xref: /openbmc/qemu/blockdev.c (revision 0db6e54a)
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 "monitor/monitor.h"
38 #include "qemu/option.h"
39 #include "qemu/config-file.h"
40 #include "qapi/qmp/types.h"
41 #include "qapi-visit.h"
42 #include "qapi/qmp-output-visitor.h"
43 #include "qapi/util.h"
44 #include "sysemu/sysemu.h"
45 #include "block/block_int.h"
46 #include "qmp-commands.h"
47 #include "trace.h"
48 #include "sysemu/arch_init.h"
49 
50 static const char *const if_name[IF_COUNT] = {
51     [IF_NONE] = "none",
52     [IF_IDE] = "ide",
53     [IF_SCSI] = "scsi",
54     [IF_FLOPPY] = "floppy",
55     [IF_PFLASH] = "pflash",
56     [IF_MTD] = "mtd",
57     [IF_SD] = "sd",
58     [IF_VIRTIO] = "virtio",
59     [IF_XEN] = "xen",
60 };
61 
62 static int if_max_devs[IF_COUNT] = {
63     /*
64      * Do not change these numbers!  They govern how drive option
65      * index maps to unit and bus.  That mapping is ABI.
66      *
67      * All controllers used to imlement if=T drives need to support
68      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
69      * Otherwise, some index values map to "impossible" bus, unit
70      * values.
71      *
72      * For instance, if you change [IF_SCSI] to 255, -drive
73      * if=scsi,index=12 no longer means bus=1,unit=5, but
74      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
75      * the drive can't be set up.  Regression.
76      */
77     [IF_IDE] = 2,
78     [IF_SCSI] = 7,
79 };
80 
81 /**
82  * Boards may call this to offer board-by-board overrides
83  * of the default, global values.
84  */
85 void override_max_devs(BlockInterfaceType type, int max_devs)
86 {
87     BlockBackend *blk;
88     DriveInfo *dinfo;
89 
90     if (max_devs <= 0) {
91         return;
92     }
93 
94     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
95         dinfo = blk_legacy_dinfo(blk);
96         if (dinfo->type == type) {
97             fprintf(stderr, "Cannot override units-per-bus property of"
98                     " the %s interface, because a drive of that type has"
99                     " already been added.\n", if_name[type]);
100             g_assert_not_reached();
101         }
102     }
103 
104     if_max_devs[type] = max_devs;
105 }
106 
107 /*
108  * We automatically delete the drive when a device using it gets
109  * unplugged.  Questionable feature, but we can't just drop it.
110  * Device models call blockdev_mark_auto_del() to schedule the
111  * automatic deletion, and generic qdev code calls blockdev_auto_del()
112  * when deletion is actually safe.
113  */
114 void blockdev_mark_auto_del(BlockBackend *blk)
115 {
116     DriveInfo *dinfo = blk_legacy_dinfo(blk);
117     BlockDriverState *bs = blk_bs(blk);
118     AioContext *aio_context;
119 
120     if (!dinfo) {
121         return;
122     }
123 
124     aio_context = bdrv_get_aio_context(bs);
125     aio_context_acquire(aio_context);
126 
127     if (bs->job) {
128         block_job_cancel(bs->job);
129     }
130 
131     aio_context_release(aio_context);
132 
133     dinfo->auto_del = 1;
134 }
135 
136 void blockdev_auto_del(BlockBackend *blk)
137 {
138     DriveInfo *dinfo = blk_legacy_dinfo(blk);
139 
140     if (dinfo && dinfo->auto_del) {
141         blk_unref(blk);
142     }
143 }
144 
145 /**
146  * Returns the current mapping of how many units per bus
147  * a particular interface can support.
148  *
149  *  A positive integer indicates n units per bus.
150  *  0 implies the mapping has not been established.
151  * -1 indicates an invalid BlockInterfaceType was given.
152  */
153 int drive_get_max_devs(BlockInterfaceType type)
154 {
155     if (type >= IF_IDE && type < IF_COUNT) {
156         return if_max_devs[type];
157     }
158 
159     return -1;
160 }
161 
162 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
163 {
164     int max_devs = if_max_devs[type];
165     return max_devs ? index / max_devs : 0;
166 }
167 
168 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
169 {
170     int max_devs = if_max_devs[type];
171     return max_devs ? index % max_devs : index;
172 }
173 
174 QemuOpts *drive_def(const char *optstr)
175 {
176     return qemu_opts_parse(qemu_find_opts("drive"), optstr, 0);
177 }
178 
179 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
180                     const char *optstr)
181 {
182     QemuOpts *opts;
183 
184     opts = drive_def(optstr);
185     if (!opts) {
186         return NULL;
187     }
188     if (type != IF_DEFAULT) {
189         qemu_opt_set(opts, "if", if_name[type], &error_abort);
190     }
191     if (index >= 0) {
192         qemu_opt_set_number(opts, "index", index, &error_abort);
193     }
194     if (file)
195         qemu_opt_set(opts, "file", file, &error_abort);
196     return opts;
197 }
198 
199 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
200 {
201     BlockBackend *blk;
202     DriveInfo *dinfo;
203 
204     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
205         dinfo = blk_legacy_dinfo(blk);
206         if (dinfo && dinfo->type == type
207             && dinfo->bus == bus && dinfo->unit == unit) {
208             return dinfo;
209         }
210     }
211 
212     return NULL;
213 }
214 
215 bool drive_check_orphaned(void)
216 {
217     BlockBackend *blk;
218     DriveInfo *dinfo;
219     bool rs = false;
220 
221     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
222         dinfo = blk_legacy_dinfo(blk);
223         /* If dinfo->bdrv->dev is NULL, it has no device attached. */
224         /* Unless this is a default drive, this may be an oversight. */
225         if (!blk_get_attached_dev(blk) && !dinfo->is_default &&
226             dinfo->type != IF_NONE) {
227             fprintf(stderr, "Warning: Orphaned drive without device: "
228                     "id=%s,file=%s,if=%s,bus=%d,unit=%d\n",
229                     blk_name(blk), blk_bs(blk)->filename, if_name[dinfo->type],
230                     dinfo->bus, dinfo->unit);
231             rs = true;
232         }
233     }
234 
235     return rs;
236 }
237 
238 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
239 {
240     return drive_get(type,
241                      drive_index_to_bus_id(type, index),
242                      drive_index_to_unit_id(type, index));
243 }
244 
245 int drive_get_max_bus(BlockInterfaceType type)
246 {
247     int max_bus;
248     BlockBackend *blk;
249     DriveInfo *dinfo;
250 
251     max_bus = -1;
252     for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
253         dinfo = blk_legacy_dinfo(blk);
254         if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
255             max_bus = dinfo->bus;
256         }
257     }
258     return max_bus;
259 }
260 
261 /* Get a block device.  This should only be used for single-drive devices
262    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
263    appropriate bus.  */
264 DriveInfo *drive_get_next(BlockInterfaceType type)
265 {
266     static int next_block_unit[IF_COUNT];
267 
268     return drive_get(type, 0, next_block_unit[type]++);
269 }
270 
271 static void bdrv_format_print(void *opaque, const char *name)
272 {
273     error_printf(" %s", name);
274 }
275 
276 typedef struct {
277     QEMUBH *bh;
278     BlockDriverState *bs;
279 } BDRVPutRefBH;
280 
281 static void bdrv_put_ref_bh(void *opaque)
282 {
283     BDRVPutRefBH *s = opaque;
284 
285     bdrv_unref(s->bs);
286     qemu_bh_delete(s->bh);
287     g_free(s);
288 }
289 
290 /*
291  * Release a BDS reference in a BH
292  *
293  * It is not safe to use bdrv_unref() from a callback function when the callers
294  * still need the BlockDriverState.  In such cases we schedule a BH to release
295  * the reference.
296  */
297 static void bdrv_put_ref_bh_schedule(BlockDriverState *bs)
298 {
299     BDRVPutRefBH *s;
300 
301     s = g_new(BDRVPutRefBH, 1);
302     s->bh = qemu_bh_new(bdrv_put_ref_bh, s);
303     s->bs = bs;
304     qemu_bh_schedule(s->bh);
305 }
306 
307 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
308 {
309     if (!strcmp(buf, "ignore")) {
310         return BLOCKDEV_ON_ERROR_IGNORE;
311     } else if (!is_read && !strcmp(buf, "enospc")) {
312         return BLOCKDEV_ON_ERROR_ENOSPC;
313     } else if (!strcmp(buf, "stop")) {
314         return BLOCKDEV_ON_ERROR_STOP;
315     } else if (!strcmp(buf, "report")) {
316         return BLOCKDEV_ON_ERROR_REPORT;
317     } else {
318         error_setg(errp, "'%s' invalid %s error action",
319                    buf, is_read ? "read" : "write");
320         return -1;
321     }
322 }
323 
324 static bool check_throttle_config(ThrottleConfig *cfg, Error **errp)
325 {
326     if (throttle_conflicting(cfg)) {
327         error_setg(errp, "bps/iops/max total values and read/write values"
328                          " cannot be used at the same time");
329         return false;
330     }
331 
332     if (!throttle_is_valid(cfg)) {
333         error_setg(errp, "bps/iops/maxs values must be 0 or greater");
334         return false;
335     }
336 
337     return true;
338 }
339 
340 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
341 
342 /* Takes the ownership of bs_opts */
343 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
344                                    Error **errp)
345 {
346     const char *buf;
347     int ro = 0;
348     int bdrv_flags = 0;
349     int on_read_error, on_write_error;
350     BlockBackend *blk;
351     BlockDriverState *bs;
352     ThrottleConfig cfg;
353     int snapshot = 0;
354     bool copy_on_read;
355     Error *error = NULL;
356     QemuOpts *opts;
357     const char *id;
358     bool has_driver_specific_opts;
359     BlockdevDetectZeroesOptions detect_zeroes;
360 
361     /* Check common options by copying from bs_opts to opts, all other options
362      * stay in bs_opts for processing by bdrv_open(). */
363     id = qdict_get_try_str(bs_opts, "id");
364     opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, &error);
365     if (error) {
366         error_propagate(errp, error);
367         goto err_no_opts;
368     }
369 
370     qemu_opts_absorb_qdict(opts, bs_opts, &error);
371     if (error) {
372         error_propagate(errp, error);
373         goto early_err;
374     }
375 
376     if (id) {
377         qdict_del(bs_opts, "id");
378     }
379 
380     has_driver_specific_opts = !!qdict_size(bs_opts);
381 
382     /* extract parameters */
383     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
384     ro = qemu_opt_get_bool(opts, "read-only", 0);
385     copy_on_read = qemu_opt_get_bool(opts, "copy-on-read", false);
386 
387     if ((buf = qemu_opt_get(opts, "discard")) != NULL) {
388         if (bdrv_parse_discard_flags(buf, &bdrv_flags) != 0) {
389             error_setg(errp, "invalid discard option");
390             goto early_err;
391         }
392     }
393 
394     if (qemu_opt_get_bool(opts, "cache.writeback", true)) {
395         bdrv_flags |= BDRV_O_CACHE_WB;
396     }
397     if (qemu_opt_get_bool(opts, "cache.direct", false)) {
398         bdrv_flags |= BDRV_O_NOCACHE;
399     }
400     if (qemu_opt_get_bool(opts, "cache.no-flush", false)) {
401         bdrv_flags |= BDRV_O_NO_FLUSH;
402     }
403 
404 #ifdef CONFIG_LINUX_AIO
405     if ((buf = qemu_opt_get(opts, "aio")) != NULL) {
406         if (!strcmp(buf, "native")) {
407             bdrv_flags |= BDRV_O_NATIVE_AIO;
408         } else if (!strcmp(buf, "threads")) {
409             /* this is the default */
410         } else {
411            error_setg(errp, "invalid aio option");
412            goto early_err;
413         }
414     }
415 #endif
416 
417     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
418         if (is_help_option(buf)) {
419             error_printf("Supported formats:");
420             bdrv_iterate_format(bdrv_format_print, NULL);
421             error_printf("\n");
422             goto early_err;
423         }
424 
425         if (qdict_haskey(bs_opts, "driver")) {
426             error_setg(errp, "Cannot specify both 'driver' and 'format'");
427             goto early_err;
428         }
429         qdict_put(bs_opts, "driver", qstring_from_str(buf));
430     }
431 
432     /* disk I/O throttling */
433     memset(&cfg, 0, sizeof(cfg));
434     cfg.buckets[THROTTLE_BPS_TOTAL].avg =
435         qemu_opt_get_number(opts, "throttling.bps-total", 0);
436     cfg.buckets[THROTTLE_BPS_READ].avg  =
437         qemu_opt_get_number(opts, "throttling.bps-read", 0);
438     cfg.buckets[THROTTLE_BPS_WRITE].avg =
439         qemu_opt_get_number(opts, "throttling.bps-write", 0);
440     cfg.buckets[THROTTLE_OPS_TOTAL].avg =
441         qemu_opt_get_number(opts, "throttling.iops-total", 0);
442     cfg.buckets[THROTTLE_OPS_READ].avg =
443         qemu_opt_get_number(opts, "throttling.iops-read", 0);
444     cfg.buckets[THROTTLE_OPS_WRITE].avg =
445         qemu_opt_get_number(opts, "throttling.iops-write", 0);
446 
447     cfg.buckets[THROTTLE_BPS_TOTAL].max =
448         qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
449     cfg.buckets[THROTTLE_BPS_READ].max  =
450         qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
451     cfg.buckets[THROTTLE_BPS_WRITE].max =
452         qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
453     cfg.buckets[THROTTLE_OPS_TOTAL].max =
454         qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
455     cfg.buckets[THROTTLE_OPS_READ].max =
456         qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
457     cfg.buckets[THROTTLE_OPS_WRITE].max =
458         qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
459 
460     cfg.op_size = qemu_opt_get_number(opts, "throttling.iops-size", 0);
461 
462     if (!check_throttle_config(&cfg, &error)) {
463         error_propagate(errp, error);
464         goto early_err;
465     }
466 
467     on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
468     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
469         on_write_error = parse_block_error_action(buf, 0, &error);
470         if (error) {
471             error_propagate(errp, error);
472             goto early_err;
473         }
474     }
475 
476     on_read_error = BLOCKDEV_ON_ERROR_REPORT;
477     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
478         on_read_error = parse_block_error_action(buf, 1, &error);
479         if (error) {
480             error_propagate(errp, error);
481             goto early_err;
482         }
483     }
484 
485     detect_zeroes =
486         qapi_enum_parse(BlockdevDetectZeroesOptions_lookup,
487                         qemu_opt_get(opts, "detect-zeroes"),
488                         BLOCKDEV_DETECT_ZEROES_OPTIONS_MAX,
489                         BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
490                         &error);
491     if (error) {
492         error_propagate(errp, error);
493         goto early_err;
494     }
495 
496     if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
497         !(bdrv_flags & BDRV_O_UNMAP)) {
498         error_setg(errp, "setting detect-zeroes to unmap is not allowed "
499                          "without setting discard operation to unmap");
500         goto early_err;
501     }
502 
503     /* init */
504     if ((!file || !*file) && !has_driver_specific_opts) {
505         blk = blk_new_with_bs(qemu_opts_id(opts), errp);
506         if (!blk) {
507             goto early_err;
508         }
509 
510         bs = blk_bs(blk);
511         bs->open_flags = snapshot ? BDRV_O_SNAPSHOT : 0;
512         bs->read_only = ro;
513 
514         QDECREF(bs_opts);
515     } else {
516         if (file && !*file) {
517             file = NULL;
518         }
519 
520         if (snapshot) {
521             /* always use cache=unsafe with snapshot */
522             bdrv_flags &= ~BDRV_O_CACHE_MASK;
523             bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
524         }
525 
526         if (copy_on_read) {
527             bdrv_flags |= BDRV_O_COPY_ON_READ;
528         }
529 
530         if (runstate_check(RUN_STATE_INMIGRATE)) {
531             bdrv_flags |= BDRV_O_INCOMING;
532         }
533 
534         bdrv_flags |= ro ? 0 : BDRV_O_RDWR;
535 
536         blk = blk_new_open(qemu_opts_id(opts), file, NULL, bs_opts, bdrv_flags,
537                            errp);
538         if (!blk) {
539             goto err_no_bs_opts;
540         }
541         bs = blk_bs(blk);
542     }
543 
544     bs->detect_zeroes = detect_zeroes;
545 
546     bdrv_set_on_error(bs, on_read_error, on_write_error);
547 
548     /* disk I/O throttling */
549     if (throttle_enabled(&cfg)) {
550         bdrv_io_limits_enable(bs);
551         bdrv_set_io_limits(bs, &cfg);
552     }
553 
554     if (bdrv_key_required(bs)) {
555         autostart = 0;
556     }
557 
558 err_no_bs_opts:
559     qemu_opts_del(opts);
560     return blk;
561 
562 early_err:
563     qemu_opts_del(opts);
564 err_no_opts:
565     QDECREF(bs_opts);
566     return NULL;
567 }
568 
569 static void qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
570                             Error **errp)
571 {
572     const char *value;
573 
574     value = qemu_opt_get(opts, from);
575     if (value) {
576         if (qemu_opt_find(opts, to)) {
577             error_setg(errp, "'%s' and its alias '%s' can't be used at the "
578                        "same time", to, from);
579             return;
580         }
581     }
582 
583     /* rename all items in opts */
584     while ((value = qemu_opt_get(opts, from))) {
585         qemu_opt_set(opts, to, value, &error_abort);
586         qemu_opt_unset(opts, from);
587     }
588 }
589 
590 QemuOptsList qemu_legacy_drive_opts = {
591     .name = "drive",
592     .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
593     .desc = {
594         {
595             .name = "bus",
596             .type = QEMU_OPT_NUMBER,
597             .help = "bus number",
598         },{
599             .name = "unit",
600             .type = QEMU_OPT_NUMBER,
601             .help = "unit number (i.e. lun for scsi)",
602         },{
603             .name = "index",
604             .type = QEMU_OPT_NUMBER,
605             .help = "index number",
606         },{
607             .name = "media",
608             .type = QEMU_OPT_STRING,
609             .help = "media type (disk, cdrom)",
610         },{
611             .name = "if",
612             .type = QEMU_OPT_STRING,
613             .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
614         },{
615             .name = "cyls",
616             .type = QEMU_OPT_NUMBER,
617             .help = "number of cylinders (ide disk geometry)",
618         },{
619             .name = "heads",
620             .type = QEMU_OPT_NUMBER,
621             .help = "number of heads (ide disk geometry)",
622         },{
623             .name = "secs",
624             .type = QEMU_OPT_NUMBER,
625             .help = "number of sectors (ide disk geometry)",
626         },{
627             .name = "trans",
628             .type = QEMU_OPT_STRING,
629             .help = "chs translation (auto, lba, none)",
630         },{
631             .name = "boot",
632             .type = QEMU_OPT_BOOL,
633             .help = "(deprecated, ignored)",
634         },{
635             .name = "addr",
636             .type = QEMU_OPT_STRING,
637             .help = "pci address (virtio only)",
638         },{
639             .name = "serial",
640             .type = QEMU_OPT_STRING,
641             .help = "disk serial number",
642         },{
643             .name = "file",
644             .type = QEMU_OPT_STRING,
645             .help = "file name",
646         },
647 
648         /* Options that are passed on, but have special semantics with -drive */
649         {
650             .name = "read-only",
651             .type = QEMU_OPT_BOOL,
652             .help = "open drive file as read-only",
653         },{
654             .name = "rerror",
655             .type = QEMU_OPT_STRING,
656             .help = "read error action",
657         },{
658             .name = "werror",
659             .type = QEMU_OPT_STRING,
660             .help = "write error action",
661         },{
662             .name = "copy-on-read",
663             .type = QEMU_OPT_BOOL,
664             .help = "copy read data from backing file into image file",
665         },
666 
667         { /* end of list */ }
668     },
669 };
670 
671 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type)
672 {
673     const char *value;
674     BlockBackend *blk;
675     DriveInfo *dinfo = NULL;
676     QDict *bs_opts;
677     QemuOpts *legacy_opts;
678     DriveMediaType media = MEDIA_DISK;
679     BlockInterfaceType type;
680     int cyls, heads, secs, translation;
681     int max_devs, bus_id, unit_id, index;
682     const char *devaddr;
683     const char *werror, *rerror;
684     bool read_only = false;
685     bool copy_on_read;
686     const char *serial;
687     const char *filename;
688     Error *local_err = NULL;
689     int i;
690 
691     /* Change legacy command line options into QMP ones */
692     static const struct {
693         const char *from;
694         const char *to;
695     } opt_renames[] = {
696         { "iops",           "throttling.iops-total" },
697         { "iops_rd",        "throttling.iops-read" },
698         { "iops_wr",        "throttling.iops-write" },
699 
700         { "bps",            "throttling.bps-total" },
701         { "bps_rd",         "throttling.bps-read" },
702         { "bps_wr",         "throttling.bps-write" },
703 
704         { "iops_max",       "throttling.iops-total-max" },
705         { "iops_rd_max",    "throttling.iops-read-max" },
706         { "iops_wr_max",    "throttling.iops-write-max" },
707 
708         { "bps_max",        "throttling.bps-total-max" },
709         { "bps_rd_max",     "throttling.bps-read-max" },
710         { "bps_wr_max",     "throttling.bps-write-max" },
711 
712         { "iops_size",      "throttling.iops-size" },
713 
714         { "readonly",       "read-only" },
715     };
716 
717     for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
718         qemu_opt_rename(all_opts, opt_renames[i].from, opt_renames[i].to,
719                         &local_err);
720         if (local_err) {
721             error_report_err(local_err);
722             return NULL;
723         }
724     }
725 
726     value = qemu_opt_get(all_opts, "cache");
727     if (value) {
728         int flags = 0;
729 
730         if (bdrv_parse_cache_flags(value, &flags) != 0) {
731             error_report("invalid cache option");
732             return NULL;
733         }
734 
735         /* Specific options take precedence */
736         if (!qemu_opt_get(all_opts, "cache.writeback")) {
737             qemu_opt_set_bool(all_opts, "cache.writeback",
738                               !!(flags & BDRV_O_CACHE_WB), &error_abort);
739         }
740         if (!qemu_opt_get(all_opts, "cache.direct")) {
741             qemu_opt_set_bool(all_opts, "cache.direct",
742                               !!(flags & BDRV_O_NOCACHE), &error_abort);
743         }
744         if (!qemu_opt_get(all_opts, "cache.no-flush")) {
745             qemu_opt_set_bool(all_opts, "cache.no-flush",
746                               !!(flags & BDRV_O_NO_FLUSH), &error_abort);
747         }
748         qemu_opt_unset(all_opts, "cache");
749     }
750 
751     /* Get a QDict for processing the options */
752     bs_opts = qdict_new();
753     qemu_opts_to_qdict(all_opts, bs_opts);
754 
755     legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
756                                    &error_abort);
757     qemu_opts_absorb_qdict(legacy_opts, bs_opts, &local_err);
758     if (local_err) {
759         error_report_err(local_err);
760         goto fail;
761     }
762 
763     /* Deprecated option boot=[on|off] */
764     if (qemu_opt_get(legacy_opts, "boot") != NULL) {
765         fprintf(stderr, "qemu-kvm: boot=on|off is deprecated and will be "
766                 "ignored. Future versions will reject this parameter. Please "
767                 "update your scripts.\n");
768     }
769 
770     /* Media type */
771     value = qemu_opt_get(legacy_opts, "media");
772     if (value) {
773         if (!strcmp(value, "disk")) {
774             media = MEDIA_DISK;
775         } else if (!strcmp(value, "cdrom")) {
776             media = MEDIA_CDROM;
777             read_only = true;
778         } else {
779             error_report("'%s' invalid media", value);
780             goto fail;
781         }
782     }
783 
784     /* copy-on-read is disabled with a warning for read-only devices */
785     read_only |= qemu_opt_get_bool(legacy_opts, "read-only", false);
786     copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
787 
788     if (read_only && copy_on_read) {
789         error_report("warning: disabling copy-on-read on read-only drive");
790         copy_on_read = false;
791     }
792 
793     qdict_put(bs_opts, "read-only",
794               qstring_from_str(read_only ? "on" : "off"));
795     qdict_put(bs_opts, "copy-on-read",
796               qstring_from_str(copy_on_read ? "on" :"off"));
797 
798     /* Controller type */
799     value = qemu_opt_get(legacy_opts, "if");
800     if (value) {
801         for (type = 0;
802              type < IF_COUNT && strcmp(value, if_name[type]);
803              type++) {
804         }
805         if (type == IF_COUNT) {
806             error_report("unsupported bus type '%s'", value);
807             goto fail;
808         }
809     } else {
810         type = block_default_type;
811     }
812 
813     /* Geometry */
814     cyls  = qemu_opt_get_number(legacy_opts, "cyls", 0);
815     heads = qemu_opt_get_number(legacy_opts, "heads", 0);
816     secs  = qemu_opt_get_number(legacy_opts, "secs", 0);
817 
818     if (cyls || heads || secs) {
819         if (cyls < 1) {
820             error_report("invalid physical cyls number");
821             goto fail;
822         }
823         if (heads < 1) {
824             error_report("invalid physical heads number");
825             goto fail;
826         }
827         if (secs < 1) {
828             error_report("invalid physical secs number");
829             goto fail;
830         }
831     }
832 
833     translation = BIOS_ATA_TRANSLATION_AUTO;
834     value = qemu_opt_get(legacy_opts, "trans");
835     if (value != NULL) {
836         if (!cyls) {
837             error_report("'%s' trans must be used with cyls, heads and secs",
838                          value);
839             goto fail;
840         }
841         if (!strcmp(value, "none")) {
842             translation = BIOS_ATA_TRANSLATION_NONE;
843         } else if (!strcmp(value, "lba")) {
844             translation = BIOS_ATA_TRANSLATION_LBA;
845         } else if (!strcmp(value, "large")) {
846             translation = BIOS_ATA_TRANSLATION_LARGE;
847         } else if (!strcmp(value, "rechs")) {
848             translation = BIOS_ATA_TRANSLATION_RECHS;
849         } else if (!strcmp(value, "auto")) {
850             translation = BIOS_ATA_TRANSLATION_AUTO;
851         } else {
852             error_report("'%s' invalid translation type", value);
853             goto fail;
854         }
855     }
856 
857     if (media == MEDIA_CDROM) {
858         if (cyls || secs || heads) {
859             error_report("CHS can't be set with media=cdrom");
860             goto fail;
861         }
862     }
863 
864     /* Device address specified by bus/unit or index.
865      * If none was specified, try to find the first free one. */
866     bus_id  = qemu_opt_get_number(legacy_opts, "bus", 0);
867     unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
868     index   = qemu_opt_get_number(legacy_opts, "index", -1);
869 
870     max_devs = if_max_devs[type];
871 
872     if (index != -1) {
873         if (bus_id != 0 || unit_id != -1) {
874             error_report("index cannot be used with bus and unit");
875             goto fail;
876         }
877         bus_id = drive_index_to_bus_id(type, index);
878         unit_id = drive_index_to_unit_id(type, index);
879     }
880 
881     if (unit_id == -1) {
882        unit_id = 0;
883        while (drive_get(type, bus_id, unit_id) != NULL) {
884            unit_id++;
885            if (max_devs && unit_id >= max_devs) {
886                unit_id -= max_devs;
887                bus_id++;
888            }
889        }
890     }
891 
892     if (max_devs && unit_id >= max_devs) {
893         error_report("unit %d too big (max is %d)", unit_id, max_devs - 1);
894         goto fail;
895     }
896 
897     if (drive_get(type, bus_id, unit_id) != NULL) {
898         error_report("drive with bus=%d, unit=%d (index=%d) exists",
899                      bus_id, unit_id, index);
900         goto fail;
901     }
902 
903     /* Serial number */
904     serial = qemu_opt_get(legacy_opts, "serial");
905 
906     /* no id supplied -> create one */
907     if (qemu_opts_id(all_opts) == NULL) {
908         char *new_id;
909         const char *mediastr = "";
910         if (type == IF_IDE || type == IF_SCSI) {
911             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
912         }
913         if (max_devs) {
914             new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
915                                      mediastr, unit_id);
916         } else {
917             new_id = g_strdup_printf("%s%s%i", if_name[type],
918                                      mediastr, unit_id);
919         }
920         qdict_put(bs_opts, "id", qstring_from_str(new_id));
921         g_free(new_id);
922     }
923 
924     /* Add virtio block device */
925     devaddr = qemu_opt_get(legacy_opts, "addr");
926     if (devaddr && type != IF_VIRTIO) {
927         error_report("addr is not supported by this bus type");
928         goto fail;
929     }
930 
931     if (type == IF_VIRTIO) {
932         QemuOpts *devopts;
933         devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
934                                    &error_abort);
935         if (arch_type == QEMU_ARCH_S390X) {
936             qemu_opt_set(devopts, "driver", "virtio-blk-s390", &error_abort);
937         } else {
938             qemu_opt_set(devopts, "driver", "virtio-blk-pci", &error_abort);
939         }
940         qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
941                      &error_abort);
942         if (devaddr) {
943             qemu_opt_set(devopts, "addr", devaddr, &error_abort);
944         }
945     }
946 
947     filename = qemu_opt_get(legacy_opts, "file");
948 
949     /* Check werror/rerror compatibility with if=... */
950     werror = qemu_opt_get(legacy_opts, "werror");
951     if (werror != NULL) {
952         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
953             type != IF_NONE) {
954             error_report("werror is not supported by this bus type");
955             goto fail;
956         }
957         qdict_put(bs_opts, "werror", qstring_from_str(werror));
958     }
959 
960     rerror = qemu_opt_get(legacy_opts, "rerror");
961     if (rerror != NULL) {
962         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
963             type != IF_NONE) {
964             error_report("rerror is not supported by this bus type");
965             goto fail;
966         }
967         qdict_put(bs_opts, "rerror", qstring_from_str(rerror));
968     }
969 
970     /* Actual block device init: Functionality shared with blockdev-add */
971     blk = blockdev_init(filename, bs_opts, &local_err);
972     bs_opts = NULL;
973     if (!blk) {
974         if (local_err) {
975             error_report_err(local_err);
976         }
977         goto fail;
978     } else {
979         assert(!local_err);
980     }
981 
982     /* Create legacy DriveInfo */
983     dinfo = g_malloc0(sizeof(*dinfo));
984     dinfo->opts = all_opts;
985 
986     dinfo->cyls = cyls;
987     dinfo->heads = heads;
988     dinfo->secs = secs;
989     dinfo->trans = translation;
990 
991     dinfo->type = type;
992     dinfo->bus = bus_id;
993     dinfo->unit = unit_id;
994     dinfo->devaddr = devaddr;
995     dinfo->serial = g_strdup(serial);
996 
997     blk_set_legacy_dinfo(blk, dinfo);
998 
999     switch(type) {
1000     case IF_IDE:
1001     case IF_SCSI:
1002     case IF_XEN:
1003     case IF_NONE:
1004         dinfo->media_cd = media == MEDIA_CDROM;
1005         break;
1006     default:
1007         break;
1008     }
1009 
1010 fail:
1011     qemu_opts_del(legacy_opts);
1012     QDECREF(bs_opts);
1013     return dinfo;
1014 }
1015 
1016 void hmp_commit(Monitor *mon, const QDict *qdict)
1017 {
1018     const char *device = qdict_get_str(qdict, "device");
1019     BlockBackend *blk;
1020     int ret;
1021 
1022     if (!strcmp(device, "all")) {
1023         ret = bdrv_commit_all();
1024     } else {
1025         blk = blk_by_name(device);
1026         if (!blk) {
1027             monitor_printf(mon, "Device '%s' not found\n", device);
1028             return;
1029         }
1030         ret = bdrv_commit(blk_bs(blk));
1031     }
1032     if (ret < 0) {
1033         monitor_printf(mon, "'commit' error for '%s': %s\n", device,
1034                        strerror(-ret));
1035     }
1036 }
1037 
1038 static void blockdev_do_action(int kind, void *data, Error **errp)
1039 {
1040     TransactionAction action;
1041     TransactionActionList list;
1042 
1043     action.kind = kind;
1044     action.data = data;
1045     list.value = &action;
1046     list.next = NULL;
1047     qmp_transaction(&list, errp);
1048 }
1049 
1050 void qmp_blockdev_snapshot_sync(bool has_device, const char *device,
1051                                 bool has_node_name, const char *node_name,
1052                                 const char *snapshot_file,
1053                                 bool has_snapshot_node_name,
1054                                 const char *snapshot_node_name,
1055                                 bool has_format, const char *format,
1056                                 bool has_mode, NewImageMode mode, Error **errp)
1057 {
1058     BlockdevSnapshot snapshot = {
1059         .has_device = has_device,
1060         .device = (char *) device,
1061         .has_node_name = has_node_name,
1062         .node_name = (char *) node_name,
1063         .snapshot_file = (char *) snapshot_file,
1064         .has_snapshot_node_name = has_snapshot_node_name,
1065         .snapshot_node_name = (char *) snapshot_node_name,
1066         .has_format = has_format,
1067         .format = (char *) format,
1068         .has_mode = has_mode,
1069         .mode = mode,
1070     };
1071     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1072                        &snapshot, errp);
1073 }
1074 
1075 void qmp_blockdev_snapshot_internal_sync(const char *device,
1076                                          const char *name,
1077                                          Error **errp)
1078 {
1079     BlockdevSnapshotInternal snapshot = {
1080         .device = (char *) device,
1081         .name = (char *) name
1082     };
1083 
1084     blockdev_do_action(TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1085                        &snapshot, errp);
1086 }
1087 
1088 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1089                                                          bool has_id,
1090                                                          const char *id,
1091                                                          bool has_name,
1092                                                          const char *name,
1093                                                          Error **errp)
1094 {
1095     BlockDriverState *bs;
1096     BlockBackend *blk;
1097     AioContext *aio_context;
1098     QEMUSnapshotInfo sn;
1099     Error *local_err = NULL;
1100     SnapshotInfo *info = NULL;
1101     int ret;
1102 
1103     blk = blk_by_name(device);
1104     if (!blk) {
1105         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1106         return NULL;
1107     }
1108     bs = blk_bs(blk);
1109 
1110     if (!has_id) {
1111         id = NULL;
1112     }
1113 
1114     if (!has_name) {
1115         name = NULL;
1116     }
1117 
1118     if (!id && !name) {
1119         error_setg(errp, "Name or id must be provided");
1120         return NULL;
1121     }
1122 
1123     aio_context = bdrv_get_aio_context(bs);
1124     aio_context_acquire(aio_context);
1125 
1126     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1127         goto out_aio_context;
1128     }
1129 
1130     ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1131     if (local_err) {
1132         error_propagate(errp, local_err);
1133         goto out_aio_context;
1134     }
1135     if (!ret) {
1136         error_setg(errp,
1137                    "Snapshot with id '%s' and name '%s' does not exist on "
1138                    "device '%s'",
1139                    STR_OR_NULL(id), STR_OR_NULL(name), device);
1140         goto out_aio_context;
1141     }
1142 
1143     bdrv_snapshot_delete(bs, id, name, &local_err);
1144     if (local_err) {
1145         error_propagate(errp, local_err);
1146         goto out_aio_context;
1147     }
1148 
1149     aio_context_release(aio_context);
1150 
1151     info = g_new0(SnapshotInfo, 1);
1152     info->id = g_strdup(sn.id_str);
1153     info->name = g_strdup(sn.name);
1154     info->date_nsec = sn.date_nsec;
1155     info->date_sec = sn.date_sec;
1156     info->vm_state_size = sn.vm_state_size;
1157     info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1158     info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1159 
1160     return info;
1161 
1162 out_aio_context:
1163     aio_context_release(aio_context);
1164     return NULL;
1165 }
1166 
1167 /* New and old BlockDriverState structs for atomic group operations */
1168 
1169 typedef struct BlkTransactionState BlkTransactionState;
1170 
1171 /* Only prepare() may fail. In a single transaction, only one of commit() or
1172    abort() will be called, clean() will always be called if it present. */
1173 typedef struct BdrvActionOps {
1174     /* Size of state struct, in bytes. */
1175     size_t instance_size;
1176     /* Prepare the work, must NOT be NULL. */
1177     void (*prepare)(BlkTransactionState *common, Error **errp);
1178     /* Commit the changes, can be NULL. */
1179     void (*commit)(BlkTransactionState *common);
1180     /* Abort the changes on fail, can be NULL. */
1181     void (*abort)(BlkTransactionState *common);
1182     /* Clean up resource in the end, can be NULL. */
1183     void (*clean)(BlkTransactionState *common);
1184 } BdrvActionOps;
1185 
1186 /*
1187  * This structure must be arranged as first member in child type, assuming
1188  * that compiler will also arrange it to the same address with parent instance.
1189  * Later it will be used in free().
1190  */
1191 struct BlkTransactionState {
1192     TransactionAction *action;
1193     const BdrvActionOps *ops;
1194     QSIMPLEQ_ENTRY(BlkTransactionState) entry;
1195 };
1196 
1197 /* internal snapshot private data */
1198 typedef struct InternalSnapshotState {
1199     BlkTransactionState common;
1200     BlockDriverState *bs;
1201     AioContext *aio_context;
1202     QEMUSnapshotInfo sn;
1203 } InternalSnapshotState;
1204 
1205 static void internal_snapshot_prepare(BlkTransactionState *common,
1206                                       Error **errp)
1207 {
1208     Error *local_err = NULL;
1209     const char *device;
1210     const char *name;
1211     BlockBackend *blk;
1212     BlockDriverState *bs;
1213     QEMUSnapshotInfo old_sn, *sn;
1214     bool ret;
1215     qemu_timeval tv;
1216     BlockdevSnapshotInternal *internal;
1217     InternalSnapshotState *state;
1218     int ret1;
1219 
1220     g_assert(common->action->kind ==
1221              TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1222     internal = common->action->blockdev_snapshot_internal_sync;
1223     state = DO_UPCAST(InternalSnapshotState, common, common);
1224 
1225     /* 1. parse input */
1226     device = internal->device;
1227     name = internal->name;
1228 
1229     /* 2. check for validation */
1230     blk = blk_by_name(device);
1231     if (!blk) {
1232         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1233         return;
1234     }
1235     bs = blk_bs(blk);
1236 
1237     /* AioContext is released in .clean() */
1238     state->aio_context = bdrv_get_aio_context(bs);
1239     aio_context_acquire(state->aio_context);
1240 
1241     if (!bdrv_is_inserted(bs)) {
1242         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1243         return;
1244     }
1245 
1246     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1247         return;
1248     }
1249 
1250     if (bdrv_is_read_only(bs)) {
1251         error_setg(errp, "Device '%s' is read only", device);
1252         return;
1253     }
1254 
1255     if (!bdrv_can_snapshot(bs)) {
1256         error_setg(errp, "Block format '%s' used by device '%s' "
1257                    "does not support internal snapshots",
1258                    bs->drv->format_name, device);
1259         return;
1260     }
1261 
1262     if (!strlen(name)) {
1263         error_setg(errp, "Name is empty");
1264         return;
1265     }
1266 
1267     /* check whether a snapshot with name exist */
1268     ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1269                                             &local_err);
1270     if (local_err) {
1271         error_propagate(errp, local_err);
1272         return;
1273     } else if (ret) {
1274         error_setg(errp,
1275                    "Snapshot with name '%s' already exists on device '%s'",
1276                    name, device);
1277         return;
1278     }
1279 
1280     /* 3. take the snapshot */
1281     sn = &state->sn;
1282     pstrcpy(sn->name, sizeof(sn->name), name);
1283     qemu_gettimeofday(&tv);
1284     sn->date_sec = tv.tv_sec;
1285     sn->date_nsec = tv.tv_usec * 1000;
1286     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1287 
1288     ret1 = bdrv_snapshot_create(bs, sn);
1289     if (ret1 < 0) {
1290         error_setg_errno(errp, -ret1,
1291                          "Failed to create snapshot '%s' on device '%s'",
1292                          name, device);
1293         return;
1294     }
1295 
1296     /* 4. succeed, mark a snapshot is created */
1297     state->bs = bs;
1298 }
1299 
1300 static void internal_snapshot_abort(BlkTransactionState *common)
1301 {
1302     InternalSnapshotState *state =
1303                              DO_UPCAST(InternalSnapshotState, common, common);
1304     BlockDriverState *bs = state->bs;
1305     QEMUSnapshotInfo *sn = &state->sn;
1306     Error *local_error = NULL;
1307 
1308     if (!bs) {
1309         return;
1310     }
1311 
1312     if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1313         error_report("Failed to delete snapshot with id '%s' and name '%s' on "
1314                      "device '%s' in abort: %s",
1315                      sn->id_str,
1316                      sn->name,
1317                      bdrv_get_device_name(bs),
1318                      error_get_pretty(local_error));
1319         error_free(local_error);
1320     }
1321 }
1322 
1323 static void internal_snapshot_clean(BlkTransactionState *common)
1324 {
1325     InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1326                                              common, common);
1327 
1328     if (state->aio_context) {
1329         aio_context_release(state->aio_context);
1330     }
1331 }
1332 
1333 /* external snapshot private data */
1334 typedef struct ExternalSnapshotState {
1335     BlkTransactionState common;
1336     BlockDriverState *old_bs;
1337     BlockDriverState *new_bs;
1338     AioContext *aio_context;
1339 } ExternalSnapshotState;
1340 
1341 static void external_snapshot_prepare(BlkTransactionState *common,
1342                                       Error **errp)
1343 {
1344     BlockDriver *drv;
1345     int flags, ret;
1346     QDict *options = NULL;
1347     Error *local_err = NULL;
1348     bool has_device = false;
1349     const char *device;
1350     bool has_node_name = false;
1351     const char *node_name;
1352     bool has_snapshot_node_name = false;
1353     const char *snapshot_node_name;
1354     const char *new_image_file;
1355     const char *format = "qcow2";
1356     enum NewImageMode mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1357     ExternalSnapshotState *state =
1358                              DO_UPCAST(ExternalSnapshotState, common, common);
1359     TransactionAction *action = common->action;
1360 
1361     /* get parameters */
1362     g_assert(action->kind == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC);
1363 
1364     has_device = action->blockdev_snapshot_sync->has_device;
1365     device = action->blockdev_snapshot_sync->device;
1366     has_node_name = action->blockdev_snapshot_sync->has_node_name;
1367     node_name = action->blockdev_snapshot_sync->node_name;
1368     has_snapshot_node_name =
1369         action->blockdev_snapshot_sync->has_snapshot_node_name;
1370     snapshot_node_name = action->blockdev_snapshot_sync->snapshot_node_name;
1371 
1372     new_image_file = action->blockdev_snapshot_sync->snapshot_file;
1373     if (action->blockdev_snapshot_sync->has_format) {
1374         format = action->blockdev_snapshot_sync->format;
1375     }
1376     if (action->blockdev_snapshot_sync->has_mode) {
1377         mode = action->blockdev_snapshot_sync->mode;
1378     }
1379 
1380     /* start processing */
1381     drv = bdrv_find_format(format);
1382     if (!drv) {
1383         error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1384         return;
1385     }
1386 
1387     state->old_bs = bdrv_lookup_bs(has_device ? device : NULL,
1388                                    has_node_name ? node_name : NULL,
1389                                    &local_err);
1390     if (local_err) {
1391         error_propagate(errp, local_err);
1392         return;
1393     }
1394 
1395     if (has_node_name && !has_snapshot_node_name) {
1396         error_setg(errp, "New snapshot node name missing");
1397         return;
1398     }
1399 
1400     if (has_snapshot_node_name && bdrv_find_node(snapshot_node_name)) {
1401         error_setg(errp, "New snapshot node name already existing");
1402         return;
1403     }
1404 
1405     /* Acquire AioContext now so any threads operating on old_bs stop */
1406     state->aio_context = bdrv_get_aio_context(state->old_bs);
1407     aio_context_acquire(state->aio_context);
1408 
1409     if (!bdrv_is_inserted(state->old_bs)) {
1410         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1411         return;
1412     }
1413 
1414     if (bdrv_op_is_blocked(state->old_bs,
1415                            BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1416         return;
1417     }
1418 
1419     if (!bdrv_is_read_only(state->old_bs)) {
1420         if (bdrv_flush(state->old_bs)) {
1421             error_set(errp, QERR_IO_ERROR);
1422             return;
1423         }
1424     }
1425 
1426     if (!bdrv_is_first_non_filter(state->old_bs)) {
1427         error_set(errp, QERR_FEATURE_DISABLED, "snapshot");
1428         return;
1429     }
1430 
1431     flags = state->old_bs->open_flags;
1432 
1433     /* create new image w/backing file */
1434     if (mode != NEW_IMAGE_MODE_EXISTING) {
1435         bdrv_img_create(new_image_file, format,
1436                         state->old_bs->filename,
1437                         state->old_bs->drv->format_name,
1438                         NULL, -1, flags, &local_err, false);
1439         if (local_err) {
1440             error_propagate(errp, local_err);
1441             return;
1442         }
1443     }
1444 
1445     if (has_snapshot_node_name) {
1446         options = qdict_new();
1447         qdict_put(options, "node-name",
1448                   qstring_from_str(snapshot_node_name));
1449     }
1450 
1451     /* TODO Inherit bs->options or only take explicit options with an
1452      * extended QMP command? */
1453     assert(state->new_bs == NULL);
1454     ret = bdrv_open(&state->new_bs, new_image_file, NULL, options,
1455                     flags | BDRV_O_NO_BACKING, drv, &local_err);
1456     /* We will manually add the backing_hd field to the bs later */
1457     if (ret != 0) {
1458         error_propagate(errp, local_err);
1459     }
1460 }
1461 
1462 static void external_snapshot_commit(BlkTransactionState *common)
1463 {
1464     ExternalSnapshotState *state =
1465                              DO_UPCAST(ExternalSnapshotState, common, common);
1466 
1467     bdrv_set_aio_context(state->new_bs, state->aio_context);
1468 
1469     /* This removes our old bs and adds the new bs */
1470     bdrv_append(state->new_bs, state->old_bs);
1471     /* We don't need (or want) to use the transactional
1472      * bdrv_reopen_multiple() across all the entries at once, because we
1473      * don't want to abort all of them if one of them fails the reopen */
1474     bdrv_reopen(state->new_bs, state->new_bs->open_flags & ~BDRV_O_RDWR,
1475                 NULL);
1476 
1477     aio_context_release(state->aio_context);
1478 }
1479 
1480 static void external_snapshot_abort(BlkTransactionState *common)
1481 {
1482     ExternalSnapshotState *state =
1483                              DO_UPCAST(ExternalSnapshotState, common, common);
1484     if (state->new_bs) {
1485         bdrv_unref(state->new_bs);
1486     }
1487     if (state->aio_context) {
1488         aio_context_release(state->aio_context);
1489     }
1490 }
1491 
1492 typedef struct DriveBackupState {
1493     BlkTransactionState common;
1494     BlockDriverState *bs;
1495     AioContext *aio_context;
1496     BlockJob *job;
1497 } DriveBackupState;
1498 
1499 static void drive_backup_prepare(BlkTransactionState *common, Error **errp)
1500 {
1501     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1502     BlockDriverState *bs;
1503     BlockBackend *blk;
1504     DriveBackup *backup;
1505     Error *local_err = NULL;
1506 
1507     assert(common->action->kind == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1508     backup = common->action->drive_backup;
1509 
1510     blk = blk_by_name(backup->device);
1511     if (!blk) {
1512         error_set(errp, QERR_DEVICE_NOT_FOUND, backup->device);
1513         return;
1514     }
1515     bs = blk_bs(blk);
1516 
1517     /* AioContext is released in .clean() */
1518     state->aio_context = bdrv_get_aio_context(bs);
1519     aio_context_acquire(state->aio_context);
1520 
1521     qmp_drive_backup(backup->device, backup->target,
1522                      backup->has_format, backup->format,
1523                      backup->sync,
1524                      backup->has_mode, backup->mode,
1525                      backup->has_speed, backup->speed,
1526                      backup->has_on_source_error, backup->on_source_error,
1527                      backup->has_on_target_error, backup->on_target_error,
1528                      &local_err);
1529     if (local_err) {
1530         error_propagate(errp, local_err);
1531         return;
1532     }
1533 
1534     state->bs = bs;
1535     state->job = state->bs->job;
1536 }
1537 
1538 static void drive_backup_abort(BlkTransactionState *common)
1539 {
1540     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1541     BlockDriverState *bs = state->bs;
1542 
1543     /* Only cancel if it's the job we started */
1544     if (bs && bs->job && bs->job == state->job) {
1545         block_job_cancel_sync(bs->job);
1546     }
1547 }
1548 
1549 static void drive_backup_clean(BlkTransactionState *common)
1550 {
1551     DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1552 
1553     if (state->aio_context) {
1554         aio_context_release(state->aio_context);
1555     }
1556 }
1557 
1558 typedef struct BlockdevBackupState {
1559     BlkTransactionState common;
1560     BlockDriverState *bs;
1561     BlockJob *job;
1562     AioContext *aio_context;
1563 } BlockdevBackupState;
1564 
1565 static void blockdev_backup_prepare(BlkTransactionState *common, Error **errp)
1566 {
1567     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1568     BlockdevBackup *backup;
1569     BlockDriverState *bs, *target;
1570     BlockBackend *blk;
1571     Error *local_err = NULL;
1572 
1573     assert(common->action->kind == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1574     backup = common->action->blockdev_backup;
1575 
1576     blk = blk_by_name(backup->device);
1577     if (!blk) {
1578         error_setg(errp, "Device '%s' not found", backup->device);
1579         return;
1580     }
1581     bs = blk_bs(blk);
1582 
1583     blk = blk_by_name(backup->target);
1584     if (!blk) {
1585         error_setg(errp, "Device '%s' not found", backup->target);
1586         return;
1587     }
1588     target = blk_bs(blk);
1589 
1590     /* AioContext is released in .clean() */
1591     state->aio_context = bdrv_get_aio_context(bs);
1592     if (state->aio_context != bdrv_get_aio_context(target)) {
1593         state->aio_context = NULL;
1594         error_setg(errp, "Backup between two IO threads is not implemented");
1595         return;
1596     }
1597     aio_context_acquire(state->aio_context);
1598 
1599     qmp_blockdev_backup(backup->device, backup->target,
1600                         backup->sync,
1601                         backup->has_speed, backup->speed,
1602                         backup->has_on_source_error, backup->on_source_error,
1603                         backup->has_on_target_error, backup->on_target_error,
1604                         &local_err);
1605     if (local_err) {
1606         error_propagate(errp, local_err);
1607         return;
1608     }
1609 
1610     state->bs = bs;
1611     state->job = state->bs->job;
1612 }
1613 
1614 static void blockdev_backup_abort(BlkTransactionState *common)
1615 {
1616     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1617     BlockDriverState *bs = state->bs;
1618 
1619     /* Only cancel if it's the job we started */
1620     if (bs && bs->job && bs->job == state->job) {
1621         block_job_cancel_sync(bs->job);
1622     }
1623 }
1624 
1625 static void blockdev_backup_clean(BlkTransactionState *common)
1626 {
1627     BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1628 
1629     if (state->aio_context) {
1630         aio_context_release(state->aio_context);
1631     }
1632 }
1633 
1634 static void abort_prepare(BlkTransactionState *common, Error **errp)
1635 {
1636     error_setg(errp, "Transaction aborted using Abort action");
1637 }
1638 
1639 static void abort_commit(BlkTransactionState *common)
1640 {
1641     g_assert_not_reached(); /* this action never succeeds */
1642 }
1643 
1644 static const BdrvActionOps actions[] = {
1645     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
1646         .instance_size = sizeof(ExternalSnapshotState),
1647         .prepare  = external_snapshot_prepare,
1648         .commit   = external_snapshot_commit,
1649         .abort = external_snapshot_abort,
1650     },
1651     [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
1652         .instance_size = sizeof(DriveBackupState),
1653         .prepare = drive_backup_prepare,
1654         .abort = drive_backup_abort,
1655         .clean = drive_backup_clean,
1656     },
1657     [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
1658         .instance_size = sizeof(BlockdevBackupState),
1659         .prepare = blockdev_backup_prepare,
1660         .abort = blockdev_backup_abort,
1661         .clean = blockdev_backup_clean,
1662     },
1663     [TRANSACTION_ACTION_KIND_ABORT] = {
1664         .instance_size = sizeof(BlkTransactionState),
1665         .prepare = abort_prepare,
1666         .commit = abort_commit,
1667     },
1668     [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
1669         .instance_size = sizeof(InternalSnapshotState),
1670         .prepare  = internal_snapshot_prepare,
1671         .abort = internal_snapshot_abort,
1672         .clean = internal_snapshot_clean,
1673     },
1674 };
1675 
1676 /*
1677  * 'Atomic' group operations.  The operations are performed as a set, and if
1678  * any fail then we roll back all operations in the group.
1679  */
1680 void qmp_transaction(TransactionActionList *dev_list, Error **errp)
1681 {
1682     TransactionActionList *dev_entry = dev_list;
1683     BlkTransactionState *state, *next;
1684     Error *local_err = NULL;
1685 
1686     QSIMPLEQ_HEAD(snap_bdrv_states, BlkTransactionState) snap_bdrv_states;
1687     QSIMPLEQ_INIT(&snap_bdrv_states);
1688 
1689     /* drain all i/o before any operations */
1690     bdrv_drain_all();
1691 
1692     /* We don't do anything in this loop that commits us to the operations */
1693     while (NULL != dev_entry) {
1694         TransactionAction *dev_info = NULL;
1695         const BdrvActionOps *ops;
1696 
1697         dev_info = dev_entry->value;
1698         dev_entry = dev_entry->next;
1699 
1700         assert(dev_info->kind < ARRAY_SIZE(actions));
1701 
1702         ops = &actions[dev_info->kind];
1703         assert(ops->instance_size > 0);
1704 
1705         state = g_malloc0(ops->instance_size);
1706         state->ops = ops;
1707         state->action = dev_info;
1708         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
1709 
1710         state->ops->prepare(state, &local_err);
1711         if (local_err) {
1712             error_propagate(errp, local_err);
1713             goto delete_and_fail;
1714         }
1715     }
1716 
1717     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1718         if (state->ops->commit) {
1719             state->ops->commit(state);
1720         }
1721     }
1722 
1723     /* success */
1724     goto exit;
1725 
1726 delete_and_fail:
1727     /* failure, and it is all-or-none; roll back all operations */
1728     QSIMPLEQ_FOREACH(state, &snap_bdrv_states, entry) {
1729         if (state->ops->abort) {
1730             state->ops->abort(state);
1731         }
1732     }
1733 exit:
1734     QSIMPLEQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
1735         if (state->ops->clean) {
1736             state->ops->clean(state);
1737         }
1738         g_free(state);
1739     }
1740 }
1741 
1742 
1743 static void eject_device(BlockBackend *blk, int force, Error **errp)
1744 {
1745     BlockDriverState *bs = blk_bs(blk);
1746     AioContext *aio_context;
1747 
1748     aio_context = bdrv_get_aio_context(bs);
1749     aio_context_acquire(aio_context);
1750 
1751     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_EJECT, errp)) {
1752         goto out;
1753     }
1754     if (!blk_dev_has_removable_media(blk)) {
1755         error_setg(errp, "Device '%s' is not removable",
1756                    bdrv_get_device_name(bs));
1757         goto out;
1758     }
1759 
1760     if (blk_dev_is_medium_locked(blk) && !blk_dev_is_tray_open(blk)) {
1761         blk_dev_eject_request(blk, force);
1762         if (!force) {
1763             error_setg(errp, "Device '%s' is locked",
1764                        bdrv_get_device_name(bs));
1765             goto out;
1766         }
1767     }
1768 
1769     bdrv_close(bs);
1770 
1771 out:
1772     aio_context_release(aio_context);
1773 }
1774 
1775 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
1776 {
1777     BlockBackend *blk;
1778 
1779     blk = blk_by_name(device);
1780     if (!blk) {
1781         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1782         return;
1783     }
1784 
1785     eject_device(blk, force, errp);
1786 }
1787 
1788 void qmp_block_passwd(bool has_device, const char *device,
1789                       bool has_node_name, const char *node_name,
1790                       const char *password, Error **errp)
1791 {
1792     Error *local_err = NULL;
1793     BlockDriverState *bs;
1794     AioContext *aio_context;
1795 
1796     bs = bdrv_lookup_bs(has_device ? device : NULL,
1797                         has_node_name ? node_name : NULL,
1798                         &local_err);
1799     if (local_err) {
1800         error_propagate(errp, local_err);
1801         return;
1802     }
1803 
1804     aio_context = bdrv_get_aio_context(bs);
1805     aio_context_acquire(aio_context);
1806 
1807     bdrv_add_key(bs, password, errp);
1808 
1809     aio_context_release(aio_context);
1810 }
1811 
1812 /* Assumes AioContext is held */
1813 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename,
1814                                     int bdrv_flags, BlockDriver *drv,
1815                                     const char *password, Error **errp)
1816 {
1817     Error *local_err = NULL;
1818     int ret;
1819 
1820     ret = bdrv_open(&bs, filename, NULL, NULL, bdrv_flags, drv, &local_err);
1821     if (ret < 0) {
1822         error_propagate(errp, local_err);
1823         return;
1824     }
1825 
1826     bdrv_add_key(bs, password, errp);
1827 }
1828 
1829 void qmp_change_blockdev(const char *device, const char *filename,
1830                          const char *format, Error **errp)
1831 {
1832     BlockBackend *blk;
1833     BlockDriverState *bs;
1834     AioContext *aio_context;
1835     BlockDriver *drv = NULL;
1836     int bdrv_flags;
1837     Error *err = NULL;
1838 
1839     blk = blk_by_name(device);
1840     if (!blk) {
1841         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1842         return;
1843     }
1844     bs = blk_bs(blk);
1845 
1846     aio_context = bdrv_get_aio_context(bs);
1847     aio_context_acquire(aio_context);
1848 
1849     if (format) {
1850         drv = bdrv_find_whitelisted_format(format, bs->read_only);
1851         if (!drv) {
1852             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
1853             goto out;
1854         }
1855     }
1856 
1857     eject_device(blk, 0, &err);
1858     if (err) {
1859         error_propagate(errp, err);
1860         goto out;
1861     }
1862 
1863     bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
1864     bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
1865 
1866     qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp);
1867 
1868 out:
1869     aio_context_release(aio_context);
1870 }
1871 
1872 /* throttling disk I/O limits */
1873 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
1874                                int64_t bps_wr,
1875                                int64_t iops,
1876                                int64_t iops_rd,
1877                                int64_t iops_wr,
1878                                bool has_bps_max,
1879                                int64_t bps_max,
1880                                bool has_bps_rd_max,
1881                                int64_t bps_rd_max,
1882                                bool has_bps_wr_max,
1883                                int64_t bps_wr_max,
1884                                bool has_iops_max,
1885                                int64_t iops_max,
1886                                bool has_iops_rd_max,
1887                                int64_t iops_rd_max,
1888                                bool has_iops_wr_max,
1889                                int64_t iops_wr_max,
1890                                bool has_iops_size,
1891                                int64_t iops_size, Error **errp)
1892 {
1893     ThrottleConfig cfg;
1894     BlockDriverState *bs;
1895     BlockBackend *blk;
1896     AioContext *aio_context;
1897 
1898     blk = blk_by_name(device);
1899     if (!blk) {
1900         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1901         return;
1902     }
1903     bs = blk_bs(blk);
1904 
1905     memset(&cfg, 0, sizeof(cfg));
1906     cfg.buckets[THROTTLE_BPS_TOTAL].avg = bps;
1907     cfg.buckets[THROTTLE_BPS_READ].avg  = bps_rd;
1908     cfg.buckets[THROTTLE_BPS_WRITE].avg = bps_wr;
1909 
1910     cfg.buckets[THROTTLE_OPS_TOTAL].avg = iops;
1911     cfg.buckets[THROTTLE_OPS_READ].avg  = iops_rd;
1912     cfg.buckets[THROTTLE_OPS_WRITE].avg = iops_wr;
1913 
1914     if (has_bps_max) {
1915         cfg.buckets[THROTTLE_BPS_TOTAL].max = bps_max;
1916     }
1917     if (has_bps_rd_max) {
1918         cfg.buckets[THROTTLE_BPS_READ].max = bps_rd_max;
1919     }
1920     if (has_bps_wr_max) {
1921         cfg.buckets[THROTTLE_BPS_WRITE].max = bps_wr_max;
1922     }
1923     if (has_iops_max) {
1924         cfg.buckets[THROTTLE_OPS_TOTAL].max = iops_max;
1925     }
1926     if (has_iops_rd_max) {
1927         cfg.buckets[THROTTLE_OPS_READ].max = iops_rd_max;
1928     }
1929     if (has_iops_wr_max) {
1930         cfg.buckets[THROTTLE_OPS_WRITE].max = iops_wr_max;
1931     }
1932 
1933     if (has_iops_size) {
1934         cfg.op_size = iops_size;
1935     }
1936 
1937     if (!check_throttle_config(&cfg, errp)) {
1938         return;
1939     }
1940 
1941     aio_context = bdrv_get_aio_context(bs);
1942     aio_context_acquire(aio_context);
1943 
1944     if (!bs->io_limits_enabled && throttle_enabled(&cfg)) {
1945         bdrv_io_limits_enable(bs);
1946     } else if (bs->io_limits_enabled && !throttle_enabled(&cfg)) {
1947         bdrv_io_limits_disable(bs);
1948     }
1949 
1950     if (bs->io_limits_enabled) {
1951         bdrv_set_io_limits(bs, &cfg);
1952     }
1953 
1954     aio_context_release(aio_context);
1955 }
1956 
1957 int hmp_drive_del(Monitor *mon, const QDict *qdict, QObject **ret_data)
1958 {
1959     const char *id = qdict_get_str(qdict, "id");
1960     BlockBackend *blk;
1961     BlockDriverState *bs;
1962     AioContext *aio_context;
1963     Error *local_err = NULL;
1964 
1965     blk = blk_by_name(id);
1966     if (!blk) {
1967         error_report("Device '%s' not found", id);
1968         return -1;
1969     }
1970     bs = blk_bs(blk);
1971 
1972     if (!blk_legacy_dinfo(blk)) {
1973         error_report("Deleting device added with blockdev-add"
1974                      " is not supported");
1975         return -1;
1976     }
1977 
1978     aio_context = bdrv_get_aio_context(bs);
1979     aio_context_acquire(aio_context);
1980 
1981     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, &local_err)) {
1982         error_report_err(local_err);
1983         aio_context_release(aio_context);
1984         return -1;
1985     }
1986 
1987     /* quiesce block driver; prevent further io */
1988     bdrv_drain_all();
1989     bdrv_flush(bs);
1990     bdrv_close(bs);
1991 
1992     /* if we have a device attached to this BlockDriverState
1993      * then we need to make the drive anonymous until the device
1994      * can be removed.  If this is a drive with no device backing
1995      * then we can just get rid of the block driver state right here.
1996      */
1997     if (blk_get_attached_dev(blk)) {
1998         blk_hide_on_behalf_of_hmp_drive_del(blk);
1999         /* Further I/O must not pause the guest */
2000         bdrv_set_on_error(bs, BLOCKDEV_ON_ERROR_REPORT,
2001                           BLOCKDEV_ON_ERROR_REPORT);
2002     } else {
2003         blk_unref(blk);
2004     }
2005 
2006     aio_context_release(aio_context);
2007     return 0;
2008 }
2009 
2010 void qmp_block_resize(bool has_device, const char *device,
2011                       bool has_node_name, const char *node_name,
2012                       int64_t size, Error **errp)
2013 {
2014     Error *local_err = NULL;
2015     BlockDriverState *bs;
2016     AioContext *aio_context;
2017     int ret;
2018 
2019     bs = bdrv_lookup_bs(has_device ? device : NULL,
2020                         has_node_name ? node_name : NULL,
2021                         &local_err);
2022     if (local_err) {
2023         error_propagate(errp, local_err);
2024         return;
2025     }
2026 
2027     aio_context = bdrv_get_aio_context(bs);
2028     aio_context_acquire(aio_context);
2029 
2030     if (!bdrv_is_first_non_filter(bs)) {
2031         error_set(errp, QERR_FEATURE_DISABLED, "resize");
2032         goto out;
2033     }
2034 
2035     if (size < 0) {
2036         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2037         goto out;
2038     }
2039 
2040     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2041         error_set(errp, QERR_DEVICE_IN_USE, device);
2042         goto out;
2043     }
2044 
2045     /* complete all in-flight operations before resizing the device */
2046     bdrv_drain_all();
2047 
2048     ret = bdrv_truncate(bs, size);
2049     switch (ret) {
2050     case 0:
2051         break;
2052     case -ENOMEDIUM:
2053         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2054         break;
2055     case -ENOTSUP:
2056         error_set(errp, QERR_UNSUPPORTED);
2057         break;
2058     case -EACCES:
2059         error_setg(errp, "Device '%s' is read only", device);
2060         break;
2061     case -EBUSY:
2062         error_set(errp, QERR_DEVICE_IN_USE, device);
2063         break;
2064     default:
2065         error_setg_errno(errp, -ret, "Could not resize");
2066         break;
2067     }
2068 
2069 out:
2070     aio_context_release(aio_context);
2071 }
2072 
2073 static void block_job_cb(void *opaque, int ret)
2074 {
2075     /* Note that this function may be executed from another AioContext besides
2076      * the QEMU main loop.  If you need to access anything that assumes the
2077      * QEMU global mutex, use a BH or introduce a mutex.
2078      */
2079 
2080     BlockDriverState *bs = opaque;
2081     const char *msg = NULL;
2082 
2083     trace_block_job_cb(bs, bs->job, ret);
2084 
2085     assert(bs->job);
2086 
2087     if (ret < 0) {
2088         msg = strerror(-ret);
2089     }
2090 
2091     if (block_job_is_cancelled(bs->job)) {
2092         block_job_event_cancelled(bs->job);
2093     } else {
2094         block_job_event_completed(bs->job, msg);
2095     }
2096 
2097     bdrv_put_ref_bh_schedule(bs);
2098 }
2099 
2100 void qmp_block_stream(const char *device,
2101                       bool has_base, const char *base,
2102                       bool has_backing_file, const char *backing_file,
2103                       bool has_speed, int64_t speed,
2104                       bool has_on_error, BlockdevOnError on_error,
2105                       Error **errp)
2106 {
2107     BlockBackend *blk;
2108     BlockDriverState *bs;
2109     BlockDriverState *base_bs = NULL;
2110     AioContext *aio_context;
2111     Error *local_err = NULL;
2112     const char *base_name = NULL;
2113 
2114     if (!has_on_error) {
2115         on_error = BLOCKDEV_ON_ERROR_REPORT;
2116     }
2117 
2118     blk = blk_by_name(device);
2119     if (!blk) {
2120         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2121         return;
2122     }
2123     bs = blk_bs(blk);
2124 
2125     aio_context = bdrv_get_aio_context(bs);
2126     aio_context_acquire(aio_context);
2127 
2128     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_STREAM, errp)) {
2129         goto out;
2130     }
2131 
2132     if (has_base) {
2133         base_bs = bdrv_find_backing_image(bs, base);
2134         if (base_bs == NULL) {
2135             error_set(errp, QERR_BASE_NOT_FOUND, base);
2136             goto out;
2137         }
2138         assert(bdrv_get_aio_context(base_bs) == aio_context);
2139         base_name = base;
2140     }
2141 
2142     /* if we are streaming the entire chain, the result will have no backing
2143      * file, and specifying one is therefore an error */
2144     if (base_bs == NULL && has_backing_file) {
2145         error_setg(errp, "backing file specified, but streaming the "
2146                          "entire chain");
2147         goto out;
2148     }
2149 
2150     /* backing_file string overrides base bs filename */
2151     base_name = has_backing_file ? backing_file : base_name;
2152 
2153     stream_start(bs, base_bs, base_name, has_speed ? speed : 0,
2154                  on_error, block_job_cb, bs, &local_err);
2155     if (local_err) {
2156         error_propagate(errp, local_err);
2157         goto out;
2158     }
2159 
2160     trace_qmp_block_stream(bs, bs->job);
2161 
2162 out:
2163     aio_context_release(aio_context);
2164 }
2165 
2166 void qmp_block_commit(const char *device,
2167                       bool has_base, const char *base,
2168                       bool has_top, const char *top,
2169                       bool has_backing_file, const char *backing_file,
2170                       bool has_speed, int64_t speed,
2171                       Error **errp)
2172 {
2173     BlockBackend *blk;
2174     BlockDriverState *bs;
2175     BlockDriverState *base_bs, *top_bs;
2176     AioContext *aio_context;
2177     Error *local_err = NULL;
2178     /* This will be part of the QMP command, if/when the
2179      * BlockdevOnError change for blkmirror makes it in
2180      */
2181     BlockdevOnError on_error = BLOCKDEV_ON_ERROR_REPORT;
2182 
2183     if (!has_speed) {
2184         speed = 0;
2185     }
2186 
2187     /* Important Note:
2188      *  libvirt relies on the DeviceNotFound error class in order to probe for
2189      *  live commit feature versions; for this to work, we must make sure to
2190      *  perform the device lookup before any generic errors that may occur in a
2191      *  scenario in which all optional arguments are omitted. */
2192     blk = blk_by_name(device);
2193     if (!blk) {
2194         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2195         return;
2196     }
2197     bs = blk_bs(blk);
2198 
2199     aio_context = bdrv_get_aio_context(bs);
2200     aio_context_acquire(aio_context);
2201 
2202     /* drain all i/o before commits */
2203     bdrv_drain_all();
2204 
2205     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2206         goto out;
2207     }
2208 
2209     /* default top_bs is the active layer */
2210     top_bs = bs;
2211 
2212     if (has_top && top) {
2213         if (strcmp(bs->filename, top) != 0) {
2214             top_bs = bdrv_find_backing_image(bs, top);
2215         }
2216     }
2217 
2218     if (top_bs == NULL) {
2219         error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2220         goto out;
2221     }
2222 
2223     assert(bdrv_get_aio_context(top_bs) == aio_context);
2224 
2225     if (has_base && base) {
2226         base_bs = bdrv_find_backing_image(top_bs, base);
2227     } else {
2228         base_bs = bdrv_find_base(top_bs);
2229     }
2230 
2231     if (base_bs == NULL) {
2232         error_set(errp, QERR_BASE_NOT_FOUND, base ? base : "NULL");
2233         goto out;
2234     }
2235 
2236     assert(bdrv_get_aio_context(base_bs) == aio_context);
2237 
2238     if (bdrv_op_is_blocked(base_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2239         goto out;
2240     }
2241 
2242     /* Do not allow attempts to commit an image into itself */
2243     if (top_bs == base_bs) {
2244         error_setg(errp, "cannot commit an image into itself");
2245         goto out;
2246     }
2247 
2248     if (top_bs == bs) {
2249         if (has_backing_file) {
2250             error_setg(errp, "'backing-file' specified,"
2251                              " but 'top' is the active layer");
2252             goto out;
2253         }
2254         commit_active_start(bs, base_bs, speed, on_error, block_job_cb,
2255                             bs, &local_err);
2256     } else {
2257         commit_start(bs, base_bs, top_bs, speed, on_error, block_job_cb, bs,
2258                      has_backing_file ? backing_file : NULL, &local_err);
2259     }
2260     if (local_err != NULL) {
2261         error_propagate(errp, local_err);
2262         goto out;
2263     }
2264 
2265 out:
2266     aio_context_release(aio_context);
2267 }
2268 
2269 void qmp_drive_backup(const char *device, const char *target,
2270                       bool has_format, const char *format,
2271                       enum MirrorSyncMode sync,
2272                       bool has_mode, enum NewImageMode mode,
2273                       bool has_speed, int64_t speed,
2274                       bool has_on_source_error, BlockdevOnError on_source_error,
2275                       bool has_on_target_error, BlockdevOnError on_target_error,
2276                       Error **errp)
2277 {
2278     BlockBackend *blk;
2279     BlockDriverState *bs;
2280     BlockDriverState *target_bs;
2281     BlockDriverState *source = NULL;
2282     AioContext *aio_context;
2283     BlockDriver *drv = NULL;
2284     Error *local_err = NULL;
2285     int flags;
2286     int64_t size;
2287     int ret;
2288 
2289     if (!has_speed) {
2290         speed = 0;
2291     }
2292     if (!has_on_source_error) {
2293         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2294     }
2295     if (!has_on_target_error) {
2296         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2297     }
2298     if (!has_mode) {
2299         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2300     }
2301 
2302     blk = blk_by_name(device);
2303     if (!blk) {
2304         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2305         return;
2306     }
2307     bs = blk_bs(blk);
2308 
2309     aio_context = bdrv_get_aio_context(bs);
2310     aio_context_acquire(aio_context);
2311 
2312     /* Although backup_run has this check too, we need to use bs->drv below, so
2313      * do an early check redundantly. */
2314     if (!bdrv_is_inserted(bs)) {
2315         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2316         goto out;
2317     }
2318 
2319     if (!has_format) {
2320         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
2321     }
2322     if (format) {
2323         drv = bdrv_find_format(format);
2324         if (!drv) {
2325             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
2326             goto out;
2327         }
2328     }
2329 
2330     /* Early check to avoid creating target */
2331     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
2332         goto out;
2333     }
2334 
2335     flags = bs->open_flags | BDRV_O_RDWR;
2336 
2337     /* See if we have a backing HD we can use to create our new image
2338      * on top of. */
2339     if (sync == MIRROR_SYNC_MODE_TOP) {
2340         source = bs->backing_hd;
2341         if (!source) {
2342             sync = MIRROR_SYNC_MODE_FULL;
2343         }
2344     }
2345     if (sync == MIRROR_SYNC_MODE_NONE) {
2346         source = bs;
2347     }
2348 
2349     size = bdrv_getlength(bs);
2350     if (size < 0) {
2351         error_setg_errno(errp, -size, "bdrv_getlength failed");
2352         goto out;
2353     }
2354 
2355     if (mode != NEW_IMAGE_MODE_EXISTING) {
2356         assert(format && drv);
2357         if (source) {
2358             bdrv_img_create(target, format, source->filename,
2359                             source->drv->format_name, NULL,
2360                             size, flags, &local_err, false);
2361         } else {
2362             bdrv_img_create(target, format, NULL, NULL, NULL,
2363                             size, flags, &local_err, false);
2364         }
2365     }
2366 
2367     if (local_err) {
2368         error_propagate(errp, local_err);
2369         goto out;
2370     }
2371 
2372     target_bs = NULL;
2373     ret = bdrv_open(&target_bs, target, NULL, NULL, flags, drv, &local_err);
2374     if (ret < 0) {
2375         error_propagate(errp, local_err);
2376         goto out;
2377     }
2378 
2379     bdrv_set_aio_context(target_bs, aio_context);
2380 
2381     backup_start(bs, target_bs, speed, sync, on_source_error, on_target_error,
2382                  block_job_cb, bs, &local_err);
2383     if (local_err != NULL) {
2384         bdrv_unref(target_bs);
2385         error_propagate(errp, local_err);
2386         goto out;
2387     }
2388 
2389 out:
2390     aio_context_release(aio_context);
2391 }
2392 
2393 BlockDeviceInfoList *qmp_query_named_block_nodes(Error **errp)
2394 {
2395     return bdrv_named_nodes_list(errp);
2396 }
2397 
2398 void qmp_blockdev_backup(const char *device, const char *target,
2399                          enum MirrorSyncMode sync,
2400                          bool has_speed, int64_t speed,
2401                          bool has_on_source_error,
2402                          BlockdevOnError on_source_error,
2403                          bool has_on_target_error,
2404                          BlockdevOnError on_target_error,
2405                          Error **errp)
2406 {
2407     BlockBackend *blk;
2408     BlockDriverState *bs;
2409     BlockDriverState *target_bs;
2410     Error *local_err = NULL;
2411     AioContext *aio_context;
2412 
2413     if (!has_speed) {
2414         speed = 0;
2415     }
2416     if (!has_on_source_error) {
2417         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2418     }
2419     if (!has_on_target_error) {
2420         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2421     }
2422 
2423     blk = blk_by_name(device);
2424     if (!blk) {
2425         error_setg(errp, "Device '%s' not found", device);
2426         return;
2427     }
2428     bs = blk_bs(blk);
2429 
2430     aio_context = bdrv_get_aio_context(bs);
2431     aio_context_acquire(aio_context);
2432 
2433     blk = blk_by_name(target);
2434     if (!blk) {
2435         error_setg(errp, "Device '%s' not found", target);
2436         goto out;
2437     }
2438     target_bs = blk_bs(blk);
2439 
2440     bdrv_ref(target_bs);
2441     bdrv_set_aio_context(target_bs, aio_context);
2442     backup_start(bs, target_bs, speed, sync, on_source_error, on_target_error,
2443                  block_job_cb, bs, &local_err);
2444     if (local_err != NULL) {
2445         bdrv_unref(target_bs);
2446         error_propagate(errp, local_err);
2447     }
2448 out:
2449     aio_context_release(aio_context);
2450 }
2451 
2452 #define DEFAULT_MIRROR_BUF_SIZE   (10 << 20)
2453 
2454 void qmp_drive_mirror(const char *device, const char *target,
2455                       bool has_format, const char *format,
2456                       bool has_node_name, const char *node_name,
2457                       bool has_replaces, const char *replaces,
2458                       enum MirrorSyncMode sync,
2459                       bool has_mode, enum NewImageMode mode,
2460                       bool has_speed, int64_t speed,
2461                       bool has_granularity, uint32_t granularity,
2462                       bool has_buf_size, int64_t buf_size,
2463                       bool has_on_source_error, BlockdevOnError on_source_error,
2464                       bool has_on_target_error, BlockdevOnError on_target_error,
2465                       Error **errp)
2466 {
2467     BlockBackend *blk;
2468     BlockDriverState *bs;
2469     BlockDriverState *source, *target_bs;
2470     AioContext *aio_context;
2471     BlockDriver *drv = NULL;
2472     Error *local_err = NULL;
2473     QDict *options = NULL;
2474     int flags;
2475     int64_t size;
2476     int ret;
2477 
2478     if (!has_speed) {
2479         speed = 0;
2480     }
2481     if (!has_on_source_error) {
2482         on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2483     }
2484     if (!has_on_target_error) {
2485         on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2486     }
2487     if (!has_mode) {
2488         mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2489     }
2490     if (!has_granularity) {
2491         granularity = 0;
2492     }
2493     if (!has_buf_size) {
2494         buf_size = DEFAULT_MIRROR_BUF_SIZE;
2495     }
2496 
2497     if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2498         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2499                   "a value in range [512B, 64MB]");
2500         return;
2501     }
2502     if (granularity & (granularity - 1)) {
2503         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "granularity", "power of 2");
2504         return;
2505     }
2506 
2507     blk = blk_by_name(device);
2508     if (!blk) {
2509         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2510         return;
2511     }
2512     bs = blk_bs(blk);
2513 
2514     aio_context = bdrv_get_aio_context(bs);
2515     aio_context_acquire(aio_context);
2516 
2517     if (!bdrv_is_inserted(bs)) {
2518         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
2519         goto out;
2520     }
2521 
2522     if (!has_format) {
2523         format = mode == NEW_IMAGE_MODE_EXISTING ? NULL : bs->drv->format_name;
2524     }
2525     if (format) {
2526         drv = bdrv_find_format(format);
2527         if (!drv) {
2528             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
2529             goto out;
2530         }
2531     }
2532 
2533     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR, errp)) {
2534         goto out;
2535     }
2536 
2537     flags = bs->open_flags | BDRV_O_RDWR;
2538     source = bs->backing_hd;
2539     if (!source && sync == MIRROR_SYNC_MODE_TOP) {
2540         sync = MIRROR_SYNC_MODE_FULL;
2541     }
2542     if (sync == MIRROR_SYNC_MODE_NONE) {
2543         source = bs;
2544     }
2545 
2546     size = bdrv_getlength(bs);
2547     if (size < 0) {
2548         error_setg_errno(errp, -size, "bdrv_getlength failed");
2549         goto out;
2550     }
2551 
2552     if (has_replaces) {
2553         BlockDriverState *to_replace_bs;
2554         AioContext *replace_aio_context;
2555         int64_t replace_size;
2556 
2557         if (!has_node_name) {
2558             error_setg(errp, "a node-name must be provided when replacing a"
2559                              " named node of the graph");
2560             goto out;
2561         }
2562 
2563         to_replace_bs = check_to_replace_node(replaces, &local_err);
2564 
2565         if (!to_replace_bs) {
2566             error_propagate(errp, local_err);
2567             goto out;
2568         }
2569 
2570         replace_aio_context = bdrv_get_aio_context(to_replace_bs);
2571         aio_context_acquire(replace_aio_context);
2572         replace_size = bdrv_getlength(to_replace_bs);
2573         aio_context_release(replace_aio_context);
2574 
2575         if (size != replace_size) {
2576             error_setg(errp, "cannot replace image with a mirror image of "
2577                              "different size");
2578             goto out;
2579         }
2580     }
2581 
2582     if ((sync == MIRROR_SYNC_MODE_FULL || !source)
2583         && mode != NEW_IMAGE_MODE_EXISTING)
2584     {
2585         /* create new image w/o backing file */
2586         assert(format && drv);
2587         bdrv_img_create(target, format,
2588                         NULL, NULL, NULL, size, flags, &local_err, false);
2589     } else {
2590         switch (mode) {
2591         case NEW_IMAGE_MODE_EXISTING:
2592             break;
2593         case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
2594             /* create new image with backing file */
2595             bdrv_img_create(target, format,
2596                             source->filename,
2597                             source->drv->format_name,
2598                             NULL, size, flags, &local_err, false);
2599             break;
2600         default:
2601             abort();
2602         }
2603     }
2604 
2605     if (local_err) {
2606         error_propagate(errp, local_err);
2607         goto out;
2608     }
2609 
2610     if (has_node_name) {
2611         options = qdict_new();
2612         qdict_put(options, "node-name", qstring_from_str(node_name));
2613     }
2614 
2615     /* Mirroring takes care of copy-on-write using the source's backing
2616      * file.
2617      */
2618     target_bs = NULL;
2619     ret = bdrv_open(&target_bs, target, NULL, options,
2620                     flags | BDRV_O_NO_BACKING, drv, &local_err);
2621     if (ret < 0) {
2622         error_propagate(errp, local_err);
2623         goto out;
2624     }
2625 
2626     bdrv_set_aio_context(target_bs, aio_context);
2627 
2628     /* pass the node name to replace to mirror start since it's loose coupling
2629      * and will allow to check whether the node still exist at mirror completion
2630      */
2631     mirror_start(bs, target_bs,
2632                  has_replaces ? replaces : NULL,
2633                  speed, granularity, buf_size, sync,
2634                  on_source_error, on_target_error,
2635                  block_job_cb, bs, &local_err);
2636     if (local_err != NULL) {
2637         bdrv_unref(target_bs);
2638         error_propagate(errp, local_err);
2639         goto out;
2640     }
2641 
2642 out:
2643     aio_context_release(aio_context);
2644 }
2645 
2646 /* Get the block job for a given device name and acquire its AioContext */
2647 static BlockJob *find_block_job(const char *device, AioContext **aio_context,
2648                                 Error **errp)
2649 {
2650     BlockBackend *blk;
2651     BlockDriverState *bs;
2652 
2653     blk = blk_by_name(device);
2654     if (!blk) {
2655         goto notfound;
2656     }
2657     bs = blk_bs(blk);
2658 
2659     *aio_context = bdrv_get_aio_context(bs);
2660     aio_context_acquire(*aio_context);
2661 
2662     if (!bs->job) {
2663         aio_context_release(*aio_context);
2664         goto notfound;
2665     }
2666 
2667     return bs->job;
2668 
2669 notfound:
2670     error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
2671               "No active block job on device '%s'", device);
2672     *aio_context = NULL;
2673     return NULL;
2674 }
2675 
2676 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
2677 {
2678     AioContext *aio_context;
2679     BlockJob *job = find_block_job(device, &aio_context, errp);
2680 
2681     if (!job) {
2682         return;
2683     }
2684 
2685     block_job_set_speed(job, speed, errp);
2686     aio_context_release(aio_context);
2687 }
2688 
2689 void qmp_block_job_cancel(const char *device,
2690                           bool has_force, bool force, Error **errp)
2691 {
2692     AioContext *aio_context;
2693     BlockJob *job = find_block_job(device, &aio_context, errp);
2694 
2695     if (!job) {
2696         return;
2697     }
2698 
2699     if (!has_force) {
2700         force = false;
2701     }
2702 
2703     if (job->user_paused && !force) {
2704         error_setg(errp, "The block job for device '%s' is currently paused",
2705                    device);
2706         goto out;
2707     }
2708 
2709     trace_qmp_block_job_cancel(job);
2710     block_job_cancel(job);
2711 out:
2712     aio_context_release(aio_context);
2713 }
2714 
2715 void qmp_block_job_pause(const char *device, Error **errp)
2716 {
2717     AioContext *aio_context;
2718     BlockJob *job = find_block_job(device, &aio_context, errp);
2719 
2720     if (!job || job->user_paused) {
2721         return;
2722     }
2723 
2724     job->user_paused = true;
2725     trace_qmp_block_job_pause(job);
2726     block_job_pause(job);
2727     aio_context_release(aio_context);
2728 }
2729 
2730 void qmp_block_job_resume(const char *device, Error **errp)
2731 {
2732     AioContext *aio_context;
2733     BlockJob *job = find_block_job(device, &aio_context, errp);
2734 
2735     if (!job || !job->user_paused) {
2736         return;
2737     }
2738 
2739     job->user_paused = false;
2740     trace_qmp_block_job_resume(job);
2741     block_job_resume(job);
2742     aio_context_release(aio_context);
2743 }
2744 
2745 void qmp_block_job_complete(const char *device, Error **errp)
2746 {
2747     AioContext *aio_context;
2748     BlockJob *job = find_block_job(device, &aio_context, errp);
2749 
2750     if (!job) {
2751         return;
2752     }
2753 
2754     trace_qmp_block_job_complete(job);
2755     block_job_complete(job, errp);
2756     aio_context_release(aio_context);
2757 }
2758 
2759 void qmp_change_backing_file(const char *device,
2760                              const char *image_node_name,
2761                              const char *backing_file,
2762                              Error **errp)
2763 {
2764     BlockBackend *blk;
2765     BlockDriverState *bs = NULL;
2766     AioContext *aio_context;
2767     BlockDriverState *image_bs = NULL;
2768     Error *local_err = NULL;
2769     bool ro;
2770     int open_flags;
2771     int ret;
2772 
2773     blk = blk_by_name(device);
2774     if (!blk) {
2775         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
2776         return;
2777     }
2778     bs = blk_bs(blk);
2779 
2780     aio_context = bdrv_get_aio_context(bs);
2781     aio_context_acquire(aio_context);
2782 
2783     image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
2784     if (local_err) {
2785         error_propagate(errp, local_err);
2786         goto out;
2787     }
2788 
2789     if (!image_bs) {
2790         error_setg(errp, "image file not found");
2791         goto out;
2792     }
2793 
2794     if (bdrv_find_base(image_bs) == image_bs) {
2795         error_setg(errp, "not allowing backing file change on an image "
2796                          "without a backing file");
2797         goto out;
2798     }
2799 
2800     /* even though we are not necessarily operating on bs, we need it to
2801      * determine if block ops are currently prohibited on the chain */
2802     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
2803         goto out;
2804     }
2805 
2806     /* final sanity check */
2807     if (!bdrv_chain_contains(bs, image_bs)) {
2808         error_setg(errp, "'%s' and image file are not in the same chain",
2809                    device);
2810         goto out;
2811     }
2812 
2813     /* if not r/w, reopen to make r/w */
2814     open_flags = image_bs->open_flags;
2815     ro = bdrv_is_read_only(image_bs);
2816 
2817     if (ro) {
2818         bdrv_reopen(image_bs, open_flags | BDRV_O_RDWR, &local_err);
2819         if (local_err) {
2820             error_propagate(errp, local_err);
2821             goto out;
2822         }
2823     }
2824 
2825     ret = bdrv_change_backing_file(image_bs, backing_file,
2826                                image_bs->drv ? image_bs->drv->format_name : "");
2827 
2828     if (ret < 0) {
2829         error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
2830                          backing_file);
2831         /* don't exit here, so we can try to restore open flags if
2832          * appropriate */
2833     }
2834 
2835     if (ro) {
2836         bdrv_reopen(image_bs, open_flags, &local_err);
2837         if (local_err) {
2838             error_propagate(errp, local_err); /* will preserve prior errp */
2839         }
2840     }
2841 
2842 out:
2843     aio_context_release(aio_context);
2844 }
2845 
2846 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
2847 {
2848     QmpOutputVisitor *ov = qmp_output_visitor_new();
2849     BlockBackend *blk;
2850     QObject *obj;
2851     QDict *qdict;
2852     Error *local_err = NULL;
2853 
2854     /* Require an ID in the top level */
2855     if (!options->has_id) {
2856         error_setg(errp, "Block device needs an ID");
2857         goto fail;
2858     }
2859 
2860     /* TODO Sort it out in raw-posix and drive_new(): Reject aio=native with
2861      * cache.direct=false instead of silently switching to aio=threads, except
2862      * when called from drive_new().
2863      *
2864      * For now, simply forbidding the combination for all drivers will do. */
2865     if (options->has_aio && options->aio == BLOCKDEV_AIO_OPTIONS_NATIVE) {
2866         bool direct = options->has_cache &&
2867                       options->cache->has_direct &&
2868                       options->cache->direct;
2869         if (!direct) {
2870             error_setg(errp, "aio=native requires cache.direct=true");
2871             goto fail;
2872         }
2873     }
2874 
2875     visit_type_BlockdevOptions(qmp_output_get_visitor(ov),
2876                                &options, NULL, &local_err);
2877     if (local_err) {
2878         error_propagate(errp, local_err);
2879         goto fail;
2880     }
2881 
2882     obj = qmp_output_get_qobject(ov);
2883     qdict = qobject_to_qdict(obj);
2884 
2885     qdict_flatten(qdict);
2886 
2887     blk = blockdev_init(NULL, qdict, &local_err);
2888     if (local_err) {
2889         error_propagate(errp, local_err);
2890         goto fail;
2891     }
2892 
2893     if (bdrv_key_required(blk_bs(blk))) {
2894         blk_unref(blk);
2895         error_setg(errp, "blockdev-add doesn't support encrypted devices");
2896         goto fail;
2897     }
2898 
2899 fail:
2900     qmp_output_visitor_cleanup(ov);
2901 }
2902 
2903 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
2904 {
2905     BlockJobInfoList *head = NULL, **p_next = &head;
2906     BlockDriverState *bs;
2907 
2908     for (bs = bdrv_next(NULL); bs; bs = bdrv_next(bs)) {
2909         AioContext *aio_context = bdrv_get_aio_context(bs);
2910 
2911         aio_context_acquire(aio_context);
2912 
2913         if (bs->job) {
2914             BlockJobInfoList *elem = g_new0(BlockJobInfoList, 1);
2915             elem->value = block_job_query(bs->job);
2916             *p_next = elem;
2917             p_next = &elem->next;
2918         }
2919 
2920         aio_context_release(aio_context);
2921     }
2922 
2923     return head;
2924 }
2925 
2926 QemuOptsList qemu_common_drive_opts = {
2927     .name = "drive",
2928     .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
2929     .desc = {
2930         {
2931             .name = "snapshot",
2932             .type = QEMU_OPT_BOOL,
2933             .help = "enable/disable snapshot mode",
2934         },{
2935             .name = "discard",
2936             .type = QEMU_OPT_STRING,
2937             .help = "discard operation (ignore/off, unmap/on)",
2938         },{
2939             .name = "cache.writeback",
2940             .type = QEMU_OPT_BOOL,
2941             .help = "enables writeback mode for any caches",
2942         },{
2943             .name = "cache.direct",
2944             .type = QEMU_OPT_BOOL,
2945             .help = "enables use of O_DIRECT (bypass the host page cache)",
2946         },{
2947             .name = "cache.no-flush",
2948             .type = QEMU_OPT_BOOL,
2949             .help = "ignore any flush requests for the device",
2950         },{
2951             .name = "aio",
2952             .type = QEMU_OPT_STRING,
2953             .help = "host AIO implementation (threads, native)",
2954         },{
2955             .name = "format",
2956             .type = QEMU_OPT_STRING,
2957             .help = "disk format (raw, qcow2, ...)",
2958         },{
2959             .name = "rerror",
2960             .type = QEMU_OPT_STRING,
2961             .help = "read error action",
2962         },{
2963             .name = "werror",
2964             .type = QEMU_OPT_STRING,
2965             .help = "write error action",
2966         },{
2967             .name = "read-only",
2968             .type = QEMU_OPT_BOOL,
2969             .help = "open drive file as read-only",
2970         },{
2971             .name = "throttling.iops-total",
2972             .type = QEMU_OPT_NUMBER,
2973             .help = "limit total I/O operations per second",
2974         },{
2975             .name = "throttling.iops-read",
2976             .type = QEMU_OPT_NUMBER,
2977             .help = "limit read operations per second",
2978         },{
2979             .name = "throttling.iops-write",
2980             .type = QEMU_OPT_NUMBER,
2981             .help = "limit write operations per second",
2982         },{
2983             .name = "throttling.bps-total",
2984             .type = QEMU_OPT_NUMBER,
2985             .help = "limit total bytes per second",
2986         },{
2987             .name = "throttling.bps-read",
2988             .type = QEMU_OPT_NUMBER,
2989             .help = "limit read bytes per second",
2990         },{
2991             .name = "throttling.bps-write",
2992             .type = QEMU_OPT_NUMBER,
2993             .help = "limit write bytes per second",
2994         },{
2995             .name = "throttling.iops-total-max",
2996             .type = QEMU_OPT_NUMBER,
2997             .help = "I/O operations burst",
2998         },{
2999             .name = "throttling.iops-read-max",
3000             .type = QEMU_OPT_NUMBER,
3001             .help = "I/O operations read burst",
3002         },{
3003             .name = "throttling.iops-write-max",
3004             .type = QEMU_OPT_NUMBER,
3005             .help = "I/O operations write burst",
3006         },{
3007             .name = "throttling.bps-total-max",
3008             .type = QEMU_OPT_NUMBER,
3009             .help = "total bytes burst",
3010         },{
3011             .name = "throttling.bps-read-max",
3012             .type = QEMU_OPT_NUMBER,
3013             .help = "total bytes read burst",
3014         },{
3015             .name = "throttling.bps-write-max",
3016             .type = QEMU_OPT_NUMBER,
3017             .help = "total bytes write burst",
3018         },{
3019             .name = "throttling.iops-size",
3020             .type = QEMU_OPT_NUMBER,
3021             .help = "when limiting by iops max size of an I/O in bytes",
3022         },{
3023             .name = "copy-on-read",
3024             .type = QEMU_OPT_BOOL,
3025             .help = "copy read data from backing file into image file",
3026         },{
3027             .name = "detect-zeroes",
3028             .type = QEMU_OPT_STRING,
3029             .help = "try to optimize zero writes (off, on, unmap)",
3030         },
3031         { /* end of list */ }
3032     },
3033 };
3034 
3035 QemuOptsList qemu_drive_opts = {
3036     .name = "drive",
3037     .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3038     .desc = {
3039         /*
3040          * no elements => accept any params
3041          * validation will happen later
3042          */
3043         { /* end of list */ }
3044     },
3045 };
3046