Merge tag 'hw-misc-20240309' of https://github.com/philmd/qemu into staging
[qemu/ar7.git] / blockdev.c
blobd8fb3399f5683f76fb6b56a839286039845e9c6e
1 /*
2 * QEMU host block devices
4 * Copyright (c) 2003-2008 Fabrice Bellard
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.
9 * This file incorporates work covered by the following copyright and
10 * permission notice:
12 * Copyright (c) 2003-2008 Fabrice Bellard
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:
21 * The above copyright notice and this permission notice shall be included in
22 * all copies or substantial portions of the Software.
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.
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/dirty-bitmap.h"
39 #include "block/qdict.h"
40 #include "block/throttle-groups.h"
41 #include "monitor/monitor.h"
42 #include "qemu/error-report.h"
43 #include "qemu/option.h"
44 #include "qemu/qemu-print.h"
45 #include "qemu/config-file.h"
46 #include "qapi/qapi-commands-block.h"
47 #include "qapi/qapi-commands-transaction.h"
48 #include "qapi/qapi-visit-block-core.h"
49 #include "qapi/qmp/qdict.h"
50 #include "qapi/qmp/qnum.h"
51 #include "qapi/qmp/qstring.h"
52 #include "qapi/error.h"
53 #include "qapi/qmp/qerror.h"
54 #include "qapi/qmp/qlist.h"
55 #include "qapi/qobject-output-visitor.h"
56 #include "sysemu/sysemu.h"
57 #include "sysemu/iothread.h"
58 #include "block/block_int.h"
59 #include "block/trace.h"
60 #include "sysemu/runstate.h"
61 #include "sysemu/replay.h"
62 #include "qemu/cutils.h"
63 #include "qemu/help_option.h"
64 #include "qemu/main-loop.h"
65 #include "qemu/throttle-options.h"
67 /* Protected by BQL */
68 QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
69 QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
71 void bdrv_set_monitor_owned(BlockDriverState *bs)
73 GLOBAL_STATE_CODE();
74 QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
77 static const char *const if_name[IF_COUNT] = {
78 [IF_NONE] = "none",
79 [IF_IDE] = "ide",
80 [IF_SCSI] = "scsi",
81 [IF_FLOPPY] = "floppy",
82 [IF_PFLASH] = "pflash",
83 [IF_MTD] = "mtd",
84 [IF_SD] = "sd",
85 [IF_VIRTIO] = "virtio",
86 [IF_XEN] = "xen",
89 static int if_max_devs[IF_COUNT] = {
91 * Do not change these numbers! They govern how drive option
92 * index maps to unit and bus. That mapping is ABI.
94 * All controllers used to implement if=T drives need to support
95 * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
96 * Otherwise, some index values map to "impossible" bus, unit
97 * values.
99 * For instance, if you change [IF_SCSI] to 255, -drive
100 * if=scsi,index=12 no longer means bus=1,unit=5, but
101 * bus=0,unit=12. With an lsi53c895a controller (7 units max),
102 * the drive can't be set up. Regression.
104 [IF_IDE] = 2,
105 [IF_SCSI] = 7,
109 * Boards may call this to offer board-by-board overrides
110 * of the default, global values.
112 void override_max_devs(BlockInterfaceType type, int max_devs)
114 BlockBackend *blk;
115 DriveInfo *dinfo;
117 GLOBAL_STATE_CODE();
119 if (max_devs <= 0) {
120 return;
123 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
124 dinfo = blk_legacy_dinfo(blk);
125 if (dinfo->type == type) {
126 fprintf(stderr, "Cannot override units-per-bus property of"
127 " the %s interface, because a drive of that type has"
128 " already been added.\n", if_name[type]);
129 g_assert_not_reached();
133 if_max_devs[type] = max_devs;
137 * We automatically delete the drive when a device using it gets
138 * unplugged. Questionable feature, but we can't just drop it.
139 * Device models call blockdev_mark_auto_del() to schedule the
140 * automatic deletion, and generic qdev code calls blockdev_auto_del()
141 * when deletion is actually safe.
143 void blockdev_mark_auto_del(BlockBackend *blk)
145 DriveInfo *dinfo = blk_legacy_dinfo(blk);
146 BlockJob *job;
148 GLOBAL_STATE_CODE();
150 if (!dinfo) {
151 return;
154 JOB_LOCK_GUARD();
156 do {
157 job = block_job_next_locked(NULL);
158 while (job && (job->job.cancelled ||
159 job->job.deferred_to_main_loop ||
160 !block_job_has_bdrv(job, blk_bs(blk))))
162 job = block_job_next_locked(job);
164 if (job) {
166 * This drops the job lock temporarily and polls, so we need to
167 * restart processing the list from the start after this.
169 job_cancel_locked(&job->job, false);
171 } while (job);
173 dinfo->auto_del = 1;
176 void blockdev_auto_del(BlockBackend *blk)
178 DriveInfo *dinfo = blk_legacy_dinfo(blk);
179 GLOBAL_STATE_CODE();
181 if (dinfo && dinfo->auto_del) {
182 monitor_remove_blk(blk);
183 blk_unref(blk);
187 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
189 int max_devs = if_max_devs[type];
190 return max_devs ? index / max_devs : 0;
193 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
195 int max_devs = if_max_devs[type];
196 return max_devs ? index % max_devs : index;
199 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
200 const char *optstr)
202 QemuOpts *opts;
204 GLOBAL_STATE_CODE();
206 opts = qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
207 if (!opts) {
208 return NULL;
210 if (type != IF_DEFAULT) {
211 qemu_opt_set(opts, "if", if_name[type], &error_abort);
213 if (index >= 0) {
214 qemu_opt_set_number(opts, "index", index, &error_abort);
216 if (file)
217 qemu_opt_set(opts, "file", file, &error_abort);
218 return opts;
221 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
223 BlockBackend *blk;
224 DriveInfo *dinfo;
226 GLOBAL_STATE_CODE();
228 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
229 dinfo = blk_legacy_dinfo(blk);
230 if (dinfo && dinfo->type == type
231 && dinfo->bus == bus && dinfo->unit == unit) {
232 return dinfo;
236 return NULL;
240 * Check board claimed all -drive that are meant to be claimed.
241 * Fatal error if any remain unclaimed.
243 void drive_check_orphaned(void)
245 BlockBackend *blk;
246 DriveInfo *dinfo;
247 Location loc;
248 bool orphans = false;
250 GLOBAL_STATE_CODE();
252 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
253 dinfo = blk_legacy_dinfo(blk);
255 * Ignore default drives, because we create certain default
256 * drives unconditionally, then leave them unclaimed. Not the
257 * users fault.
258 * Ignore IF_VIRTIO or IF_XEN, because it gets desugared into
259 * -device, so we can leave failing to -device.
260 * Ignore IF_NONE, because leaving unclaimed IF_NONE remains
261 * available for device_add is a feature.
263 if (dinfo->is_default || dinfo->type == IF_VIRTIO
264 || dinfo->type == IF_XEN || dinfo->type == IF_NONE) {
265 continue;
267 if (!blk_get_attached_dev(blk)) {
268 loc_push_none(&loc);
269 qemu_opts_loc_restore(dinfo->opts);
270 error_report("machine type does not support"
271 " if=%s,bus=%d,unit=%d",
272 if_name[dinfo->type], dinfo->bus, dinfo->unit);
273 loc_pop(&loc);
274 orphans = true;
278 if (orphans) {
279 exit(1);
283 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
285 GLOBAL_STATE_CODE();
286 return drive_get(type,
287 drive_index_to_bus_id(type, index),
288 drive_index_to_unit_id(type, index));
291 int drive_get_max_bus(BlockInterfaceType type)
293 int max_bus;
294 BlockBackend *blk;
295 DriveInfo *dinfo;
297 GLOBAL_STATE_CODE();
299 max_bus = -1;
300 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
301 dinfo = blk_legacy_dinfo(blk);
302 if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
303 max_bus = dinfo->bus;
306 return max_bus;
309 static void bdrv_format_print(void *opaque, const char *name)
311 qemu_printf(" %s", name);
314 typedef struct {
315 QEMUBH *bh;
316 BlockDriverState *bs;
317 } BDRVPutRefBH;
319 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
321 if (!strcmp(buf, "ignore")) {
322 return BLOCKDEV_ON_ERROR_IGNORE;
323 } else if (!is_read && !strcmp(buf, "enospc")) {
324 return BLOCKDEV_ON_ERROR_ENOSPC;
325 } else if (!strcmp(buf, "stop")) {
326 return BLOCKDEV_ON_ERROR_STOP;
327 } else if (!strcmp(buf, "report")) {
328 return BLOCKDEV_ON_ERROR_REPORT;
329 } else {
330 error_setg(errp, "'%s' invalid %s error action",
331 buf, is_read ? "read" : "write");
332 return -1;
336 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
337 Error **errp)
339 const QListEntry *entry;
340 for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
341 switch (qobject_type(entry->value)) {
343 case QTYPE_QSTRING: {
344 uint64_t length;
345 const char *str = qstring_get_str(qobject_to(QString,
346 entry->value));
347 if (parse_uint_full(str, 10, &length) == 0 &&
348 length > 0 && length <= UINT_MAX) {
349 block_acct_add_interval(stats, (unsigned) length);
350 } else {
351 error_setg(errp, "Invalid interval length: %s", str);
352 return false;
354 break;
357 case QTYPE_QNUM: {
358 int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
360 if (length > 0 && length <= UINT_MAX) {
361 block_acct_add_interval(stats, (unsigned) length);
362 } else {
363 error_setg(errp, "Invalid interval length: %" PRId64, length);
364 return false;
366 break;
369 default:
370 error_setg(errp, "The specification of stats-intervals is invalid");
371 return false;
374 return true;
377 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
379 /* All parameters but @opts are optional and may be set to NULL. */
380 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
381 const char **throttling_group, ThrottleConfig *throttle_cfg,
382 BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
384 Error *local_error = NULL;
385 const char *aio;
387 if (bdrv_flags) {
388 if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
389 *bdrv_flags |= BDRV_O_COPY_ON_READ;
392 if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
393 if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
394 error_setg(errp, "invalid aio option");
395 return;
400 /* disk I/O throttling */
401 if (throttling_group) {
402 *throttling_group = qemu_opt_get(opts, "throttling.group");
405 if (throttle_cfg) {
406 throttle_config_init(throttle_cfg);
407 throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
408 qemu_opt_get_number(opts, "throttling.bps-total", 0);
409 throttle_cfg->buckets[THROTTLE_BPS_READ].avg =
410 qemu_opt_get_number(opts, "throttling.bps-read", 0);
411 throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
412 qemu_opt_get_number(opts, "throttling.bps-write", 0);
413 throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
414 qemu_opt_get_number(opts, "throttling.iops-total", 0);
415 throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
416 qemu_opt_get_number(opts, "throttling.iops-read", 0);
417 throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
418 qemu_opt_get_number(opts, "throttling.iops-write", 0);
420 throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
421 qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
422 throttle_cfg->buckets[THROTTLE_BPS_READ].max =
423 qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
424 throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
425 qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
426 throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
427 qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
428 throttle_cfg->buckets[THROTTLE_OPS_READ].max =
429 qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
430 throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
431 qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
433 throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
434 qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
435 throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length =
436 qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
437 throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
438 qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
439 throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
440 qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
441 throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
442 qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
443 throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
444 qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
446 throttle_cfg->op_size =
447 qemu_opt_get_number(opts, "throttling.iops-size", 0);
449 if (!throttle_is_valid(throttle_cfg, errp)) {
450 return;
454 if (detect_zeroes) {
455 *detect_zeroes =
456 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
457 qemu_opt_get(opts, "detect-zeroes"),
458 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
459 &local_error);
460 if (local_error) {
461 error_propagate(errp, local_error);
462 return;
467 static OnOffAuto account_get_opt(QemuOpts *opts, const char *name)
469 if (!qemu_opt_find(opts, name)) {
470 return ON_OFF_AUTO_AUTO;
472 if (qemu_opt_get_bool(opts, name, true)) {
473 return ON_OFF_AUTO_ON;
475 return ON_OFF_AUTO_OFF;
478 /* Takes the ownership of bs_opts */
479 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
480 Error **errp)
482 const char *buf;
483 int bdrv_flags = 0;
484 int on_read_error, on_write_error;
485 OnOffAuto account_invalid, account_failed;
486 bool writethrough, read_only;
487 BlockBackend *blk;
488 BlockDriverState *bs;
489 ThrottleConfig cfg;
490 int snapshot = 0;
491 Error *error = NULL;
492 QemuOpts *opts;
493 QDict *interval_dict = NULL;
494 QList *interval_list = NULL;
495 const char *id;
496 BlockdevDetectZeroesOptions detect_zeroes =
497 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
498 const char *throttling_group = NULL;
500 /* Check common options by copying from bs_opts to opts, all other options
501 * stay in bs_opts for processing by bdrv_open(). */
502 id = qdict_get_try_str(bs_opts, "id");
503 opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
504 if (!opts) {
505 goto err_no_opts;
508 if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
509 goto early_err;
512 if (id) {
513 qdict_del(bs_opts, "id");
516 /* extract parameters */
517 snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
519 account_invalid = account_get_opt(opts, "stats-account-invalid");
520 account_failed = account_get_opt(opts, "stats-account-failed");
522 writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
524 id = qemu_opts_id(opts);
526 qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
527 qdict_array_split(interval_dict, &interval_list);
529 if (qdict_size(interval_dict) != 0) {
530 error_setg(errp, "Invalid option stats-intervals.%s",
531 qdict_first(interval_dict)->key);
532 goto early_err;
535 extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
536 &detect_zeroes, &error);
537 if (error) {
538 error_propagate(errp, error);
539 goto early_err;
542 if ((buf = qemu_opt_get(opts, "format")) != NULL) {
543 if (is_help_option(buf)) {
544 qemu_printf("Supported formats:");
545 bdrv_iterate_format(bdrv_format_print, NULL, false);
546 qemu_printf("\nSupported formats (read-only):");
547 bdrv_iterate_format(bdrv_format_print, NULL, true);
548 qemu_printf("\n");
549 goto early_err;
552 if (qdict_haskey(bs_opts, "driver")) {
553 error_setg(errp, "Cannot specify both 'driver' and 'format'");
554 goto early_err;
556 qdict_put_str(bs_opts, "driver", buf);
559 on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
560 if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
561 on_write_error = parse_block_error_action(buf, 0, &error);
562 if (error) {
563 error_propagate(errp, error);
564 goto early_err;
568 on_read_error = BLOCKDEV_ON_ERROR_REPORT;
569 if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
570 on_read_error = parse_block_error_action(buf, 1, &error);
571 if (error) {
572 error_propagate(errp, error);
573 goto early_err;
577 if (snapshot) {
578 bdrv_flags |= BDRV_O_SNAPSHOT;
581 read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
583 /* init */
584 if ((!file || !*file) && !qdict_size(bs_opts)) {
585 BlockBackendRootState *blk_rs;
587 blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
588 blk_rs = blk_get_root_state(blk);
589 blk_rs->open_flags = bdrv_flags | (read_only ? 0 : BDRV_O_RDWR);
590 blk_rs->detect_zeroes = detect_zeroes;
592 qobject_unref(bs_opts);
593 } else {
594 if (file && !*file) {
595 file = NULL;
598 /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
599 * with other callers) rather than what we want as the real defaults.
600 * Apply the defaults here instead. */
601 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
602 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
603 qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
604 read_only ? "on" : "off");
605 qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
606 assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
608 if (runstate_check(RUN_STATE_INMIGRATE)) {
609 bdrv_flags |= BDRV_O_INACTIVE;
612 blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
613 if (!blk) {
614 goto err_no_bs_opts;
616 bs = blk_bs(blk);
618 bs->detect_zeroes = detect_zeroes;
620 block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
622 if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
623 blk_unref(blk);
624 blk = NULL;
625 goto err_no_bs_opts;
629 /* disk I/O throttling */
630 if (throttle_enabled(&cfg)) {
631 if (!throttling_group) {
632 throttling_group = id;
634 blk_io_limits_enable(blk, throttling_group);
635 blk_set_io_limits(blk, &cfg);
638 blk_set_enable_write_cache(blk, !writethrough);
639 blk_set_on_error(blk, on_read_error, on_write_error);
641 if (!monitor_add_blk(blk, id, errp)) {
642 blk_unref(blk);
643 blk = NULL;
644 goto err_no_bs_opts;
647 err_no_bs_opts:
648 qemu_opts_del(opts);
649 qobject_unref(interval_dict);
650 qobject_unref(interval_list);
651 return blk;
653 early_err:
654 qemu_opts_del(opts);
655 qobject_unref(interval_dict);
656 qobject_unref(interval_list);
657 err_no_opts:
658 qobject_unref(bs_opts);
659 return NULL;
662 /* Takes the ownership of bs_opts */
663 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
665 int bdrv_flags = 0;
667 GLOBAL_STATE_CODE();
668 /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
669 * with other callers) rather than what we want as the real defaults.
670 * Apply the defaults here instead. */
671 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
672 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
673 qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
675 if (runstate_check(RUN_STATE_INMIGRATE)) {
676 bdrv_flags |= BDRV_O_INACTIVE;
679 return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
682 void blockdev_close_all_bdrv_states(void)
684 BlockDriverState *bs, *next_bs;
686 GLOBAL_STATE_CODE();
687 QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
688 bdrv_unref(bs);
692 /* Iterates over the list of monitor-owned BlockDriverStates */
693 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
695 GLOBAL_STATE_CODE();
696 return bs ? QTAILQ_NEXT(bs, monitor_list)
697 : QTAILQ_FIRST(&monitor_bdrv_states);
700 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
701 Error **errp)
703 const char *value;
705 value = qemu_opt_get(opts, from);
706 if (value) {
707 if (qemu_opt_find(opts, to)) {
708 error_setg(errp, "'%s' and its alias '%s' can't be used at the "
709 "same time", to, from);
710 return false;
714 /* rename all items in opts */
715 while ((value = qemu_opt_get(opts, from))) {
716 qemu_opt_set(opts, to, value, &error_abort);
717 qemu_opt_unset(opts, from);
719 return true;
722 QemuOptsList qemu_legacy_drive_opts = {
723 .name = "drive",
724 .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
725 .desc = {
727 .name = "bus",
728 .type = QEMU_OPT_NUMBER,
729 .help = "bus number",
731 .name = "unit",
732 .type = QEMU_OPT_NUMBER,
733 .help = "unit number (i.e. lun for scsi)",
735 .name = "index",
736 .type = QEMU_OPT_NUMBER,
737 .help = "index number",
739 .name = "media",
740 .type = QEMU_OPT_STRING,
741 .help = "media type (disk, cdrom)",
743 .name = "if",
744 .type = QEMU_OPT_STRING,
745 .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
747 .name = "file",
748 .type = QEMU_OPT_STRING,
749 .help = "file name",
752 /* Options that are passed on, but have special semantics with -drive */
754 .name = BDRV_OPT_READ_ONLY,
755 .type = QEMU_OPT_BOOL,
756 .help = "open drive file as read-only",
758 .name = "rerror",
759 .type = QEMU_OPT_STRING,
760 .help = "read error action",
762 .name = "werror",
763 .type = QEMU_OPT_STRING,
764 .help = "write error action",
766 .name = "copy-on-read",
767 .type = QEMU_OPT_BOOL,
768 .help = "copy read data from backing file into image file",
771 { /* end of list */ }
775 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
776 Error **errp)
778 const char *value;
779 BlockBackend *blk;
780 DriveInfo *dinfo = NULL;
781 QDict *bs_opts;
782 QemuOpts *legacy_opts;
783 DriveMediaType media = MEDIA_DISK;
784 BlockInterfaceType type;
785 int max_devs, bus_id, unit_id, index;
786 const char *werror, *rerror;
787 bool read_only = false;
788 bool copy_on_read;
789 const char *filename;
790 int i;
792 GLOBAL_STATE_CODE();
794 /* Change legacy command line options into QMP ones */
795 static const struct {
796 const char *from;
797 const char *to;
798 } opt_renames[] = {
799 { "iops", "throttling.iops-total" },
800 { "iops_rd", "throttling.iops-read" },
801 { "iops_wr", "throttling.iops-write" },
803 { "bps", "throttling.bps-total" },
804 { "bps_rd", "throttling.bps-read" },
805 { "bps_wr", "throttling.bps-write" },
807 { "iops_max", "throttling.iops-total-max" },
808 { "iops_rd_max", "throttling.iops-read-max" },
809 { "iops_wr_max", "throttling.iops-write-max" },
811 { "bps_max", "throttling.bps-total-max" },
812 { "bps_rd_max", "throttling.bps-read-max" },
813 { "bps_wr_max", "throttling.bps-write-max" },
815 { "iops_size", "throttling.iops-size" },
817 { "group", "throttling.group" },
819 { "readonly", BDRV_OPT_READ_ONLY },
822 for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
823 if (!qemu_opt_rename(all_opts, opt_renames[i].from,
824 opt_renames[i].to, errp)) {
825 return NULL;
829 value = qemu_opt_get(all_opts, "cache");
830 if (value) {
831 int flags = 0;
832 bool writethrough;
834 if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
835 error_setg(errp, "invalid cache option");
836 return NULL;
839 /* Specific options take precedence */
840 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
841 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
842 !writethrough, &error_abort);
844 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
845 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
846 !!(flags & BDRV_O_NOCACHE), &error_abort);
848 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
849 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
850 !!(flags & BDRV_O_NO_FLUSH), &error_abort);
852 qemu_opt_unset(all_opts, "cache");
855 /* Get a QDict for processing the options */
856 bs_opts = qdict_new();
857 qemu_opts_to_qdict(all_opts, bs_opts);
859 legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
860 &error_abort);
861 if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
862 goto fail;
865 /* Media type */
866 value = qemu_opt_get(legacy_opts, "media");
867 if (value) {
868 if (!strcmp(value, "disk")) {
869 media = MEDIA_DISK;
870 } else if (!strcmp(value, "cdrom")) {
871 media = MEDIA_CDROM;
872 read_only = true;
873 } else {
874 error_setg(errp, "'%s' invalid media", value);
875 goto fail;
879 /* copy-on-read is disabled with a warning for read-only devices */
880 read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
881 copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
883 if (read_only && copy_on_read) {
884 warn_report("disabling copy-on-read on read-only drive");
885 copy_on_read = false;
888 qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
889 qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
891 /* Controller type */
892 value = qemu_opt_get(legacy_opts, "if");
893 if (value) {
894 for (type = 0;
895 type < IF_COUNT && strcmp(value, if_name[type]);
896 type++) {
898 if (type == IF_COUNT) {
899 error_setg(errp, "unsupported bus type '%s'", value);
900 goto fail;
902 } else {
903 type = block_default_type;
906 /* Device address specified by bus/unit or index.
907 * If none was specified, try to find the first free one. */
908 bus_id = qemu_opt_get_number(legacy_opts, "bus", 0);
909 unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
910 index = qemu_opt_get_number(legacy_opts, "index", -1);
912 max_devs = if_max_devs[type];
914 if (index != -1) {
915 if (bus_id != 0 || unit_id != -1) {
916 error_setg(errp, "index cannot be used with bus and unit");
917 goto fail;
919 bus_id = drive_index_to_bus_id(type, index);
920 unit_id = drive_index_to_unit_id(type, index);
923 if (unit_id == -1) {
924 unit_id = 0;
925 while (drive_get(type, bus_id, unit_id) != NULL) {
926 unit_id++;
927 if (max_devs && unit_id >= max_devs) {
928 unit_id -= max_devs;
929 bus_id++;
934 if (max_devs && unit_id >= max_devs) {
935 error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
936 goto fail;
939 if (drive_get(type, bus_id, unit_id) != NULL) {
940 error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
941 bus_id, unit_id, index);
942 goto fail;
945 /* no id supplied -> create one */
946 if (qemu_opts_id(all_opts) == NULL) {
947 char *new_id;
948 const char *mediastr = "";
949 if (type == IF_IDE || type == IF_SCSI) {
950 mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
952 if (max_devs) {
953 new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
954 mediastr, unit_id);
955 } else {
956 new_id = g_strdup_printf("%s%s%i", if_name[type],
957 mediastr, unit_id);
959 qdict_put_str(bs_opts, "id", new_id);
960 g_free(new_id);
963 /* Add virtio block device */
964 if (type == IF_VIRTIO) {
965 QemuOpts *devopts;
966 devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
967 &error_abort);
968 qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
969 qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
970 &error_abort);
971 } else if (type == IF_XEN) {
972 QemuOpts *devopts;
973 devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
974 &error_abort);
975 qemu_opt_set(devopts, "driver",
976 (media == MEDIA_CDROM) ? "xen-cdrom" : "xen-disk",
977 &error_abort);
978 qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
979 &error_abort);
982 filename = qemu_opt_get(legacy_opts, "file");
984 /* Check werror/rerror compatibility with if=... */
985 werror = qemu_opt_get(legacy_opts, "werror");
986 if (werror != NULL) {
987 if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
988 type != IF_NONE) {
989 error_setg(errp, "werror is not supported by this bus type");
990 goto fail;
992 qdict_put_str(bs_opts, "werror", werror);
995 rerror = qemu_opt_get(legacy_opts, "rerror");
996 if (rerror != NULL) {
997 if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
998 type != IF_NONE) {
999 error_setg(errp, "rerror is not supported by this bus type");
1000 goto fail;
1002 qdict_put_str(bs_opts, "rerror", rerror);
1005 /* Actual block device init: Functionality shared with blockdev-add */
1006 blk = blockdev_init(filename, bs_opts, errp);
1007 bs_opts = NULL;
1008 if (!blk) {
1009 goto fail;
1012 /* Create legacy DriveInfo */
1013 dinfo = g_malloc0(sizeof(*dinfo));
1014 dinfo->opts = all_opts;
1016 dinfo->type = type;
1017 dinfo->bus = bus_id;
1018 dinfo->unit = unit_id;
1020 blk_set_legacy_dinfo(blk, dinfo);
1022 switch(type) {
1023 case IF_IDE:
1024 case IF_SCSI:
1025 case IF_XEN:
1026 case IF_NONE:
1027 dinfo->media_cd = media == MEDIA_CDROM;
1028 break;
1029 default:
1030 break;
1033 fail:
1034 qemu_opts_del(legacy_opts);
1035 qobject_unref(bs_opts);
1036 return dinfo;
1039 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1041 BlockDriverState *bs;
1043 GRAPH_RDLOCK_GUARD_MAINLOOP();
1045 bs = bdrv_lookup_bs(name, name, errp);
1046 if (bs == NULL) {
1047 return NULL;
1050 if (!bdrv_is_root_node(bs)) {
1051 error_setg(errp, "Need a root block node");
1052 return NULL;
1055 if (!bdrv_is_inserted(bs)) {
1056 error_setg(errp, "Device has no medium");
1057 bs = NULL;
1060 return bs;
1063 static void blockdev_do_action(TransactionAction *action, Error **errp)
1065 TransactionActionList list;
1067 list.value = action;
1068 list.next = NULL;
1069 qmp_transaction(&list, NULL, errp);
1072 void qmp_blockdev_snapshot_sync(const char *device, const char *node_name,
1073 const char *snapshot_file,
1074 const char *snapshot_node_name,
1075 const char *format,
1076 bool has_mode, NewImageMode mode, Error **errp)
1078 BlockdevSnapshotSync snapshot = {
1079 .device = (char *) device,
1080 .node_name = (char *) node_name,
1081 .snapshot_file = (char *) snapshot_file,
1082 .snapshot_node_name = (char *) snapshot_node_name,
1083 .format = (char *) format,
1084 .has_mode = has_mode,
1085 .mode = mode,
1087 TransactionAction action = {
1088 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1089 .u.blockdev_snapshot_sync.data = &snapshot,
1091 blockdev_do_action(&action, errp);
1094 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1095 Error **errp)
1097 BlockdevSnapshot snapshot_data = {
1098 .node = (char *) node,
1099 .overlay = (char *) overlay
1101 TransactionAction action = {
1102 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1103 .u.blockdev_snapshot.data = &snapshot_data,
1105 blockdev_do_action(&action, errp);
1108 void qmp_blockdev_snapshot_internal_sync(const char *device,
1109 const char *name,
1110 Error **errp)
1112 BlockdevSnapshotInternal snapshot = {
1113 .device = (char *) device,
1114 .name = (char *) name
1116 TransactionAction action = {
1117 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1118 .u.blockdev_snapshot_internal_sync.data = &snapshot,
1120 blockdev_do_action(&action, errp);
1123 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1124 const char *id,
1125 const char *name,
1126 Error **errp)
1128 BlockDriverState *bs;
1129 QEMUSnapshotInfo sn;
1130 Error *local_err = NULL;
1131 SnapshotInfo *info = NULL;
1132 int ret;
1134 GLOBAL_STATE_CODE();
1135 GRAPH_RDLOCK_GUARD_MAINLOOP();
1137 bs = qmp_get_root_bs(device, errp);
1138 if (!bs) {
1139 return NULL;
1142 if (!id && !name) {
1143 error_setg(errp, "Name or id must be provided");
1144 return NULL;
1147 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1148 return NULL;
1151 ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1152 if (local_err) {
1153 error_propagate(errp, local_err);
1154 return NULL;
1156 if (!ret) {
1157 error_setg(errp,
1158 "Snapshot with id '%s' and name '%s' does not exist on "
1159 "device '%s'",
1160 STR_OR_NULL(id), STR_OR_NULL(name), device);
1161 return NULL;
1164 bdrv_snapshot_delete(bs, id, name, &local_err);
1165 if (local_err) {
1166 error_propagate(errp, local_err);
1167 return NULL;
1170 info = g_new0(SnapshotInfo, 1);
1171 info->id = g_strdup(sn.id_str);
1172 info->name = g_strdup(sn.name);
1173 info->date_nsec = sn.date_nsec;
1174 info->date_sec = sn.date_sec;
1175 info->vm_state_size = sn.vm_state_size;
1176 info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1177 info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1178 if (sn.icount != -1ULL) {
1179 info->icount = sn.icount;
1180 info->has_icount = true;
1183 return info;
1186 /* internal snapshot private data */
1187 typedef struct InternalSnapshotState {
1188 BlockDriverState *bs;
1189 QEMUSnapshotInfo sn;
1190 bool created;
1191 } InternalSnapshotState;
1193 static void internal_snapshot_abort(void *opaque);
1194 static void internal_snapshot_clean(void *opaque);
1195 TransactionActionDrv internal_snapshot_drv = {
1196 .abort = internal_snapshot_abort,
1197 .clean = internal_snapshot_clean,
1200 static void internal_snapshot_action(BlockdevSnapshotInternal *internal,
1201 Transaction *tran, Error **errp)
1203 Error *local_err = NULL;
1204 const char *device;
1205 const char *name;
1206 BlockDriverState *bs;
1207 QEMUSnapshotInfo old_sn, *sn;
1208 bool ret;
1209 int64_t rt;
1210 InternalSnapshotState *state = g_new0(InternalSnapshotState, 1);
1211 int ret1;
1213 GLOBAL_STATE_CODE();
1214 GRAPH_RDLOCK_GUARD_MAINLOOP();
1216 tran_add(tran, &internal_snapshot_drv, state);
1218 device = internal->device;
1219 name = internal->name;
1221 bs = qmp_get_root_bs(device, errp);
1222 if (!bs) {
1223 return;
1226 state->bs = bs;
1228 /* Paired with .clean() */
1229 bdrv_drained_begin(bs);
1231 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1232 return;
1235 if (bdrv_is_read_only(bs)) {
1236 error_setg(errp, "Device '%s' is read only", device);
1237 return;
1240 if (!bdrv_can_snapshot(bs)) {
1241 error_setg(errp, "Block format '%s' used by device '%s' "
1242 "does not support internal snapshots",
1243 bs->drv->format_name, device);
1244 return;
1247 if (!strlen(name)) {
1248 error_setg(errp, "Name is empty");
1249 return;
1252 /* check whether a snapshot with name exist */
1253 ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1254 &local_err);
1255 if (local_err) {
1256 error_propagate(errp, local_err);
1257 return;
1258 } else if (ret) {
1259 error_setg(errp,
1260 "Snapshot with name '%s' already exists on device '%s'",
1261 name, device);
1262 return;
1265 /* 3. take the snapshot */
1266 sn = &state->sn;
1267 pstrcpy(sn->name, sizeof(sn->name), name);
1268 rt = g_get_real_time();
1269 sn->date_sec = rt / G_USEC_PER_SEC;
1270 sn->date_nsec = (rt % G_USEC_PER_SEC) * 1000;
1271 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1272 if (replay_mode != REPLAY_MODE_NONE) {
1273 sn->icount = replay_get_current_icount();
1274 } else {
1275 sn->icount = -1ULL;
1278 ret1 = bdrv_snapshot_create(bs, sn);
1279 if (ret1 < 0) {
1280 error_setg_errno(errp, -ret1,
1281 "Failed to create snapshot '%s' on device '%s'",
1282 name, device);
1283 return;
1286 /* 4. succeed, mark a snapshot is created */
1287 state->created = true;
1290 static void internal_snapshot_abort(void *opaque)
1292 InternalSnapshotState *state = opaque;
1293 BlockDriverState *bs = state->bs;
1294 QEMUSnapshotInfo *sn = &state->sn;
1295 Error *local_error = NULL;
1297 GLOBAL_STATE_CODE();
1298 GRAPH_RDLOCK_GUARD_MAINLOOP();
1300 if (!state->created) {
1301 return;
1304 if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1305 error_reportf_err(local_error,
1306 "Failed to delete snapshot with id '%s' and "
1307 "name '%s' on device '%s' in abort: ",
1308 sn->id_str, sn->name,
1309 bdrv_get_device_name(bs));
1313 static void internal_snapshot_clean(void *opaque)
1315 g_autofree InternalSnapshotState *state = opaque;
1317 if (!state->bs) {
1318 return;
1321 bdrv_drained_end(state->bs);
1324 /* external snapshot private data */
1325 typedef struct ExternalSnapshotState {
1326 BlockDriverState *old_bs;
1327 BlockDriverState *new_bs;
1328 bool overlay_appended;
1329 } ExternalSnapshotState;
1331 static void external_snapshot_commit(void *opaque);
1332 static void external_snapshot_abort(void *opaque);
1333 static void external_snapshot_clean(void *opaque);
1334 TransactionActionDrv external_snapshot_drv = {
1335 .commit = external_snapshot_commit,
1336 .abort = external_snapshot_abort,
1337 .clean = external_snapshot_clean,
1340 static void external_snapshot_action(TransactionAction *action,
1341 Transaction *tran, Error **errp)
1343 int ret;
1344 int flags = 0;
1345 QDict *options = NULL;
1346 Error *local_err = NULL;
1347 /* Device and node name of the image to generate the snapshot from */
1348 const char *device;
1349 const char *node_name;
1350 /* Reference to the new image (for 'blockdev-snapshot') */
1351 const char *snapshot_ref;
1352 /* File name of the new image (for 'blockdev-snapshot-sync') */
1353 const char *new_image_file;
1354 ExternalSnapshotState *state = g_new0(ExternalSnapshotState, 1);
1355 uint64_t perm, shared;
1357 /* TODO We'll eventually have to take a writer lock in this function */
1358 GRAPH_RDLOCK_GUARD_MAINLOOP();
1360 tran_add(tran, &external_snapshot_drv, state);
1362 /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1363 * purpose but a different set of parameters */
1364 switch (action->type) {
1365 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1367 BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1368 device = s->node;
1369 node_name = s->node;
1370 new_image_file = NULL;
1371 snapshot_ref = s->overlay;
1373 break;
1374 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1376 BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1377 device = s->device;
1378 node_name = s->node_name;
1379 new_image_file = s->snapshot_file;
1380 snapshot_ref = NULL;
1382 break;
1383 default:
1384 g_assert_not_reached();
1387 /* start processing */
1389 state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1390 if (!state->old_bs) {
1391 return;
1394 /* Paired with .clean() */
1395 bdrv_drained_begin(state->old_bs);
1397 if (!bdrv_is_inserted(state->old_bs)) {
1398 error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1399 return;
1402 if (bdrv_op_is_blocked(state->old_bs,
1403 BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1404 return;
1407 if (!bdrv_is_read_only(state->old_bs)) {
1408 if (bdrv_flush(state->old_bs)) {
1409 error_setg(errp, QERR_IO_ERROR);
1410 return;
1414 if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1415 BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1416 const char *format = s->format ?: "qcow2";
1417 enum NewImageMode mode;
1418 const char *snapshot_node_name = s->snapshot_node_name;
1420 if (node_name && !snapshot_node_name) {
1421 error_setg(errp, "New overlay node-name missing");
1422 return;
1425 if (snapshot_node_name &&
1426 bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1427 error_setg(errp, "New overlay node-name already in use");
1428 return;
1431 flags = state->old_bs->open_flags;
1432 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1433 flags |= BDRV_O_NO_BACKING;
1435 /* create new image w/backing file */
1436 mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1437 if (mode != NEW_IMAGE_MODE_EXISTING) {
1438 int64_t size = bdrv_getlength(state->old_bs);
1439 if (size < 0) {
1440 error_setg_errno(errp, -size, "bdrv_getlength failed");
1441 return;
1443 bdrv_refresh_filename(state->old_bs);
1445 bdrv_img_create(new_image_file, format,
1446 state->old_bs->filename,
1447 state->old_bs->drv->format_name,
1448 NULL, size, flags, false, &local_err);
1450 if (local_err) {
1451 error_propagate(errp, local_err);
1452 return;
1456 options = qdict_new();
1457 if (snapshot_node_name) {
1458 qdict_put_str(options, "node-name", snapshot_node_name);
1460 qdict_put_str(options, "driver", format);
1463 state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1464 errp);
1466 /* We will manually add the backing_hd field to the bs later */
1467 if (!state->new_bs) {
1468 return;
1472 * Allow attaching a backing file to an overlay that's already in use only
1473 * if the parents don't assume that they are already seeing a valid image.
1474 * (Specifically, allow it as a mirror target, which is write-only access.)
1476 bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1477 if (perm & BLK_PERM_CONSISTENT_READ) {
1478 error_setg(errp, "The overlay is already in use");
1479 return;
1482 if (state->new_bs->drv->is_filter) {
1483 error_setg(errp, "Filters cannot be used as overlays");
1484 return;
1487 if (bdrv_cow_child(state->new_bs)) {
1488 error_setg(errp, "The overlay already has a backing image");
1489 return;
1492 if (!state->new_bs->drv->supports_backing) {
1493 error_setg(errp, "The overlay does not support backing images");
1494 return;
1497 ret = bdrv_append(state->new_bs, state->old_bs, errp);
1498 if (ret < 0) {
1499 return;
1501 state->overlay_appended = true;
1504 static void external_snapshot_commit(void *opaque)
1506 ExternalSnapshotState *state = opaque;
1508 /* We don't need (or want) to use the transactional
1509 * bdrv_reopen_multiple() across all the entries at once, because we
1510 * don't want to abort all of them if one of them fails the reopen */
1511 if (!qatomic_read(&state->old_bs->copy_on_read)) {
1512 bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1516 static void external_snapshot_abort(void *opaque)
1518 ExternalSnapshotState *state = opaque;
1519 if (state->new_bs) {
1520 if (state->overlay_appended) {
1521 AioContext *aio_context;
1522 AioContext *tmp_context;
1523 int ret;
1525 aio_context = bdrv_get_aio_context(state->old_bs);
1527 bdrv_ref(state->old_bs); /* we can't let bdrv_set_backind_hd()
1528 close state->old_bs; we need it */
1529 bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1532 * The call to bdrv_set_backing_hd() above returns state->old_bs to
1533 * the main AioContext. As we're still going to be using it, return
1534 * it to the AioContext it was before.
1536 tmp_context = bdrv_get_aio_context(state->old_bs);
1537 if (aio_context != tmp_context) {
1538 ret = bdrv_try_change_aio_context(state->old_bs,
1539 aio_context, NULL, NULL);
1540 assert(ret == 0);
1543 bdrv_drained_begin(state->new_bs);
1544 bdrv_graph_wrlock();
1545 bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1546 bdrv_graph_wrunlock();
1547 bdrv_drained_end(state->new_bs);
1549 bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1554 static void external_snapshot_clean(void *opaque)
1556 g_autofree ExternalSnapshotState *state = opaque;
1558 if (!state->old_bs) {
1559 return;
1562 bdrv_drained_end(state->old_bs);
1563 bdrv_unref(state->new_bs);
1566 typedef struct DriveBackupState {
1567 BlockDriverState *bs;
1568 BlockJob *job;
1569 } DriveBackupState;
1571 static BlockJob *do_backup_common(BackupCommon *backup,
1572 BlockDriverState *bs,
1573 BlockDriverState *target_bs,
1574 AioContext *aio_context,
1575 JobTxn *txn, Error **errp);
1577 static void drive_backup_commit(void *opaque);
1578 static void drive_backup_abort(void *opaque);
1579 static void drive_backup_clean(void *opaque);
1580 TransactionActionDrv drive_backup_drv = {
1581 .commit = drive_backup_commit,
1582 .abort = drive_backup_abort,
1583 .clean = drive_backup_clean,
1586 static void drive_backup_action(DriveBackup *backup,
1587 JobTxn *block_job_txn,
1588 Transaction *tran, Error **errp)
1590 DriveBackupState *state = g_new0(DriveBackupState, 1);
1591 BlockDriverState *bs;
1592 BlockDriverState *target_bs;
1593 BlockDriverState *source = NULL;
1594 AioContext *aio_context;
1595 const char *format;
1596 QDict *options;
1597 Error *local_err = NULL;
1598 int flags;
1599 int64_t size;
1600 bool set_backing_hd = false;
1601 int ret;
1603 GLOBAL_STATE_CODE();
1605 tran_add(tran, &drive_backup_drv, state);
1607 if (!backup->has_mode) {
1608 backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1611 bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1612 if (!bs) {
1613 return;
1616 if (!bs->drv) {
1617 error_setg(errp, "Device has no medium");
1618 return;
1621 aio_context = bdrv_get_aio_context(bs);
1623 state->bs = bs;
1624 /* Paired with .clean() */
1625 bdrv_drained_begin(bs);
1627 format = backup->format;
1628 if (!format && backup->mode != NEW_IMAGE_MODE_EXISTING) {
1629 format = bs->drv->format_name;
1632 /* Early check to avoid creating target */
1633 bdrv_graph_rdlock_main_loop();
1634 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1635 bdrv_graph_rdunlock_main_loop();
1636 return;
1639 flags = bs->open_flags | BDRV_O_RDWR;
1642 * See if we have a backing HD we can use to create our new image
1643 * on top of.
1645 if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1647 * Backup will not replace the source by the target, so none
1648 * of the filters skipped here will be removed (in contrast to
1649 * mirror). Therefore, we can skip all of them when looking
1650 * for the first COW relationship.
1652 source = bdrv_cow_bs(bdrv_skip_filters(bs));
1653 if (!source) {
1654 backup->sync = MIRROR_SYNC_MODE_FULL;
1657 if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1658 source = bs;
1659 flags |= BDRV_O_NO_BACKING;
1660 set_backing_hd = true;
1662 bdrv_graph_rdunlock_main_loop();
1664 size = bdrv_getlength(bs);
1665 if (size < 0) {
1666 error_setg_errno(errp, -size, "bdrv_getlength failed");
1667 return;
1670 if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1671 assert(format);
1672 if (source) {
1673 /* Implicit filters should not appear in the filename */
1674 BlockDriverState *explicit_backing;
1676 bdrv_graph_rdlock_main_loop();
1677 explicit_backing = bdrv_skip_implicit_filters(source);
1678 bdrv_refresh_filename(explicit_backing);
1679 bdrv_graph_rdunlock_main_loop();
1681 bdrv_img_create(backup->target, format,
1682 explicit_backing->filename,
1683 explicit_backing->drv->format_name, NULL,
1684 size, flags, false, &local_err);
1685 } else {
1686 bdrv_img_create(backup->target, format, NULL, NULL, NULL,
1687 size, flags, false, &local_err);
1691 if (local_err) {
1692 error_propagate(errp, local_err);
1693 return;
1696 options = qdict_new();
1697 qdict_put_str(options, "discard", "unmap");
1698 qdict_put_str(options, "detect-zeroes", "unmap");
1699 if (format) {
1700 qdict_put_str(options, "driver", format);
1703 target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1704 if (!target_bs) {
1705 return;
1708 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1709 if (ret < 0) {
1710 bdrv_unref(target_bs);
1711 return;
1714 if (set_backing_hd) {
1715 if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1716 goto unref;
1720 state->job = do_backup_common(qapi_DriveBackup_base(backup),
1721 bs, target_bs, aio_context,
1722 block_job_txn, errp);
1724 unref:
1725 bdrv_unref(target_bs);
1728 static void drive_backup_commit(void *opaque)
1730 DriveBackupState *state = opaque;
1732 assert(state->job);
1733 job_start(&state->job->job);
1736 static void drive_backup_abort(void *opaque)
1738 DriveBackupState *state = opaque;
1740 if (state->job) {
1741 job_cancel_sync(&state->job->job, true);
1745 static void drive_backup_clean(void *opaque)
1747 g_autofree DriveBackupState *state = opaque;
1749 if (!state->bs) {
1750 return;
1753 bdrv_drained_end(state->bs);
1756 typedef struct BlockdevBackupState {
1757 BlockDriverState *bs;
1758 BlockJob *job;
1759 } BlockdevBackupState;
1761 static void blockdev_backup_commit(void *opaque);
1762 static void blockdev_backup_abort(void *opaque);
1763 static void blockdev_backup_clean(void *opaque);
1764 TransactionActionDrv blockdev_backup_drv = {
1765 .commit = blockdev_backup_commit,
1766 .abort = blockdev_backup_abort,
1767 .clean = blockdev_backup_clean,
1770 static void blockdev_backup_action(BlockdevBackup *backup,
1771 JobTxn *block_job_txn,
1772 Transaction *tran, Error **errp)
1774 BlockdevBackupState *state = g_new0(BlockdevBackupState, 1);
1775 BlockDriverState *bs;
1776 BlockDriverState *target_bs;
1777 AioContext *aio_context;
1778 int ret;
1780 tran_add(tran, &blockdev_backup_drv, state);
1782 bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1783 if (!bs) {
1784 return;
1787 target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1788 if (!target_bs) {
1789 return;
1792 /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1793 aio_context = bdrv_get_aio_context(bs);
1795 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1796 if (ret < 0) {
1797 return;
1800 state->bs = bs;
1802 /* Paired with .clean() */
1803 bdrv_drained_begin(state->bs);
1805 state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1806 bs, target_bs, aio_context,
1807 block_job_txn, errp);
1810 static void blockdev_backup_commit(void *opaque)
1812 BlockdevBackupState *state = opaque;
1814 assert(state->job);
1815 job_start(&state->job->job);
1818 static void blockdev_backup_abort(void *opaque)
1820 BlockdevBackupState *state = opaque;
1822 if (state->job) {
1823 job_cancel_sync(&state->job->job, true);
1827 static void blockdev_backup_clean(void *opaque)
1829 g_autofree BlockdevBackupState *state = opaque;
1831 if (!state->bs) {
1832 return;
1835 bdrv_drained_end(state->bs);
1838 typedef struct BlockDirtyBitmapState {
1839 BdrvDirtyBitmap *bitmap;
1840 BlockDriverState *bs;
1841 HBitmap *backup;
1842 bool was_enabled;
1843 } BlockDirtyBitmapState;
1845 static void block_dirty_bitmap_add_abort(void *opaque);
1846 TransactionActionDrv block_dirty_bitmap_add_drv = {
1847 .abort = block_dirty_bitmap_add_abort,
1848 .clean = g_free,
1851 static void block_dirty_bitmap_add_action(BlockDirtyBitmapAdd *action,
1852 Transaction *tran, Error **errp)
1854 Error *local_err = NULL;
1855 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1857 tran_add(tran, &block_dirty_bitmap_add_drv, state);
1859 /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1860 qmp_block_dirty_bitmap_add(action->node, action->name,
1861 action->has_granularity, action->granularity,
1862 action->has_persistent, action->persistent,
1863 action->has_disabled, action->disabled,
1864 &local_err);
1866 if (!local_err) {
1867 state->bitmap = block_dirty_bitmap_lookup(action->node, action->name,
1868 NULL, &error_abort);
1869 } else {
1870 error_propagate(errp, local_err);
1874 static void block_dirty_bitmap_add_abort(void *opaque)
1876 BlockDirtyBitmapState *state = opaque;
1878 if (state->bitmap) {
1879 bdrv_release_dirty_bitmap(state->bitmap);
1883 static void block_dirty_bitmap_restore(void *opaque);
1884 static void block_dirty_bitmap_free_backup(void *opaque);
1885 TransactionActionDrv block_dirty_bitmap_clear_drv = {
1886 .abort = block_dirty_bitmap_restore,
1887 .commit = block_dirty_bitmap_free_backup,
1888 .clean = g_free,
1891 static void block_dirty_bitmap_clear_action(BlockDirtyBitmap *action,
1892 Transaction *tran, Error **errp)
1894 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1896 tran_add(tran, &block_dirty_bitmap_clear_drv, state);
1898 state->bitmap = block_dirty_bitmap_lookup(action->node,
1899 action->name,
1900 &state->bs,
1901 errp);
1902 if (!state->bitmap) {
1903 return;
1906 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
1907 return;
1910 bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
1913 static void block_dirty_bitmap_restore(void *opaque)
1915 BlockDirtyBitmapState *state = opaque;
1917 if (state->backup) {
1918 bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
1922 static void block_dirty_bitmap_free_backup(void *opaque)
1924 BlockDirtyBitmapState *state = opaque;
1926 hbitmap_free(state->backup);
1929 static void block_dirty_bitmap_enable_abort(void *opaque);
1930 TransactionActionDrv block_dirty_bitmap_enable_drv = {
1931 .abort = block_dirty_bitmap_enable_abort,
1932 .clean = g_free,
1935 static void block_dirty_bitmap_enable_action(BlockDirtyBitmap *action,
1936 Transaction *tran, Error **errp)
1938 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1940 tran_add(tran, &block_dirty_bitmap_enable_drv, state);
1942 state->bitmap = block_dirty_bitmap_lookup(action->node,
1943 action->name,
1944 NULL,
1945 errp);
1946 if (!state->bitmap) {
1947 return;
1950 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
1951 return;
1954 state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
1955 bdrv_enable_dirty_bitmap(state->bitmap);
1958 static void block_dirty_bitmap_enable_abort(void *opaque)
1960 BlockDirtyBitmapState *state = opaque;
1962 if (!state->was_enabled) {
1963 bdrv_disable_dirty_bitmap(state->bitmap);
1967 static void block_dirty_bitmap_disable_abort(void *opaque);
1968 TransactionActionDrv block_dirty_bitmap_disable_drv = {
1969 .abort = block_dirty_bitmap_disable_abort,
1970 .clean = g_free,
1973 static void block_dirty_bitmap_disable_action(BlockDirtyBitmap *action,
1974 Transaction *tran, Error **errp)
1976 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1978 tran_add(tran, &block_dirty_bitmap_disable_drv, state);
1980 state->bitmap = block_dirty_bitmap_lookup(action->node,
1981 action->name,
1982 NULL,
1983 errp);
1984 if (!state->bitmap) {
1985 return;
1988 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
1989 return;
1992 state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
1993 bdrv_disable_dirty_bitmap(state->bitmap);
1996 static void block_dirty_bitmap_disable_abort(void *opaque)
1998 BlockDirtyBitmapState *state = opaque;
2000 if (state->was_enabled) {
2001 bdrv_enable_dirty_bitmap(state->bitmap);
2005 TransactionActionDrv block_dirty_bitmap_merge_drv = {
2006 .commit = block_dirty_bitmap_free_backup,
2007 .abort = block_dirty_bitmap_restore,
2008 .clean = g_free,
2011 static void block_dirty_bitmap_merge_action(BlockDirtyBitmapMerge *action,
2012 Transaction *tran, Error **errp)
2014 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2016 tran_add(tran, &block_dirty_bitmap_merge_drv, state);
2018 state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2019 action->bitmaps, &state->backup,
2020 errp);
2023 static void block_dirty_bitmap_remove_commit(void *opaque);
2024 static void block_dirty_bitmap_remove_abort(void *opaque);
2025 TransactionActionDrv block_dirty_bitmap_remove_drv = {
2026 .commit = block_dirty_bitmap_remove_commit,
2027 .abort = block_dirty_bitmap_remove_abort,
2028 .clean = g_free,
2031 static void block_dirty_bitmap_remove_action(BlockDirtyBitmap *action,
2032 Transaction *tran, Error **errp)
2034 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2036 tran_add(tran, &block_dirty_bitmap_remove_drv, state);
2039 state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2040 false, &state->bs, errp);
2041 if (state->bitmap) {
2042 bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2043 bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2047 static void block_dirty_bitmap_remove_abort(void *opaque)
2049 BlockDirtyBitmapState *state = opaque;
2051 if (state->bitmap) {
2052 bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2053 bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2057 static void block_dirty_bitmap_remove_commit(void *opaque)
2059 BlockDirtyBitmapState *state = opaque;
2061 bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2062 bdrv_release_dirty_bitmap(state->bitmap);
2065 static void abort_commit(void *opaque);
2066 TransactionActionDrv abort_drv = {
2067 .commit = abort_commit,
2070 static void abort_action(Transaction *tran, Error **errp)
2072 tran_add(tran, &abort_drv, NULL);
2073 error_setg(errp, "Transaction aborted using Abort action");
2076 static void abort_commit(void *opaque)
2078 g_assert_not_reached(); /* this action never succeeds */
2081 static void transaction_action(TransactionAction *act, JobTxn *block_job_txn,
2082 Transaction *tran, Error **errp)
2084 switch (act->type) {
2085 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
2086 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
2087 external_snapshot_action(act, tran, errp);
2088 return;
2089 case TRANSACTION_ACTION_KIND_DRIVE_BACKUP:
2090 drive_backup_action(act->u.drive_backup.data,
2091 block_job_txn, tran, errp);
2092 return;
2093 case TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP:
2094 blockdev_backup_action(act->u.blockdev_backup.data,
2095 block_job_txn, tran, errp);
2096 return;
2097 case TRANSACTION_ACTION_KIND_ABORT:
2098 abort_action(tran, errp);
2099 return;
2100 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC:
2101 internal_snapshot_action(act->u.blockdev_snapshot_internal_sync.data,
2102 tran, errp);
2103 return;
2104 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD:
2105 block_dirty_bitmap_add_action(act->u.block_dirty_bitmap_add.data,
2106 tran, errp);
2107 return;
2108 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR:
2109 block_dirty_bitmap_clear_action(act->u.block_dirty_bitmap_clear.data,
2110 tran, errp);
2111 return;
2112 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE:
2113 block_dirty_bitmap_enable_action(act->u.block_dirty_bitmap_enable.data,
2114 tran, errp);
2115 return;
2116 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE:
2117 block_dirty_bitmap_disable_action(
2118 act->u.block_dirty_bitmap_disable.data, tran, errp);
2119 return;
2120 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE:
2121 block_dirty_bitmap_merge_action(act->u.block_dirty_bitmap_merge.data,
2122 tran, errp);
2123 return;
2124 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE:
2125 block_dirty_bitmap_remove_action(act->u.block_dirty_bitmap_remove.data,
2126 tran, errp);
2127 return;
2129 * Where are transactions for MIRROR, COMMIT and STREAM?
2130 * Although these blockjobs use transaction callbacks like the backup job,
2131 * these jobs do not necessarily adhere to transaction semantics.
2132 * These jobs may not fully undo all of their actions on abort, nor do they
2133 * necessarily work in transactions with more than one job in them.
2135 case TRANSACTION_ACTION_KIND__MAX:
2136 default:
2137 g_assert_not_reached();
2143 * 'Atomic' group operations. The operations are performed as a set, and if
2144 * any fail then we roll back all operations in the group.
2146 * Always run under BQL.
2148 void qmp_transaction(TransactionActionList *actions,
2149 struct TransactionProperties *properties,
2150 Error **errp)
2152 TransactionActionList *act;
2153 JobTxn *block_job_txn = NULL;
2154 Error *local_err = NULL;
2155 Transaction *tran;
2156 ActionCompletionMode comp_mode =
2157 properties ? properties->completion_mode :
2158 ACTION_COMPLETION_MODE_INDIVIDUAL;
2160 GLOBAL_STATE_CODE();
2162 /* Does this transaction get canceled as a group on failure?
2163 * If not, we don't really need to make a JobTxn.
2165 if (comp_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2166 for (act = actions; act; act = act->next) {
2167 TransactionActionKind type = act->value->type;
2169 if (type != TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP &&
2170 type != TRANSACTION_ACTION_KIND_DRIVE_BACKUP)
2172 error_setg(errp,
2173 "Action '%s' does not support transaction property "
2174 "completion-mode = %s",
2175 TransactionActionKind_str(type),
2176 ActionCompletionMode_str(comp_mode));
2177 return;
2181 block_job_txn = job_txn_new();
2184 /* drain all i/o before any operations */
2185 bdrv_drain_all();
2187 tran = tran_new();
2189 /* We don't do anything in this loop that commits us to the operations */
2190 for (act = actions; act; act = act->next) {
2191 transaction_action(act->value, block_job_txn, tran, &local_err);
2192 if (local_err) {
2193 error_propagate(errp, local_err);
2194 goto delete_and_fail;
2198 tran_commit(tran);
2200 /* success */
2201 goto exit;
2203 delete_and_fail:
2204 /* failure, and it is all-or-none; roll back all operations */
2205 tran_abort(tran);
2206 exit:
2207 job_txn_unref(block_job_txn);
2210 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2211 const char *name,
2212 Error **errp)
2214 BdrvDirtyBitmap *bitmap;
2215 BlockDriverState *bs;
2216 BlockDirtyBitmapSha256 *ret = NULL;
2217 char *sha256;
2219 bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2220 if (!bitmap || !bs) {
2221 return NULL;
2224 sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2225 if (sha256 == NULL) {
2226 return NULL;
2229 ret = g_new(BlockDirtyBitmapSha256, 1);
2230 ret->sha256 = sha256;
2232 return ret;
2235 void coroutine_fn qmp_block_resize(const char *device, const char *node_name,
2236 int64_t size, Error **errp)
2238 Error *local_err = NULL;
2239 BlockBackend *blk;
2240 BlockDriverState *bs;
2241 AioContext *old_ctx;
2243 bs = bdrv_lookup_bs(device, node_name, &local_err);
2244 if (local_err) {
2245 error_propagate(errp, local_err);
2246 return;
2249 if (size < 0) {
2250 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2251 return;
2254 bdrv_graph_co_rdlock();
2255 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, errp)) {
2256 bdrv_graph_co_rdunlock();
2257 return;
2259 bdrv_graph_co_rdunlock();
2261 blk = blk_co_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2262 if (!blk) {
2263 return;
2266 bdrv_drained_begin(bs);
2268 old_ctx = bdrv_co_enter(bs);
2269 blk_co_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2270 bdrv_co_leave(bs, old_ctx);
2272 bdrv_drained_end(bs);
2273 blk_co_unref(blk);
2276 void qmp_block_stream(const char *job_id, const char *device,
2277 const char *base,
2278 const char *base_node,
2279 const char *backing_file,
2280 bool has_backing_mask_protocol,
2281 bool backing_mask_protocol,
2282 const char *bottom,
2283 bool has_speed, int64_t speed,
2284 bool has_on_error, BlockdevOnError on_error,
2285 const char *filter_node_name,
2286 bool has_auto_finalize, bool auto_finalize,
2287 bool has_auto_dismiss, bool auto_dismiss,
2288 Error **errp)
2290 BlockDriverState *bs, *iter, *iter_end;
2291 BlockDriverState *base_bs = NULL;
2292 BlockDriverState *bottom_bs = NULL;
2293 AioContext *aio_context;
2294 Error *local_err = NULL;
2295 int job_flags = JOB_DEFAULT;
2297 GLOBAL_STATE_CODE();
2299 if (base && base_node) {
2300 error_setg(errp, "'base' and 'base-node' cannot be specified "
2301 "at the same time");
2302 return;
2305 if (base && bottom) {
2306 error_setg(errp, "'base' and 'bottom' cannot be specified "
2307 "at the same time");
2308 return;
2311 if (bottom && base_node) {
2312 error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2313 "at the same time");
2314 return;
2317 if (!has_backing_mask_protocol) {
2318 backing_mask_protocol = false;
2321 if (!has_on_error) {
2322 on_error = BLOCKDEV_ON_ERROR_REPORT;
2325 bs = bdrv_lookup_bs(device, device, errp);
2326 if (!bs) {
2327 return;
2330 aio_context = bdrv_get_aio_context(bs);
2332 bdrv_graph_rdlock_main_loop();
2333 if (base) {
2334 base_bs = bdrv_find_backing_image(bs, base);
2335 if (base_bs == NULL) {
2336 error_setg(errp, "Can't find '%s' in the backing chain", base);
2337 goto out_rdlock;
2339 assert(bdrv_get_aio_context(base_bs) == aio_context);
2342 if (base_node) {
2343 base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2344 if (!base_bs) {
2345 goto out_rdlock;
2347 if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2348 error_setg(errp, "Node '%s' is not a backing image of '%s'",
2349 base_node, device);
2350 goto out_rdlock;
2352 assert(bdrv_get_aio_context(base_bs) == aio_context);
2354 bdrv_refresh_filename(base_bs);
2357 if (bottom) {
2358 bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2359 if (!bottom_bs) {
2360 goto out_rdlock;
2362 if (!bottom_bs->drv) {
2363 error_setg(errp, "Node '%s' is not open", bottom);
2364 goto out_rdlock;
2366 if (bottom_bs->drv->is_filter) {
2367 error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2368 "as 'bottom'", bottom);
2369 goto out_rdlock;
2371 if (!bdrv_chain_contains(bs, bottom_bs)) {
2372 error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2373 bottom, device);
2374 goto out_rdlock;
2376 assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2380 * Check for op blockers in the whole chain between bs and base (or bottom)
2382 iter_end = bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2383 for (iter = bs; iter && iter != iter_end;
2384 iter = bdrv_filter_or_cow_bs(iter))
2386 if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2387 goto out_rdlock;
2390 bdrv_graph_rdunlock_main_loop();
2392 /* if we are streaming the entire chain, the result will have no backing
2393 * file, and specifying one is therefore an error */
2394 if (!base_bs && backing_file) {
2395 error_setg(errp, "backing file specified, but streaming the "
2396 "entire chain");
2397 return;
2400 if (has_auto_finalize && !auto_finalize) {
2401 job_flags |= JOB_MANUAL_FINALIZE;
2403 if (has_auto_dismiss && !auto_dismiss) {
2404 job_flags |= JOB_MANUAL_DISMISS;
2407 stream_start(job_id, bs, base_bs, backing_file,
2408 backing_mask_protocol,
2409 bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2410 filter_node_name, &local_err);
2411 if (local_err) {
2412 error_propagate(errp, local_err);
2413 return;
2416 trace_qmp_block_stream(bs);
2417 return;
2419 out_rdlock:
2420 bdrv_graph_rdunlock_main_loop();
2423 void qmp_block_commit(const char *job_id, const char *device,
2424 const char *base_node,
2425 const char *base,
2426 const char *top_node,
2427 const char *top,
2428 const char *backing_file,
2429 bool has_backing_mask_protocol,
2430 bool backing_mask_protocol,
2431 bool has_speed, int64_t speed,
2432 bool has_on_error, BlockdevOnError on_error,
2433 const char *filter_node_name,
2434 bool has_auto_finalize, bool auto_finalize,
2435 bool has_auto_dismiss, bool auto_dismiss,
2436 Error **errp)
2438 BlockDriverState *bs;
2439 BlockDriverState *iter;
2440 BlockDriverState *base_bs, *top_bs;
2441 AioContext *aio_context;
2442 Error *local_err = NULL;
2443 int job_flags = JOB_DEFAULT;
2444 uint64_t top_perm, top_shared;
2446 /* TODO We'll eventually have to take a writer lock in this function */
2447 GRAPH_RDLOCK_GUARD_MAINLOOP();
2449 if (!has_speed) {
2450 speed = 0;
2452 if (!has_on_error) {
2453 on_error = BLOCKDEV_ON_ERROR_REPORT;
2455 if (has_auto_finalize && !auto_finalize) {
2456 job_flags |= JOB_MANUAL_FINALIZE;
2458 if (has_auto_dismiss && !auto_dismiss) {
2459 job_flags |= JOB_MANUAL_DISMISS;
2461 if (!has_backing_mask_protocol) {
2462 backing_mask_protocol = false;
2465 /* Important Note:
2466 * libvirt relies on the DeviceNotFound error class in order to probe for
2467 * live commit feature versions; for this to work, we must make sure to
2468 * perform the device lookup before any generic errors that may occur in a
2469 * scenario in which all optional arguments are omitted. */
2470 bs = qmp_get_root_bs(device, &local_err);
2471 if (!bs) {
2472 bs = bdrv_lookup_bs(device, device, NULL);
2473 if (!bs) {
2474 error_free(local_err);
2475 error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2476 "Device '%s' not found", device);
2477 } else {
2478 error_propagate(errp, local_err);
2480 return;
2483 aio_context = bdrv_get_aio_context(bs);
2485 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2486 return;
2489 /* default top_bs is the active layer */
2490 top_bs = bs;
2492 if (top_node && top) {
2493 error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2494 return;
2495 } else if (top_node) {
2496 top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2497 if (top_bs == NULL) {
2498 return;
2500 if (!bdrv_chain_contains(bs, top_bs)) {
2501 error_setg(errp, "'%s' is not in this backing file chain",
2502 top_node);
2503 return;
2505 } else if (top) {
2506 /* This strcmp() is just a shortcut, there is no need to
2507 * refresh @bs's filename. If it mismatches,
2508 * bdrv_find_backing_image() will do the refresh and may still
2509 * return @bs. */
2510 if (strcmp(bs->filename, top) != 0) {
2511 top_bs = bdrv_find_backing_image(bs, top);
2515 if (top_bs == NULL) {
2516 error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2517 return;
2520 assert(bdrv_get_aio_context(top_bs) == aio_context);
2522 if (base_node && base) {
2523 error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2524 return;
2525 } else if (base_node) {
2526 base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2527 if (base_bs == NULL) {
2528 return;
2530 if (!bdrv_chain_contains(top_bs, base_bs)) {
2531 error_setg(errp, "'%s' is not in this backing file chain",
2532 base_node);
2533 return;
2535 } else if (base) {
2536 base_bs = bdrv_find_backing_image(top_bs, base);
2537 if (base_bs == NULL) {
2538 error_setg(errp, "Can't find '%s' in the backing chain", base);
2539 return;
2541 } else {
2542 base_bs = bdrv_find_base(top_bs);
2543 if (base_bs == NULL) {
2544 error_setg(errp, "There is no backimg image");
2545 return;
2549 assert(bdrv_get_aio_context(base_bs) == aio_context);
2551 for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2552 iter = bdrv_filter_or_cow_bs(iter))
2554 if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2555 return;
2559 /* Do not allow attempts to commit an image into itself */
2560 if (top_bs == base_bs) {
2561 error_setg(errp, "cannot commit an image into itself");
2562 return;
2566 * Active commit is required if and only if someone has taken a
2567 * WRITE permission on the top node. Historically, we have always
2568 * used active commit for top nodes, so continue that practice
2569 * lest we possibly break clients that rely on this behavior, e.g.
2570 * to later attach this node to a writing parent.
2571 * (Active commit is never really wrong.)
2573 bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2574 if (top_perm & BLK_PERM_WRITE ||
2575 bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2577 if (backing_file) {
2578 if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2579 error_setg(errp, "'backing-file' specified,"
2580 " but 'top' is the active layer");
2581 } else {
2582 error_setg(errp, "'backing-file' specified, but 'top' has a "
2583 "writer on it");
2585 return;
2587 if (!job_id) {
2589 * Emulate here what block_job_create() does, because it
2590 * is possible that @bs != @top_bs (the block job should
2591 * be named after @bs, even if @top_bs is the actual
2592 * source)
2594 job_id = bdrv_get_device_name(bs);
2596 commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2597 filter_node_name, NULL, NULL, false, &local_err);
2598 } else {
2599 BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2600 if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2601 return;
2603 commit_start(job_id, bs, base_bs, top_bs, job_flags,
2604 speed, on_error, backing_file,
2605 backing_mask_protocol,
2606 filter_node_name, &local_err);
2608 if (local_err != NULL) {
2609 error_propagate(errp, local_err);
2610 return;
2614 /* Common QMP interface for drive-backup and blockdev-backup */
2615 static BlockJob *do_backup_common(BackupCommon *backup,
2616 BlockDriverState *bs,
2617 BlockDriverState *target_bs,
2618 AioContext *aio_context,
2619 JobTxn *txn, Error **errp)
2621 BlockJob *job = NULL;
2622 BdrvDirtyBitmap *bmap = NULL;
2623 BackupPerf perf = { .max_workers = 64 };
2624 int job_flags = JOB_DEFAULT;
2626 if (!backup->has_speed) {
2627 backup->speed = 0;
2629 if (!backup->has_on_source_error) {
2630 backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2632 if (!backup->has_on_target_error) {
2633 backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2635 if (!backup->has_auto_finalize) {
2636 backup->auto_finalize = true;
2638 if (!backup->has_auto_dismiss) {
2639 backup->auto_dismiss = true;
2641 if (!backup->has_compress) {
2642 backup->compress = false;
2645 if (backup->x_perf) {
2646 if (backup->x_perf->has_use_copy_range) {
2647 perf.use_copy_range = backup->x_perf->use_copy_range;
2649 if (backup->x_perf->has_max_workers) {
2650 perf.max_workers = backup->x_perf->max_workers;
2652 if (backup->x_perf->has_max_chunk) {
2653 perf.max_chunk = backup->x_perf->max_chunk;
2657 if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2658 (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2659 /* done before desugaring 'incremental' to print the right message */
2660 if (!backup->bitmap) {
2661 error_setg(errp, "must provide a valid bitmap name for "
2662 "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2663 return NULL;
2667 if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2668 if (backup->has_bitmap_mode &&
2669 backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2670 error_setg(errp, "Bitmap sync mode must be '%s' "
2671 "when using sync mode '%s'",
2672 BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2673 MirrorSyncMode_str(backup->sync));
2674 return NULL;
2676 backup->has_bitmap_mode = true;
2677 backup->sync = MIRROR_SYNC_MODE_BITMAP;
2678 backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2681 if (backup->bitmap) {
2682 bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2683 if (!bmap) {
2684 error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2685 return NULL;
2687 if (!backup->has_bitmap_mode) {
2688 error_setg(errp, "Bitmap sync mode must be given "
2689 "when providing a bitmap");
2690 return NULL;
2692 if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2693 return NULL;
2696 /* This does not produce a useful bitmap artifact: */
2697 if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2698 error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2699 " outputs", MirrorSyncMode_str(backup->sync));
2700 return NULL;
2703 /* If the bitmap isn't used for input or output, this is useless: */
2704 if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2705 backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2706 error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2707 " when combined with sync mode '%s'",
2708 BitmapSyncMode_str(backup->bitmap_mode),
2709 MirrorSyncMode_str(backup->sync));
2710 return NULL;
2714 if (!backup->bitmap && backup->has_bitmap_mode) {
2715 error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2716 return NULL;
2719 if (!backup->auto_finalize) {
2720 job_flags |= JOB_MANUAL_FINALIZE;
2722 if (!backup->auto_dismiss) {
2723 job_flags |= JOB_MANUAL_DISMISS;
2726 job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2727 backup->sync, bmap, backup->bitmap_mode,
2728 backup->compress,
2729 backup->filter_node_name,
2730 &perf,
2731 backup->on_source_error,
2732 backup->on_target_error,
2733 job_flags, NULL, NULL, txn, errp);
2734 return job;
2737 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2739 TransactionAction action = {
2740 .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2741 .u.drive_backup.data = backup,
2743 blockdev_do_action(&action, errp);
2746 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2747 bool flat,
2748 Error **errp)
2750 bool return_flat = has_flat && flat;
2752 return bdrv_named_nodes_list(return_flat, errp);
2755 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2757 GRAPH_RDLOCK_GUARD_MAINLOOP();
2759 return bdrv_get_xdbg_block_graph(errp);
2762 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2764 TransactionAction action = {
2765 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2766 .u.blockdev_backup.data = backup,
2768 blockdev_do_action(&action, errp);
2771 /* Parameter check and block job starting for drive mirroring.
2772 * Caller should hold @device and @target's aio context (must be the same).
2774 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2775 BlockDriverState *target,
2776 const char *replaces,
2777 enum MirrorSyncMode sync,
2778 BlockMirrorBackingMode backing_mode,
2779 bool zero_target,
2780 bool has_speed, int64_t speed,
2781 bool has_granularity, uint32_t granularity,
2782 bool has_buf_size, int64_t buf_size,
2783 bool has_on_source_error,
2784 BlockdevOnError on_source_error,
2785 bool has_on_target_error,
2786 BlockdevOnError on_target_error,
2787 bool has_unmap, bool unmap,
2788 const char *filter_node_name,
2789 bool has_copy_mode, MirrorCopyMode copy_mode,
2790 bool has_auto_finalize, bool auto_finalize,
2791 bool has_auto_dismiss, bool auto_dismiss,
2792 Error **errp)
2794 BlockDriverState *unfiltered_bs;
2795 int job_flags = JOB_DEFAULT;
2797 GLOBAL_STATE_CODE();
2798 GRAPH_RDLOCK_GUARD_MAINLOOP();
2800 if (!has_speed) {
2801 speed = 0;
2803 if (!has_on_source_error) {
2804 on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2806 if (!has_on_target_error) {
2807 on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2809 if (!has_granularity) {
2810 granularity = 0;
2812 if (!has_buf_size) {
2813 buf_size = 0;
2815 if (!has_unmap) {
2816 unmap = true;
2818 if (!has_copy_mode) {
2819 copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2821 if (has_auto_finalize && !auto_finalize) {
2822 job_flags |= JOB_MANUAL_FINALIZE;
2824 if (has_auto_dismiss && !auto_dismiss) {
2825 job_flags |= JOB_MANUAL_DISMISS;
2828 if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2829 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2830 "a value in range [512B, 64MB]");
2831 return;
2833 if (granularity & (granularity - 1)) {
2834 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2835 "a power of 2");
2836 return;
2839 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2840 return;
2842 if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
2843 return;
2846 if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
2847 sync = MIRROR_SYNC_MODE_FULL;
2850 if (!replaces) {
2851 /* We want to mirror from @bs, but keep implicit filters on top */
2852 unfiltered_bs = bdrv_skip_implicit_filters(bs);
2853 if (unfiltered_bs != bs) {
2854 replaces = unfiltered_bs->node_name;
2858 if (replaces) {
2859 BlockDriverState *to_replace_bs;
2860 int64_t bs_size, replace_size;
2862 bs_size = bdrv_getlength(bs);
2863 if (bs_size < 0) {
2864 error_setg_errno(errp, -bs_size, "Failed to query device's size");
2865 return;
2868 to_replace_bs = check_to_replace_node(bs, replaces, errp);
2869 if (!to_replace_bs) {
2870 return;
2873 replace_size = bdrv_getlength(to_replace_bs);
2875 if (replace_size < 0) {
2876 error_setg_errno(errp, -replace_size,
2877 "Failed to query the replacement node's size");
2878 return;
2880 if (bs_size != replace_size) {
2881 error_setg(errp, "cannot replace image with a mirror image of "
2882 "different size");
2883 return;
2887 /* pass the node name to replace to mirror start since it's loose coupling
2888 * and will allow to check whether the node still exist at mirror completion
2890 mirror_start(job_id, bs, target,
2891 replaces, job_flags,
2892 speed, granularity, buf_size, sync, backing_mode, zero_target,
2893 on_source_error, on_target_error, unmap, filter_node_name,
2894 copy_mode, errp);
2897 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
2899 BlockDriverState *bs;
2900 BlockDriverState *target_backing_bs, *target_bs;
2901 AioContext *aio_context;
2902 BlockMirrorBackingMode backing_mode;
2903 Error *local_err = NULL;
2904 QDict *options = NULL;
2905 int flags;
2906 int64_t size;
2907 const char *format = arg->format;
2908 bool zero_target;
2909 int ret;
2911 bs = qmp_get_root_bs(arg->device, errp);
2912 if (!bs) {
2913 return;
2916 /* Early check to avoid creating target */
2917 bdrv_graph_rdlock_main_loop();
2918 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2919 bdrv_graph_rdunlock_main_loop();
2920 return;
2923 aio_context = bdrv_get_aio_context(bs);
2925 if (!arg->has_mode) {
2926 arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2929 if (!arg->format) {
2930 format = (arg->mode == NEW_IMAGE_MODE_EXISTING
2931 ? NULL : bs->drv->format_name);
2934 flags = bs->open_flags | BDRV_O_RDWR;
2935 target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
2936 if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
2937 arg->sync = MIRROR_SYNC_MODE_FULL;
2939 if (arg->sync == MIRROR_SYNC_MODE_NONE) {
2940 target_backing_bs = bs;
2942 bdrv_graph_rdunlock_main_loop();
2944 size = bdrv_getlength(bs);
2945 if (size < 0) {
2946 error_setg_errno(errp, -size, "bdrv_getlength failed");
2947 return;
2950 if (arg->replaces) {
2951 if (!arg->node_name) {
2952 error_setg(errp, "a node-name must be provided when replacing a"
2953 " named node of the graph");
2954 return;
2958 if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
2959 backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
2960 } else {
2961 backing_mode = MIRROR_OPEN_BACKING_CHAIN;
2964 /* Don't open backing image in create() */
2965 flags |= BDRV_O_NO_BACKING;
2967 if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
2968 && arg->mode != NEW_IMAGE_MODE_EXISTING)
2970 /* create new image w/o backing file */
2971 assert(format);
2972 bdrv_img_create(arg->target, format,
2973 NULL, NULL, NULL, size, flags, false, &local_err);
2974 } else {
2975 BlockDriverState *explicit_backing;
2977 switch (arg->mode) {
2978 case NEW_IMAGE_MODE_EXISTING:
2979 break;
2980 case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
2982 * Create new image with backing file.
2983 * Implicit filters should not appear in the filename.
2985 bdrv_graph_rdlock_main_loop();
2986 explicit_backing = bdrv_skip_implicit_filters(target_backing_bs);
2987 bdrv_refresh_filename(explicit_backing);
2988 bdrv_graph_rdunlock_main_loop();
2990 bdrv_img_create(arg->target, format,
2991 explicit_backing->filename,
2992 explicit_backing->drv->format_name,
2993 NULL, size, flags, false, &local_err);
2994 break;
2995 default:
2996 abort();
3000 if (local_err) {
3001 error_propagate(errp, local_err);
3002 return;
3005 options = qdict_new();
3006 if (arg->node_name) {
3007 qdict_put_str(options, "node-name", arg->node_name);
3009 if (format) {
3010 qdict_put_str(options, "driver", format);
3013 /* Mirroring takes care of copy-on-write using the source's backing
3014 * file.
3016 target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3017 if (!target_bs) {
3018 return;
3021 bdrv_graph_rdlock_main_loop();
3022 zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3023 (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3024 !bdrv_has_zero_init(target_bs)));
3025 bdrv_graph_rdunlock_main_loop();
3028 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3029 if (ret < 0) {
3030 bdrv_unref(target_bs);
3031 return;
3034 blockdev_mirror_common(arg->job_id, bs, target_bs,
3035 arg->replaces, arg->sync,
3036 backing_mode, zero_target,
3037 arg->has_speed, arg->speed,
3038 arg->has_granularity, arg->granularity,
3039 arg->has_buf_size, arg->buf_size,
3040 arg->has_on_source_error, arg->on_source_error,
3041 arg->has_on_target_error, arg->on_target_error,
3042 arg->has_unmap, arg->unmap,
3043 NULL,
3044 arg->has_copy_mode, arg->copy_mode,
3045 arg->has_auto_finalize, arg->auto_finalize,
3046 arg->has_auto_dismiss, arg->auto_dismiss,
3047 errp);
3048 bdrv_unref(target_bs);
3051 void qmp_blockdev_mirror(const char *job_id,
3052 const char *device, const char *target,
3053 const char *replaces,
3054 MirrorSyncMode sync,
3055 bool has_speed, int64_t speed,
3056 bool has_granularity, uint32_t granularity,
3057 bool has_buf_size, int64_t buf_size,
3058 bool has_on_source_error,
3059 BlockdevOnError on_source_error,
3060 bool has_on_target_error,
3061 BlockdevOnError on_target_error,
3062 const char *filter_node_name,
3063 bool has_copy_mode, MirrorCopyMode copy_mode,
3064 bool has_auto_finalize, bool auto_finalize,
3065 bool has_auto_dismiss, bool auto_dismiss,
3066 Error **errp)
3068 BlockDriverState *bs;
3069 BlockDriverState *target_bs;
3070 AioContext *aio_context;
3071 BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3072 bool zero_target;
3073 int ret;
3075 bs = qmp_get_root_bs(device, errp);
3076 if (!bs) {
3077 return;
3080 target_bs = bdrv_lookup_bs(target, target, errp);
3081 if (!target_bs) {
3082 return;
3085 zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3087 aio_context = bdrv_get_aio_context(bs);
3089 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3090 if (ret < 0) {
3091 return;
3094 blockdev_mirror_common(job_id, bs, target_bs,
3095 replaces, sync, backing_mode,
3096 zero_target, has_speed, speed,
3097 has_granularity, granularity,
3098 has_buf_size, buf_size,
3099 has_on_source_error, on_source_error,
3100 has_on_target_error, on_target_error,
3101 true, true, filter_node_name,
3102 has_copy_mode, copy_mode,
3103 has_auto_finalize, auto_finalize,
3104 has_auto_dismiss, auto_dismiss,
3105 errp);
3109 * Get a block job using its ID. Called with job_mutex held.
3111 static BlockJob *find_block_job_locked(const char *id, Error **errp)
3113 BlockJob *job;
3115 assert(id != NULL);
3117 job = block_job_get_locked(id);
3119 if (!job) {
3120 error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3121 "Block job '%s' not found", id);
3122 return NULL;
3125 return job;
3128 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3130 BlockJob *job;
3132 JOB_LOCK_GUARD();
3133 job = find_block_job_locked(device, errp);
3135 if (!job) {
3136 return;
3139 block_job_set_speed_locked(job, speed, errp);
3142 void qmp_block_job_cancel(const char *device,
3143 bool has_force, bool force, Error **errp)
3145 BlockJob *job;
3147 JOB_LOCK_GUARD();
3148 job = find_block_job_locked(device, errp);
3150 if (!job) {
3151 return;
3154 if (!has_force) {
3155 force = false;
3158 if (job_user_paused_locked(&job->job) && !force) {
3159 error_setg(errp, "The block job for device '%s' is currently paused",
3160 device);
3161 return;
3164 trace_qmp_block_job_cancel(job);
3165 job_user_cancel_locked(&job->job, force, errp);
3168 void qmp_block_job_pause(const char *device, Error **errp)
3170 BlockJob *job;
3172 JOB_LOCK_GUARD();
3173 job = find_block_job_locked(device, errp);
3175 if (!job) {
3176 return;
3179 trace_qmp_block_job_pause(job);
3180 job_user_pause_locked(&job->job, errp);
3183 void qmp_block_job_resume(const char *device, Error **errp)
3185 BlockJob *job;
3187 JOB_LOCK_GUARD();
3188 job = find_block_job_locked(device, errp);
3190 if (!job) {
3191 return;
3194 trace_qmp_block_job_resume(job);
3195 job_user_resume_locked(&job->job, errp);
3198 void qmp_block_job_complete(const char *device, Error **errp)
3200 BlockJob *job;
3202 JOB_LOCK_GUARD();
3203 job = find_block_job_locked(device, errp);
3205 if (!job) {
3206 return;
3209 trace_qmp_block_job_complete(job);
3210 job_complete_locked(&job->job, errp);
3213 void qmp_block_job_finalize(const char *id, Error **errp)
3215 BlockJob *job;
3217 JOB_LOCK_GUARD();
3218 job = find_block_job_locked(id, errp);
3220 if (!job) {
3221 return;
3224 trace_qmp_block_job_finalize(job);
3225 job_ref_locked(&job->job);
3226 job_finalize_locked(&job->job, errp);
3228 job_unref_locked(&job->job);
3231 void qmp_block_job_dismiss(const char *id, Error **errp)
3233 BlockJob *bjob;
3234 Job *job;
3236 JOB_LOCK_GUARD();
3237 bjob = find_block_job_locked(id, errp);
3239 if (!bjob) {
3240 return;
3243 trace_qmp_block_job_dismiss(bjob);
3244 job = &bjob->job;
3245 job_dismiss_locked(&job, errp);
3248 void qmp_block_job_change(BlockJobChangeOptions *opts, Error **errp)
3250 BlockJob *job;
3252 JOB_LOCK_GUARD();
3253 job = find_block_job_locked(opts->id, errp);
3255 if (!job) {
3256 return;
3259 block_job_change_locked(job, opts, errp);
3262 void qmp_change_backing_file(const char *device,
3263 const char *image_node_name,
3264 const char *backing_file,
3265 Error **errp)
3267 BlockDriverState *bs = NULL;
3268 BlockDriverState *image_bs = NULL;
3269 Error *local_err = NULL;
3270 bool ro;
3271 int ret;
3273 bs = qmp_get_root_bs(device, errp);
3274 if (!bs) {
3275 return;
3278 bdrv_graph_rdlock_main_loop();
3280 image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3281 if (local_err) {
3282 error_propagate(errp, local_err);
3283 goto out_rdlock;
3286 if (!image_bs) {
3287 error_setg(errp, "image file not found");
3288 goto out_rdlock;
3291 if (bdrv_find_base(image_bs) == image_bs) {
3292 error_setg(errp, "not allowing backing file change on an image "
3293 "without a backing file");
3294 goto out_rdlock;
3297 /* even though we are not necessarily operating on bs, we need it to
3298 * determine if block ops are currently prohibited on the chain */
3299 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3300 goto out_rdlock;
3303 /* final sanity check */
3304 if (!bdrv_chain_contains(bs, image_bs)) {
3305 error_setg(errp, "'%s' and image file are not in the same chain",
3306 device);
3307 goto out_rdlock;
3309 bdrv_graph_rdunlock_main_loop();
3311 /* if not r/w, reopen to make r/w */
3312 ro = bdrv_is_read_only(image_bs);
3314 if (ro) {
3315 if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3316 return;
3320 ret = bdrv_change_backing_file(image_bs, backing_file,
3321 image_bs->drv ? image_bs->drv->format_name : "",
3322 false);
3324 if (ret < 0) {
3325 error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3326 backing_file);
3327 /* don't exit here, so we can try to restore open flags if
3328 * appropriate */
3331 if (ro) {
3332 bdrv_reopen_set_read_only(image_bs, true, errp);
3334 return;
3336 out_rdlock:
3337 bdrv_graph_rdunlock_main_loop();
3340 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3342 BlockDriverState *bs;
3343 QObject *obj;
3344 Visitor *v = qobject_output_visitor_new(&obj);
3345 QDict *qdict;
3347 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3348 visit_complete(v, &obj);
3349 qdict = qobject_to(QDict, obj);
3351 qdict_flatten(qdict);
3353 if (!qdict_get_try_str(qdict, "node-name")) {
3354 error_setg(errp, "'node-name' must be specified for the root node");
3355 goto fail;
3358 bs = bds_tree_init(qdict, errp);
3359 if (!bs) {
3360 goto fail;
3363 bdrv_set_monitor_owned(bs);
3365 fail:
3366 visit_free(v);
3369 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3371 BlockReopenQueue *queue = NULL;
3373 /* Add each one of the BDS that we want to reopen to the queue */
3374 for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3375 BlockdevOptions *options = reopen_list->value;
3376 BlockDriverState *bs;
3377 QObject *obj;
3378 Visitor *v;
3379 QDict *qdict;
3381 /* Check for the selected node name */
3382 if (!options->node_name) {
3383 error_setg(errp, "node-name not specified");
3384 goto fail;
3387 bs = bdrv_find_node(options->node_name);
3388 if (!bs) {
3389 error_setg(errp, "Failed to find node with node-name='%s'",
3390 options->node_name);
3391 goto fail;
3394 /* Put all options in a QDict and flatten it */
3395 v = qobject_output_visitor_new(&obj);
3396 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3397 visit_complete(v, &obj);
3398 visit_free(v);
3400 qdict = qobject_to(QDict, obj);
3402 qdict_flatten(qdict);
3404 queue = bdrv_reopen_queue(queue, bs, qdict, false);
3407 /* Perform the reopen operation */
3408 bdrv_reopen_multiple(queue, errp);
3409 queue = NULL;
3411 fail:
3412 bdrv_reopen_queue_free(queue);
3415 void qmp_blockdev_del(const char *node_name, Error **errp)
3417 BlockDriverState *bs;
3419 GLOBAL_STATE_CODE();
3420 GRAPH_RDLOCK_GUARD_MAINLOOP();
3422 bs = bdrv_find_node(node_name);
3423 if (!bs) {
3424 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3425 return;
3427 if (bdrv_has_blk(bs)) {
3428 error_setg(errp, "Node %s is in use", node_name);
3429 return;
3432 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3433 return;
3436 if (!QTAILQ_IN_USE(bs, monitor_list)) {
3437 error_setg(errp, "Node %s is not owned by the monitor",
3438 bs->node_name);
3439 return;
3442 if (bs->refcnt > 1) {
3443 error_setg(errp, "Block device %s is in use",
3444 bdrv_get_device_or_node_name(bs));
3445 return;
3448 QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3449 bdrv_unref(bs);
3452 static BdrvChild * GRAPH_RDLOCK
3453 bdrv_find_child(BlockDriverState *parent_bs, const char *child_name)
3455 BdrvChild *child;
3457 QLIST_FOREACH(child, &parent_bs->children, next) {
3458 if (strcmp(child->name, child_name) == 0) {
3459 return child;
3463 return NULL;
3466 void qmp_x_blockdev_change(const char *parent, const char *child,
3467 const char *node, Error **errp)
3469 BlockDriverState *parent_bs, *new_bs = NULL;
3470 BdrvChild *p_child;
3472 bdrv_graph_wrlock();
3474 parent_bs = bdrv_lookup_bs(parent, parent, errp);
3475 if (!parent_bs) {
3476 goto out;
3479 if (!child == !node) {
3480 if (child) {
3481 error_setg(errp, "The parameters child and node are in conflict");
3482 } else {
3483 error_setg(errp, "Either child or node must be specified");
3485 goto out;
3488 if (child) {
3489 p_child = bdrv_find_child(parent_bs, child);
3490 if (!p_child) {
3491 error_setg(errp, "Node '%s' does not have child '%s'",
3492 parent, child);
3493 goto out;
3495 bdrv_del_child(parent_bs, p_child, errp);
3498 if (node) {
3499 new_bs = bdrv_find_node(node);
3500 if (!new_bs) {
3501 error_setg(errp, "Node '%s' not found", node);
3502 goto out;
3504 bdrv_add_child(parent_bs, new_bs, errp);
3507 out:
3508 bdrv_graph_wrunlock();
3511 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3513 BlockJobInfoList *head = NULL, **tail = &head;
3514 BlockJob *job;
3516 JOB_LOCK_GUARD();
3518 for (job = block_job_next_locked(NULL); job;
3519 job = block_job_next_locked(job)) {
3520 BlockJobInfo *value;
3522 if (block_job_is_internal(job)) {
3523 continue;
3525 value = block_job_query_locked(job, errp);
3526 if (!value) {
3527 qapi_free_BlockJobInfoList(head);
3528 return NULL;
3530 QAPI_LIST_APPEND(tail, value);
3533 return head;
3536 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3537 bool has_force, bool force, Error **errp)
3539 AioContext *new_context;
3540 BlockDriverState *bs;
3542 GRAPH_RDLOCK_GUARD_MAINLOOP();
3544 bs = bdrv_find_node(node_name);
3545 if (!bs) {
3546 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3547 return;
3550 /* Protects against accidents. */
3551 if (!(has_force && force) && bdrv_has_blk(bs)) {
3552 error_setg(errp, "Node %s is associated with a BlockBackend and could "
3553 "be in use (use force=true to override this check)",
3554 node_name);
3555 return;
3558 if (iothread->type == QTYPE_QSTRING) {
3559 IOThread *obj = iothread_by_id(iothread->u.s);
3560 if (!obj) {
3561 error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3562 return;
3565 new_context = iothread_get_aio_context(obj);
3566 } else {
3567 new_context = qemu_get_aio_context();
3570 bdrv_try_change_aio_context(bs, new_context, NULL, errp);
3573 QemuOptsList qemu_common_drive_opts = {
3574 .name = "drive",
3575 .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3576 .desc = {
3578 .name = "snapshot",
3579 .type = QEMU_OPT_BOOL,
3580 .help = "enable/disable snapshot mode",
3582 .name = "aio",
3583 .type = QEMU_OPT_STRING,
3584 .help = "host AIO implementation (threads, native, io_uring)",
3586 .name = BDRV_OPT_CACHE_WB,
3587 .type = QEMU_OPT_BOOL,
3588 .help = "Enable writeback mode",
3590 .name = "format",
3591 .type = QEMU_OPT_STRING,
3592 .help = "disk format (raw, qcow2, ...)",
3594 .name = "rerror",
3595 .type = QEMU_OPT_STRING,
3596 .help = "read error action",
3598 .name = "werror",
3599 .type = QEMU_OPT_STRING,
3600 .help = "write error action",
3602 .name = BDRV_OPT_READ_ONLY,
3603 .type = QEMU_OPT_BOOL,
3604 .help = "open drive file as read-only",
3607 THROTTLE_OPTS,
3610 .name = "throttling.group",
3611 .type = QEMU_OPT_STRING,
3612 .help = "name of the block throttling group",
3614 .name = "copy-on-read",
3615 .type = QEMU_OPT_BOOL,
3616 .help = "copy read data from backing file into image file",
3618 .name = "detect-zeroes",
3619 .type = QEMU_OPT_STRING,
3620 .help = "try to optimize zero writes (off, on, unmap)",
3622 .name = "stats-account-invalid",
3623 .type = QEMU_OPT_BOOL,
3624 .help = "whether to account for invalid I/O operations "
3625 "in the statistics",
3627 .name = "stats-account-failed",
3628 .type = QEMU_OPT_BOOL,
3629 .help = "whether to account for failed I/O operations "
3630 "in the statistics",
3632 { /* end of list */ }
3636 QemuOptsList qemu_drive_opts = {
3637 .name = "drive",
3638 .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3639 .desc = {
3641 * no elements => accept any params
3642 * validation will happen later
3644 { /* end of list */ }