net/net_rx_pkt: Use iovec for net_rx_pkt_set_protocols()
[qemu/kevin.git] / blockdev.c
blob5d56b79df4ad97e18b9c967b55ae551033dc1b01
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, because it gets desugared into -device,
259 * 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_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 unsigned long long length;
345 const char *str = qstring_get_str(qobject_to(QString,
346 entry->value));
347 if (parse_uint_full(str, &length, 10) == 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 AioContext *ctx = bdrv_get_aio_context(bs);
690 aio_context_acquire(ctx);
691 bdrv_unref(bs);
692 aio_context_release(ctx);
696 /* Iterates over the list of monitor-owned BlockDriverStates */
697 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
699 GLOBAL_STATE_CODE();
700 return bs ? QTAILQ_NEXT(bs, monitor_list)
701 : QTAILQ_FIRST(&monitor_bdrv_states);
704 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
705 Error **errp)
707 const char *value;
709 value = qemu_opt_get(opts, from);
710 if (value) {
711 if (qemu_opt_find(opts, to)) {
712 error_setg(errp, "'%s' and its alias '%s' can't be used at the "
713 "same time", to, from);
714 return false;
718 /* rename all items in opts */
719 while ((value = qemu_opt_get(opts, from))) {
720 qemu_opt_set(opts, to, value, &error_abort);
721 qemu_opt_unset(opts, from);
723 return true;
726 QemuOptsList qemu_legacy_drive_opts = {
727 .name = "drive",
728 .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
729 .desc = {
731 .name = "bus",
732 .type = QEMU_OPT_NUMBER,
733 .help = "bus number",
735 .name = "unit",
736 .type = QEMU_OPT_NUMBER,
737 .help = "unit number (i.e. lun for scsi)",
739 .name = "index",
740 .type = QEMU_OPT_NUMBER,
741 .help = "index number",
743 .name = "media",
744 .type = QEMU_OPT_STRING,
745 .help = "media type (disk, cdrom)",
747 .name = "if",
748 .type = QEMU_OPT_STRING,
749 .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
751 .name = "file",
752 .type = QEMU_OPT_STRING,
753 .help = "file name",
756 /* Options that are passed on, but have special semantics with -drive */
758 .name = BDRV_OPT_READ_ONLY,
759 .type = QEMU_OPT_BOOL,
760 .help = "open drive file as read-only",
762 .name = "rerror",
763 .type = QEMU_OPT_STRING,
764 .help = "read error action",
766 .name = "werror",
767 .type = QEMU_OPT_STRING,
768 .help = "write error action",
770 .name = "copy-on-read",
771 .type = QEMU_OPT_BOOL,
772 .help = "copy read data from backing file into image file",
775 { /* end of list */ }
779 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
780 Error **errp)
782 const char *value;
783 BlockBackend *blk;
784 DriveInfo *dinfo = NULL;
785 QDict *bs_opts;
786 QemuOpts *legacy_opts;
787 DriveMediaType media = MEDIA_DISK;
788 BlockInterfaceType type;
789 int max_devs, bus_id, unit_id, index;
790 const char *werror, *rerror;
791 bool read_only = false;
792 bool copy_on_read;
793 const char *filename;
794 int i;
796 GLOBAL_STATE_CODE();
798 /* Change legacy command line options into QMP ones */
799 static const struct {
800 const char *from;
801 const char *to;
802 } opt_renames[] = {
803 { "iops", "throttling.iops-total" },
804 { "iops_rd", "throttling.iops-read" },
805 { "iops_wr", "throttling.iops-write" },
807 { "bps", "throttling.bps-total" },
808 { "bps_rd", "throttling.bps-read" },
809 { "bps_wr", "throttling.bps-write" },
811 { "iops_max", "throttling.iops-total-max" },
812 { "iops_rd_max", "throttling.iops-read-max" },
813 { "iops_wr_max", "throttling.iops-write-max" },
815 { "bps_max", "throttling.bps-total-max" },
816 { "bps_rd_max", "throttling.bps-read-max" },
817 { "bps_wr_max", "throttling.bps-write-max" },
819 { "iops_size", "throttling.iops-size" },
821 { "group", "throttling.group" },
823 { "readonly", BDRV_OPT_READ_ONLY },
826 for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
827 if (!qemu_opt_rename(all_opts, opt_renames[i].from,
828 opt_renames[i].to, errp)) {
829 return NULL;
833 value = qemu_opt_get(all_opts, "cache");
834 if (value) {
835 int flags = 0;
836 bool writethrough;
838 if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
839 error_setg(errp, "invalid cache option");
840 return NULL;
843 /* Specific options take precedence */
844 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
845 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
846 !writethrough, &error_abort);
848 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
849 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
850 !!(flags & BDRV_O_NOCACHE), &error_abort);
852 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
853 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
854 !!(flags & BDRV_O_NO_FLUSH), &error_abort);
856 qemu_opt_unset(all_opts, "cache");
859 /* Get a QDict for processing the options */
860 bs_opts = qdict_new();
861 qemu_opts_to_qdict(all_opts, bs_opts);
863 legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
864 &error_abort);
865 if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
866 goto fail;
869 /* Media type */
870 value = qemu_opt_get(legacy_opts, "media");
871 if (value) {
872 if (!strcmp(value, "disk")) {
873 media = MEDIA_DISK;
874 } else if (!strcmp(value, "cdrom")) {
875 media = MEDIA_CDROM;
876 read_only = true;
877 } else {
878 error_setg(errp, "'%s' invalid media", value);
879 goto fail;
883 /* copy-on-read is disabled with a warning for read-only devices */
884 read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
885 copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
887 if (read_only && copy_on_read) {
888 warn_report("disabling copy-on-read on read-only drive");
889 copy_on_read = false;
892 qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
893 qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
895 /* Controller type */
896 value = qemu_opt_get(legacy_opts, "if");
897 if (value) {
898 for (type = 0;
899 type < IF_COUNT && strcmp(value, if_name[type]);
900 type++) {
902 if (type == IF_COUNT) {
903 error_setg(errp, "unsupported bus type '%s'", value);
904 goto fail;
906 } else {
907 type = block_default_type;
910 /* Device address specified by bus/unit or index.
911 * If none was specified, try to find the first free one. */
912 bus_id = qemu_opt_get_number(legacy_opts, "bus", 0);
913 unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
914 index = qemu_opt_get_number(legacy_opts, "index", -1);
916 max_devs = if_max_devs[type];
918 if (index != -1) {
919 if (bus_id != 0 || unit_id != -1) {
920 error_setg(errp, "index cannot be used with bus and unit");
921 goto fail;
923 bus_id = drive_index_to_bus_id(type, index);
924 unit_id = drive_index_to_unit_id(type, index);
927 if (unit_id == -1) {
928 unit_id = 0;
929 while (drive_get(type, bus_id, unit_id) != NULL) {
930 unit_id++;
931 if (max_devs && unit_id >= max_devs) {
932 unit_id -= max_devs;
933 bus_id++;
938 if (max_devs && unit_id >= max_devs) {
939 error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
940 goto fail;
943 if (drive_get(type, bus_id, unit_id) != NULL) {
944 error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
945 bus_id, unit_id, index);
946 goto fail;
949 /* no id supplied -> create one */
950 if (qemu_opts_id(all_opts) == NULL) {
951 char *new_id;
952 const char *mediastr = "";
953 if (type == IF_IDE || type == IF_SCSI) {
954 mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
956 if (max_devs) {
957 new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
958 mediastr, unit_id);
959 } else {
960 new_id = g_strdup_printf("%s%s%i", if_name[type],
961 mediastr, unit_id);
963 qdict_put_str(bs_opts, "id", new_id);
964 g_free(new_id);
967 /* Add virtio block device */
968 if (type == IF_VIRTIO) {
969 QemuOpts *devopts;
970 devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
971 &error_abort);
972 qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
973 qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
974 &error_abort);
977 filename = qemu_opt_get(legacy_opts, "file");
979 /* Check werror/rerror compatibility with if=... */
980 werror = qemu_opt_get(legacy_opts, "werror");
981 if (werror != NULL) {
982 if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
983 type != IF_NONE) {
984 error_setg(errp, "werror is not supported by this bus type");
985 goto fail;
987 qdict_put_str(bs_opts, "werror", werror);
990 rerror = qemu_opt_get(legacy_opts, "rerror");
991 if (rerror != NULL) {
992 if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
993 type != IF_NONE) {
994 error_setg(errp, "rerror is not supported by this bus type");
995 goto fail;
997 qdict_put_str(bs_opts, "rerror", rerror);
1000 /* Actual block device init: Functionality shared with blockdev-add */
1001 blk = blockdev_init(filename, bs_opts, errp);
1002 bs_opts = NULL;
1003 if (!blk) {
1004 goto fail;
1007 /* Create legacy DriveInfo */
1008 dinfo = g_malloc0(sizeof(*dinfo));
1009 dinfo->opts = all_opts;
1011 dinfo->type = type;
1012 dinfo->bus = bus_id;
1013 dinfo->unit = unit_id;
1015 blk_set_legacy_dinfo(blk, dinfo);
1017 switch(type) {
1018 case IF_IDE:
1019 case IF_SCSI:
1020 case IF_XEN:
1021 case IF_NONE:
1022 dinfo->media_cd = media == MEDIA_CDROM;
1023 break;
1024 default:
1025 break;
1028 fail:
1029 qemu_opts_del(legacy_opts);
1030 qobject_unref(bs_opts);
1031 return dinfo;
1034 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1036 BlockDriverState *bs;
1037 AioContext *aio_context;
1039 bs = bdrv_lookup_bs(name, name, errp);
1040 if (bs == NULL) {
1041 return NULL;
1044 if (!bdrv_is_root_node(bs)) {
1045 error_setg(errp, "Need a root block node");
1046 return NULL;
1049 aio_context = bdrv_get_aio_context(bs);
1050 aio_context_acquire(aio_context);
1052 if (!bdrv_is_inserted(bs)) {
1053 error_setg(errp, "Device has no medium");
1054 bs = NULL;
1057 aio_context_release(aio_context);
1059 return bs;
1062 static void blockdev_do_action(TransactionAction *action, Error **errp)
1064 TransactionActionList list;
1066 list.value = action;
1067 list.next = NULL;
1068 qmp_transaction(&list, NULL, errp);
1071 void qmp_blockdev_snapshot_sync(const char *device, const char *node_name,
1072 const char *snapshot_file,
1073 const char *snapshot_node_name,
1074 const char *format,
1075 bool has_mode, NewImageMode mode, Error **errp)
1077 BlockdevSnapshotSync snapshot = {
1078 .device = (char *) device,
1079 .node_name = (char *) node_name,
1080 .snapshot_file = (char *) snapshot_file,
1081 .snapshot_node_name = (char *) snapshot_node_name,
1082 .format = (char *) format,
1083 .has_mode = has_mode,
1084 .mode = mode,
1086 TransactionAction action = {
1087 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1088 .u.blockdev_snapshot_sync.data = &snapshot,
1090 blockdev_do_action(&action, errp);
1093 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1094 Error **errp)
1096 BlockdevSnapshot snapshot_data = {
1097 .node = (char *) node,
1098 .overlay = (char *) overlay
1100 TransactionAction action = {
1101 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1102 .u.blockdev_snapshot.data = &snapshot_data,
1104 blockdev_do_action(&action, errp);
1107 void qmp_blockdev_snapshot_internal_sync(const char *device,
1108 const char *name,
1109 Error **errp)
1111 BlockdevSnapshotInternal snapshot = {
1112 .device = (char *) device,
1113 .name = (char *) name
1115 TransactionAction action = {
1116 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1117 .u.blockdev_snapshot_internal_sync.data = &snapshot,
1119 blockdev_do_action(&action, errp);
1122 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1123 const char *id,
1124 const char *name,
1125 Error **errp)
1127 BlockDriverState *bs;
1128 AioContext *aio_context;
1129 QEMUSnapshotInfo sn;
1130 Error *local_err = NULL;
1131 SnapshotInfo *info = NULL;
1132 int ret;
1134 bs = qmp_get_root_bs(device, errp);
1135 if (!bs) {
1136 return NULL;
1138 aio_context = bdrv_get_aio_context(bs);
1139 aio_context_acquire(aio_context);
1141 if (!id && !name) {
1142 error_setg(errp, "Name or id must be provided");
1143 goto out_aio_context;
1146 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1147 goto out_aio_context;
1150 ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1151 if (local_err) {
1152 error_propagate(errp, local_err);
1153 goto out_aio_context;
1155 if (!ret) {
1156 error_setg(errp,
1157 "Snapshot with id '%s' and name '%s' does not exist on "
1158 "device '%s'",
1159 STR_OR_NULL(id), STR_OR_NULL(name), device);
1160 goto out_aio_context;
1163 bdrv_snapshot_delete(bs, id, name, &local_err);
1164 if (local_err) {
1165 error_propagate(errp, local_err);
1166 goto out_aio_context;
1169 aio_context_release(aio_context);
1171 info = g_new0(SnapshotInfo, 1);
1172 info->id = g_strdup(sn.id_str);
1173 info->name = g_strdup(sn.name);
1174 info->date_nsec = sn.date_nsec;
1175 info->date_sec = sn.date_sec;
1176 info->vm_state_size = sn.vm_state_size;
1177 info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1178 info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1179 if (sn.icount != -1ULL) {
1180 info->icount = sn.icount;
1181 info->has_icount = true;
1184 return info;
1186 out_aio_context:
1187 aio_context_release(aio_context);
1188 return NULL;
1191 /* internal snapshot private data */
1192 typedef struct InternalSnapshotState {
1193 BlockDriverState *bs;
1194 QEMUSnapshotInfo sn;
1195 bool created;
1196 } InternalSnapshotState;
1198 static void internal_snapshot_abort(void *opaque);
1199 static void internal_snapshot_clean(void *opaque);
1200 TransactionActionDrv internal_snapshot_drv = {
1201 .abort = internal_snapshot_abort,
1202 .clean = internal_snapshot_clean,
1205 static void internal_snapshot_action(BlockdevSnapshotInternal *internal,
1206 Transaction *tran, Error **errp)
1208 Error *local_err = NULL;
1209 const char *device;
1210 const char *name;
1211 BlockDriverState *bs;
1212 QEMUSnapshotInfo old_sn, *sn;
1213 bool ret;
1214 int64_t rt;
1215 InternalSnapshotState *state = g_new0(InternalSnapshotState, 1);
1216 AioContext *aio_context;
1217 int ret1;
1219 tran_add(tran, &internal_snapshot_drv, state);
1221 device = internal->device;
1222 name = internal->name;
1224 bs = qmp_get_root_bs(device, errp);
1225 if (!bs) {
1226 return;
1229 aio_context = bdrv_get_aio_context(bs);
1230 aio_context_acquire(aio_context);
1232 state->bs = bs;
1234 /* Paired with .clean() */
1235 bdrv_drained_begin(bs);
1237 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1238 goto out;
1241 if (bdrv_is_read_only(bs)) {
1242 error_setg(errp, "Device '%s' is read only", device);
1243 goto out;
1246 if (!bdrv_can_snapshot(bs)) {
1247 error_setg(errp, "Block format '%s' used by device '%s' "
1248 "does not support internal snapshots",
1249 bs->drv->format_name, device);
1250 goto out;
1253 if (!strlen(name)) {
1254 error_setg(errp, "Name is empty");
1255 goto out;
1258 /* check whether a snapshot with name exist */
1259 ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1260 &local_err);
1261 if (local_err) {
1262 error_propagate(errp, local_err);
1263 goto out;
1264 } else if (ret) {
1265 error_setg(errp,
1266 "Snapshot with name '%s' already exists on device '%s'",
1267 name, device);
1268 goto out;
1271 /* 3. take the snapshot */
1272 sn = &state->sn;
1273 pstrcpy(sn->name, sizeof(sn->name), name);
1274 rt = g_get_real_time();
1275 sn->date_sec = rt / G_USEC_PER_SEC;
1276 sn->date_nsec = (rt % G_USEC_PER_SEC) * 1000;
1277 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1278 if (replay_mode != REPLAY_MODE_NONE) {
1279 sn->icount = replay_get_current_icount();
1280 } else {
1281 sn->icount = -1ULL;
1284 ret1 = bdrv_snapshot_create(bs, sn);
1285 if (ret1 < 0) {
1286 error_setg_errno(errp, -ret1,
1287 "Failed to create snapshot '%s' on device '%s'",
1288 name, device);
1289 goto out;
1292 /* 4. succeed, mark a snapshot is created */
1293 state->created = true;
1295 out:
1296 aio_context_release(aio_context);
1299 static void internal_snapshot_abort(void *opaque)
1301 InternalSnapshotState *state = opaque;
1302 BlockDriverState *bs = state->bs;
1303 QEMUSnapshotInfo *sn = &state->sn;
1304 AioContext *aio_context;
1305 Error *local_error = NULL;
1307 if (!state->created) {
1308 return;
1311 aio_context = bdrv_get_aio_context(state->bs);
1312 aio_context_acquire(aio_context);
1314 if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1315 error_reportf_err(local_error,
1316 "Failed to delete snapshot with id '%s' and "
1317 "name '%s' on device '%s' in abort: ",
1318 sn->id_str, sn->name,
1319 bdrv_get_device_name(bs));
1322 aio_context_release(aio_context);
1325 static void internal_snapshot_clean(void *opaque)
1327 g_autofree InternalSnapshotState *state = opaque;
1328 AioContext *aio_context;
1330 if (!state->bs) {
1331 return;
1334 aio_context = bdrv_get_aio_context(state->bs);
1335 aio_context_acquire(aio_context);
1337 bdrv_drained_end(state->bs);
1339 aio_context_release(aio_context);
1342 /* external snapshot private data */
1343 typedef struct ExternalSnapshotState {
1344 BlockDriverState *old_bs;
1345 BlockDriverState *new_bs;
1346 bool overlay_appended;
1347 } ExternalSnapshotState;
1349 static void external_snapshot_commit(void *opaque);
1350 static void external_snapshot_abort(void *opaque);
1351 static void external_snapshot_clean(void *opaque);
1352 TransactionActionDrv external_snapshot_drv = {
1353 .commit = external_snapshot_commit,
1354 .abort = external_snapshot_abort,
1355 .clean = external_snapshot_clean,
1358 static void external_snapshot_action(TransactionAction *action,
1359 Transaction *tran, Error **errp)
1361 int ret;
1362 int flags = 0;
1363 QDict *options = NULL;
1364 Error *local_err = NULL;
1365 /* Device and node name of the image to generate the snapshot from */
1366 const char *device;
1367 const char *node_name;
1368 /* Reference to the new image (for 'blockdev-snapshot') */
1369 const char *snapshot_ref;
1370 /* File name of the new image (for 'blockdev-snapshot-sync') */
1371 const char *new_image_file;
1372 ExternalSnapshotState *state = g_new0(ExternalSnapshotState, 1);
1373 AioContext *aio_context;
1374 uint64_t perm, shared;
1376 tran_add(tran, &external_snapshot_drv, state);
1378 /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1379 * purpose but a different set of parameters */
1380 switch (action->type) {
1381 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1383 BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1384 device = s->node;
1385 node_name = s->node;
1386 new_image_file = NULL;
1387 snapshot_ref = s->overlay;
1389 break;
1390 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1392 BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1393 device = s->device;
1394 node_name = s->node_name;
1395 new_image_file = s->snapshot_file;
1396 snapshot_ref = NULL;
1398 break;
1399 default:
1400 g_assert_not_reached();
1403 /* start processing */
1405 state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1406 if (!state->old_bs) {
1407 return;
1410 aio_context = bdrv_get_aio_context(state->old_bs);
1411 aio_context_acquire(aio_context);
1413 /* Paired with .clean() */
1414 bdrv_drained_begin(state->old_bs);
1416 if (!bdrv_is_inserted(state->old_bs)) {
1417 error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1418 goto out;
1421 if (bdrv_op_is_blocked(state->old_bs,
1422 BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1423 goto out;
1426 if (!bdrv_is_read_only(state->old_bs)) {
1427 if (bdrv_flush(state->old_bs)) {
1428 error_setg(errp, QERR_IO_ERROR);
1429 goto out;
1433 if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1434 BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1435 const char *format = s->format ?: "qcow2";
1436 enum NewImageMode mode;
1437 const char *snapshot_node_name = s->snapshot_node_name;
1439 if (node_name && !snapshot_node_name) {
1440 error_setg(errp, "New overlay node-name missing");
1441 goto out;
1444 if (snapshot_node_name &&
1445 bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1446 error_setg(errp, "New overlay node-name already in use");
1447 goto out;
1450 flags = state->old_bs->open_flags;
1451 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1452 flags |= BDRV_O_NO_BACKING;
1454 /* create new image w/backing file */
1455 mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1456 if (mode != NEW_IMAGE_MODE_EXISTING) {
1457 int64_t size = bdrv_getlength(state->old_bs);
1458 if (size < 0) {
1459 error_setg_errno(errp, -size, "bdrv_getlength failed");
1460 goto out;
1462 bdrv_refresh_filename(state->old_bs);
1464 aio_context_release(aio_context);
1465 bdrv_img_create(new_image_file, format,
1466 state->old_bs->filename,
1467 state->old_bs->drv->format_name,
1468 NULL, size, flags, false, &local_err);
1469 aio_context_acquire(aio_context);
1471 if (local_err) {
1472 error_propagate(errp, local_err);
1473 goto out;
1477 options = qdict_new();
1478 if (snapshot_node_name) {
1479 qdict_put_str(options, "node-name", snapshot_node_name);
1481 qdict_put_str(options, "driver", format);
1484 state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1485 errp);
1486 /* We will manually add the backing_hd field to the bs later */
1487 if (!state->new_bs) {
1488 goto out;
1492 * Allow attaching a backing file to an overlay that's already in use only
1493 * if the parents don't assume that they are already seeing a valid image.
1494 * (Specifically, allow it as a mirror target, which is write-only access.)
1496 bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1497 if (perm & BLK_PERM_CONSISTENT_READ) {
1498 error_setg(errp, "The overlay is already in use");
1499 goto out;
1502 if (state->new_bs->drv->is_filter) {
1503 error_setg(errp, "Filters cannot be used as overlays");
1504 goto out;
1507 if (bdrv_cow_child(state->new_bs)) {
1508 error_setg(errp, "The overlay already has a backing image");
1509 goto out;
1512 if (!state->new_bs->drv->supports_backing) {
1513 error_setg(errp, "The overlay does not support backing images");
1514 goto out;
1517 ret = bdrv_append(state->new_bs, state->old_bs, errp);
1518 if (ret < 0) {
1519 goto out;
1521 state->overlay_appended = true;
1523 out:
1524 aio_context_release(aio_context);
1527 static void external_snapshot_commit(void *opaque)
1529 ExternalSnapshotState *state = opaque;
1530 AioContext *aio_context;
1532 aio_context = bdrv_get_aio_context(state->old_bs);
1533 aio_context_acquire(aio_context);
1535 /* We don't need (or want) to use the transactional
1536 * bdrv_reopen_multiple() across all the entries at once, because we
1537 * don't want to abort all of them if one of them fails the reopen */
1538 if (!qatomic_read(&state->old_bs->copy_on_read)) {
1539 bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1542 aio_context_release(aio_context);
1545 static void external_snapshot_abort(void *opaque)
1547 ExternalSnapshotState *state = opaque;
1548 if (state->new_bs) {
1549 if (state->overlay_appended) {
1550 AioContext *aio_context;
1551 AioContext *tmp_context;
1552 int ret;
1554 aio_context = bdrv_get_aio_context(state->old_bs);
1555 aio_context_acquire(aio_context);
1557 bdrv_ref(state->old_bs); /* we can't let bdrv_set_backind_hd()
1558 close state->old_bs; we need it */
1559 bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1562 * The call to bdrv_set_backing_hd() above returns state->old_bs to
1563 * the main AioContext. As we're still going to be using it, return
1564 * it to the AioContext it was before.
1566 tmp_context = bdrv_get_aio_context(state->old_bs);
1567 if (aio_context != tmp_context) {
1568 aio_context_release(aio_context);
1569 aio_context_acquire(tmp_context);
1571 ret = bdrv_try_change_aio_context(state->old_bs,
1572 aio_context, NULL, NULL);
1573 assert(ret == 0);
1575 aio_context_release(tmp_context);
1576 aio_context_acquire(aio_context);
1579 bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1580 bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1582 aio_context_release(aio_context);
1587 static void external_snapshot_clean(void *opaque)
1589 g_autofree ExternalSnapshotState *state = opaque;
1590 AioContext *aio_context;
1592 if (!state->old_bs) {
1593 return;
1596 aio_context = bdrv_get_aio_context(state->old_bs);
1597 aio_context_acquire(aio_context);
1599 bdrv_drained_end(state->old_bs);
1600 bdrv_unref(state->new_bs);
1602 aio_context_release(aio_context);
1605 typedef struct DriveBackupState {
1606 BlockDriverState *bs;
1607 BlockJob *job;
1608 } DriveBackupState;
1610 static BlockJob *do_backup_common(BackupCommon *backup,
1611 BlockDriverState *bs,
1612 BlockDriverState *target_bs,
1613 AioContext *aio_context,
1614 JobTxn *txn, Error **errp);
1616 static void drive_backup_commit(void *opaque);
1617 static void drive_backup_abort(void *opaque);
1618 static void drive_backup_clean(void *opaque);
1619 TransactionActionDrv drive_backup_drv = {
1620 .commit = drive_backup_commit,
1621 .abort = drive_backup_abort,
1622 .clean = drive_backup_clean,
1625 static void drive_backup_action(DriveBackup *backup,
1626 JobTxn *block_job_txn,
1627 Transaction *tran, Error **errp)
1629 DriveBackupState *state = g_new0(DriveBackupState, 1);
1630 BlockDriverState *bs;
1631 BlockDriverState *target_bs;
1632 BlockDriverState *source = NULL;
1633 AioContext *aio_context;
1634 AioContext *old_context;
1635 const char *format;
1636 QDict *options;
1637 Error *local_err = NULL;
1638 int flags;
1639 int64_t size;
1640 bool set_backing_hd = false;
1641 int ret;
1643 tran_add(tran, &drive_backup_drv, state);
1645 if (!backup->has_mode) {
1646 backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1649 bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1650 if (!bs) {
1651 return;
1654 if (!bs->drv) {
1655 error_setg(errp, "Device has no medium");
1656 return;
1659 aio_context = bdrv_get_aio_context(bs);
1660 aio_context_acquire(aio_context);
1662 state->bs = bs;
1663 /* Paired with .clean() */
1664 bdrv_drained_begin(bs);
1666 format = backup->format;
1667 if (!format && backup->mode != NEW_IMAGE_MODE_EXISTING) {
1668 format = bs->drv->format_name;
1671 /* Early check to avoid creating target */
1672 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1673 goto out;
1676 flags = bs->open_flags | BDRV_O_RDWR;
1679 * See if we have a backing HD we can use to create our new image
1680 * on top of.
1682 if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1684 * Backup will not replace the source by the target, so none
1685 * of the filters skipped here will be removed (in contrast to
1686 * mirror). Therefore, we can skip all of them when looking
1687 * for the first COW relationship.
1689 source = bdrv_cow_bs(bdrv_skip_filters(bs));
1690 if (!source) {
1691 backup->sync = MIRROR_SYNC_MODE_FULL;
1694 if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1695 source = bs;
1696 flags |= BDRV_O_NO_BACKING;
1697 set_backing_hd = true;
1700 size = bdrv_getlength(bs);
1701 if (size < 0) {
1702 error_setg_errno(errp, -size, "bdrv_getlength failed");
1703 goto out;
1706 if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1707 assert(format);
1708 if (source) {
1709 /* Implicit filters should not appear in the filename */
1710 BlockDriverState *explicit_backing =
1711 bdrv_skip_implicit_filters(source);
1713 bdrv_refresh_filename(explicit_backing);
1714 bdrv_img_create(backup->target, format,
1715 explicit_backing->filename,
1716 explicit_backing->drv->format_name, NULL,
1717 size, flags, false, &local_err);
1718 } else {
1719 bdrv_img_create(backup->target, format, NULL, NULL, NULL,
1720 size, flags, false, &local_err);
1724 if (local_err) {
1725 error_propagate(errp, local_err);
1726 goto out;
1729 options = qdict_new();
1730 qdict_put_str(options, "discard", "unmap");
1731 qdict_put_str(options, "detect-zeroes", "unmap");
1732 if (format) {
1733 qdict_put_str(options, "driver", format);
1736 target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1737 if (!target_bs) {
1738 goto out;
1741 /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1742 old_context = bdrv_get_aio_context(target_bs);
1743 aio_context_release(aio_context);
1744 aio_context_acquire(old_context);
1746 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1747 if (ret < 0) {
1748 bdrv_unref(target_bs);
1749 aio_context_release(old_context);
1750 return;
1753 aio_context_release(old_context);
1754 aio_context_acquire(aio_context);
1756 if (set_backing_hd) {
1757 if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1758 goto unref;
1762 state->job = do_backup_common(qapi_DriveBackup_base(backup),
1763 bs, target_bs, aio_context,
1764 block_job_txn, errp);
1766 unref:
1767 bdrv_unref(target_bs);
1768 out:
1769 aio_context_release(aio_context);
1772 static void drive_backup_commit(void *opaque)
1774 DriveBackupState *state = opaque;
1775 AioContext *aio_context;
1777 aio_context = bdrv_get_aio_context(state->bs);
1778 aio_context_acquire(aio_context);
1780 assert(state->job);
1781 job_start(&state->job->job);
1783 aio_context_release(aio_context);
1786 static void drive_backup_abort(void *opaque)
1788 DriveBackupState *state = opaque;
1790 if (state->job) {
1791 job_cancel_sync(&state->job->job, true);
1795 static void drive_backup_clean(void *opaque)
1797 g_autofree DriveBackupState *state = opaque;
1798 AioContext *aio_context;
1800 if (!state->bs) {
1801 return;
1804 aio_context = bdrv_get_aio_context(state->bs);
1805 aio_context_acquire(aio_context);
1807 bdrv_drained_end(state->bs);
1809 aio_context_release(aio_context);
1812 typedef struct BlockdevBackupState {
1813 BlockDriverState *bs;
1814 BlockJob *job;
1815 } BlockdevBackupState;
1817 static void blockdev_backup_commit(void *opaque);
1818 static void blockdev_backup_abort(void *opaque);
1819 static void blockdev_backup_clean(void *opaque);
1820 TransactionActionDrv blockdev_backup_drv = {
1821 .commit = blockdev_backup_commit,
1822 .abort = blockdev_backup_abort,
1823 .clean = blockdev_backup_clean,
1826 static void blockdev_backup_action(BlockdevBackup *backup,
1827 JobTxn *block_job_txn,
1828 Transaction *tran, Error **errp)
1830 BlockdevBackupState *state = g_new0(BlockdevBackupState, 1);
1831 BlockDriverState *bs;
1832 BlockDriverState *target_bs;
1833 AioContext *aio_context;
1834 AioContext *old_context;
1835 int ret;
1837 tran_add(tran, &blockdev_backup_drv, state);
1839 bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1840 if (!bs) {
1841 return;
1844 target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1845 if (!target_bs) {
1846 return;
1849 /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1850 aio_context = bdrv_get_aio_context(bs);
1851 old_context = bdrv_get_aio_context(target_bs);
1852 aio_context_acquire(old_context);
1854 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1855 if (ret < 0) {
1856 aio_context_release(old_context);
1857 return;
1860 aio_context_release(old_context);
1861 aio_context_acquire(aio_context);
1862 state->bs = bs;
1864 /* Paired with .clean() */
1865 bdrv_drained_begin(state->bs);
1867 state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1868 bs, target_bs, aio_context,
1869 block_job_txn, errp);
1871 aio_context_release(aio_context);
1874 static void blockdev_backup_commit(void *opaque)
1876 BlockdevBackupState *state = opaque;
1877 AioContext *aio_context;
1879 aio_context = bdrv_get_aio_context(state->bs);
1880 aio_context_acquire(aio_context);
1882 assert(state->job);
1883 job_start(&state->job->job);
1885 aio_context_release(aio_context);
1888 static void blockdev_backup_abort(void *opaque)
1890 BlockdevBackupState *state = opaque;
1892 if (state->job) {
1893 job_cancel_sync(&state->job->job, true);
1897 static void blockdev_backup_clean(void *opaque)
1899 g_autofree BlockdevBackupState *state = opaque;
1900 AioContext *aio_context;
1902 if (!state->bs) {
1903 return;
1906 aio_context = bdrv_get_aio_context(state->bs);
1907 aio_context_acquire(aio_context);
1909 bdrv_drained_end(state->bs);
1911 aio_context_release(aio_context);
1914 typedef struct BlockDirtyBitmapState {
1915 BdrvDirtyBitmap *bitmap;
1916 BlockDriverState *bs;
1917 HBitmap *backup;
1918 bool was_enabled;
1919 } BlockDirtyBitmapState;
1921 static void block_dirty_bitmap_add_abort(void *opaque);
1922 TransactionActionDrv block_dirty_bitmap_add_drv = {
1923 .abort = block_dirty_bitmap_add_abort,
1924 .clean = g_free,
1927 static void block_dirty_bitmap_add_action(BlockDirtyBitmapAdd *action,
1928 Transaction *tran, Error **errp)
1930 Error *local_err = NULL;
1931 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1933 tran_add(tran, &block_dirty_bitmap_add_drv, state);
1935 /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1936 qmp_block_dirty_bitmap_add(action->node, action->name,
1937 action->has_granularity, action->granularity,
1938 action->has_persistent, action->persistent,
1939 action->has_disabled, action->disabled,
1940 &local_err);
1942 if (!local_err) {
1943 state->bitmap = block_dirty_bitmap_lookup(action->node, action->name,
1944 NULL, &error_abort);
1945 } else {
1946 error_propagate(errp, local_err);
1950 static void block_dirty_bitmap_add_abort(void *opaque)
1952 BlockDirtyBitmapState *state = opaque;
1954 if (state->bitmap) {
1955 bdrv_release_dirty_bitmap(state->bitmap);
1959 static void block_dirty_bitmap_restore(void *opaque);
1960 static void block_dirty_bitmap_free_backup(void *opaque);
1961 TransactionActionDrv block_dirty_bitmap_clear_drv = {
1962 .abort = block_dirty_bitmap_restore,
1963 .commit = block_dirty_bitmap_free_backup,
1964 .clean = g_free,
1967 static void block_dirty_bitmap_clear_action(BlockDirtyBitmap *action,
1968 Transaction *tran, Error **errp)
1970 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
1972 tran_add(tran, &block_dirty_bitmap_clear_drv, state);
1974 state->bitmap = block_dirty_bitmap_lookup(action->node,
1975 action->name,
1976 &state->bs,
1977 errp);
1978 if (!state->bitmap) {
1979 return;
1982 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
1983 return;
1986 bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
1989 static void block_dirty_bitmap_restore(void *opaque)
1991 BlockDirtyBitmapState *state = opaque;
1993 if (state->backup) {
1994 bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
1998 static void block_dirty_bitmap_free_backup(void *opaque)
2000 BlockDirtyBitmapState *state = opaque;
2002 hbitmap_free(state->backup);
2005 static void block_dirty_bitmap_enable_abort(void *opaque);
2006 TransactionActionDrv block_dirty_bitmap_enable_drv = {
2007 .abort = block_dirty_bitmap_enable_abort,
2008 .clean = g_free,
2011 static void block_dirty_bitmap_enable_action(BlockDirtyBitmap *action,
2012 Transaction *tran, Error **errp)
2014 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2016 tran_add(tran, &block_dirty_bitmap_enable_drv, state);
2018 state->bitmap = block_dirty_bitmap_lookup(action->node,
2019 action->name,
2020 NULL,
2021 errp);
2022 if (!state->bitmap) {
2023 return;
2026 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2027 return;
2030 state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2031 bdrv_enable_dirty_bitmap(state->bitmap);
2034 static void block_dirty_bitmap_enable_abort(void *opaque)
2036 BlockDirtyBitmapState *state = opaque;
2038 if (!state->was_enabled) {
2039 bdrv_disable_dirty_bitmap(state->bitmap);
2043 static void block_dirty_bitmap_disable_abort(void *opaque);
2044 TransactionActionDrv block_dirty_bitmap_disable_drv = {
2045 .abort = block_dirty_bitmap_disable_abort,
2046 .clean = g_free,
2049 static void block_dirty_bitmap_disable_action(BlockDirtyBitmap *action,
2050 Transaction *tran, Error **errp)
2052 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2054 tran_add(tran, &block_dirty_bitmap_disable_drv, state);
2056 state->bitmap = block_dirty_bitmap_lookup(action->node,
2057 action->name,
2058 NULL,
2059 errp);
2060 if (!state->bitmap) {
2061 return;
2064 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2065 return;
2068 state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2069 bdrv_disable_dirty_bitmap(state->bitmap);
2072 static void block_dirty_bitmap_disable_abort(void *opaque)
2074 BlockDirtyBitmapState *state = opaque;
2076 if (state->was_enabled) {
2077 bdrv_enable_dirty_bitmap(state->bitmap);
2081 TransactionActionDrv block_dirty_bitmap_merge_drv = {
2082 .commit = block_dirty_bitmap_free_backup,
2083 .abort = block_dirty_bitmap_restore,
2084 .clean = g_free,
2087 static void block_dirty_bitmap_merge_action(BlockDirtyBitmapMerge *action,
2088 Transaction *tran, Error **errp)
2090 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2092 tran_add(tran, &block_dirty_bitmap_merge_drv, state);
2094 state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2095 action->bitmaps, &state->backup,
2096 errp);
2099 static void block_dirty_bitmap_remove_commit(void *opaque);
2100 static void block_dirty_bitmap_remove_abort(void *opaque);
2101 TransactionActionDrv block_dirty_bitmap_remove_drv = {
2102 .commit = block_dirty_bitmap_remove_commit,
2103 .abort = block_dirty_bitmap_remove_abort,
2104 .clean = g_free,
2107 static void block_dirty_bitmap_remove_action(BlockDirtyBitmap *action,
2108 Transaction *tran, Error **errp)
2110 BlockDirtyBitmapState *state = g_new0(BlockDirtyBitmapState, 1);
2112 tran_add(tran, &block_dirty_bitmap_remove_drv, state);
2115 state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2116 false, &state->bs, errp);
2117 if (state->bitmap) {
2118 bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2119 bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2123 static void block_dirty_bitmap_remove_abort(void *opaque)
2125 BlockDirtyBitmapState *state = opaque;
2127 if (state->bitmap) {
2128 bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2129 bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2133 static void block_dirty_bitmap_remove_commit(void *opaque)
2135 BlockDirtyBitmapState *state = opaque;
2137 bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2138 bdrv_release_dirty_bitmap(state->bitmap);
2141 static void abort_commit(void *opaque);
2142 TransactionActionDrv abort_drv = {
2143 .commit = abort_commit,
2146 static void abort_action(Transaction *tran, Error **errp)
2148 tran_add(tran, &abort_drv, NULL);
2149 error_setg(errp, "Transaction aborted using Abort action");
2152 static void abort_commit(void *opaque)
2154 g_assert_not_reached(); /* this action never succeeds */
2157 static void transaction_action(TransactionAction *act, JobTxn *block_job_txn,
2158 Transaction *tran, Error **errp)
2160 switch (act->type) {
2161 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
2162 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
2163 external_snapshot_action(act, tran, errp);
2164 return;
2165 case TRANSACTION_ACTION_KIND_DRIVE_BACKUP:
2166 drive_backup_action(act->u.drive_backup.data,
2167 block_job_txn, tran, errp);
2168 return;
2169 case TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP:
2170 blockdev_backup_action(act->u.blockdev_backup.data,
2171 block_job_txn, tran, errp);
2172 return;
2173 case TRANSACTION_ACTION_KIND_ABORT:
2174 abort_action(tran, errp);
2175 return;
2176 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC:
2177 internal_snapshot_action(act->u.blockdev_snapshot_internal_sync.data,
2178 tran, errp);
2179 return;
2180 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD:
2181 block_dirty_bitmap_add_action(act->u.block_dirty_bitmap_add.data,
2182 tran, errp);
2183 return;
2184 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR:
2185 block_dirty_bitmap_clear_action(act->u.block_dirty_bitmap_clear.data,
2186 tran, errp);
2187 return;
2188 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE:
2189 block_dirty_bitmap_enable_action(act->u.block_dirty_bitmap_enable.data,
2190 tran, errp);
2191 return;
2192 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE:
2193 block_dirty_bitmap_disable_action(
2194 act->u.block_dirty_bitmap_disable.data, tran, errp);
2195 return;
2196 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE:
2197 block_dirty_bitmap_merge_action(act->u.block_dirty_bitmap_merge.data,
2198 tran, errp);
2199 return;
2200 case TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE:
2201 block_dirty_bitmap_remove_action(act->u.block_dirty_bitmap_remove.data,
2202 tran, errp);
2203 return;
2205 * Where are transactions for MIRROR, COMMIT and STREAM?
2206 * Although these blockjobs use transaction callbacks like the backup job,
2207 * these jobs do not necessarily adhere to transaction semantics.
2208 * These jobs may not fully undo all of their actions on abort, nor do they
2209 * necessarily work in transactions with more than one job in them.
2211 case TRANSACTION_ACTION_KIND__MAX:
2212 default:
2213 g_assert_not_reached();
2219 * 'Atomic' group operations. The operations are performed as a set, and if
2220 * any fail then we roll back all operations in the group.
2222 * Always run under BQL.
2224 void qmp_transaction(TransactionActionList *actions,
2225 struct TransactionProperties *properties,
2226 Error **errp)
2228 TransactionActionList *act;
2229 JobTxn *block_job_txn = NULL;
2230 Error *local_err = NULL;
2231 Transaction *tran;
2232 ActionCompletionMode comp_mode =
2233 properties ? properties->completion_mode :
2234 ACTION_COMPLETION_MODE_INDIVIDUAL;
2236 GLOBAL_STATE_CODE();
2238 /* Does this transaction get canceled as a group on failure?
2239 * If not, we don't really need to make a JobTxn.
2241 if (comp_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2242 for (act = actions; act; act = act->next) {
2243 TransactionActionKind type = act->value->type;
2245 if (type != TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP &&
2246 type != TRANSACTION_ACTION_KIND_DRIVE_BACKUP)
2248 error_setg(errp,
2249 "Action '%s' does not support transaction property "
2250 "completion-mode = %s",
2251 TransactionActionKind_str(type),
2252 ActionCompletionMode_str(comp_mode));
2253 return;
2257 block_job_txn = job_txn_new();
2260 /* drain all i/o before any operations */
2261 bdrv_drain_all();
2263 tran = tran_new();
2265 /* We don't do anything in this loop that commits us to the operations */
2266 for (act = actions; act; act = act->next) {
2267 transaction_action(act->value, block_job_txn, tran, &local_err);
2268 if (local_err) {
2269 error_propagate(errp, local_err);
2270 goto delete_and_fail;
2274 tran_commit(tran);
2276 /* success */
2277 goto exit;
2279 delete_and_fail:
2280 /* failure, and it is all-or-none; roll back all operations */
2281 tran_abort(tran);
2282 exit:
2283 job_txn_unref(block_job_txn);
2286 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2287 const char *name,
2288 Error **errp)
2290 BdrvDirtyBitmap *bitmap;
2291 BlockDriverState *bs;
2292 BlockDirtyBitmapSha256 *ret = NULL;
2293 char *sha256;
2295 bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2296 if (!bitmap || !bs) {
2297 return NULL;
2300 sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2301 if (sha256 == NULL) {
2302 return NULL;
2305 ret = g_new(BlockDirtyBitmapSha256, 1);
2306 ret->sha256 = sha256;
2308 return ret;
2311 void coroutine_fn qmp_block_resize(const char *device, const char *node_name,
2312 int64_t size, Error **errp)
2314 Error *local_err = NULL;
2315 BlockBackend *blk;
2316 BlockDriverState *bs;
2317 AioContext *old_ctx;
2319 bs = bdrv_lookup_bs(device, node_name, &local_err);
2320 if (local_err) {
2321 error_propagate(errp, local_err);
2322 return;
2325 if (size < 0) {
2326 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2327 return;
2330 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2331 error_setg(errp, QERR_DEVICE_IN_USE, device);
2332 return;
2335 blk = blk_co_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2336 if (!blk) {
2337 return;
2340 bdrv_co_lock(bs);
2341 bdrv_drained_begin(bs);
2342 bdrv_co_unlock(bs);
2344 old_ctx = bdrv_co_enter(bs);
2345 blk_co_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2346 bdrv_co_leave(bs, old_ctx);
2348 bdrv_co_lock(bs);
2349 bdrv_drained_end(bs);
2350 blk_co_unref(blk);
2351 bdrv_co_unlock(bs);
2354 void qmp_block_stream(const char *job_id, const char *device,
2355 const char *base,
2356 const char *base_node,
2357 const char *backing_file,
2358 const char *bottom,
2359 bool has_speed, int64_t speed,
2360 bool has_on_error, BlockdevOnError on_error,
2361 const char *filter_node_name,
2362 bool has_auto_finalize, bool auto_finalize,
2363 bool has_auto_dismiss, bool auto_dismiss,
2364 Error **errp)
2366 BlockDriverState *bs, *iter, *iter_end;
2367 BlockDriverState *base_bs = NULL;
2368 BlockDriverState *bottom_bs = NULL;
2369 AioContext *aio_context;
2370 Error *local_err = NULL;
2371 int job_flags = JOB_DEFAULT;
2373 if (base && base_node) {
2374 error_setg(errp, "'base' and 'base-node' cannot be specified "
2375 "at the same time");
2376 return;
2379 if (base && bottom) {
2380 error_setg(errp, "'base' and 'bottom' cannot be specified "
2381 "at the same time");
2382 return;
2385 if (bottom && base_node) {
2386 error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2387 "at the same time");
2388 return;
2391 if (!has_on_error) {
2392 on_error = BLOCKDEV_ON_ERROR_REPORT;
2395 bs = bdrv_lookup_bs(device, device, errp);
2396 if (!bs) {
2397 return;
2400 aio_context = bdrv_get_aio_context(bs);
2401 aio_context_acquire(aio_context);
2403 if (base) {
2404 base_bs = bdrv_find_backing_image(bs, base);
2405 if (base_bs == NULL) {
2406 error_setg(errp, "Can't find '%s' in the backing chain", base);
2407 goto out;
2409 assert(bdrv_get_aio_context(base_bs) == aio_context);
2412 if (base_node) {
2413 base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2414 if (!base_bs) {
2415 goto out;
2417 if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2418 error_setg(errp, "Node '%s' is not a backing image of '%s'",
2419 base_node, device);
2420 goto out;
2422 assert(bdrv_get_aio_context(base_bs) == aio_context);
2423 bdrv_refresh_filename(base_bs);
2426 if (bottom) {
2427 bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2428 if (!bottom_bs) {
2429 goto out;
2431 if (!bottom_bs->drv) {
2432 error_setg(errp, "Node '%s' is not open", bottom);
2433 goto out;
2435 if (bottom_bs->drv->is_filter) {
2436 error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2437 "as 'bottom'", bottom);
2438 goto out;
2440 if (!bdrv_chain_contains(bs, bottom_bs)) {
2441 error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2442 bottom, device);
2443 goto out;
2445 assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2449 * Check for op blockers in the whole chain between bs and base (or bottom)
2451 iter_end = bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2452 for (iter = bs; iter && iter != iter_end;
2453 iter = bdrv_filter_or_cow_bs(iter))
2455 if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2456 goto out;
2460 /* if we are streaming the entire chain, the result will have no backing
2461 * file, and specifying one is therefore an error */
2462 if (!base_bs && backing_file) {
2463 error_setg(errp, "backing file specified, but streaming the "
2464 "entire chain");
2465 goto out;
2468 if (has_auto_finalize && !auto_finalize) {
2469 job_flags |= JOB_MANUAL_FINALIZE;
2471 if (has_auto_dismiss && !auto_dismiss) {
2472 job_flags |= JOB_MANUAL_DISMISS;
2475 stream_start(job_id, bs, base_bs, backing_file,
2476 bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2477 filter_node_name, &local_err);
2478 if (local_err) {
2479 error_propagate(errp, local_err);
2480 goto out;
2483 trace_qmp_block_stream(bs);
2485 out:
2486 aio_context_release(aio_context);
2489 void qmp_block_commit(const char *job_id, const char *device,
2490 const char *base_node,
2491 const char *base,
2492 const char *top_node,
2493 const char *top,
2494 const char *backing_file,
2495 bool has_speed, int64_t speed,
2496 bool has_on_error, BlockdevOnError on_error,
2497 const char *filter_node_name,
2498 bool has_auto_finalize, bool auto_finalize,
2499 bool has_auto_dismiss, bool auto_dismiss,
2500 Error **errp)
2502 BlockDriverState *bs;
2503 BlockDriverState *iter;
2504 BlockDriverState *base_bs, *top_bs;
2505 AioContext *aio_context;
2506 Error *local_err = NULL;
2507 int job_flags = JOB_DEFAULT;
2508 uint64_t top_perm, top_shared;
2510 if (!has_speed) {
2511 speed = 0;
2513 if (!has_on_error) {
2514 on_error = BLOCKDEV_ON_ERROR_REPORT;
2516 if (has_auto_finalize && !auto_finalize) {
2517 job_flags |= JOB_MANUAL_FINALIZE;
2519 if (has_auto_dismiss && !auto_dismiss) {
2520 job_flags |= JOB_MANUAL_DISMISS;
2523 /* Important Note:
2524 * libvirt relies on the DeviceNotFound error class in order to probe for
2525 * live commit feature versions; for this to work, we must make sure to
2526 * perform the device lookup before any generic errors that may occur in a
2527 * scenario in which all optional arguments are omitted. */
2528 bs = qmp_get_root_bs(device, &local_err);
2529 if (!bs) {
2530 bs = bdrv_lookup_bs(device, device, NULL);
2531 if (!bs) {
2532 error_free(local_err);
2533 error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2534 "Device '%s' not found", device);
2535 } else {
2536 error_propagate(errp, local_err);
2538 return;
2541 aio_context = bdrv_get_aio_context(bs);
2542 aio_context_acquire(aio_context);
2544 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2545 goto out;
2548 /* default top_bs is the active layer */
2549 top_bs = bs;
2551 if (top_node && top) {
2552 error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2553 goto out;
2554 } else if (top_node) {
2555 top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2556 if (top_bs == NULL) {
2557 goto out;
2559 if (!bdrv_chain_contains(bs, top_bs)) {
2560 error_setg(errp, "'%s' is not in this backing file chain",
2561 top_node);
2562 goto out;
2564 } else if (top) {
2565 /* This strcmp() is just a shortcut, there is no need to
2566 * refresh @bs's filename. If it mismatches,
2567 * bdrv_find_backing_image() will do the refresh and may still
2568 * return @bs. */
2569 if (strcmp(bs->filename, top) != 0) {
2570 top_bs = bdrv_find_backing_image(bs, top);
2574 if (top_bs == NULL) {
2575 error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2576 goto out;
2579 assert(bdrv_get_aio_context(top_bs) == aio_context);
2581 if (base_node && base) {
2582 error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2583 goto out;
2584 } else if (base_node) {
2585 base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2586 if (base_bs == NULL) {
2587 goto out;
2589 if (!bdrv_chain_contains(top_bs, base_bs)) {
2590 error_setg(errp, "'%s' is not in this backing file chain",
2591 base_node);
2592 goto out;
2594 } else if (base) {
2595 base_bs = bdrv_find_backing_image(top_bs, base);
2596 if (base_bs == NULL) {
2597 error_setg(errp, "Can't find '%s' in the backing chain", base);
2598 goto out;
2600 } else {
2601 base_bs = bdrv_find_base(top_bs);
2602 if (base_bs == NULL) {
2603 error_setg(errp, "There is no backimg image");
2604 goto out;
2608 assert(bdrv_get_aio_context(base_bs) == aio_context);
2610 for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2611 iter = bdrv_filter_or_cow_bs(iter))
2613 if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2614 goto out;
2618 /* Do not allow attempts to commit an image into itself */
2619 if (top_bs == base_bs) {
2620 error_setg(errp, "cannot commit an image into itself");
2621 goto out;
2625 * Active commit is required if and only if someone has taken a
2626 * WRITE permission on the top node. Historically, we have always
2627 * used active commit for top nodes, so continue that practice
2628 * lest we possibly break clients that rely on this behavior, e.g.
2629 * to later attach this node to a writing parent.
2630 * (Active commit is never really wrong.)
2632 bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2633 if (top_perm & BLK_PERM_WRITE ||
2634 bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2636 if (backing_file) {
2637 if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2638 error_setg(errp, "'backing-file' specified,"
2639 " but 'top' is the active layer");
2640 } else {
2641 error_setg(errp, "'backing-file' specified, but 'top' has a "
2642 "writer on it");
2644 goto out;
2646 if (!job_id) {
2648 * Emulate here what block_job_create() does, because it
2649 * is possible that @bs != @top_bs (the block job should
2650 * be named after @bs, even if @top_bs is the actual
2651 * source)
2653 job_id = bdrv_get_device_name(bs);
2655 commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2656 filter_node_name, NULL, NULL, false, &local_err);
2657 } else {
2658 BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2659 if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2660 goto out;
2662 commit_start(job_id, bs, base_bs, top_bs, job_flags,
2663 speed, on_error, backing_file,
2664 filter_node_name, &local_err);
2666 if (local_err != NULL) {
2667 error_propagate(errp, local_err);
2668 goto out;
2671 out:
2672 aio_context_release(aio_context);
2675 /* Common QMP interface for drive-backup and blockdev-backup */
2676 static BlockJob *do_backup_common(BackupCommon *backup,
2677 BlockDriverState *bs,
2678 BlockDriverState *target_bs,
2679 AioContext *aio_context,
2680 JobTxn *txn, Error **errp)
2682 BlockJob *job = NULL;
2683 BdrvDirtyBitmap *bmap = NULL;
2684 BackupPerf perf = { .max_workers = 64 };
2685 int job_flags = JOB_DEFAULT;
2687 if (!backup->has_speed) {
2688 backup->speed = 0;
2690 if (!backup->has_on_source_error) {
2691 backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2693 if (!backup->has_on_target_error) {
2694 backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2696 if (!backup->has_auto_finalize) {
2697 backup->auto_finalize = true;
2699 if (!backup->has_auto_dismiss) {
2700 backup->auto_dismiss = true;
2702 if (!backup->has_compress) {
2703 backup->compress = false;
2706 if (backup->x_perf) {
2707 if (backup->x_perf->has_use_copy_range) {
2708 perf.use_copy_range = backup->x_perf->use_copy_range;
2710 if (backup->x_perf->has_max_workers) {
2711 perf.max_workers = backup->x_perf->max_workers;
2713 if (backup->x_perf->has_max_chunk) {
2714 perf.max_chunk = backup->x_perf->max_chunk;
2718 if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2719 (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2720 /* done before desugaring 'incremental' to print the right message */
2721 if (!backup->bitmap) {
2722 error_setg(errp, "must provide a valid bitmap name for "
2723 "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2724 return NULL;
2728 if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2729 if (backup->has_bitmap_mode &&
2730 backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2731 error_setg(errp, "Bitmap sync mode must be '%s' "
2732 "when using sync mode '%s'",
2733 BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2734 MirrorSyncMode_str(backup->sync));
2735 return NULL;
2737 backup->has_bitmap_mode = true;
2738 backup->sync = MIRROR_SYNC_MODE_BITMAP;
2739 backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2742 if (backup->bitmap) {
2743 bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2744 if (!bmap) {
2745 error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2746 return NULL;
2748 if (!backup->has_bitmap_mode) {
2749 error_setg(errp, "Bitmap sync mode must be given "
2750 "when providing a bitmap");
2751 return NULL;
2753 if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2754 return NULL;
2757 /* This does not produce a useful bitmap artifact: */
2758 if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2759 error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2760 " outputs", MirrorSyncMode_str(backup->sync));
2761 return NULL;
2764 /* If the bitmap isn't used for input or output, this is useless: */
2765 if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2766 backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2767 error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2768 " when combined with sync mode '%s'",
2769 BitmapSyncMode_str(backup->bitmap_mode),
2770 MirrorSyncMode_str(backup->sync));
2771 return NULL;
2775 if (!backup->bitmap && backup->has_bitmap_mode) {
2776 error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2777 return NULL;
2780 if (!backup->auto_finalize) {
2781 job_flags |= JOB_MANUAL_FINALIZE;
2783 if (!backup->auto_dismiss) {
2784 job_flags |= JOB_MANUAL_DISMISS;
2787 job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2788 backup->sync, bmap, backup->bitmap_mode,
2789 backup->compress,
2790 backup->filter_node_name,
2791 &perf,
2792 backup->on_source_error,
2793 backup->on_target_error,
2794 job_flags, NULL, NULL, txn, errp);
2795 return job;
2798 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2800 TransactionAction action = {
2801 .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2802 .u.drive_backup.data = backup,
2804 blockdev_do_action(&action, errp);
2807 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2808 bool flat,
2809 Error **errp)
2811 bool return_flat = has_flat && flat;
2813 return bdrv_named_nodes_list(return_flat, errp);
2816 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2818 return bdrv_get_xdbg_block_graph(errp);
2821 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2823 TransactionAction action = {
2824 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2825 .u.blockdev_backup.data = backup,
2827 blockdev_do_action(&action, errp);
2830 /* Parameter check and block job starting for drive mirroring.
2831 * Caller should hold @device and @target's aio context (must be the same).
2833 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2834 BlockDriverState *target,
2835 const char *replaces,
2836 enum MirrorSyncMode sync,
2837 BlockMirrorBackingMode backing_mode,
2838 bool zero_target,
2839 bool has_speed, int64_t speed,
2840 bool has_granularity, uint32_t granularity,
2841 bool has_buf_size, int64_t buf_size,
2842 bool has_on_source_error,
2843 BlockdevOnError on_source_error,
2844 bool has_on_target_error,
2845 BlockdevOnError on_target_error,
2846 bool has_unmap, bool unmap,
2847 const char *filter_node_name,
2848 bool has_copy_mode, MirrorCopyMode copy_mode,
2849 bool has_auto_finalize, bool auto_finalize,
2850 bool has_auto_dismiss, bool auto_dismiss,
2851 Error **errp)
2853 BlockDriverState *unfiltered_bs;
2854 int job_flags = JOB_DEFAULT;
2856 GLOBAL_STATE_CODE();
2857 GRAPH_RDLOCK_GUARD_MAINLOOP();
2859 if (!has_speed) {
2860 speed = 0;
2862 if (!has_on_source_error) {
2863 on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2865 if (!has_on_target_error) {
2866 on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2868 if (!has_granularity) {
2869 granularity = 0;
2871 if (!has_buf_size) {
2872 buf_size = 0;
2874 if (!has_unmap) {
2875 unmap = true;
2877 if (!has_copy_mode) {
2878 copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2880 if (has_auto_finalize && !auto_finalize) {
2881 job_flags |= JOB_MANUAL_FINALIZE;
2883 if (has_auto_dismiss && !auto_dismiss) {
2884 job_flags |= JOB_MANUAL_DISMISS;
2887 if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2888 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2889 "a value in range [512B, 64MB]");
2890 return;
2892 if (granularity & (granularity - 1)) {
2893 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2894 "a power of 2");
2895 return;
2898 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2899 return;
2901 if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
2902 return;
2905 if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
2906 sync = MIRROR_SYNC_MODE_FULL;
2909 if (!replaces) {
2910 /* We want to mirror from @bs, but keep implicit filters on top */
2911 unfiltered_bs = bdrv_skip_implicit_filters(bs);
2912 if (unfiltered_bs != bs) {
2913 replaces = unfiltered_bs->node_name;
2917 if (replaces) {
2918 BlockDriverState *to_replace_bs;
2919 AioContext *replace_aio_context;
2920 int64_t bs_size, replace_size;
2922 bs_size = bdrv_getlength(bs);
2923 if (bs_size < 0) {
2924 error_setg_errno(errp, -bs_size, "Failed to query device's size");
2925 return;
2928 to_replace_bs = check_to_replace_node(bs, replaces, errp);
2929 if (!to_replace_bs) {
2930 return;
2933 replace_aio_context = bdrv_get_aio_context(to_replace_bs);
2934 aio_context_acquire(replace_aio_context);
2935 replace_size = bdrv_getlength(to_replace_bs);
2936 aio_context_release(replace_aio_context);
2938 if (replace_size < 0) {
2939 error_setg_errno(errp, -replace_size,
2940 "Failed to query the replacement node's size");
2941 return;
2943 if (bs_size != replace_size) {
2944 error_setg(errp, "cannot replace image with a mirror image of "
2945 "different size");
2946 return;
2950 /* pass the node name to replace to mirror start since it's loose coupling
2951 * and will allow to check whether the node still exist at mirror completion
2953 mirror_start(job_id, bs, target,
2954 replaces, job_flags,
2955 speed, granularity, buf_size, sync, backing_mode, zero_target,
2956 on_source_error, on_target_error, unmap, filter_node_name,
2957 copy_mode, errp);
2960 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
2962 BlockDriverState *bs;
2963 BlockDriverState *target_backing_bs, *target_bs;
2964 AioContext *aio_context;
2965 AioContext *old_context;
2966 BlockMirrorBackingMode backing_mode;
2967 Error *local_err = NULL;
2968 QDict *options = NULL;
2969 int flags;
2970 int64_t size;
2971 const char *format = arg->format;
2972 bool zero_target;
2973 int ret;
2975 bs = qmp_get_root_bs(arg->device, errp);
2976 if (!bs) {
2977 return;
2980 /* Early check to avoid creating target */
2981 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2982 return;
2985 aio_context = bdrv_get_aio_context(bs);
2986 aio_context_acquire(aio_context);
2988 if (!arg->has_mode) {
2989 arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
2992 if (!arg->format) {
2993 format = (arg->mode == NEW_IMAGE_MODE_EXISTING
2994 ? NULL : bs->drv->format_name);
2997 flags = bs->open_flags | BDRV_O_RDWR;
2998 target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
2999 if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3000 arg->sync = MIRROR_SYNC_MODE_FULL;
3002 if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3003 target_backing_bs = bs;
3006 size = bdrv_getlength(bs);
3007 if (size < 0) {
3008 error_setg_errno(errp, -size, "bdrv_getlength failed");
3009 goto out;
3012 if (arg->replaces) {
3013 if (!arg->node_name) {
3014 error_setg(errp, "a node-name must be provided when replacing a"
3015 " named node of the graph");
3016 goto out;
3020 if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3021 backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3022 } else {
3023 backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3026 /* Don't open backing image in create() */
3027 flags |= BDRV_O_NO_BACKING;
3029 if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3030 && arg->mode != NEW_IMAGE_MODE_EXISTING)
3032 /* create new image w/o backing file */
3033 assert(format);
3034 bdrv_img_create(arg->target, format,
3035 NULL, NULL, NULL, size, flags, false, &local_err);
3036 } else {
3037 /* Implicit filters should not appear in the filename */
3038 BlockDriverState *explicit_backing =
3039 bdrv_skip_implicit_filters(target_backing_bs);
3041 switch (arg->mode) {
3042 case NEW_IMAGE_MODE_EXISTING:
3043 break;
3044 case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3045 /* create new image with backing file */
3046 bdrv_refresh_filename(explicit_backing);
3047 bdrv_img_create(arg->target, format,
3048 explicit_backing->filename,
3049 explicit_backing->drv->format_name,
3050 NULL, size, flags, false, &local_err);
3051 break;
3052 default:
3053 abort();
3057 if (local_err) {
3058 error_propagate(errp, local_err);
3059 goto out;
3062 options = qdict_new();
3063 if (arg->node_name) {
3064 qdict_put_str(options, "node-name", arg->node_name);
3066 if (format) {
3067 qdict_put_str(options, "driver", format);
3070 /* Mirroring takes care of copy-on-write using the source's backing
3071 * file.
3073 target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3074 if (!target_bs) {
3075 goto out;
3078 zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3079 (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3080 !bdrv_has_zero_init(target_bs)));
3083 /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3084 old_context = bdrv_get_aio_context(target_bs);
3085 aio_context_release(aio_context);
3086 aio_context_acquire(old_context);
3088 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3089 if (ret < 0) {
3090 bdrv_unref(target_bs);
3091 aio_context_release(old_context);
3092 return;
3095 aio_context_release(old_context);
3096 aio_context_acquire(aio_context);
3098 blockdev_mirror_common(arg->job_id, bs, target_bs,
3099 arg->replaces, arg->sync,
3100 backing_mode, zero_target,
3101 arg->has_speed, arg->speed,
3102 arg->has_granularity, arg->granularity,
3103 arg->has_buf_size, arg->buf_size,
3104 arg->has_on_source_error, arg->on_source_error,
3105 arg->has_on_target_error, arg->on_target_error,
3106 arg->has_unmap, arg->unmap,
3107 NULL,
3108 arg->has_copy_mode, arg->copy_mode,
3109 arg->has_auto_finalize, arg->auto_finalize,
3110 arg->has_auto_dismiss, arg->auto_dismiss,
3111 errp);
3112 bdrv_unref(target_bs);
3113 out:
3114 aio_context_release(aio_context);
3117 void qmp_blockdev_mirror(const char *job_id,
3118 const char *device, const char *target,
3119 const char *replaces,
3120 MirrorSyncMode sync,
3121 bool has_speed, int64_t speed,
3122 bool has_granularity, uint32_t granularity,
3123 bool has_buf_size, int64_t buf_size,
3124 bool has_on_source_error,
3125 BlockdevOnError on_source_error,
3126 bool has_on_target_error,
3127 BlockdevOnError on_target_error,
3128 const char *filter_node_name,
3129 bool has_copy_mode, MirrorCopyMode copy_mode,
3130 bool has_auto_finalize, bool auto_finalize,
3131 bool has_auto_dismiss, bool auto_dismiss,
3132 Error **errp)
3134 BlockDriverState *bs;
3135 BlockDriverState *target_bs;
3136 AioContext *aio_context;
3137 AioContext *old_context;
3138 BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3139 bool zero_target;
3140 int ret;
3142 bs = qmp_get_root_bs(device, errp);
3143 if (!bs) {
3144 return;
3147 target_bs = bdrv_lookup_bs(target, target, errp);
3148 if (!target_bs) {
3149 return;
3152 zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3154 /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3155 old_context = bdrv_get_aio_context(target_bs);
3156 aio_context = bdrv_get_aio_context(bs);
3157 aio_context_acquire(old_context);
3159 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3161 aio_context_release(old_context);
3162 aio_context_acquire(aio_context);
3164 if (ret < 0) {
3165 goto out;
3168 blockdev_mirror_common(job_id, bs, target_bs,
3169 replaces, sync, backing_mode,
3170 zero_target, has_speed, speed,
3171 has_granularity, granularity,
3172 has_buf_size, buf_size,
3173 has_on_source_error, on_source_error,
3174 has_on_target_error, on_target_error,
3175 true, true, filter_node_name,
3176 has_copy_mode, copy_mode,
3177 has_auto_finalize, auto_finalize,
3178 has_auto_dismiss, auto_dismiss,
3179 errp);
3180 out:
3181 aio_context_release(aio_context);
3185 * Get a block job using its ID. Called with job_mutex held.
3187 static BlockJob *find_block_job_locked(const char *id, Error **errp)
3189 BlockJob *job;
3191 assert(id != NULL);
3193 job = block_job_get_locked(id);
3195 if (!job) {
3196 error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3197 "Block job '%s' not found", id);
3198 return NULL;
3201 return job;
3204 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3206 BlockJob *job;
3208 JOB_LOCK_GUARD();
3209 job = find_block_job_locked(device, errp);
3211 if (!job) {
3212 return;
3215 block_job_set_speed_locked(job, speed, errp);
3218 void qmp_block_job_cancel(const char *device,
3219 bool has_force, bool force, Error **errp)
3221 BlockJob *job;
3223 JOB_LOCK_GUARD();
3224 job = find_block_job_locked(device, errp);
3226 if (!job) {
3227 return;
3230 if (!has_force) {
3231 force = false;
3234 if (job_user_paused_locked(&job->job) && !force) {
3235 error_setg(errp, "The block job for device '%s' is currently paused",
3236 device);
3237 return;
3240 trace_qmp_block_job_cancel(job);
3241 job_user_cancel_locked(&job->job, force, errp);
3244 void qmp_block_job_pause(const char *device, Error **errp)
3246 BlockJob *job;
3248 JOB_LOCK_GUARD();
3249 job = find_block_job_locked(device, errp);
3251 if (!job) {
3252 return;
3255 trace_qmp_block_job_pause(job);
3256 job_user_pause_locked(&job->job, errp);
3259 void qmp_block_job_resume(const char *device, Error **errp)
3261 BlockJob *job;
3263 JOB_LOCK_GUARD();
3264 job = find_block_job_locked(device, errp);
3266 if (!job) {
3267 return;
3270 trace_qmp_block_job_resume(job);
3271 job_user_resume_locked(&job->job, errp);
3274 void qmp_block_job_complete(const char *device, Error **errp)
3276 BlockJob *job;
3278 JOB_LOCK_GUARD();
3279 job = find_block_job_locked(device, errp);
3281 if (!job) {
3282 return;
3285 trace_qmp_block_job_complete(job);
3286 job_complete_locked(&job->job, errp);
3289 void qmp_block_job_finalize(const char *id, Error **errp)
3291 BlockJob *job;
3293 JOB_LOCK_GUARD();
3294 job = find_block_job_locked(id, errp);
3296 if (!job) {
3297 return;
3300 trace_qmp_block_job_finalize(job);
3301 job_ref_locked(&job->job);
3302 job_finalize_locked(&job->job, errp);
3304 job_unref_locked(&job->job);
3307 void qmp_block_job_dismiss(const char *id, Error **errp)
3309 BlockJob *bjob;
3310 Job *job;
3312 JOB_LOCK_GUARD();
3313 bjob = find_block_job_locked(id, errp);
3315 if (!bjob) {
3316 return;
3319 trace_qmp_block_job_dismiss(bjob);
3320 job = &bjob->job;
3321 job_dismiss_locked(&job, errp);
3324 void qmp_change_backing_file(const char *device,
3325 const char *image_node_name,
3326 const char *backing_file,
3327 Error **errp)
3329 BlockDriverState *bs = NULL;
3330 AioContext *aio_context;
3331 BlockDriverState *image_bs = NULL;
3332 Error *local_err = NULL;
3333 bool ro;
3334 int ret;
3336 bs = qmp_get_root_bs(device, errp);
3337 if (!bs) {
3338 return;
3341 aio_context = bdrv_get_aio_context(bs);
3342 aio_context_acquire(aio_context);
3344 image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3345 if (local_err) {
3346 error_propagate(errp, local_err);
3347 goto out;
3350 if (!image_bs) {
3351 error_setg(errp, "image file not found");
3352 goto out;
3355 if (bdrv_find_base(image_bs) == image_bs) {
3356 error_setg(errp, "not allowing backing file change on an image "
3357 "without a backing file");
3358 goto out;
3361 /* even though we are not necessarily operating on bs, we need it to
3362 * determine if block ops are currently prohibited on the chain */
3363 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3364 goto out;
3367 /* final sanity check */
3368 if (!bdrv_chain_contains(bs, image_bs)) {
3369 error_setg(errp, "'%s' and image file are not in the same chain",
3370 device);
3371 goto out;
3374 /* if not r/w, reopen to make r/w */
3375 ro = bdrv_is_read_only(image_bs);
3377 if (ro) {
3378 if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3379 goto out;
3383 ret = bdrv_change_backing_file(image_bs, backing_file,
3384 image_bs->drv ? image_bs->drv->format_name : "",
3385 false);
3387 if (ret < 0) {
3388 error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3389 backing_file);
3390 /* don't exit here, so we can try to restore open flags if
3391 * appropriate */
3394 if (ro) {
3395 bdrv_reopen_set_read_only(image_bs, true, errp);
3398 out:
3399 aio_context_release(aio_context);
3402 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3404 BlockDriverState *bs;
3405 QObject *obj;
3406 Visitor *v = qobject_output_visitor_new(&obj);
3407 QDict *qdict;
3409 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3410 visit_complete(v, &obj);
3411 qdict = qobject_to(QDict, obj);
3413 qdict_flatten(qdict);
3415 if (!qdict_get_try_str(qdict, "node-name")) {
3416 error_setg(errp, "'node-name' must be specified for the root node");
3417 goto fail;
3420 bs = bds_tree_init(qdict, errp);
3421 if (!bs) {
3422 goto fail;
3425 bdrv_set_monitor_owned(bs);
3427 fail:
3428 visit_free(v);
3431 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3433 BlockReopenQueue *queue = NULL;
3435 /* Add each one of the BDS that we want to reopen to the queue */
3436 for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3437 BlockdevOptions *options = reopen_list->value;
3438 BlockDriverState *bs;
3439 AioContext *ctx;
3440 QObject *obj;
3441 Visitor *v;
3442 QDict *qdict;
3444 /* Check for the selected node name */
3445 if (!options->node_name) {
3446 error_setg(errp, "node-name not specified");
3447 goto fail;
3450 bs = bdrv_find_node(options->node_name);
3451 if (!bs) {
3452 error_setg(errp, "Failed to find node with node-name='%s'",
3453 options->node_name);
3454 goto fail;
3457 /* Put all options in a QDict and flatten it */
3458 v = qobject_output_visitor_new(&obj);
3459 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3460 visit_complete(v, &obj);
3461 visit_free(v);
3463 qdict = qobject_to(QDict, obj);
3465 qdict_flatten(qdict);
3467 ctx = bdrv_get_aio_context(bs);
3468 aio_context_acquire(ctx);
3470 queue = bdrv_reopen_queue(queue, bs, qdict, false);
3472 aio_context_release(ctx);
3475 /* Perform the reopen operation */
3476 bdrv_reopen_multiple(queue, errp);
3477 queue = NULL;
3479 fail:
3480 bdrv_reopen_queue_free(queue);
3483 void qmp_blockdev_del(const char *node_name, Error **errp)
3485 AioContext *aio_context;
3486 BlockDriverState *bs;
3488 GLOBAL_STATE_CODE();
3490 bs = bdrv_find_node(node_name);
3491 if (!bs) {
3492 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3493 return;
3495 if (bdrv_has_blk(bs)) {
3496 error_setg(errp, "Node %s is in use", node_name);
3497 return;
3499 aio_context = bdrv_get_aio_context(bs);
3500 aio_context_acquire(aio_context);
3502 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3503 goto out;
3506 if (!QTAILQ_IN_USE(bs, monitor_list)) {
3507 error_setg(errp, "Node %s is not owned by the monitor",
3508 bs->node_name);
3509 goto out;
3512 if (bs->refcnt > 1) {
3513 error_setg(errp, "Block device %s is in use",
3514 bdrv_get_device_or_node_name(bs));
3515 goto out;
3518 QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3519 bdrv_unref(bs);
3521 out:
3522 aio_context_release(aio_context);
3525 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3526 const char *child_name)
3528 BdrvChild *child;
3530 QLIST_FOREACH(child, &parent_bs->children, next) {
3531 if (strcmp(child->name, child_name) == 0) {
3532 return child;
3536 return NULL;
3539 void qmp_x_blockdev_change(const char *parent, const char *child,
3540 const char *node, Error **errp)
3542 BlockDriverState *parent_bs, *new_bs = NULL;
3543 BdrvChild *p_child;
3545 parent_bs = bdrv_lookup_bs(parent, parent, errp);
3546 if (!parent_bs) {
3547 return;
3550 if (!child == !node) {
3551 if (child) {
3552 error_setg(errp, "The parameters child and node are in conflict");
3553 } else {
3554 error_setg(errp, "Either child or node must be specified");
3556 return;
3559 if (child) {
3560 p_child = bdrv_find_child(parent_bs, child);
3561 if (!p_child) {
3562 error_setg(errp, "Node '%s' does not have child '%s'",
3563 parent, child);
3564 return;
3566 bdrv_del_child(parent_bs, p_child, errp);
3569 if (node) {
3570 new_bs = bdrv_find_node(node);
3571 if (!new_bs) {
3572 error_setg(errp, "Node '%s' not found", node);
3573 return;
3575 bdrv_add_child(parent_bs, new_bs, errp);
3579 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3581 BlockJobInfoList *head = NULL, **tail = &head;
3582 BlockJob *job;
3584 JOB_LOCK_GUARD();
3586 for (job = block_job_next_locked(NULL); job;
3587 job = block_job_next_locked(job)) {
3588 BlockJobInfo *value;
3590 if (block_job_is_internal(job)) {
3591 continue;
3593 value = block_job_query_locked(job, errp);
3594 if (!value) {
3595 qapi_free_BlockJobInfoList(head);
3596 return NULL;
3598 QAPI_LIST_APPEND(tail, value);
3601 return head;
3604 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3605 bool has_force, bool force, Error **errp)
3607 AioContext *old_context;
3608 AioContext *new_context;
3609 BlockDriverState *bs;
3611 bs = bdrv_find_node(node_name);
3612 if (!bs) {
3613 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3614 return;
3617 /* Protects against accidents. */
3618 if (!(has_force && force) && bdrv_has_blk(bs)) {
3619 error_setg(errp, "Node %s is associated with a BlockBackend and could "
3620 "be in use (use force=true to override this check)",
3621 node_name);
3622 return;
3625 if (iothread->type == QTYPE_QSTRING) {
3626 IOThread *obj = iothread_by_id(iothread->u.s);
3627 if (!obj) {
3628 error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3629 return;
3632 new_context = iothread_get_aio_context(obj);
3633 } else {
3634 new_context = qemu_get_aio_context();
3637 old_context = bdrv_get_aio_context(bs);
3638 aio_context_acquire(old_context);
3640 bdrv_try_change_aio_context(bs, new_context, NULL, errp);
3642 aio_context_release(old_context);
3645 QemuOptsList qemu_common_drive_opts = {
3646 .name = "drive",
3647 .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3648 .desc = {
3650 .name = "snapshot",
3651 .type = QEMU_OPT_BOOL,
3652 .help = "enable/disable snapshot mode",
3654 .name = "aio",
3655 .type = QEMU_OPT_STRING,
3656 .help = "host AIO implementation (threads, native, io_uring)",
3658 .name = BDRV_OPT_CACHE_WB,
3659 .type = QEMU_OPT_BOOL,
3660 .help = "Enable writeback mode",
3662 .name = "format",
3663 .type = QEMU_OPT_STRING,
3664 .help = "disk format (raw, qcow2, ...)",
3666 .name = "rerror",
3667 .type = QEMU_OPT_STRING,
3668 .help = "read error action",
3670 .name = "werror",
3671 .type = QEMU_OPT_STRING,
3672 .help = "write error action",
3674 .name = BDRV_OPT_READ_ONLY,
3675 .type = QEMU_OPT_BOOL,
3676 .help = "open drive file as read-only",
3679 THROTTLE_OPTS,
3682 .name = "throttling.group",
3683 .type = QEMU_OPT_STRING,
3684 .help = "name of the block throttling group",
3686 .name = "copy-on-read",
3687 .type = QEMU_OPT_BOOL,
3688 .help = "copy read data from backing file into image file",
3690 .name = "detect-zeroes",
3691 .type = QEMU_OPT_STRING,
3692 .help = "try to optimize zero writes (off, on, unmap)",
3694 .name = "stats-account-invalid",
3695 .type = QEMU_OPT_BOOL,
3696 .help = "whether to account for invalid I/O operations "
3697 "in the statistics",
3699 .name = "stats-account-failed",
3700 .type = QEMU_OPT_BOOL,
3701 .help = "whether to account for failed I/O operations "
3702 "in the statistics",
3704 { /* end of list */ }
3708 QemuOptsList qemu_drive_opts = {
3709 .name = "drive",
3710 .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3711 .desc = {
3713 * no elements => accept any params
3714 * validation will happen later
3716 { /* end of list */ }