?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
s390x/pci: shrink DMA aperture to be bound by vfio DMA limit
[qemu/armbru.git] / blockdev.c
blob75eef8535ebdd636301e9927ab00f1df7871f54a
1 /*
2 * QEMU host block devices
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 *
6 * This work is licensed under the terms of the GNU GPL, version 2 or
7 * later. See the COPYING file in the top-level directory.
8 *
9 * This file incorporates work covered by the following copyright and
10 * permission notice:
11 *
12 * Copyright (c) 2003-2008 Fabrice Bellard
13 *
14 * Permission is hereby granted, free of charge, to any person obtaining a copy
15 * of this software and associated documentation files (the "Software"), to deal
16 * in the Software without restriction, including without limitation the rights
17 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
18 * copies of the Software, and to permit persons to whom the Software is
19 * furnished to do so, subject to the following conditions:
20 *
21 * The above copyright notice and this permission notice shall be included in
22 * all copies or substantial portions of the Software.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
27 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
29 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
30 * THE SOFTWARE.
31 */
32
33 #include "qemu/osdep.h"
34 #include "sysemu/block-backend.h"
35 #include "sysemu/blockdev.h"
36 #include "hw/block/block.h"
37 #include "block/blockjob.h"
38 #include "block/qdict.h"
39 #include "block/throttle-groups.h"
40 #include "monitor/monitor.h"
41 #include "qemu/error-report.h"
42 #include "qemu/option.h"
43 #include "qemu/qemu-print.h"
44 #include "qemu/config-file.h"
45 #include "qapi/qapi-commands-block.h"
46 #include "qapi/qapi-commands-transaction.h"
47 #include "qapi/qapi-visit-block-core.h"
48 #include "qapi/qmp/qdict.h"
49 #include "qapi/qmp/qnum.h"
50 #include "qapi/qmp/qstring.h"
51 #include "qapi/error.h"
52 #include "qapi/qmp/qerror.h"
53 #include "qapi/qmp/qlist.h"
54 #include "qapi/qobject-output-visitor.h"
55 #include "sysemu/sysemu.h"
56 #include "sysemu/iothread.h"
57 #include "block/block_int.h"
58 #include "block/trace.h"
59 #include "sysemu/runstate.h"
60 #include "sysemu/replay.h"
61 #include "qemu/cutils.h"
62 #include "qemu/help_option.h"
63 #include "qemu/main-loop.h"
64 #include "qemu/throttle-options.h"
65
66 /* Protected by BQL */
67 QTAILQ_HEAD(, BlockDriverState) monitor_bdrv_states =
68 QTAILQ_HEAD_INITIALIZER(monitor_bdrv_states);
69
70 void bdrv_set_monitor_owned(BlockDriverState *bs)
71 {
72 GLOBAL_STATE_CODE();
73 QTAILQ_INSERT_TAIL(&monitor_bdrv_states, bs, monitor_list);
74 }
75
76 static const char *const if_name[IF_COUNT] = {
77 [IF_NONE] = "none",
78 [IF_IDE] = "ide",
79 [IF_SCSI] = "scsi",
80 [IF_FLOPPY] = "floppy",
81 [IF_PFLASH] = "pflash",
82 [IF_MTD] = "mtd",
83 [IF_SD] = "sd",
84 [IF_VIRTIO] = "virtio",
85 [IF_XEN] = "xen",
86 };
87
88 static int if_max_devs[IF_COUNT] = {
89 /*
90 * Do not change these numbers! They govern how drive option
91 * index maps to unit and bus. That mapping is ABI.
92 *
93 * All controllers used to implement if=T drives need to support
94 * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
95 * Otherwise, some index values map to "impossible" bus, unit
96 * values.
97 *
98 * For instance, if you change [IF_SCSI] to 255, -drive
99 * if=scsi,index=12 no longer means bus=1,unit=5, but
100 * bus=0,unit=12. With an lsi53c895a controller (7 units max),
101 * the drive can't be set up. Regression.
102 */
103 [IF_IDE] = 2,
104 [IF_SCSI] = 7,
105 };
106
107 /**
108 * Boards may call this to offer board-by-board overrides
109 * of the default, global values.
110 */
111 void override_max_devs(BlockInterfaceType type, int max_devs)
112 {
113 BlockBackend *blk;
114 DriveInfo *dinfo;
115
116 GLOBAL_STATE_CODE();
117
118 if (max_devs <= 0) {
119 return;
120 }
121
122 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
123 dinfo = blk_legacy_dinfo(blk);
124 if (dinfo->type == type) {
125 fprintf(stderr, "Cannot override units-per-bus property of"
126 " the %s interface, because a drive of that type has"
127 " already been added.\n", if_name[type]);
128 g_assert_not_reached();
129 }
130 }
131
132 if_max_devs[type] = max_devs;
133 }
134
135 /*
136 * We automatically delete the drive when a device using it gets
137 * unplugged. Questionable feature, but we can't just drop it.
138 * Device models call blockdev_mark_auto_del() to schedule the
139 * automatic deletion, and generic qdev code calls blockdev_auto_del()
140 * when deletion is actually safe.
141 */
142 void blockdev_mark_auto_del(BlockBackend *blk)
143 {
144 DriveInfo *dinfo = blk_legacy_dinfo(blk);
145 BlockJob *job;
146
147 GLOBAL_STATE_CODE();
148
149 if (!dinfo) {
150 return;
151 }
152
153 JOB_LOCK_GUARD();
154
155 for (job = block_job_next_locked(NULL); job;
156 job = block_job_next_locked(job)) {
157 if (block_job_has_bdrv(job, blk_bs(blk))) {
158 job_cancel_locked(&job->job, false);
159 }
160 }
161
162 dinfo->auto_del = 1;
163 }
164
165 void blockdev_auto_del(BlockBackend *blk)
166 {
167 DriveInfo *dinfo = blk_legacy_dinfo(blk);
168 GLOBAL_STATE_CODE();
169
170 if (dinfo && dinfo->auto_del) {
171 monitor_remove_blk(blk);
172 blk_unref(blk);
173 }
174 }
175
176 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
177 {
178 int max_devs = if_max_devs[type];
179 return max_devs ? index / max_devs : 0;
180 }
181
182 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
183 {
184 int max_devs = if_max_devs[type];
185 return max_devs ? index % max_devs : index;
186 }
187
188 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
189 const char *optstr)
190 {
191 QemuOpts *opts;
192
193 GLOBAL_STATE_CODE();
194
195 opts = qemu_opts_parse_noisily(qemu_find_opts("drive"), optstr, false);
196 if (!opts) {
197 return NULL;
198 }
199 if (type != IF_DEFAULT) {
200 qemu_opt_set(opts, "if", if_name[type], &error_abort);
201 }
202 if (index >= 0) {
203 qemu_opt_set_number(opts, "index", index, &error_abort);
204 }
205 if (file)
206 qemu_opt_set(opts, "file", file, &error_abort);
207 return opts;
208 }
209
210 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
211 {
212 BlockBackend *blk;
213 DriveInfo *dinfo;
214
215 GLOBAL_STATE_CODE();
216
217 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
218 dinfo = blk_legacy_dinfo(blk);
219 if (dinfo && dinfo->type == type
220 && dinfo->bus == bus && dinfo->unit == unit) {
221 return dinfo;
222 }
223 }
224
225 return NULL;
226 }
227
228 /*
229 * Check board claimed all -drive that are meant to be claimed.
230 * Fatal error if any remain unclaimed.
231 */
232 void drive_check_orphaned(void)
233 {
234 BlockBackend *blk;
235 DriveInfo *dinfo;
236 Location loc;
237 bool orphans = false;
238
239 GLOBAL_STATE_CODE();
240
241 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
242 dinfo = blk_legacy_dinfo(blk);
243 /*
244 * Ignore default drives, because we create certain default
245 * drives unconditionally, then leave them unclaimed. Not the
246 * users fault.
247 * Ignore IF_VIRTIO, because it gets desugared into -device,
248 * so we can leave failing to -device.
249 * Ignore IF_NONE, because leaving unclaimed IF_NONE remains
250 * available for device_add is a feature.
251 */
252 if (dinfo->is_default || dinfo->type == IF_VIRTIO
253 || dinfo->type == IF_NONE) {
254 continue;
255 }
256 if (!blk_get_attached_dev(blk)) {
257 loc_push_none(&loc);
258 qemu_opts_loc_restore(dinfo->opts);
259 error_report("machine type does not support"
260 " if=%s,bus=%d,unit=%d",
261 if_name[dinfo->type], dinfo->bus, dinfo->unit);
262 loc_pop(&loc);
263 orphans = true;
264 }
265 }
266
267 if (orphans) {
268 exit(1);
269 }
270 }
271
272 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
273 {
274 GLOBAL_STATE_CODE();
275 return drive_get(type,
276 drive_index_to_bus_id(type, index),
277 drive_index_to_unit_id(type, index));
278 }
279
280 int drive_get_max_bus(BlockInterfaceType type)
281 {
282 int max_bus;
283 BlockBackend *blk;
284 DriveInfo *dinfo;
285
286 GLOBAL_STATE_CODE();
287
288 max_bus = -1;
289 for (blk = blk_next(NULL); blk; blk = blk_next(blk)) {
290 dinfo = blk_legacy_dinfo(blk);
291 if (dinfo && dinfo->type == type && dinfo->bus > max_bus) {
292 max_bus = dinfo->bus;
293 }
294 }
295 return max_bus;
296 }
297
298 static void bdrv_format_print(void *opaque, const char *name)
299 {
300 qemu_printf(" %s", name);
301 }
302
303 typedef struct {
304 QEMUBH *bh;
305 BlockDriverState *bs;
306 } BDRVPutRefBH;
307
308 static int parse_block_error_action(const char *buf, bool is_read, Error **errp)
309 {
310 if (!strcmp(buf, "ignore")) {
311 return BLOCKDEV_ON_ERROR_IGNORE;
312 } else if (!is_read && !strcmp(buf, "enospc")) {
313 return BLOCKDEV_ON_ERROR_ENOSPC;
314 } else if (!strcmp(buf, "stop")) {
315 return BLOCKDEV_ON_ERROR_STOP;
316 } else if (!strcmp(buf, "report")) {
317 return BLOCKDEV_ON_ERROR_REPORT;
318 } else {
319 error_setg(errp, "'%s' invalid %s error action",
320 buf, is_read ? "read" : "write");
321 return -1;
322 }
323 }
324
325 static bool parse_stats_intervals(BlockAcctStats *stats, QList *intervals,
326 Error **errp)
327 {
328 const QListEntry *entry;
329 for (entry = qlist_first(intervals); entry; entry = qlist_next(entry)) {
330 switch (qobject_type(entry->value)) {
331
332 case QTYPE_QSTRING: {
333 unsigned long long length;
334 const char *str = qstring_get_str(qobject_to(QString,
335 entry->value));
336 if (parse_uint_full(str, &length, 10) == 0 &&
337 length > 0 && length <= UINT_MAX) {
338 block_acct_add_interval(stats, (unsigned) length);
339 } else {
340 error_setg(errp, "Invalid interval length: %s", str);
341 return false;
342 }
343 break;
344 }
345
346 case QTYPE_QNUM: {
347 int64_t length = qnum_get_int(qobject_to(QNum, entry->value));
348
349 if (length > 0 && length <= UINT_MAX) {
350 block_acct_add_interval(stats, (unsigned) length);
351 } else {
352 error_setg(errp, "Invalid interval length: %" PRId64, length);
353 return false;
354 }
355 break;
356 }
357
358 default:
359 error_setg(errp, "The specification of stats-intervals is invalid");
360 return false;
361 }
362 }
363 return true;
364 }
365
366 typedef enum { MEDIA_DISK, MEDIA_CDROM } DriveMediaType;
367
368 /* All parameters but @opts are optional and may be set to NULL. */
369 static void extract_common_blockdev_options(QemuOpts *opts, int *bdrv_flags,
370 const char **throttling_group, ThrottleConfig *throttle_cfg,
371 BlockdevDetectZeroesOptions *detect_zeroes, Error **errp)
372 {
373 Error *local_error = NULL;
374 const char *aio;
375
376 if (bdrv_flags) {
377 if (qemu_opt_get_bool(opts, "copy-on-read", false)) {
378 *bdrv_flags |= BDRV_O_COPY_ON_READ;
379 }
380
381 if ((aio = qemu_opt_get(opts, "aio")) != NULL) {
382 if (bdrv_parse_aio(aio, bdrv_flags) < 0) {
383 error_setg(errp, "invalid aio option");
384 return;
385 }
386 }
387 }
388
389 /* disk I/O throttling */
390 if (throttling_group) {
391 *throttling_group = qemu_opt_get(opts, "throttling.group");
392 }
393
394 if (throttle_cfg) {
395 throttle_config_init(throttle_cfg);
396 throttle_cfg->buckets[THROTTLE_BPS_TOTAL].avg =
397 qemu_opt_get_number(opts, "throttling.bps-total", 0);
398 throttle_cfg->buckets[THROTTLE_BPS_READ].avg =
399 qemu_opt_get_number(opts, "throttling.bps-read", 0);
400 throttle_cfg->buckets[THROTTLE_BPS_WRITE].avg =
401 qemu_opt_get_number(opts, "throttling.bps-write", 0);
402 throttle_cfg->buckets[THROTTLE_OPS_TOTAL].avg =
403 qemu_opt_get_number(opts, "throttling.iops-total", 0);
404 throttle_cfg->buckets[THROTTLE_OPS_READ].avg =
405 qemu_opt_get_number(opts, "throttling.iops-read", 0);
406 throttle_cfg->buckets[THROTTLE_OPS_WRITE].avg =
407 qemu_opt_get_number(opts, "throttling.iops-write", 0);
408
409 throttle_cfg->buckets[THROTTLE_BPS_TOTAL].max =
410 qemu_opt_get_number(opts, "throttling.bps-total-max", 0);
411 throttle_cfg->buckets[THROTTLE_BPS_READ].max =
412 qemu_opt_get_number(opts, "throttling.bps-read-max", 0);
413 throttle_cfg->buckets[THROTTLE_BPS_WRITE].max =
414 qemu_opt_get_number(opts, "throttling.bps-write-max", 0);
415 throttle_cfg->buckets[THROTTLE_OPS_TOTAL].max =
416 qemu_opt_get_number(opts, "throttling.iops-total-max", 0);
417 throttle_cfg->buckets[THROTTLE_OPS_READ].max =
418 qemu_opt_get_number(opts, "throttling.iops-read-max", 0);
419 throttle_cfg->buckets[THROTTLE_OPS_WRITE].max =
420 qemu_opt_get_number(opts, "throttling.iops-write-max", 0);
421
422 throttle_cfg->buckets[THROTTLE_BPS_TOTAL].burst_length =
423 qemu_opt_get_number(opts, "throttling.bps-total-max-length", 1);
424 throttle_cfg->buckets[THROTTLE_BPS_READ].burst_length =
425 qemu_opt_get_number(opts, "throttling.bps-read-max-length", 1);
426 throttle_cfg->buckets[THROTTLE_BPS_WRITE].burst_length =
427 qemu_opt_get_number(opts, "throttling.bps-write-max-length", 1);
428 throttle_cfg->buckets[THROTTLE_OPS_TOTAL].burst_length =
429 qemu_opt_get_number(opts, "throttling.iops-total-max-length", 1);
430 throttle_cfg->buckets[THROTTLE_OPS_READ].burst_length =
431 qemu_opt_get_number(opts, "throttling.iops-read-max-length", 1);
432 throttle_cfg->buckets[THROTTLE_OPS_WRITE].burst_length =
433 qemu_opt_get_number(opts, "throttling.iops-write-max-length", 1);
434
435 throttle_cfg->op_size =
436 qemu_opt_get_number(opts, "throttling.iops-size", 0);
437
438 if (!throttle_is_valid(throttle_cfg, errp)) {
439 return;
440 }
441 }
442
443 if (detect_zeroes) {
444 *detect_zeroes =
445 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup,
446 qemu_opt_get(opts, "detect-zeroes"),
447 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF,
448 &local_error);
449 if (local_error) {
450 error_propagate(errp, local_error);
451 return;
452 }
453 }
454 }
455
456 static OnOffAuto account_get_opt(QemuOpts *opts, const char *name)
457 {
458 if (!qemu_opt_find(opts, name)) {
459 return ON_OFF_AUTO_AUTO;
460 }
461 if (qemu_opt_get_bool(opts, name, true)) {
462 return ON_OFF_AUTO_ON;
463 }
464 return ON_OFF_AUTO_OFF;
465 }
466
467 /* Takes the ownership of bs_opts */
468 static BlockBackend *blockdev_init(const char *file, QDict *bs_opts,
469 Error **errp)
470 {
471 const char *buf;
472 int bdrv_flags = 0;
473 int on_read_error, on_write_error;
474 OnOffAuto account_invalid, account_failed;
475 bool writethrough, read_only;
476 BlockBackend *blk;
477 BlockDriverState *bs;
478 ThrottleConfig cfg;
479 int snapshot = 0;
480 Error *error = NULL;
481 QemuOpts *opts;
482 QDict *interval_dict = NULL;
483 QList *interval_list = NULL;
484 const char *id;
485 BlockdevDetectZeroesOptions detect_zeroes =
486 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF;
487 const char *throttling_group = NULL;
488
489 /* Check common options by copying from bs_opts to opts, all other options
490 * stay in bs_opts for processing by bdrv_open(). */
491 id = qdict_get_try_str(bs_opts, "id");
492 opts = qemu_opts_create(&qemu_common_drive_opts, id, 1, errp);
493 if (!opts) {
494 goto err_no_opts;
495 }
496
497 if (!qemu_opts_absorb_qdict(opts, bs_opts, errp)) {
498 goto early_err;
499 }
500
501 if (id) {
502 qdict_del(bs_opts, "id");
503 }
504
505 /* extract parameters */
506 snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
507
508 account_invalid = account_get_opt(opts, "stats-account-invalid");
509 account_failed = account_get_opt(opts, "stats-account-failed");
510
511 writethrough = !qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, true);
512
513 id = qemu_opts_id(opts);
514
515 qdict_extract_subqdict(bs_opts, &interval_dict, "stats-intervals.");
516 qdict_array_split(interval_dict, &interval_list);
517
518 if (qdict_size(interval_dict) != 0) {
519 error_setg(errp, "Invalid option stats-intervals.%s",
520 qdict_first(interval_dict)->key);
521 goto early_err;
522 }
523
524 extract_common_blockdev_options(opts, &bdrv_flags, &throttling_group, &cfg,
525 &detect_zeroes, &error);
526 if (error) {
527 error_propagate(errp, error);
528 goto early_err;
529 }
530
531 if ((buf = qemu_opt_get(opts, "format")) != NULL) {
532 if (is_help_option(buf)) {
533 qemu_printf("Supported formats:");
534 bdrv_iterate_format(bdrv_format_print, NULL, false);
535 qemu_printf("\nSupported formats (read-only):");
536 bdrv_iterate_format(bdrv_format_print, NULL, true);
537 qemu_printf("\n");
538 goto early_err;
539 }
540
541 if (qdict_haskey(bs_opts, "driver")) {
542 error_setg(errp, "Cannot specify both 'driver' and 'format'");
543 goto early_err;
544 }
545 qdict_put_str(bs_opts, "driver", buf);
546 }
547
548 on_write_error = BLOCKDEV_ON_ERROR_ENOSPC;
549 if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
550 on_write_error = parse_block_error_action(buf, 0, &error);
551 if (error) {
552 error_propagate(errp, error);
553 goto early_err;
554 }
555 }
556
557 on_read_error = BLOCKDEV_ON_ERROR_REPORT;
558 if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
559 on_read_error = parse_block_error_action(buf, 1, &error);
560 if (error) {
561 error_propagate(errp, error);
562 goto early_err;
563 }
564 }
565
566 if (snapshot) {
567 bdrv_flags |= BDRV_O_SNAPSHOT;
568 }
569
570 read_only = qemu_opt_get_bool(opts, BDRV_OPT_READ_ONLY, false);
571
572 /* init */
573 if ((!file || !*file) && !qdict_size(bs_opts)) {
574 BlockBackendRootState *blk_rs;
575
576 blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
577 blk_rs = blk_get_root_state(blk);
578 blk_rs->open_flags = bdrv_flags | (read_only ? 0 : BDRV_O_RDWR);
579 blk_rs->detect_zeroes = detect_zeroes;
580
581 qobject_unref(bs_opts);
582 } else {
583 if (file && !*file) {
584 file = NULL;
585 }
586
587 /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
588 * with other callers) rather than what we want as the real defaults.
589 * Apply the defaults here instead. */
590 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
591 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
592 qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY,
593 read_only ? "on" : "off");
594 qdict_set_default_str(bs_opts, BDRV_OPT_AUTO_READ_ONLY, "on");
595 assert((bdrv_flags & BDRV_O_CACHE_MASK) == 0);
596
597 if (runstate_check(RUN_STATE_INMIGRATE)) {
598 bdrv_flags |= BDRV_O_INACTIVE;
599 }
600
601 blk = blk_new_open(file, NULL, bs_opts, bdrv_flags, errp);
602 if (!blk) {
603 goto err_no_bs_opts;
604 }
605 bs = blk_bs(blk);
606
607 bs->detect_zeroes = detect_zeroes;
608
609 block_acct_setup(blk_get_stats(blk), account_invalid, account_failed);
610
611 if (!parse_stats_intervals(blk_get_stats(blk), interval_list, errp)) {
612 blk_unref(blk);
613 blk = NULL;
614 goto err_no_bs_opts;
615 }
616 }
617
618 /* disk I/O throttling */
619 if (throttle_enabled(&cfg)) {
620 if (!throttling_group) {
621 throttling_group = id;
622 }
623 blk_io_limits_enable(blk, throttling_group);
624 blk_set_io_limits(blk, &cfg);
625 }
626
627 blk_set_enable_write_cache(blk, !writethrough);
628 blk_set_on_error(blk, on_read_error, on_write_error);
629
630 if (!monitor_add_blk(blk, id, errp)) {
631 blk_unref(blk);
632 blk = NULL;
633 goto err_no_bs_opts;
634 }
635
636 err_no_bs_opts:
637 qemu_opts_del(opts);
638 qobject_unref(interval_dict);
639 qobject_unref(interval_list);
640 return blk;
641
642 early_err:
643 qemu_opts_del(opts);
644 qobject_unref(interval_dict);
645 qobject_unref(interval_list);
646 err_no_opts:
647 qobject_unref(bs_opts);
648 return NULL;
649 }
650
651 /* Takes the ownership of bs_opts */
652 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp)
653 {
654 int bdrv_flags = 0;
655
656 GLOBAL_STATE_CODE();
657 /* bdrv_open() defaults to the values in bdrv_flags (for compatibility
658 * with other callers) rather than what we want as the real defaults.
659 * Apply the defaults here instead. */
660 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_DIRECT, "off");
661 qdict_set_default_str(bs_opts, BDRV_OPT_CACHE_NO_FLUSH, "off");
662 qdict_set_default_str(bs_opts, BDRV_OPT_READ_ONLY, "off");
663
664 if (runstate_check(RUN_STATE_INMIGRATE)) {
665 bdrv_flags |= BDRV_O_INACTIVE;
666 }
667
668 return bdrv_open(NULL, NULL, bs_opts, bdrv_flags, errp);
669 }
670
671 void blockdev_close_all_bdrv_states(void)
672 {
673 BlockDriverState *bs, *next_bs;
674
675 GLOBAL_STATE_CODE();
676 QTAILQ_FOREACH_SAFE(bs, &monitor_bdrv_states, monitor_list, next_bs) {
677 AioContext *ctx = bdrv_get_aio_context(bs);
678
679 aio_context_acquire(ctx);
680 bdrv_unref(bs);
681 aio_context_release(ctx);
682 }
683 }
684
685 /* Iterates over the list of monitor-owned BlockDriverStates */
686 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs)
687 {
688 GLOBAL_STATE_CODE();
689 return bs ? QTAILQ_NEXT(bs, monitor_list)
690 : QTAILQ_FIRST(&monitor_bdrv_states);
691 }
692
693 static bool qemu_opt_rename(QemuOpts *opts, const char *from, const char *to,
694 Error **errp)
695 {
696 const char *value;
697
698 value = qemu_opt_get(opts, from);
699 if (value) {
700 if (qemu_opt_find(opts, to)) {
701 error_setg(errp, "'%s' and its alias '%s' can't be used at the "
702 "same time", to, from);
703 return false;
704 }
705 }
706
707 /* rename all items in opts */
708 while ((value = qemu_opt_get(opts, from))) {
709 qemu_opt_set(opts, to, value, &error_abort);
710 qemu_opt_unset(opts, from);
711 }
712 return true;
713 }
714
715 QemuOptsList qemu_legacy_drive_opts = {
716 .name = "drive",
717 .head = QTAILQ_HEAD_INITIALIZER(qemu_legacy_drive_opts.head),
718 .desc = {
719 {
720 .name = "bus",
721 .type = QEMU_OPT_NUMBER,
722 .help = "bus number",
723 },{
724 .name = "unit",
725 .type = QEMU_OPT_NUMBER,
726 .help = "unit number (i.e. lun for scsi)",
727 },{
728 .name = "index",
729 .type = QEMU_OPT_NUMBER,
730 .help = "index number",
731 },{
732 .name = "media",
733 .type = QEMU_OPT_STRING,
734 .help = "media type (disk, cdrom)",
735 },{
736 .name = "if",
737 .type = QEMU_OPT_STRING,
738 .help = "interface (ide, scsi, sd, mtd, floppy, pflash, virtio)",
739 },{
740 .name = "file",
741 .type = QEMU_OPT_STRING,
742 .help = "file name",
743 },
744
745 /* Options that are passed on, but have special semantics with -drive */
746 {
747 .name = BDRV_OPT_READ_ONLY,
748 .type = QEMU_OPT_BOOL,
749 .help = "open drive file as read-only",
750 },{
751 .name = "rerror",
752 .type = QEMU_OPT_STRING,
753 .help = "read error action",
754 },{
755 .name = "werror",
756 .type = QEMU_OPT_STRING,
757 .help = "write error action",
758 },{
759 .name = "copy-on-read",
760 .type = QEMU_OPT_BOOL,
761 .help = "copy read data from backing file into image file",
762 },
763
764 { /* end of list */ }
765 },
766 };
767
768 DriveInfo *drive_new(QemuOpts *all_opts, BlockInterfaceType block_default_type,
769 Error **errp)
770 {
771 const char *value;
772 BlockBackend *blk;
773 DriveInfo *dinfo = NULL;
774 QDict *bs_opts;
775 QemuOpts *legacy_opts;
776 DriveMediaType media = MEDIA_DISK;
777 BlockInterfaceType type;
778 int max_devs, bus_id, unit_id, index;
779 const char *werror, *rerror;
780 bool read_only = false;
781 bool copy_on_read;
782 const char *filename;
783 int i;
784
785 GLOBAL_STATE_CODE();
786
787 /* Change legacy command line options into QMP ones */
788 static const struct {
789 const char *from;
790 const char *to;
791 } opt_renames[] = {
792 { "iops", "throttling.iops-total" },
793 { "iops_rd", "throttling.iops-read" },
794 { "iops_wr", "throttling.iops-write" },
795
796 { "bps", "throttling.bps-total" },
797 { "bps_rd", "throttling.bps-read" },
798 { "bps_wr", "throttling.bps-write" },
799
800 { "iops_max", "throttling.iops-total-max" },
801 { "iops_rd_max", "throttling.iops-read-max" },
802 { "iops_wr_max", "throttling.iops-write-max" },
803
804 { "bps_max", "throttling.bps-total-max" },
805 { "bps_rd_max", "throttling.bps-read-max" },
806 { "bps_wr_max", "throttling.bps-write-max" },
807
808 { "iops_size", "throttling.iops-size" },
809
810 { "group", "throttling.group" },
811
812 { "readonly", BDRV_OPT_READ_ONLY },
813 };
814
815 for (i = 0; i < ARRAY_SIZE(opt_renames); i++) {
816 if (!qemu_opt_rename(all_opts, opt_renames[i].from,
817 opt_renames[i].to, errp)) {
818 return NULL;
819 }
820 }
821
822 value = qemu_opt_get(all_opts, "cache");
823 if (value) {
824 int flags = 0;
825 bool writethrough;
826
827 if (bdrv_parse_cache_mode(value, &flags, &writethrough) != 0) {
828 error_setg(errp, "invalid cache option");
829 return NULL;
830 }
831
832 /* Specific options take precedence */
833 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_WB)) {
834 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_WB,
835 !writethrough, &error_abort);
836 }
837 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_DIRECT)) {
838 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_DIRECT,
839 !!(flags & BDRV_O_NOCACHE), &error_abort);
840 }
841 if (!qemu_opt_get(all_opts, BDRV_OPT_CACHE_NO_FLUSH)) {
842 qemu_opt_set_bool(all_opts, BDRV_OPT_CACHE_NO_FLUSH,
843 !!(flags & BDRV_O_NO_FLUSH), &error_abort);
844 }
845 qemu_opt_unset(all_opts, "cache");
846 }
847
848 /* Get a QDict for processing the options */
849 bs_opts = qdict_new();
850 qemu_opts_to_qdict(all_opts, bs_opts);
851
852 legacy_opts = qemu_opts_create(&qemu_legacy_drive_opts, NULL, 0,
853 &error_abort);
854 if (!qemu_opts_absorb_qdict(legacy_opts, bs_opts, errp)) {
855 goto fail;
856 }
857
858 /* Media type */
859 value = qemu_opt_get(legacy_opts, "media");
860 if (value) {
861 if (!strcmp(value, "disk")) {
862 media = MEDIA_DISK;
863 } else if (!strcmp(value, "cdrom")) {
864 media = MEDIA_CDROM;
865 read_only = true;
866 } else {
867 error_setg(errp, "'%s' invalid media", value);
868 goto fail;
869 }
870 }
871
872 /* copy-on-read is disabled with a warning for read-only devices */
873 read_only |= qemu_opt_get_bool(legacy_opts, BDRV_OPT_READ_ONLY, false);
874 copy_on_read = qemu_opt_get_bool(legacy_opts, "copy-on-read", false);
875
876 if (read_only && copy_on_read) {
877 warn_report("disabling copy-on-read on read-only drive");
878 copy_on_read = false;
879 }
880
881 qdict_put_str(bs_opts, BDRV_OPT_READ_ONLY, read_only ? "on" : "off");
882 qdict_put_str(bs_opts, "copy-on-read", copy_on_read ? "on" : "off");
883
884 /* Controller type */
885 value = qemu_opt_get(legacy_opts, "if");
886 if (value) {
887 for (type = 0;
888 type < IF_COUNT && strcmp(value, if_name[type]);
889 type++) {
890 }
891 if (type == IF_COUNT) {
892 error_setg(errp, "unsupported bus type '%s'", value);
893 goto fail;
894 }
895 } else {
896 type = block_default_type;
897 }
898
899 /* Device address specified by bus/unit or index.
900 * If none was specified, try to find the first free one. */
901 bus_id = qemu_opt_get_number(legacy_opts, "bus", 0);
902 unit_id = qemu_opt_get_number(legacy_opts, "unit", -1);
903 index = qemu_opt_get_number(legacy_opts, "index", -1);
904
905 max_devs = if_max_devs[type];
906
907 if (index != -1) {
908 if (bus_id != 0 || unit_id != -1) {
909 error_setg(errp, "index cannot be used with bus and unit");
910 goto fail;
911 }
912 bus_id = drive_index_to_bus_id(type, index);
913 unit_id = drive_index_to_unit_id(type, index);
914 }
915
916 if (unit_id == -1) {
917 unit_id = 0;
918 while (drive_get(type, bus_id, unit_id) != NULL) {
919 unit_id++;
920 if (max_devs && unit_id >= max_devs) {
921 unit_id -= max_devs;
922 bus_id++;
923 }
924 }
925 }
926
927 if (max_devs && unit_id >= max_devs) {
928 error_setg(errp, "unit %d too big (max is %d)", unit_id, max_devs - 1);
929 goto fail;
930 }
931
932 if (drive_get(type, bus_id, unit_id) != NULL) {
933 error_setg(errp, "drive with bus=%d, unit=%d (index=%d) exists",
934 bus_id, unit_id, index);
935 goto fail;
936 }
937
938 /* no id supplied -> create one */
939 if (qemu_opts_id(all_opts) == NULL) {
940 char *new_id;
941 const char *mediastr = "";
942 if (type == IF_IDE || type == IF_SCSI) {
943 mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
944 }
945 if (max_devs) {
946 new_id = g_strdup_printf("%s%i%s%i", if_name[type], bus_id,
947 mediastr, unit_id);
948 } else {
949 new_id = g_strdup_printf("%s%s%i", if_name[type],
950 mediastr, unit_id);
951 }
952 qdict_put_str(bs_opts, "id", new_id);
953 g_free(new_id);
954 }
955
956 /* Add virtio block device */
957 if (type == IF_VIRTIO) {
958 QemuOpts *devopts;
959 devopts = qemu_opts_create(qemu_find_opts("device"), NULL, 0,
960 &error_abort);
961 qemu_opt_set(devopts, "driver", "virtio-blk", &error_abort);
962 qemu_opt_set(devopts, "drive", qdict_get_str(bs_opts, "id"),
963 &error_abort);
964 }
965
966 filename = qemu_opt_get(legacy_opts, "file");
967
968 /* Check werror/rerror compatibility with if=... */
969 werror = qemu_opt_get(legacy_opts, "werror");
970 if (werror != NULL) {
971 if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO &&
972 type != IF_NONE) {
973 error_setg(errp, "werror is not supported by this bus type");
974 goto fail;
975 }
976 qdict_put_str(bs_opts, "werror", werror);
977 }
978
979 rerror = qemu_opt_get(legacy_opts, "rerror");
980 if (rerror != NULL) {
981 if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI &&
982 type != IF_NONE) {
983 error_setg(errp, "rerror is not supported by this bus type");
984 goto fail;
985 }
986 qdict_put_str(bs_opts, "rerror", rerror);
987 }
988
989 /* Actual block device init: Functionality shared with blockdev-add */
990 blk = blockdev_init(filename, bs_opts, errp);
991 bs_opts = NULL;
992 if (!blk) {
993 goto fail;
994 }
995
996 /* Create legacy DriveInfo */
997 dinfo = g_malloc0(sizeof(*dinfo));
998 dinfo->opts = all_opts;
999
1000 dinfo->type = type;
1001 dinfo->bus = bus_id;
1002 dinfo->unit = unit_id;
1003
1004 blk_set_legacy_dinfo(blk, dinfo);
1005
1006 switch(type) {
1007 case IF_IDE:
1008 case IF_SCSI:
1009 case IF_XEN:
1010 case IF_NONE:
1011 dinfo->media_cd = media == MEDIA_CDROM;
1012 break;
1013 default:
1014 break;
1015 }
1016
1017 fail:
1018 qemu_opts_del(legacy_opts);
1019 qobject_unref(bs_opts);
1020 return dinfo;
1021 }
1022
1023 static BlockDriverState *qmp_get_root_bs(const char *name, Error **errp)
1024 {
1025 BlockDriverState *bs;
1026
1027 bs = bdrv_lookup_bs(name, name, errp);
1028 if (bs == NULL) {
1029 return NULL;
1030 }
1031
1032 if (!bdrv_is_root_node(bs)) {
1033 error_setg(errp, "Need a root block node");
1034 return NULL;
1035 }
1036
1037 if (!bdrv_is_inserted(bs)) {
1038 error_setg(errp, "Device has no medium");
1039 return NULL;
1040 }
1041
1042 return bs;
1043 }
1044
1045 static void blockdev_do_action(TransactionAction *action, Error **errp)
1046 {
1047 TransactionActionList list;
1048
1049 list.value = action;
1050 list.next = NULL;
1051 qmp_transaction(&list, NULL, errp);
1052 }
1053
1054 void qmp_blockdev_snapshot_sync(const char *device, const char *node_name,
1055 const char *snapshot_file,
1056 const char *snapshot_node_name,
1057 const char *format,
1058 bool has_mode, NewImageMode mode, Error **errp)
1059 {
1060 BlockdevSnapshotSync snapshot = {
1061 .device = (char *) device,
1062 .node_name = (char *) node_name,
1063 .snapshot_file = (char *) snapshot_file,
1064 .snapshot_node_name = (char *) snapshot_node_name,
1065 .format = (char *) format,
1066 .has_mode = has_mode,
1067 .mode = mode,
1068 };
1069 TransactionAction action = {
1070 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC,
1071 .u.blockdev_snapshot_sync.data = &snapshot,
1072 };
1073 blockdev_do_action(&action, errp);
1074 }
1075
1076 void qmp_blockdev_snapshot(const char *node, const char *overlay,
1077 Error **errp)
1078 {
1079 BlockdevSnapshot snapshot_data = {
1080 .node = (char *) node,
1081 .overlay = (char *) overlay
1082 };
1083 TransactionAction action = {
1084 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT,
1085 .u.blockdev_snapshot.data = &snapshot_data,
1086 };
1087 blockdev_do_action(&action, errp);
1088 }
1089
1090 void qmp_blockdev_snapshot_internal_sync(const char *device,
1091 const char *name,
1092 Error **errp)
1093 {
1094 BlockdevSnapshotInternal snapshot = {
1095 .device = (char *) device,
1096 .name = (char *) name
1097 };
1098 TransactionAction action = {
1099 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC,
1100 .u.blockdev_snapshot_internal_sync.data = &snapshot,
1101 };
1102 blockdev_do_action(&action, errp);
1103 }
1104
1105 SnapshotInfo *qmp_blockdev_snapshot_delete_internal_sync(const char *device,
1106 const char *id,
1107 const char *name,
1108 Error **errp)
1109 {
1110 BlockDriverState *bs;
1111 AioContext *aio_context;
1112 QEMUSnapshotInfo sn;
1113 Error *local_err = NULL;
1114 SnapshotInfo *info = NULL;
1115 int ret;
1116
1117 bs = qmp_get_root_bs(device, errp);
1118 if (!bs) {
1119 return NULL;
1120 }
1121 aio_context = bdrv_get_aio_context(bs);
1122 aio_context_acquire(aio_context);
1123
1124 if (!id && !name) {
1125 error_setg(errp, "Name or id must be provided");
1126 goto out_aio_context;
1127 }
1128
1129 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE, errp)) {
1130 goto out_aio_context;
1131 }
1132
1133 ret = bdrv_snapshot_find_by_id_and_name(bs, id, name, &sn, &local_err);
1134 if (local_err) {
1135 error_propagate(errp, local_err);
1136 goto out_aio_context;
1137 }
1138 if (!ret) {
1139 error_setg(errp,
1140 "Snapshot with id '%s' and name '%s' does not exist on "
1141 "device '%s'",
1142 STR_OR_NULL(id), STR_OR_NULL(name), device);
1143 goto out_aio_context;
1144 }
1145
1146 bdrv_snapshot_delete(bs, id, name, &local_err);
1147 if (local_err) {
1148 error_propagate(errp, local_err);
1149 goto out_aio_context;
1150 }
1151
1152 aio_context_release(aio_context);
1153
1154 info = g_new0(SnapshotInfo, 1);
1155 info->id = g_strdup(sn.id_str);
1156 info->name = g_strdup(sn.name);
1157 info->date_nsec = sn.date_nsec;
1158 info->date_sec = sn.date_sec;
1159 info->vm_state_size = sn.vm_state_size;
1160 info->vm_clock_nsec = sn.vm_clock_nsec % 1000000000;
1161 info->vm_clock_sec = sn.vm_clock_nsec / 1000000000;
1162 if (sn.icount != -1ULL) {
1163 info->icount = sn.icount;
1164 info->has_icount = true;
1165 }
1166
1167 return info;
1168
1169 out_aio_context:
1170 aio_context_release(aio_context);
1171 return NULL;
1172 }
1173
1174 /* New and old BlockDriverState structs for atomic group operations */
1175
1176 typedef struct BlkActionState BlkActionState;
1177
1178 /**
1179 * BlkActionOps:
1180 * Table of operations that define an Action.
1181 *
1182 * @instance_size: Size of state struct, in bytes.
1183 * @prepare: Prepare the work, must NOT be NULL.
1184 * @commit: Commit the changes, can be NULL.
1185 * @abort: Abort the changes on fail, can be NULL.
1186 * @clean: Clean up resources after all transaction actions have called
1187 * commit() or abort(). Can be NULL.
1188 *
1189 * Only prepare() may fail. In a single transaction, only one of commit() or
1190 * abort() will be called. clean() will always be called if it is present.
1191 *
1192 * Always run under BQL.
1193 */
1194 typedef struct BlkActionOps {
1195 size_t instance_size;
1196 void (*prepare)(BlkActionState *common, Error **errp);
1197 void (*commit)(BlkActionState *common);
1198 void (*abort)(BlkActionState *common);
1199 void (*clean)(BlkActionState *common);
1200 } BlkActionOps;
1201
1202 /**
1203 * BlkActionState:
1204 * Describes one Action's state within a Transaction.
1205 *
1206 * @action: QAPI-defined enum identifying which Action to perform.
1207 * @ops: Table of ActionOps this Action can perform.
1208 * @block_job_txn: Transaction which this action belongs to.
1209 * @entry: List membership for all Actions in this Transaction.
1210 *
1211 * This structure must be arranged as first member in a subclassed type,
1212 * assuming that the compiler will also arrange it to the same offsets as the
1213 * base class.
1214 */
1215 struct BlkActionState {
1216 TransactionAction *action;
1217 const BlkActionOps *ops;
1218 JobTxn *block_job_txn;
1219 TransactionProperties *txn_props;
1220 QTAILQ_ENTRY(BlkActionState) entry;
1221 };
1222
1223 /* internal snapshot private data */
1224 typedef struct InternalSnapshotState {
1225 BlkActionState common;
1226 BlockDriverState *bs;
1227 QEMUSnapshotInfo sn;
1228 bool created;
1229 } InternalSnapshotState;
1230
1231
1232 static int action_check_completion_mode(BlkActionState *s, Error **errp)
1233 {
1234 if (s->txn_props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
1235 error_setg(errp,
1236 "Action '%s' does not support Transaction property "
1237 "completion-mode = %s",
1238 TransactionActionKind_str(s->action->type),
1239 ActionCompletionMode_str(s->txn_props->completion_mode));
1240 return -1;
1241 }
1242 return 0;
1243 }
1244
1245 static void internal_snapshot_prepare(BlkActionState *common,
1246 Error **errp)
1247 {
1248 Error *local_err = NULL;
1249 const char *device;
1250 const char *name;
1251 BlockDriverState *bs;
1252 QEMUSnapshotInfo old_sn, *sn;
1253 bool ret;
1254 int64_t rt;
1255 BlockdevSnapshotInternal *internal;
1256 InternalSnapshotState *state;
1257 AioContext *aio_context;
1258 int ret1;
1259
1260 g_assert(common->action->type ==
1261 TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC);
1262 internal = common->action->u.blockdev_snapshot_internal_sync.data;
1263 state = DO_UPCAST(InternalSnapshotState, common, common);
1264
1265 /* 1. parse input */
1266 device = internal->device;
1267 name = internal->name;
1268
1269 /* 2. check for validation */
1270 if (action_check_completion_mode(common, errp) < 0) {
1271 return;
1272 }
1273
1274 bs = qmp_get_root_bs(device, errp);
1275 if (!bs) {
1276 return;
1277 }
1278
1279 aio_context = bdrv_get_aio_context(bs);
1280 aio_context_acquire(aio_context);
1281
1282 state->bs = bs;
1283
1284 /* Paired with .clean() */
1285 bdrv_drained_begin(bs);
1286
1287 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
1288 goto out;
1289 }
1290
1291 if (bdrv_is_read_only(bs)) {
1292 error_setg(errp, "Device '%s' is read only", device);
1293 goto out;
1294 }
1295
1296 if (!bdrv_can_snapshot(bs)) {
1297 error_setg(errp, "Block format '%s' used by device '%s' "
1298 "does not support internal snapshots",
1299 bs->drv->format_name, device);
1300 goto out;
1301 }
1302
1303 if (!strlen(name)) {
1304 error_setg(errp, "Name is empty");
1305 goto out;
1306 }
1307
1308 /* check whether a snapshot with name exist */
1309 ret = bdrv_snapshot_find_by_id_and_name(bs, NULL, name, &old_sn,
1310 &local_err);
1311 if (local_err) {
1312 error_propagate(errp, local_err);
1313 goto out;
1314 } else if (ret) {
1315 error_setg(errp,
1316 "Snapshot with name '%s' already exists on device '%s'",
1317 name, device);
1318 goto out;
1319 }
1320
1321 /* 3. take the snapshot */
1322 sn = &state->sn;
1323 pstrcpy(sn->name, sizeof(sn->name), name);
1324 rt = g_get_real_time();
1325 sn->date_sec = rt / G_USEC_PER_SEC;
1326 sn->date_nsec = (rt % G_USEC_PER_SEC) * 1000;
1327 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1328 if (replay_mode != REPLAY_MODE_NONE) {
1329 sn->icount = replay_get_current_icount();
1330 } else {
1331 sn->icount = -1ULL;
1332 }
1333
1334 ret1 = bdrv_snapshot_create(bs, sn);
1335 if (ret1 < 0) {
1336 error_setg_errno(errp, -ret1,
1337 "Failed to create snapshot '%s' on device '%s'",
1338 name, device);
1339 goto out;
1340 }
1341
1342 /* 4. succeed, mark a snapshot is created */
1343 state->created = true;
1344
1345 out:
1346 aio_context_release(aio_context);
1347 }
1348
1349 static void internal_snapshot_abort(BlkActionState *common)
1350 {
1351 InternalSnapshotState *state =
1352 DO_UPCAST(InternalSnapshotState, common, common);
1353 BlockDriverState *bs = state->bs;
1354 QEMUSnapshotInfo *sn = &state->sn;
1355 AioContext *aio_context;
1356 Error *local_error = NULL;
1357
1358 if (!state->created) {
1359 return;
1360 }
1361
1362 aio_context = bdrv_get_aio_context(state->bs);
1363 aio_context_acquire(aio_context);
1364
1365 if (bdrv_snapshot_delete(bs, sn->id_str, sn->name, &local_error) < 0) {
1366 error_reportf_err(local_error,
1367 "Failed to delete snapshot with id '%s' and "
1368 "name '%s' on device '%s' in abort: ",
1369 sn->id_str, sn->name,
1370 bdrv_get_device_name(bs));
1371 }
1372
1373 aio_context_release(aio_context);
1374 }
1375
1376 static void internal_snapshot_clean(BlkActionState *common)
1377 {
1378 InternalSnapshotState *state = DO_UPCAST(InternalSnapshotState,
1379 common, common);
1380 AioContext *aio_context;
1381
1382 if (!state->bs) {
1383 return;
1384 }
1385
1386 aio_context = bdrv_get_aio_context(state->bs);
1387 aio_context_acquire(aio_context);
1388
1389 bdrv_drained_end(state->bs);
1390
1391 aio_context_release(aio_context);
1392 }
1393
1394 /* external snapshot private data */
1395 typedef struct ExternalSnapshotState {
1396 BlkActionState common;
1397 BlockDriverState *old_bs;
1398 BlockDriverState *new_bs;
1399 bool overlay_appended;
1400 } ExternalSnapshotState;
1401
1402 static void external_snapshot_prepare(BlkActionState *common,
1403 Error **errp)
1404 {
1405 int ret;
1406 int flags = 0;
1407 QDict *options = NULL;
1408 Error *local_err = NULL;
1409 /* Device and node name of the image to generate the snapshot from */
1410 const char *device;
1411 const char *node_name;
1412 /* Reference to the new image (for 'blockdev-snapshot') */
1413 const char *snapshot_ref;
1414 /* File name of the new image (for 'blockdev-snapshot-sync') */
1415 const char *new_image_file;
1416 ExternalSnapshotState *state =
1417 DO_UPCAST(ExternalSnapshotState, common, common);
1418 TransactionAction *action = common->action;
1419 AioContext *aio_context;
1420 uint64_t perm, shared;
1421
1422 /* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
1423 * purpose but a different set of parameters */
1424 switch (action->type) {
1425 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT:
1426 {
1427 BlockdevSnapshot *s = action->u.blockdev_snapshot.data;
1428 device = s->node;
1429 node_name = s->node;
1430 new_image_file = NULL;
1431 snapshot_ref = s->overlay;
1432 }
1433 break;
1434 case TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC:
1435 {
1436 BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1437 device = s->device;
1438 node_name = s->node_name;
1439 new_image_file = s->snapshot_file;
1440 snapshot_ref = NULL;
1441 }
1442 break;
1443 default:
1444 g_assert_not_reached();
1445 }
1446
1447 /* start processing */
1448 if (action_check_completion_mode(common, errp) < 0) {
1449 return;
1450 }
1451
1452 state->old_bs = bdrv_lookup_bs(device, node_name, errp);
1453 if (!state->old_bs) {
1454 return;
1455 }
1456
1457 aio_context = bdrv_get_aio_context(state->old_bs);
1458 aio_context_acquire(aio_context);
1459
1460 /* Paired with .clean() */
1461 bdrv_drained_begin(state->old_bs);
1462
1463 if (!bdrv_is_inserted(state->old_bs)) {
1464 error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1465 goto out;
1466 }
1467
1468 if (bdrv_op_is_blocked(state->old_bs,
1469 BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT, errp)) {
1470 goto out;
1471 }
1472
1473 if (!bdrv_is_read_only(state->old_bs)) {
1474 if (bdrv_flush(state->old_bs)) {
1475 error_setg(errp, QERR_IO_ERROR);
1476 goto out;
1477 }
1478 }
1479
1480 if (action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC) {
1481 BlockdevSnapshotSync *s = action->u.blockdev_snapshot_sync.data;
1482 const char *format = s->format ?: "qcow2";
1483 enum NewImageMode mode;
1484 const char *snapshot_node_name = s->snapshot_node_name;
1485
1486 if (node_name && !snapshot_node_name) {
1487 error_setg(errp, "New overlay node-name missing");
1488 goto out;
1489 }
1490
1491 if (snapshot_node_name &&
1492 bdrv_lookup_bs(snapshot_node_name, snapshot_node_name, NULL)) {
1493 error_setg(errp, "New overlay node-name already in use");
1494 goto out;
1495 }
1496
1497 flags = state->old_bs->open_flags;
1498 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_COPY_ON_READ);
1499 flags |= BDRV_O_NO_BACKING;
1500
1501 /* create new image w/backing file */
1502 mode = s->has_mode ? s->mode : NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1503 if (mode != NEW_IMAGE_MODE_EXISTING) {
1504 int64_t size = bdrv_getlength(state->old_bs);
1505 if (size < 0) {
1506 error_setg_errno(errp, -size, "bdrv_getlength failed");
1507 goto out;
1508 }
1509 bdrv_refresh_filename(state->old_bs);
1510 bdrv_img_create(new_image_file, format,
1511 state->old_bs->filename,
1512 state->old_bs->drv->format_name,
1513 NULL, size, flags, false, &local_err);
1514 if (local_err) {
1515 error_propagate(errp, local_err);
1516 goto out;
1517 }
1518 }
1519
1520 options = qdict_new();
1521 if (snapshot_node_name) {
1522 qdict_put_str(options, "node-name", snapshot_node_name);
1523 }
1524 qdict_put_str(options, "driver", format);
1525 }
1526
1527 state->new_bs = bdrv_open(new_image_file, snapshot_ref, options, flags,
1528 errp);
1529 /* We will manually add the backing_hd field to the bs later */
1530 if (!state->new_bs) {
1531 goto out;
1532 }
1533
1534 /*
1535 * Allow attaching a backing file to an overlay that's already in use only
1536 * if the parents don't assume that they are already seeing a valid image.
1537 * (Specifically, allow it as a mirror target, which is write-only access.)
1538 */
1539 bdrv_get_cumulative_perm(state->new_bs, &perm, &shared);
1540 if (perm & BLK_PERM_CONSISTENT_READ) {
1541 error_setg(errp, "The overlay is already in use");
1542 goto out;
1543 }
1544
1545 if (state->new_bs->drv->is_filter) {
1546 error_setg(errp, "Filters cannot be used as overlays");
1547 goto out;
1548 }
1549
1550 if (bdrv_cow_child(state->new_bs)) {
1551 error_setg(errp, "The overlay already has a backing image");
1552 goto out;
1553 }
1554
1555 if (!state->new_bs->drv->supports_backing) {
1556 error_setg(errp, "The overlay does not support backing images");
1557 goto out;
1558 }
1559
1560 ret = bdrv_append(state->new_bs, state->old_bs, errp);
1561 if (ret < 0) {
1562 goto out;
1563 }
1564 state->overlay_appended = true;
1565
1566 out:
1567 aio_context_release(aio_context);
1568 }
1569
1570 static void external_snapshot_commit(BlkActionState *common)
1571 {
1572 ExternalSnapshotState *state =
1573 DO_UPCAST(ExternalSnapshotState, common, common);
1574 AioContext *aio_context;
1575
1576 aio_context = bdrv_get_aio_context(state->old_bs);
1577 aio_context_acquire(aio_context);
1578
1579 /* We don't need (or want) to use the transactional
1580 * bdrv_reopen_multiple() across all the entries at once, because we
1581 * don't want to abort all of them if one of them fails the reopen */
1582 if (!qatomic_read(&state->old_bs->copy_on_read)) {
1583 bdrv_reopen_set_read_only(state->old_bs, true, NULL);
1584 }
1585
1586 aio_context_release(aio_context);
1587 }
1588
1589 static void external_snapshot_abort(BlkActionState *common)
1590 {
1591 ExternalSnapshotState *state =
1592 DO_UPCAST(ExternalSnapshotState, common, common);
1593 if (state->new_bs) {
1594 if (state->overlay_appended) {
1595 AioContext *aio_context;
1596 AioContext *tmp_context;
1597 int ret;
1598
1599 aio_context = bdrv_get_aio_context(state->old_bs);
1600 aio_context_acquire(aio_context);
1601
1602 bdrv_ref(state->old_bs); /* we can't let bdrv_set_backind_hd()
1603 close state->old_bs; we need it */
1604 bdrv_set_backing_hd(state->new_bs, NULL, &error_abort);
1605
1606 /*
1607 * The call to bdrv_set_backing_hd() above returns state->old_bs to
1608 * the main AioContext. As we're still going to be using it, return
1609 * it to the AioContext it was before.
1610 */
1611 tmp_context = bdrv_get_aio_context(state->old_bs);
1612 if (aio_context != tmp_context) {
1613 aio_context_release(aio_context);
1614 aio_context_acquire(tmp_context);
1615
1616 ret = bdrv_try_change_aio_context(state->old_bs,
1617 aio_context, NULL, NULL);
1618 assert(ret == 0);
1619
1620 aio_context_release(tmp_context);
1621 aio_context_acquire(aio_context);
1622 }
1623
1624 bdrv_replace_node(state->new_bs, state->old_bs, &error_abort);
1625 bdrv_unref(state->old_bs); /* bdrv_replace_node() ref'ed old_bs */
1626
1627 aio_context_release(aio_context);
1628 }
1629 }
1630 }
1631
1632 static void external_snapshot_clean(BlkActionState *common)
1633 {
1634 ExternalSnapshotState *state =
1635 DO_UPCAST(ExternalSnapshotState, common, common);
1636 AioContext *aio_context;
1637
1638 if (!state->old_bs) {
1639 return;
1640 }
1641
1642 aio_context = bdrv_get_aio_context(state->old_bs);
1643 aio_context_acquire(aio_context);
1644
1645 bdrv_drained_end(state->old_bs);
1646 bdrv_unref(state->new_bs);
1647
1648 aio_context_release(aio_context);
1649 }
1650
1651 typedef struct DriveBackupState {
1652 BlkActionState common;
1653 BlockDriverState *bs;
1654 BlockJob *job;
1655 } DriveBackupState;
1656
1657 static BlockJob *do_backup_common(BackupCommon *backup,
1658 BlockDriverState *bs,
1659 BlockDriverState *target_bs,
1660 AioContext *aio_context,
1661 JobTxn *txn, Error **errp);
1662
1663 static void drive_backup_prepare(BlkActionState *common, Error **errp)
1664 {
1665 DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1666 DriveBackup *backup;
1667 BlockDriverState *bs;
1668 BlockDriverState *target_bs;
1669 BlockDriverState *source = NULL;
1670 AioContext *aio_context;
1671 AioContext *old_context;
1672 const char *format;
1673 QDict *options;
1674 Error *local_err = NULL;
1675 int flags;
1676 int64_t size;
1677 bool set_backing_hd = false;
1678 int ret;
1679
1680 assert(common->action->type == TRANSACTION_ACTION_KIND_DRIVE_BACKUP);
1681 backup = common->action->u.drive_backup.data;
1682
1683 if (!backup->has_mode) {
1684 backup->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
1685 }
1686
1687 bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1688 if (!bs) {
1689 return;
1690 }
1691
1692 if (!bs->drv) {
1693 error_setg(errp, "Device has no medium");
1694 return;
1695 }
1696
1697 aio_context = bdrv_get_aio_context(bs);
1698 aio_context_acquire(aio_context);
1699
1700 state->bs = bs;
1701 /* Paired with .clean() */
1702 bdrv_drained_begin(bs);
1703
1704 format = backup->format;
1705 if (!format && backup->mode != NEW_IMAGE_MODE_EXISTING) {
1706 format = bs->drv->format_name;
1707 }
1708
1709 /* Early check to avoid creating target */
1710 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
1711 goto out;
1712 }
1713
1714 flags = bs->open_flags | BDRV_O_RDWR;
1715
1716 /*
1717 * See if we have a backing HD we can use to create our new image
1718 * on top of.
1719 */
1720 if (backup->sync == MIRROR_SYNC_MODE_TOP) {
1721 /*
1722 * Backup will not replace the source by the target, so none
1723 * of the filters skipped here will be removed (in contrast to
1724 * mirror). Therefore, we can skip all of them when looking
1725 * for the first COW relationship.
1726 */
1727 source = bdrv_cow_bs(bdrv_skip_filters(bs));
1728 if (!source) {
1729 backup->sync = MIRROR_SYNC_MODE_FULL;
1730 }
1731 }
1732 if (backup->sync == MIRROR_SYNC_MODE_NONE) {
1733 source = bs;
1734 flags |= BDRV_O_NO_BACKING;
1735 set_backing_hd = true;
1736 }
1737
1738 size = bdrv_getlength(bs);
1739 if (size < 0) {
1740 error_setg_errno(errp, -size, "bdrv_getlength failed");
1741 goto out;
1742 }
1743
1744 if (backup->mode != NEW_IMAGE_MODE_EXISTING) {
1745 assert(format);
1746 if (source) {
1747 /* Implicit filters should not appear in the filename */
1748 BlockDriverState *explicit_backing =
1749 bdrv_skip_implicit_filters(source);
1750
1751 bdrv_refresh_filename(explicit_backing);
1752 bdrv_img_create(backup->target, format,
1753 explicit_backing->filename,
1754 explicit_backing->drv->format_name, NULL,
1755 size, flags, false, &local_err);
1756 } else {
1757 bdrv_img_create(backup->target, format, NULL, NULL, NULL,
1758 size, flags, false, &local_err);
1759 }
1760 }
1761
1762 if (local_err) {
1763 error_propagate(errp, local_err);
1764 goto out;
1765 }
1766
1767 options = qdict_new();
1768 qdict_put_str(options, "discard", "unmap");
1769 qdict_put_str(options, "detect-zeroes", "unmap");
1770 if (format) {
1771 qdict_put_str(options, "driver", format);
1772 }
1773
1774 target_bs = bdrv_open(backup->target, NULL, options, flags, errp);
1775 if (!target_bs) {
1776 goto out;
1777 }
1778
1779 /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1780 old_context = bdrv_get_aio_context(target_bs);
1781 aio_context_release(aio_context);
1782 aio_context_acquire(old_context);
1783
1784 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1785 if (ret < 0) {
1786 bdrv_unref(target_bs);
1787 aio_context_release(old_context);
1788 return;
1789 }
1790
1791 aio_context_release(old_context);
1792 aio_context_acquire(aio_context);
1793
1794 if (set_backing_hd) {
1795 if (bdrv_set_backing_hd(target_bs, source, errp) < 0) {
1796 goto unref;
1797 }
1798 }
1799
1800 state->job = do_backup_common(qapi_DriveBackup_base(backup),
1801 bs, target_bs, aio_context,
1802 common->block_job_txn, errp);
1803
1804 unref:
1805 bdrv_unref(target_bs);
1806 out:
1807 aio_context_release(aio_context);
1808 }
1809
1810 static void drive_backup_commit(BlkActionState *common)
1811 {
1812 DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1813 AioContext *aio_context;
1814
1815 aio_context = bdrv_get_aio_context(state->bs);
1816 aio_context_acquire(aio_context);
1817
1818 assert(state->job);
1819 job_start(&state->job->job);
1820
1821 aio_context_release(aio_context);
1822 }
1823
1824 static void drive_backup_abort(BlkActionState *common)
1825 {
1826 DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1827
1828 if (state->job) {
1829 job_cancel_sync(&state->job->job, true);
1830 }
1831 }
1832
1833 static void drive_backup_clean(BlkActionState *common)
1834 {
1835 DriveBackupState *state = DO_UPCAST(DriveBackupState, common, common);
1836 AioContext *aio_context;
1837
1838 if (!state->bs) {
1839 return;
1840 }
1841
1842 aio_context = bdrv_get_aio_context(state->bs);
1843 aio_context_acquire(aio_context);
1844
1845 bdrv_drained_end(state->bs);
1846
1847 aio_context_release(aio_context);
1848 }
1849
1850 typedef struct BlockdevBackupState {
1851 BlkActionState common;
1852 BlockDriverState *bs;
1853 BlockJob *job;
1854 } BlockdevBackupState;
1855
1856 static void blockdev_backup_prepare(BlkActionState *common, Error **errp)
1857 {
1858 BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1859 BlockdevBackup *backup;
1860 BlockDriverState *bs;
1861 BlockDriverState *target_bs;
1862 AioContext *aio_context;
1863 AioContext *old_context;
1864 int ret;
1865
1866 assert(common->action->type == TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP);
1867 backup = common->action->u.blockdev_backup.data;
1868
1869 bs = bdrv_lookup_bs(backup->device, backup->device, errp);
1870 if (!bs) {
1871 return;
1872 }
1873
1874 target_bs = bdrv_lookup_bs(backup->target, backup->target, errp);
1875 if (!target_bs) {
1876 return;
1877 }
1878
1879 /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
1880 aio_context = bdrv_get_aio_context(bs);
1881 old_context = bdrv_get_aio_context(target_bs);
1882 aio_context_acquire(old_context);
1883
1884 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
1885 if (ret < 0) {
1886 aio_context_release(old_context);
1887 return;
1888 }
1889
1890 aio_context_release(old_context);
1891 aio_context_acquire(aio_context);
1892 state->bs = bs;
1893
1894 /* Paired with .clean() */
1895 bdrv_drained_begin(state->bs);
1896
1897 state->job = do_backup_common(qapi_BlockdevBackup_base(backup),
1898 bs, target_bs, aio_context,
1899 common->block_job_txn, errp);
1900
1901 aio_context_release(aio_context);
1902 }
1903
1904 static void blockdev_backup_commit(BlkActionState *common)
1905 {
1906 BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1907 AioContext *aio_context;
1908
1909 aio_context = bdrv_get_aio_context(state->bs);
1910 aio_context_acquire(aio_context);
1911
1912 assert(state->job);
1913 job_start(&state->job->job);
1914
1915 aio_context_release(aio_context);
1916 }
1917
1918 static void blockdev_backup_abort(BlkActionState *common)
1919 {
1920 BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1921
1922 if (state->job) {
1923 job_cancel_sync(&state->job->job, true);
1924 }
1925 }
1926
1927 static void blockdev_backup_clean(BlkActionState *common)
1928 {
1929 BlockdevBackupState *state = DO_UPCAST(BlockdevBackupState, common, common);
1930 AioContext *aio_context;
1931
1932 if (!state->bs) {
1933 return;
1934 }
1935
1936 aio_context = bdrv_get_aio_context(state->bs);
1937 aio_context_acquire(aio_context);
1938
1939 bdrv_drained_end(state->bs);
1940
1941 aio_context_release(aio_context);
1942 }
1943
1944 typedef struct BlockDirtyBitmapState {
1945 BlkActionState common;
1946 BdrvDirtyBitmap *bitmap;
1947 BlockDriverState *bs;
1948 HBitmap *backup;
1949 bool prepared;
1950 bool was_enabled;
1951 } BlockDirtyBitmapState;
1952
1953 static void block_dirty_bitmap_add_prepare(BlkActionState *common,
1954 Error **errp)
1955 {
1956 Error *local_err = NULL;
1957 BlockDirtyBitmapAdd *action;
1958 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1959 common, common);
1960
1961 if (action_check_completion_mode(common, errp) < 0) {
1962 return;
1963 }
1964
1965 action = common->action->u.block_dirty_bitmap_add.data;
1966 /* AIO context taken and released within qmp_block_dirty_bitmap_add */
1967 qmp_block_dirty_bitmap_add(action->node, action->name,
1968 action->has_granularity, action->granularity,
1969 action->has_persistent, action->persistent,
1970 action->has_disabled, action->disabled,
1971 &local_err);
1972
1973 if (!local_err) {
1974 state->prepared = true;
1975 } else {
1976 error_propagate(errp, local_err);
1977 }
1978 }
1979
1980 static void block_dirty_bitmap_add_abort(BlkActionState *common)
1981 {
1982 BlockDirtyBitmapAdd *action;
1983 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1984 common, common);
1985
1986 action = common->action->u.block_dirty_bitmap_add.data;
1987 /* Should not be able to fail: IF the bitmap was added via .prepare(),
1988 * then the node reference and bitmap name must have been valid.
1989 */
1990 if (state->prepared) {
1991 qmp_block_dirty_bitmap_remove(action->node, action->name, &error_abort);
1992 }
1993 }
1994
1995 static void block_dirty_bitmap_clear_prepare(BlkActionState *common,
1996 Error **errp)
1997 {
1998 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
1999 common, common);
2000 BlockDirtyBitmap *action;
2001
2002 if (action_check_completion_mode(common, errp) < 0) {
2003 return;
2004 }
2005
2006 action = common->action->u.block_dirty_bitmap_clear.data;
2007 state->bitmap = block_dirty_bitmap_lookup(action->node,
2008 action->name,
2009 &state->bs,
2010 errp);
2011 if (!state->bitmap) {
2012 return;
2013 }
2014
2015 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_DEFAULT, errp)) {
2016 return;
2017 }
2018
2019 bdrv_clear_dirty_bitmap(state->bitmap, &state->backup);
2020 }
2021
2022 static void block_dirty_bitmap_restore(BlkActionState *common)
2023 {
2024 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2025 common, common);
2026
2027 if (state->backup) {
2028 bdrv_restore_dirty_bitmap(state->bitmap, state->backup);
2029 }
2030 }
2031
2032 static void block_dirty_bitmap_free_backup(BlkActionState *common)
2033 {
2034 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2035 common, common);
2036
2037 hbitmap_free(state->backup);
2038 }
2039
2040 static void block_dirty_bitmap_enable_prepare(BlkActionState *common,
2041 Error **errp)
2042 {
2043 BlockDirtyBitmap *action;
2044 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2045 common, common);
2046
2047 if (action_check_completion_mode(common, errp) < 0) {
2048 return;
2049 }
2050
2051 action = common->action->u.block_dirty_bitmap_enable.data;
2052 state->bitmap = block_dirty_bitmap_lookup(action->node,
2053 action->name,
2054 NULL,
2055 errp);
2056 if (!state->bitmap) {
2057 return;
2058 }
2059
2060 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2061 return;
2062 }
2063
2064 state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2065 bdrv_enable_dirty_bitmap(state->bitmap);
2066 }
2067
2068 static void block_dirty_bitmap_enable_abort(BlkActionState *common)
2069 {
2070 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2071 common, common);
2072
2073 if (!state->was_enabled) {
2074 bdrv_disable_dirty_bitmap(state->bitmap);
2075 }
2076 }
2077
2078 static void block_dirty_bitmap_disable_prepare(BlkActionState *common,
2079 Error **errp)
2080 {
2081 BlockDirtyBitmap *action;
2082 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2083 common, common);
2084
2085 if (action_check_completion_mode(common, errp) < 0) {
2086 return;
2087 }
2088
2089 action = common->action->u.block_dirty_bitmap_disable.data;
2090 state->bitmap = block_dirty_bitmap_lookup(action->node,
2091 action->name,
2092 NULL,
2093 errp);
2094 if (!state->bitmap) {
2095 return;
2096 }
2097
2098 if (bdrv_dirty_bitmap_check(state->bitmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2099 return;
2100 }
2101
2102 state->was_enabled = bdrv_dirty_bitmap_enabled(state->bitmap);
2103 bdrv_disable_dirty_bitmap(state->bitmap);
2104 }
2105
2106 static void block_dirty_bitmap_disable_abort(BlkActionState *common)
2107 {
2108 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2109 common, common);
2110
2111 if (state->was_enabled) {
2112 bdrv_enable_dirty_bitmap(state->bitmap);
2113 }
2114 }
2115
2116 static void block_dirty_bitmap_merge_prepare(BlkActionState *common,
2117 Error **errp)
2118 {
2119 BlockDirtyBitmapMerge *action;
2120 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2121 common, common);
2122
2123 if (action_check_completion_mode(common, errp) < 0) {
2124 return;
2125 }
2126
2127 action = common->action->u.block_dirty_bitmap_merge.data;
2128
2129 state->bitmap = block_dirty_bitmap_merge(action->node, action->target,
2130 action->bitmaps, &state->backup,
2131 errp);
2132 }
2133
2134 static void block_dirty_bitmap_remove_prepare(BlkActionState *common,
2135 Error **errp)
2136 {
2137 BlockDirtyBitmap *action;
2138 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2139 common, common);
2140
2141 if (action_check_completion_mode(common, errp) < 0) {
2142 return;
2143 }
2144
2145 action = common->action->u.block_dirty_bitmap_remove.data;
2146
2147 state->bitmap = block_dirty_bitmap_remove(action->node, action->name,
2148 false, &state->bs, errp);
2149 if (state->bitmap) {
2150 bdrv_dirty_bitmap_skip_store(state->bitmap, true);
2151 bdrv_dirty_bitmap_set_busy(state->bitmap, true);
2152 }
2153 }
2154
2155 static void block_dirty_bitmap_remove_abort(BlkActionState *common)
2156 {
2157 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2158 common, common);
2159
2160 if (state->bitmap) {
2161 bdrv_dirty_bitmap_skip_store(state->bitmap, false);
2162 bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2163 }
2164 }
2165
2166 static void block_dirty_bitmap_remove_commit(BlkActionState *common)
2167 {
2168 BlockDirtyBitmapState *state = DO_UPCAST(BlockDirtyBitmapState,
2169 common, common);
2170
2171 bdrv_dirty_bitmap_set_busy(state->bitmap, false);
2172 bdrv_release_dirty_bitmap(state->bitmap);
2173 }
2174
2175 static void abort_prepare(BlkActionState *common, Error **errp)
2176 {
2177 error_setg(errp, "Transaction aborted using Abort action");
2178 }
2179
2180 static void abort_commit(BlkActionState *common)
2181 {
2182 g_assert_not_reached(); /* this action never succeeds */
2183 }
2184
2185 static const BlkActionOps actions[] = {
2186 [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT] = {
2187 .instance_size = sizeof(ExternalSnapshotState),
2188 .prepare = external_snapshot_prepare,
2189 .commit = external_snapshot_commit,
2190 .abort = external_snapshot_abort,
2191 .clean = external_snapshot_clean,
2192 },
2193 [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_SYNC] = {
2194 .instance_size = sizeof(ExternalSnapshotState),
2195 .prepare = external_snapshot_prepare,
2196 .commit = external_snapshot_commit,
2197 .abort = external_snapshot_abort,
2198 .clean = external_snapshot_clean,
2199 },
2200 [TRANSACTION_ACTION_KIND_DRIVE_BACKUP] = {
2201 .instance_size = sizeof(DriveBackupState),
2202 .prepare = drive_backup_prepare,
2203 .commit = drive_backup_commit,
2204 .abort = drive_backup_abort,
2205 .clean = drive_backup_clean,
2206 },
2207 [TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP] = {
2208 .instance_size = sizeof(BlockdevBackupState),
2209 .prepare = blockdev_backup_prepare,
2210 .commit = blockdev_backup_commit,
2211 .abort = blockdev_backup_abort,
2212 .clean = blockdev_backup_clean,
2213 },
2214 [TRANSACTION_ACTION_KIND_ABORT] = {
2215 .instance_size = sizeof(BlkActionState),
2216 .prepare = abort_prepare,
2217 .commit = abort_commit,
2218 },
2219 [TRANSACTION_ACTION_KIND_BLOCKDEV_SNAPSHOT_INTERNAL_SYNC] = {
2220 .instance_size = sizeof(InternalSnapshotState),
2221 .prepare = internal_snapshot_prepare,
2222 .abort = internal_snapshot_abort,
2223 .clean = internal_snapshot_clean,
2224 },
2225 [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ADD] = {
2226 .instance_size = sizeof(BlockDirtyBitmapState),
2227 .prepare = block_dirty_bitmap_add_prepare,
2228 .abort = block_dirty_bitmap_add_abort,
2229 },
2230 [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_CLEAR] = {
2231 .instance_size = sizeof(BlockDirtyBitmapState),
2232 .prepare = block_dirty_bitmap_clear_prepare,
2233 .commit = block_dirty_bitmap_free_backup,
2234 .abort = block_dirty_bitmap_restore,
2235 },
2236 [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_ENABLE] = {
2237 .instance_size = sizeof(BlockDirtyBitmapState),
2238 .prepare = block_dirty_bitmap_enable_prepare,
2239 .abort = block_dirty_bitmap_enable_abort,
2240 },
2241 [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_DISABLE] = {
2242 .instance_size = sizeof(BlockDirtyBitmapState),
2243 .prepare = block_dirty_bitmap_disable_prepare,
2244 .abort = block_dirty_bitmap_disable_abort,
2245 },
2246 [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_MERGE] = {
2247 .instance_size = sizeof(BlockDirtyBitmapState),
2248 .prepare = block_dirty_bitmap_merge_prepare,
2249 .commit = block_dirty_bitmap_free_backup,
2250 .abort = block_dirty_bitmap_restore,
2251 },
2252 [TRANSACTION_ACTION_KIND_BLOCK_DIRTY_BITMAP_REMOVE] = {
2253 .instance_size = sizeof(BlockDirtyBitmapState),
2254 .prepare = block_dirty_bitmap_remove_prepare,
2255 .commit = block_dirty_bitmap_remove_commit,
2256 .abort = block_dirty_bitmap_remove_abort,
2257 },
2258 /* Where are transactions for MIRROR, COMMIT and STREAM?
2259 * Although these blockjobs use transaction callbacks like the backup job,
2260 * these jobs do not necessarily adhere to transaction semantics.
2261 * These jobs may not fully undo all of their actions on abort, nor do they
2262 * necessarily work in transactions with more than one job in them.
2263 */
2264 };
2265
2266 /**
2267 * Allocate a TransactionProperties structure if necessary, and fill
2268 * that structure with desired defaults if they are unset.
2269 */
2270 static TransactionProperties *get_transaction_properties(
2271 TransactionProperties *props)
2272 {
2273 if (!props) {
2274 props = g_new0(TransactionProperties, 1);
2275 }
2276
2277 if (!props->has_completion_mode) {
2278 props->has_completion_mode = true;
2279 props->completion_mode = ACTION_COMPLETION_MODE_INDIVIDUAL;
2280 }
2281
2282 return props;
2283 }
2284
2285 /*
2286 * 'Atomic' group operations. The operations are performed as a set, and if
2287 * any fail then we roll back all operations in the group.
2288 *
2289 * Always run under BQL.
2290 */
2291 void qmp_transaction(TransactionActionList *dev_list,
2292 struct TransactionProperties *props,
2293 Error **errp)
2294 {
2295 TransactionActionList *dev_entry = dev_list;
2296 bool has_props = !!props;
2297 JobTxn *block_job_txn = NULL;
2298 BlkActionState *state, *next;
2299 Error *local_err = NULL;
2300
2301 GLOBAL_STATE_CODE();
2302
2303 QTAILQ_HEAD(, BlkActionState) snap_bdrv_states;
2304 QTAILQ_INIT(&snap_bdrv_states);
2305
2306 /* Does this transaction get canceled as a group on failure?
2307 * If not, we don't really need to make a JobTxn.
2308 */
2309 props = get_transaction_properties(props);
2310 if (props->completion_mode != ACTION_COMPLETION_MODE_INDIVIDUAL) {
2311 block_job_txn = job_txn_new();
2312 }
2313
2314 /* drain all i/o before any operations */
2315 bdrv_drain_all();
2316
2317 /* We don't do anything in this loop that commits us to the operations */
2318 while (NULL != dev_entry) {
2319 TransactionAction *dev_info = NULL;
2320 const BlkActionOps *ops;
2321
2322 dev_info = dev_entry->value;
2323 dev_entry = dev_entry->next;
2324
2325 assert(dev_info->type < ARRAY_SIZE(actions));
2326
2327 ops = &actions[dev_info->type];
2328 assert(ops->instance_size > 0);
2329
2330 state = g_malloc0(ops->instance_size);
2331 state->ops = ops;
2332 state->action = dev_info;
2333 state->block_job_txn = block_job_txn;
2334 state->txn_props = props;
2335 QTAILQ_INSERT_TAIL(&snap_bdrv_states, state, entry);
2336
2337 state->ops->prepare(state, &local_err);
2338 if (local_err) {
2339 error_propagate(errp, local_err);
2340 goto delete_and_fail;
2341 }
2342 }
2343
2344 QTAILQ_FOREACH(state, &snap_bdrv_states, entry) {
2345 if (state->ops->commit) {
2346 state->ops->commit(state);
2347 }
2348 }
2349
2350 /* success */
2351 goto exit;
2352
2353 delete_and_fail:
2354 /* failure, and it is all-or-none; roll back all operations */
2355 QTAILQ_FOREACH_REVERSE(state, &snap_bdrv_states, entry) {
2356 if (state->ops->abort) {
2357 state->ops->abort(state);
2358 }
2359 }
2360 exit:
2361 QTAILQ_FOREACH_SAFE(state, &snap_bdrv_states, entry, next) {
2362 if (state->ops->clean) {
2363 state->ops->clean(state);
2364 }
2365 g_free(state);
2366 }
2367 if (!has_props) {
2368 qapi_free_TransactionProperties(props);
2369 }
2370 job_txn_unref(block_job_txn);
2371 }
2372
2373 BlockDirtyBitmapSha256 *qmp_x_debug_block_dirty_bitmap_sha256(const char *node,
2374 const char *name,
2375 Error **errp)
2376 {
2377 BdrvDirtyBitmap *bitmap;
2378 BlockDriverState *bs;
2379 BlockDirtyBitmapSha256 *ret = NULL;
2380 char *sha256;
2381
2382 bitmap = block_dirty_bitmap_lookup(node, name, &bs, errp);
2383 if (!bitmap || !bs) {
2384 return NULL;
2385 }
2386
2387 sha256 = bdrv_dirty_bitmap_sha256(bitmap, errp);
2388 if (sha256 == NULL) {
2389 return NULL;
2390 }
2391
2392 ret = g_new(BlockDirtyBitmapSha256, 1);
2393 ret->sha256 = sha256;
2394
2395 return ret;
2396 }
2397
2398 void coroutine_fn qmp_block_resize(const char *device, const char *node_name,
2399 int64_t size, Error **errp)
2400 {
2401 Error *local_err = NULL;
2402 BlockBackend *blk;
2403 BlockDriverState *bs;
2404 AioContext *old_ctx;
2405
2406 bs = bdrv_lookup_bs(device, node_name, &local_err);
2407 if (local_err) {
2408 error_propagate(errp, local_err);
2409 return;
2410 }
2411
2412 if (size < 0) {
2413 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
2414 return;
2415 }
2416
2417 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
2418 error_setg(errp, QERR_DEVICE_IN_USE, device);
2419 return;
2420 }
2421
2422 blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL, errp);
2423 if (!blk) {
2424 return;
2425 }
2426
2427 bdrv_co_lock(bs);
2428 bdrv_drained_begin(bs);
2429 bdrv_co_unlock(bs);
2430
2431 old_ctx = bdrv_co_enter(bs);
2432 blk_co_truncate(blk, size, false, PREALLOC_MODE_OFF, 0, errp);
2433 bdrv_co_leave(bs, old_ctx);
2434
2435 bdrv_co_lock(bs);
2436 bdrv_drained_end(bs);
2437 blk_unref(blk);
2438 bdrv_co_unlock(bs);
2439 }
2440
2441 void qmp_block_stream(const char *job_id, const char *device,
2442 const char *base,
2443 const char *base_node,
2444 const char *backing_file,
2445 const char *bottom,
2446 bool has_speed, int64_t speed,
2447 bool has_on_error, BlockdevOnError on_error,
2448 const char *filter_node_name,
2449 bool has_auto_finalize, bool auto_finalize,
2450 bool has_auto_dismiss, bool auto_dismiss,
2451 Error **errp)
2452 {
2453 BlockDriverState *bs, *iter, *iter_end;
2454 BlockDriverState *base_bs = NULL;
2455 BlockDriverState *bottom_bs = NULL;
2456 AioContext *aio_context;
2457 Error *local_err = NULL;
2458 int job_flags = JOB_DEFAULT;
2459
2460 if (base && base_node) {
2461 error_setg(errp, "'base' and 'base-node' cannot be specified "
2462 "at the same time");
2463 return;
2464 }
2465
2466 if (base && bottom) {
2467 error_setg(errp, "'base' and 'bottom' cannot be specified "
2468 "at the same time");
2469 return;
2470 }
2471
2472 if (bottom && base_node) {
2473 error_setg(errp, "'bottom' and 'base-node' cannot be specified "
2474 "at the same time");
2475 return;
2476 }
2477
2478 if (!has_on_error) {
2479 on_error = BLOCKDEV_ON_ERROR_REPORT;
2480 }
2481
2482 bs = bdrv_lookup_bs(device, device, errp);
2483 if (!bs) {
2484 return;
2485 }
2486
2487 aio_context = bdrv_get_aio_context(bs);
2488 aio_context_acquire(aio_context);
2489
2490 if (base) {
2491 base_bs = bdrv_find_backing_image(bs, base);
2492 if (base_bs == NULL) {
2493 error_setg(errp, "Can't find '%s' in the backing chain", base);
2494 goto out;
2495 }
2496 assert(bdrv_get_aio_context(base_bs) == aio_context);
2497 }
2498
2499 if (base_node) {
2500 base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2501 if (!base_bs) {
2502 goto out;
2503 }
2504 if (bs == base_bs || !bdrv_chain_contains(bs, base_bs)) {
2505 error_setg(errp, "Node '%s' is not a backing image of '%s'",
2506 base_node, device);
2507 goto out;
2508 }
2509 assert(bdrv_get_aio_context(base_bs) == aio_context);
2510 bdrv_refresh_filename(base_bs);
2511 }
2512
2513 if (bottom) {
2514 bottom_bs = bdrv_lookup_bs(NULL, bottom, errp);
2515 if (!bottom_bs) {
2516 goto out;
2517 }
2518 if (!bottom_bs->drv) {
2519 error_setg(errp, "Node '%s' is not open", bottom);
2520 goto out;
2521 }
2522 if (bottom_bs->drv->is_filter) {
2523 error_setg(errp, "Node '%s' is a filter, use a non-filter node "
2524 "as 'bottom'", bottom);
2525 goto out;
2526 }
2527 if (!bdrv_chain_contains(bs, bottom_bs)) {
2528 error_setg(errp, "Node '%s' is not in a chain starting from '%s'",
2529 bottom, device);
2530 goto out;
2531 }
2532 assert(bdrv_get_aio_context(bottom_bs) == aio_context);
2533 }
2534
2535 /*
2536 * Check for op blockers in the whole chain between bs and base (or bottom)
2537 */
2538 iter_end = bottom ? bdrv_filter_or_cow_bs(bottom_bs) : base_bs;
2539 for (iter = bs; iter && iter != iter_end;
2540 iter = bdrv_filter_or_cow_bs(iter))
2541 {
2542 if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_STREAM, errp)) {
2543 goto out;
2544 }
2545 }
2546
2547 /* if we are streaming the entire chain, the result will have no backing
2548 * file, and specifying one is therefore an error */
2549 if (!base_bs && backing_file) {
2550 error_setg(errp, "backing file specified, but streaming the "
2551 "entire chain");
2552 goto out;
2553 }
2554
2555 if (has_auto_finalize && !auto_finalize) {
2556 job_flags |= JOB_MANUAL_FINALIZE;
2557 }
2558 if (has_auto_dismiss && !auto_dismiss) {
2559 job_flags |= JOB_MANUAL_DISMISS;
2560 }
2561
2562 stream_start(job_id, bs, base_bs, backing_file,
2563 bottom_bs, job_flags, has_speed ? speed : 0, on_error,
2564 filter_node_name, &local_err);
2565 if (local_err) {
2566 error_propagate(errp, local_err);
2567 goto out;
2568 }
2569
2570 trace_qmp_block_stream(bs);
2571
2572 out:
2573 aio_context_release(aio_context);
2574 }
2575
2576 void qmp_block_commit(const char *job_id, const char *device,
2577 const char *base_node,
2578 const char *base,
2579 const char *top_node,
2580 const char *top,
2581 const char *backing_file,
2582 bool has_speed, int64_t speed,
2583 bool has_on_error, BlockdevOnError on_error,
2584 const char *filter_node_name,
2585 bool has_auto_finalize, bool auto_finalize,
2586 bool has_auto_dismiss, bool auto_dismiss,
2587 Error **errp)
2588 {
2589 BlockDriverState *bs;
2590 BlockDriverState *iter;
2591 BlockDriverState *base_bs, *top_bs;
2592 AioContext *aio_context;
2593 Error *local_err = NULL;
2594 int job_flags = JOB_DEFAULT;
2595 uint64_t top_perm, top_shared;
2596
2597 if (!has_speed) {
2598 speed = 0;
2599 }
2600 if (!has_on_error) {
2601 on_error = BLOCKDEV_ON_ERROR_REPORT;
2602 }
2603 if (has_auto_finalize && !auto_finalize) {
2604 job_flags |= JOB_MANUAL_FINALIZE;
2605 }
2606 if (has_auto_dismiss && !auto_dismiss) {
2607 job_flags |= JOB_MANUAL_DISMISS;
2608 }
2609
2610 /* Important Note:
2611 * libvirt relies on the DeviceNotFound error class in order to probe for
2612 * live commit feature versions; for this to work, we must make sure to
2613 * perform the device lookup before any generic errors that may occur in a
2614 * scenario in which all optional arguments are omitted. */
2615 bs = qmp_get_root_bs(device, &local_err);
2616 if (!bs) {
2617 bs = bdrv_lookup_bs(device, device, NULL);
2618 if (!bs) {
2619 error_free(local_err);
2620 error_set(errp, ERROR_CLASS_DEVICE_NOT_FOUND,
2621 "Device '%s' not found", device);
2622 } else {
2623 error_propagate(errp, local_err);
2624 }
2625 return;
2626 }
2627
2628 aio_context = bdrv_get_aio_context(bs);
2629 aio_context_acquire(aio_context);
2630
2631 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, errp)) {
2632 goto out;
2633 }
2634
2635 /* default top_bs is the active layer */
2636 top_bs = bs;
2637
2638 if (top_node && top) {
2639 error_setg(errp, "'top-node' and 'top' are mutually exclusive");
2640 goto out;
2641 } else if (top_node) {
2642 top_bs = bdrv_lookup_bs(NULL, top_node, errp);
2643 if (top_bs == NULL) {
2644 goto out;
2645 }
2646 if (!bdrv_chain_contains(bs, top_bs)) {
2647 error_setg(errp, "'%s' is not in this backing file chain",
2648 top_node);
2649 goto out;
2650 }
2651 } else if (top) {
2652 /* This strcmp() is just a shortcut, there is no need to
2653 * refresh @bs's filename. If it mismatches,
2654 * bdrv_find_backing_image() will do the refresh and may still
2655 * return @bs. */
2656 if (strcmp(bs->filename, top) != 0) {
2657 top_bs = bdrv_find_backing_image(bs, top);
2658 }
2659 }
2660
2661 if (top_bs == NULL) {
2662 error_setg(errp, "Top image file %s not found", top ? top : "NULL");
2663 goto out;
2664 }
2665
2666 assert(bdrv_get_aio_context(top_bs) == aio_context);
2667
2668 if (base_node && base) {
2669 error_setg(errp, "'base-node' and 'base' are mutually exclusive");
2670 goto out;
2671 } else if (base_node) {
2672 base_bs = bdrv_lookup_bs(NULL, base_node, errp);
2673 if (base_bs == NULL) {
2674 goto out;
2675 }
2676 if (!bdrv_chain_contains(top_bs, base_bs)) {
2677 error_setg(errp, "'%s' is not in this backing file chain",
2678 base_node);
2679 goto out;
2680 }
2681 } else if (base) {
2682 base_bs = bdrv_find_backing_image(top_bs, base);
2683 if (base_bs == NULL) {
2684 error_setg(errp, "Can't find '%s' in the backing chain", base);
2685 goto out;
2686 }
2687 } else {
2688 base_bs = bdrv_find_base(top_bs);
2689 if (base_bs == NULL) {
2690 error_setg(errp, "There is no backimg image");
2691 goto out;
2692 }
2693 }
2694
2695 assert(bdrv_get_aio_context(base_bs) == aio_context);
2696
2697 for (iter = top_bs; iter != bdrv_filter_or_cow_bs(base_bs);
2698 iter = bdrv_filter_or_cow_bs(iter))
2699 {
2700 if (bdrv_op_is_blocked(iter, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2701 goto out;
2702 }
2703 }
2704
2705 /* Do not allow attempts to commit an image into itself */
2706 if (top_bs == base_bs) {
2707 error_setg(errp, "cannot commit an image into itself");
2708 goto out;
2709 }
2710
2711 /*
2712 * Active commit is required if and only if someone has taken a
2713 * WRITE permission on the top node. Historically, we have always
2714 * used active commit for top nodes, so continue that practice
2715 * lest we possibly break clients that rely on this behavior, e.g.
2716 * to later attach this node to a writing parent.
2717 * (Active commit is never really wrong.)
2718 */
2719 bdrv_get_cumulative_perm(top_bs, &top_perm, &top_shared);
2720 if (top_perm & BLK_PERM_WRITE ||
2721 bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs))
2722 {
2723 if (backing_file) {
2724 if (bdrv_skip_filters(top_bs) == bdrv_skip_filters(bs)) {
2725 error_setg(errp, "'backing-file' specified,"
2726 " but 'top' is the active layer");
2727 } else {
2728 error_setg(errp, "'backing-file' specified, but 'top' has a "
2729 "writer on it");
2730 }
2731 goto out;
2732 }
2733 if (!job_id) {
2734 /*
2735 * Emulate here what block_job_create() does, because it
2736 * is possible that @bs != @top_bs (the block job should
2737 * be named after @bs, even if @top_bs is the actual
2738 * source)
2739 */
2740 job_id = bdrv_get_device_name(bs);
2741 }
2742 commit_active_start(job_id, top_bs, base_bs, job_flags, speed, on_error,
2743 filter_node_name, NULL, NULL, false, &local_err);
2744 } else {
2745 BlockDriverState *overlay_bs = bdrv_find_overlay(bs, top_bs);
2746 if (bdrv_op_is_blocked(overlay_bs, BLOCK_OP_TYPE_COMMIT_TARGET, errp)) {
2747 goto out;
2748 }
2749 commit_start(job_id, bs, base_bs, top_bs, job_flags,
2750 speed, on_error, backing_file,
2751 filter_node_name, &local_err);
2752 }
2753 if (local_err != NULL) {
2754 error_propagate(errp, local_err);
2755 goto out;
2756 }
2757
2758 out:
2759 aio_context_release(aio_context);
2760 }
2761
2762 /* Common QMP interface for drive-backup and blockdev-backup */
2763 static BlockJob *do_backup_common(BackupCommon *backup,
2764 BlockDriverState *bs,
2765 BlockDriverState *target_bs,
2766 AioContext *aio_context,
2767 JobTxn *txn, Error **errp)
2768 {
2769 BlockJob *job = NULL;
2770 BdrvDirtyBitmap *bmap = NULL;
2771 BackupPerf perf = { .max_workers = 64 };
2772 int job_flags = JOB_DEFAULT;
2773
2774 if (!backup->has_speed) {
2775 backup->speed = 0;
2776 }
2777 if (!backup->has_on_source_error) {
2778 backup->on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2779 }
2780 if (!backup->has_on_target_error) {
2781 backup->on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2782 }
2783 if (!backup->has_auto_finalize) {
2784 backup->auto_finalize = true;
2785 }
2786 if (!backup->has_auto_dismiss) {
2787 backup->auto_dismiss = true;
2788 }
2789 if (!backup->has_compress) {
2790 backup->compress = false;
2791 }
2792
2793 if (backup->x_perf) {
2794 if (backup->x_perf->has_use_copy_range) {
2795 perf.use_copy_range = backup->x_perf->use_copy_range;
2796 }
2797 if (backup->x_perf->has_max_workers) {
2798 perf.max_workers = backup->x_perf->max_workers;
2799 }
2800 if (backup->x_perf->has_max_chunk) {
2801 perf.max_chunk = backup->x_perf->max_chunk;
2802 }
2803 }
2804
2805 if ((backup->sync == MIRROR_SYNC_MODE_BITMAP) ||
2806 (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL)) {
2807 /* done before desugaring 'incremental' to print the right message */
2808 if (!backup->bitmap) {
2809 error_setg(errp, "must provide a valid bitmap name for "
2810 "'%s' sync mode", MirrorSyncMode_str(backup->sync));
2811 return NULL;
2812 }
2813 }
2814
2815 if (backup->sync == MIRROR_SYNC_MODE_INCREMENTAL) {
2816 if (backup->has_bitmap_mode &&
2817 backup->bitmap_mode != BITMAP_SYNC_MODE_ON_SUCCESS) {
2818 error_setg(errp, "Bitmap sync mode must be '%s' "
2819 "when using sync mode '%s'",
2820 BitmapSyncMode_str(BITMAP_SYNC_MODE_ON_SUCCESS),
2821 MirrorSyncMode_str(backup->sync));
2822 return NULL;
2823 }
2824 backup->has_bitmap_mode = true;
2825 backup->sync = MIRROR_SYNC_MODE_BITMAP;
2826 backup->bitmap_mode = BITMAP_SYNC_MODE_ON_SUCCESS;
2827 }
2828
2829 if (backup->bitmap) {
2830 bmap = bdrv_find_dirty_bitmap(bs, backup->bitmap);
2831 if (!bmap) {
2832 error_setg(errp, "Bitmap '%s' could not be found", backup->bitmap);
2833 return NULL;
2834 }
2835 if (!backup->has_bitmap_mode) {
2836 error_setg(errp, "Bitmap sync mode must be given "
2837 "when providing a bitmap");
2838 return NULL;
2839 }
2840 if (bdrv_dirty_bitmap_check(bmap, BDRV_BITMAP_ALLOW_RO, errp)) {
2841 return NULL;
2842 }
2843
2844 /* This does not produce a useful bitmap artifact: */
2845 if (backup->sync == MIRROR_SYNC_MODE_NONE) {
2846 error_setg(errp, "sync mode '%s' does not produce meaningful bitmap"
2847 " outputs", MirrorSyncMode_str(backup->sync));
2848 return NULL;
2849 }
2850
2851 /* If the bitmap isn't used for input or output, this is useless: */
2852 if (backup->bitmap_mode == BITMAP_SYNC_MODE_NEVER &&
2853 backup->sync != MIRROR_SYNC_MODE_BITMAP) {
2854 error_setg(errp, "Bitmap sync mode '%s' has no meaningful effect"
2855 " when combined with sync mode '%s'",
2856 BitmapSyncMode_str(backup->bitmap_mode),
2857 MirrorSyncMode_str(backup->sync));
2858 return NULL;
2859 }
2860 }
2861
2862 if (!backup->bitmap && backup->has_bitmap_mode) {
2863 error_setg(errp, "Cannot specify bitmap sync mode without a bitmap");
2864 return NULL;
2865 }
2866
2867 if (!backup->auto_finalize) {
2868 job_flags |= JOB_MANUAL_FINALIZE;
2869 }
2870 if (!backup->auto_dismiss) {
2871 job_flags |= JOB_MANUAL_DISMISS;
2872 }
2873
2874 job = backup_job_create(backup->job_id, bs, target_bs, backup->speed,
2875 backup->sync, bmap, backup->bitmap_mode,
2876 backup->compress,
2877 backup->filter_node_name,
2878 &perf,
2879 backup->on_source_error,
2880 backup->on_target_error,
2881 job_flags, NULL, NULL, txn, errp);
2882 return job;
2883 }
2884
2885 void qmp_drive_backup(DriveBackup *backup, Error **errp)
2886 {
2887 TransactionAction action = {
2888 .type = TRANSACTION_ACTION_KIND_DRIVE_BACKUP,
2889 .u.drive_backup.data = backup,
2890 };
2891 blockdev_do_action(&action, errp);
2892 }
2893
2894 BlockDeviceInfoList *qmp_query_named_block_nodes(bool has_flat,
2895 bool flat,
2896 Error **errp)
2897 {
2898 bool return_flat = has_flat && flat;
2899
2900 return bdrv_named_nodes_list(return_flat, errp);
2901 }
2902
2903 XDbgBlockGraph *qmp_x_debug_query_block_graph(Error **errp)
2904 {
2905 return bdrv_get_xdbg_block_graph(errp);
2906 }
2907
2908 void qmp_blockdev_backup(BlockdevBackup *backup, Error **errp)
2909 {
2910 TransactionAction action = {
2911 .type = TRANSACTION_ACTION_KIND_BLOCKDEV_BACKUP,
2912 .u.blockdev_backup.data = backup,
2913 };
2914 blockdev_do_action(&action, errp);
2915 }
2916
2917 /* Parameter check and block job starting for drive mirroring.
2918 * Caller should hold @device and @target's aio context (must be the same).
2919 **/
2920 static void blockdev_mirror_common(const char *job_id, BlockDriverState *bs,
2921 BlockDriverState *target,
2922 const char *replaces,
2923 enum MirrorSyncMode sync,
2924 BlockMirrorBackingMode backing_mode,
2925 bool zero_target,
2926 bool has_speed, int64_t speed,
2927 bool has_granularity, uint32_t granularity,
2928 bool has_buf_size, int64_t buf_size,
2929 bool has_on_source_error,
2930 BlockdevOnError on_source_error,
2931 bool has_on_target_error,
2932 BlockdevOnError on_target_error,
2933 bool has_unmap, bool unmap,
2934 const char *filter_node_name,
2935 bool has_copy_mode, MirrorCopyMode copy_mode,
2936 bool has_auto_finalize, bool auto_finalize,
2937 bool has_auto_dismiss, bool auto_dismiss,
2938 Error **errp)
2939 {
2940 BlockDriverState *unfiltered_bs;
2941 int job_flags = JOB_DEFAULT;
2942
2943 if (!has_speed) {
2944 speed = 0;
2945 }
2946 if (!has_on_source_error) {
2947 on_source_error = BLOCKDEV_ON_ERROR_REPORT;
2948 }
2949 if (!has_on_target_error) {
2950 on_target_error = BLOCKDEV_ON_ERROR_REPORT;
2951 }
2952 if (!has_granularity) {
2953 granularity = 0;
2954 }
2955 if (!has_buf_size) {
2956 buf_size = 0;
2957 }
2958 if (!has_unmap) {
2959 unmap = true;
2960 }
2961 if (!has_copy_mode) {
2962 copy_mode = MIRROR_COPY_MODE_BACKGROUND;
2963 }
2964 if (has_auto_finalize && !auto_finalize) {
2965 job_flags |= JOB_MANUAL_FINALIZE;
2966 }
2967 if (has_auto_dismiss && !auto_dismiss) {
2968 job_flags |= JOB_MANUAL_DISMISS;
2969 }
2970
2971 if (granularity != 0 && (granularity < 512 || granularity > 1048576 * 64)) {
2972 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2973 "a value in range [512B, 64MB]");
2974 return;
2975 }
2976 if (granularity & (granularity - 1)) {
2977 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "granularity",
2978 "a power of 2");
2979 return;
2980 }
2981
2982 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
2983 return;
2984 }
2985 if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_MIRROR_TARGET, errp)) {
2986 return;
2987 }
2988
2989 if (!bdrv_backing_chain_next(bs) && sync == MIRROR_SYNC_MODE_TOP) {
2990 sync = MIRROR_SYNC_MODE_FULL;
2991 }
2992
2993 if (!replaces) {
2994 /* We want to mirror from @bs, but keep implicit filters on top */
2995 unfiltered_bs = bdrv_skip_implicit_filters(bs);
2996 if (unfiltered_bs != bs) {
2997 replaces = unfiltered_bs->node_name;
2998 }
2999 }
3000
3001 if (replaces) {
3002 BlockDriverState *to_replace_bs;
3003 AioContext *replace_aio_context;
3004 int64_t bs_size, replace_size;
3005
3006 bs_size = bdrv_getlength(bs);
3007 if (bs_size < 0) {
3008 error_setg_errno(errp, -bs_size, "Failed to query device's size");
3009 return;
3010 }
3011
3012 to_replace_bs = check_to_replace_node(bs, replaces, errp);
3013 if (!to_replace_bs) {
3014 return;
3015 }
3016
3017 replace_aio_context = bdrv_get_aio_context(to_replace_bs);
3018 aio_context_acquire(replace_aio_context);
3019 replace_size = bdrv_getlength(to_replace_bs);
3020 aio_context_release(replace_aio_context);
3021
3022 if (replace_size < 0) {
3023 error_setg_errno(errp, -replace_size,
3024 "Failed to query the replacement node's size");
3025 return;
3026 }
3027 if (bs_size != replace_size) {
3028 error_setg(errp, "cannot replace image with a mirror image of "
3029 "different size");
3030 return;
3031 }
3032 }
3033
3034 /* pass the node name to replace to mirror start since it's loose coupling
3035 * and will allow to check whether the node still exist at mirror completion
3036 */
3037 mirror_start(job_id, bs, target,
3038 replaces, job_flags,
3039 speed, granularity, buf_size, sync, backing_mode, zero_target,
3040 on_source_error, on_target_error, unmap, filter_node_name,
3041 copy_mode, errp);
3042 }
3043
3044 void qmp_drive_mirror(DriveMirror *arg, Error **errp)
3045 {
3046 BlockDriverState *bs;
3047 BlockDriverState *target_backing_bs, *target_bs;
3048 AioContext *aio_context;
3049 AioContext *old_context;
3050 BlockMirrorBackingMode backing_mode;
3051 Error *local_err = NULL;
3052 QDict *options = NULL;
3053 int flags;
3054 int64_t size;
3055 const char *format = arg->format;
3056 bool zero_target;
3057 int ret;
3058
3059 bs = qmp_get_root_bs(arg->device, errp);
3060 if (!bs) {
3061 return;
3062 }
3063
3064 /* Early check to avoid creating target */
3065 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_MIRROR_SOURCE, errp)) {
3066 return;
3067 }
3068
3069 aio_context = bdrv_get_aio_context(bs);
3070 aio_context_acquire(aio_context);
3071
3072 if (!arg->has_mode) {
3073 arg->mode = NEW_IMAGE_MODE_ABSOLUTE_PATHS;
3074 }
3075
3076 if (!arg->format) {
3077 format = (arg->mode == NEW_IMAGE_MODE_EXISTING
3078 ? NULL : bs->drv->format_name);
3079 }
3080
3081 flags = bs->open_flags | BDRV_O_RDWR;
3082 target_backing_bs = bdrv_cow_bs(bdrv_skip_filters(bs));
3083 if (!target_backing_bs && arg->sync == MIRROR_SYNC_MODE_TOP) {
3084 arg->sync = MIRROR_SYNC_MODE_FULL;
3085 }
3086 if (arg->sync == MIRROR_SYNC_MODE_NONE) {
3087 target_backing_bs = bs;
3088 }
3089
3090 size = bdrv_getlength(bs);
3091 if (size < 0) {
3092 error_setg_errno(errp, -size, "bdrv_getlength failed");
3093 goto out;
3094 }
3095
3096 if (arg->replaces) {
3097 if (!arg->node_name) {
3098 error_setg(errp, "a node-name must be provided when replacing a"
3099 " named node of the graph");
3100 goto out;
3101 }
3102 }
3103
3104 if (arg->mode == NEW_IMAGE_MODE_ABSOLUTE_PATHS) {
3105 backing_mode = MIRROR_SOURCE_BACKING_CHAIN;
3106 } else {
3107 backing_mode = MIRROR_OPEN_BACKING_CHAIN;
3108 }
3109
3110 /* Don't open backing image in create() */
3111 flags |= BDRV_O_NO_BACKING;
3112
3113 if ((arg->sync == MIRROR_SYNC_MODE_FULL || !target_backing_bs)
3114 && arg->mode != NEW_IMAGE_MODE_EXISTING)
3115 {
3116 /* create new image w/o backing file */
3117 assert(format);
3118 bdrv_img_create(arg->target, format,
3119 NULL, NULL, NULL, size, flags, false, &local_err);
3120 } else {
3121 /* Implicit filters should not appear in the filename */
3122 BlockDriverState *explicit_backing =
3123 bdrv_skip_implicit_filters(target_backing_bs);
3124
3125 switch (arg->mode) {
3126 case NEW_IMAGE_MODE_EXISTING:
3127 break;
3128 case NEW_IMAGE_MODE_ABSOLUTE_PATHS:
3129 /* create new image with backing file */
3130 bdrv_refresh_filename(explicit_backing);
3131 bdrv_img_create(arg->target, format,
3132 explicit_backing->filename,
3133 explicit_backing->drv->format_name,
3134 NULL, size, flags, false, &local_err);
3135 break;
3136 default:
3137 abort();
3138 }
3139 }
3140
3141 if (local_err) {
3142 error_propagate(errp, local_err);
3143 goto out;
3144 }
3145
3146 options = qdict_new();
3147 if (arg->node_name) {
3148 qdict_put_str(options, "node-name", arg->node_name);
3149 }
3150 if (format) {
3151 qdict_put_str(options, "driver", format);
3152 }
3153
3154 /* Mirroring takes care of copy-on-write using the source's backing
3155 * file.
3156 */
3157 target_bs = bdrv_open(arg->target, NULL, options, flags, errp);
3158 if (!target_bs) {
3159 goto out;
3160 }
3161
3162 zero_target = (arg->sync == MIRROR_SYNC_MODE_FULL &&
3163 (arg->mode == NEW_IMAGE_MODE_EXISTING ||
3164 !bdrv_has_zero_init(target_bs)));
3165
3166
3167 /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3168 old_context = bdrv_get_aio_context(target_bs);
3169 aio_context_release(aio_context);
3170 aio_context_acquire(old_context);
3171
3172 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3173 if (ret < 0) {
3174 bdrv_unref(target_bs);
3175 aio_context_release(old_context);
3176 return;
3177 }
3178
3179 aio_context_release(old_context);
3180 aio_context_acquire(aio_context);
3181
3182 blockdev_mirror_common(arg->job_id, bs, target_bs,
3183 arg->replaces, arg->sync,
3184 backing_mode, zero_target,
3185 arg->has_speed, arg->speed,
3186 arg->has_granularity, arg->granularity,
3187 arg->has_buf_size, arg->buf_size,
3188 arg->has_on_source_error, arg->on_source_error,
3189 arg->has_on_target_error, arg->on_target_error,
3190 arg->has_unmap, arg->unmap,
3191 NULL,
3192 arg->has_copy_mode, arg->copy_mode,
3193 arg->has_auto_finalize, arg->auto_finalize,
3194 arg->has_auto_dismiss, arg->auto_dismiss,
3195 errp);
3196 bdrv_unref(target_bs);
3197 out:
3198 aio_context_release(aio_context);
3199 }
3200
3201 void qmp_blockdev_mirror(const char *job_id,
3202 const char *device, const char *target,
3203 const char *replaces,
3204 MirrorSyncMode sync,
3205 bool has_speed, int64_t speed,
3206 bool has_granularity, uint32_t granularity,
3207 bool has_buf_size, int64_t buf_size,
3208 bool has_on_source_error,
3209 BlockdevOnError on_source_error,
3210 bool has_on_target_error,
3211 BlockdevOnError on_target_error,
3212 const char *filter_node_name,
3213 bool has_copy_mode, MirrorCopyMode copy_mode,
3214 bool has_auto_finalize, bool auto_finalize,
3215 bool has_auto_dismiss, bool auto_dismiss,
3216 Error **errp)
3217 {
3218 BlockDriverState *bs;
3219 BlockDriverState *target_bs;
3220 AioContext *aio_context;
3221 AioContext *old_context;
3222 BlockMirrorBackingMode backing_mode = MIRROR_LEAVE_BACKING_CHAIN;
3223 bool zero_target;
3224 int ret;
3225
3226 bs = qmp_get_root_bs(device, errp);
3227 if (!bs) {
3228 return;
3229 }
3230
3231 target_bs = bdrv_lookup_bs(target, target, errp);
3232 if (!target_bs) {
3233 return;
3234 }
3235
3236 zero_target = (sync == MIRROR_SYNC_MODE_FULL);
3237
3238 /* Honor bdrv_try_change_aio_context() context acquisition requirements. */
3239 old_context = bdrv_get_aio_context(target_bs);
3240 aio_context = bdrv_get_aio_context(bs);
3241 aio_context_acquire(old_context);
3242
3243 ret = bdrv_try_change_aio_context(target_bs, aio_context, NULL, errp);
3244
3245 aio_context_release(old_context);
3246 aio_context_acquire(aio_context);
3247
3248 if (ret < 0) {
3249 goto out;
3250 }
3251
3252 blockdev_mirror_common(job_id, bs, target_bs,
3253 replaces, sync, backing_mode,
3254 zero_target, has_speed, speed,
3255 has_granularity, granularity,
3256 has_buf_size, buf_size,
3257 has_on_source_error, on_source_error,
3258 has_on_target_error, on_target_error,
3259 true, true, filter_node_name,
3260 has_copy_mode, copy_mode,
3261 has_auto_finalize, auto_finalize,
3262 has_auto_dismiss, auto_dismiss,
3263 errp);
3264 out:
3265 aio_context_release(aio_context);
3266 }
3267
3268 /*
3269 * Get a block job using its ID. Called with job_mutex held.
3270 */
3271 static BlockJob *find_block_job_locked(const char *id, Error **errp)
3272 {
3273 BlockJob *job;
3274
3275 assert(id != NULL);
3276
3277 job = block_job_get_locked(id);
3278
3279 if (!job) {
3280 error_set(errp, ERROR_CLASS_DEVICE_NOT_ACTIVE,
3281 "Block job '%s' not found", id);
3282 return NULL;
3283 }
3284
3285 return job;
3286 }
3287
3288 void qmp_block_job_set_speed(const char *device, int64_t speed, Error **errp)
3289 {
3290 BlockJob *job;
3291
3292 JOB_LOCK_GUARD();
3293 job = find_block_job_locked(device, errp);
3294
3295 if (!job) {
3296 return;
3297 }
3298
3299 block_job_set_speed_locked(job, speed, errp);
3300 }
3301
3302 void qmp_block_job_cancel(const char *device,
3303 bool has_force, bool force, Error **errp)
3304 {
3305 BlockJob *job;
3306
3307 JOB_LOCK_GUARD();
3308 job = find_block_job_locked(device, errp);
3309
3310 if (!job) {
3311 return;
3312 }
3313
3314 if (!has_force) {
3315 force = false;
3316 }
3317
3318 if (job_user_paused_locked(&job->job) && !force) {
3319 error_setg(errp, "The block job for device '%s' is currently paused",
3320 device);
3321 return;
3322 }
3323
3324 trace_qmp_block_job_cancel(job);
3325 job_user_cancel_locked(&job->job, force, errp);
3326 }
3327
3328 void qmp_block_job_pause(const char *device, Error **errp)
3329 {
3330 BlockJob *job;
3331
3332 JOB_LOCK_GUARD();
3333 job = find_block_job_locked(device, errp);
3334
3335 if (!job) {
3336 return;
3337 }
3338
3339 trace_qmp_block_job_pause(job);
3340 job_user_pause_locked(&job->job, errp);
3341 }
3342
3343 void qmp_block_job_resume(const char *device, Error **errp)
3344 {
3345 BlockJob *job;
3346
3347 JOB_LOCK_GUARD();
3348 job = find_block_job_locked(device, errp);
3349
3350 if (!job) {
3351 return;
3352 }
3353
3354 trace_qmp_block_job_resume(job);
3355 job_user_resume_locked(&job->job, errp);
3356 }
3357
3358 void qmp_block_job_complete(const char *device, Error **errp)
3359 {
3360 BlockJob *job;
3361
3362 JOB_LOCK_GUARD();
3363 job = find_block_job_locked(device, errp);
3364
3365 if (!job) {
3366 return;
3367 }
3368
3369 trace_qmp_block_job_complete(job);
3370 job_complete_locked(&job->job, errp);
3371 }
3372
3373 void qmp_block_job_finalize(const char *id, Error **errp)
3374 {
3375 BlockJob *job;
3376
3377 JOB_LOCK_GUARD();
3378 job = find_block_job_locked(id, errp);
3379
3380 if (!job) {
3381 return;
3382 }
3383
3384 trace_qmp_block_job_finalize(job);
3385 job_ref_locked(&job->job);
3386 job_finalize_locked(&job->job, errp);
3387
3388 job_unref_locked(&job->job);
3389 }
3390
3391 void qmp_block_job_dismiss(const char *id, Error **errp)
3392 {
3393 BlockJob *bjob;
3394 Job *job;
3395
3396 JOB_LOCK_GUARD();
3397 bjob = find_block_job_locked(id, errp);
3398
3399 if (!bjob) {
3400 return;
3401 }
3402
3403 trace_qmp_block_job_dismiss(bjob);
3404 job = &bjob->job;
3405 job_dismiss_locked(&job, errp);
3406 }
3407
3408 void qmp_change_backing_file(const char *device,
3409 const char *image_node_name,
3410 const char *backing_file,
3411 Error **errp)
3412 {
3413 BlockDriverState *bs = NULL;
3414 AioContext *aio_context;
3415 BlockDriverState *image_bs = NULL;
3416 Error *local_err = NULL;
3417 bool ro;
3418 int ret;
3419
3420 bs = qmp_get_root_bs(device, errp);
3421 if (!bs) {
3422 return;
3423 }
3424
3425 aio_context = bdrv_get_aio_context(bs);
3426 aio_context_acquire(aio_context);
3427
3428 image_bs = bdrv_lookup_bs(NULL, image_node_name, &local_err);
3429 if (local_err) {
3430 error_propagate(errp, local_err);
3431 goto out;
3432 }
3433
3434 if (!image_bs) {
3435 error_setg(errp, "image file not found");
3436 goto out;
3437 }
3438
3439 if (bdrv_find_base(image_bs) == image_bs) {
3440 error_setg(errp, "not allowing backing file change on an image "
3441 "without a backing file");
3442 goto out;
3443 }
3444
3445 /* even though we are not necessarily operating on bs, we need it to
3446 * determine if block ops are currently prohibited on the chain */
3447 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_CHANGE, errp)) {
3448 goto out;
3449 }
3450
3451 /* final sanity check */
3452 if (!bdrv_chain_contains(bs, image_bs)) {
3453 error_setg(errp, "'%s' and image file are not in the same chain",
3454 device);
3455 goto out;
3456 }
3457
3458 /* if not r/w, reopen to make r/w */
3459 ro = bdrv_is_read_only(image_bs);
3460
3461 if (ro) {
3462 if (bdrv_reopen_set_read_only(image_bs, false, errp) != 0) {
3463 goto out;
3464 }
3465 }
3466
3467 ret = bdrv_change_backing_file(image_bs, backing_file,
3468 image_bs->drv ? image_bs->drv->format_name : "",
3469 false);
3470
3471 if (ret < 0) {
3472 error_setg_errno(errp, -ret, "Could not change backing file to '%s'",
3473 backing_file);
3474 /* don't exit here, so we can try to restore open flags if
3475 * appropriate */
3476 }
3477
3478 if (ro) {
3479 bdrv_reopen_set_read_only(image_bs, true, errp);
3480 }
3481
3482 out:
3483 aio_context_release(aio_context);
3484 }
3485
3486 void qmp_blockdev_add(BlockdevOptions *options, Error **errp)
3487 {
3488 BlockDriverState *bs;
3489 QObject *obj;
3490 Visitor *v = qobject_output_visitor_new(&obj);
3491 QDict *qdict;
3492
3493 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3494 visit_complete(v, &obj);
3495 qdict = qobject_to(QDict, obj);
3496
3497 qdict_flatten(qdict);
3498
3499 if (!qdict_get_try_str(qdict, "node-name")) {
3500 error_setg(errp, "'node-name' must be specified for the root node");
3501 goto fail;
3502 }
3503
3504 bs = bds_tree_init(qdict, errp);
3505 if (!bs) {
3506 goto fail;
3507 }
3508
3509 bdrv_set_monitor_owned(bs);
3510
3511 fail:
3512 visit_free(v);
3513 }
3514
3515 void qmp_blockdev_reopen(BlockdevOptionsList *reopen_list, Error **errp)
3516 {
3517 BlockReopenQueue *queue = NULL;
3518 GSList *drained = NULL;
3519 GSList *p;
3520
3521 /* Add each one of the BDS that we want to reopen to the queue */
3522 for (; reopen_list != NULL; reopen_list = reopen_list->next) {
3523 BlockdevOptions *options = reopen_list->value;
3524 BlockDriverState *bs;
3525 AioContext *ctx;
3526 QObject *obj;
3527 Visitor *v;
3528 QDict *qdict;
3529
3530 /* Check for the selected node name */
3531 if (!options->node_name) {
3532 error_setg(errp, "node-name not specified");
3533 goto fail;
3534 }
3535
3536 bs = bdrv_find_node(options->node_name);
3537 if (!bs) {
3538 error_setg(errp, "Failed to find node with node-name='%s'",
3539 options->node_name);
3540 goto fail;
3541 }
3542
3543 /* Put all options in a QDict and flatten it */
3544 v = qobject_output_visitor_new(&obj);
3545 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3546 visit_complete(v, &obj);
3547 visit_free(v);
3548
3549 qdict = qobject_to(QDict, obj);
3550
3551 qdict_flatten(qdict);
3552
3553 ctx = bdrv_get_aio_context(bs);
3554 aio_context_acquire(ctx);
3555
3556 bdrv_subtree_drained_begin(bs);
3557 queue = bdrv_reopen_queue(queue, bs, qdict, false);
3558 drained = g_slist_prepend(drained, bs);
3559
3560 aio_context_release(ctx);
3561 }
3562
3563 /* Perform the reopen operation */
3564 bdrv_reopen_multiple(queue, errp);
3565 queue = NULL;
3566
3567 fail:
3568 bdrv_reopen_queue_free(queue);
3569 for (p = drained; p; p = p->next) {
3570 BlockDriverState *bs = p->data;
3571 AioContext *ctx = bdrv_get_aio_context(bs);
3572
3573 aio_context_acquire(ctx);
3574 bdrv_subtree_drained_end(bs);
3575 aio_context_release(ctx);
3576 }
3577 g_slist_free(drained);
3578 }
3579
3580 void qmp_blockdev_del(const char *node_name, Error **errp)
3581 {
3582 AioContext *aio_context;
3583 BlockDriverState *bs;
3584
3585 GLOBAL_STATE_CODE();
3586
3587 bs = bdrv_find_node(node_name);
3588 if (!bs) {
3589 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3590 return;
3591 }
3592 if (bdrv_has_blk(bs)) {
3593 error_setg(errp, "Node %s is in use", node_name);
3594 return;
3595 }
3596 aio_context = bdrv_get_aio_context(bs);
3597 aio_context_acquire(aio_context);
3598
3599 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_DRIVE_DEL, errp)) {
3600 goto out;
3601 }
3602
3603 if (!QTAILQ_IN_USE(bs, monitor_list)) {
3604 error_setg(errp, "Node %s is not owned by the monitor",
3605 bs->node_name);
3606 goto out;
3607 }
3608
3609 if (bs->refcnt > 1) {
3610 error_setg(errp, "Block device %s is in use",
3611 bdrv_get_device_or_node_name(bs));
3612 goto out;
3613 }
3614
3615 QTAILQ_REMOVE(&monitor_bdrv_states, bs, monitor_list);
3616 bdrv_unref(bs);
3617
3618 out:
3619 aio_context_release(aio_context);
3620 }
3621
3622 static BdrvChild *bdrv_find_child(BlockDriverState *parent_bs,
3623 const char *child_name)
3624 {
3625 BdrvChild *child;
3626
3627 QLIST_FOREACH(child, &parent_bs->children, next) {
3628 if (strcmp(child->name, child_name) == 0) {
3629 return child;
3630 }
3631 }
3632
3633 return NULL;
3634 }
3635
3636 void qmp_x_blockdev_change(const char *parent, const char *child,
3637 const char *node, Error **errp)
3638 {
3639 BlockDriverState *parent_bs, *new_bs = NULL;
3640 BdrvChild *p_child;
3641
3642 parent_bs = bdrv_lookup_bs(parent, parent, errp);
3643 if (!parent_bs) {
3644 return;
3645 }
3646
3647 if (!child == !node) {
3648 if (child) {
3649 error_setg(errp, "The parameters child and node are in conflict");
3650 } else {
3651 error_setg(errp, "Either child or node must be specified");
3652 }
3653 return;
3654 }
3655
3656 if (child) {
3657 p_child = bdrv_find_child(parent_bs, child);
3658 if (!p_child) {
3659 error_setg(errp, "Node '%s' does not have child '%s'",
3660 parent, child);
3661 return;
3662 }
3663 bdrv_del_child(parent_bs, p_child, errp);
3664 }
3665
3666 if (node) {
3667 new_bs = bdrv_find_node(node);
3668 if (!new_bs) {
3669 error_setg(errp, "Node '%s' not found", node);
3670 return;
3671 }
3672 bdrv_add_child(parent_bs, new_bs, errp);
3673 }
3674 }
3675
3676 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
3677 {
3678 BlockJobInfoList *head = NULL, **tail = &head;
3679 BlockJob *job;
3680
3681 JOB_LOCK_GUARD();
3682
3683 for (job = block_job_next_locked(NULL); job;
3684 job = block_job_next_locked(job)) {
3685 BlockJobInfo *value;
3686
3687 if (block_job_is_internal(job)) {
3688 continue;
3689 }
3690 value = block_job_query_locked(job, errp);
3691 if (!value) {
3692 qapi_free_BlockJobInfoList(head);
3693 return NULL;
3694 }
3695 QAPI_LIST_APPEND(tail, value);
3696 }
3697
3698 return head;
3699 }
3700
3701 void qmp_x_blockdev_set_iothread(const char *node_name, StrOrNull *iothread,
3702 bool has_force, bool force, Error **errp)
3703 {
3704 AioContext *old_context;
3705 AioContext *new_context;
3706 BlockDriverState *bs;
3707
3708 bs = bdrv_find_node(node_name);
3709 if (!bs) {
3710 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
3711 return;
3712 }
3713
3714 /* Protects against accidents. */
3715 if (!(has_force && force) && bdrv_has_blk(bs)) {
3716 error_setg(errp, "Node %s is associated with a BlockBackend and could "
3717 "be in use (use force=true to override this check)",
3718 node_name);
3719 return;
3720 }
3721
3722 if (iothread->type == QTYPE_QSTRING) {
3723 IOThread *obj = iothread_by_id(iothread->u.s);
3724 if (!obj) {
3725 error_setg(errp, "Cannot find iothread %s", iothread->u.s);
3726 return;
3727 }
3728
3729 new_context = iothread_get_aio_context(obj);
3730 } else {
3731 new_context = qemu_get_aio_context();
3732 }
3733
3734 old_context = bdrv_get_aio_context(bs);
3735 aio_context_acquire(old_context);
3736
3737 bdrv_try_change_aio_context(bs, new_context, NULL, errp);
3738
3739 aio_context_release(old_context);
3740 }
3741
3742 QemuOptsList qemu_common_drive_opts = {
3743 .name = "drive",
3744 .head = QTAILQ_HEAD_INITIALIZER(qemu_common_drive_opts.head),
3745 .desc = {
3746 {
3747 .name = "snapshot",
3748 .type = QEMU_OPT_BOOL,
3749 .help = "enable/disable snapshot mode",
3750 },{
3751 .name = "aio",
3752 .type = QEMU_OPT_STRING,
3753 .help = "host AIO implementation (threads, native, io_uring)",
3754 },{
3755 .name = BDRV_OPT_CACHE_WB,
3756 .type = QEMU_OPT_BOOL,
3757 .help = "Enable writeback mode",
3758 },{
3759 .name = "format",
3760 .type = QEMU_OPT_STRING,
3761 .help = "disk format (raw, qcow2, ...)",
3762 },{
3763 .name = "rerror",
3764 .type = QEMU_OPT_STRING,
3765 .help = "read error action",
3766 },{
3767 .name = "werror",
3768 .type = QEMU_OPT_STRING,
3769 .help = "write error action",
3770 },{
3771 .name = BDRV_OPT_READ_ONLY,
3772 .type = QEMU_OPT_BOOL,
3773 .help = "open drive file as read-only",
3774 },
3775
3776 THROTTLE_OPTS,
3777
3778 {
3779 .name = "throttling.group",
3780 .type = QEMU_OPT_STRING,
3781 .help = "name of the block throttling group",
3782 },{
3783 .name = "copy-on-read",
3784 .type = QEMU_OPT_BOOL,
3785 .help = "copy read data from backing file into image file",
3786 },{
3787 .name = "detect-zeroes",
3788 .type = QEMU_OPT_STRING,
3789 .help = "try to optimize zero writes (off, on, unmap)",
3790 },{
3791 .name = "stats-account-invalid",
3792 .type = QEMU_OPT_BOOL,
3793 .help = "whether to account for invalid I/O operations "
3794 "in the statistics",
3795 },{
3796 .name = "stats-account-failed",
3797 .type = QEMU_OPT_BOOL,
3798 .help = "whether to account for failed I/O operations "
3799 "in the statistics",
3800 },
3801 { /* end of list */ }
3802 },
3803 };
3804
3805 QemuOptsList qemu_drive_opts = {
3806 .name = "drive",
3807 .head = QTAILQ_HEAD_INITIALIZER(qemu_drive_opts.head),
3808 .desc = {
3809 /*
3810 * no elements => accept any params
3811 * validation will happen later
3812 */
3813 { /* end of list */ }
3814 },
3815 };
armbru's QEMU hackery