337     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1338     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1339 
1340     return info;
1341 
1342 out_aio_context:
1343     aio_context_release(aio_context);
1344     return NULL;
1345 }
1346 
1347 /**
1348  * block_dirty_bitmap_lookup:
1349  * Return a dirty bitmap (if present), after validating
1350  * the node reference and bitmap names.
1351  *
1352  * @node: The name of the BDS node to search for bitmaps
1353  * @name: The name of the bitmap to search for
1354  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1355  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1356  * @errp: Output pointer for error information. Can be NULL.
1357  *
1358  * @return: A bitmap object on success, or NULL on failure.
1359  */
1360 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1361                                                   const char *name,
1362                                                   BlockDriverState **pbs,
1363                                                   AioContext **paio,
1364                                                   Error **errp)
1365 {
1366     BlockDriverState *bs;
1367     BdrvDirtyBitmap *bitmap;
1368     AioContext *aio_context;
1369 
1370     if (!node) {
1371         error_setg(errp, "Node cannot be NULL");
1372         return NULL;
1373     }
1374     if (!name) {
1375         error_setg(errp, "Bitmap name cannot be NULL");
1376         return NULL;
1377     }
1378     bs = bdrv_lookup_bs(node, node, NULL);
1379     if (!bs) {
1380         error_setg(errp, "Node '%s' not found", node);
1381         return NULL;
1382     }
1383 
1384     aio_context = bdrv_get_aio_context(bs);
1385     aio_context_acquire(aio_context);
1386 
1387     bitmap = bdrv_find_dirty_bitmap(bs, name);
1388     if (!bitmap) {
1389         error_setg(errp, "Dirty bitmap '%s' not found", name);
1390         goto fail;
1391     }
1392 
1393     if (pbs) {
1394         *pbs = bs;
1395     }
1396     if (paio) {
1397         *paio = aio_context;
1398     } else {
1399         aio_context_release(aio_context);
1400     }
1401 
1402     return bitmap;
1403 
1404  fail:
1405     aio_context_release(aio_context);
1406     return NULL;
1407 }
1408 
1409 /* New and old BlockDriverState structs for atomic group operations */
1410 
1411 typedef struct BlkActionState BlkActionState;
1412 
1413 /**
1414  * BlkActionOps:
1415  * Table of operations that define an Action.
1416  *
1417  * @instance_size: Size of state struct, in bytes.
1418  * @prepare: Prepare the work, must NOT be NULL.
1419  * @commit: Commit the changes, can be NULL.
1420  * @abort: Abort the changes on fail, can be NULL.
1421  * @clean: Clean up resources after all transaction actions have called
1422  *         commit() or abort(). Can be NULL.
1423  *
1424  * Only prepare() may fail. In a single transaction, only one of commit() or
1425  * abort() will be called. clean() will always be called if it is present.
1426  */
1427 typedef struct BlkActionOps {
1428     size_t instance_size;
1429     void (*prepare)(BlkActionState *common, Error **errp);
1430     void (*commit)(BlkActionState *common);
1431     void (*abort)(BlkActionState *common);
1432     void (*clean)(BlkActionState *common);
1433 } BlkActionOps;
1434 
1435 /**
1436  * BlkActionState:
1437  * Describes one Action's state within a Transaction.
1438  *
1439  * @action: QAPI-defined enum identifying which Action to perform.
1440  * @ops: Table of ActionOps this Action can perform.
1441  * @block_job_txn: Transaction which this action belongs to.
1442  * @entry: List membership for all Actions in this Transaction.
1443  *
1444  * This structure must be arranged as first member in a subclassed type,
1445  * assuming that the compiler will also arrange it to the same offsets as the
1446  * base class.
1447  */
1448 struct BlkActionState {
1449     TransactionAction *action;
1450     const BlkActionOps *ops;
1451     BlockJobTxn *block_job_txn;
1452     TransactionProperties *txn_props;
1453     QSIMPLEQ_ENTRY(BlkActionState) entry;
1454 };
1455 
1456 /* internal snapshot private data */
1457 typedef struct InternalSnapshotState {
1458     BlkActionState common;
1459     BlockDriverState *bs;
1460     AioContext *aio_context;
1461     QEMUSnapshotInfo sn;
1462     bool created;
1463 } InternalSnapshotState;
1464 
1465 
1466 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1467 {
1468     if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1469         error_setg(errp,
1470                    "Action '%s' does not support Transaction property "
1471                    "completion-mode = %s",
1472                    TransactionActionKind_lookup[s->action->type],
1473                    ActionCompletionMode_lookup[s->txn_props->completion_mode]);
1474         return -1;
1475     }
1476     return 0;
1477 }
1478 
1479 static void internal_snapshot_prepare(BlkActionState *common,
1480                                       Error **errp)
1481 {
1482     Error *local_err = NULL;
1483     const char *device;
1484     const char *name;
1485     BlockDriverState *bs;
1486     QEMUSnapshotInfo old_sn, *sn;
1487     bool ret;
1488     qemu_timeval tv;
1489     BlockdevSnapshotInternal *internal;
1490     InternalSnapshotState *state;
1491     int ret1;
1492 
1493     g_assert(common->action->type ==
1494              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1495     internal = common->action->u.blockdev_snapshot_internal_sync.data;
1496     state = DO_UPCAST(InternalSnapshotState, common, common);
1497 
1498     /* 1. parse input */
1499     device = internal->device;
1500     name = internal->name;
1501 
1502     /* 2. check for validation */
1503     if (action_check_completion_mode(common, errp) < 0) {
1504         return;
1505     }
1506 
1507     bs = qmp_get_root_bs(device, errp);
1508     if (!bs) {
1509         return;
1510     }
1511 
1512     /* AioContext is released in .clean() */
1513     state->aio_context = bdrv_get_aio_context(bs);
1514     aio_context_acquire(state->aio_context);
1515 
1516     state->bs = bs;
1517     bdrv_drained_begin(bs);
1518 
1519     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1520         return;
1521     }
1522 
1523     if (bdrv_is_read_only(bs)) {
1524         error_setg(errp, "Device '%s' is read only", device);
1525         return;
1526     }
1527 
1528     if (!bdrv_can_snapshot(bs)) {
1529         error_setg(errp, "Block format '%s' used by device '%s' "
1530                    "does not support internal snapshots",
1531                    bs->drv->format_name, device);
1532         return;
1533     }
1534 
1535     if (!strlen(name)) {
1536         error_setg(errp, "Name is empty");
1537         return;
1538     }
1539 
1540     /* check whether a snapshot with name exist */
1541     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1542                                             &local_err);
1543     if (local_err) {
1544         error_propagate(errp, local_err);
1545         return;
1546     } else if (ret) {
1547         error_setg(errp,
1548                    "Snapshot with name '%s' already exists on device '%s'",
1549                    name, device);
1550         return;
1551     }
1552 
1553     /* 3. take the snapshot */
1554     sn = &state->sn;
1555     pstrcpy(sn->name, sizeof(sn->name), name);
1556     qemu_gettimeofday(&tv);
1557     sn->date_sec = tv.tv_sec;
1558     sn->date_nsec = tv.tv_usec * 1000;
1559     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1560 
1561     ret1 = bdrv_snapshot_create(bs, sn);
1562     if (ret1 < 0) {
1563         error_setg_errno(errp, -ret1,
1564                          "Failed to create snapshot '%s' on device '%s'",
1565                          name, device);
1566         return;
1567     }
1568 
1569     /* 4. succeed, mark a snapshot is created */
1570     state->created = true;
1571 }
1572 
1573 static void internal_snapshot_abort(BlkActionState *common)
1574 {
1575     InternalSnapshotState *state =
1576                              DO_UPCAST(InternalSnapshotState, common, common);
1577     BlockDriverState *bs = state->bs;
1578     QEMUSnapshotInfo *sn = &state->sn;
1579     Error *local_error = NULL;
1580 
1581     if (!state->created) {
1582         return;
1583     }
1584 
1585     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1586         error_reportf_err(local_error,
1587                           "Failed to delete snapshot with id '%s' and "
1588                           "name '%s' on device '%s' in abort: ",
1589                           sn->id_str, sn->name,
1590                           bdrv_get_device_name(bs));
1591     }
1592 }
1593 
1594 static void internal_snapshot_clean(BlkActionState *common)
1595 {
1596     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1597                                              common, common);
1598 
1599     if (state->aio_context) {
1600         if (state->bs) {
1601             bdrv_drained_end(state->bs);
1602         }
1603         aio_context_release(state->aio_context);
1604     }
1605 }
1606 
1607 /* external snapshot private data */
1608 typedef struct ExternalSnapshotState {
1609     BlkActionState common;
1610     BlockDriverState *old_bs;
1611     BlockDriverState *new_bs;
1612     AioContext *aio_context;
1613 } ExternalSnapshotState;
1614 
1615 static void external_snapshot_prepare(BlkActionState *common,
1616                                       Error **errp)
1617 {
1618     int flags = 0;
1619     QDict *options = NULL;
1620     Error *local_err = NULL;
1621     /* Device and node name of the image to generate the snapshot from */
1622     const char *device;
1623     const char *node_name;
1624     /* Reference to the new image (for 'blockdev-snapshot') */
1625     const char *snapshot_ref;
1626     /* File name of the new image (for 'blockdev-snapshot-sync') */
1627     const char *new_image_file;
1628     ExternalSnapshotState *state =
1629                              DO_UPCAST(ExternalSnapshotState, common, common);
1630     TransactionAction *action = common->action;
1631 
1632     /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1633      * purpose but a different set of parameters */
1634     switch (action->type) {
1635     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1636         {
1637             BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1638             device = s->node;
1639             node_name = s->node;
1640             new_image_file = NULL;
1641             snapshot_ref = s->overlay;
1642         }
1643         break;
1644     case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1645         {
1646             BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1647             device = s->has_device ? s->device : NULL;
1648             node_name = s->has_node_name ? s->node_name : NULL;
1649             new_image_file = s->snapshot_file;
1650             snapshot_ref = NULL;
1651         }
1652         break;
1653     default:
1654         g_assert_not_reached();
1655     }
1656 
1657     /* start processing */
1658     if (action_check_completion_mode(common, errp) < 0) {
1659         return;
1660     }
1661 
1662     state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1663     if (!state->old_bs) {
1664         return;
1665     }
1666 
1667     /* Acquire AioContext now so any threads operating on old_bs stop */
1668     state->aio_context = bdrv_get_aio_context(state->old_bs);
1669     aio_context_acquire(state->aio_context);
1670     bdrv_drained_begin(state->old_bs);
1671 
1672     if (!bdrv_is_inserted(state->old_bs)) {
1673         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1674         return;
1675     }
1676 
1677     if (bdrv_op_is_blocked(state->old_bs,
1678                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1679         return;
1680     }
1681 
1682     if (!bdrv_is_read_only(state->old_bs)) {
1683         if (bdrv_flush(state->old_bs)) {
1684             error_setg(errp, QERR_IO_ERROR);
1685             return;
1686         }
1687     }
1688 
1689     if (!bdrv_is_first_non_filter(state->old_bs)) {
1690         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1691         return;
1692     }
1693 
1694     if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1695         BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1696         const char *format = s->has_format ? s->format : "qcow2";
1697         enum NewImageMode mode;
1698         const char *snapshot_node_name =
1699             s->has_snapshot_node_name ? s->snapshot_node_name : NULL;
1700 
1701         if (node_name && !snapshot_node_name) {
1702             error_setg(errp, "New snapshot node name missing");
1703             return;
1704         }
1705 
1706         if (snapshot_node_name &&
1707             bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1708             error_setg(errp, "New snapshot node name already in use");
1709             return;
1710         }
1711 
1712         flags = state->old_bs->open_flags;
1713         flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1714 
1715         /* create new image w/backing file */
1716         mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1717         if (mode != NEW_IMAGE_MODE_EXISTING) {
1718             int64_t size = bdrv_getlength(state->old_bs);
1719             if (size < 0) {
1720                 error_setg_errno(errp, -size, "bdrv_getlength failed");
1721                 return;
1722             }
1723             bdrv_img_create(new_image_file, format,
1724                             state->old_bs->filename,
1725                             state->old_bs->drv->format_name,
1726                             NULL, size, flags, &local_err, false);
1727             if (local_err) {
1728                 error_propagate(errp, local_err);
1729                 return;
1730             }
1731         }
1732 
1733         options = qdict_new();
1734         if (s->has_snapshot_node_name) {
1735             qdict_put(options, "node-name",
1736                       qstring_from_str(snapshot_node_name));
1737         }
1738         qdict_put(options, "driver", qstring_from_str(format));
1739 
1740         flags |= BDRV_O_NO_BACKING;
1741     }
1742 
1743     state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1744                               errp);
1745     /* We will manually add the backing_hd field to the bs later */
1746     if (!state->new_bs) {
1747         return;
1748     }
1749 
1750     if (bdrv_has_blk(state->new_bs)) {
1751         error_setg(errp, "The snapshot is already in use");
1752         return;
1753     }
1754 
1755     if (bdrv_op_is_blocked(state->new_bs, BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
1756                            errp)) {
1757         return;
1758     }
1759 
1760     if (state->new_bs->backing != NULL) {
1761         error_setg(errp, "The snapshot already has a backing image");
1762         return;
1763     }
1764 
1765     if (!state->new_bs->drv->supports_backing) {
1766         error_setg(errp, "The snapshot does not support backing images");
1767     }
1768 }
1769 
1770 static void external_snapshot_commit(BlkActionState *common)
1771 {
1772     ExternalSnapshotState *state =
1773                              DO_UPCAST(ExternalSnapshotState, common, common);
1774 
1775     bdrv_set_aio_context(state->new_bs, state->aio_context);
1776 
1777     /* This removes our old bs and adds the new bs */
1778     bdrv_append(state->new_bs, state->old_bs);
1779     /* We don't need (or want) to use the transactional
1780      * bdrv_reopen_multiple() across all the entries at once, because we
1781      * don't want to abort all of them if one of them fails the reopen */
1782     if (!state->old_bs->copy_on_read) {
1783         bdrv_reopen(state->old_bs, state->old_bs->open_flags & ~BDRV_O_RDWR,
1784                     NULL);
1785     }
1786 }
1787 
1788 static void external_snapshot_abort(BlkActionState *common)
1789 {
1790     ExternalSnapshotState *state =
1791                              DO_UPCAST(ExternalSnapshotState, common, common);
1792     if (state->new_bs) {
1793         bdrv_unref(state->new_bs);
1794     }
1795 }
1796 
1797 static void external_snapshot_clean(BlkActionState *common)
1798 {
1799     ExternalSnapshotState *state =
1800                              DO_UPCAST(ExternalSnapshotState, common, common);
1801     if (state->aio_context) {
1802         bdrv_drained_end(state->old_bs);
1803         aio_context_release(state->aio_context);
1804     }
1805 }
1806 
1807 typedef struct DriveBackupState {
1808     BlkActionState common;
1809     BlockDriverState *bs;
1810     AioContext *aio_context;
1811     BlockJob *job;
1812 } DriveBackupState;
1813 
1814 static void do_drive_backup(DriveBackup *backup, BlockJobTxn *txn,
1815                             Error **errp);
1816 
1817 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1818 {
1819     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1820     BlockDriverState *bs;
1821     DriveBackup *backup;
1822     Error *local_err = NULL;
1823 
1824     assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1825     backup = common->action->u.drive_backup.data;
1826 
1827     bs = qmp_get_root_bs(backup->device, errp);
1828     if (!bs) {
1829         return;
1830     }
1831 
1832     /* AioContext is released in .clean() */
1833     state->aio_context = bdrv_get_aio_context(bs);
1834     aio_context_acquire(state->aio_context);
1835     bdrv_drained_begin(bs);
1836     state->bs = bs;
1837 
1838     do_drive_backup(backup, common->block_job_txn, &local_err);
1839     if (local_err) {
1840         error_propagate(errp, local_err);
1841         return;
1842     }
1843 
1844     state->job = state->bs->job;
1845 }
1846 
1847 static void drive_backup_abort(BlkActionState *common)
1848 {
1849     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1850     BlockDriverState *bs = state->bs;
1851 
1852     /* Only cancel if it's the job we started */
1853     if (bs && bs->job && bs->job == state->job) {
1854         block_job_cancel_sync(bs->job);
1855     }
1856 }
1857 
1858 static void drive_backup_clean(BlkActionState *common)
1859 {
1860     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1861 
1862     if (state->aio_context) {
1863         bdrv_drained_end(state->bs);
1864         aio_context_release(state->aio_context);
1865     }
1866 }
1867 
1868 typedef struct BlockdevBackupState {
1869     BlkActionState common;
1870     BlockDriverState *bs;
1871     BlockJob *job;
1872     AioContext *aio_context;
1873 } BlockdevBackupState;
1874 
1875 static void do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn,
1876                                Error **errp);
1877 
1878 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1879 {
1880     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1881     BlockdevBackup *backup;
1882     BlockDriverState *bs, *target;
1883     Error *local_err = NULL;
1884 
1885     assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1886     backup = common->action->u.blockdev_backup.data;
1887 
1888     bs = qmp_get_root_bs(backup->device, errp);
1889     if (!bs) {
1890         return;
1891     }
1892 
1893     target = bdrv_lookup_bs(backup->target, backup->target, errp);
1894     if (!target) {
1895         return;
1896     }
1897 
1898     /* AioContext is released in .clean() */
1899     state->aio_context = bdrv_get_aio_context(bs);
1900     if (state->aio_context != bdrv_get_aio_context(target)) {
1901         state->aio_context = NULL;
1902         error_setg(errp, "Backup between two IO threads is not implemented");
1903         return;
1904     }
1905     aio_context_acquire(state->aio_context);
1906     state->bs = bs;
1907     bdrv_drained_begin(state->bs);
1908 
1909     do_blockdev_backup(backup, common->block_job_txn, &local_err);
1910     if (local_err) {
1911         error_propagate(errp, local_err);
1912         return;
1913     }
1914 
1915     state->job = state->bs->job;
1916 }
1917 
1918 static void blockdev_backup_abort(BlkActionState *common)
1919 {
1920     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1921     BlockDriverState *bs = state->bs;
1922 
1923     /* Only cancel if it's the job we started */
1924     if (bs && bs->job && bs->job == state->job) {
1925         block_job_cancel_sync(bs->job);
1926     }
1927 }
1928 
1929 static void blockdev_backup_clean(BlkActionState *common)
1930 {
1931     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1932 
1933     if (state->aio_context) {
1934         bdrv_drained_end(state->bs);
1935         aio_context_release(state->aio_context);
1936     }
1937 }
1938 
1939 typedef struct BlockDirtyBitmapState {
1940     BlkActionState common;
1941     BdrvDirtyBitmap *bitmap;
1942     BlockDriverState *bs;
1943     AioContext *aio_context;
1944     HBitmap *backup;
1945     bool prepared;
1946 } BlockDirtyBitmapState;
1947 
1948 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1949                                            Error **errp)
1950 {
1951     Error *local_err = NULL;
1952     BlockDirtyBitmapAdd *action;
1953     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1954                                              common, common);
1955 
1956     if (action_check_completion_mode(common, errp) < 0) {
1957         return;
1958     }
1959 
1960     action = common->action->u.block_dirty_bitmap_add.data;
1961     /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1962     qmp_block_dirty_bitmap_add(action->node, action->name,
1963                                action->has_granularity, action->granularity,
1964                                &local_err);
1965 
1966     if (!local_err) {
1967         state->prepared = true;
1968     } else {
1969         error_propagate(errp, local_err);
1970     }
1971 }
1972 
1973 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1974 {
1975     BlockDirtyBitmapAdd *action;
1976     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1977                                              common, common);
1978 
1979     action = common->action->u.block_dirty_bitmap_add.data;
1980     /* Should not be able to fail: IF the bitmap was added via .prepare(),
1981      * then the node reference and bitmap name must have been valid.
1982      */
1983     if (state->prepared) {
1984         qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
1985     }
1986 }
1987 
1988 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
1989                                              Error **errp)
1990 {
1991     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1992                                              common, common);
1993     BlockDirtyBitmap *action;
1994 
1995     if (action_check_completion_mode(common, errp) < 0) {
1996         return;
1997     }
1998 
1999     action = common->action->u.block_dirty_bitmap_clear.data;
2000     state->bitmap = block_dirty_bitmap_lookup(action->node,
2001                                               action->name,
2002                                               &state->bs,
2003                                               &state->aio_context,
2004                                               errp);
2005     if (!state->bitmap) {
2006         return;
2007     }
2008 
2009     if (bdrv_dirty_bitmap_frozen(state->bitmap)) {
2010         error_setg(errp, "Cannot modify a frozen bitmap");
2011         return;
2012     } else if (!bdrv_dirty_bitmap_enabled(state->bitmap)) {
2013         error_setg(errp, "Cannot clear a disabled bitmap");
2014         return;
2015     }
2016 
2017     bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2018     /* AioContext is released in .clean() */
2019 }
2020 
2021 static void block_dirty_bitmap_clear_abort(BlkActionState *common)
2022 {
2023     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2024                                              common, common);
2025 
2026     bdrv_undo_clear_dirty_bitmap(state->bitmap, state->backup);
2027 }
2028 
2029 static void block_dirty_bitmap_clear_commit(BlkActionState *common)
2030 {
2031     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2032                                              common, common);
2033 
2034     hbitmap_free(state->backup);
2035 }
2036 
2037 static void block_dirty_bitmap_clear_clean(BlkActionState *common)
2038 {
2039     BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2040                                              common, common);
2041 
2042     if (state->aio_context) {
2043         aio_context_release(state->aio_context);
2044     }
2045 }
2046 
2047 static void abort_prepare(BlkActionState *common, Error **errp)
2048 {
2049     error_setg(errp, "Transaction aborted using Abort action");
2050 }
2051 
2052 static void abort_commit(BlkActionState *common)
2053 {
2054     g_assert_not_reached(); /* this action never succeeds */
2055 }
2056 
2057 static const BlkActionOps actions[] = {
2058     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2059         .instance_size = sizeof(ExternalSnapshotState),
2060         .prepare  = external_snapshot_prepare,
2061         .commit   = external_snapshot_commit,
2062         .abort = external_snapshot_abort,
2063         .clean = external_snapshot_clean,
2064     },
2065     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2066         .instance_size = sizeof(ExternalSnapshotState),
2067         .prepare  = external_snapshot_prepare,
2068         .commit   = external_snapshot_commit,
2069         .abort = external_snapshot_abort,
2070         .clean = external_snapshot_clean,
2071     },
2072     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2073         .instance_size = sizeof(DriveBackupState),
2074         .prepare = drive_backup_prepare,
2075         .abort = drive_backup_abort,
2076         .clean = drive_backup_clean,
2077     },
2078     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2079         .instance_size = sizeof(BlockdevBackupState),
2080         .prepare = blockdev_backup_prepare,
2081         .abort = blockdev_backup_abort,
2082         .clean = blockdev_backup_clean,
2083     },
2084     [TRANSACTION_ACTION_KIND_ABORT] = {
2085         .instance_size = sizeof(BlkActionState),
2086         .prepare = abort_prepare,
2087         .commit = abort_commit,
2088     },
2089     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2090         .instance_size = sizeof(InternalSnapshotState),
2091         .prepare  = internal_snapshot_prepare,
2092         .abort = internal_snapshot_abort,
2093         .clean = internal_snapshot_clean,
2094     },
2095     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2096         .instance_size = sizeof(BlockDirtyBitmapState),
2097         .prepare = block_dirty_bitmap_add_prepare,
2098         .abort = block_dirty_bitmap_add_abort,
2099     },
2100     [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2101         .instance_size = sizeof(BlockDirtyBitmapState),
2102         .prepare = block_dirty_bitmap_clear_prepare,
2103         .commit = block_dirty_bitmap_clear_commit,
2104         .abort = block_dirty_bitmap_clear_abort,
2105         .clean = block_dirty_bitmap_clear_clean,
2106     }
2107 };
2108 
2109 /**
2110  * Allocate a TransactionProperties structure if necessary, and fill
2111  * that structure with desired defaults if they are unset.
2112  */
2113 static TransactionProperties *get_transaction_properties(
2114     TransactionProperties *props)
2115 {
2116     if (!props) {
2117         props = g_new0(TransactionProperties, 1);
2118     }
2119 
2120     if (!props->has_completion_mode) {
2121         props->has_completion_mode = true;
2122         props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2123     }
2124 
2125     return props;
2126 }
2127 
2128 /*
2129  * 'Atomic' group operations.  The operations are performed as a set, and if
2130  * any fail then we roll back all operations in the group.
2131  */
2132 void qmp_transaction(TransactionActionList *dev_list,
2133                      bool has_props,
2134                      struct TransactionProperties *props,
2135                      Error **errp)
2136 {
2137     TransactionActionList *dev_entry = dev_list;
2138     BlockJobTxn *block_job_txn = NULL;
2139     BlkActionState *state, *next;
2140     Error *local_err = NULL;
2141 
2142     QSIMPLEQ_HEAD(snap_bdrv_states, BlkActionState) snap_bdrv_states;
2143     QSIMPLEQ_INIT(&snap_bdrv_states);
2144 
2145     /* Does this transaction get canceled as a group on failure?
2146      * If not, we don't really need to make a BlockJobTxn.
2147      */
2148     props = get_transaction_properties(props);
2149     if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2150         block_job_txn = block_job_txn_new();
2151     }
2152 
2153     /* drain all i/o before any operations */
2154     bdrv_drain_all();
2155 
2156     /* We don't do anything in this loop that commits us to the operations */
2157     while (NULL != dev_entry) {
2158         TransactionAction *dev_info = NULL;
2159         const BlkActionOps *ops;
2160 
2161         dev_info = dev_entry->value;
2162         dev_entry = dev_entry->next;
2163 
2164         assert(dev_info->type < ARRAY_SIZE(actions));
2165 
2166         ops = &actions[dev_info->type];
2167         assert(ops->instance_size > 0);
2168 
2169         state = g_malloc0(ops->instance_size);
2170         state->ops = ops;
2171         state->action = dev_info;
2172         state->block_job_txn = block_job_txn;
2173         state->txn_props = props;
2174         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2175 
2176         state->ops->prepare(state, &local_err);
2177         if (local_err) {
2178             error_propagate(errp, local_err);
2179             goto delete_and_fail;
2180         }
2181     }
2182 
2183     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2184         if (state->ops->commit) {
2185             state->ops->commit(state);
2186         }
2187     }
2188 
2189     /* success */
2190     goto exit;
2191 
2192 delete_and_fail:
2193     /* failure, and it is all-or-none; roll back all operations */
2194     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
2195         if (state->ops->abort) {
2196             state->ops->abort(state);
2197         }
2198     }
2199 exit:
2200     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2201         if (state->ops->clean) {
2202             state->ops->clean(state);
2203         }
2204         g_free(state);
2205     }
2206     if (!has_props) {
2207         qapi_free_TransactionProperties(props);
2208     }
2209     block_job_txn_unref(block_job_txn);
2210 }
2211 
2212 void qmp_eject(bool has_device, const char *device,
2213                bool has_id, const char *id,
2214                bool has_force, bool force, Error **errp)
2215 {
2216     Error *local_err = NULL;
2217     int rc;
2218 
2219     if (!has_force) {
2220         force = false;
2221     }
2222 
2223     rc = do_open_tray(has_device ? device : NULL,
2224                       has_id ? id : NULL,
2225                       force, &local_err);
2226     if (rc && rc != -ENOSYS) {
2227         error_propagate(errp, local_err);
2228         return;
2229     }
2230     error_free(local_err);
2231 
2232     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, errp);
2233 }
2234 
2235 void qmp_block_passwd(bool has_device, const char *device,
2236                       bool has_node_name, const char *node_name,
2237                       const char *password, Error **errp)
2238 {
2239     Error *local_err = NULL;
2240     BlockDriverState *bs;
2241     AioContext *aio_context;
2242 
2243     bs = bdrv_lookup_bs(has_device ? device : NULL,
2244                         has_node_name ? node_name : NULL,
2245                         &local_err);
2246     if (local_err) {
2247         error_propagate(errp, local_err);
2248         return;
2249     }
2250 
2251     aio_context = bdrv_get_aio_context(bs);
2252     aio_context_acquire(aio_context);
2253 
2254     bdrv_add_key(bs, password, errp);
2255 
2256     aio_context_release(aio_context);
2257 }
2258 
2259 /*
2260  * Attempt to open the tray of @device.
2261  * If @force, ignore its tray lock.
2262  * Else, if the tray is locked, don't open it, but ask the guest to open it.
2263  * On error, store an error through @errp and return -errno.
2264  * If @device does not exist, return -ENODEV.
2265  * If it has no removable media, return -ENOTSUP.
2266  * If it has no tray, return -ENOSYS.
2267  * If the guest was asked to open the tray, return -EINPROGRESS.
2268  * Else, return 0.
2269  */
2270 static int do_open_tray(const char *blk_name, const char *qdev_id,
2271                         bool force, Error **errp)
2272 {
2273     BlockBackend *blk;
2274     const char *device = qdev_id ?: blk_name;
2275     bool locked;
2276 
2277     blk = qmp_get_blk(blk_name, qdev_id, errp);
2278     if (!blk) {
2279         return -ENODEV;
2280     }
2281 
2282     if (!blk_dev_has_removable_media(blk)) {
2283         error_setg(errp, "Device '%s' is not removable", device);
2284         return -ENOTSUP;
2285     }
2286 
2287     if (!blk_dev_has_tray(blk)) {
2288         error_setg(errp, "Device '%s' does not have a tray", device);
2289         return -ENOSYS;
2290     }
2291 
2292     if (blk_dev_is_tray_open(blk)) {
2293         return 0;
2294     }
2295 
2296     locked = blk_dev_is_medium_locked(blk);
2297     if (locked) {
2298         blk_dev_eject_request(blk, force);
2299     }
2300 
2301     if (!locked || force) {
2302         blk_dev_change_media_cb(blk, false);
2303     }
2304 
2305     if (locked && !force) {
2306         error_setg(errp, "Device '%s' is locked and force was not specified, "
2307                    "wait for tray to open and try again", device);
2308         return -EINPROGRESS;
2309     }
2310 
2311     return 0;
2312 }
2313 
2314 void qmp_blockdev_open_tray(bool has_device, const char *device,
2315                             bool has_id, const char *id,
2316                             bool has_force, bool force,
2317                             Error **errp)
2318 {
2319     Error *local_err = NULL;
2320     int rc;
2321 
2322     if (!has_force) {
2323         force = false;
2324     }
2325     rc = do_open_tray(has_device ? device : NULL,
2326                       has_id ? id : NULL,
2327                       force, &local_err);
2328     if (rc && rc != -ENOSYS && rc != -EINPROGRESS) {
2329         error_propagate(errp, local_err);
2330         return;
2331     }
2332     error_free(local_err);
2333 }
2334 
2335 void qmp_blockdev_close_tray(bool has_device, const char *device,
2336                              bool has_id, const char *id,
2337                              Error **errp)
2338 {
2339     BlockBackend *blk;
2340 
2341     device = has_device ? device : NULL;
2342     id = has_id ? id : NULL;
2343 
2344     blk = qmp_get_blk(device, id, errp);
2345     if (!blk) {
2346         return;
2347     }
2348 
2349     if (!blk_dev_has_removable_media(blk)) {
2350         error_setg(errp, "Device '%s' is not removable", device ?: id);
2351         return;
2352     }
2353 
2354     if (!blk_dev_has_tray(blk)) {
2355         /* Ignore this command on tray-less devices */
2356         return;
2357     }
2358 
2359     if (!blk_dev_is_tray_open(blk)) {
2360         return;
2361     }
2362 
2363     blk_dev_change_media_cb(blk, true);
2364 }
2365 
2366 void qmp_x_blockdev_remove_medium(bool has_device, const char *device,
2367                                   bool has_id, const char *id, Error **errp)
2368 {
2369     BlockBackend *blk;
2370     BlockDriverState *bs;
2371     AioContext *aio_context;
2372     bool has_attached_device;
2373 
2374     device = has_device ? device : NULL;
2375     id = has_id ? id : NULL;
2376 
2377     blk = qmp_get_blk(device, id, errp);
2378     if (!blk) {
2379         return;
2380     }
2381 
2382     /* For BBs without a device, we can exchange the BDS tree at will */
2383     has_attached_device = blk_get_attached_dev(blk);
2384 
2385     if (has_attached_device && !blk_dev_has_removable_media(blk)) {
2386         error_setg(errp, "Device '%s' is not removable", device ?: id);
2387         return;
2388     }
2389 
2390     if (has_attached_device && blk_dev_has_tray(blk) &&
2391         !blk_dev_is_tray_open(blk))
2392     {
2393         error_setg(errp, "Tray of device '%s' is not open", device ?: id);
2394         return;
2395     }
2396 
2397     bs = blk_bs(blk);
2398     if (!bs) {
2399         return;
2400     }
2401 
2402     aio_context = bdrv_get_aio_context(bs);
2403     aio_context_acquire(aio_context);
2404 
2405     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
2406         goto out;
2407     }
2408 
2409     blk_remove_bs(blk);
2410 
2411     if (!blk_dev_has_tray(blk)) {
2412         /* For tray-less devices, blockdev-open-tray is a no-op (or may not be
2413          * called at all); therefore, the medium needs to be ejected here.
2414          * Do it after blk_remove_bs() so blk_is_inserted(blk) returns the @load
2415          * value passed here (i.e. false). */
2416         blk_dev_change_media_cb(blk, false);
2417     }
2418 
2419 out:
2420     aio_context_release(aio_context);
2421 }
2422 
2423 static void qmp_blockdev_insert_anon_medium(BlockBackend *blk,
2424                                             BlockDriverState *bs, Error **errp)
2425 {
2426     bool has_device;
2427 
2428     /* For BBs without a device, we can exchange the BDS tree at will */
2429     has_device = blk_get_attached_dev(blk);
2430 
2431     if (has_device && !blk_dev_has_removable_media(blk)) {
2432         error_setg(errp, "Device is not removable");
2433         return;
2434     }
2435 
2436     if (has_device && blk_dev_has_tray(blk) && !blk_dev_is_tray_open(blk)) {
2437         error_setg(errp, "Tray of the device is not open");
2438         return;
2439     }
2440 
2441     if (blk_bs(blk)) {
2442         error_setg(errp, "There already is a medium in the device");
2443         return;
2444     }
2445 
2446     blk_insert_bs(blk, bs);
2447 
2448     if (!blk_dev_has_tray(blk)) {
2449         /* For tray-less devices, blockdev-close-tray is a no-op (or may not be
2450          * called at all); therefore, the medium needs to be pushed into the
2451          * slot here.
2452          * Do it after blk_insert_bs() so blk_is_inserted(blk) returns the @load
2453          * value passed here (i.e. true). */
2454         blk_dev_change_media_cb(blk, true);
2455     }
2456 }
2457 
2458 void qmp_x_blockdev_insert_medium(bool has_device, const char *device,
2459                                   bool has_id, const char *id,
2460                                   const char *node_name, Error **errp)
2461 {
2462     BlockBackend *blk;
2463     BlockDriverState *bs;
2464 
2465     blk = qmp_get_blk(has_device ? device : NULL,
2466                       has_id ? id : NULL,
2467                       errp);
2468     if (!blk) {
2469         return;
2470     }
2471 
2472     bs = bdrv_find_node(node_name);
2473     if (!bs) {
2474         error_setg(errp, "Node '%s' not found", node_name);
2475         return;
2476     }
2477 
2478     if (bdrv_has_blk(bs)) {
2479         error_setg(errp, "Node '%s' is already in use", node_name);
2480         return;
2481     }
2482 
2483     qmp_blockdev_insert_anon_medium(blk, bs, errp);
2484 }
2485 
2486 void qmp_blockdev_change_medium(bool has_device, const char *device,
2487                                 bool has_id, const char *id,
2488                                 const char *filename,
2489                                 bool has_format, const char *format,
2490                                 bool has_read_only,
2491                                 BlockdevChangeReadOnlyMode read_only,
2492                                 Error **errp)
2493 {
2494     BlockBackend *blk;
2495     BlockDriverState *medium_bs = NULL;
2496     int bdrv_flags;
2497     bool detect_zeroes;
2498     int rc;
2499     QDict *options = NULL;
2500     Error *err = NULL;
2501 
2502     blk = qmp_get_blk(has_device ? device : NULL,
2503                       has_id ? id : NULL,
2504                       errp);
2505     if (!blk) {
2506         goto fail;
2507     }
2508 
2509     if (blk_bs(blk)) {
2510         blk_update_root_state(blk);
2511     }
2512 
2513     bdrv_flags = blk_get_open_flags_from_root_state(blk);
2514     bdrv_flags &= ~(BDRV_O_TEMPORARY | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING |
2515         BDRV_O_PROTOCOL);
2516 
2517     if (!has_read_only) {
2518         read_only = BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN;
2519     }
2520 
2521     switch (read_only) {
2522     case BLOCKDEV_CHANGE_READ_ONLY_MODE_RETAIN:
2523         break;
2524 
2525     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_ONLY:
2526         bdrv_flags &= ~BDRV_O_RDWR;
2527         break;
2528 
2529     case BLOCKDEV_CHANGE_READ_ONLY_MODE_READ_WRITE:
2530         bdrv_flags |= BDRV_O_RDWR;
2531         break;
2532 
2533     default:
2534         abort();
2535     }
2536 
2537     options = qdict_new();
2538     detect_zeroes = blk_get_detect_zeroes_from_root_state(blk);
2539     qdict_put(options, "detect-zeroes",
2540               qstring_from_str(detect_zeroes ? "on" : "off"));
2541 
2542     if (has_format) {
2543         qdict_put(options, "driver", qstring_from_str(format));
2544     }
2545 
2546     medium_bs = bdrv_open(filename, NULL, options, bdrv_flags, errp);
2547     if (!medium_bs) {
2548         goto fail;
2549     }
2550 
2551     bdrv_add_key(medium_bs, NULL, &err);
2552     if (err) {
2553         error_propagate(errp, err);
2554         goto fail;
2555     }
2556 
2557     rc = do_open_tray(has_device ? device : NULL,
2558                       has_id ? id : NULL,
2559                       false, &err);
2560     if (rc && rc != -ENOSYS) {
2561         error_propagate(errp, err);
2562         goto fail;
2563     }
2564     error_free(err);
2565     err = NULL;
2566 
2567     qmp_x_blockdev_remove_medium(has_device, device, has_id, id, &err);
2568     if (err) {
2569         error_propagate(errp, err);
2570         goto fail;
2571     }
2572 
2573     qmp_blockdev_insert_anon_medium(blk, medium_bs, &err);
2574     if (err) {
2575         error_propagate(errp, err);
2576         goto fail;
2577     }
2578 
2579     qmp_blockdev_close_tray(has_device, device, has_id, id, errp);
2580 
2581 fail:
2582     /* If the medium has been inserted, the device has its own reference, so
2583      * ours must be relinquished; and if it has not been inserted successfully,
2584      * the reference must be relinquished anyway */
2585     bdrv_unref(medium_bs);
2586 }
2587 
2588 /* throttling disk I/O limits */
2589 void qmp_block_set_io_throttle(BlockIOThrottle *arg, Error **errp)
2590 {
2591     ThrottleConfig cfg;
2592     BlockDriverState *bs;
2593     BlockBackend *blk;
2594     AioContext *aio_context;
2595 
2596     blk = qmp_get_blk(arg->has_device ? arg->device : NULL,
2597                       arg->has_id ? arg->id : NULL,
2598                       errp);
2599     if (!blk) {
2600         return;
2601     }
2602 
2603     aio_context = blk_get_aio_context(blk);
2604     aio_context_acquire(aio_context);
2605 
2606     bs = blk_bs(blk);
2607     if (!bs) {
2608         error_setg(errp, "Device has no medium");
2609         goto out;
2610     }
2611 
2612     throttle_config_init(&cfg);
2613     cfg.buckets[THROTTLE_BPS_TOTAL].avg = arg->bps;
2614     cfg.buckets[THROTTLE_BPS_READ].avg  = arg->bps_rd;
2615     cfg.buckets[THROTTLE_BPS_WRITE].avg = arg->bps_wr;
2616 
2617     cfg.buckets[THROTTLE_OPS_TOTAL].avg = arg->iops;
2618     cfg.buckets[THROTTLE_OPS_READ].avg  = arg->iops_rd;
2619     cfg.buckets[THROTTLE_OPS_WRITE].avg = arg->iops_wr;
2620 
2621     if (arg->has_bps_max) {
2622         cfg.buckets[THROTTLE_BPS_TOTAL].max = arg->bps_max;
2623     }
2624     if (arg->has_bps_rd_max) {
2625         cfg.buckets[THROTTLE_BPS_READ].max = arg->bps_rd_max;
2626     }
2627     if (arg->has_bps_wr_max) {
2628         cfg.buckets[THROTTLE_BPS_WRITE].max = arg->bps_wr_max;
2629     }
2630     if (arg->has_iops_max) {
2631         cfg.buckets[THROTTLE_OPS_TOTAL].max = arg->iops_max;
2632     }
2633     if (arg->has_iops_rd_max) {
2634         cfg.buckets[THROTTLE_OPS_READ].max = arg->iops_rd_max;
2635     }
2636     if (arg->has_iops_wr_max) {
2637         cfg.buckets[THROTTLE_OPS_WRITE].max = arg->iops_wr_max;
2638     }
2639 
2640     if (arg->has_bps_max_length) {
2641         cfg.buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_max_length;
2642     }
2643     if (arg->has_bps_rd_max_length) {
2644         cfg.buckets[THROTTLE_BPS_READ].burst_length = arg->bps_rd_max_length;
2645     }
2646     if (arg->has_bps_wr_max_length) {
2647         cfg.buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_wr_max_length;
2648     }
2649     if (arg->has_iops_max_length) {
2650         cfg.buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_max_length;
2651     }
2652     if (arg->has_iops_rd_max_length) {
2653         cfg.buckets[THROTTLE_OPS_READ].burst_length = arg->iops_rd_max_length;
2654     }
2655     if (arg->has_iops_wr_max_length) {
2656         cfg.buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_wr_max_length;
2657     }
2658 
2659     if (arg->has_iops_size) {
2660         cfg.op_size = arg->iops_size;
2661     }
2662 
2663     if (!throttle_is_valid(&cfg, errp)) {
2664         goto out;
2665     }
2666 
2667     if (throttle_enabled(&cfg)) {
2668         /* Enable I/O limits if they're not enabled yet, otherwise
2669          * just update the throttling group. */
2670         if (!blk_get_public(blk)->throttle_state) {
2671             blk_io_limits_enable(blk,
2672                                  arg->has_group ? arg->group :
2673                                  arg->has_device ? arg->device :
2674                                  arg->id);
2675         } else if (arg->has_group) {
2676             blk_io_limits_update_group(blk, arg->group);
2677         }
2678         /* Set the new throttling configuration */
2679         blk_set_io_limits(blk, &cfg);
2680     } else if (blk_get_public(blk)->throttle_state) {
2681         /* If all throttling settings are set to 0, disable I/O limits */
2682         blk_io_limits_disable(blk);
2683     }
2684 
2685 out:
2686     aio_context_release(aio_context);
2687 }
2688 
2689 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2690                                 bool has_granularity, uint32_t granularity,
2691                                 Error **errp)
2692 {
2693     AioContext *aio_context;
2694     BlockDriverState *bs;
2695 
2696     if (!name || name[0] == '\0') {
2697         error_setg(errp, "Bitmap name cannot be empty");
2698         return;
2699     }
2700 
2701     bs = bdrv_lookup_bs(node, node, errp);
2702     if (!bs) {
2703         return;
2704     }
2705 
2706     aio_context = bdrv_get_aio_context(bs);
2707     aio_context_acquire(aio_context);
2708 
2709     if (has_granularity) {
2710         if (granularity < 512 || !is_power_of_2(granularity)) {
2711             error_setg(errp, "Granularity must be power of 2 "
2712                              "and at least 512");
2713             goto out;
2714         }
2715     } else {
2716         /* Default to cluster size, if available: */
2717         granularity = bdrv_get_default_bitmap_granularity(bs);
2718     }
2719 
2720     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2721 
2722  out:
2723     aio_context_release(aio_context);
2724 }
2725 
2726 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2727                                    Error **errp)
2728 {
2729     AioContext *aio_context;
2730     BlockDriverState *bs;
2731     BdrvDirtyBitmap *bitmap;
2732 
2733     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2734     if (!bitmap || !bs) {
2735         return;
2736     }
2737 
2738     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2739         error_setg(errp,
2740                    "Bitmap '%s' is currently frozen and cannot be removed",
2741                    name);
2742         goto out;
2743     }
2744     bdrv_dirty_bitmap_make_anon(bitmap);
2745     bdrv_release_dirty_bitmap(bs, bitmap);
2746 
2747  out:
2748     aio_context_release(aio_context);
2749 }
2750 
2751 /**
2752  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2753  * immediately after a full backup operation.
2754  */
2755 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2756                                   Error **errp)
2757 {
2758     AioContext *aio_context;
2759     BdrvDirtyBitmap *bitmap;
2760     BlockDriverState *bs;
2761 
2762     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2763     if (!bitmap || !bs) {
2764         return;
2765     }
2766 
2767     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2768         error_setg(errp,
2769                    "Bitmap '%s' is currently frozen and cannot be modified",
2770                    name);
2771         goto out;
2772     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2773         error_setg(errp,
2774                    "Bitmap '%s' is currently disabled and cannot be cleared",
2775                    name);
2776         goto out;
2777     }
2778 
2779     bdrv_clear_dirty_bitmap(bitmap, NULL);
2780 
2781  out:
2782     aio_context_release(aio_context);
2783 }
2784 
2785 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2786 {
2787     const char *id = qdict_get_str(qdict, "id");
2788     BlockBackend *blk;
2789     BlockDriverState *bs;
2790     AioContext *aio_context;
2791     Error *local_err = NULL;
2792 
2793     bs = bdrv_find_node(id);
2794     if (bs) {
2795         qmp_x_blockdev_del(id, &local_err);
2796         if (local_err) {
2797             error_report_err(local_err);
2798         }
2799         return;
2800     }
2801 
2802     blk = blk_by_name(id);
2803     if (!blk) {
2804         error_report("Device '%s' not found", id);
2805         return;
2806     }
2807 
2808     if (!blk_legacy_dinfo(blk)) {
2809         error_report("Deleting device added with blockdev-add"
2810                      " is not supported");
2811         return;
2812     }
2813 
2814     aio_context = blk_get_aio_context(blk);
2815     aio_context_acquire(aio_context);
2816 
2817     bs = blk_bs(blk);
2818     if (bs) {
2819         if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2820             error_report_err(local_err);
2821             aio_context_release(aio_context);
2822             return;
2823         }
2824 
2825         blk_remove_bs(blk);
2826     }
2827 
2828     /* Make the BlockBackend and the attached BlockDriverState anonymous */
2829     monitor_remove_blk(blk);
2830 
2831     /* If this BlockBackend has a device attached to it, its refcount will be
2832      * decremented when the device is removed; otherwise we have to do so here.
2833      */
2834     if (blk_get_attached_dev(blk)) {
2835         /* Further I/O must not pause the guest */
2836         blk_set_on_error(blk, BLOCKDEV_ON_ERROR_REPORT,
2837                          BLOCKDEV_ON_ERROR_REPORT);
2838     } else {
2839         blk_unref(blk);
2840     }
2841 
2842     aio_context_release(aio_context);
2843 }
2844 
2845 void qmp_block_resize(bool has_device, const char *device,
2846                       bool has_node_name, const char *node_name,
2847                       int64_t size, Error **errp)
2848 {
2849     Error *local_err = NULL;
2850     BlockDriverState *bs;
2851     AioContext *aio_context;
2852     int ret;
2853 
2854     bs = bdrv_lookup_bs(has_device ? device : NULL,
2855                         has_node_name ? node_name : NULL,
2856                         &local_err);
2857     if (local_err) {
2858         error_propagate(errp, local_err);
2859         return;
2860     }
2861 
2862     aio_context = bdrv_get_aio_context(bs);
2863     aio_context_acquire(aio_context);
2864 
2865     if (!bdrv_is_first_non_filter(bs)) {
2866         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2867         goto out;
2868     }
2869 
2870     if (size < 0) {
2871         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2872         goto out;
2873     }
2874 
2875     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2876         error_setg(errp, QERR_DEVICE_IN_USE, device);
2877         goto out;
2878     }
2879 
2880     /* complete all in-flight operations before resizing the device */
2881     bdrv_drain_all();
2882 
2883     ret = bdrv_truncate(bs, size);
2884     switch (ret) {
2885     case 0:
2886         break;
2887     case -ENOMEDIUM:
2888         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2889         break;
2890     case -ENOTSUP:
2891         error_setg(errp, QERR_UNSUPPORTED);
2892         break;
2893     case -EACCES:
2894         error_setg(errp, "Device '%s' is read only", device);
2895         break;
2896     case -EBUSY:
2897         error_setg(errp, QERR_DEVICE_IN_USE, device);
2898         break;
2899     default:
2900         error_setg_errno(errp, -ret, "Could not resize");
2901         break;
2902     }
2903 
2904 out:
2905     aio_context_release(aio_context);
2906 }
2907 
2908 static void block_job_cb(void *opaque, int ret)
2909 {
2910     /* Note that this function may be executed from another AioContext besides
2911      * the QEMU main loop.  If you need to access anything that assumes the
2912      * QEMU global mutex, use a BH or introduce a mutex.
2913      */
2914 
2915     BlockDriverState *bs = opaque;
2916     const char *msg = NULL;
2917 
2918     trace_block_job_cb(bs, bs->job, ret);
2919 
2920     assert(bs->job);
2921 
2922     if (ret < 0) {
2923         msg = strerror(-ret);
2924     }
2925 
2926     if (block_job_is_cancelled(bs->job)) {
2927         block_job_event_cancelled(bs->job);
2928     } else {
2929         block_job_event_completed(bs->job, msg);
2930     }
2931 }
2932 
2933 void qmp_block_stream(bool has_job_id, const char *job_id, const char *device,
2934                       bool has_base, const char *base,
2935                       bool has_backing_file, const char *backing_file,
2936                       bool has_speed, int64_t speed,
2937                       bool has_on_error, BlockdevOnError on_error,
2938                       Error **errp)
2939 {
2940     BlockDriverState *bs;
2941     BlockDriverState *base_bs = NULL;
2942     AioContext *aio_context;
2943     Error *local_err = NULL;
2944     const char *base_name = NULL;
2945 
2946     if (!has_on_error) {
2947         on_error = BLOCKDEV_ON_ERROR_REPORT;
2948     }
2949 
2950     bs = qmp_get_root_bs(device, errp);
2951     if (!bs) {
2952         return;
2953     }
2954 
2955     aio_context = bdrv_get_aio_context(bs);
2956     aio_context_acquire(aio_context);
2957 
2958     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_STREAM, errp)) {
2959         goto out;
2960     }
2961 
2962     if (has_base) {
2963         base_bs = bdrv_find_backing_image(bs, base);
2964         if (base_bs == NULL) {
2965             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2966             goto out;
2967         }
2968         assert(bdrv_get_aio_context(base_bs) == aio_context);
2969         base_name = base;
2970     }
2971 
2972     /* if we are streaming the entire chain, the result will have no backing
2973      * file, and specifying one is therefore an error */
2974     if (base_bs == NULL && has_backing_file) {
2975         error_setg(errp, "backing file specified, but streaming the "
2976                          "entire chain");
2977         goto out;
2978     }
2979 
2980     /* backing_file string overrides base bs filename */
2981     base_name = has_backing_file ? backing_file : base_name;
2982 
2983     stream_start(has_job_id ? job_id : NULL, bs, base_bs, base_name,
2984                  has_speed ? speed : 0, on_error, block_job_cb, bs, &local_err);
2985     if (local_err) {
2986         error_propagate(errp, local_err);
2987         goto out;
2988     }
2989 
2990     trace_qmp_block_stream(bs, bs->job);
2991 
2992 out:
2993     aio_context_release(aio_context);
2994 }
2995 
2996 void qmp_block_commit(bool has_job_id, const char *job_id, const char *device,
2997                       bool has_base, const char *base,
2998                       bool has_top, const char *top,
2999                       bool has_backing_file, const char *backing_file,
3000                       bool has_speed, int64_t speed,
3001                       Error **errp)
3002 {
3003     BlockDriverState *bs;
3004     BlockDriverState *base_bs, *top_bs;
3005     AioContext *aio_context;
3006     Error *local_err = NULL;
3007     /* This will be part of the QMP command, if/when the
3008      * BlockdevOnError change for blkmirror makes it in
3009      */
3010     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
3011 
3012     if (!has_speed) {
3013         speed = 0;
3014     }
3015 
3016     /* Important Note:
3017      *  libvirt relies on the DeviceNotFound error class in order to probe for
3018      *  live commit feature versions; for this to work, we must make sure to
3019      *  perform the device lookup before any generic errors that may occur in a
3020      *  scenario in which all optional arguments are omitted. */
3021     bs = qmp_get_root_bs(device, &local_err);
3022     if (!bs) {
3023         bs = bdrv_lookup_bs(device, device, NULL);
3024         if (!bs) {
3025             error_free(local_err);
3026             error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
3027                       "Device '%s' not found", device);
3028         } else {
3029             error_propagate(errp, local_err);
3030         }
3031         return;
3032     }
3033 
3034     aio_context = bdrv_get_aio_context(bs);
3035     aio_context_acquire(aio_context);
3036 
3037     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
3038         goto out;
3039     }
3040 
3041     /* default top_bs is the active layer */
3042     top_bs = bs;
3043 
3044     if (has_top && top) {
3045         if (strcmp(bs->filename, top) != 0) {
3046             top_bs = bdrv_find_backing_image(bs, top);
3047         }
3048     }
3049 
3050     if (top_bs == NULL) {
3051         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
3052         goto out;
3053     }
3054 
3055     assert(bdrv_get_aio_context(top_bs) == aio_context);
3056 
3057     if (has_base && base) {
3058         base_bs = bdrv_find_backing_image(top_bs, base);
3059     } else {
3060         base_bs = bdrv_find_base(top_bs);
3061     }
3062 
3063     if (base_bs == NULL) {
3064         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
3065         goto out;
3066     }
3067 
3068     assert(bdrv_get_aio_context(base_bs) == aio_context);
3069 
3070     if (bdrv_op_is_blocked(base_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
3071         goto out;
3072     }
3073 
3074     /* Do not allow attempts to commit an image into itself */
3075     if (top_bs == base_bs) {
3076         error_setg(errp, "cannot commit an image into itself");
3077         goto out;
3078     }
3079 
3080     if (top_bs == bs) {
3081         if (has_backing_file) {
3082             error_setg(errp, "'backing-file' specified,"
3083                              " but 'top' is the active layer");
3084             goto out;
3085         }
3086         commit_active_start(has_job_id ? job_id : NULL, bs, base_bs, speed,
3087                             on_error, block_job_cb, bs, &local_err, false);
3088     } else {
3089         commit_start(has_job_id ? job_id : NULL, bs, base_bs, top_bs, speed,
3090                      on_error, block_job_cb, bs,
3091                      has_backing_file ? backing_file : NULL, &local_err);
3092     }
3093     if (local_err != NULL) {
3094         error_propagate(errp, local_err);
3095         goto out;
3096     }
3097 
3098 out:
3099     aio_context_release(aio_context);
3100 }
3101 
3102 static void do_drive_backup(DriveBackup *backup, BlockJobTxn *txn, Error **errp)
3103 {
3104     BlockDriverState *bs;
3105     BlockDriverState *target_bs;
3106     BlockDriverState *source = NULL;
3107     BdrvDirtyBitmap *bmap = NULL;
3108     AioContext *aio_context;
3109     QDict *options = NULL;
3110     Error *local_err = NULL;
3111     int flags;
3112     int64_t size;
3113 
3114     if (!backup->has_speed) {
3115         backup->speed = 0;
3116     }
3117     if (!backup->has_on_source_error) {
3118         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3119     }
3120     if (!backup->has_on_target_error) {
3121         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3122     }
3123     if (!backup->has_mode) {
3124         backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3125     }
3126     if (!backup->has_job_id) {
3127         backup->job_id = NULL;
3128     }
3129     if (!backup->has_compress) {
3130         backup->compress = false;
3131     }
3132 
3133     bs = qmp_get_root_bs(backup->device, errp);
3134     if (!bs) {
3135         return;
3136     }
3137 
3138     aio_context = bdrv_get_aio_context(bs);
3139     aio_context_acquire(aio_context);
3140 
3141     if (!backup->has_format) {
3142         backup->format = backup->mode == NEW_IMAGE_MODE_EXISTING ?
3143                          NULL : (char*) bs->drv->format_name;
3144     }
3145 
3146     /* Early check to avoid creating target */
3147     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
3148         goto out;
3149     }
3150 
3151     flags = bs->open_flags | BDRV_O_RDWR;
3152 
3153     /* See if we have a backing HD we can use to create our new image
3154      * on top of. */
3155     if (backup->sync == MIRROR_SYNC_MODE_TOP) {
3156         source = backing_bs(bs);
3157         if (!source) {
3158             backup->sync = MIRROR_SYNC_MODE_FULL;
3159         }
3160     }
3161     if (backup->sync == MIRROR_SYNC_MODE_NONE) {
3162         source = bs;
3163     }
3164 
3165     size = bdrv_getlength(bs);
3166     if (size < 0) {
3167         error_setg_errno(errp, -size, "bdrv_getlength failed");
3168         goto out;
3169     }
3170 
3171     if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
3172         assert(backup->format);
3173         if (source) {
3174             bdrv_img_create(backup->target, backup->format, source->filename,
3175                             source->drv->format_name, NULL,
3176                             size, flags, &local_err, false);
3177         } else {
3178             bdrv_img_create(backup->target, backup->format, NULL, NULL, NULL,
3179                             size, flags, &local_err, false);
3180         }
3181     }
3182 
3183     if (local_err) {
3184         error_propagate(errp, local_err);
3185         goto out;
3186     }
3187 
3188     if (backup->format) {
3189         options = qdict_new();
3190         qdict_put(options, "driver", qstring_from_str(backup->format));
3191     }
3192 
3193     target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
3194     if (!target_bs) {
3195         goto out;
3196     }
3197 
3198     bdrv_set_aio_context(target_bs, aio_context);
3199 
3200     if (backup->has_bitmap) {
3201         bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
3202         if (!bmap) {
3203             error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
3204             bdrv_unref(target_bs);
3205             goto out;
3206         }
3207     }
3208 
3209     backup_start(backup->job_id, bs, target_bs, backup->speed, backup->sync,
3210                  bmap, backup->compress, backup->on_source_error,
3211                  backup->on_target_error, block_job_cb, bs, txn, &local_err);
3212     bdrv_unref(target_bs);
3213     if (local_err != NULL) {
3214         error_propagate(errp, local_err);
3215         goto out;
3216     }
3217 
3218 out:
3219     aio_context_release(aio_context);
3220 }
3221 
3222 void qmp_drive_backup(DriveBackup *arg, Error **errp)
3223 {
3224     return do_drive_backup(arg, NULL, errp);
3225 }
3226 
3227 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
3228 {
3229     return bdrv_named_nodes_list(errp);
3230 }
3231 
3232 void do_blockdev_backup(BlockdevBackup *backup, BlockJobTxn *txn, Error **errp)
3233 {
3234     BlockDriverState *bs;
3235     BlockDriverState *target_bs;
3236     Error *local_err = NULL;
3237     AioContext *aio_context;
3238 
3239     if (!backup->has_speed) {
3240         backup->speed = 0;
3241     }
3242     if (!backup->has_on_source_error) {
3243         backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3244     }
3245     if (!backup->has_on_target_error) {
3246         backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3247     }
3248     if (!backup->has_job_id) {
3249         backup->job_id = NULL;
3250     }
3251     if (!backup->has_compress) {
3252         backup->compress = false;
3253     }
3254 
3255     bs = qmp_get_root_bs(backup->device, errp);
3256     if (!bs) {
3257         return;
3258     }
3259 
3260     aio_context = bdrv_get_aio_context(bs);
3261     aio_context_acquire(aio_context);
3262 
3263     target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
3264     if (!target_bs) {
3265         goto out;
3266     }
3267 
3268     if (bdrv_get_aio_context(target_bs) != aio_context) {
3269         if (!bdrv_has_blk(target_bs)) {
3270             /* The target BDS is not attached, we can safely move it to another
3271              * AioContext. */
3272             bdrv_set_aio_context(target_bs, aio_context);
3273         } else {
3274             error_setg(errp, "Target is attached to a different thread from "
3275                              "source.");
3276             goto out;
3277         }
3278     }
3279     backup_start(backup->job_id, bs, target_bs, backup->speed, backup->sync,
3280                  NULL, backup->compress, backup->on_source_error,
3281                  backup->on_target_error, block_job_cb, bs, txn, &local_err);
3282     if (local_err != NULL) {
3283         error_propagate(errp, local_err);
3284     }
3285 out:
3286     aio_context_release(aio_context);
3287 }
3288 
3289 void qmp_blockdev_backup(BlockdevBackup *arg, Error **errp)
3290 {
3291     do_blockdev_backup(arg, NULL, errp);
3292 }
3293 
3294 /* Parameter check and block job starting for drive mirroring.
3295  * Caller should hold @device and @target's aio context (must be the same).
3296  **/
3297 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
3298                                    BlockDriverState *target,
3299                                    bool has_replaces, const char *replaces,
3300                                    enum MirrorSyncMode sync,
3301                                    BlockMirrorBackingMode backing_mode,
3302                                    bool has_speed, int64_t speed,
3303                                    bool has_granularity, uint32_t granularity,
3304                                    bool has_buf_size, int64_t buf_size,
3305                                    bool has_on_source_error,
3306                                    BlockdevOnError on_source_error,
3307                                    bool has_on_target_error,
3308                                    BlockdevOnError on_target_error,
3309                                    bool has_unmap, bool unmap,
3310                                    Error **errp)
3311 {
3312 
3313     if (!has_speed) {
3314         speed = 0;
3315     }
3316     if (!has_on_source_error) {
3317         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
3318     }
3319     if (!has_on_target_error) {
3320         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
3321     }
3322     if (!has_granularity) {
3323         granularity = 0;
3324     }
3325     if (!has_buf_size) {
3326         buf_size = 0;
3327     }
3328     if (!has_unmap) {
3329         unmap = true;
3330     }
3331 
3332     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
3333         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3334                    "a value in range [512B, 64MB]");
3335         return;
3336     }
3337     if (granularity & (granularity - 1)) {
3338         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
3339                    "power of 2");
3340         return;
3341     }
3342 
3343     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3344         return;
3345     }
3346     if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
3347         return;
3348     }
3349 
3350     if (!bs->backing && sync == MIRROR_SYNC_MODE_TOP) {
3351         sync = MIRROR_SYNC_MODE_FULL;
3352     }
3353 
3354     /* pass the node name to replace to mirror start since it's loose coupling
3355      * and will allow to check whether the node still exist at mirror completion
3356      */
3357     mirror_start(job_id, bs, target,
3358                  has_replaces ? replaces : NULL,
3359                  speed, granularity, buf_size, sync, backing_mode,
3360                  on_source_error, on_target_error, unmap,
3361                  block_job_cb, bs, errp);
3362 }
3363 
3364 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3365 {
3366     BlockDriverState *bs;
3367     BlockDriverState *source, *target_bs;
3368     AioContext *aio_context;
3369     BlockMirrorBackingMode backing_mode;
3370     Error *local_err = NULL;
3371     QDict *options = NULL;
3372     int flags;
3373     int64_t size;
3374     const char *format = arg->format;
3375 
3376     bs = qmp_get_root_bs(arg->device, errp);
3377     if (!bs) {
3378         return;
3379     }
3380 
3381     aio_context = bdrv_get_aio_context(bs);
3382     aio_context_acquire(aio_context);
3383 
3384     if (!arg->has_mode) {
3385         arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3386     }
3387 
3388     if (!arg->has_format) {
3389         format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3390                   ? NULL : bs->drv->format_name);
3391     }
3392 
3393     flags = bs->open_flags | BDRV_O_RDWR;
3394     source = backing_bs(bs);
3395     if (!source && arg->sync == MIRROR_SYNC_MODE_TOP) {
3396         arg->sync = MIRROR_SYNC_MODE_FULL;
3397     }
3398     if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3399         source = bs;
3400     }
3401 
3402     size = bdrv_getlength(bs);
3403     if (size < 0) {
3404         error_setg_errno(errp, -size, "bdrv_getlength failed");
3405         goto out;
3406     }
3407 
3408     if (arg->has_replaces) {
3409         BlockDriverState *to_replace_bs;
3410         AioContext *replace_aio_context;
3411         int64_t replace_size;
3412 
3413         if (!arg->has_node_name) {
3414             error_setg(errp, "a node-name must be provided when replacing a"
3415                              " named node of the graph");
3416             goto out;
3417         }
3418 
3419         to_replace_bs = check_to_replace_node(bs, arg->replaces, &local_err);
3420 
3421         if (!to_replace_bs) {
3422             error_propagate(errp, local_err);
3423             goto out;
3424         }
3425 
3426         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3427         aio_context_acquire(replace_aio_context);
3428         replace_size = bdrv_getlength(to_replace_bs);
3429         aio_context_release(replace_aio_context);
3430 
3431         if (size != replace_size) {
3432             error_setg(errp, "cannot replace image with a mirror image of "
3433                              "different size");
3434             goto out;
3435         }
3436     }
3437 
3438     if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3439         backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3440     } else {
3441         backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3442     }
3443 
3444     if ((arg->sync == MIRROR_SYNC_MODE_FULL || !source)
3445         && arg->mode != NEW_IMAGE_MODE_EXISTING)
3446     {
3447         /* create new image w/o backing file */
3448         assert(format);
3449         bdrv_img_create(arg->target, format,
3450                         NULL, NULL, NULL, size, flags, &local_err, false);
3451     } else {
3452         switch (arg->mode) {
3453         case NEW_IMAGE_MODE_EXISTING:
3454             break;
3455         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3456             /* create new image with backing file */
3457             bdrv_img_create(arg->target, format,
3458                             source->filename,
3459                             source->drv->format_name,
3460                             NULL, size, flags, &local_err, false);
3461             break;
3462         default:
3463             abort();
3464         }
3465     }
3466 
3467     if (local_err) {
3468         error_propagate(errp, local_err);
3469         goto out;
3470     }
3471 
3472     options = qdict_new();
3473     if (arg->has_node_name) {
3474         qdict_put(options, "node-name", qstring_from_str(arg->node_name));
3475     }
3476     if (format) {
3477         qdict_put(options, "driver", qstring_from_str(format));
3478     }
3479 
3480     /* Mirroring takes care of copy-on-write using the source's backing
3481      * file.
3482      */
3483     target_bs = bdrv_open(arg->target, NULL, options,
3484                           flags | BDRV_O_NO_BACKING, errp);
3485     if (!target_bs) {
3486         goto out;
3487     }
3488 
3489     bdrv_set_aio_context(target_bs, aio_context);
3490 
3491     blockdev_mirror_common(arg->has_job_id ? arg->job_id : NULL, bs, target_bs,
3492                            arg->has_replaces, arg->replaces, arg->sync,
3493                            backing_mode, arg->has_speed, arg->speed,
3494                            arg->has_granularity, arg->granularity,
3495                            arg->has_buf_size, arg->buf_size,
3496                            arg->has_on_source_error, arg->on_source_error,
3497                            arg->has_on_target_error, arg->on_target_error,
3498                            arg->has_unmap, arg->unmap,
3499                            &local_err);
3500     bdrv_unref(target_bs);
3501     error_propagate(errp, local_err);
3502 out:
3503     aio_context_release(aio_context);
3504 }
3505 
3506 void qmp_blockdev_mirror(bool has_job_id, const char *job_id,
3507                          const char *device, const char *target,
3508                          bool has_replaces, const char *replaces,
3509                          MirrorSyncMode sync,
3510                          bool has_speed, int64_t speed,
3511                          bool has_granularity, uint32_t granularity,
3512                          bool has_buf_size, int64_t buf_size,
3513                          bool has_on_source_error,
3514                          BlockdevOnError on_source_error,
3515                          bool has_on_target_error,
3516                          BlockdevOnError on_target_error,
3517                          Error **errp)
3518 {
3519     BlockDriverState *bs;
3520     BlockDriverState *target_bs;
3521     AioContext *aio_context;
3522     BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3523     Error *local_err = NULL;
3524 
3525     bs = qmp_get_root_bs(device, errp);
3526     if (!bs) {
3527         return;
3528     }
3529 
3530     target_bs = bdrv_lookup_bs(target, target, errp);
3531     if (!target_bs) {
3532         return;
3533     }
3534 
3535     aio_context = bdrv_get_aio_context(bs);
3536     aio_context_acquire(aio_context);
3537 
3538     bdrv_set_aio_context(target_bs, aio_context);
3539 
3540     blockdev_mirror_common(has_job_id ? job_id : NULL, bs, target_bs,
3541                            has_replaces, replaces, sync, backing_mode,
3542                            has_speed, speed,
3543                            has_granularity, granularity,
3544                            has_buf_size, buf_size,
3545                            has_on_source_error, on_source_error,
3546                            has_on_target_error, on_target_error,
3547                            true, true,
3548                            &local_err);
3549     error_propagate(errp, local_err);
3550 
3551     aio_context_release(aio_context);
3552 }
3553 
3554 /* Get a block job using its ID and acquire its AioContext */
3555 static BlockJob *find_block_job(const char *id, AioContext **aio_context,
3556                                 Error **errp)
3557 {
3558     BlockJob *job;
3559 
3560     assert(id != NULL);
3561 
3562     *aio_context = NULL;
3563 
3564     job = block_job_get(id);
3565 
3566     if (!job) {
3567         error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3568                   "Block job '%s' not found", id);
3569         return NULL;
3570     }
3571 
3572     *aio_context = blk_get_aio_context(job->blk);
3573     aio_context_acquire(*aio_context);
3574 
3575     return job;
3576 }
3577 
3578 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3579 {
3580     AioContext *aio_context;
3581     BlockJob *job = find_block_job(device, &aio_context, errp);
3582 
3583     if (!job) {
3584         return;
3585     }
3586 
3587     block_job_set_speed(job, speed, errp);
3588     aio_context_release(aio_context);
3589 }
3590 
3591 void qmp_block_job_cancel(const char *device,
3592                           bool has_force, bool force, Error **errp)
3593 {
3594     AioContext *aio_context;
3595     BlockJob *job = find_block_job(device, &aio_context, errp);
3596 
3597     if (!job) {
3598         return;
3599     }
3600 
3601     if (!has_force) {
3602         force = false;
3603     }
3604 
3605     if (job->user_paused && !force) {
3606         error_setg(errp, "The block job for device '%s' is currently paused",
3607                    device);
3608         goto out;
3609     }
3610 
3611     trace_qmp_block_job_cancel(job);
3612     block_job_cancel(job);
3613 out:
3614     aio_context_release(aio_context);
3615 }
3616 
3617 void qmp_block_job_pause(const char *device, Error **errp)
3618 {
3619     AioContext *aio_context;
3620     BlockJob *job = find_block_job(device, &aio_context, errp);
3621 
3622     if (!job || job->user_paused) {
3623         return;
3624     }
3625 
3626     job->user_paused = true;
3627     trace_qmp_block_job_pause(job);
3628     block_job_pause(job);
3629     aio_context_release(aio_context);
3630 }
3631 
3632 void qmp_block_job_resume(const char *device, Error **errp)
3633 {
3634     AioContext *aio_context;
3635     BlockJob *job = find_block_job(device, &aio_context, errp);
3636 
3637     if (!job || !job->user_paused) {
3638         return;
3639     }
3640 
3641     job->user_paused = false;
3642     trace_qmp_block_job_resume(job);
3643     block_job_iostatus_reset(job);
3644     block_job_resume(job);
3645     aio_context_release(aio_context);
3646 }
3647 
3648 void qmp_block_job_complete(const char *device, Error **errp)
3649 {
3650     AioContext *aio_context;
3651     BlockJob *job = find_block_job(device, &aio_context, errp);
3652 
3653     if (!job) {
3654         return;
3655     }
3656 
3657     trace_qmp_block_job_complete(job);
3658     block_job_complete(job, errp);
3659     aio_context_release(aio_context);
3660 }
3661 
3662 void qmp_change_backing_file(const char *device,
3663                              const char *image_node_name,
3664                              const char *backing_file,
3665                              Error **errp)
3666 {
3667     BlockDriverState *bs = NULL;
3668     AioContext *aio_context;
3669     BlockDriverState *image_bs = NULL;
3670     Error *local_err = NULL;
3671     bool ro;
3672     int open_flags;
3673     int ret;
3674 
3675     bs = qmp_get_root_bs(device, errp);
3676     if (!bs) {
3677         return;
3678     }
3679 
3680     aio_context = bdrv_get_aio_context(bs);
3681     aio_context_acquire(aio_context);
3682 
3683     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3684     if (local_err) {
3685         error_propagate(errp, local_err);
3686         goto out;
3687     }
3688 
3689     if (!image_bs) {
3690         error_setg(errp, "image file not found");
3691         goto out;
3692     }
3693 
3694     if (bdrv_find_base(image_bs) == image_bs) {
3695         error_setg(errp, "not allowing backing file change on an image "
3696                          "without a backing file");
3697         goto out;
3698     }
3699 
3700     /* even though we are not necessarily operating on bs, we need it to
3701      * determine if block ops are currently prohibited on the chain */
3702     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3703         goto out;
3704     }
3705 
3706     /* final sanity check */
3707     if (!bdrv_chain_contains(bs, image_bs)) {
3708         error_setg(errp, "'%s' and image file are not in the same chain",
3709                    device);
3710         goto out;
3711     }
3712 
3713     /* if not r/w, reopen to make r/w */
3714     open_flags = image_bs->open_flags;
3715     ro = bdrv_is_read_only(image_bs);
3716 
3717     if (ro) {
3718         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3719         if (local_err) {
3720             error_propagate(errp, local_err);
3721             goto out;
3722         }
3723     }
3724 
3725     ret = bdrv_change_backing_file(image_bs, backing_file,
3726                                image_bs->drv ? image_bs->drv->format_name : "");
3727 
3728     if (ret < 0) {
3729         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3730                          backing_file);
3731         /* don't exit here, so we can try to restore open flags if
3732          * appropriate */
3733     }
3734 
3735     if (ro) {
3736         bdrv_reopen(image_bs, open_flags, &local_err);
3737         error_propagate(errp, local_err);
3738     }
3739 
3740 out:
3741     aio_context_release(aio_context);
3742 }
3743 
3744 void hmp_drive_add_node(Monitor *mon, const char *optstr)
3745 {
3746     QemuOpts *opts;
3747     QDict *qdict;
3748     Error *local_err = NULL;
3749 
3750     opts = qemu_opts_parse_noisily(&qemu_drive_opts, optstr, false);
3751     if (!opts) {
3752         return;
3753     }
3754 
3755     qdict = qemu_opts_to_qdict(opts, NULL);
3756 
3757     if (!qdict_get_try_str(qdict, "node-name")) {
3758         QDECREF(qdict);
3759         error_report("'node-name' needs to be specified");
3760         goto out;
3761     }
3762 
3763     BlockDriverState *bs = bds_tree_init(qdict, &local_err);
3764     if (!bs) {
3765         error_report_err(local_err);
3766         goto out;
3767     }
3768 
3769     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3770 
3771 out:
3772     qemu_opts_del(opts);
3773 }
3774 
3775 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3776 {
3777     BlockDriverState *bs;
3778     QObject *obj;
3779     Visitor *v = qobject_output_visitor_new(&obj);
3780     QDict *qdict;
3781     Error *local_err = NULL;
3782 
3783     visit_type_BlockdevOptions(v, NULL, &options, &local_err);
3784     if (local_err) {
3785         error_propagate(errp, local_err);
3786         goto fail;
3787     }
3788 
3789     visit_complete(v, &obj);
3790     qdict = qobject_to_qdict(obj);
3791 
3792     qdict_flatten(qdict);
3793 
3794     if (!qdict_get_try_str(qdict, "node-name")) {
3795         error_setg(errp, "'node-name' must be specified for the root node");
3796         goto fail;
3797     }
3798 
3799     bs = bds_tree_init(qdict, errp);
3800     if (!bs) {
3801         goto fail;
3802     }
3803 
3804     QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
3805 
3806     if (bs && bdrv_key_required(bs)) {
3807         QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3808         bdrv_unref(bs);
3809         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3810         goto fail;
3811     }
3812 
3813 fail:
3814     visit_free(v);
3815 }
3816 
3817 void qmp_x_blockdev_del(const char *node_name, Error **errp)
3818 {
3819     AioContext *aio_context;
3820     BlockDriverState *bs;
3821 
3822     bs = bdrv_find_node(node_name);
3823     if (!bs) {
3824         error_setg(errp, "Cannot find node %s", node_name);
3825         return;
3826     }
3827     if (bdrv_has_blk(bs)) {
3828         error_setg(errp, "Node %s is in use", node_name);
3829         return;
3830     }
3831     aio_context = bdrv_get_aio_context(bs);
3832     aio_context_acquire(aio_context);
3833 
3834     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3835         goto out;
3836     }
3837 
3838     if (!bs->monitor_list.tqe_prev) {
3839         error_setg(errp, "Node %s is not owned by the monitor",
3840                    bs->node_name);
3841         goto out;
3842     }
3843 
3844     if (bs->refcnt > 1) {
3845         error_setg(errp, "Block device %s is in use",
3846                    bdrv_get_device_or_node_name(bs));
3847         goto out;
3848     }
3849 
3850     QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3851     bdrv_unref(bs);
3852 
3853 out:
3854     aio_context_release(aio_context);
3855 }
3856 
3857 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3858                                   const char *child_name)
3859 {
3860     BdrvChild *child;
3861 
3862     QLIST_FOREACH(child, &parent_bs->children, next) {
3863         if (strcmp(child->name, child_name) == 0) {
3864             return child;
3865         }
3866     }
3867 
3868     return NULL;
3869 }
3870 
3871 void qmp_x_blockdev_change(const char *parent, bool has_child,
3872                            const char *child, bool has_node,
3873                            const char *node, Error **errp)
3874 {
3875     BlockDriverState *parent_bs, *new_bs = NULL;
3876     BdrvChild *p_child;
3877 
3878     parent_bs = bdrv_lookup_bs(parent, parent, errp);
3879     if (!parent_bs) {
3880         return;
3881     }
3882 
3883     if (has_child == has_node) {
3884         if (has_child) {
3885             error_setg(errp, "The parameters child and node are in conflict");
3886         } else {
3887             error_setg(errp, "Either child or node must be specified");
3888         }
3889         return;
3890     }
3891 
3892     if (has_child) {
3893         p_child = bdrv_find_child(parent_bs, child);
3894         if (!p_child) {
3895             error_setg(errp, "Node '%s' does not have child '%s'",
3896                        parent, child);
3897             return;
3898         }
3899         bdrv_del_child(parent_bs, p_child, errp);
3900     }
3901 
3902     if (has_node) {
3903         new_bs = bdrv_find_node(node);
3904         if (!new_bs) {
3905             error_setg(errp, "Node '%s' not found", node);
3906             return;
3907         }
3908         bdrv_add_child(parent_bs, new_bs, errp);
3909     }
3910 }
3911 
3912 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3913 {
3914     BlockJobInfoList *head = NULL, **p_next = &head;
3915     BlockJob *job;
3916 
3917     for (job = block_job_next(NULL); job; job = block_job_next(job)) {
3918         BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
3919         AioContext *aio_context = blk_get_aio_context(job->blk);
3920 
3921         aio_context_acquire(aio_context);
3922         elem->value = block_job_query(job);
3923         aio_context_release(aio_context);
3924 
3925         *p_next = elem;
3926         p_next = &elem->next;
3927     }
3928 
3929     return head;
3930 }
3931 
3932 QemuOptsList qemu_common_drive_opts = {
3933     .name = "drive",
3934     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3935     .desc = {
3936         {
3937             .name = "snapshot",
3938             .type = QEMU_OPT_BOOL,
3939             .help = "enable/disable snapshot mode",
3940         },{
3941             .name = "aio",
3942             .type = QEMU_OPT_STRING,
3943             .help = "host AIO implementation (threads, native)",
3944         },{
3945             .name = BDRV_OPT_CACHE_WB,
3946             .type = QEMU_OPT_BOOL,
3947             .help = "Enable writeback mode",
3948         },{
3949             .name = "format",
3950             .type = QEMU_OPT_STRING,
3951             .help = "disk format (raw, qcow2, ...)",
3952         },{
3953             .name = "rerror",
3954             .type = QEMU_OPT_STRING,
3955             .help = "read error action",
3956         },{
3957             .name = "werror",
3958             .type = QEMU_OPT_STRING,
3959             .help = "write error action",
3960         },{
3961             .name = BDRV_OPT_READ_ONLY,
3962             .type = QEMU_OPT_BOOL,
3963             .help = "open drive file as read-only",
3964         },{
3965             .name = "throttling.iops-total",
3966             .type = QEMU_OPT_NUMBER,
3967             .help = "limit total I/O operations per second",
3968         },{
3969             .name = "throttling.iops-read",
3970             .type = QEMU_OPT_NUMBER,
3971             .help = "limit read operations per second",
3972         },{
3973             .name = "throttling.iops-write",
3974             .type = QEMU_OPT_NUMBER,
3975             .help = "limit write operations per second",
3976         },{
3977             .name = "throttling.bps-total",
3978             .type = QEMU_OPT_NUMBER,
3979             .help = "limit total bytes per second",
3980         },{
3981             .name = "throttling.bps-read",
3982             .type = QEMU_OPT_NUMBER,
3983             .help = "limit read bytes per second",
3984         },{
3985             .name = "throttling.bps-write",
3986             .type = QEMU_OPT_NUMBER,
3987             .help = "limit write bytes per second",
3988         },{
3989             .name = "throttling.iops-total-max",
3990             .type = QEMU_OPT_NUMBER,
3991             .help = "I/O operations burst",
3992         },{
3993             .name = "throttling.iops-read-max",
3994             .type = QEMU_OPT_NUMBER,
3995             .help = "I/O operations read burst",
3996         },{
3997             .name = "throttling.iops-write-max",
3998             .type = QEMU_OPT_NUMBER,
3999             .help = "I/O operations write burst",
4000         },{
4001             .name = "throttling.bps-total-max",
4002             .type = QEMU_OPT_NUMBER,
4003             .help = "total bytes burst",
4004         },{
4005             .name = "throttling.bps-read-max",
4006             .type = QEMU_OPT_NUMBER,
4007             .help = "total bytes read burst",
4008         },{
4009             .name = "throttling.bps-write-max",
4010             .type = QEMU_OPT_NUMBER,
4011             .help = "total bytes write burst",
4012         },{
4013             .name = "throttling.iops-total-max-length",
4014             .type = QEMU_OPT_NUMBER,
4015             .help = "length of the iops-total-max burst period, in seconds",
4016         },{
4017             .name = "throttling.iops-read-max-length",
4018             .type = QEMU_OPT_NUMBER,
4019             .help = "length of the iops-read-max burst period, in seconds",
4020         },{
4021             .name = "throttling.iops-write-max-length",
4022             .type = QEMU_OPT_NUMBER,
4023             .help = "length of the iops-write-max burst period, in seconds",
4024         },{
4025             .name = "throttling.bps-total-max-length",
4026             .type = QEMU_OPT_NUMBER,
4027             .help = "length of the bps-total-max burst period, in seconds",
4028         },{
4029             .name = "throttling.bps-read-max-length",
4030             .type = QEMU_OPT_NUMBER,
4031             .help = "length of the bps-read-max burst period, in seconds",
4032         },{
4033             .name = "throttling.bps-write-max-length",
4034             .type = QEMU_OPT_NUMBER,
4035             .help = "length of the bps-write-max burst period, in seconds",
4036         },{
4037             .name = "throttling.iops-size",
4038             .type = QEMU_OPT_NUMBER,
4039             .help = "when limiting by iops max size of an I/O in bytes",
4040         },{
4041             .name = "throttling.group",
4042             .type = QEMU_OPT_STRING,
4043             .help = "name of the block throttling group",
4044         },{
4045             .name = "copy-on-read",
4046             .type = QEMU_OPT_BOOL,
4047             .help = "copy read data from backing file into image file",
4048         },{
4049             .name = "detect-zeroes",
4050             .type = QEMU_OPT_STRING,
4051             .help = "try to optimize zero writes (off, on, unmap)",
4052         },{
4053             .name = "stats-account-invalid",
4054             .type = QEMU_OPT_BOOL,
4055             .help = "whether to account for invalid I/O operations "
4056                     "in the statistics",
4057         },{
4058             .name = "stats-account-failed",
4059             .type = QEMU_OPT_BOOL,
4060             .help = "whether to account for failed I/O operations "
4061                     "in the statistics",
4062         },
4063         { /* end of list */ }
4064     },
4065 };
4066 
4067 QemuOptsList qemu_drive_opts = {
4068     .name = "drive",
4069     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
4070     .desc = {
4071         /*
4072          * no elements => accept any params
4073          * validation will happen later
4074          */
4075         { /* end of list */ }
4076     },
4077 };
4078