xref: /openbmc/qemu/blockdev.c (revision c6bd8c70)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  *
9  * This file incorporates work covered by the following copyright and
10  * permission notice:
11  *
12  * Copyright (c) 2003-2008 Fabrice Bellard
13  *
14  * Permission is hereby granted, free of charge, to any person obtaining a copy
15  * of this software and associated documentation files (the "Software"), to deal
16  * in the Software without restriction, including without limitation the rights
17  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18  * copies of the Software, and to permit persons to whom the Software is
19  * furnished to do so, subject to the following conditions:
20  *
21  * The above copyright notice and this permission notice shall be included in
22  * all copies or substantial portions of the Software.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30  * THE SOFTWARE.
31  */
32 
33 #include "sysemu/block-backend.h"
34 #include "sysemu/blockdev.h"
35 #include "hw/block/block.h"
36 #include "block/blockjob.h"
37 #include "block/throttle-groups.h"
38 #include "monitor/monitor.h"
39 #include "qemu/option.h"
40 #include "qemu/config-file.h"
41 #include "qapi/qmp/types.h"
42 #include "qapi-visit.h"
43 #include "qapi/qmp-output-visitor.h"
44 #include "qapi/util.h"
45 #include "sysemu/sysemu.h"
46 #include "block/block_int.h"
47 #include "qmp-commands.h"
48 #include "trace.h"
49 #include "sysemu/arch_init.h"
50 
51 static const char *const if_name[IF_COUNT] = {
52     [IF_NONE] = "none",
53     [IF_IDE] = "ide",
54     [IF_SCSI] = "scsi",
55     [IF_FLOPPY] = "floppy",
56     [IF_PFLASH] = "pflash",
57     [IF_MTD] = "mtd",
58     [IF_SD] = "sd",
59     [IF_VIRTIO] = "virtio",
60     [IF_XEN] = "xen",
61 };
62 
63 static int if_max_devs[IF_COUNT] = {
64     /*
65      * Do not change these numbers!  They govern how drive option
66      * index maps to unit and bus.  That mapping is ABI.
67      *
68      * All controllers used to imlement if=T drives need to support
69      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
70      * Otherwise, some index values map to "impossible" bus, unit
71      * values.
72      *
73      * For instance, if you change [IF_SCSI] to 255, -drive
74      * if=scsi,index=12 no longer means bus=1,unit=5, but
75      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
76      * the drive can't be set up.  Regression.
77      */
78     [IF_IDE] = 2,
79     [IF_SCSI] = 7,
80 };
81 
82 /**
83  * Boards may call this to offer board-by-board overrides
84  * of the default, global values.
85  */
86 void override_max_devs(BlockInterfaceType type, int max_devs)
87 {
88     BlockBackend *blk;
89     DriveInfo *dinfo;
90 
91     if (max_devs <= 0) {
92         return;
93     }
94 
95     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
96         dinfo = blk_legacy_dinfo(blk);
97         if (dinfo->type == type) {
98             fprintf(stderr, "Cannot override units-per-bus property of"
99                     " the %s interface, because a drive of that type has"
100                     " already been added.\n", if_name[type]);
101             g_assert_not_reached();
102         }
103     }
104 
105     if_max_devs[type] = max_devs;
106 }
107 
108 /*
109  * We automatically delete the drive when a device using it gets
110  * unplugged.  Questionable feature, but we can't just drop it.
111  * Device models call blockdev_mark_auto_del() to schedule the
112  * automatic deletion, and generic qdev code calls blockdev_auto_del()
113  * when deletion is actually safe.
114  */
115 void blockdev_mark_auto_del(BlockBackend *blk)
116 {
117     DriveInfo *dinfo = blk_legacy_dinfo(blk);
118     BlockDriverState *bs = blk_bs(blk);
119     AioContext *aio_context;
120 
121     if (!dinfo) {
122         return;
123     }
124 
125     aio_context = bdrv_get_aio_context(bs);
126     aio_context_acquire(aio_context);
127 
128     if (bs->job) {
129         block_job_cancel(bs->job);
130     }
131 
132     aio_context_release(aio_context);
133 
134     dinfo->auto_del = 1;
135 }
136 
137 void blockdev_auto_del(BlockBackend *blk)
138 {
139     DriveInfo *dinfo = blk_legacy_dinfo(blk);
140 
141     if (dinfo && dinfo->auto_del) {
142         blk_unref(blk);
143     }
144 }
145 
146 /**
147  * Returns the current mapping of how many units per bus
148  * a particular interface can support.
149  *
150  *  A positive integer indicates n units per bus.
151  *  0 implies the mapping has not been established.
152  * -1 indicates an invalid BlockInterfaceType was given.
153  */
154 int drive_get_max_devs(BlockInterfaceType type)
155 {
156     if (type >= IF_IDE && type < IF_COUNT) {
157         return if_max_devs[type];
158     }
159 
160     return -1;
161 }
162 
163 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
164 {
165     int max_devs = if_max_devs[type];
166     return max_devs ? index / max_devs : 0;
167 }
168 
169 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
170 {
171     int max_devs = if_max_devs[type];
172     return max_devs ? index % max_devs : index;
173 }
174 
175 QemuOpts *drive_def(const char *optstr)
176 {
177     return qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
178 }
179 
180 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
181                     const char *optstr)
182 {
183     QemuOpts *opts;
184 
185     opts = drive_def(optstr);
186     if (!opts) {
187         return NULL;
188     }
189     if (type != IF_DEFAULT) {
190         qemu_opt_set(opts, "if", if_name[type], &error_abort);
191     }
192     if (index >= 0) {
193         qemu_opt_set_number(opts, "index", index, &error_abort);
194     }
195     if (file)
196         qemu_opt_set(opts, "file", file, &error_abort);
197     return opts;
198 }
199 
200 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
201 {
202     BlockBackend *blk;
203     DriveInfo *dinfo;
204 
205     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
206         dinfo = blk_legacy_dinfo(blk);
207         if (dinfo && dinfo->type == type
208             && dinfo->bus == bus && dinfo->unit == unit) {
209             return dinfo;
210         }
211     }
212 
213     return NULL;
214 }
215 
216 bool drive_check_orphaned(void)
217 {
218     BlockBackend *blk;
219     DriveInfo *dinfo;
220     bool rs = false;
221 
222     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
223         dinfo = blk_legacy_dinfo(blk);
224         /* If dinfo->bdrv->dev is NULL, it has no device attached. */
225         /* Unless this is a default drive, this may be an oversight. */
226         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
227             dinfo->type != IF_NONE) {
228             fprintf(stderr, "Warning: Orphaned drive without device: "
229                     "id=%s,file=%s,if=%s,bus=%d,unit=%d\n",
230                     blk_name(blk), blk_bs(blk)->filename, if_name[dinfo->type],
231                     dinfo->bus, dinfo->unit);
232             rs = true;
233         }
234     }
235 
236     return rs;
237 }
238 
239 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
240 {
241     return drive_get(type,
242                      drive_index_to_bus_id(type, index),
243                      drive_index_to_unit_id(type, index));
244 }
245 
246 int drive_get_max_bus(BlockInterfaceType type)
247 {
248     int max_bus;
249     BlockBackend *blk;
250     DriveInfo *dinfo;
251 
252     max_bus = -1;
253     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
254         dinfo = blk_legacy_dinfo(blk);
255         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
256             max_bus = dinfo->bus;
257         }
258     }
259     return max_bus;
260 }
261 
262 /* Get a block device.  This should only be used for single-drive devices
263    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
264    appropriate bus.  */
265 DriveInfo *drive_get_next(BlockInterfaceType type)
266 {
267     static int next_block_unit[IF_COUNT];
268 
269     return drive_get(type, 0, next_block_unit[type]++);
270 }
271 
272 static void bdrv_format_print(void *opaque, const char *name)
273 {
274     error_printf(" %s", name);
275 }
276 
277 typedef struct {
278     QEMUBH *bh;
279     BlockDriverState *bs;
280 } BDRVPutRefBH;
281 
282 static void bdrv_put_ref_bh(void *opaque)
283 {
284     BDRVPutRefBH *s = opaque;
285 
286     bdrv_unref(s->bs);
287     qemu_bh_delete(s->bh);
288     g_free(s);
289 }
290 
291 /*
292  * Release a BDS reference in a BH
293  *
294  * It is not safe to use bdrv_unref() from a callback function when the callers
295  * still need the BlockDriverState.  In such cases we schedule a BH to release
296  * the reference.
297  */
298 static void bdrv_put_ref_bh_schedule(BlockDriverState *bs)
299 {
300     BDRVPutRefBH *s;
301 
302     s = g_new(BDRVPutRefBH, 1);
303     s->bh = qemu_bh_new(bdrv_put_ref_bh, s);
304     s->bs = bs;
305     qemu_bh_schedule(s->bh);
306 }
307 
308 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
309 {
310     if (!strcmp(buf, "ignore")) {
311         return BLOCKDEV_ON_ERROR_IGNORE;
312     } else if (!is_read && !strcmp(buf, "enospc")) {
313         return BLOCKDEV_ON_ERROR_ENOSPC;
314     } else if (!strcmp(buf, "stop")) {
315         return BLOCKDEV_ON_ERROR_STOP;
316     } else if (!strcmp(buf, "report")) {
317         return BLOCKDEV_ON_ERROR_REPORT;
318     } else {
319         error_setg(errp, "'%s' invalid %s error action",
320                    buf, is_read ? "read" : "write");
321         return -1;
322     }
323 }
324 
325 static bool check_throttle_config(ThrottleConfig *cfg, Error **errp)
326 {
327     if (throttle_conflicting(cfg)) {
328         error_setg(errp, "bps/iops/max total values and read/write values"
329                          " cannot be used at the same time");
330         return false;
331     }
332 
333     if (!throttle_is_valid(cfg)) {
334         error_setg(errp, "bps/iops/maxs values must be 0 or greater");
335         return false;
336     }
337 
338     return true;
339 }
340 
341 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
342 
343 /* Takes the ownership of bs_opts */
344 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
345                                    Error **errp)
346 {
347     const char *buf;
348     int ro = 0;
349     int bdrv_flags = 0;
350     int on_read_error, on_write_error;
351     BlockBackend *blk;
352     BlockDriverState *bs;
353     ThrottleConfig cfg;
354     int snapshot = 0;
355     bool copy_on_read;
356     Error *error = NULL;
357     QemuOpts *opts;
358     const char *id;
359     bool has_driver_specific_opts;
360     BlockdevDetectZeroesOptions detect_zeroes;
361     const char *throttling_group;
362 
363     /* Check common options by copying from bs_opts to opts, all other options
364      * stay in bs_opts for processing by bdrv_open(). */
365     id = qdict_get_try_str(bs_opts, "id");
366     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
367     if (error) {
368         error_propagate(errp, error);
369         goto err_no_opts;
370     }
371 
372     qemu_opts_absorb_qdict(opts, bs_opts, &error);
373     if (error) {
374         error_propagate(errp, error);
375         goto early_err;
376     }
377 
378     if (id) {
379         qdict_del(bs_opts, "id");
380     }
381 
382     has_driver_specific_opts = !!qdict_size(bs_opts);
383 
384     /* extract parameters */
385     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
386     ro = qemu_opt_get_bool(opts, "read-only", 0);
387     copy_on_read = qemu_opt_get_bool(opts, "copy-on-read", false);
388 
389     if ((buf = qemu_opt_get(opts, "discard")) != NULL) {
390         if (bdrv_parse_discard_flags(buf, &bdrv_flags) != 0) {
391             error_setg(errp, "invalid discard option");
392             goto early_err;
393         }
394     }
395 
396     if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true)) {
397         bdrv_flags |= BDRV_O_CACHE_WB;
398     }
399     if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
400         bdrv_flags |= BDRV_O_NOCACHE;
401     }
402     if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
403         bdrv_flags |= BDRV_O_NO_FLUSH;
404     }
405 
406 #ifdef CONFIG_LINUX_AIO
407     if ((buf = qemu_opt_get(opts, "aio")) != NULL) {
408         if (!strcmp(buf, "native")) {
409             bdrv_flags |= BDRV_O_NATIVE_AIO;
410         } else if (!strcmp(buf, "threads")) {
411             /* this is the default */
412         } else {
413            error_setg(errp, "invalid aio option");
414            goto early_err;
415         }
416     }
417 #endif
418 
419     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
420         if (is_help_option(buf)) {
421             error_printf("Supported formats:");
422             bdrv_iterate_format(bdrv_format_print, NULL);
423             error_printf("\n");
424             goto early_err;
425         }
426 
427         if (qdict_haskey(bs_opts, "driver")) {
428             error_setg(errp, "Cannot specify both 'driver' and 'format'");
429             goto early_err;
430         }
431         qdict_put(bs_opts, "driver", qstring_from_str(buf));
432     }
433 
434     /* disk I/O throttling */
435     memset(&cfg, 0, sizeof(cfg));
436     cfg.buckets[THROTTLE_BPS_TOTAL].avg =
437         qemu_opt_get_number(opts, "throttling.bps-total", 0);
438     cfg.buckets[THROTTLE_BPS_READ].avg  =
439         qemu_opt_get_number(opts, "throttling.bps-read", 0);
440     cfg.buckets[THROTTLE_BPS_WRITE].avg =
441         qemu_opt_get_number(opts, "throttling.bps-write", 0);
442     cfg.buckets[THROTTLE_OPS_TOTAL].avg =
443         qemu_opt_get_number(opts, "throttling.iops-total", 0);
444     cfg.buckets[THROTTLE_OPS_READ].avg =
445         qemu_opt_get_number(opts, "throttling.iops-read", 0);
446     cfg.buckets[THROTTLE_OPS_WRITE].avg =
447         qemu_opt_get_number(opts, "throttling.iops-write", 0);
448 
449     cfg.buckets[THROTTLE_BPS_TOTAL].max =
450         qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
451     cfg.buckets[THROTTLE_BPS_READ].max  =
452         qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
453     cfg.buckets[THROTTLE_BPS_WRITE].max =
454         qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
455     cfg.buckets[THROTTLE_OPS_TOTAL].max =
456         qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
457     cfg.buckets[THROTTLE_OPS_READ].max =
458         qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
459     cfg.buckets[THROTTLE_OPS_WRITE].max =
460         qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
461 
462     cfg.op_size = qemu_opt_get_number(opts, "throttling.iops-size", 0);
463 
464     throttling_group = qemu_opt_get(opts, "throttling.group");
465 
466     if (!check_throttle_config(&cfg, &error)) {
467         error_propagate(errp, error);
468         goto early_err;
469     }
470 
471     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
472     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
473         on_write_error = parse_block_error_action(buf, 0, &error);
474         if (error) {
475             error_propagate(errp, error);
476             goto early_err;
477         }
478     }
479 
480     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
481     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
482         on_read_error = parse_block_error_action(buf, 1, &error);
483         if (error) {
484             error_propagate(errp, error);
485             goto early_err;
486         }
487     }
488 
489     detect_zeroes =
490         qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
491                         qemu_opt_get(opts, "detect-zeroes"),
492                         BLOCKDEV_DETECT_ZEROES_OPTIONS_MAX,
493                         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
494                         &error);
495     if (error) {
496         error_propagate(errp, error);
497         goto early_err;
498     }
499 
500     if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
501         !(bdrv_flags & BDRV_O_UNMAP)) {
502         error_setg(errp, "setting detect-zeroes to unmap is not allowed "
503                          "without setting discard operation to unmap");
504         goto early_err;
505     }
506 
507     /* init */
508     if ((!file || !*file) && !has_driver_specific_opts) {
509         blk = blk_new_with_bs(qemu_opts_id(opts), errp);
510         if (!blk) {
511             goto early_err;
512         }
513 
514         bs = blk_bs(blk);
515         bs->open_flags = snapshot ? BDRV_O_SNAPSHOT : 0;
516         bs->read_only = ro;
517 
518         QDECREF(bs_opts);
519     } else {
520         if (file && !*file) {
521             file = NULL;
522         }
523 
524         if (snapshot) {
525             /* always use cache=unsafe with snapshot */
526             bdrv_flags &= ~BDRV_O_CACHE_MASK;
527             bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
528         }
529 
530         if (copy_on_read) {
531             bdrv_flags |= BDRV_O_COPY_ON_READ;
532         }
533 
534         if (runstate_check(RUN_STATE_INMIGRATE)) {
535             bdrv_flags |= BDRV_O_INCOMING;
536         }
537 
538         bdrv_flags |= ro ? 0 : BDRV_O_RDWR;
539 
540         blk = blk_new_open(qemu_opts_id(opts), file, NULL, bs_opts, bdrv_flags,
541                            errp);
542         if (!blk) {
543             goto err_no_bs_opts;
544         }
545         bs = blk_bs(blk);
546     }
547 
548     bs->detect_zeroes = detect_zeroes;
549 
550     bdrv_set_on_error(bs, on_read_error, on_write_error);
551 
552     /* disk I/O throttling */
553     if (throttle_enabled(&cfg)) {
554         if (!throttling_group) {
555             throttling_group = blk_name(blk);
556         }
557         bdrv_io_limits_enable(bs, throttling_group);
558         bdrv_set_io_limits(bs, &cfg);
559     }
560 
561     if (bdrv_key_required(bs)) {
562         autostart = 0;
563     }
564 
565 err_no_bs_opts:
566     qemu_opts_del(opts);
567     return blk;
568 
569 early_err:
570     qemu_opts_del(opts);
571 err_no_opts:
572     QDECREF(bs_opts);
573     return NULL;
574 }
575 
576 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
577                             Error **errp)
578 {
579     const char *value;
580 
581     value = qemu_opt_get(opts, from);
582     if (value) {
583         if (qemu_opt_find(opts, to)) {
584             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
585                        "same time", to, from);
586             return;
587         }
588     }
589 
590     /* rename all items in opts */
591     while ((value = qemu_opt_get(opts, from))) {
592         qemu_opt_set(opts, to, value, &error_abort);
593         qemu_opt_unset(opts, from);
594     }
595 }
596 
597 QemuOptsList qemu_legacy_drive_opts = {
598     .name = "drive",
599     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
600     .desc = {
601         {
602             .name = "bus",
603             .type = QEMU_OPT_NUMBER,
604             .help = "bus number",
605         },{
606             .name = "unit",
607             .type = QEMU_OPT_NUMBER,
608             .help = "unit number (i.e. lun for scsi)",
609         },{
610             .name = "index",
611             .type = QEMU_OPT_NUMBER,
612             .help = "index number",
613         },{
614             .name = "media",
615             .type = QEMU_OPT_STRING,
616             .help = "media type (disk, cdrom)",
617         },{
618             .name = "if",
619             .type = QEMU_OPT_STRING,
620             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
621         },{
622             .name = "cyls",
623             .type = QEMU_OPT_NUMBER,
624             .help = "number of cylinders (ide disk geometry)",
625         },{
626             .name = "heads",
627             .type = QEMU_OPT_NUMBER,
628             .help = "number of heads (ide disk geometry)",
629         },{
630             .name = "secs",
631             .type = QEMU_OPT_NUMBER,
632             .help = "number of sectors (ide disk geometry)",
633         },{
634             .name = "trans",
635             .type = QEMU_OPT_STRING,
636             .help = "chs translation (auto, lba, none)",
637         },{
638             .name = "boot",
639             .type = QEMU_OPT_BOOL,
640             .help = "(deprecated, ignored)",
641         },{
642             .name = "addr",
643             .type = QEMU_OPT_STRING,
644             .help = "pci address (virtio only)",
645         },{
646             .name = "serial",
647             .type = QEMU_OPT_STRING,
648             .help = "disk serial number",
649         },{
650             .name = "file",
651             .type = QEMU_OPT_STRING,
652             .help = "file name",
653         },
654 
655         /* Options that are passed on, but have special semantics with -drive */
656         {
657             .name = "read-only",
658             .type = QEMU_OPT_BOOL,
659             .help = "open drive file as read-only",
660         },{
661             .name = "rerror",
662             .type = QEMU_OPT_STRING,
663             .help = "read error action",
664         },{
665             .name = "werror",
666             .type = QEMU_OPT_STRING,
667             .help = "write error action",
668         },{
669             .name = "copy-on-read",
670             .type = QEMU_OPT_BOOL,
671             .help = "copy read data from backing file into image file",
672         },
673 
674         { /* end of list */ }
675     },
676 };
677 
678 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
679 {
680     const char *value;
681     BlockBackend *blk;
682     DriveInfo *dinfo = NULL;
683     QDict *bs_opts;
684     QemuOpts *legacy_opts;
685     DriveMediaType media = MEDIA_DISK;
686     BlockInterfaceType type;
687     int cyls, heads, secs, translation;
688     int max_devs, bus_id, unit_id, index;
689     const char *devaddr;
690     const char *werror, *rerror;
691     bool read_only = false;
692     bool copy_on_read;
693     const char *serial;
694     const char *filename;
695     Error *local_err = NULL;
696     int i;
697 
698     /* Change legacy command line options into QMP ones */
699     static const struct {
700         const char *from;
701         const char *to;
702     } opt_renames[] = {
703         { "iops",           "throttling.iops-total" },
704         { "iops_rd",        "throttling.iops-read" },
705         { "iops_wr",        "throttling.iops-write" },
706 
707         { "bps",            "throttling.bps-total" },
708         { "bps_rd",         "throttling.bps-read" },
709         { "bps_wr",         "throttling.bps-write" },
710 
711         { "iops_max",       "throttling.iops-total-max" },
712         { "iops_rd_max",    "throttling.iops-read-max" },
713         { "iops_wr_max",    "throttling.iops-write-max" },
714 
715         { "bps_max",        "throttling.bps-total-max" },
716         { "bps_rd_max",     "throttling.bps-read-max" },
717         { "bps_wr_max",     "throttling.bps-write-max" },
718 
719         { "iops_size",      "throttling.iops-size" },
720 
721         { "group",          "throttling.group" },
722 
723         { "readonly",       "read-only" },
724     };
725 
726     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
727         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
728                         &local_err);
729         if (local_err) {
730             error_report_err(local_err);
731             return NULL;
732         }
733     }
734 
735     value = qemu_opt_get(all_opts, "cache");
736     if (value) {
737         int flags = 0;
738 
739         if (bdrv_parse_cache_flags(value, &flags) != 0) {
740             error_report("invalid cache option");
741             return NULL;
742         }
743 
744         /* Specific options take precedence */
745         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
746             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
747                               !!(flags & BDRV_O_CACHE_WB), &error_abort);
748         }
749         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
750             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
751                               !!(flags & BDRV_O_NOCACHE), &error_abort);
752         }
753         if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
754             qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
755                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
756         }
757         qemu_opt_unset(all_opts, "cache");
758     }
759 
760     /* Get a QDict for processing the options */
761     bs_opts = qdict_new();
762     qemu_opts_to_qdict(all_opts, bs_opts);
763 
764     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
765                                    &error_abort);
766     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
767     if (local_err) {
768         error_report_err(local_err);
769         goto fail;
770     }
771 
772     /* Deprecated option boot=[on|off] */
773     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
774         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
775                 "ignored. Future versions will reject this parameter. Please "
776                 "update your scripts.\n");
777     }
778 
779     /* Media type */
780     value = qemu_opt_get(legacy_opts, "media");
781     if (value) {
782         if (!strcmp(value, "disk")) {
783             media = MEDIA_DISK;
784         } else if (!strcmp(value, "cdrom")) {
785             media = MEDIA_CDROM;
786             read_only = true;
787         } else {
788             error_report("'%s' invalid media", value);
789             goto fail;
790         }
791     }
792 
793     /* copy-on-read is disabled with a warning for read-only devices */
794     read_only |= qemu_opt_get_bool(legacy_opts, "read-only", false);
795     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
796 
797     if (read_only && copy_on_read) {
798         error_report("warning: disabling copy-on-read on read-only drive");
799         copy_on_read = false;
800     }
801 
802     qdict_put(bs_opts, "read-only",
803               qstring_from_str(read_only ? "on" : "off"));
804     qdict_put(bs_opts, "copy-on-read",
805               qstring_from_str(copy_on_read ? "on" :"off"));
806 
807     /* Controller type */
808     value = qemu_opt_get(legacy_opts, "if");
809     if (value) {
810         for (type = 0;
811              type < IF_COUNT && strcmp(value, if_name[type]);
812              type++) {
813         }
814         if (type == IF_COUNT) {
815             error_report("unsupported bus type '%s'", value);
816             goto fail;
817         }
818     } else {
819         type = block_default_type;
820     }
821 
822     /* Geometry */
823     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
824     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
825     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
826 
827     if (cyls || heads || secs) {
828         if (cyls < 1) {
829             error_report("invalid physical cyls number");
830             goto fail;
831         }
832         if (heads < 1) {
833             error_report("invalid physical heads number");
834             goto fail;
835         }
836         if (secs < 1) {
837             error_report("invalid physical secs number");
838             goto fail;
839         }
840     }
841 
842     translation = BIOS_ATA_TRANSLATION_AUTO;
843     value = qemu_opt_get(legacy_opts, "trans");
844     if (value != NULL) {
845         if (!cyls) {
846             error_report("'%s' trans must be used with cyls, heads and secs",
847                          value);
848             goto fail;
849         }
850         if (!strcmp(value, "none")) {
851             translation = BIOS_ATA_TRANSLATION_NONE;
852         } else if (!strcmp(value, "lba")) {
853             translation = BIOS_ATA_TRANSLATION_LBA;
854         } else if (!strcmp(value, "large")) {
855             translation = BIOS_ATA_TRANSLATION_LARGE;
856         } else if (!strcmp(value, "rechs")) {
857             translation = BIOS_ATA_TRANSLATION_RECHS;
858         } else if (!strcmp(value, "auto")) {
859             translation = BIOS_ATA_TRANSLATION_AUTO;
860         } else {
861             error_report("'%s' invalid translation type", value);
862             goto fail;
863         }
864     }
865 
866     if (media == MEDIA_CDROM) {
867         if (cyls || secs || heads) {
868             error_report("CHS can't be set with media=cdrom");
869             goto fail;
870         }
871     }
872 
873     /* Device address specified by bus/unit or index.
874      * If none was specified, try to find the first free one. */
875     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
876     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
877     index   = qemu_opt_get_number(legacy_opts, "index", -1);
878 
879     max_devs = if_max_devs[type];
880 
881     if (index != -1) {
882         if (bus_id != 0 || unit_id != -1) {
883             error_report("index cannot be used with bus and unit");
884             goto fail;
885         }
886         bus_id = drive_index_to_bus_id(type, index);
887         unit_id = drive_index_to_unit_id(type, index);
888     }
889 
890     if (unit_id == -1) {
891        unit_id = 0;
892        while (drive_get(type, bus_id, unit_id) != NULL) {
893            unit_id++;
894            if (max_devs && unit_id >= max_devs) {
895                unit_id -= max_devs;
896                bus_id++;
897            }
898        }
899     }
900 
901     if (max_devs && unit_id >= max_devs) {
902         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
903         goto fail;
904     }
905 
906     if (drive_get(type, bus_id, unit_id) != NULL) {
907         error_report("drive with bus=%d, unit=%d (index=%d) exists",
908                      bus_id, unit_id, index);
909         goto fail;
910     }
911 
912     /* Serial number */
913     serial = qemu_opt_get(legacy_opts, "serial");
914 
915     /* no id supplied -> create one */
916     if (qemu_opts_id(all_opts) == NULL) {
917         char *new_id;
918         const char *mediastr = "";
919         if (type == IF_IDE || type == IF_SCSI) {
920             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
921         }
922         if (max_devs) {
923             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
924                                      mediastr, unit_id);
925         } else {
926             new_id = g_strdup_printf("%s%s%i", if_name[type],
927                                      mediastr, unit_id);
928         }
929         qdict_put(bs_opts, "id", qstring_from_str(new_id));
930         g_free(new_id);
931     }
932 
933     /* Add virtio block device */
934     devaddr = qemu_opt_get(legacy_opts, "addr");
935     if (devaddr && type != IF_VIRTIO) {
936         error_report("addr is not supported by this bus type");
937         goto fail;
938     }
939 
940     if (type == IF_VIRTIO) {
941         QemuOpts *devopts;
942         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
943                                    &error_abort);
944         if (arch_type == QEMU_ARCH_S390X) {
945             qemu_opt_set(devopts, "driver", "virtio-blk-ccw", &error_abort);
946         } else {
947             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
948         }
949         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
950                      &error_abort);
951         if (devaddr) {
952             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
953         }
954     }
955 
956     filename = qemu_opt_get(legacy_opts, "file");
957 
958     /* Check werror/rerror compatibility with if=... */
959     werror = qemu_opt_get(legacy_opts, "werror");
960     if (werror != NULL) {
961         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
962             type != IF_NONE) {
963             error_report("werror is not supported by this bus type");
964             goto fail;
965         }
966         qdict_put(bs_opts, "werror", qstring_from_str(werror));
967     }
968 
969     rerror = qemu_opt_get(legacy_opts, "rerror");
970     if (rerror != NULL) {
971         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
972             type != IF_NONE) {
973             error_report("rerror is not supported by this bus type");
974             goto fail;
975         }
976         qdict_put(bs_opts, "rerror", qstring_from_str(rerror));
977     }
978 
979     /* Actual block device init: Functionality shared with blockdev-add */
980     blk = blockdev_init(filename, bs_opts, &local_err);
981     bs_opts = NULL;
982     if (!blk) {
983         if (local_err) {
984             error_report_err(local_err);
985         }
986         goto fail;
987     } else {
988         assert(!local_err);
989     }
990 
991     /* Create legacy DriveInfo */
992     dinfo = g_malloc0(sizeof(*dinfo));
993     dinfo->opts = all_opts;
994 
995     dinfo->cyls = cyls;
996     dinfo->heads = heads;
997     dinfo->secs = secs;
998     dinfo->trans = translation;
999 
1000     dinfo->type = type;
1001     dinfo->bus = bus_id;
1002     dinfo->unit = unit_id;
1003     dinfo->devaddr = devaddr;
1004     dinfo->serial = g_strdup(serial);
1005 
1006     blk_set_legacy_dinfo(blk, dinfo);
1007 
1008     switch(type) {
1009     case IF_IDE:
1010     case IF_SCSI:
1011     case IF_XEN:
1012     case IF_NONE:
1013         dinfo->media_cd = media == MEDIA_CDROM;
1014         break;
1015     default:
1016         break;
1017     }
1018 
1019 fail:
1020     qemu_opts_del(legacy_opts);
1021     QDECREF(bs_opts);
1022     return dinfo;
1023 }
1024 
1025 void hmp_commit(Monitor *mon, const QDict *qdict)
1026 {
1027     const char *device = qdict_get_str(qdict, "device");
1028     BlockBackend *blk;
1029     int ret;
1030 
1031     if (!strcmp(device, "all")) {
1032         ret = bdrv_commit_all();
1033     } else {
1034         blk = blk_by_name(device);
1035         if (!blk) {
1036             monitor_printf(mon, "Device '%s' not found\n", device);
1037             return;
1038         }
1039         ret = bdrv_commit(blk_bs(blk));
1040     }
1041     if (ret < 0) {
1042         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1043                        strerror(-ret));
1044     }
1045 }
1046 
1047 static void blockdev_do_action(int kind, void *data, Error **errp)
1048 {
1049     TransactionAction action;
1050     TransactionActionList list;
1051 
1052     action.kind = kind;
1053     action.data = data;
1054     list.value = &action;
1055     list.next = NULL;
1056     qmp_transaction(&list, errp);
1057 }
1058 
1059 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1060                                 bool has_node_name, const char *node_name,
1061                                 const char *snapshot_file,
1062                                 bool has_snapshot_node_name,
1063                                 const char *snapshot_node_name,
1064                                 bool has_format, const char *format,
1065                                 bool has_mode, NewImageMode mode, Error **errp)
1066 {
1067     BlockdevSnapshot snapshot = {
1068         .has_device = has_device,
1069         .device = (char *) device,
1070         .has_node_name = has_node_name,
1071         .node_name = (char *) node_name,
1072         .snapshot_file = (char *) snapshot_file,
1073         .has_snapshot_node_name = has_snapshot_node_name,
1074         .snapshot_node_name = (char *) snapshot_node_name,
1075         .has_format = has_format,
1076         .format = (char *) format,
1077         .has_mode = has_mode,
1078         .mode = mode,
1079     };
1080     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1081                        &snapshot, errp);
1082 }
1083 
1084 void qmp_blockdev_snapshot_internal_sync(const char *device,
1085                                          const char *name,
1086                                          Error **errp)
1087 {
1088     BlockdevSnapshotInternal snapshot = {
1089         .device = (char *) device,
1090         .name = (char *) name
1091     };
1092 
1093     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1094                        &snapshot, errp);
1095 }
1096 
1097 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1098                                                          bool has_id,
1099                                                          const char *id,
1100                                                          bool has_name,
1101                                                          const char *name,
1102                                                          Error **errp)
1103 {
1104     BlockDriverState *bs;
1105     BlockBackend *blk;
1106     AioContext *aio_context;
1107     QEMUSnapshotInfo sn;
1108     Error *local_err = NULL;
1109     SnapshotInfo *info = NULL;
1110     int ret;
1111 
1112     blk = blk_by_name(device);
1113     if (!blk) {
1114         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1115                   "Device '%s' not found", device);
1116         return NULL;
1117     }
1118     bs = blk_bs(blk);
1119 
1120     if (!has_id) {
1121         id = NULL;
1122     }
1123 
1124     if (!has_name) {
1125         name = NULL;
1126     }
1127 
1128     if (!id && !name) {
1129         error_setg(errp, "Name or id must be provided");
1130         return NULL;
1131     }
1132 
1133     aio_context = bdrv_get_aio_context(bs);
1134     aio_context_acquire(aio_context);
1135 
1136     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1137         goto out_aio_context;
1138     }
1139 
1140     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1141     if (local_err) {
1142         error_propagate(errp, local_err);
1143         goto out_aio_context;
1144     }
1145     if (!ret) {
1146         error_setg(errp,
1147                    "Snapshot with id '%s' and name '%s' does not exist on "
1148                    "device '%s'",
1149                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1150         goto out_aio_context;
1151     }
1152 
1153     bdrv_snapshot_delete(bs, id, name, &local_err);
1154     if (local_err) {
1155         error_propagate(errp, local_err);
1156         goto out_aio_context;
1157     }
1158 
1159     aio_context_release(aio_context);
1160 
1161     info = g_new0(SnapshotInfo, 1);
1162     info->id = g_strdup(sn.id_str);
1163     info->name = g_strdup(sn.name);
1164     info->date_nsec = sn.date_nsec;
1165     info->date_sec = sn.date_sec;
1166     info->vm_state_size = sn.vm_state_size;
1167     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1168     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1169 
1170     return info;
1171 
1172 out_aio_context:
1173     aio_context_release(aio_context);
1174     return NULL;
1175 }
1176 
1177 /**
1178  * block_dirty_bitmap_lookup:
1179  * Return a dirty bitmap (if present), after validating
1180  * the node reference and bitmap names.
1181  *
1182  * @node: The name of the BDS node to search for bitmaps
1183  * @name: The name of the bitmap to search for
1184  * @pbs: Output pointer for BDS lookup, if desired. Can be NULL.
1185  * @paio: Output pointer for aio_context acquisition, if desired. Can be NULL.
1186  * @errp: Output pointer for error information. Can be NULL.
1187  *
1188  * @return: A bitmap object on success, or NULL on failure.
1189  */
1190 static BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1191                                                   const char *name,
1192                                                   BlockDriverState **pbs,
1193                                                   AioContext **paio,
1194                                                   Error **errp)
1195 {
1196     BlockDriverState *bs;
1197     BdrvDirtyBitmap *bitmap;
1198     AioContext *aio_context;
1199 
1200     if (!node) {
1201         error_setg(errp, "Node cannot be NULL");
1202         return NULL;
1203     }
1204     if (!name) {
1205         error_setg(errp, "Bitmap name cannot be NULL");
1206         return NULL;
1207     }
1208     bs = bdrv_lookup_bs(node, node, NULL);
1209     if (!bs) {
1210         error_setg(errp, "Node '%s' not found", node);
1211         return NULL;
1212     }
1213 
1214     aio_context = bdrv_get_aio_context(bs);
1215     aio_context_acquire(aio_context);
1216 
1217     bitmap = bdrv_find_dirty_bitmap(bs, name);
1218     if (!bitmap) {
1219         error_setg(errp, "Dirty bitmap '%s' not found", name);
1220         goto fail;
1221     }
1222 
1223     if (pbs) {
1224         *pbs = bs;
1225     }
1226     if (paio) {
1227         *paio = aio_context;
1228     } else {
1229         aio_context_release(aio_context);
1230     }
1231 
1232     return bitmap;
1233 
1234  fail:
1235     aio_context_release(aio_context);
1236     return NULL;
1237 }
1238 
1239 /* New and old BlockDriverState structs for atomic group operations */
1240 
1241 typedef struct BlkTransactionState BlkTransactionState;
1242 
1243 /* Only prepare() may fail. In a single transaction, only one of commit() or
1244    abort() will be called, clean() will always be called if it present. */
1245 typedef struct BdrvActionOps {
1246     /* Size of state struct, in bytes. */
1247     size_t instance_size;
1248     /* Prepare the work, must NOT be NULL. */
1249     void (*prepare)(BlkTransactionState *common, Error **errp);
1250     /* Commit the changes, can be NULL. */
1251     void (*commit)(BlkTransactionState *common);
1252     /* Abort the changes on fail, can be NULL. */
1253     void (*abort)(BlkTransactionState *common);
1254     /* Clean up resource in the end, can be NULL. */
1255     void (*clean)(BlkTransactionState *common);
1256 } BdrvActionOps;
1257 
1258 /*
1259  * This structure must be arranged as first member in child type, assuming
1260  * that compiler will also arrange it to the same address with parent instance.
1261  * Later it will be used in free().
1262  */
1263 struct BlkTransactionState {
1264     TransactionAction *action;
1265     const BdrvActionOps *ops;
1266     QSIMPLEQ_ENTRY(BlkTransactionState) entry;
1267 };
1268 
1269 /* internal snapshot private data */
1270 typedef struct InternalSnapshotState {
1271     BlkTransactionState common;
1272     BlockDriverState *bs;
1273     AioContext *aio_context;
1274     QEMUSnapshotInfo sn;
1275 } InternalSnapshotState;
1276 
1277 static void internal_snapshot_prepare(BlkTransactionState *common,
1278                                       Error **errp)
1279 {
1280     Error *local_err = NULL;
1281     const char *device;
1282     const char *name;
1283     BlockBackend *blk;
1284     BlockDriverState *bs;
1285     QEMUSnapshotInfo old_sn, *sn;
1286     bool ret;
1287     qemu_timeval tv;
1288     BlockdevSnapshotInternal *internal;
1289     InternalSnapshotState *state;
1290     int ret1;
1291 
1292     g_assert(common->action->kind ==
1293              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1294     internal = common->action->blockdev_snapshot_internal_sync;
1295     state = DO_UPCAST(InternalSnapshotState, common, common);
1296 
1297     /* 1. parse input */
1298     device = internal->device;
1299     name = internal->name;
1300 
1301     /* 2. check for validation */
1302     blk = blk_by_name(device);
1303     if (!blk) {
1304         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1305                   "Device '%s' not found", device);
1306         return;
1307     }
1308     bs = blk_bs(blk);
1309 
1310     /* AioContext is released in .clean() */
1311     state->aio_context = bdrv_get_aio_context(bs);
1312     aio_context_acquire(state->aio_context);
1313 
1314     if (!bdrv_is_inserted(bs)) {
1315         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1316         return;
1317     }
1318 
1319     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1320         return;
1321     }
1322 
1323     if (bdrv_is_read_only(bs)) {
1324         error_setg(errp, "Device '%s' is read only", device);
1325         return;
1326     }
1327 
1328     if (!bdrv_can_snapshot(bs)) {
1329         error_setg(errp, "Block format '%s' used by device '%s' "
1330                    "does not support internal snapshots",
1331                    bs->drv->format_name, device);
1332         return;
1333     }
1334 
1335     if (!strlen(name)) {
1336         error_setg(errp, "Name is empty");
1337         return;
1338     }
1339 
1340     /* check whether a snapshot with name exist */
1341     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1342                                             &local_err);
1343     if (local_err) {
1344         error_propagate(errp, local_err);
1345         return;
1346     } else if (ret) {
1347         error_setg(errp,
1348                    "Snapshot with name '%s' already exists on device '%s'",
1349                    name, device);
1350         return;
1351     }
1352 
1353     /* 3. take the snapshot */
1354     sn = &state->sn;
1355     pstrcpy(sn->name, sizeof(sn->name), name);
1356     qemu_gettimeofday(&tv);
1357     sn->date_sec = tv.tv_sec;
1358     sn->date_nsec = tv.tv_usec * 1000;
1359     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1360 
1361     ret1 = bdrv_snapshot_create(bs, sn);
1362     if (ret1 < 0) {
1363         error_setg_errno(errp, -ret1,
1364                          "Failed to create snapshot '%s' on device '%s'",
1365                          name, device);
1366         return;
1367     }
1368 
1369     /* 4. succeed, mark a snapshot is created */
1370     state->bs = bs;
1371 }
1372 
1373 static void internal_snapshot_abort(BlkTransactionState *common)
1374 {
1375     InternalSnapshotState *state =
1376                              DO_UPCAST(InternalSnapshotState, common, common);
1377     BlockDriverState *bs = state->bs;
1378     QEMUSnapshotInfo *sn = &state->sn;
1379     Error *local_error = NULL;
1380 
1381     if (!bs) {
1382         return;
1383     }
1384 
1385     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1386         error_report("Failed to delete snapshot with id '%s' and name '%s' on "
1387                      "device '%s' in abort: %s",
1388                      sn->id_str,
1389                      sn->name,
1390                      bdrv_get_device_name(bs),
1391                      error_get_pretty(local_error));
1392         error_free(local_error);
1393     }
1394 }
1395 
1396 static void internal_snapshot_clean(BlkTransactionState *common)
1397 {
1398     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1399                                              common, common);
1400 
1401     if (state->aio_context) {
1402         aio_context_release(state->aio_context);
1403     }
1404 }
1405 
1406 /* external snapshot private data */
1407 typedef struct ExternalSnapshotState {
1408     BlkTransactionState common;
1409     BlockDriverState *old_bs;
1410     BlockDriverState *new_bs;
1411     AioContext *aio_context;
1412 } ExternalSnapshotState;
1413 
1414 static void external_snapshot_prepare(BlkTransactionState *common,
1415                                       Error **errp)
1416 {
1417     BlockDriver *drv;
1418     int flags, ret;
1419     QDict *options = NULL;
1420     Error *local_err = NULL;
1421     bool has_device = false;
1422     const char *device;
1423     bool has_node_name = false;
1424     const char *node_name;
1425     bool has_snapshot_node_name = false;
1426     const char *snapshot_node_name;
1427     const char *new_image_file;
1428     const char *format = "qcow2";
1429     enum NewImageMode mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1430     ExternalSnapshotState *state =
1431                              DO_UPCAST(ExternalSnapshotState, common, common);
1432     TransactionAction *action = common->action;
1433 
1434     /* get parameters */
1435     g_assert(action->kind == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC);
1436 
1437     has_device = action->blockdev_snapshot_sync->has_device;
1438     device = action->blockdev_snapshot_sync->device;
1439     has_node_name = action->blockdev_snapshot_sync->has_node_name;
1440     node_name = action->blockdev_snapshot_sync->node_name;
1441     has_snapshot_node_name =
1442         action->blockdev_snapshot_sync->has_snapshot_node_name;
1443     snapshot_node_name = action->blockdev_snapshot_sync->snapshot_node_name;
1444 
1445     new_image_file = action->blockdev_snapshot_sync->snapshot_file;
1446     if (action->blockdev_snapshot_sync->has_format) {
1447         format = action->blockdev_snapshot_sync->format;
1448     }
1449     if (action->blockdev_snapshot_sync->has_mode) {
1450         mode = action->blockdev_snapshot_sync->mode;
1451     }
1452 
1453     /* start processing */
1454     drv = bdrv_find_format(format);
1455     if (!drv) {
1456         error_setg(errp, QERR_INVALID_BLOCK_FORMAT, format);
1457         return;
1458     }
1459 
1460     state->old_bs = bdrv_lookup_bs(has_device ? device : NULL,
1461                                    has_node_name ? node_name : NULL,
1462                                    &local_err);
1463     if (local_err) {
1464         error_propagate(errp, local_err);
1465         return;
1466     }
1467 
1468     if (has_node_name && !has_snapshot_node_name) {
1469         error_setg(errp, "New snapshot node name missing");
1470         return;
1471     }
1472 
1473     if (has_snapshot_node_name && bdrv_find_node(snapshot_node_name)) {
1474         error_setg(errp, "New snapshot node name already existing");
1475         return;
1476     }
1477 
1478     /* Acquire AioContext now so any threads operating on old_bs stop */
1479     state->aio_context = bdrv_get_aio_context(state->old_bs);
1480     aio_context_acquire(state->aio_context);
1481 
1482     if (!bdrv_is_inserted(state->old_bs)) {
1483         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1484         return;
1485     }
1486 
1487     if (bdrv_op_is_blocked(state->old_bs,
1488                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1489         return;
1490     }
1491 
1492     if (!bdrv_is_read_only(state->old_bs)) {
1493         if (bdrv_flush(state->old_bs)) {
1494             error_setg(errp, QERR_IO_ERROR);
1495             return;
1496         }
1497     }
1498 
1499     if (!bdrv_is_first_non_filter(state->old_bs)) {
1500         error_setg(errp, QERR_FEATURE_DISABLED, "snapshot");
1501         return;
1502     }
1503 
1504     flags = state->old_bs->open_flags;
1505 
1506     /* create new image w/backing file */
1507     if (mode != NEW_IMAGE_MODE_EXISTING) {
1508         bdrv_img_create(new_image_file, format,
1509                         state->old_bs->filename,
1510                         state->old_bs->drv->format_name,
1511                         NULL, -1, flags, &local_err, false);
1512         if (local_err) {
1513             error_propagate(errp, local_err);
1514             return;
1515         }
1516     }
1517 
1518     if (has_snapshot_node_name) {
1519         options = qdict_new();
1520         qdict_put(options, "node-name",
1521                   qstring_from_str(snapshot_node_name));
1522     }
1523 
1524     /* TODO Inherit bs->options or only take explicit options with an
1525      * extended QMP command? */
1526     assert(state->new_bs == NULL);
1527     ret = bdrv_open(&state->new_bs, new_image_file, NULL, options,
1528                     flags | BDRV_O_NO_BACKING, drv, &local_err);
1529     /* We will manually add the backing_hd field to the bs later */
1530     if (ret != 0) {
1531         error_propagate(errp, local_err);
1532     }
1533 }
1534 
1535 static void external_snapshot_commit(BlkTransactionState *common)
1536 {
1537     ExternalSnapshotState *state =
1538                              DO_UPCAST(ExternalSnapshotState, common, common);
1539 
1540     bdrv_set_aio_context(state->new_bs, state->aio_context);
1541 
1542     /* This removes our old bs and adds the new bs */
1543     bdrv_append(state->new_bs, state->old_bs);
1544     /* We don't need (or want) to use the transactional
1545      * bdrv_reopen_multiple() across all the entries at once, because we
1546      * don't want to abort all of them if one of them fails the reopen */
1547     bdrv_reopen(state->new_bs, state->new_bs->open_flags & ~BDRV_O_RDWR,
1548                 NULL);
1549 
1550     aio_context_release(state->aio_context);
1551 }
1552 
1553 static void external_snapshot_abort(BlkTransactionState *common)
1554 {
1555     ExternalSnapshotState *state =
1556                              DO_UPCAST(ExternalSnapshotState, common, common);
1557     if (state->new_bs) {
1558         bdrv_unref(state->new_bs);
1559     }
1560     if (state->aio_context) {
1561         aio_context_release(state->aio_context);
1562     }
1563 }
1564 
1565 typedef struct DriveBackupState {
1566     BlkTransactionState common;
1567     BlockDriverState *bs;
1568     AioContext *aio_context;
1569     BlockJob *job;
1570 } DriveBackupState;
1571 
1572 static void drive_backup_prepare(BlkTransactionState *common, Error **errp)
1573 {
1574     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1575     BlockDriverState *bs;
1576     BlockBackend *blk;
1577     DriveBackup *backup;
1578     Error *local_err = NULL;
1579 
1580     assert(common->action->kind == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1581     backup = common->action->drive_backup;
1582 
1583     blk = blk_by_name(backup->device);
1584     if (!blk) {
1585         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1586                   "Device '%s' not found", backup->device);
1587         return;
1588     }
1589     bs = blk_bs(blk);
1590 
1591     /* AioContext is released in .clean() */
1592     state->aio_context = bdrv_get_aio_context(bs);
1593     aio_context_acquire(state->aio_context);
1594 
1595     qmp_drive_backup(backup->device, backup->target,
1596                      backup->has_format, backup->format,
1597                      backup->sync,
1598                      backup->has_mode, backup->mode,
1599                      backup->has_speed, backup->speed,
1600                      backup->has_bitmap, backup->bitmap,
1601                      backup->has_on_source_error, backup->on_source_error,
1602                      backup->has_on_target_error, backup->on_target_error,
1603                      &local_err);
1604     if (local_err) {
1605         error_propagate(errp, local_err);
1606         return;
1607     }
1608 
1609     state->bs = bs;
1610     state->job = state->bs->job;
1611 }
1612 
1613 static void drive_backup_abort(BlkTransactionState *common)
1614 {
1615     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1616     BlockDriverState *bs = state->bs;
1617 
1618     /* Only cancel if it's the job we started */
1619     if (bs && bs->job && bs->job == state->job) {
1620         block_job_cancel_sync(bs->job);
1621     }
1622 }
1623 
1624 static void drive_backup_clean(BlkTransactionState *common)
1625 {
1626     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1627 
1628     if (state->aio_context) {
1629         aio_context_release(state->aio_context);
1630     }
1631 }
1632 
1633 typedef struct BlockdevBackupState {
1634     BlkTransactionState common;
1635     BlockDriverState *bs;
1636     BlockJob *job;
1637     AioContext *aio_context;
1638 } BlockdevBackupState;
1639 
1640 static void blockdev_backup_prepare(BlkTransactionState *common, Error **errp)
1641 {
1642     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1643     BlockdevBackup *backup;
1644     BlockDriverState *bs, *target;
1645     BlockBackend *blk;
1646     Error *local_err = NULL;
1647 
1648     assert(common->action->kind == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1649     backup = common->action->blockdev_backup;
1650 
1651     blk = blk_by_name(backup->device);
1652     if (!blk) {
1653         error_setg(errp, "Device '%s' not found", backup->device);
1654         return;
1655     }
1656     bs = blk_bs(blk);
1657 
1658     blk = blk_by_name(backup->target);
1659     if (!blk) {
1660         error_setg(errp, "Device '%s' not found", backup->target);
1661         return;
1662     }
1663     target = blk_bs(blk);
1664 
1665     /* AioContext is released in .clean() */
1666     state->aio_context = bdrv_get_aio_context(bs);
1667     if (state->aio_context != bdrv_get_aio_context(target)) {
1668         state->aio_context = NULL;
1669         error_setg(errp, "Backup between two IO threads is not implemented");
1670         return;
1671     }
1672     aio_context_acquire(state->aio_context);
1673 
1674     qmp_blockdev_backup(backup->device, backup->target,
1675                         backup->sync,
1676                         backup->has_speed, backup->speed,
1677                         backup->has_on_source_error, backup->on_source_error,
1678                         backup->has_on_target_error, backup->on_target_error,
1679                         &local_err);
1680     if (local_err) {
1681         error_propagate(errp, local_err);
1682         return;
1683     }
1684 
1685     state->bs = bs;
1686     state->job = state->bs->job;
1687 }
1688 
1689 static void blockdev_backup_abort(BlkTransactionState *common)
1690 {
1691     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1692     BlockDriverState *bs = state->bs;
1693 
1694     /* Only cancel if it's the job we started */
1695     if (bs && bs->job && bs->job == state->job) {
1696         block_job_cancel_sync(bs->job);
1697     }
1698 }
1699 
1700 static void blockdev_backup_clean(BlkTransactionState *common)
1701 {
1702     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1703 
1704     if (state->aio_context) {
1705         aio_context_release(state->aio_context);
1706     }
1707 }
1708 
1709 static void abort_prepare(BlkTransactionState *common, Error **errp)
1710 {
1711     error_setg(errp, "Transaction aborted using Abort action");
1712 }
1713 
1714 static void abort_commit(BlkTransactionState *common)
1715 {
1716     g_assert_not_reached(); /* this action never succeeds */
1717 }
1718 
1719 static const BdrvActionOps actions[] = {
1720     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
1721         .instance_size = sizeof(ExternalSnapshotState),
1722         .prepare  = external_snapshot_prepare,
1723         .commit   = external_snapshot_commit,
1724         .abort = external_snapshot_abort,
1725     },
1726     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
1727         .instance_size = sizeof(DriveBackupState),
1728         .prepare = drive_backup_prepare,
1729         .abort = drive_backup_abort,
1730         .clean = drive_backup_clean,
1731     },
1732     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
1733         .instance_size = sizeof(BlockdevBackupState),
1734         .prepare = blockdev_backup_prepare,
1735         .abort = blockdev_backup_abort,
1736         .clean = blockdev_backup_clean,
1737     },
1738     [TRANSACTION_ACTION_KIND_ABORT] = {
1739         .instance_size = sizeof(BlkTransactionState),
1740         .prepare = abort_prepare,
1741         .commit = abort_commit,
1742     },
1743     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
1744         .instance_size = sizeof(InternalSnapshotState),
1745         .prepare  = internal_snapshot_prepare,
1746         .abort = internal_snapshot_abort,
1747         .clean = internal_snapshot_clean,
1748     },
1749 };
1750 
1751 /*
1752  * 'Atomic' group operations.  The operations are performed as a set, and if
1753  * any fail then we roll back all operations in the group.
1754  */
1755 void qmp_transaction(TransactionActionList *dev_list, Error **errp)
1756 {
1757     TransactionActionList *dev_entry = dev_list;
1758     BlkTransactionState *state, *next;
1759     Error *local_err = NULL;
1760 
1761     QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionState) snap_bdrv_states;
1762     QSIMPLEQ_INIT(&snap_bdrv_states);
1763 
1764     /* drain all i/o before any operations */
1765     bdrv_drain_all();
1766 
1767     /* We don't do anything in this loop that commits us to the operations */
1768     while (NULL != dev_entry) {
1769         TransactionAction *dev_info = NULL;
1770         const BdrvActionOps *ops;
1771 
1772         dev_info = dev_entry->value;
1773         dev_entry = dev_entry->next;
1774 
1775         assert(dev_info->kind < ARRAY_SIZE(actions));
1776 
1777         ops = &actions[dev_info->kind];
1778         assert(ops->instance_size > 0);
1779 
1780         state = g_malloc0(ops->instance_size);
1781         state->ops = ops;
1782         state->action = dev_info;
1783         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
1784 
1785         state->ops->prepare(state, &local_err);
1786         if (local_err) {
1787             error_propagate(errp, local_err);
1788             goto delete_and_fail;
1789         }
1790     }
1791 
1792     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1793         if (state->ops->commit) {
1794             state->ops->commit(state);
1795         }
1796     }
1797 
1798     /* success */
1799     goto exit;
1800 
1801 delete_and_fail:
1802     /* failure, and it is all-or-none; roll back all operations */
1803     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1804         if (state->ops->abort) {
1805             state->ops->abort(state);
1806         }
1807     }
1808 exit:
1809     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
1810         if (state->ops->clean) {
1811             state->ops->clean(state);
1812         }
1813         g_free(state);
1814     }
1815 }
1816 
1817 
1818 static void eject_device(BlockBackend *blk, int force, Error **errp)
1819 {
1820     BlockDriverState *bs = blk_bs(blk);
1821     AioContext *aio_context;
1822 
1823     aio_context = bdrv_get_aio_context(bs);
1824     aio_context_acquire(aio_context);
1825 
1826     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
1827         goto out;
1828     }
1829     if (!blk_dev_has_removable_media(blk)) {
1830         error_setg(errp, "Device '%s' is not removable",
1831                    bdrv_get_device_name(bs));
1832         goto out;
1833     }
1834 
1835     if (blk_dev_is_medium_locked(blk) && !blk_dev_is_tray_open(blk)) {
1836         blk_dev_eject_request(blk, force);
1837         if (!force) {
1838             error_setg(errp, "Device '%s' is locked",
1839                        bdrv_get_device_name(bs));
1840             goto out;
1841         }
1842     }
1843 
1844     bdrv_close(bs);
1845 
1846 out:
1847     aio_context_release(aio_context);
1848 }
1849 
1850 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
1851 {
1852     BlockBackend *blk;
1853 
1854     blk = blk_by_name(device);
1855     if (!blk) {
1856         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1857                   "Device '%s' not found", device);
1858         return;
1859     }
1860 
1861     eject_device(blk, force, errp);
1862 }
1863 
1864 void qmp_block_passwd(bool has_device, const char *device,
1865                       bool has_node_name, const char *node_name,
1866                       const char *password, Error **errp)
1867 {
1868     Error *local_err = NULL;
1869     BlockDriverState *bs;
1870     AioContext *aio_context;
1871 
1872     bs = bdrv_lookup_bs(has_device ? device : NULL,
1873                         has_node_name ? node_name : NULL,
1874                         &local_err);
1875     if (local_err) {
1876         error_propagate(errp, local_err);
1877         return;
1878     }
1879 
1880     aio_context = bdrv_get_aio_context(bs);
1881     aio_context_acquire(aio_context);
1882 
1883     bdrv_add_key(bs, password, errp);
1884 
1885     aio_context_release(aio_context);
1886 }
1887 
1888 /* Assumes AioContext is held */
1889 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename,
1890                                     int bdrv_flags, BlockDriver *drv,
1891                                     const char *password, Error **errp)
1892 {
1893     Error *local_err = NULL;
1894     int ret;
1895 
1896     ret = bdrv_open(&bs, filename, NULL, NULL, bdrv_flags, drv, &local_err);
1897     if (ret < 0) {
1898         error_propagate(errp, local_err);
1899         return;
1900     }
1901 
1902     bdrv_add_key(bs, password, errp);
1903 }
1904 
1905 void qmp_change_blockdev(const char *device, const char *filename,
1906                          const char *format, Error **errp)
1907 {
1908     BlockBackend *blk;
1909     BlockDriverState *bs;
1910     AioContext *aio_context;
1911     BlockDriver *drv = NULL;
1912     int bdrv_flags;
1913     Error *err = NULL;
1914 
1915     blk = blk_by_name(device);
1916     if (!blk) {
1917         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1918                   "Device '%s' not found", device);
1919         return;
1920     }
1921     bs = blk_bs(blk);
1922 
1923     aio_context = bdrv_get_aio_context(bs);
1924     aio_context_acquire(aio_context);
1925 
1926     if (format) {
1927         drv = bdrv_find_whitelisted_format(format, bs->read_only);
1928         if (!drv) {
1929             error_setg(errp, QERR_INVALID_BLOCK_FORMAT, format);
1930             goto out;
1931         }
1932     }
1933 
1934     eject_device(blk, 0, &err);
1935     if (err) {
1936         error_propagate(errp, err);
1937         goto out;
1938     }
1939 
1940     bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
1941     bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
1942 
1943     qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp);
1944 
1945 out:
1946     aio_context_release(aio_context);
1947 }
1948 
1949 /* throttling disk I/O limits */
1950 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
1951                                int64_t bps_wr,
1952                                int64_t iops,
1953                                int64_t iops_rd,
1954                                int64_t iops_wr,
1955                                bool has_bps_max,
1956                                int64_t bps_max,
1957                                bool has_bps_rd_max,
1958                                int64_t bps_rd_max,
1959                                bool has_bps_wr_max,
1960                                int64_t bps_wr_max,
1961                                bool has_iops_max,
1962                                int64_t iops_max,
1963                                bool has_iops_rd_max,
1964                                int64_t iops_rd_max,
1965                                bool has_iops_wr_max,
1966                                int64_t iops_wr_max,
1967                                bool has_iops_size,
1968                                int64_t iops_size,
1969                                bool has_group,
1970                                const char *group, Error **errp)
1971 {
1972     ThrottleConfig cfg;
1973     BlockDriverState *bs;
1974     BlockBackend *blk;
1975     AioContext *aio_context;
1976 
1977     blk = blk_by_name(device);
1978     if (!blk) {
1979         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
1980                   "Device '%s' not found", device);
1981         return;
1982     }
1983     bs = blk_bs(blk);
1984 
1985     memset(&cfg, 0, sizeof(cfg));
1986     cfg.buckets[THROTTLE_BPS_TOTAL].avg = bps;
1987     cfg.buckets[THROTTLE_BPS_READ].avg  = bps_rd;
1988     cfg.buckets[THROTTLE_BPS_WRITE].avg = bps_wr;
1989 
1990     cfg.buckets[THROTTLE_OPS_TOTAL].avg = iops;
1991     cfg.buckets[THROTTLE_OPS_READ].avg  = iops_rd;
1992     cfg.buckets[THROTTLE_OPS_WRITE].avg = iops_wr;
1993 
1994     if (has_bps_max) {
1995         cfg.buckets[THROTTLE_BPS_TOTAL].max = bps_max;
1996     }
1997     if (has_bps_rd_max) {
1998         cfg.buckets[THROTTLE_BPS_READ].max = bps_rd_max;
1999     }
2000     if (has_bps_wr_max) {
2001         cfg.buckets[THROTTLE_BPS_WRITE].max = bps_wr_max;
2002     }
2003     if (has_iops_max) {
2004         cfg.buckets[THROTTLE_OPS_TOTAL].max = iops_max;
2005     }
2006     if (has_iops_rd_max) {
2007         cfg.buckets[THROTTLE_OPS_READ].max = iops_rd_max;
2008     }
2009     if (has_iops_wr_max) {
2010         cfg.buckets[THROTTLE_OPS_WRITE].max = iops_wr_max;
2011     }
2012 
2013     if (has_iops_size) {
2014         cfg.op_size = iops_size;
2015     }
2016 
2017     if (!check_throttle_config(&cfg, errp)) {
2018         return;
2019     }
2020 
2021     aio_context = bdrv_get_aio_context(bs);
2022     aio_context_acquire(aio_context);
2023 
2024     if (throttle_enabled(&cfg)) {
2025         /* Enable I/O limits if they're not enabled yet, otherwise
2026          * just update the throttling group. */
2027         if (!bs->io_limits_enabled) {
2028             bdrv_io_limits_enable(bs, has_group ? group : device);
2029         } else if (has_group) {
2030             bdrv_io_limits_update_group(bs, group);
2031         }
2032         /* Set the new throttling configuration */
2033         bdrv_set_io_limits(bs, &cfg);
2034     } else if (bs->io_limits_enabled) {
2035         /* If all throttling settings are set to 0, disable I/O limits */
2036         bdrv_io_limits_disable(bs);
2037     }
2038 
2039     aio_context_release(aio_context);
2040 }
2041 
2042 void qmp_block_dirty_bitmap_add(const char *node, const char *name,
2043                                 bool has_granularity, uint32_t granularity,
2044                                 Error **errp)
2045 {
2046     AioContext *aio_context;
2047     BlockDriverState *bs;
2048 
2049     if (!name || name[0] == '\0') {
2050         error_setg(errp, "Bitmap name cannot be empty");
2051         return;
2052     }
2053 
2054     bs = bdrv_lookup_bs(node, node, errp);
2055     if (!bs) {
2056         return;
2057     }
2058 
2059     aio_context = bdrv_get_aio_context(bs);
2060     aio_context_acquire(aio_context);
2061 
2062     if (has_granularity) {
2063         if (granularity < 512 || !is_power_of_2(granularity)) {
2064             error_setg(errp, "Granularity must be power of 2 "
2065                              "and at least 512");
2066             goto out;
2067         }
2068     } else {
2069         /* Default to cluster size, if available: */
2070         granularity = bdrv_get_default_bitmap_granularity(bs);
2071     }
2072 
2073     bdrv_create_dirty_bitmap(bs, granularity, name, errp);
2074 
2075  out:
2076     aio_context_release(aio_context);
2077 }
2078 
2079 void qmp_block_dirty_bitmap_remove(const char *node, const char *name,
2080                                    Error **errp)
2081 {
2082     AioContext *aio_context;
2083     BlockDriverState *bs;
2084     BdrvDirtyBitmap *bitmap;
2085 
2086     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2087     if (!bitmap || !bs) {
2088         return;
2089     }
2090 
2091     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2092         error_setg(errp,
2093                    "Bitmap '%s' is currently frozen and cannot be removed",
2094                    name);
2095         goto out;
2096     }
2097     bdrv_dirty_bitmap_make_anon(bitmap);
2098     bdrv_release_dirty_bitmap(bs, bitmap);
2099 
2100  out:
2101     aio_context_release(aio_context);
2102 }
2103 
2104 /**
2105  * Completely clear a bitmap, for the purposes of synchronizing a bitmap
2106  * immediately after a full backup operation.
2107  */
2108 void qmp_block_dirty_bitmap_clear(const char *node, const char *name,
2109                                   Error **errp)
2110 {
2111     AioContext *aio_context;
2112     BdrvDirtyBitmap *bitmap;
2113     BlockDriverState *bs;
2114 
2115     bitmap = block_dirty_bitmap_lookup(node, name, &bs, &aio_context, errp);
2116     if (!bitmap || !bs) {
2117         return;
2118     }
2119 
2120     if (bdrv_dirty_bitmap_frozen(bitmap)) {
2121         error_setg(errp,
2122                    "Bitmap '%s' is currently frozen and cannot be modified",
2123                    name);
2124         goto out;
2125     } else if (!bdrv_dirty_bitmap_enabled(bitmap)) {
2126         error_setg(errp,
2127                    "Bitmap '%s' is currently disabled and cannot be cleared",
2128                    name);
2129         goto out;
2130     }
2131 
2132     bdrv_clear_dirty_bitmap(bitmap);
2133 
2134  out:
2135     aio_context_release(aio_context);
2136 }
2137 
2138 void hmp_drive_del(Monitor *mon, const QDict *qdict)
2139 {
2140     const char *id = qdict_get_str(qdict, "id");
2141     BlockBackend *blk;
2142     BlockDriverState *bs;
2143     AioContext *aio_context;
2144     Error *local_err = NULL;
2145 
2146     blk = blk_by_name(id);
2147     if (!blk) {
2148         error_report("Device '%s' not found", id);
2149         return;
2150     }
2151     bs = blk_bs(blk);
2152 
2153     if (!blk_legacy_dinfo(blk)) {
2154         error_report("Deleting device added with blockdev-add"
2155                      " is not supported");
2156         return;
2157     }
2158 
2159     aio_context = bdrv_get_aio_context(bs);
2160     aio_context_acquire(aio_context);
2161 
2162     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
2163         error_report_err(local_err);
2164         aio_context_release(aio_context);
2165         return;
2166     }
2167 
2168     /* quiesce block driver; prevent further io */
2169     bdrv_drain_all();
2170     bdrv_flush(bs);
2171     bdrv_close(bs);
2172 
2173     /* if we have a device attached to this BlockDriverState
2174      * then we need to make the drive anonymous until the device
2175      * can be removed.  If this is a drive with no device backing
2176      * then we can just get rid of the block driver state right here.
2177      */
2178     if (blk_get_attached_dev(blk)) {
2179         blk_hide_on_behalf_of_hmp_drive_del(blk);
2180         /* Further I/O must not pause the guest */
2181         bdrv_set_on_error(bs, BLOCKDEV_ON_ERROR_REPORT,
2182                           BLOCKDEV_ON_ERROR_REPORT);
2183     } else {
2184         blk_unref(blk);
2185     }
2186 
2187     aio_context_release(aio_context);
2188 }
2189 
2190 void qmp_block_resize(bool has_device, const char *device,
2191                       bool has_node_name, const char *node_name,
2192                       int64_t size, Error **errp)
2193 {
2194     Error *local_err = NULL;
2195     BlockDriverState *bs;
2196     AioContext *aio_context;
2197     int ret;
2198 
2199     bs = bdrv_lookup_bs(has_device ? device : NULL,
2200                         has_node_name ? node_name : NULL,
2201                         &local_err);
2202     if (local_err) {
2203         error_propagate(errp, local_err);
2204         return;
2205     }
2206 
2207     aio_context = bdrv_get_aio_context(bs);
2208     aio_context_acquire(aio_context);
2209 
2210     if (!bdrv_is_first_non_filter(bs)) {
2211         error_setg(errp, QERR_FEATURE_DISABLED, "resize");
2212         goto out;
2213     }
2214 
2215     if (size < 0) {
2216         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2217         goto out;
2218     }
2219 
2220     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2221         error_setg(errp, QERR_DEVICE_IN_USE, device);
2222         goto out;
2223     }
2224 
2225     /* complete all in-flight operations before resizing the device */
2226     bdrv_drain_all();
2227 
2228     ret = bdrv_truncate(bs, size);
2229     switch (ret) {
2230     case 0:
2231         break;
2232     case -ENOMEDIUM:
2233         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2234         break;
2235     case -ENOTSUP:
2236         error_setg(errp, QERR_UNSUPPORTED);
2237         break;
2238     case -EACCES:
2239         error_setg(errp, "Device '%s' is read only", device);
2240         break;
2241     case -EBUSY:
2242         error_setg(errp, QERR_DEVICE_IN_USE, device);
2243         break;
2244     default:
2245         error_setg_errno(errp, -ret, "Could not resize");
2246         break;
2247     }
2248 
2249 out:
2250     aio_context_release(aio_context);
2251 }
2252 
2253 static void block_job_cb(void *opaque, int ret)
2254 {
2255     /* Note that this function may be executed from another AioContext besides
2256      * the QEMU main loop.  If you need to access anything that assumes the
2257      * QEMU global mutex, use a BH or introduce a mutex.
2258      */
2259 
2260     BlockDriverState *bs = opaque;
2261     const char *msg = NULL;
2262 
2263     trace_block_job_cb(bs, bs->job, ret);
2264 
2265     assert(bs->job);
2266 
2267     if (ret < 0) {
2268         msg = strerror(-ret);
2269     }
2270 
2271     if (block_job_is_cancelled(bs->job)) {
2272         block_job_event_cancelled(bs->job);
2273     } else {
2274         block_job_event_completed(bs->job, msg);
2275     }
2276 
2277     bdrv_put_ref_bh_schedule(bs);
2278 }
2279 
2280 void qmp_block_stream(const char *device,
2281                       bool has_base, const char *base,
2282                       bool has_backing_file, const char *backing_file,
2283                       bool has_speed, int64_t speed,
2284                       bool has_on_error, BlockdevOnError on_error,
2285                       Error **errp)
2286 {
2287     BlockBackend *blk;
2288     BlockDriverState *bs;
2289     BlockDriverState *base_bs = NULL;
2290     AioContext *aio_context;
2291     Error *local_err = NULL;
2292     const char *base_name = NULL;
2293 
2294     if (!has_on_error) {
2295         on_error = BLOCKDEV_ON_ERROR_REPORT;
2296     }
2297 
2298     blk = blk_by_name(device);
2299     if (!blk) {
2300         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2301                   "Device '%s' not found", device);
2302         return;
2303     }
2304     bs = blk_bs(blk);
2305 
2306     aio_context = bdrv_get_aio_context(bs);
2307     aio_context_acquire(aio_context);
2308 
2309     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_STREAM, errp)) {
2310         goto out;
2311     }
2312 
2313     if (has_base) {
2314         base_bs = bdrv_find_backing_image(bs, base);
2315         if (base_bs == NULL) {
2316             error_setg(errp, QERR_BASE_NOT_FOUND, base);
2317             goto out;
2318         }
2319         assert(bdrv_get_aio_context(base_bs) == aio_context);
2320         base_name = base;
2321     }
2322 
2323     /* if we are streaming the entire chain, the result will have no backing
2324      * file, and specifying one is therefore an error */
2325     if (base_bs == NULL && has_backing_file) {
2326         error_setg(errp, "backing file specified, but streaming the "
2327                          "entire chain");
2328         goto out;
2329     }
2330 
2331     /* backing_file string overrides base bs filename */
2332     base_name = has_backing_file ? backing_file : base_name;
2333 
2334     stream_start(bs, base_bs, base_name, has_speed ? speed : 0,
2335                  on_error, block_job_cb, bs, &local_err);
2336     if (local_err) {
2337         error_propagate(errp, local_err);
2338         goto out;
2339     }
2340 
2341     trace_qmp_block_stream(bs, bs->job);
2342 
2343 out:
2344     aio_context_release(aio_context);
2345 }
2346 
2347 void qmp_block_commit(const char *device,
2348                       bool has_base, const char *base,
2349                       bool has_top, const char *top,
2350                       bool has_backing_file, const char *backing_file,
2351                       bool has_speed, int64_t speed,
2352                       Error **errp)
2353 {
2354     BlockBackend *blk;
2355     BlockDriverState *bs;
2356     BlockDriverState *base_bs, *top_bs;
2357     AioContext *aio_context;
2358     Error *local_err = NULL;
2359     /* This will be part of the QMP command, if/when the
2360      * BlockdevOnError change for blkmirror makes it in
2361      */
2362     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
2363 
2364     if (!has_speed) {
2365         speed = 0;
2366     }
2367 
2368     /* Important Note:
2369      *  libvirt relies on the DeviceNotFound error class in order to probe for
2370      *  live commit feature versions; for this to work, we must make sure to
2371      *  perform the device lookup before any generic errors that may occur in a
2372      *  scenario in which all optional arguments are omitted. */
2373     blk = blk_by_name(device);
2374     if (!blk) {
2375         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2376                   "Device '%s' not found", device);
2377         return;
2378     }
2379     bs = blk_bs(blk);
2380 
2381     aio_context = bdrv_get_aio_context(bs);
2382     aio_context_acquire(aio_context);
2383 
2384     /* drain all i/o before commits */
2385     bdrv_drain_all();
2386 
2387     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2388         goto out;
2389     }
2390 
2391     /* default top_bs is the active layer */
2392     top_bs = bs;
2393 
2394     if (has_top && top) {
2395         if (strcmp(bs->filename, top) != 0) {
2396             top_bs = bdrv_find_backing_image(bs, top);
2397         }
2398     }
2399 
2400     if (top_bs == NULL) {
2401         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2402         goto out;
2403     }
2404 
2405     assert(bdrv_get_aio_context(top_bs) == aio_context);
2406 
2407     if (has_base && base) {
2408         base_bs = bdrv_find_backing_image(top_bs, base);
2409     } else {
2410         base_bs = bdrv_find_base(top_bs);
2411     }
2412 
2413     if (base_bs == NULL) {
2414         error_setg(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
2415         goto out;
2416     }
2417 
2418     assert(bdrv_get_aio_context(base_bs) == aio_context);
2419 
2420     if (bdrv_op_is_blocked(base_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2421         goto out;
2422     }
2423 
2424     /* Do not allow attempts to commit an image into itself */
2425     if (top_bs == base_bs) {
2426         error_setg(errp, "cannot commit an image into itself");
2427         goto out;
2428     }
2429 
2430     if (top_bs == bs) {
2431         if (has_backing_file) {
2432             error_setg(errp, "'backing-file' specified,"
2433                              " but 'top' is the active layer");
2434             goto out;
2435         }
2436         commit_active_start(bs, base_bs, speed, on_error, block_job_cb,
2437                             bs, &local_err);
2438     } else {
2439         commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
2440                      has_backing_file ? backing_file : NULL, &local_err);
2441     }
2442     if (local_err != NULL) {
2443         error_propagate(errp, local_err);
2444         goto out;
2445     }
2446 
2447 out:
2448     aio_context_release(aio_context);
2449 }
2450 
2451 void qmp_drive_backup(const char *device, const char *target,
2452                       bool has_format, const char *format,
2453                       enum MirrorSyncMode sync,
2454                       bool has_mode, enum NewImageMode mode,
2455                       bool has_speed, int64_t speed,
2456                       bool has_bitmap, const char *bitmap,
2457                       bool has_on_source_error, BlockdevOnError on_source_error,
2458                       bool has_on_target_error, BlockdevOnError on_target_error,
2459                       Error **errp)
2460 {
2461     BlockBackend *blk;
2462     BlockDriverState *bs;
2463     BlockDriverState *target_bs;
2464     BlockDriverState *source = NULL;
2465     BdrvDirtyBitmap *bmap = NULL;
2466     AioContext *aio_context;
2467     BlockDriver *drv = NULL;
2468     Error *local_err = NULL;
2469     int flags;
2470     int64_t size;
2471     int ret;
2472 
2473     if (!has_speed) {
2474         speed = 0;
2475     }
2476     if (!has_on_source_error) {
2477         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2478     }
2479     if (!has_on_target_error) {
2480         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2481     }
2482     if (!has_mode) {
2483         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2484     }
2485 
2486     blk = blk_by_name(device);
2487     if (!blk) {
2488         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2489                   "Device '%s' not found", device);
2490         return;
2491     }
2492     bs = blk_bs(blk);
2493 
2494     aio_context = bdrv_get_aio_context(bs);
2495     aio_context_acquire(aio_context);
2496 
2497     /* Although backup_run has this check too, we need to use bs->drv below, so
2498      * do an early check redundantly. */
2499     if (!bdrv_is_inserted(bs)) {
2500         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2501         goto out;
2502     }
2503 
2504     if (!has_format) {
2505         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
2506     }
2507     if (format) {
2508         drv = bdrv_find_format(format);
2509         if (!drv) {
2510             error_setg(errp, QERR_INVALID_BLOCK_FORMAT, format);
2511             goto out;
2512         }
2513     }
2514 
2515     /* Early check to avoid creating target */
2516     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
2517         goto out;
2518     }
2519 
2520     flags = bs->open_flags | BDRV_O_RDWR;
2521 
2522     /* See if we have a backing HD we can use to create our new image
2523      * on top of. */
2524     if (sync == MIRROR_SYNC_MODE_TOP) {
2525         source = bs->backing_hd;
2526         if (!source) {
2527             sync = MIRROR_SYNC_MODE_FULL;
2528         }
2529     }
2530     if (sync == MIRROR_SYNC_MODE_NONE) {
2531         source = bs;
2532     }
2533 
2534     size = bdrv_getlength(bs);
2535     if (size < 0) {
2536         error_setg_errno(errp, -size, "bdrv_getlength failed");
2537         goto out;
2538     }
2539 
2540     if (mode != NEW_IMAGE_MODE_EXISTING) {
2541         assert(format && drv);
2542         if (source) {
2543             bdrv_img_create(target, format, source->filename,
2544                             source->drv->format_name, NULL,
2545                             size, flags, &local_err, false);
2546         } else {
2547             bdrv_img_create(target, format, NULL, NULL, NULL,
2548                             size, flags, &local_err, false);
2549         }
2550     }
2551 
2552     if (local_err) {
2553         error_propagate(errp, local_err);
2554         goto out;
2555     }
2556 
2557     target_bs = NULL;
2558     ret = bdrv_open(&target_bs, target, NULL, NULL, flags, drv, &local_err);
2559     if (ret < 0) {
2560         error_propagate(errp, local_err);
2561         goto out;
2562     }
2563 
2564     bdrv_set_aio_context(target_bs, aio_context);
2565 
2566     if (has_bitmap) {
2567         bmap = bdrv_find_dirty_bitmap(bs, bitmap);
2568         if (!bmap) {
2569             error_setg(errp, "Bitmap '%s' could not be found", bitmap);
2570             goto out;
2571         }
2572     }
2573 
2574     backup_start(bs, target_bs, speed, sync, bmap,
2575                  on_source_error, on_target_error,
2576                  block_job_cb, bs, &local_err);
2577     if (local_err != NULL) {
2578         bdrv_unref(target_bs);
2579         error_propagate(errp, local_err);
2580         goto out;
2581     }
2582 
2583 out:
2584     aio_context_release(aio_context);
2585 }
2586 
2587 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
2588 {
2589     return bdrv_named_nodes_list(errp);
2590 }
2591 
2592 void qmp_blockdev_backup(const char *device, const char *target,
2593                          enum MirrorSyncMode sync,
2594                          bool has_speed, int64_t speed,
2595                          bool has_on_source_error,
2596                          BlockdevOnError on_source_error,
2597                          bool has_on_target_error,
2598                          BlockdevOnError on_target_error,
2599                          Error **errp)
2600 {
2601     BlockBackend *blk;
2602     BlockDriverState *bs;
2603     BlockDriverState *target_bs;
2604     Error *local_err = NULL;
2605     AioContext *aio_context;
2606 
2607     if (!has_speed) {
2608         speed = 0;
2609     }
2610     if (!has_on_source_error) {
2611         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2612     }
2613     if (!has_on_target_error) {
2614         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2615     }
2616 
2617     blk = blk_by_name(device);
2618     if (!blk) {
2619         error_setg(errp, "Device '%s' not found", device);
2620         return;
2621     }
2622     bs = blk_bs(blk);
2623 
2624     aio_context = bdrv_get_aio_context(bs);
2625     aio_context_acquire(aio_context);
2626 
2627     blk = blk_by_name(target);
2628     if (!blk) {
2629         error_setg(errp, "Device '%s' not found", target);
2630         goto out;
2631     }
2632     target_bs = blk_bs(blk);
2633 
2634     bdrv_ref(target_bs);
2635     bdrv_set_aio_context(target_bs, aio_context);
2636     backup_start(bs, target_bs, speed, sync, NULL, on_source_error,
2637                  on_target_error, block_job_cb, bs, &local_err);
2638     if (local_err != NULL) {
2639         bdrv_unref(target_bs);
2640         error_propagate(errp, local_err);
2641     }
2642 out:
2643     aio_context_release(aio_context);
2644 }
2645 
2646 #define DEFAULT_MIRROR_BUF_SIZE   (10 << 20)
2647 
2648 void qmp_drive_mirror(const char *device, const char *target,
2649                       bool has_format, const char *format,
2650                       bool has_node_name, const char *node_name,
2651                       bool has_replaces, const char *replaces,
2652                       enum MirrorSyncMode sync,
2653                       bool has_mode, enum NewImageMode mode,
2654                       bool has_speed, int64_t speed,
2655                       bool has_granularity, uint32_t granularity,
2656                       bool has_buf_size, int64_t buf_size,
2657                       bool has_on_source_error, BlockdevOnError on_source_error,
2658                       bool has_on_target_error, BlockdevOnError on_target_error,
2659                       Error **errp)
2660 {
2661     BlockBackend *blk;
2662     BlockDriverState *bs;
2663     BlockDriverState *source, *target_bs;
2664     AioContext *aio_context;
2665     BlockDriver *drv = NULL;
2666     Error *local_err = NULL;
2667     QDict *options = NULL;
2668     int flags;
2669     int64_t size;
2670     int ret;
2671 
2672     if (!has_speed) {
2673         speed = 0;
2674     }
2675     if (!has_on_source_error) {
2676         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2677     }
2678     if (!has_on_target_error) {
2679         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2680     }
2681     if (!has_mode) {
2682         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2683     }
2684     if (!has_granularity) {
2685         granularity = 0;
2686     }
2687     if (!has_buf_size) {
2688         buf_size = DEFAULT_MIRROR_BUF_SIZE;
2689     }
2690 
2691     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2692         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2693                    "a value in range [512B, 64MB]");
2694         return;
2695     }
2696     if (granularity & (granularity - 1)) {
2697         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2698                    "power of 2");
2699         return;
2700     }
2701 
2702     blk = blk_by_name(device);
2703     if (!blk) {
2704         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2705                   "Device '%s' not found", device);
2706         return;
2707     }
2708     bs = blk_bs(blk);
2709 
2710     aio_context = bdrv_get_aio_context(bs);
2711     aio_context_acquire(aio_context);
2712 
2713     if (!bdrv_is_inserted(bs)) {
2714         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2715         goto out;
2716     }
2717 
2718     if (!has_format) {
2719         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
2720     }
2721     if (format) {
2722         drv = bdrv_find_format(format);
2723         if (!drv) {
2724             error_setg(errp, QERR_INVALID_BLOCK_FORMAT, format);
2725             goto out;
2726         }
2727     }
2728 
2729     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR, errp)) {
2730         goto out;
2731     }
2732 
2733     flags = bs->open_flags | BDRV_O_RDWR;
2734     source = bs->backing_hd;
2735     if (!source && sync == MIRROR_SYNC_MODE_TOP) {
2736         sync = MIRROR_SYNC_MODE_FULL;
2737     }
2738     if (sync == MIRROR_SYNC_MODE_NONE) {
2739         source = bs;
2740     }
2741 
2742     size = bdrv_getlength(bs);
2743     if (size < 0) {
2744         error_setg_errno(errp, -size, "bdrv_getlength failed");
2745         goto out;
2746     }
2747 
2748     if (has_replaces) {
2749         BlockDriverState *to_replace_bs;
2750         AioContext *replace_aio_context;
2751         int64_t replace_size;
2752 
2753         if (!has_node_name) {
2754             error_setg(errp, "a node-name must be provided when replacing a"
2755                              " named node of the graph");
2756             goto out;
2757         }
2758 
2759         to_replace_bs = check_to_replace_node(replaces, &local_err);
2760 
2761         if (!to_replace_bs) {
2762             error_propagate(errp, local_err);
2763             goto out;
2764         }
2765 
2766         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
2767         aio_context_acquire(replace_aio_context);
2768         replace_size = bdrv_getlength(to_replace_bs);
2769         aio_context_release(replace_aio_context);
2770 
2771         if (size != replace_size) {
2772             error_setg(errp, "cannot replace image with a mirror image of "
2773                              "different size");
2774             goto out;
2775         }
2776     }
2777 
2778     if ((sync == MIRROR_SYNC_MODE_FULL || !source)
2779         && mode != NEW_IMAGE_MODE_EXISTING)
2780     {
2781         /* create new image w/o backing file */
2782         assert(format && drv);
2783         bdrv_img_create(target, format,
2784                         NULL, NULL, NULL, size, flags, &local_err, false);
2785     } else {
2786         switch (mode) {
2787         case NEW_IMAGE_MODE_EXISTING:
2788             break;
2789         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
2790             /* create new image with backing file */
2791             bdrv_img_create(target, format,
2792                             source->filename,
2793                             source->drv->format_name,
2794                             NULL, size, flags, &local_err, false);
2795             break;
2796         default:
2797             abort();
2798         }
2799     }
2800 
2801     if (local_err) {
2802         error_propagate(errp, local_err);
2803         goto out;
2804     }
2805 
2806     if (has_node_name) {
2807         options = qdict_new();
2808         qdict_put(options, "node-name", qstring_from_str(node_name));
2809     }
2810 
2811     /* Mirroring takes care of copy-on-write using the source's backing
2812      * file.
2813      */
2814     target_bs = NULL;
2815     ret = bdrv_open(&target_bs, target, NULL, options,
2816                     flags | BDRV_O_NO_BACKING, drv, &local_err);
2817     if (ret < 0) {
2818         error_propagate(errp, local_err);
2819         goto out;
2820     }
2821 
2822     bdrv_set_aio_context(target_bs, aio_context);
2823 
2824     /* pass the node name to replace to mirror start since it's loose coupling
2825      * and will allow to check whether the node still exist at mirror completion
2826      */
2827     mirror_start(bs, target_bs,
2828                  has_replaces ? replaces : NULL,
2829                  speed, granularity, buf_size, sync,
2830                  on_source_error, on_target_error,
2831                  block_job_cb, bs, &local_err);
2832     if (local_err != NULL) {
2833         bdrv_unref(target_bs);
2834         error_propagate(errp, local_err);
2835         goto out;
2836     }
2837 
2838 out:
2839     aio_context_release(aio_context);
2840 }
2841 
2842 /* Get the block job for a given device name and acquire its AioContext */
2843 static BlockJob *find_block_job(const char *device, AioContext **aio_context,
2844                                 Error **errp)
2845 {
2846     BlockBackend *blk;
2847     BlockDriverState *bs;
2848 
2849     blk = blk_by_name(device);
2850     if (!blk) {
2851         goto notfound;
2852     }
2853     bs = blk_bs(blk);
2854 
2855     *aio_context = bdrv_get_aio_context(bs);
2856     aio_context_acquire(*aio_context);
2857 
2858     if (!bs->job) {
2859         aio_context_release(*aio_context);
2860         goto notfound;
2861     }
2862 
2863     return bs->job;
2864 
2865 notfound:
2866     error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
2867               "No active block job on device '%s'", device);
2868     *aio_context = NULL;
2869     return NULL;
2870 }
2871 
2872 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
2873 {
2874     AioContext *aio_context;
2875     BlockJob *job = find_block_job(device, &aio_context, errp);
2876 
2877     if (!job) {
2878         return;
2879     }
2880 
2881     block_job_set_speed(job, speed, errp);
2882     aio_context_release(aio_context);
2883 }
2884 
2885 void qmp_block_job_cancel(const char *device,
2886                           bool has_force, bool force, Error **errp)
2887 {
2888     AioContext *aio_context;
2889     BlockJob *job = find_block_job(device, &aio_context, errp);
2890 
2891     if (!job) {
2892         return;
2893     }
2894 
2895     if (!has_force) {
2896         force = false;
2897     }
2898 
2899     if (job->user_paused && !force) {
2900         error_setg(errp, "The block job for device '%s' is currently paused",
2901                    device);
2902         goto out;
2903     }
2904 
2905     trace_qmp_block_job_cancel(job);
2906     block_job_cancel(job);
2907 out:
2908     aio_context_release(aio_context);
2909 }
2910 
2911 void qmp_block_job_pause(const char *device, Error **errp)
2912 {
2913     AioContext *aio_context;
2914     BlockJob *job = find_block_job(device, &aio_context, errp);
2915 
2916     if (!job || job->user_paused) {
2917         return;
2918     }
2919 
2920     job->user_paused = true;
2921     trace_qmp_block_job_pause(job);
2922     block_job_pause(job);
2923     aio_context_release(aio_context);
2924 }
2925 
2926 void qmp_block_job_resume(const char *device, Error **errp)
2927 {
2928     AioContext *aio_context;
2929     BlockJob *job = find_block_job(device, &aio_context, errp);
2930 
2931     if (!job || !job->user_paused) {
2932         return;
2933     }
2934 
2935     job->user_paused = false;
2936     trace_qmp_block_job_resume(job);
2937     block_job_resume(job);
2938     aio_context_release(aio_context);
2939 }
2940 
2941 void qmp_block_job_complete(const char *device, Error **errp)
2942 {
2943     AioContext *aio_context;
2944     BlockJob *job = find_block_job(device, &aio_context, errp);
2945 
2946     if (!job) {
2947         return;
2948     }
2949 
2950     trace_qmp_block_job_complete(job);
2951     block_job_complete(job, errp);
2952     aio_context_release(aio_context);
2953 }
2954 
2955 void qmp_change_backing_file(const char *device,
2956                              const char *image_node_name,
2957                              const char *backing_file,
2958                              Error **errp)
2959 {
2960     BlockBackend *blk;
2961     BlockDriverState *bs = NULL;
2962     AioContext *aio_context;
2963     BlockDriverState *image_bs = NULL;
2964     Error *local_err = NULL;
2965     bool ro;
2966     int open_flags;
2967     int ret;
2968 
2969     blk = blk_by_name(device);
2970     if (!blk) {
2971         error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2972                   "Device '%s' not found", device);
2973         return;
2974     }
2975     bs = blk_bs(blk);
2976 
2977     aio_context = bdrv_get_aio_context(bs);
2978     aio_context_acquire(aio_context);
2979 
2980     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
2981     if (local_err) {
2982         error_propagate(errp, local_err);
2983         goto out;
2984     }
2985 
2986     if (!image_bs) {
2987         error_setg(errp, "image file not found");
2988         goto out;
2989     }
2990 
2991     if (bdrv_find_base(image_bs) == image_bs) {
2992         error_setg(errp, "not allowing backing file change on an image "
2993                          "without a backing file");
2994         goto out;
2995     }
2996 
2997     /* even though we are not necessarily operating on bs, we need it to
2998      * determine if block ops are currently prohibited on the chain */
2999     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3000         goto out;
3001     }
3002 
3003     /* final sanity check */
3004     if (!bdrv_chain_contains(bs, image_bs)) {
3005         error_setg(errp, "'%s' and image file are not in the same chain",
3006                    device);
3007         goto out;
3008     }
3009 
3010     /* if not r/w, reopen to make r/w */
3011     open_flags = image_bs->open_flags;
3012     ro = bdrv_is_read_only(image_bs);
3013 
3014     if (ro) {
3015         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
3016         if (local_err) {
3017             error_propagate(errp, local_err);
3018             goto out;
3019         }
3020     }
3021 
3022     ret = bdrv_change_backing_file(image_bs, backing_file,
3023                                image_bs->drv ? image_bs->drv->format_name : "");
3024 
3025     if (ret < 0) {
3026         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3027                          backing_file);
3028         /* don't exit here, so we can try to restore open flags if
3029          * appropriate */
3030     }
3031 
3032     if (ro) {
3033         bdrv_reopen(image_bs, open_flags, &local_err);
3034         if (local_err) {
3035             error_propagate(errp, local_err); /* will preserve prior errp */
3036         }
3037     }
3038 
3039 out:
3040     aio_context_release(aio_context);
3041 }
3042 
3043 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3044 {
3045     QmpOutputVisitor *ov = qmp_output_visitor_new();
3046     BlockBackend *blk;
3047     QObject *obj;
3048     QDict *qdict;
3049     Error *local_err = NULL;
3050 
3051     /* Require an ID in the top level */
3052     if (!options->has_id) {
3053         error_setg(errp, "Block device needs an ID");
3054         goto fail;
3055     }
3056 
3057     /* TODO Sort it out in raw-posix and drive_new(): Reject aio=native with
3058      * cache.direct=false instead of silently switching to aio=threads, except
3059      * when called from drive_new().
3060      *
3061      * For now, simply forbidding the combination for all drivers will do. */
3062     if (options->has_aio && options->aio == BLOCKDEV_AIO_OPTIONS_NATIVE) {
3063         bool direct = options->has_cache &&
3064                       options->cache->has_direct &&
3065                       options->cache->direct;
3066         if (!direct) {
3067             error_setg(errp, "aio=native requires cache.direct=true");
3068             goto fail;
3069         }
3070     }
3071 
3072     visit_type_BlockdevOptions(qmp_output_get_visitor(ov),
3073                                &options, NULL, &local_err);
3074     if (local_err) {
3075         error_propagate(errp, local_err);
3076         goto fail;
3077     }
3078 
3079     obj = qmp_output_get_qobject(ov);
3080     qdict = qobject_to_qdict(obj);
3081 
3082     qdict_flatten(qdict);
3083 
3084     blk = blockdev_init(NULL, qdict, &local_err);
3085     if (local_err) {
3086         error_propagate(errp, local_err);
3087         goto fail;
3088     }
3089 
3090     if (bdrv_key_required(blk_bs(blk))) {
3091         blk_unref(blk);
3092         error_setg(errp, "blockdev-add doesn't support encrypted devices");
3093         goto fail;
3094     }
3095 
3096 fail:
3097     qmp_output_visitor_cleanup(ov);
3098 }
3099 
3100 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3101 {
3102     BlockJobInfoList *head = NULL, **p_next = &head;
3103     BlockDriverState *bs;
3104 
3105     for (bs = bdrv_next(NULL); bs; bs = bdrv_next(bs)) {
3106         AioContext *aio_context = bdrv_get_aio_context(bs);
3107 
3108         aio_context_acquire(aio_context);
3109 
3110         if (bs->job) {
3111             BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
3112             elem->value = block_job_query(bs->job);
3113             *p_next = elem;
3114             p_next = &elem->next;
3115         }
3116 
3117         aio_context_release(aio_context);
3118     }
3119 
3120     return head;
3121 }
3122 
3123 QemuOptsList qemu_common_drive_opts = {
3124     .name = "drive",
3125     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3126     .desc = {
3127         {
3128             .name = "snapshot",
3129             .type = QEMU_OPT_BOOL,
3130             .help = "enable/disable snapshot mode",
3131         },{
3132             .name = "discard",
3133             .type = QEMU_OPT_STRING,
3134             .help = "discard operation (ignore/off, unmap/on)",
3135         },{
3136             .name = BDRV_OPT_CACHE_WB,
3137             .type = QEMU_OPT_BOOL,
3138             .help = "enables writeback mode for any caches",
3139         },{
3140             .name = BDRV_OPT_CACHE_DIRECT,
3141             .type = QEMU_OPT_BOOL,
3142             .help = "enables use of O_DIRECT (bypass the host page cache)",
3143         },{
3144             .name = BDRV_OPT_CACHE_NO_FLUSH,
3145             .type = QEMU_OPT_BOOL,
3146             .help = "ignore any flush requests for the device",
3147         },{
3148             .name = "aio",
3149             .type = QEMU_OPT_STRING,
3150             .help = "host AIO implementation (threads, native)",
3151         },{
3152             .name = "format",
3153             .type = QEMU_OPT_STRING,
3154             .help = "disk format (raw, qcow2, ...)",
3155         },{
3156             .name = "rerror",
3157             .type = QEMU_OPT_STRING,
3158             .help = "read error action",
3159         },{
3160             .name = "werror",
3161             .type = QEMU_OPT_STRING,
3162             .help = "write error action",
3163         },{
3164             .name = "read-only",
3165             .type = QEMU_OPT_BOOL,
3166             .help = "open drive file as read-only",
3167         },{
3168             .name = "throttling.iops-total",
3169             .type = QEMU_OPT_NUMBER,
3170             .help = "limit total I/O operations per second",
3171         },{
3172             .name = "throttling.iops-read",
3173             .type = QEMU_OPT_NUMBER,
3174             .help = "limit read operations per second",
3175         },{
3176             .name = "throttling.iops-write",
3177             .type = QEMU_OPT_NUMBER,
3178             .help = "limit write operations per second",
3179         },{
3180             .name = "throttling.bps-total",
3181             .type = QEMU_OPT_NUMBER,
3182             .help = "limit total bytes per second",
3183         },{
3184             .name = "throttling.bps-read",
3185             .type = QEMU_OPT_NUMBER,
3186             .help = "limit read bytes per second",
3187         },{
3188             .name = "throttling.bps-write",
3189             .type = QEMU_OPT_NUMBER,
3190             .help = "limit write bytes per second",
3191         },{
3192             .name = "throttling.iops-total-max",
3193             .type = QEMU_OPT_NUMBER,
3194             .help = "I/O operations burst",
3195         },{
3196             .name = "throttling.iops-read-max",
3197             .type = QEMU_OPT_NUMBER,
3198             .help = "I/O operations read burst",
3199         },{
3200             .name = "throttling.iops-write-max",
3201             .type = QEMU_OPT_NUMBER,
3202             .help = "I/O operations write burst",
3203         },{
3204             .name = "throttling.bps-total-max",
3205             .type = QEMU_OPT_NUMBER,
3206             .help = "total bytes burst",
3207         },{
3208             .name = "throttling.bps-read-max",
3209             .type = QEMU_OPT_NUMBER,
3210             .help = "total bytes read burst",
3211         },{
3212             .name = "throttling.bps-write-max",
3213             .type = QEMU_OPT_NUMBER,
3214             .help = "total bytes write burst",
3215         },{
3216             .name = "throttling.iops-size",
3217             .type = QEMU_OPT_NUMBER,
3218             .help = "when limiting by iops max size of an I/O in bytes",
3219         },{
3220             .name = "throttling.group",
3221             .type = QEMU_OPT_STRING,
3222             .help = "name of the block throttling group",
3223         },{
3224             .name = "copy-on-read",
3225             .type = QEMU_OPT_BOOL,
3226             .help = "copy read data from backing file into image file",
3227         },{
3228             .name = "detect-zeroes",
3229             .type = QEMU_OPT_STRING,
3230             .help = "try to optimize zero writes (off, on, unmap)",
3231         },
3232         { /* end of list */ }
3233     },
3234 };
3235 
3236 QemuOptsList qemu_drive_opts = {
3237     .name = "drive",
3238     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3239     .desc = {
3240         /*
3241          * no elements => accept any params
3242          * validation will happen later
3243          */
3244         { /* end of list */ }
3245     },
3246 };
3247