]> git.proxmox.com Git - mirror_qemu.git/blob - block.c
qapi: Reformat recent doc comments to conform to current conventions
[mirror_qemu.git] / block.c
1 /*
2 * QEMU System Emulator block driver
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2020 Virtuozzo International GmbH.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23 * THE SOFTWARE.
24 */
25
26 #include "qemu/osdep.h"
27 #include "block/trace.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "block/dirty-bitmap.h"
31 #include "block/fuse.h"
32 #include "block/nbd.h"
33 #include "block/qdict.h"
34 #include "qemu/error-report.h"
35 #include "block/module_block.h"
36 #include "qemu/main-loop.h"
37 #include "qemu/module.h"
38 #include "qapi/error.h"
39 #include "qapi/qmp/qdict.h"
40 #include "qapi/qmp/qjson.h"
41 #include "qapi/qmp/qnull.h"
42 #include "qapi/qmp/qstring.h"
43 #include "qapi/qobject-output-visitor.h"
44 #include "qapi/qapi-visit-block-core.h"
45 #include "sysemu/block-backend.h"
46 #include "qemu/notify.h"
47 #include "qemu/option.h"
48 #include "qemu/coroutine.h"
49 #include "block/qapi.h"
50 #include "qemu/timer.h"
51 #include "qemu/cutils.h"
52 #include "qemu/id.h"
53 #include "qemu/range.h"
54 #include "qemu/rcu.h"
55 #include "block/coroutines.h"
56
57 #ifdef CONFIG_BSD
58 #include <sys/ioctl.h>
59 #include <sys/queue.h>
60 #if defined(HAVE_SYS_DISK_H)
61 #include <sys/disk.h>
62 #endif
63 #endif
64
65 #ifdef _WIN32
66 #include <windows.h>
67 #endif
68
69 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
70
71 /* Protected by BQL */
72 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
73 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
74
75 /* Protected by BQL */
76 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
77 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
78
79 /* Protected by BQL */
80 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
81 QLIST_HEAD_INITIALIZER(bdrv_drivers);
82
83 static BlockDriverState *bdrv_open_inherit(const char *filename,
84 const char *reference,
85 QDict *options, int flags,
86 BlockDriverState *parent,
87 const BdrvChildClass *child_class,
88 BdrvChildRole child_role,
89 Error **errp);
90
91 static bool bdrv_recurse_has_child(BlockDriverState *bs,
92 BlockDriverState *child);
93
94 static void bdrv_replace_child_noperm(BdrvChild *child,
95 BlockDriverState *new_bs);
96 static void bdrv_remove_child(BdrvChild *child, Transaction *tran);
97
98 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
99 BlockReopenQueue *queue,
100 Transaction *change_child_tran, Error **errp);
101 static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
102 static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
103
104 static bool bdrv_backing_overridden(BlockDriverState *bs);
105
106 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
107 GHashTable *visited, Transaction *tran,
108 Error **errp);
109
110 /* If non-zero, use only whitelisted block drivers */
111 static int use_bdrv_whitelist;
112
113 #ifdef _WIN32
114 static int is_windows_drive_prefix(const char *filename)
115 {
116 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
117 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
118 filename[1] == ':');
119 }
120
121 int is_windows_drive(const char *filename)
122 {
123 if (is_windows_drive_prefix(filename) &&
124 filename[2] == '\0')
125 return 1;
126 if (strstart(filename, "\\\\.\\", NULL) ||
127 strstart(filename, "//./", NULL))
128 return 1;
129 return 0;
130 }
131 #endif
132
133 size_t bdrv_opt_mem_align(BlockDriverState *bs)
134 {
135 if (!bs || !bs->drv) {
136 /* page size or 4k (hdd sector size) should be on the safe side */
137 return MAX(4096, qemu_real_host_page_size());
138 }
139 IO_CODE();
140
141 return bs->bl.opt_mem_alignment;
142 }
143
144 size_t bdrv_min_mem_align(BlockDriverState *bs)
145 {
146 if (!bs || !bs->drv) {
147 /* page size or 4k (hdd sector size) should be on the safe side */
148 return MAX(4096, qemu_real_host_page_size());
149 }
150 IO_CODE();
151
152 return bs->bl.min_mem_alignment;
153 }
154
155 /* check if the path starts with "<protocol>:" */
156 int path_has_protocol(const char *path)
157 {
158 const char *p;
159
160 #ifdef _WIN32
161 if (is_windows_drive(path) ||
162 is_windows_drive_prefix(path)) {
163 return 0;
164 }
165 p = path + strcspn(path, ":/\\");
166 #else
167 p = path + strcspn(path, ":/");
168 #endif
169
170 return *p == ':';
171 }
172
173 int path_is_absolute(const char *path)
174 {
175 #ifdef _WIN32
176 /* specific case for names like: "\\.\d:" */
177 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
178 return 1;
179 }
180 return (*path == '/' || *path == '\\');
181 #else
182 return (*path == '/');
183 #endif
184 }
185
186 /* if filename is absolute, just return its duplicate. Otherwise, build a
187 path to it by considering it is relative to base_path. URL are
188 supported. */
189 char *path_combine(const char *base_path, const char *filename)
190 {
191 const char *protocol_stripped = NULL;
192 const char *p, *p1;
193 char *result;
194 int len;
195
196 if (path_is_absolute(filename)) {
197 return g_strdup(filename);
198 }
199
200 if (path_has_protocol(base_path)) {
201 protocol_stripped = strchr(base_path, ':');
202 if (protocol_stripped) {
203 protocol_stripped++;
204 }
205 }
206 p = protocol_stripped ?: base_path;
207
208 p1 = strrchr(base_path, '/');
209 #ifdef _WIN32
210 {
211 const char *p2;
212 p2 = strrchr(base_path, '\\');
213 if (!p1 || p2 > p1) {
214 p1 = p2;
215 }
216 }
217 #endif
218 if (p1) {
219 p1++;
220 } else {
221 p1 = base_path;
222 }
223 if (p1 > p) {
224 p = p1;
225 }
226 len = p - base_path;
227
228 result = g_malloc(len + strlen(filename) + 1);
229 memcpy(result, base_path, len);
230 strcpy(result + len, filename);
231
232 return result;
233 }
234
235 /*
236 * Helper function for bdrv_parse_filename() implementations to remove optional
237 * protocol prefixes (especially "file:") from a filename and for putting the
238 * stripped filename into the options QDict if there is such a prefix.
239 */
240 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
241 QDict *options)
242 {
243 if (strstart(filename, prefix, &filename)) {
244 /* Stripping the explicit protocol prefix may result in a protocol
245 * prefix being (wrongly) detected (if the filename contains a colon) */
246 if (path_has_protocol(filename)) {
247 GString *fat_filename;
248
249 /* This means there is some colon before the first slash; therefore,
250 * this cannot be an absolute path */
251 assert(!path_is_absolute(filename));
252
253 /* And we can thus fix the protocol detection issue by prefixing it
254 * by "./" */
255 fat_filename = g_string_new("./");
256 g_string_append(fat_filename, filename);
257
258 assert(!path_has_protocol(fat_filename->str));
259
260 qdict_put(options, "filename",
261 qstring_from_gstring(fat_filename));
262 } else {
263 /* If no protocol prefix was detected, we can use the shortened
264 * filename as-is */
265 qdict_put_str(options, "filename", filename);
266 }
267 }
268 }
269
270
271 /* Returns whether the image file is opened as read-only. Note that this can
272 * return false and writing to the image file is still not possible because the
273 * image is inactivated. */
274 bool bdrv_is_read_only(BlockDriverState *bs)
275 {
276 IO_CODE();
277 return !(bs->open_flags & BDRV_O_RDWR);
278 }
279
280 static int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
281 bool ignore_allow_rdw, Error **errp)
282 {
283 IO_CODE();
284
285 /* Do not set read_only if copy_on_read is enabled */
286 if (bs->copy_on_read && read_only) {
287 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
288 bdrv_get_device_or_node_name(bs));
289 return -EINVAL;
290 }
291
292 /* Do not clear read_only if it is prohibited */
293 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
294 !ignore_allow_rdw)
295 {
296 error_setg(errp, "Node '%s' is read only",
297 bdrv_get_device_or_node_name(bs));
298 return -EPERM;
299 }
300
301 return 0;
302 }
303
304 /*
305 * Called by a driver that can only provide a read-only image.
306 *
307 * Returns 0 if the node is already read-only or it could switch the node to
308 * read-only because BDRV_O_AUTO_RDONLY is set.
309 *
310 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
311 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
312 * is not NULL, it is used as the error message for the Error object.
313 */
314 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
315 Error **errp)
316 {
317 int ret = 0;
318 IO_CODE();
319
320 if (!(bs->open_flags & BDRV_O_RDWR)) {
321 return 0;
322 }
323 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
324 goto fail;
325 }
326
327 ret = bdrv_can_set_read_only(bs, true, false, NULL);
328 if (ret < 0) {
329 goto fail;
330 }
331
332 bs->open_flags &= ~BDRV_O_RDWR;
333
334 return 0;
335
336 fail:
337 error_setg(errp, "%s", errmsg ?: "Image is read-only");
338 return -EACCES;
339 }
340
341 /*
342 * If @backing is empty, this function returns NULL without setting
343 * @errp. In all other cases, NULL will only be returned with @errp
344 * set.
345 *
346 * Therefore, a return value of NULL without @errp set means that
347 * there is no backing file; if @errp is set, there is one but its
348 * absolute filename cannot be generated.
349 */
350 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
351 const char *backing,
352 Error **errp)
353 {
354 if (backing[0] == '\0') {
355 return NULL;
356 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
357 return g_strdup(backing);
358 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
359 error_setg(errp, "Cannot use relative backing file names for '%s'",
360 backed);
361 return NULL;
362 } else {
363 return path_combine(backed, backing);
364 }
365 }
366
367 /*
368 * If @filename is empty or NULL, this function returns NULL without
369 * setting @errp. In all other cases, NULL will only be returned with
370 * @errp set.
371 */
372 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
373 const char *filename, Error **errp)
374 {
375 char *dir, *full_name;
376
377 if (!filename || filename[0] == '\0') {
378 return NULL;
379 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
380 return g_strdup(filename);
381 }
382
383 dir = bdrv_dirname(relative_to, errp);
384 if (!dir) {
385 return NULL;
386 }
387
388 full_name = g_strconcat(dir, filename, NULL);
389 g_free(dir);
390 return full_name;
391 }
392
393 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
394 {
395 GLOBAL_STATE_CODE();
396 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
397 }
398
399 void bdrv_register(BlockDriver *bdrv)
400 {
401 assert(bdrv->format_name);
402 GLOBAL_STATE_CODE();
403 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
404 }
405
406 BlockDriverState *bdrv_new(void)
407 {
408 BlockDriverState *bs;
409 int i;
410
411 GLOBAL_STATE_CODE();
412
413 bs = g_new0(BlockDriverState, 1);
414 QLIST_INIT(&bs->dirty_bitmaps);
415 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
416 QLIST_INIT(&bs->op_blockers[i]);
417 }
418 qemu_co_mutex_init(&bs->reqs_lock);
419 qemu_mutex_init(&bs->dirty_bitmap_mutex);
420 bs->refcnt = 1;
421 bs->aio_context = qemu_get_aio_context();
422
423 qemu_co_queue_init(&bs->flush_queue);
424
425 qemu_co_mutex_init(&bs->bsc_modify_lock);
426 bs->block_status_cache = g_new0(BdrvBlockStatusCache, 1);
427
428 for (i = 0; i < bdrv_drain_all_count; i++) {
429 bdrv_drained_begin(bs);
430 }
431
432 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
433
434 return bs;
435 }
436
437 static BlockDriver *bdrv_do_find_format(const char *format_name)
438 {
439 BlockDriver *drv1;
440 GLOBAL_STATE_CODE();
441
442 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
443 if (!strcmp(drv1->format_name, format_name)) {
444 return drv1;
445 }
446 }
447
448 return NULL;
449 }
450
451 BlockDriver *bdrv_find_format(const char *format_name)
452 {
453 BlockDriver *drv1;
454 int i;
455
456 GLOBAL_STATE_CODE();
457
458 drv1 = bdrv_do_find_format(format_name);
459 if (drv1) {
460 return drv1;
461 }
462
463 /* The driver isn't registered, maybe we need to load a module */
464 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
465 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
466 Error *local_err = NULL;
467 int rv = block_module_load(block_driver_modules[i].library_name,
468 &local_err);
469 if (rv > 0) {
470 return bdrv_do_find_format(format_name);
471 } else if (rv < 0) {
472 error_report_err(local_err);
473 }
474 break;
475 }
476 }
477 return NULL;
478 }
479
480 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
481 {
482 static const char *whitelist_rw[] = {
483 CONFIG_BDRV_RW_WHITELIST
484 NULL
485 };
486 static const char *whitelist_ro[] = {
487 CONFIG_BDRV_RO_WHITELIST
488 NULL
489 };
490 const char **p;
491
492 if (!whitelist_rw[0] && !whitelist_ro[0]) {
493 return 1; /* no whitelist, anything goes */
494 }
495
496 for (p = whitelist_rw; *p; p++) {
497 if (!strcmp(format_name, *p)) {
498 return 1;
499 }
500 }
501 if (read_only) {
502 for (p = whitelist_ro; *p; p++) {
503 if (!strcmp(format_name, *p)) {
504 return 1;
505 }
506 }
507 }
508 return 0;
509 }
510
511 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
512 {
513 GLOBAL_STATE_CODE();
514 return bdrv_format_is_whitelisted(drv->format_name, read_only);
515 }
516
517 bool bdrv_uses_whitelist(void)
518 {
519 return use_bdrv_whitelist;
520 }
521
522 typedef struct CreateCo {
523 BlockDriver *drv;
524 char *filename;
525 QemuOpts *opts;
526 int ret;
527 Error *err;
528 } CreateCo;
529
530 int coroutine_fn bdrv_co_create(BlockDriver *drv, const char *filename,
531 QemuOpts *opts, Error **errp)
532 {
533 int ret;
534 GLOBAL_STATE_CODE();
535 ERRP_GUARD();
536
537 if (!drv->bdrv_co_create_opts) {
538 error_setg(errp, "Driver '%s' does not support image creation",
539 drv->format_name);
540 return -ENOTSUP;
541 }
542
543 ret = drv->bdrv_co_create_opts(drv, filename, opts, errp);
544 if (ret < 0 && !*errp) {
545 error_setg_errno(errp, -ret, "Could not create image");
546 }
547
548 return ret;
549 }
550
551 /**
552 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
553 * least the given @minimum_size.
554 *
555 * On success, return @blk's actual length.
556 * Otherwise, return -errno.
557 */
558 static int64_t coroutine_fn GRAPH_UNLOCKED
559 create_file_fallback_truncate(BlockBackend *blk, int64_t minimum_size,
560 Error **errp)
561 {
562 Error *local_err = NULL;
563 int64_t size;
564 int ret;
565
566 GLOBAL_STATE_CODE();
567
568 ret = blk_co_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
569 &local_err);
570 if (ret < 0 && ret != -ENOTSUP) {
571 error_propagate(errp, local_err);
572 return ret;
573 }
574
575 size = blk_co_getlength(blk);
576 if (size < 0) {
577 error_free(local_err);
578 error_setg_errno(errp, -size,
579 "Failed to inquire the new image file's length");
580 return size;
581 }
582
583 if (size < minimum_size) {
584 /* Need to grow the image, but we failed to do that */
585 error_propagate(errp, local_err);
586 return -ENOTSUP;
587 }
588
589 error_free(local_err);
590 local_err = NULL;
591
592 return size;
593 }
594
595 /**
596 * Helper function for bdrv_create_file_fallback(): Zero the first
597 * sector to remove any potentially pre-existing image header.
598 */
599 static int coroutine_fn
600 create_file_fallback_zero_first_sector(BlockBackend *blk,
601 int64_t current_size,
602 Error **errp)
603 {
604 int64_t bytes_to_clear;
605 int ret;
606
607 GLOBAL_STATE_CODE();
608
609 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
610 if (bytes_to_clear) {
611 ret = blk_co_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
612 if (ret < 0) {
613 error_setg_errno(errp, -ret,
614 "Failed to clear the new image's first sector");
615 return ret;
616 }
617 }
618
619 return 0;
620 }
621
622 /**
623 * Simple implementation of bdrv_co_create_opts for protocol drivers
624 * which only support creation via opening a file
625 * (usually existing raw storage device)
626 */
627 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
628 const char *filename,
629 QemuOpts *opts,
630 Error **errp)
631 {
632 BlockBackend *blk;
633 QDict *options;
634 int64_t size = 0;
635 char *buf = NULL;
636 PreallocMode prealloc;
637 Error *local_err = NULL;
638 int ret;
639
640 GLOBAL_STATE_CODE();
641
642 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
643 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
644 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
645 PREALLOC_MODE_OFF, &local_err);
646 g_free(buf);
647 if (local_err) {
648 error_propagate(errp, local_err);
649 return -EINVAL;
650 }
651
652 if (prealloc != PREALLOC_MODE_OFF) {
653 error_setg(errp, "Unsupported preallocation mode '%s'",
654 PreallocMode_str(prealloc));
655 return -ENOTSUP;
656 }
657
658 options = qdict_new();
659 qdict_put_str(options, "driver", drv->format_name);
660
661 blk = blk_co_new_open(filename, NULL, options,
662 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
663 if (!blk) {
664 error_prepend(errp, "Protocol driver '%s' does not support image "
665 "creation, and opening the image failed: ",
666 drv->format_name);
667 return -EINVAL;
668 }
669
670 size = create_file_fallback_truncate(blk, size, errp);
671 if (size < 0) {
672 ret = size;
673 goto out;
674 }
675
676 ret = create_file_fallback_zero_first_sector(blk, size, errp);
677 if (ret < 0) {
678 goto out;
679 }
680
681 ret = 0;
682 out:
683 blk_co_unref(blk);
684 return ret;
685 }
686
687 int coroutine_fn bdrv_co_create_file(const char *filename, QemuOpts *opts,
688 Error **errp)
689 {
690 QemuOpts *protocol_opts;
691 BlockDriver *drv;
692 QDict *qdict;
693 int ret;
694
695 GLOBAL_STATE_CODE();
696
697 drv = bdrv_find_protocol(filename, true, errp);
698 if (drv == NULL) {
699 return -ENOENT;
700 }
701
702 if (!drv->create_opts) {
703 error_setg(errp, "Driver '%s' does not support image creation",
704 drv->format_name);
705 return -ENOTSUP;
706 }
707
708 /*
709 * 'opts' contains a QemuOptsList with a combination of format and protocol
710 * default values.
711 *
712 * The format properly removes its options, but the default values remain
713 * in 'opts->list'. So if the protocol has options with the same name
714 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
715 * of the format, since for overlapping options, the format wins.
716 *
717 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
718 * only the set options, and then convert it back to QemuOpts, using the
719 * create_opts of the protocol. So the new QemuOpts, will contain only the
720 * protocol defaults.
721 */
722 qdict = qemu_opts_to_qdict(opts, NULL);
723 protocol_opts = qemu_opts_from_qdict(drv->create_opts, qdict, errp);
724 if (protocol_opts == NULL) {
725 ret = -EINVAL;
726 goto out;
727 }
728
729 ret = bdrv_co_create(drv, filename, protocol_opts, errp);
730 out:
731 qemu_opts_del(protocol_opts);
732 qobject_unref(qdict);
733 return ret;
734 }
735
736 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
737 {
738 Error *local_err = NULL;
739 int ret;
740
741 IO_CODE();
742 assert(bs != NULL);
743 assert_bdrv_graph_readable();
744
745 if (!bs->drv) {
746 error_setg(errp, "Block node '%s' is not opened", bs->filename);
747 return -ENOMEDIUM;
748 }
749
750 if (!bs->drv->bdrv_co_delete_file) {
751 error_setg(errp, "Driver '%s' does not support image deletion",
752 bs->drv->format_name);
753 return -ENOTSUP;
754 }
755
756 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
757 if (ret < 0) {
758 error_propagate(errp, local_err);
759 }
760
761 return ret;
762 }
763
764 void coroutine_fn bdrv_co_delete_file_noerr(BlockDriverState *bs)
765 {
766 Error *local_err = NULL;
767 int ret;
768 IO_CODE();
769
770 if (!bs) {
771 return;
772 }
773
774 ret = bdrv_co_delete_file(bs, &local_err);
775 /*
776 * ENOTSUP will happen if the block driver doesn't support
777 * the 'bdrv_co_delete_file' interface. This is a predictable
778 * scenario and shouldn't be reported back to the user.
779 */
780 if (ret == -ENOTSUP) {
781 error_free(local_err);
782 } else if (ret < 0) {
783 error_report_err(local_err);
784 }
785 }
786
787 /**
788 * Try to get @bs's logical and physical block size.
789 * On success, store them in @bsz struct and return 0.
790 * On failure return -errno.
791 * @bs must not be empty.
792 */
793 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
794 {
795 BlockDriver *drv = bs->drv;
796 BlockDriverState *filtered = bdrv_filter_bs(bs);
797 GLOBAL_STATE_CODE();
798
799 if (drv && drv->bdrv_probe_blocksizes) {
800 return drv->bdrv_probe_blocksizes(bs, bsz);
801 } else if (filtered) {
802 return bdrv_probe_blocksizes(filtered, bsz);
803 }
804
805 return -ENOTSUP;
806 }
807
808 /**
809 * Try to get @bs's geometry (cyls, heads, sectors).
810 * On success, store them in @geo struct and return 0.
811 * On failure return -errno.
812 * @bs must not be empty.
813 */
814 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
815 {
816 BlockDriver *drv = bs->drv;
817 BlockDriverState *filtered = bdrv_filter_bs(bs);
818 GLOBAL_STATE_CODE();
819
820 if (drv && drv->bdrv_probe_geometry) {
821 return drv->bdrv_probe_geometry(bs, geo);
822 } else if (filtered) {
823 return bdrv_probe_geometry(filtered, geo);
824 }
825
826 return -ENOTSUP;
827 }
828
829 /*
830 * Create a uniquely-named empty temporary file.
831 * Return the actual file name used upon success, otherwise NULL.
832 * This string should be freed with g_free() when not needed any longer.
833 *
834 * Note: creating a temporary file for the caller to (re)open is
835 * inherently racy. Use g_file_open_tmp() instead whenever practical.
836 */
837 char *create_tmp_file(Error **errp)
838 {
839 int fd;
840 const char *tmpdir;
841 g_autofree char *filename = NULL;
842
843 tmpdir = g_get_tmp_dir();
844 #ifndef _WIN32
845 /*
846 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
847 *
848 * This function is used to create temporary disk images (like -snapshot),
849 * so the files can become very large. /tmp is often a tmpfs where as
850 * /var/tmp is usually on a disk, so more appropriate for disk images.
851 */
852 if (!g_strcmp0(tmpdir, "/tmp")) {
853 tmpdir = "/var/tmp";
854 }
855 #endif
856
857 filename = g_strdup_printf("%s/vl.XXXXXX", tmpdir);
858 fd = g_mkstemp(filename);
859 if (fd < 0) {
860 error_setg_errno(errp, errno, "Could not open temporary file '%s'",
861 filename);
862 return NULL;
863 }
864 close(fd);
865
866 return g_steal_pointer(&filename);
867 }
868
869 /*
870 * Detect host devices. By convention, /dev/cdrom[N] is always
871 * recognized as a host CDROM.
872 */
873 static BlockDriver *find_hdev_driver(const char *filename)
874 {
875 int score_max = 0, score;
876 BlockDriver *drv = NULL, *d;
877 GLOBAL_STATE_CODE();
878
879 QLIST_FOREACH(d, &bdrv_drivers, list) {
880 if (d->bdrv_probe_device) {
881 score = d->bdrv_probe_device(filename);
882 if (score > score_max) {
883 score_max = score;
884 drv = d;
885 }
886 }
887 }
888
889 return drv;
890 }
891
892 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
893 {
894 BlockDriver *drv1;
895 GLOBAL_STATE_CODE();
896
897 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
898 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
899 return drv1;
900 }
901 }
902
903 return NULL;
904 }
905
906 BlockDriver *bdrv_find_protocol(const char *filename,
907 bool allow_protocol_prefix,
908 Error **errp)
909 {
910 BlockDriver *drv1;
911 char protocol[128];
912 int len;
913 const char *p;
914 int i;
915
916 GLOBAL_STATE_CODE();
917 /* TODO Drivers without bdrv_file_open must be specified explicitly */
918
919 /*
920 * XXX(hch): we really should not let host device detection
921 * override an explicit protocol specification, but moving this
922 * later breaks access to device names with colons in them.
923 * Thanks to the brain-dead persistent naming schemes on udev-
924 * based Linux systems those actually are quite common.
925 */
926 drv1 = find_hdev_driver(filename);
927 if (drv1) {
928 return drv1;
929 }
930
931 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
932 return &bdrv_file;
933 }
934
935 p = strchr(filename, ':');
936 assert(p != NULL);
937 len = p - filename;
938 if (len > sizeof(protocol) - 1)
939 len = sizeof(protocol) - 1;
940 memcpy(protocol, filename, len);
941 protocol[len] = '\0';
942
943 drv1 = bdrv_do_find_protocol(protocol);
944 if (drv1) {
945 return drv1;
946 }
947
948 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
949 if (block_driver_modules[i].protocol_name &&
950 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
951 int rv = block_module_load(block_driver_modules[i].library_name, errp);
952 if (rv > 0) {
953 drv1 = bdrv_do_find_protocol(protocol);
954 } else if (rv < 0) {
955 return NULL;
956 }
957 break;
958 }
959 }
960
961 if (!drv1) {
962 error_setg(errp, "Unknown protocol '%s'", protocol);
963 }
964 return drv1;
965 }
966
967 /*
968 * Guess image format by probing its contents.
969 * This is not a good idea when your image is raw (CVE-2008-2004), but
970 * we do it anyway for backward compatibility.
971 *
972 * @buf contains the image's first @buf_size bytes.
973 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
974 * but can be smaller if the image file is smaller)
975 * @filename is its filename.
976 *
977 * For all block drivers, call the bdrv_probe() method to get its
978 * probing score.
979 * Return the first block driver with the highest probing score.
980 */
981 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
982 const char *filename)
983 {
984 int score_max = 0, score;
985 BlockDriver *drv = NULL, *d;
986 IO_CODE();
987
988 QLIST_FOREACH(d, &bdrv_drivers, list) {
989 if (d->bdrv_probe) {
990 score = d->bdrv_probe(buf, buf_size, filename);
991 if (score > score_max) {
992 score_max = score;
993 drv = d;
994 }
995 }
996 }
997
998 return drv;
999 }
1000
1001 static int find_image_format(BlockBackend *file, const char *filename,
1002 BlockDriver **pdrv, Error **errp)
1003 {
1004 BlockDriver *drv;
1005 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
1006 int ret = 0;
1007
1008 GLOBAL_STATE_CODE();
1009
1010 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1011 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
1012 *pdrv = &bdrv_raw;
1013 return ret;
1014 }
1015
1016 ret = blk_pread(file, 0, sizeof(buf), buf, 0);
1017 if (ret < 0) {
1018 error_setg_errno(errp, -ret, "Could not read image for determining its "
1019 "format");
1020 *pdrv = NULL;
1021 return ret;
1022 }
1023
1024 drv = bdrv_probe_all(buf, sizeof(buf), filename);
1025 if (!drv) {
1026 error_setg(errp, "Could not determine image format: No compatible "
1027 "driver found");
1028 *pdrv = NULL;
1029 return -ENOENT;
1030 }
1031
1032 *pdrv = drv;
1033 return 0;
1034 }
1035
1036 /**
1037 * Set the current 'total_sectors' value
1038 * Return 0 on success, -errno on error.
1039 */
1040 int coroutine_fn bdrv_co_refresh_total_sectors(BlockDriverState *bs,
1041 int64_t hint)
1042 {
1043 BlockDriver *drv = bs->drv;
1044 IO_CODE();
1045 assert_bdrv_graph_readable();
1046
1047 if (!drv) {
1048 return -ENOMEDIUM;
1049 }
1050
1051 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
1052 if (bdrv_is_sg(bs))
1053 return 0;
1054
1055 /* query actual device if possible, otherwise just trust the hint */
1056 if (drv->bdrv_co_getlength) {
1057 int64_t length = drv->bdrv_co_getlength(bs);
1058 if (length < 0) {
1059 return length;
1060 }
1061 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
1062 }
1063
1064 bs->total_sectors = hint;
1065
1066 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
1067 return -EFBIG;
1068 }
1069
1070 return 0;
1071 }
1072
1073 /**
1074 * Combines a QDict of new block driver @options with any missing options taken
1075 * from @old_options, so that leaving out an option defaults to its old value.
1076 */
1077 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
1078 QDict *old_options)
1079 {
1080 GLOBAL_STATE_CODE();
1081 if (bs->drv && bs->drv->bdrv_join_options) {
1082 bs->drv->bdrv_join_options(options, old_options);
1083 } else {
1084 qdict_join(options, old_options, false);
1085 }
1086 }
1087
1088 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
1089 int open_flags,
1090 Error **errp)
1091 {
1092 Error *local_err = NULL;
1093 char *value = qemu_opt_get_del(opts, "detect-zeroes");
1094 BlockdevDetectZeroesOptions detect_zeroes =
1095 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
1096 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
1097 GLOBAL_STATE_CODE();
1098 g_free(value);
1099 if (local_err) {
1100 error_propagate(errp, local_err);
1101 return detect_zeroes;
1102 }
1103
1104 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1105 !(open_flags & BDRV_O_UNMAP))
1106 {
1107 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1108 "without setting discard operation to unmap");
1109 }
1110
1111 return detect_zeroes;
1112 }
1113
1114 /**
1115 * Set open flags for aio engine
1116 *
1117 * Return 0 on success, -1 if the engine specified is invalid
1118 */
1119 int bdrv_parse_aio(const char *mode, int *flags)
1120 {
1121 if (!strcmp(mode, "threads")) {
1122 /* do nothing, default */
1123 } else if (!strcmp(mode, "native")) {
1124 *flags |= BDRV_O_NATIVE_AIO;
1125 #ifdef CONFIG_LINUX_IO_URING
1126 } else if (!strcmp(mode, "io_uring")) {
1127 *flags |= BDRV_O_IO_URING;
1128 #endif
1129 } else {
1130 return -1;
1131 }
1132
1133 return 0;
1134 }
1135
1136 /**
1137 * Set open flags for a given discard mode
1138 *
1139 * Return 0 on success, -1 if the discard mode was invalid.
1140 */
1141 int bdrv_parse_discard_flags(const char *mode, int *flags)
1142 {
1143 *flags &= ~BDRV_O_UNMAP;
1144
1145 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1146 /* do nothing */
1147 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1148 *flags |= BDRV_O_UNMAP;
1149 } else {
1150 return -1;
1151 }
1152
1153 return 0;
1154 }
1155
1156 /**
1157 * Set open flags for a given cache mode
1158 *
1159 * Return 0 on success, -1 if the cache mode was invalid.
1160 */
1161 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1162 {
1163 *flags &= ~BDRV_O_CACHE_MASK;
1164
1165 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1166 *writethrough = false;
1167 *flags |= BDRV_O_NOCACHE;
1168 } else if (!strcmp(mode, "directsync")) {
1169 *writethrough = true;
1170 *flags |= BDRV_O_NOCACHE;
1171 } else if (!strcmp(mode, "writeback")) {
1172 *writethrough = false;
1173 } else if (!strcmp(mode, "unsafe")) {
1174 *writethrough = false;
1175 *flags |= BDRV_O_NO_FLUSH;
1176 } else if (!strcmp(mode, "writethrough")) {
1177 *writethrough = true;
1178 } else {
1179 return -1;
1180 }
1181
1182 return 0;
1183 }
1184
1185 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1186 {
1187 BlockDriverState *parent = c->opaque;
1188 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent));
1189 }
1190
1191 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1192 {
1193 BlockDriverState *bs = child->opaque;
1194 bdrv_do_drained_begin_quiesce(bs, NULL);
1195 }
1196
1197 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1198 {
1199 BlockDriverState *bs = child->opaque;
1200 return bdrv_drain_poll(bs, NULL, false);
1201 }
1202
1203 static void bdrv_child_cb_drained_end(BdrvChild *child)
1204 {
1205 BlockDriverState *bs = child->opaque;
1206 bdrv_drained_end(bs);
1207 }
1208
1209 static int bdrv_child_cb_inactivate(BdrvChild *child)
1210 {
1211 BlockDriverState *bs = child->opaque;
1212 GLOBAL_STATE_CODE();
1213 assert(bs->open_flags & BDRV_O_INACTIVE);
1214 return 0;
1215 }
1216
1217 static bool bdrv_child_cb_change_aio_ctx(BdrvChild *child, AioContext *ctx,
1218 GHashTable *visited, Transaction *tran,
1219 Error **errp)
1220 {
1221 BlockDriverState *bs = child->opaque;
1222 return bdrv_change_aio_context(bs, ctx, visited, tran, errp);
1223 }
1224
1225 /*
1226 * Returns the options and flags that a temporary snapshot should get, based on
1227 * the originally requested flags (the originally requested image will have
1228 * flags like a backing file)
1229 */
1230 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1231 int parent_flags, QDict *parent_options)
1232 {
1233 GLOBAL_STATE_CODE();
1234 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1235
1236 /* For temporary files, unconditional cache=unsafe is fine */
1237 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1238 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1239
1240 /* Copy the read-only and discard options from the parent */
1241 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1242 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1243
1244 /* aio=native doesn't work for cache.direct=off, so disable it for the
1245 * temporary snapshot */
1246 *child_flags &= ~BDRV_O_NATIVE_AIO;
1247 }
1248
1249 static void bdrv_backing_attach(BdrvChild *c)
1250 {
1251 BlockDriverState *parent = c->opaque;
1252 BlockDriverState *backing_hd = c->bs;
1253
1254 GLOBAL_STATE_CODE();
1255 assert(!parent->backing_blocker);
1256 error_setg(&parent->backing_blocker,
1257 "node is used as backing hd of '%s'",
1258 bdrv_get_device_or_node_name(parent));
1259
1260 bdrv_refresh_filename(backing_hd);
1261
1262 parent->open_flags &= ~BDRV_O_NO_BACKING;
1263
1264 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1265 /* Otherwise we won't be able to commit or stream */
1266 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1267 parent->backing_blocker);
1268 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1269 parent->backing_blocker);
1270 /*
1271 * We do backup in 3 ways:
1272 * 1. drive backup
1273 * The target bs is new opened, and the source is top BDS
1274 * 2. blockdev backup
1275 * Both the source and the target are top BDSes.
1276 * 3. internal backup(used for block replication)
1277 * Both the source and the target are backing file
1278 *
1279 * In case 1 and 2, neither the source nor the target is the backing file.
1280 * In case 3, we will block the top BDS, so there is only one block job
1281 * for the top BDS and its backing chain.
1282 */
1283 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1284 parent->backing_blocker);
1285 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1286 parent->backing_blocker);
1287 }
1288
1289 static void bdrv_backing_detach(BdrvChild *c)
1290 {
1291 BlockDriverState *parent = c->opaque;
1292
1293 GLOBAL_STATE_CODE();
1294 assert(parent->backing_blocker);
1295 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1296 error_free(parent->backing_blocker);
1297 parent->backing_blocker = NULL;
1298 }
1299
1300 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1301 const char *filename, Error **errp)
1302 {
1303 BlockDriverState *parent = c->opaque;
1304 bool read_only = bdrv_is_read_only(parent);
1305 int ret;
1306 GLOBAL_STATE_CODE();
1307
1308 if (read_only) {
1309 ret = bdrv_reopen_set_read_only(parent, false, errp);
1310 if (ret < 0) {
1311 return ret;
1312 }
1313 }
1314
1315 ret = bdrv_change_backing_file(parent, filename,
1316 base->drv ? base->drv->format_name : "",
1317 false);
1318 if (ret < 0) {
1319 error_setg_errno(errp, -ret, "Could not update backing file link");
1320 }
1321
1322 if (read_only) {
1323 bdrv_reopen_set_read_only(parent, true, NULL);
1324 }
1325
1326 return ret;
1327 }
1328
1329 /*
1330 * Returns the options and flags that a generic child of a BDS should
1331 * get, based on the given options and flags for the parent BDS.
1332 */
1333 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1334 int *child_flags, QDict *child_options,
1335 int parent_flags, QDict *parent_options)
1336 {
1337 int flags = parent_flags;
1338 GLOBAL_STATE_CODE();
1339
1340 /*
1341 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1342 * Generally, the question to answer is: Should this child be
1343 * format-probed by default?
1344 */
1345
1346 /*
1347 * Pure and non-filtered data children of non-format nodes should
1348 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1349 * set). This only affects a very limited set of drivers (namely
1350 * quorum and blkverify when this comment was written).
1351 * Force-clear BDRV_O_PROTOCOL then.
1352 */
1353 if (!parent_is_format &&
1354 (role & BDRV_CHILD_DATA) &&
1355 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1356 {
1357 flags &= ~BDRV_O_PROTOCOL;
1358 }
1359
1360 /*
1361 * All children of format nodes (except for COW children) and all
1362 * metadata children in general should never be format-probed.
1363 * Force-set BDRV_O_PROTOCOL then.
1364 */
1365 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1366 (role & BDRV_CHILD_METADATA))
1367 {
1368 flags |= BDRV_O_PROTOCOL;
1369 }
1370
1371 /*
1372 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1373 * the parent.
1374 */
1375 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1376 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1377 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1378
1379 if (role & BDRV_CHILD_COW) {
1380 /* backing files are opened read-only by default */
1381 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1382 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1383 } else {
1384 /* Inherit the read-only option from the parent if it's not set */
1385 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1386 qdict_copy_default(child_options, parent_options,
1387 BDRV_OPT_AUTO_READ_ONLY);
1388 }
1389
1390 /*
1391 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1392 * can default to enable it on lower layers regardless of the
1393 * parent option.
1394 */
1395 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1396
1397 /* Clear flags that only apply to the top layer */
1398 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1399
1400 if (role & BDRV_CHILD_METADATA) {
1401 flags &= ~BDRV_O_NO_IO;
1402 }
1403 if (role & BDRV_CHILD_COW) {
1404 flags &= ~BDRV_O_TEMPORARY;
1405 }
1406
1407 *child_flags = flags;
1408 }
1409
1410 static void GRAPH_WRLOCK bdrv_child_cb_attach(BdrvChild *child)
1411 {
1412 BlockDriverState *bs = child->opaque;
1413
1414 assert_bdrv_graph_writable();
1415 QLIST_INSERT_HEAD(&bs->children, child, next);
1416 if (bs->drv->is_filter || (child->role & BDRV_CHILD_FILTERED)) {
1417 /*
1418 * Here we handle filters and block/raw-format.c when it behave like
1419 * filter. They generally have a single PRIMARY child, which is also the
1420 * FILTERED child, and that they may have multiple more children, which
1421 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1422 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1423 * into bs->backing on exceptional cases; and bs->backing will be
1424 * nothing else.
1425 */
1426 assert(!(child->role & BDRV_CHILD_COW));
1427 if (child->role & BDRV_CHILD_PRIMARY) {
1428 assert(child->role & BDRV_CHILD_FILTERED);
1429 assert(!bs->backing);
1430 assert(!bs->file);
1431
1432 if (bs->drv->filtered_child_is_backing) {
1433 bs->backing = child;
1434 } else {
1435 bs->file = child;
1436 }
1437 } else {
1438 assert(!(child->role & BDRV_CHILD_FILTERED));
1439 }
1440 } else if (child->role & BDRV_CHILD_COW) {
1441 assert(bs->drv->supports_backing);
1442 assert(!(child->role & BDRV_CHILD_PRIMARY));
1443 assert(!bs->backing);
1444 bs->backing = child;
1445 bdrv_backing_attach(child);
1446 } else if (child->role & BDRV_CHILD_PRIMARY) {
1447 assert(!bs->file);
1448 bs->file = child;
1449 }
1450 }
1451
1452 static void GRAPH_WRLOCK bdrv_child_cb_detach(BdrvChild *child)
1453 {
1454 BlockDriverState *bs = child->opaque;
1455
1456 if (child->role & BDRV_CHILD_COW) {
1457 bdrv_backing_detach(child);
1458 }
1459
1460 assert_bdrv_graph_writable();
1461 QLIST_REMOVE(child, next);
1462 if (child == bs->backing) {
1463 assert(child != bs->file);
1464 bs->backing = NULL;
1465 } else if (child == bs->file) {
1466 bs->file = NULL;
1467 }
1468 }
1469
1470 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1471 const char *filename, Error **errp)
1472 {
1473 if (c->role & BDRV_CHILD_COW) {
1474 return bdrv_backing_update_filename(c, base, filename, errp);
1475 }
1476 return 0;
1477 }
1478
1479 AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c)
1480 {
1481 BlockDriverState *bs = c->opaque;
1482 IO_CODE();
1483
1484 return bdrv_get_aio_context(bs);
1485 }
1486
1487 const BdrvChildClass child_of_bds = {
1488 .parent_is_bds = true,
1489 .get_parent_desc = bdrv_child_get_parent_desc,
1490 .inherit_options = bdrv_inherited_options,
1491 .drained_begin = bdrv_child_cb_drained_begin,
1492 .drained_poll = bdrv_child_cb_drained_poll,
1493 .drained_end = bdrv_child_cb_drained_end,
1494 .attach = bdrv_child_cb_attach,
1495 .detach = bdrv_child_cb_detach,
1496 .inactivate = bdrv_child_cb_inactivate,
1497 .change_aio_ctx = bdrv_child_cb_change_aio_ctx,
1498 .update_filename = bdrv_child_cb_update_filename,
1499 .get_parent_aio_context = child_of_bds_get_parent_aio_context,
1500 };
1501
1502 AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c)
1503 {
1504 IO_CODE();
1505 return c->klass->get_parent_aio_context(c);
1506 }
1507
1508 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1509 {
1510 int open_flags = flags;
1511 GLOBAL_STATE_CODE();
1512
1513 /*
1514 * Clear flags that are internal to the block layer before opening the
1515 * image.
1516 */
1517 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1518
1519 return open_flags;
1520 }
1521
1522 static void update_flags_from_options(int *flags, QemuOpts *opts)
1523 {
1524 GLOBAL_STATE_CODE();
1525
1526 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1527
1528 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1529 *flags |= BDRV_O_NO_FLUSH;
1530 }
1531
1532 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1533 *flags |= BDRV_O_NOCACHE;
1534 }
1535
1536 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1537 *flags |= BDRV_O_RDWR;
1538 }
1539
1540 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1541 *flags |= BDRV_O_AUTO_RDONLY;
1542 }
1543 }
1544
1545 static void update_options_from_flags(QDict *options, int flags)
1546 {
1547 GLOBAL_STATE_CODE();
1548 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1549 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1550 }
1551 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1552 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1553 flags & BDRV_O_NO_FLUSH);
1554 }
1555 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1556 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1557 }
1558 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1559 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1560 flags & BDRV_O_AUTO_RDONLY);
1561 }
1562 }
1563
1564 static void bdrv_assign_node_name(BlockDriverState *bs,
1565 const char *node_name,
1566 Error **errp)
1567 {
1568 char *gen_node_name = NULL;
1569 GLOBAL_STATE_CODE();
1570
1571 if (!node_name) {
1572 node_name = gen_node_name = id_generate(ID_BLOCK);
1573 } else if (!id_wellformed(node_name)) {
1574 /*
1575 * Check for empty string or invalid characters, but not if it is
1576 * generated (generated names use characters not available to the user)
1577 */
1578 error_setg(errp, "Invalid node-name: '%s'", node_name);
1579 return;
1580 }
1581
1582 /* takes care of avoiding namespaces collisions */
1583 if (blk_by_name(node_name)) {
1584 error_setg(errp, "node-name=%s is conflicting with a device id",
1585 node_name);
1586 goto out;
1587 }
1588
1589 /* takes care of avoiding duplicates node names */
1590 if (bdrv_find_node(node_name)) {
1591 error_setg(errp, "Duplicate nodes with node-name='%s'", node_name);
1592 goto out;
1593 }
1594
1595 /* Make sure that the node name isn't truncated */
1596 if (strlen(node_name) >= sizeof(bs->node_name)) {
1597 error_setg(errp, "Node name too long");
1598 goto out;
1599 }
1600
1601 /* copy node name into the bs and insert it into the graph list */
1602 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1603 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1604 out:
1605 g_free(gen_node_name);
1606 }
1607
1608 /*
1609 * The caller must always hold @bs AioContext lock, because this function calls
1610 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1611 * context.
1612 */
1613 static int no_coroutine_fn GRAPH_UNLOCKED
1614 bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv, const char *node_name,
1615 QDict *options, int open_flags, Error **errp)
1616 {
1617 AioContext *ctx;
1618 Error *local_err = NULL;
1619 int i, ret;
1620 GLOBAL_STATE_CODE();
1621
1622 bdrv_assign_node_name(bs, node_name, &local_err);
1623 if (local_err) {
1624 error_propagate(errp, local_err);
1625 return -EINVAL;
1626 }
1627
1628 bs->drv = drv;
1629 bs->opaque = g_malloc0(drv->instance_size);
1630
1631 if (drv->bdrv_file_open) {
1632 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1633 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1634 } else if (drv->bdrv_open) {
1635 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1636 } else {
1637 ret = 0;
1638 }
1639
1640 if (ret < 0) {
1641 if (local_err) {
1642 error_propagate(errp, local_err);
1643 } else if (bs->filename[0]) {
1644 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1645 } else {
1646 error_setg_errno(errp, -ret, "Could not open image");
1647 }
1648 goto open_failed;
1649 }
1650
1651 assert(!(bs->supported_read_flags & ~BDRV_REQ_MASK));
1652 assert(!(bs->supported_write_flags & ~BDRV_REQ_MASK));
1653
1654 /*
1655 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1656 * drivers that pass read/write requests through to a child the trouble of
1657 * declaring support explicitly.
1658 *
1659 * Drivers must not propagate this flag accidentally when they initiate I/O
1660 * to a bounce buffer. That case should be rare though.
1661 */
1662 bs->supported_read_flags |= BDRV_REQ_REGISTERED_BUF;
1663 bs->supported_write_flags |= BDRV_REQ_REGISTERED_BUF;
1664
1665 /* Get the context after .bdrv_open, it can change the context */
1666 ctx = bdrv_get_aio_context(bs);
1667 aio_context_acquire(ctx);
1668
1669 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
1670 if (ret < 0) {
1671 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1672 aio_context_release(ctx);
1673 return ret;
1674 }
1675
1676 bdrv_graph_rdlock_main_loop();
1677 bdrv_refresh_limits(bs, NULL, &local_err);
1678 bdrv_graph_rdunlock_main_loop();
1679 aio_context_release(ctx);
1680
1681 if (local_err) {
1682 error_propagate(errp, local_err);
1683 return -EINVAL;
1684 }
1685
1686 assert(bdrv_opt_mem_align(bs) != 0);
1687 assert(bdrv_min_mem_align(bs) != 0);
1688 assert(is_power_of_2(bs->bl.request_alignment));
1689
1690 for (i = 0; i < bs->quiesce_counter; i++) {
1691 if (drv->bdrv_drain_begin) {
1692 drv->bdrv_drain_begin(bs);
1693 }
1694 }
1695
1696 return 0;
1697 open_failed:
1698 bs->drv = NULL;
1699 if (bs->file != NULL) {
1700 bdrv_unref_child(bs, bs->file);
1701 assert(!bs->file);
1702 }
1703 g_free(bs->opaque);
1704 bs->opaque = NULL;
1705 return ret;
1706 }
1707
1708 /*
1709 * Create and open a block node.
1710 *
1711 * @options is a QDict of options to pass to the block drivers, or NULL for an
1712 * empty set of options. The reference to the QDict belongs to the block layer
1713 * after the call (even on failure), so if the caller intends to reuse the
1714 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1715 */
1716 BlockDriverState *bdrv_new_open_driver_opts(BlockDriver *drv,
1717 const char *node_name,
1718 QDict *options, int flags,
1719 Error **errp)
1720 {
1721 BlockDriverState *bs;
1722 int ret;
1723
1724 GLOBAL_STATE_CODE();
1725
1726 bs = bdrv_new();
1727 bs->open_flags = flags;
1728 bs->options = options ?: qdict_new();
1729 bs->explicit_options = qdict_clone_shallow(bs->options);
1730 bs->opaque = NULL;
1731
1732 update_options_from_flags(bs->options, flags);
1733
1734 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1735 if (ret < 0) {
1736 qobject_unref(bs->explicit_options);
1737 bs->explicit_options = NULL;
1738 qobject_unref(bs->options);
1739 bs->options = NULL;
1740 bdrv_unref(bs);
1741 return NULL;
1742 }
1743
1744 return bs;
1745 }
1746
1747 /* Create and open a block node. */
1748 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1749 int flags, Error **errp)
1750 {
1751 GLOBAL_STATE_CODE();
1752 return bdrv_new_open_driver_opts(drv, node_name, NULL, flags, errp);
1753 }
1754
1755 QemuOptsList bdrv_runtime_opts = {
1756 .name = "bdrv_common",
1757 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1758 .desc = {
1759 {
1760 .name = "node-name",
1761 .type = QEMU_OPT_STRING,
1762 .help = "Node name of the block device node",
1763 },
1764 {
1765 .name = "driver",
1766 .type = QEMU_OPT_STRING,
1767 .help = "Block driver to use for the node",
1768 },
1769 {
1770 .name = BDRV_OPT_CACHE_DIRECT,
1771 .type = QEMU_OPT_BOOL,
1772 .help = "Bypass software writeback cache on the host",
1773 },
1774 {
1775 .name = BDRV_OPT_CACHE_NO_FLUSH,
1776 .type = QEMU_OPT_BOOL,
1777 .help = "Ignore flush requests",
1778 },
1779 {
1780 .name = BDRV_OPT_READ_ONLY,
1781 .type = QEMU_OPT_BOOL,
1782 .help = "Node is opened in read-only mode",
1783 },
1784 {
1785 .name = BDRV_OPT_AUTO_READ_ONLY,
1786 .type = QEMU_OPT_BOOL,
1787 .help = "Node can become read-only if opening read-write fails",
1788 },
1789 {
1790 .name = "detect-zeroes",
1791 .type = QEMU_OPT_STRING,
1792 .help = "try to optimize zero writes (off, on, unmap)",
1793 },
1794 {
1795 .name = BDRV_OPT_DISCARD,
1796 .type = QEMU_OPT_STRING,
1797 .help = "discard operation (ignore/off, unmap/on)",
1798 },
1799 {
1800 .name = BDRV_OPT_FORCE_SHARE,
1801 .type = QEMU_OPT_BOOL,
1802 .help = "always accept other writers (default: off)",
1803 },
1804 { /* end of list */ }
1805 },
1806 };
1807
1808 QemuOptsList bdrv_create_opts_simple = {
1809 .name = "simple-create-opts",
1810 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1811 .desc = {
1812 {
1813 .name = BLOCK_OPT_SIZE,
1814 .type = QEMU_OPT_SIZE,
1815 .help = "Virtual disk size"
1816 },
1817 {
1818 .name = BLOCK_OPT_PREALLOC,
1819 .type = QEMU_OPT_STRING,
1820 .help = "Preallocation mode (allowed values: off)"
1821 },
1822 { /* end of list */ }
1823 }
1824 };
1825
1826 /*
1827 * Common part for opening disk images and files
1828 *
1829 * Removes all processed options from *options.
1830 */
1831 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1832 QDict *options, Error **errp)
1833 {
1834 int ret, open_flags;
1835 const char *filename;
1836 const char *driver_name = NULL;
1837 const char *node_name = NULL;
1838 const char *discard;
1839 QemuOpts *opts;
1840 BlockDriver *drv;
1841 Error *local_err = NULL;
1842 bool ro;
1843
1844 assert(bs->file == NULL);
1845 assert(options != NULL && bs->options != options);
1846 GLOBAL_STATE_CODE();
1847
1848 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1849 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1850 ret = -EINVAL;
1851 goto fail_opts;
1852 }
1853
1854 update_flags_from_options(&bs->open_flags, opts);
1855
1856 driver_name = qemu_opt_get(opts, "driver");
1857 drv = bdrv_find_format(driver_name);
1858 assert(drv != NULL);
1859
1860 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1861
1862 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1863 error_setg(errp,
1864 BDRV_OPT_FORCE_SHARE
1865 "=on can only be used with read-only images");
1866 ret = -EINVAL;
1867 goto fail_opts;
1868 }
1869
1870 if (file != NULL) {
1871 bdrv_refresh_filename(blk_bs(file));
1872 filename = blk_bs(file)->filename;
1873 } else {
1874 /*
1875 * Caution: while qdict_get_try_str() is fine, getting
1876 * non-string types would require more care. When @options
1877 * come from -blockdev or blockdev_add, its members are typed
1878 * according to the QAPI schema, but when they come from
1879 * -drive, they're all QString.
1880 */
1881 filename = qdict_get_try_str(options, "filename");
1882 }
1883
1884 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1885 error_setg(errp, "The '%s' block driver requires a file name",
1886 drv->format_name);
1887 ret = -EINVAL;
1888 goto fail_opts;
1889 }
1890
1891 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1892 drv->format_name);
1893
1894 ro = bdrv_is_read_only(bs);
1895
1896 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, ro)) {
1897 if (!ro && bdrv_is_whitelisted(drv, true)) {
1898 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1899 } else {
1900 ret = -ENOTSUP;
1901 }
1902 if (ret < 0) {
1903 error_setg(errp,
1904 !ro && bdrv_is_whitelisted(drv, true)
1905 ? "Driver '%s' can only be used for read-only devices"
1906 : "Driver '%s' is not whitelisted",
1907 drv->format_name);
1908 goto fail_opts;
1909 }
1910 }
1911
1912 /* bdrv_new() and bdrv_close() make it so */
1913 assert(qatomic_read(&bs->copy_on_read) == 0);
1914
1915 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1916 if (!ro) {
1917 bdrv_enable_copy_on_read(bs);
1918 } else {
1919 error_setg(errp, "Can't use copy-on-read on read-only device");
1920 ret = -EINVAL;
1921 goto fail_opts;
1922 }
1923 }
1924
1925 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1926 if (discard != NULL) {
1927 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1928 error_setg(errp, "Invalid discard option");
1929 ret = -EINVAL;
1930 goto fail_opts;
1931 }
1932 }
1933
1934 bs->detect_zeroes =
1935 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1936 if (local_err) {
1937 error_propagate(errp, local_err);
1938 ret = -EINVAL;
1939 goto fail_opts;
1940 }
1941
1942 if (filename != NULL) {
1943 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1944 } else {
1945 bs->filename[0] = '\0';
1946 }
1947 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1948
1949 /* Open the image, either directly or using a protocol */
1950 open_flags = bdrv_open_flags(bs, bs->open_flags);
1951 node_name = qemu_opt_get(opts, "node-name");
1952
1953 assert(!drv->bdrv_file_open || file == NULL);
1954 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1955 if (ret < 0) {
1956 goto fail_opts;
1957 }
1958
1959 qemu_opts_del(opts);
1960 return 0;
1961
1962 fail_opts:
1963 qemu_opts_del(opts);
1964 return ret;
1965 }
1966
1967 static QDict *parse_json_filename(const char *filename, Error **errp)
1968 {
1969 QObject *options_obj;
1970 QDict *options;
1971 int ret;
1972 GLOBAL_STATE_CODE();
1973
1974 ret = strstart(filename, "json:", &filename);
1975 assert(ret);
1976
1977 options_obj = qobject_from_json(filename, errp);
1978 if (!options_obj) {
1979 error_prepend(errp, "Could not parse the JSON options: ");
1980 return NULL;
1981 }
1982
1983 options = qobject_to(QDict, options_obj);
1984 if (!options) {
1985 qobject_unref(options_obj);
1986 error_setg(errp, "Invalid JSON object given");
1987 return NULL;
1988 }
1989
1990 qdict_flatten(options);
1991
1992 return options;
1993 }
1994
1995 static void parse_json_protocol(QDict *options, const char **pfilename,
1996 Error **errp)
1997 {
1998 QDict *json_options;
1999 Error *local_err = NULL;
2000 GLOBAL_STATE_CODE();
2001
2002 /* Parse json: pseudo-protocol */
2003 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
2004 return;
2005 }
2006
2007 json_options = parse_json_filename(*pfilename, &local_err);
2008 if (local_err) {
2009 error_propagate(errp, local_err);
2010 return;
2011 }
2012
2013 /* Options given in the filename have lower priority than options
2014 * specified directly */
2015 qdict_join(options, json_options, false);
2016 qobject_unref(json_options);
2017 *pfilename = NULL;
2018 }
2019
2020 /*
2021 * Fills in default options for opening images and converts the legacy
2022 * filename/flags pair to option QDict entries.
2023 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2024 * block driver has been specified explicitly.
2025 */
2026 static int bdrv_fill_options(QDict **options, const char *filename,
2027 int *flags, Error **errp)
2028 {
2029 const char *drvname;
2030 bool protocol = *flags & BDRV_O_PROTOCOL;
2031 bool parse_filename = false;
2032 BlockDriver *drv = NULL;
2033 Error *local_err = NULL;
2034
2035 GLOBAL_STATE_CODE();
2036
2037 /*
2038 * Caution: while qdict_get_try_str() is fine, getting non-string
2039 * types would require more care. When @options come from
2040 * -blockdev or blockdev_add, its members are typed according to
2041 * the QAPI schema, but when they come from -drive, they're all
2042 * QString.
2043 */
2044 drvname = qdict_get_try_str(*options, "driver");
2045 if (drvname) {
2046 drv = bdrv_find_format(drvname);
2047 if (!drv) {
2048 error_setg(errp, "Unknown driver '%s'", drvname);
2049 return -ENOENT;
2050 }
2051 /* If the user has explicitly specified the driver, this choice should
2052 * override the BDRV_O_PROTOCOL flag */
2053 protocol = drv->bdrv_file_open;
2054 }
2055
2056 if (protocol) {
2057 *flags |= BDRV_O_PROTOCOL;
2058 } else {
2059 *flags &= ~BDRV_O_PROTOCOL;
2060 }
2061
2062 /* Translate cache options from flags into options */
2063 update_options_from_flags(*options, *flags);
2064
2065 /* Fetch the file name from the options QDict if necessary */
2066 if (protocol && filename) {
2067 if (!qdict_haskey(*options, "filename")) {
2068 qdict_put_str(*options, "filename", filename);
2069 parse_filename = true;
2070 } else {
2071 error_setg(errp, "Can't specify 'file' and 'filename' options at "
2072 "the same time");
2073 return -EINVAL;
2074 }
2075 }
2076
2077 /* Find the right block driver */
2078 /* See cautionary note on accessing @options above */
2079 filename = qdict_get_try_str(*options, "filename");
2080
2081 if (!drvname && protocol) {
2082 if (filename) {
2083 drv = bdrv_find_protocol(filename, parse_filename, errp);
2084 if (!drv) {
2085 return -EINVAL;
2086 }
2087
2088 drvname = drv->format_name;
2089 qdict_put_str(*options, "driver", drvname);
2090 } else {
2091 error_setg(errp, "Must specify either driver or file");
2092 return -EINVAL;
2093 }
2094 }
2095
2096 assert(drv || !protocol);
2097
2098 /* Driver-specific filename parsing */
2099 if (drv && drv->bdrv_parse_filename && parse_filename) {
2100 drv->bdrv_parse_filename(filename, *options, &local_err);
2101 if (local_err) {
2102 error_propagate(errp, local_err);
2103 return -EINVAL;
2104 }
2105
2106 if (!drv->bdrv_needs_filename) {
2107 qdict_del(*options, "filename");
2108 }
2109 }
2110
2111 return 0;
2112 }
2113
2114 typedef struct BlockReopenQueueEntry {
2115 bool prepared;
2116 bool perms_checked;
2117 BDRVReopenState state;
2118 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
2119 } BlockReopenQueueEntry;
2120
2121 /*
2122 * Return the flags that @bs will have after the reopens in @q have
2123 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2124 * return the current flags.
2125 */
2126 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
2127 {
2128 BlockReopenQueueEntry *entry;
2129
2130 if (q != NULL) {
2131 QTAILQ_FOREACH(entry, q, entry) {
2132 if (entry->state.bs == bs) {
2133 return entry->state.flags;
2134 }
2135 }
2136 }
2137
2138 return bs->open_flags;
2139 }
2140
2141 /* Returns whether the image file can be written to after the reopen queue @q
2142 * has been successfully applied, or right now if @q is NULL. */
2143 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
2144 BlockReopenQueue *q)
2145 {
2146 int flags = bdrv_reopen_get_flags(q, bs);
2147
2148 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
2149 }
2150
2151 /*
2152 * Return whether the BDS can be written to. This is not necessarily
2153 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2154 * be written to but do not count as read-only images.
2155 */
2156 bool bdrv_is_writable(BlockDriverState *bs)
2157 {
2158 IO_CODE();
2159 return bdrv_is_writable_after_reopen(bs, NULL);
2160 }
2161
2162 static char *bdrv_child_user_desc(BdrvChild *c)
2163 {
2164 GLOBAL_STATE_CODE();
2165 return c->klass->get_parent_desc(c);
2166 }
2167
2168 /*
2169 * Check that @a allows everything that @b needs. @a and @b must reference same
2170 * child node.
2171 */
2172 static bool bdrv_a_allow_b(BdrvChild *a, BdrvChild *b, Error **errp)
2173 {
2174 const char *child_bs_name;
2175 g_autofree char *a_user = NULL;
2176 g_autofree char *b_user = NULL;
2177 g_autofree char *perms = NULL;
2178
2179 assert(a->bs);
2180 assert(a->bs == b->bs);
2181 GLOBAL_STATE_CODE();
2182
2183 if ((b->perm & a->shared_perm) == b->perm) {
2184 return true;
2185 }
2186
2187 child_bs_name = bdrv_get_node_name(b->bs);
2188 a_user = bdrv_child_user_desc(a);
2189 b_user = bdrv_child_user_desc(b);
2190 perms = bdrv_perm_names(b->perm & ~a->shared_perm);
2191
2192 error_setg(errp, "Permission conflict on node '%s': permissions '%s' are "
2193 "both required by %s (uses node '%s' as '%s' child) and "
2194 "unshared by %s (uses node '%s' as '%s' child).",
2195 child_bs_name, perms,
2196 b_user, child_bs_name, b->name,
2197 a_user, child_bs_name, a->name);
2198
2199 return false;
2200 }
2201
2202 static bool bdrv_parent_perms_conflict(BlockDriverState *bs, Error **errp)
2203 {
2204 BdrvChild *a, *b;
2205 GLOBAL_STATE_CODE();
2206
2207 /*
2208 * During the loop we'll look at each pair twice. That's correct because
2209 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2210 * directions.
2211 */
2212 QLIST_FOREACH(a, &bs->parents, next_parent) {
2213 QLIST_FOREACH(b, &bs->parents, next_parent) {
2214 if (a == b) {
2215 continue;
2216 }
2217
2218 if (!bdrv_a_allow_b(a, b, errp)) {
2219 return true;
2220 }
2221 }
2222 }
2223
2224 return false;
2225 }
2226
2227 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
2228 BdrvChild *c, BdrvChildRole role,
2229 BlockReopenQueue *reopen_queue,
2230 uint64_t parent_perm, uint64_t parent_shared,
2231 uint64_t *nperm, uint64_t *nshared)
2232 {
2233 assert(bs->drv && bs->drv->bdrv_child_perm);
2234 GLOBAL_STATE_CODE();
2235 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
2236 parent_perm, parent_shared,
2237 nperm, nshared);
2238 /* TODO Take force_share from reopen_queue */
2239 if (child_bs && child_bs->force_share) {
2240 *nshared = BLK_PERM_ALL;
2241 }
2242 }
2243
2244 /*
2245 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2246 * nodes that are already in the @list, of course) so that final list is
2247 * topologically sorted. Return the result (GSList @list object is updated, so
2248 * don't use old reference after function call).
2249 *
2250 * On function start @list must be already topologically sorted and for any node
2251 * in the @list the whole subtree of the node must be in the @list as well. The
2252 * simplest way to satisfy this criteria: use only result of
2253 * bdrv_topological_dfs() or NULL as @list parameter.
2254 */
2255 static GSList *bdrv_topological_dfs(GSList *list, GHashTable *found,
2256 BlockDriverState *bs)
2257 {
2258 BdrvChild *child;
2259 g_autoptr(GHashTable) local_found = NULL;
2260
2261 GLOBAL_STATE_CODE();
2262
2263 if (!found) {
2264 assert(!list);
2265 found = local_found = g_hash_table_new(NULL, NULL);
2266 }
2267
2268 if (g_hash_table_contains(found, bs)) {
2269 return list;
2270 }
2271 g_hash_table_add(found, bs);
2272
2273 QLIST_FOREACH(child, &bs->children, next) {
2274 list = bdrv_topological_dfs(list, found, child->bs);
2275 }
2276
2277 return g_slist_prepend(list, bs);
2278 }
2279
2280 typedef struct BdrvChildSetPermState {
2281 BdrvChild *child;
2282 uint64_t old_perm;
2283 uint64_t old_shared_perm;
2284 } BdrvChildSetPermState;
2285
2286 static void bdrv_child_set_perm_abort(void *opaque)
2287 {
2288 BdrvChildSetPermState *s = opaque;
2289
2290 GLOBAL_STATE_CODE();
2291
2292 s->child->perm = s->old_perm;
2293 s->child->shared_perm = s->old_shared_perm;
2294 }
2295
2296 static TransactionActionDrv bdrv_child_set_pem_drv = {
2297 .abort = bdrv_child_set_perm_abort,
2298 .clean = g_free,
2299 };
2300
2301 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm,
2302 uint64_t shared, Transaction *tran)
2303 {
2304 BdrvChildSetPermState *s = g_new(BdrvChildSetPermState, 1);
2305 GLOBAL_STATE_CODE();
2306
2307 *s = (BdrvChildSetPermState) {
2308 .child = c,
2309 .old_perm = c->perm,
2310 .old_shared_perm = c->shared_perm,
2311 };
2312
2313 c->perm = perm;
2314 c->shared_perm = shared;
2315
2316 tran_add(tran, &bdrv_child_set_pem_drv, s);
2317 }
2318
2319 static void bdrv_drv_set_perm_commit(void *opaque)
2320 {
2321 BlockDriverState *bs = opaque;
2322 uint64_t cumulative_perms, cumulative_shared_perms;
2323 GLOBAL_STATE_CODE();
2324
2325 if (bs->drv->bdrv_set_perm) {
2326 bdrv_get_cumulative_perm(bs, &cumulative_perms,
2327 &cumulative_shared_perms);
2328 bs->drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2329 }
2330 }
2331
2332 static void bdrv_drv_set_perm_abort(void *opaque)
2333 {
2334 BlockDriverState *bs = opaque;
2335 GLOBAL_STATE_CODE();
2336
2337 if (bs->drv->bdrv_abort_perm_update) {
2338 bs->drv->bdrv_abort_perm_update(bs);
2339 }
2340 }
2341
2342 TransactionActionDrv bdrv_drv_set_perm_drv = {
2343 .abort = bdrv_drv_set_perm_abort,
2344 .commit = bdrv_drv_set_perm_commit,
2345 };
2346
2347 static int bdrv_drv_set_perm(BlockDriverState *bs, uint64_t perm,
2348 uint64_t shared_perm, Transaction *tran,
2349 Error **errp)
2350 {
2351 GLOBAL_STATE_CODE();
2352 if (!bs->drv) {
2353 return 0;
2354 }
2355
2356 if (bs->drv->bdrv_check_perm) {
2357 int ret = bs->drv->bdrv_check_perm(bs, perm, shared_perm, errp);
2358 if (ret < 0) {
2359 return ret;
2360 }
2361 }
2362
2363 if (tran) {
2364 tran_add(tran, &bdrv_drv_set_perm_drv, bs);
2365 }
2366
2367 return 0;
2368 }
2369
2370 typedef struct BdrvReplaceChildState {
2371 BdrvChild *child;
2372 BlockDriverState *old_bs;
2373 } BdrvReplaceChildState;
2374
2375 static void bdrv_replace_child_commit(void *opaque)
2376 {
2377 BdrvReplaceChildState *s = opaque;
2378 GLOBAL_STATE_CODE();
2379
2380 bdrv_unref(s->old_bs);
2381 }
2382
2383 static void bdrv_replace_child_abort(void *opaque)
2384 {
2385 BdrvReplaceChildState *s = opaque;
2386 BlockDriverState *new_bs = s->child->bs;
2387
2388 GLOBAL_STATE_CODE();
2389 /* old_bs reference is transparently moved from @s to @s->child */
2390 if (!s->child->bs) {
2391 /*
2392 * The parents were undrained when removing old_bs from the child. New
2393 * requests can't have been made, though, because the child was empty.
2394 *
2395 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2396 * undraining the parent in the first place. Once this is done, having
2397 * new_bs drained when calling bdrv_replace_child_tran() is not a
2398 * requirement any more.
2399 */
2400 bdrv_parent_drained_begin_single(s->child);
2401 assert(!bdrv_parent_drained_poll_single(s->child));
2402 }
2403 assert(s->child->quiesced_parent);
2404 bdrv_replace_child_noperm(s->child, s->old_bs);
2405 bdrv_unref(new_bs);
2406 }
2407
2408 static TransactionActionDrv bdrv_replace_child_drv = {
2409 .commit = bdrv_replace_child_commit,
2410 .abort = bdrv_replace_child_abort,
2411 .clean = g_free,
2412 };
2413
2414 /*
2415 * bdrv_replace_child_tran
2416 *
2417 * Note: real unref of old_bs is done only on commit.
2418 *
2419 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2420 * kept drained until the transaction is completed.
2421 *
2422 * The function doesn't update permissions, caller is responsible for this.
2423 */
2424 static void bdrv_replace_child_tran(BdrvChild *child, BlockDriverState *new_bs,
2425 Transaction *tran)
2426 {
2427 BdrvReplaceChildState *s = g_new(BdrvReplaceChildState, 1);
2428
2429 assert(child->quiesced_parent);
2430 assert(!new_bs || new_bs->quiesce_counter);
2431
2432 *s = (BdrvReplaceChildState) {
2433 .child = child,
2434 .old_bs = child->bs,
2435 };
2436 tran_add(tran, &bdrv_replace_child_drv, s);
2437
2438 if (new_bs) {
2439 bdrv_ref(new_bs);
2440 }
2441 bdrv_replace_child_noperm(child, new_bs);
2442 /* old_bs reference is transparently moved from @child to @s */
2443 }
2444
2445 /*
2446 * Refresh permissions in @bs subtree. The function is intended to be called
2447 * after some graph modification that was done without permission update.
2448 */
2449 static int bdrv_node_refresh_perm(BlockDriverState *bs, BlockReopenQueue *q,
2450 Transaction *tran, Error **errp)
2451 {
2452 BlockDriver *drv = bs->drv;
2453 BdrvChild *c;
2454 int ret;
2455 uint64_t cumulative_perms, cumulative_shared_perms;
2456 GLOBAL_STATE_CODE();
2457
2458 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2459
2460 /* Write permissions never work with read-only images */
2461 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2462 !bdrv_is_writable_after_reopen(bs, q))
2463 {
2464 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2465 error_setg(errp, "Block node is read-only");
2466 } else {
2467 error_setg(errp, "Read-only block node '%s' cannot support "
2468 "read-write users", bdrv_get_node_name(bs));
2469 }
2470
2471 return -EPERM;
2472 }
2473
2474 /*
2475 * Unaligned requests will automatically be aligned to bl.request_alignment
2476 * and without RESIZE we can't extend requests to write to space beyond the
2477 * end of the image, so it's required that the image size is aligned.
2478 */
2479 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2480 !(cumulative_perms & BLK_PERM_RESIZE))
2481 {
2482 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2483 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2484 "Image size is not a multiple of request "
2485 "alignment");
2486 return -EPERM;
2487 }
2488 }
2489
2490 /* Check this node */
2491 if (!drv) {
2492 return 0;
2493 }
2494
2495 ret = bdrv_drv_set_perm(bs, cumulative_perms, cumulative_shared_perms, tran,
2496 errp);
2497 if (ret < 0) {
2498 return ret;
2499 }
2500
2501 /* Drivers that never have children can omit .bdrv_child_perm() */
2502 if (!drv->bdrv_child_perm) {
2503 assert(QLIST_EMPTY(&bs->children));
2504 return 0;
2505 }
2506
2507 /* Check all children */
2508 QLIST_FOREACH(c, &bs->children, next) {
2509 uint64_t cur_perm, cur_shared;
2510
2511 bdrv_child_perm(bs, c->bs, c, c->role, q,
2512 cumulative_perms, cumulative_shared_perms,
2513 &cur_perm, &cur_shared);
2514 bdrv_child_set_perm(c, cur_perm, cur_shared, tran);
2515 }
2516
2517 return 0;
2518 }
2519
2520 /*
2521 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2522 * a topologically sorted subgraph.
2523 */
2524 static int bdrv_do_refresh_perms(GSList *list, BlockReopenQueue *q,
2525 Transaction *tran, Error **errp)
2526 {
2527 int ret;
2528 BlockDriverState *bs;
2529 GLOBAL_STATE_CODE();
2530
2531 for ( ; list; list = list->next) {
2532 bs = list->data;
2533
2534 if (bdrv_parent_perms_conflict(bs, errp)) {
2535 return -EINVAL;
2536 }
2537
2538 ret = bdrv_node_refresh_perm(bs, q, tran, errp);
2539 if (ret < 0) {
2540 return ret;
2541 }
2542 }
2543
2544 return 0;
2545 }
2546
2547 /*
2548 * @list is any list of nodes. List is completed by all subtrees and
2549 * topologically sorted. It's not a problem if some node occurs in the @list
2550 * several times.
2551 */
2552 static int bdrv_list_refresh_perms(GSList *list, BlockReopenQueue *q,
2553 Transaction *tran, Error **errp)
2554 {
2555 g_autoptr(GHashTable) found = g_hash_table_new(NULL, NULL);
2556 g_autoptr(GSList) refresh_list = NULL;
2557
2558 for ( ; list; list = list->next) {
2559 refresh_list = bdrv_topological_dfs(refresh_list, found, list->data);
2560 }
2561
2562 return bdrv_do_refresh_perms(refresh_list, q, tran, errp);
2563 }
2564
2565 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2566 uint64_t *shared_perm)
2567 {
2568 BdrvChild *c;
2569 uint64_t cumulative_perms = 0;
2570 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2571
2572 GLOBAL_STATE_CODE();
2573
2574 QLIST_FOREACH(c, &bs->parents, next_parent) {
2575 cumulative_perms |= c->perm;
2576 cumulative_shared_perms &= c->shared_perm;
2577 }
2578
2579 *perm = cumulative_perms;
2580 *shared_perm = cumulative_shared_perms;
2581 }
2582
2583 char *bdrv_perm_names(uint64_t perm)
2584 {
2585 struct perm_name {
2586 uint64_t perm;
2587 const char *name;
2588 } permissions[] = {
2589 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2590 { BLK_PERM_WRITE, "write" },
2591 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2592 { BLK_PERM_RESIZE, "resize" },
2593 { 0, NULL }
2594 };
2595
2596 GString *result = g_string_sized_new(30);
2597 struct perm_name *p;
2598
2599 for (p = permissions; p->name; p++) {
2600 if (perm & p->perm) {
2601 if (result->len > 0) {
2602 g_string_append(result, ", ");
2603 }
2604 g_string_append(result, p->name);
2605 }
2606 }
2607
2608 return g_string_free(result, FALSE);
2609 }
2610
2611
2612 /* @tran is allowed to be NULL. In this case no rollback is possible */
2613 static int bdrv_refresh_perms(BlockDriverState *bs, Transaction *tran,
2614 Error **errp)
2615 {
2616 int ret;
2617 Transaction *local_tran = NULL;
2618 g_autoptr(GSList) list = bdrv_topological_dfs(NULL, NULL, bs);
2619 GLOBAL_STATE_CODE();
2620
2621 if (!tran) {
2622 tran = local_tran = tran_new();
2623 }
2624
2625 ret = bdrv_do_refresh_perms(list, NULL, tran, errp);
2626
2627 if (local_tran) {
2628 tran_finalize(local_tran, ret);
2629 }
2630
2631 return ret;
2632 }
2633
2634 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2635 Error **errp)
2636 {
2637 Error *local_err = NULL;
2638 Transaction *tran = tran_new();
2639 int ret;
2640
2641 GLOBAL_STATE_CODE();
2642
2643 bdrv_child_set_perm(c, perm, shared, tran);
2644
2645 ret = bdrv_refresh_perms(c->bs, tran, &local_err);
2646
2647 tran_finalize(tran, ret);
2648
2649 if (ret < 0) {
2650 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2651 /* tighten permissions */
2652 error_propagate(errp, local_err);
2653 } else {
2654 /*
2655 * Our caller may intend to only loosen restrictions and
2656 * does not expect this function to fail. Errors are not
2657 * fatal in such a case, so we can just hide them from our
2658 * caller.
2659 */
2660 error_free(local_err);
2661 ret = 0;
2662 }
2663 }
2664
2665 return ret;
2666 }
2667
2668 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2669 {
2670 uint64_t parent_perms, parent_shared;
2671 uint64_t perms, shared;
2672
2673 GLOBAL_STATE_CODE();
2674
2675 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2676 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2677 parent_perms, parent_shared, &perms, &shared);
2678
2679 return bdrv_child_try_set_perm(c, perms, shared, errp);
2680 }
2681
2682 /*
2683 * Default implementation for .bdrv_child_perm() for block filters:
2684 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2685 * filtered child.
2686 */
2687 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2688 BdrvChildRole role,
2689 BlockReopenQueue *reopen_queue,
2690 uint64_t perm, uint64_t shared,
2691 uint64_t *nperm, uint64_t *nshared)
2692 {
2693 GLOBAL_STATE_CODE();
2694 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2695 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2696 }
2697
2698 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2699 BdrvChildRole role,
2700 BlockReopenQueue *reopen_queue,
2701 uint64_t perm, uint64_t shared,
2702 uint64_t *nperm, uint64_t *nshared)
2703 {
2704 assert(role & BDRV_CHILD_COW);
2705 GLOBAL_STATE_CODE();
2706
2707 /*
2708 * We want consistent read from backing files if the parent needs it.
2709 * No other operations are performed on backing files.
2710 */
2711 perm &= BLK_PERM_CONSISTENT_READ;
2712
2713 /*
2714 * If the parent can deal with changing data, we're okay with a
2715 * writable and resizable backing file.
2716 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2717 */
2718 if (shared & BLK_PERM_WRITE) {
2719 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2720 } else {
2721 shared = 0;
2722 }
2723
2724 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
2725
2726 if (bs->open_flags & BDRV_O_INACTIVE) {
2727 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2728 }
2729
2730 *nperm = perm;
2731 *nshared = shared;
2732 }
2733
2734 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2735 BdrvChildRole role,
2736 BlockReopenQueue *reopen_queue,
2737 uint64_t perm, uint64_t shared,
2738 uint64_t *nperm, uint64_t *nshared)
2739 {
2740 int flags;
2741
2742 GLOBAL_STATE_CODE();
2743 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2744
2745 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2746
2747 /*
2748 * Apart from the modifications below, the same permissions are
2749 * forwarded and left alone as for filters
2750 */
2751 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2752 perm, shared, &perm, &shared);
2753
2754 if (role & BDRV_CHILD_METADATA) {
2755 /* Format drivers may touch metadata even if the guest doesn't write */
2756 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2757 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2758 }
2759
2760 /*
2761 * bs->file always needs to be consistent because of the
2762 * metadata. We can never allow other users to resize or write
2763 * to it.
2764 */
2765 if (!(flags & BDRV_O_NO_IO)) {
2766 perm |= BLK_PERM_CONSISTENT_READ;
2767 }
2768 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2769 }
2770
2771 if (role & BDRV_CHILD_DATA) {
2772 /*
2773 * Technically, everything in this block is a subset of the
2774 * BDRV_CHILD_METADATA path taken above, and so this could
2775 * be an "else if" branch. However, that is not obvious, and
2776 * this function is not performance critical, therefore we let
2777 * this be an independent "if".
2778 */
2779
2780 /*
2781 * We cannot allow other users to resize the file because the
2782 * format driver might have some assumptions about the size
2783 * (e.g. because it is stored in metadata, or because the file
2784 * is split into fixed-size data files).
2785 */
2786 shared &= ~BLK_PERM_RESIZE;
2787
2788 /*
2789 * WRITE_UNCHANGED often cannot be performed as such on the
2790 * data file. For example, the qcow2 driver may still need to
2791 * write copied clusters on copy-on-read.
2792 */
2793 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2794 perm |= BLK_PERM_WRITE;
2795 }
2796
2797 /*
2798 * If the data file is written to, the format driver may
2799 * expect to be able to resize it by writing beyond the EOF.
2800 */
2801 if (perm & BLK_PERM_WRITE) {
2802 perm |= BLK_PERM_RESIZE;
2803 }
2804 }
2805
2806 if (bs->open_flags & BDRV_O_INACTIVE) {
2807 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2808 }
2809
2810 *nperm = perm;
2811 *nshared = shared;
2812 }
2813
2814 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2815 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2816 uint64_t perm, uint64_t shared,
2817 uint64_t *nperm, uint64_t *nshared)
2818 {
2819 GLOBAL_STATE_CODE();
2820 if (role & BDRV_CHILD_FILTERED) {
2821 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2822 BDRV_CHILD_COW)));
2823 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2824 perm, shared, nperm, nshared);
2825 } else if (role & BDRV_CHILD_COW) {
2826 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2827 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2828 perm, shared, nperm, nshared);
2829 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2830 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2831 perm, shared, nperm, nshared);
2832 } else {
2833 g_assert_not_reached();
2834 }
2835 }
2836
2837 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2838 {
2839 static const uint64_t permissions[] = {
2840 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2841 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2842 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2843 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2844 };
2845
2846 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2847 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2848
2849 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2850
2851 return permissions[qapi_perm];
2852 }
2853
2854 /*
2855 * Replaces the node that a BdrvChild points to without updating permissions.
2856 *
2857 * If @new_bs is non-NULL, the parent of @child must already be drained through
2858 * @child and the caller must hold the AioContext lock for @new_bs.
2859 */
2860 static void bdrv_replace_child_noperm(BdrvChild *child,
2861 BlockDriverState *new_bs)
2862 {
2863 BlockDriverState *old_bs = child->bs;
2864 int new_bs_quiesce_counter;
2865
2866 assert(!child->frozen);
2867
2868 /*
2869 * If we want to change the BdrvChild to point to a drained node as its new
2870 * child->bs, we need to make sure that its new parent is drained, too. In
2871 * other words, either child->quiesce_parent must already be true or we must
2872 * be able to set it and keep the parent's quiesce_counter consistent with
2873 * that, but without polling or starting new requests (this function
2874 * guarantees that it doesn't poll, and starting new requests would be
2875 * against the invariants of drain sections).
2876 *
2877 * To keep things simple, we pick the first option (child->quiesce_parent
2878 * must already be true). We also generalise the rule a bit to make it
2879 * easier to verify in callers and more likely to be covered in test cases:
2880 * The parent must be quiesced through this child even if new_bs isn't
2881 * currently drained.
2882 *
2883 * The only exception is for callers that always pass new_bs == NULL. In
2884 * this case, we obviously never need to consider the case of a drained
2885 * new_bs, so we can keep the callers simpler by allowing them not to drain
2886 * the parent.
2887 */
2888 assert(!new_bs || child->quiesced_parent);
2889 assert(old_bs != new_bs);
2890 GLOBAL_STATE_CODE();
2891
2892 if (old_bs && new_bs) {
2893 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2894 }
2895
2896 /* TODO Pull this up into the callers to avoid polling here */
2897 bdrv_graph_wrlock(new_bs);
2898 if (old_bs) {
2899 if (child->klass->detach) {
2900 child->klass->detach(child);
2901 }
2902 QLIST_REMOVE(child, next_parent);
2903 }
2904
2905 child->bs = new_bs;
2906
2907 if (new_bs) {
2908 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2909 if (child->klass->attach) {
2910 child->klass->attach(child);
2911 }
2912 }
2913 bdrv_graph_wrunlock();
2914
2915 /*
2916 * If the parent was drained through this BdrvChild previously, but new_bs
2917 * is not drained, allow requests to come in only after the new node has
2918 * been attached.
2919 */
2920 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2921 if (!new_bs_quiesce_counter && child->quiesced_parent) {
2922 bdrv_parent_drained_end_single(child);
2923 }
2924 }
2925
2926 /**
2927 * Free the given @child.
2928 *
2929 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2930 * unused (i.e. not in a children list).
2931 */
2932 static void bdrv_child_free(BdrvChild *child)
2933 {
2934 assert(!child->bs);
2935 GLOBAL_STATE_CODE();
2936 assert(!child->next.le_prev); /* not in children list */
2937
2938 g_free(child->name);
2939 g_free(child);
2940 }
2941
2942 typedef struct BdrvAttachChildCommonState {
2943 BdrvChild *child;
2944 AioContext *old_parent_ctx;
2945 AioContext *old_child_ctx;
2946 } BdrvAttachChildCommonState;
2947
2948 static void bdrv_attach_child_common_abort(void *opaque)
2949 {
2950 BdrvAttachChildCommonState *s = opaque;
2951 BlockDriverState *bs = s->child->bs;
2952
2953 GLOBAL_STATE_CODE();
2954 bdrv_replace_child_noperm(s->child, NULL);
2955
2956 if (bdrv_get_aio_context(bs) != s->old_child_ctx) {
2957 bdrv_try_change_aio_context(bs, s->old_child_ctx, NULL, &error_abort);
2958 }
2959
2960 if (bdrv_child_get_parent_aio_context(s->child) != s->old_parent_ctx) {
2961 Transaction *tran;
2962 GHashTable *visited;
2963 bool ret;
2964
2965 tran = tran_new();
2966
2967 /* No need to visit `child`, because it has been detached already */
2968 visited = g_hash_table_new(NULL, NULL);
2969 ret = s->child->klass->change_aio_ctx(s->child, s->old_parent_ctx,
2970 visited, tran, &error_abort);
2971 g_hash_table_destroy(visited);
2972
2973 /* transaction is supposed to always succeed */
2974 assert(ret == true);
2975 tran_commit(tran);
2976 }
2977
2978 bdrv_unref(bs);
2979 bdrv_child_free(s->child);
2980 }
2981
2982 static TransactionActionDrv bdrv_attach_child_common_drv = {
2983 .abort = bdrv_attach_child_common_abort,
2984 .clean = g_free,
2985 };
2986
2987 /*
2988 * Common part of attaching bdrv child to bs or to blk or to job
2989 *
2990 * Function doesn't update permissions, caller is responsible for this.
2991 *
2992 * Returns new created child.
2993 *
2994 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
2995 * @child_bs can move to a different AioContext in this function. Callers must
2996 * make sure that their AioContext locking is still correct after this.
2997 */
2998 static BdrvChild *bdrv_attach_child_common(BlockDriverState *child_bs,
2999 const char *child_name,
3000 const BdrvChildClass *child_class,
3001 BdrvChildRole child_role,
3002 uint64_t perm, uint64_t shared_perm,
3003 void *opaque,
3004 Transaction *tran, Error **errp)
3005 {
3006 BdrvChild *new_child;
3007 AioContext *parent_ctx, *new_child_ctx;
3008 AioContext *child_ctx = bdrv_get_aio_context(child_bs);
3009
3010 assert(child_class->get_parent_desc);
3011 GLOBAL_STATE_CODE();
3012
3013 new_child = g_new(BdrvChild, 1);
3014 *new_child = (BdrvChild) {
3015 .bs = NULL,
3016 .name = g_strdup(child_name),
3017 .klass = child_class,
3018 .role = child_role,
3019 .perm = perm,
3020 .shared_perm = shared_perm,
3021 .opaque = opaque,
3022 };
3023
3024 /*
3025 * If the AioContexts don't match, first try to move the subtree of
3026 * child_bs into the AioContext of the new parent. If this doesn't work,
3027 * try moving the parent into the AioContext of child_bs instead.
3028 */
3029 parent_ctx = bdrv_child_get_parent_aio_context(new_child);
3030 if (child_ctx != parent_ctx) {
3031 Error *local_err = NULL;
3032 int ret = bdrv_try_change_aio_context(child_bs, parent_ctx, NULL,
3033 &local_err);
3034
3035 if (ret < 0 && child_class->change_aio_ctx) {
3036 Transaction *tran = tran_new();
3037 GHashTable *visited = g_hash_table_new(NULL, NULL);
3038 bool ret_child;
3039
3040 g_hash_table_add(visited, new_child);
3041 ret_child = child_class->change_aio_ctx(new_child, child_ctx,
3042 visited, tran, NULL);
3043 if (ret_child == true) {
3044 error_free(local_err);
3045 ret = 0;
3046 }
3047 tran_finalize(tran, ret_child == true ? 0 : -1);
3048 g_hash_table_destroy(visited);
3049 }
3050
3051 if (ret < 0) {
3052 error_propagate(errp, local_err);
3053 bdrv_child_free(new_child);
3054 return NULL;
3055 }
3056 }
3057
3058 new_child_ctx = bdrv_get_aio_context(child_bs);
3059 if (new_child_ctx != child_ctx) {
3060 aio_context_release(child_ctx);
3061 aio_context_acquire(new_child_ctx);
3062 }
3063
3064 bdrv_ref(child_bs);
3065 /*
3066 * Let every new BdrvChild start with a drained parent. Inserting the child
3067 * in the graph with bdrv_replace_child_noperm() will undrain it if
3068 * @child_bs is not drained.
3069 *
3070 * The child was only just created and is not yet visible in global state
3071 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3072 * could have sent requests and polling is not necessary.
3073 *
3074 * Note that this means that the parent isn't fully drained yet, we only
3075 * stop new requests from coming in. This is fine, we don't care about the
3076 * old requests here, they are not for this child. If another place enters a
3077 * drain section for the same parent, but wants it to be fully quiesced, it
3078 * will not run most of the the code in .drained_begin() again (which is not
3079 * a problem, we already did this), but it will still poll until the parent
3080 * is fully quiesced, so it will not be negatively affected either.
3081 */
3082 bdrv_parent_drained_begin_single(new_child);
3083 bdrv_replace_child_noperm(new_child, child_bs);
3084
3085 BdrvAttachChildCommonState *s = g_new(BdrvAttachChildCommonState, 1);
3086 *s = (BdrvAttachChildCommonState) {
3087 .child = new_child,
3088 .old_parent_ctx = parent_ctx,
3089 .old_child_ctx = child_ctx,
3090 };
3091 tran_add(tran, &bdrv_attach_child_common_drv, s);
3092
3093 if (new_child_ctx != child_ctx) {
3094 aio_context_release(new_child_ctx);
3095 aio_context_acquire(child_ctx);
3096 }
3097
3098 return new_child;
3099 }
3100
3101 /*
3102 * Function doesn't update permissions, caller is responsible for this.
3103 *
3104 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3105 * @child_bs can move to a different AioContext in this function. Callers must
3106 * make sure that their AioContext locking is still correct after this.
3107 */
3108 static BdrvChild *bdrv_attach_child_noperm(BlockDriverState *parent_bs,
3109 BlockDriverState *child_bs,
3110 const char *child_name,
3111 const BdrvChildClass *child_class,
3112 BdrvChildRole child_role,
3113 Transaction *tran,
3114 Error **errp)
3115 {
3116 uint64_t perm, shared_perm;
3117
3118 assert(parent_bs->drv);
3119 GLOBAL_STATE_CODE();
3120
3121 if (bdrv_recurse_has_child(child_bs, parent_bs)) {
3122 error_setg(errp, "Making '%s' a %s child of '%s' would create a cycle",
3123 child_bs->node_name, child_name, parent_bs->node_name);
3124 return NULL;
3125 }
3126
3127 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
3128 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
3129 perm, shared_perm, &perm, &shared_perm);
3130
3131 return bdrv_attach_child_common(child_bs, child_name, child_class,
3132 child_role, perm, shared_perm, parent_bs,
3133 tran, errp);
3134 }
3135
3136 /*
3137 * This function steals the reference to child_bs from the caller.
3138 * That reference is later dropped by bdrv_root_unref_child().
3139 *
3140 * On failure NULL is returned, errp is set and the reference to
3141 * child_bs is also dropped.
3142 *
3143 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3144 * (unless @child_bs is already in @ctx).
3145 */
3146 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
3147 const char *child_name,
3148 const BdrvChildClass *child_class,
3149 BdrvChildRole child_role,
3150 uint64_t perm, uint64_t shared_perm,
3151 void *opaque, Error **errp)
3152 {
3153 int ret;
3154 BdrvChild *child;
3155 Transaction *tran = tran_new();
3156
3157 GLOBAL_STATE_CODE();
3158
3159 child = bdrv_attach_child_common(child_bs, child_name, child_class,
3160 child_role, perm, shared_perm, opaque,
3161 tran, errp);
3162 if (!child) {
3163 ret = -EINVAL;
3164 goto out;
3165 }
3166
3167 ret = bdrv_refresh_perms(child_bs, tran, errp);
3168
3169 out:
3170 tran_finalize(tran, ret);
3171
3172 bdrv_unref(child_bs);
3173
3174 return ret < 0 ? NULL : child;
3175 }
3176
3177 /*
3178 * This function transfers the reference to child_bs from the caller
3179 * to parent_bs. That reference is later dropped by parent_bs on
3180 * bdrv_close() or if someone calls bdrv_unref_child().
3181 *
3182 * On failure NULL is returned, errp is set and the reference to
3183 * child_bs is also dropped.
3184 *
3185 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3186 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3187 */
3188 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
3189 BlockDriverState *child_bs,
3190 const char *child_name,
3191 const BdrvChildClass *child_class,
3192 BdrvChildRole child_role,
3193 Error **errp)
3194 {
3195 int ret;
3196 BdrvChild *child;
3197 Transaction *tran = tran_new();
3198
3199 GLOBAL_STATE_CODE();
3200
3201 child = bdrv_attach_child_noperm(parent_bs, child_bs, child_name,
3202 child_class, child_role, tran, errp);
3203 if (!child) {
3204 ret = -EINVAL;
3205 goto out;
3206 }
3207
3208 ret = bdrv_refresh_perms(parent_bs, tran, errp);
3209 if (ret < 0) {
3210 goto out;
3211 }
3212
3213 out:
3214 tran_finalize(tran, ret);
3215
3216 bdrv_unref(child_bs);
3217
3218 return ret < 0 ? NULL : child;
3219 }
3220
3221 /* Callers must ensure that child->frozen is false. */
3222 void bdrv_root_unref_child(BdrvChild *child)
3223 {
3224 BlockDriverState *child_bs = child->bs;
3225
3226 GLOBAL_STATE_CODE();
3227 bdrv_replace_child_noperm(child, NULL);
3228 bdrv_child_free(child);
3229
3230 if (child_bs) {
3231 /*
3232 * Update permissions for old node. We're just taking a parent away, so
3233 * we're loosening restrictions. Errors of permission update are not
3234 * fatal in this case, ignore them.
3235 */
3236 bdrv_refresh_perms(child_bs, NULL, NULL);
3237
3238 /*
3239 * When the parent requiring a non-default AioContext is removed, the
3240 * node moves back to the main AioContext
3241 */
3242 bdrv_try_change_aio_context(child_bs, qemu_get_aio_context(), NULL,
3243 NULL);
3244 }
3245
3246 bdrv_unref(child_bs);
3247 }
3248
3249 typedef struct BdrvSetInheritsFrom {
3250 BlockDriverState *bs;
3251 BlockDriverState *old_inherits_from;
3252 } BdrvSetInheritsFrom;
3253
3254 static void bdrv_set_inherits_from_abort(void *opaque)
3255 {
3256 BdrvSetInheritsFrom *s = opaque;
3257
3258 s->bs->inherits_from = s->old_inherits_from;
3259 }
3260
3261 static TransactionActionDrv bdrv_set_inherits_from_drv = {
3262 .abort = bdrv_set_inherits_from_abort,
3263 .clean = g_free,
3264 };
3265
3266 /* @tran is allowed to be NULL. In this case no rollback is possible */
3267 static void bdrv_set_inherits_from(BlockDriverState *bs,
3268 BlockDriverState *new_inherits_from,
3269 Transaction *tran)
3270 {
3271 if (tran) {
3272 BdrvSetInheritsFrom *s = g_new(BdrvSetInheritsFrom, 1);
3273
3274 *s = (BdrvSetInheritsFrom) {
3275 .bs = bs,
3276 .old_inherits_from = bs->inherits_from,
3277 };
3278
3279 tran_add(tran, &bdrv_set_inherits_from_drv, s);
3280 }
3281
3282 bs->inherits_from = new_inherits_from;
3283 }
3284
3285 /**
3286 * Clear all inherits_from pointers from children and grandchildren of
3287 * @root that point to @root, where necessary.
3288 * @tran is allowed to be NULL. In this case no rollback is possible
3289 */
3290 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child,
3291 Transaction *tran)
3292 {
3293 BdrvChild *c;
3294
3295 if (child->bs->inherits_from == root) {
3296 /*
3297 * Remove inherits_from only when the last reference between root and
3298 * child->bs goes away.
3299 */
3300 QLIST_FOREACH(c, &root->children, next) {
3301 if (c != child && c->bs == child->bs) {
3302 break;
3303 }
3304 }
3305 if (c == NULL) {
3306 bdrv_set_inherits_from(child->bs, NULL, tran);
3307 }
3308 }
3309
3310 QLIST_FOREACH(c, &child->bs->children, next) {
3311 bdrv_unset_inherits_from(root, c, tran);
3312 }
3313 }
3314
3315 /* Callers must ensure that child->frozen is false. */
3316 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
3317 {
3318 GLOBAL_STATE_CODE();
3319 if (child == NULL) {
3320 return;
3321 }
3322
3323 bdrv_unset_inherits_from(parent, child, NULL);
3324 bdrv_root_unref_child(child);
3325 }
3326
3327
3328 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
3329 {
3330 BdrvChild *c;
3331 GLOBAL_STATE_CODE();
3332 QLIST_FOREACH(c, &bs->parents, next_parent) {
3333 if (c->klass->change_media) {
3334 c->klass->change_media(c, load);
3335 }
3336 }
3337 }
3338
3339 /* Return true if you can reach parent going through child->inherits_from
3340 * recursively. If parent or child are NULL, return false */
3341 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
3342 BlockDriverState *parent)
3343 {
3344 while (child && child != parent) {
3345 child = child->inherits_from;
3346 }
3347
3348 return child != NULL;
3349 }
3350
3351 /*
3352 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3353 * mostly used for COW backing children (role = COW), but also for
3354 * filtered children (role = FILTERED | PRIMARY).
3355 */
3356 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
3357 {
3358 if (bs->drv && bs->drv->is_filter) {
3359 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3360 } else {
3361 return BDRV_CHILD_COW;
3362 }
3363 }
3364
3365 /*
3366 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3367 * callers which don't need their own reference any more must call bdrv_unref().
3368 *
3369 * Function doesn't update permissions, caller is responsible for this.
3370 *
3371 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3372 * @child_bs can move to a different AioContext in this function. Callers must
3373 * make sure that their AioContext locking is still correct after this.
3374 */
3375 static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
3376 BlockDriverState *child_bs,
3377 bool is_backing,
3378 Transaction *tran, Error **errp)
3379 {
3380 bool update_inherits_from =
3381 bdrv_inherits_from_recursive(child_bs, parent_bs);
3382 BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
3383 BdrvChildRole role;
3384
3385 GLOBAL_STATE_CODE();
3386
3387 if (!parent_bs->drv) {
3388 /*
3389 * Node without drv is an object without a class :/. TODO: finally fix
3390 * qcow2 driver to never clear bs->drv and implement format corruption
3391 * handling in other way.
3392 */
3393 error_setg(errp, "Node corrupted");
3394 return -EINVAL;
3395 }
3396
3397 if (child && child->frozen) {
3398 error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
3399 child->name, parent_bs->node_name, child->bs->node_name);
3400 return -EPERM;
3401 }
3402
3403 if (is_backing && !parent_bs->drv->is_filter &&
3404 !parent_bs->drv->supports_backing)
3405 {
3406 error_setg(errp, "Driver '%s' of node '%s' does not support backing "
3407 "files", parent_bs->drv->format_name, parent_bs->node_name);
3408 return -EINVAL;
3409 }
3410
3411 if (parent_bs->drv->is_filter) {
3412 role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
3413 } else if (is_backing) {
3414 role = BDRV_CHILD_COW;
3415 } else {
3416 /*
3417 * We only can use same role as it is in existing child. We don't have
3418 * infrastructure to determine role of file child in generic way
3419 */
3420 if (!child) {
3421 error_setg(errp, "Cannot set file child to format node without "
3422 "file child");
3423 return -EINVAL;
3424 }
3425 role = child->role;
3426 }
3427
3428 if (child) {
3429 bdrv_unset_inherits_from(parent_bs, child, tran);
3430 bdrv_remove_child(child, tran);
3431 }
3432
3433 if (!child_bs) {
3434 goto out;
3435 }
3436
3437 child = bdrv_attach_child_noperm(parent_bs, child_bs,
3438 is_backing ? "backing" : "file",
3439 &child_of_bds, role,
3440 tran, errp);
3441 if (!child) {
3442 return -EINVAL;
3443 }
3444
3445
3446 /*
3447 * If inherits_from pointed recursively to bs then let's update it to
3448 * point directly to bs (else it will become NULL).
3449 */
3450 if (update_inherits_from) {
3451 bdrv_set_inherits_from(child_bs, parent_bs, tran);
3452 }
3453
3454 out:
3455 bdrv_graph_rdlock_main_loop();
3456 bdrv_refresh_limits(parent_bs, tran, NULL);
3457 bdrv_graph_rdunlock_main_loop();
3458
3459 return 0;
3460 }
3461
3462 /*
3463 * The caller must hold the AioContext lock for @backing_hd. Both @bs and
3464 * @backing_hd can move to a different AioContext in this function. Callers must
3465 * make sure that their AioContext locking is still correct after this.
3466 */
3467 static int bdrv_set_backing_noperm(BlockDriverState *bs,
3468 BlockDriverState *backing_hd,
3469 Transaction *tran, Error **errp)
3470 {
3471 GLOBAL_STATE_CODE();
3472 return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
3473 }
3474
3475 int bdrv_set_backing_hd_drained(BlockDriverState *bs,
3476 BlockDriverState *backing_hd,
3477 Error **errp)
3478 {
3479 int ret;
3480 Transaction *tran = tran_new();
3481
3482 GLOBAL_STATE_CODE();
3483 assert(bs->quiesce_counter > 0);
3484
3485 ret = bdrv_set_backing_noperm(bs, backing_hd, tran, errp);
3486 if (ret < 0) {
3487 goto out;
3488 }
3489
3490 ret = bdrv_refresh_perms(bs, tran, errp);
3491 out:
3492 tran_finalize(tran, ret);
3493 return ret;
3494 }
3495
3496 int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
3497 Error **errp)
3498 {
3499 int ret;
3500 GLOBAL_STATE_CODE();
3501
3502 bdrv_drained_begin(bs);
3503 ret = bdrv_set_backing_hd_drained(bs, backing_hd, errp);
3504 bdrv_drained_end(bs);
3505
3506 return ret;
3507 }
3508
3509 /*
3510 * Opens the backing file for a BlockDriverState if not yet open
3511 *
3512 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3513 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3514 * itself, all options starting with "${bdref_key}." are considered part of the
3515 * BlockdevRef.
3516 *
3517 * The caller must hold the main AioContext lock.
3518 *
3519 * TODO Can this be unified with bdrv_open_image()?
3520 */
3521 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
3522 const char *bdref_key, Error **errp)
3523 {
3524 char *backing_filename = NULL;
3525 char *bdref_key_dot;
3526 const char *reference = NULL;
3527 int ret = 0;
3528 bool implicit_backing = false;
3529 BlockDriverState *backing_hd;
3530 AioContext *backing_hd_ctx;
3531 QDict *options;
3532 QDict *tmp_parent_options = NULL;
3533 Error *local_err = NULL;
3534
3535 GLOBAL_STATE_CODE();
3536
3537 if (bs->backing != NULL) {
3538 goto free_exit;
3539 }
3540
3541 /* NULL means an empty set of options */
3542 if (parent_options == NULL) {
3543 tmp_parent_options = qdict_new();
3544 parent_options = tmp_parent_options;
3545 }
3546
3547 bs->open_flags &= ~BDRV_O_NO_BACKING;
3548
3549 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3550 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
3551 g_free(bdref_key_dot);
3552
3553 /*
3554 * Caution: while qdict_get_try_str() is fine, getting non-string
3555 * types would require more care. When @parent_options come from
3556 * -blockdev or blockdev_add, its members are typed according to
3557 * the QAPI schema, but when they come from -drive, they're all
3558 * QString.
3559 */
3560 reference = qdict_get_try_str(parent_options, bdref_key);
3561 if (reference || qdict_haskey(options, "file.filename")) {
3562 /* keep backing_filename NULL */
3563 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
3564 qobject_unref(options);
3565 goto free_exit;
3566 } else {
3567 if (qdict_size(options) == 0) {
3568 /* If the user specifies options that do not modify the
3569 * backing file's behavior, we might still consider it the
3570 * implicit backing file. But it's easier this way, and
3571 * just specifying some of the backing BDS's options is
3572 * only possible with -drive anyway (otherwise the QAPI
3573 * schema forces the user to specify everything). */
3574 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
3575 }
3576
3577 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
3578 if (local_err) {
3579 ret = -EINVAL;
3580 error_propagate(errp, local_err);
3581 qobject_unref(options);
3582 goto free_exit;
3583 }
3584 }
3585
3586 if (!bs->drv || !bs->drv->supports_backing) {
3587 ret = -EINVAL;
3588 error_setg(errp, "Driver doesn't support backing files");
3589 qobject_unref(options);
3590 goto free_exit;
3591 }
3592
3593 if (!reference &&
3594 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
3595 qdict_put_str(options, "driver", bs->backing_format);
3596 }
3597
3598 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
3599 &child_of_bds, bdrv_backing_role(bs), errp);
3600 if (!backing_hd) {
3601 bs->open_flags |= BDRV_O_NO_BACKING;
3602 error_prepend(errp, "Could not open backing file: ");
3603 ret = -EINVAL;
3604 goto free_exit;
3605 }
3606
3607 if (implicit_backing) {
3608 bdrv_refresh_filename(backing_hd);
3609 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3610 backing_hd->filename);
3611 }
3612
3613 /* Hook up the backing file link; drop our reference, bs owns the
3614 * backing_hd reference now */
3615 backing_hd_ctx = bdrv_get_aio_context(backing_hd);
3616 aio_context_acquire(backing_hd_ctx);
3617 ret = bdrv_set_backing_hd(bs, backing_hd, errp);
3618 bdrv_unref(backing_hd);
3619 aio_context_release(backing_hd_ctx);
3620
3621 if (ret < 0) {
3622 goto free_exit;
3623 }
3624
3625 qdict_del(parent_options, bdref_key);
3626
3627 free_exit:
3628 g_free(backing_filename);
3629 qobject_unref(tmp_parent_options);
3630 return ret;
3631 }
3632
3633 static BlockDriverState *
3634 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3635 BlockDriverState *parent, const BdrvChildClass *child_class,
3636 BdrvChildRole child_role, bool allow_none, Error **errp)
3637 {
3638 BlockDriverState *bs = NULL;
3639 QDict *image_options;
3640 char *bdref_key_dot;
3641 const char *reference;
3642
3643 assert(child_class != NULL);
3644
3645 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3646 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3647 g_free(bdref_key_dot);
3648
3649 /*
3650 * Caution: while qdict_get_try_str() is fine, getting non-string
3651 * types would require more care. When @options come from
3652 * -blockdev or blockdev_add, its members are typed according to
3653 * the QAPI schema, but when they come from -drive, they're all
3654 * QString.
3655 */
3656 reference = qdict_get_try_str(options, bdref_key);
3657 if (!filename && !reference && !qdict_size(image_options)) {
3658 if (!allow_none) {
3659 error_setg(errp, "A block device must be specified for \"%s\"",
3660 bdref_key);
3661 }
3662 qobject_unref(image_options);
3663 goto done;
3664 }
3665
3666 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3667 parent, child_class, child_role, errp);
3668 if (!bs) {
3669 goto done;
3670 }
3671
3672 done:
3673 qdict_del(options, bdref_key);
3674 return bs;
3675 }
3676
3677 /*
3678 * Opens a disk image whose options are given as BlockdevRef in another block
3679 * device's options.
3680 *
3681 * If allow_none is true, no image will be opened if filename is false and no
3682 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3683 *
3684 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3685 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3686 * itself, all options starting with "${bdref_key}." are considered part of the
3687 * BlockdevRef.
3688 *
3689 * The BlockdevRef will be removed from the options QDict.
3690 *
3691 * The caller must hold the lock of the main AioContext and no other AioContext.
3692 * @parent can move to a different AioContext in this function. Callers must
3693 * make sure that their AioContext locking is still correct after this.
3694 */
3695 BdrvChild *bdrv_open_child(const char *filename,
3696 QDict *options, const char *bdref_key,
3697 BlockDriverState *parent,
3698 const BdrvChildClass *child_class,
3699 BdrvChildRole child_role,
3700 bool allow_none, Error **errp)
3701 {
3702 BlockDriverState *bs;
3703 BdrvChild *child;
3704 AioContext *ctx;
3705
3706 GLOBAL_STATE_CODE();
3707
3708 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3709 child_role, allow_none, errp);
3710 if (bs == NULL) {
3711 return NULL;
3712 }
3713
3714 ctx = bdrv_get_aio_context(bs);
3715 aio_context_acquire(ctx);
3716 child = bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3717 errp);
3718 aio_context_release(ctx);
3719
3720 return child;
3721 }
3722
3723 /*
3724 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3725 *
3726 * The caller must hold the lock of the main AioContext and no other AioContext.
3727 * @parent can move to a different AioContext in this function. Callers must
3728 * make sure that their AioContext locking is still correct after this.
3729 */
3730 int bdrv_open_file_child(const char *filename,
3731 QDict *options, const char *bdref_key,
3732 BlockDriverState *parent, Error **errp)
3733 {
3734 BdrvChildRole role;
3735
3736 /* commit_top and mirror_top don't use this function */
3737 assert(!parent->drv->filtered_child_is_backing);
3738 role = parent->drv->is_filter ?
3739 (BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY) : BDRV_CHILD_IMAGE;
3740
3741 if (!bdrv_open_child(filename, options, bdref_key, parent,
3742 &child_of_bds, role, false, errp))
3743 {
3744 return -EINVAL;
3745 }
3746
3747 return 0;
3748 }
3749
3750 /*
3751 * TODO Future callers may need to specify parent/child_class in order for
3752 * option inheritance to work. Existing callers use it for the root node.
3753 */
3754 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3755 {
3756 BlockDriverState *bs = NULL;
3757 QObject *obj = NULL;
3758 QDict *qdict = NULL;
3759 const char *reference = NULL;
3760 Visitor *v = NULL;
3761
3762 GLOBAL_STATE_CODE();
3763
3764 if (ref->type == QTYPE_QSTRING) {
3765 reference = ref->u.reference;
3766 } else {
3767 BlockdevOptions *options = &ref->u.definition;
3768 assert(ref->type == QTYPE_QDICT);
3769
3770 v = qobject_output_visitor_new(&obj);
3771 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3772 visit_complete(v, &obj);
3773
3774 qdict = qobject_to(QDict, obj);
3775 qdict_flatten(qdict);
3776
3777 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3778 * compatibility with other callers) rather than what we want as the
3779 * real defaults. Apply the defaults here instead. */
3780 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3781 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3782 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3783 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3784
3785 }
3786
3787 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3788 obj = NULL;
3789 qobject_unref(obj);
3790 visit_free(v);
3791 return bs;
3792 }
3793
3794 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3795 int flags,
3796 QDict *snapshot_options,
3797 Error **errp)
3798 {
3799 g_autofree char *tmp_filename = NULL;
3800 int64_t total_size;
3801 QemuOpts *opts = NULL;
3802 BlockDriverState *bs_snapshot = NULL;
3803 AioContext *ctx = bdrv_get_aio_context(bs);
3804 int ret;
3805
3806 GLOBAL_STATE_CODE();
3807
3808 /* if snapshot, we create a temporary backing file and open it
3809 instead of opening 'filename' directly */
3810
3811 /* Get the required size from the image */
3812 aio_context_acquire(ctx);
3813 total_size = bdrv_getlength(bs);
3814 aio_context_release(ctx);
3815
3816 if (total_size < 0) {
3817 error_setg_errno(errp, -total_size, "Could not get image size");
3818 goto out;
3819 }
3820
3821 /* Create the temporary image */
3822 tmp_filename = create_tmp_file(errp);
3823 if (!tmp_filename) {
3824 goto out;
3825 }
3826
3827 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3828 &error_abort);
3829 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3830 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3831 qemu_opts_del(opts);
3832 if (ret < 0) {
3833 error_prepend(errp, "Could not create temporary overlay '%s': ",
3834 tmp_filename);
3835 goto out;
3836 }
3837
3838 /* Prepare options QDict for the temporary file */
3839 qdict_put_str(snapshot_options, "file.driver", "file");
3840 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3841 qdict_put_str(snapshot_options, "driver", "qcow2");
3842
3843 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3844 snapshot_options = NULL;
3845 if (!bs_snapshot) {
3846 goto out;
3847 }
3848
3849 aio_context_acquire(ctx);
3850 ret = bdrv_append(bs_snapshot, bs, errp);
3851 aio_context_release(ctx);
3852
3853 if (ret < 0) {
3854 bs_snapshot = NULL;
3855 goto out;
3856 }
3857
3858 out:
3859 qobject_unref(snapshot_options);
3860 return bs_snapshot;
3861 }
3862
3863 /*
3864 * Opens a disk image (raw, qcow2, vmdk, ...)
3865 *
3866 * options is a QDict of options to pass to the block drivers, or NULL for an
3867 * empty set of options. The reference to the QDict belongs to the block layer
3868 * after the call (even on failure), so if the caller intends to reuse the
3869 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3870 *
3871 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3872 * If it is not NULL, the referenced BDS will be reused.
3873 *
3874 * The reference parameter may be used to specify an existing block device which
3875 * should be opened. If specified, neither options nor a filename may be given,
3876 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3877 *
3878 * The caller must always hold the main AioContext lock.
3879 */
3880 static BlockDriverState * no_coroutine_fn
3881 bdrv_open_inherit(const char *filename, const char *reference, QDict *options,
3882 int flags, BlockDriverState *parent,
3883 const BdrvChildClass *child_class, BdrvChildRole child_role,
3884 Error **errp)
3885 {
3886 int ret;
3887 BlockBackend *file = NULL;
3888 BlockDriverState *bs;
3889 BlockDriver *drv = NULL;
3890 BdrvChild *child;
3891 const char *drvname;
3892 const char *backing;
3893 Error *local_err = NULL;
3894 QDict *snapshot_options = NULL;
3895 int snapshot_flags = 0;
3896 AioContext *ctx = qemu_get_aio_context();
3897
3898 assert(!child_class || !flags);
3899 assert(!child_class == !parent);
3900 GLOBAL_STATE_CODE();
3901 assert(!qemu_in_coroutine());
3902
3903 if (reference) {
3904 bool options_non_empty = options ? qdict_size(options) : false;
3905 qobject_unref(options);
3906
3907 if (filename || options_non_empty) {
3908 error_setg(errp, "Cannot reference an existing block device with "
3909 "additional options or a new filename");
3910 return NULL;
3911 }
3912
3913 bs = bdrv_lookup_bs(reference, reference, errp);
3914 if (!bs) {
3915 return NULL;
3916 }
3917
3918 bdrv_ref(bs);
3919 return bs;
3920 }
3921
3922 bs = bdrv_new();
3923
3924 /* NULL means an empty set of options */
3925 if (options == NULL) {
3926 options = qdict_new();
3927 }
3928
3929 /* json: syntax counts as explicit options, as if in the QDict */
3930 parse_json_protocol(options, &filename, &local_err);
3931 if (local_err) {
3932 goto fail;
3933 }
3934
3935 bs->explicit_options = qdict_clone_shallow(options);
3936
3937 if (child_class) {
3938 bool parent_is_format;
3939
3940 if (parent->drv) {
3941 parent_is_format = parent->drv->is_format;
3942 } else {
3943 /*
3944 * parent->drv is not set yet because this node is opened for
3945 * (potential) format probing. That means that @parent is going
3946 * to be a format node.
3947 */
3948 parent_is_format = true;
3949 }
3950
3951 bs->inherits_from = parent;
3952 child_class->inherit_options(child_role, parent_is_format,
3953 &flags, options,
3954 parent->open_flags, parent->options);
3955 }
3956
3957 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3958 if (ret < 0) {
3959 goto fail;
3960 }
3961
3962 /*
3963 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3964 * Caution: getting a boolean member of @options requires care.
3965 * When @options come from -blockdev or blockdev_add, members are
3966 * typed according to the QAPI schema, but when they come from
3967 * -drive, they're all QString.
3968 */
3969 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3970 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3971 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3972 } else {
3973 flags &= ~BDRV_O_RDWR;
3974 }
3975
3976 if (flags & BDRV_O_SNAPSHOT) {
3977 snapshot_options = qdict_new();
3978 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3979 flags, options);
3980 /* Let bdrv_backing_options() override "read-only" */
3981 qdict_del(options, BDRV_OPT_READ_ONLY);
3982 bdrv_inherited_options(BDRV_CHILD_COW, true,
3983 &flags, options, flags, options);
3984 }
3985
3986 bs->open_flags = flags;
3987 bs->options = options;
3988 options = qdict_clone_shallow(options);
3989
3990 /* Find the right image format driver */
3991 /* See cautionary note on accessing @options above */
3992 drvname = qdict_get_try_str(options, "driver");
3993 if (drvname) {
3994 drv = bdrv_find_format(drvname);
3995 if (!drv) {
3996 error_setg(errp, "Unknown driver: '%s'", drvname);
3997 goto fail;
3998 }
3999 }
4000
4001 assert(drvname || !(flags & BDRV_O_PROTOCOL));
4002
4003 /* See cautionary note on accessing @options above */
4004 backing = qdict_get_try_str(options, "backing");
4005 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
4006 (backing && *backing == '\0'))
4007 {
4008 if (backing) {
4009 warn_report("Use of \"backing\": \"\" is deprecated; "
4010 "use \"backing\": null instead");
4011 }
4012 flags |= BDRV_O_NO_BACKING;
4013 qdict_del(bs->explicit_options, "backing");
4014 qdict_del(bs->options, "backing");
4015 qdict_del(options, "backing");
4016 }
4017
4018 /* Open image file without format layer. This BlockBackend is only used for
4019 * probing, the block drivers will do their own bdrv_open_child() for the
4020 * same BDS, which is why we put the node name back into options. */
4021 if ((flags & BDRV_O_PROTOCOL) == 0) {
4022 BlockDriverState *file_bs;
4023
4024 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
4025 &child_of_bds, BDRV_CHILD_IMAGE,
4026 true, &local_err);
4027 if (local_err) {
4028 goto fail;
4029 }
4030 if (file_bs != NULL) {
4031 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
4032 * looking at the header to guess the image format. This works even
4033 * in cases where a guest would not see a consistent state. */
4034 ctx = bdrv_get_aio_context(file_bs);
4035 aio_context_acquire(ctx);
4036 file = blk_new(ctx, 0, BLK_PERM_ALL);
4037 blk_insert_bs(file, file_bs, &local_err);
4038 bdrv_unref(file_bs);
4039 aio_context_release(ctx);
4040
4041 if (local_err) {
4042 goto fail;
4043 }
4044
4045 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
4046 }
4047 }
4048
4049 /* Image format probing */
4050 bs->probed = !drv;
4051 if (!drv && file) {
4052 ret = find_image_format(file, filename, &drv, &local_err);
4053 if (ret < 0) {
4054 goto fail;
4055 }
4056 /*
4057 * This option update would logically belong in bdrv_fill_options(),
4058 * but we first need to open bs->file for the probing to work, while
4059 * opening bs->file already requires the (mostly) final set of options
4060 * so that cache mode etc. can be inherited.
4061 *
4062 * Adding the driver later is somewhat ugly, but it's not an option
4063 * that would ever be inherited, so it's correct. We just need to make
4064 * sure to update both bs->options (which has the full effective
4065 * options for bs) and options (which has file.* already removed).
4066 */
4067 qdict_put_str(bs->options, "driver", drv->format_name);
4068 qdict_put_str(options, "driver", drv->format_name);
4069 } else if (!drv) {
4070 error_setg(errp, "Must specify either driver or file");
4071 goto fail;
4072 }
4073
4074 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4075 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
4076 /* file must be NULL if a protocol BDS is about to be created
4077 * (the inverse results in an error message from bdrv_open_common()) */
4078 assert(!(flags & BDRV_O_PROTOCOL) || !file);
4079
4080 /* Open the image */
4081 ret = bdrv_open_common(bs, file, options, &local_err);
4082 if (ret < 0) {
4083 goto fail;
4084 }
4085
4086 /* The AioContext could have changed during bdrv_open_common() */
4087 ctx = bdrv_get_aio_context(bs);
4088
4089 if (file) {
4090 aio_context_acquire(ctx);
4091 blk_unref(file);
4092 aio_context_release(ctx);
4093 file = NULL;
4094 }
4095
4096 /* If there is a backing file, use it */
4097 if ((flags & BDRV_O_NO_BACKING) == 0) {
4098 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
4099 if (ret < 0) {
4100 goto close_and_fail;
4101 }
4102 }
4103
4104 /* Remove all children options and references
4105 * from bs->options and bs->explicit_options */
4106 QLIST_FOREACH(child, &bs->children, next) {
4107 char *child_key_dot;
4108 child_key_dot = g_strdup_printf("%s.", child->name);
4109 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
4110 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
4111 qdict_del(bs->explicit_options, child->name);
4112 qdict_del(bs->options, child->name);
4113 g_free(child_key_dot);
4114 }
4115
4116 /* Check if any unknown options were used */
4117 if (qdict_size(options) != 0) {
4118 const QDictEntry *entry = qdict_first(options);
4119 if (flags & BDRV_O_PROTOCOL) {
4120 error_setg(errp, "Block protocol '%s' doesn't support the option "
4121 "'%s'", drv->format_name, entry->key);
4122 } else {
4123 error_setg(errp,
4124 "Block format '%s' does not support the option '%s'",
4125 drv->format_name, entry->key);
4126 }
4127
4128 goto close_and_fail;
4129 }
4130
4131 bdrv_parent_cb_change_media(bs, true);
4132
4133 qobject_unref(options);
4134 options = NULL;
4135
4136 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4137 * temporary snapshot afterwards. */
4138 if (snapshot_flags) {
4139 BlockDriverState *snapshot_bs;
4140 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
4141 snapshot_options, &local_err);
4142 snapshot_options = NULL;
4143 if (local_err) {
4144 goto close_and_fail;
4145 }
4146 /* We are not going to return bs but the overlay on top of it
4147 * (snapshot_bs); thus, we have to drop the strong reference to bs
4148 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4149 * though, because the overlay still has a reference to it. */
4150 aio_context_acquire(ctx);
4151 bdrv_unref(bs);
4152 aio_context_release(ctx);
4153 bs = snapshot_bs;
4154 }
4155
4156 return bs;
4157
4158 fail:
4159 aio_context_acquire(ctx);
4160 blk_unref(file);
4161 qobject_unref(snapshot_options);
4162 qobject_unref(bs->explicit_options);
4163 qobject_unref(bs->options);
4164 qobject_unref(options);
4165 bs->options = NULL;
4166 bs->explicit_options = NULL;
4167 bdrv_unref(bs);
4168 aio_context_release(ctx);
4169 error_propagate(errp, local_err);
4170 return NULL;
4171
4172 close_and_fail:
4173 aio_context_acquire(ctx);
4174 bdrv_unref(bs);
4175 aio_context_release(ctx);
4176 qobject_unref(snapshot_options);
4177 qobject_unref(options);
4178 error_propagate(errp, local_err);
4179 return NULL;
4180 }
4181
4182 /* The caller must always hold the main AioContext lock. */
4183 BlockDriverState *bdrv_open(const char *filename, const char *reference,
4184 QDict *options, int flags, Error **errp)
4185 {
4186 GLOBAL_STATE_CODE();
4187
4188 return bdrv_open_inherit(filename, reference, options, flags, NULL,
4189 NULL, 0, errp);
4190 }
4191
4192 /* Return true if the NULL-terminated @list contains @str */
4193 static bool is_str_in_list(const char *str, const char *const *list)
4194 {
4195 if (str && list) {
4196 int i;
4197 for (i = 0; list[i] != NULL; i++) {
4198 if (!strcmp(str, list[i])) {
4199 return true;
4200 }
4201 }
4202 }
4203 return false;
4204 }
4205
4206 /*
4207 * Check that every option set in @bs->options is also set in
4208 * @new_opts.
4209 *
4210 * Options listed in the common_options list and in
4211 * @bs->drv->mutable_opts are skipped.
4212 *
4213 * Return 0 on success, otherwise return -EINVAL and set @errp.
4214 */
4215 static int bdrv_reset_options_allowed(BlockDriverState *bs,
4216 const QDict *new_opts, Error **errp)
4217 {
4218 const QDictEntry *e;
4219 /* These options are common to all block drivers and are handled
4220 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4221 const char *const common_options[] = {
4222 "node-name", "discard", "cache.direct", "cache.no-flush",
4223 "read-only", "auto-read-only", "detect-zeroes", NULL
4224 };
4225
4226 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
4227 if (!qdict_haskey(new_opts, e->key) &&
4228 !is_str_in_list(e->key, common_options) &&
4229 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
4230 error_setg(errp, "Option '%s' cannot be reset "
4231 "to its default value", e->key);
4232 return -EINVAL;
4233 }
4234 }
4235
4236 return 0;
4237 }
4238
4239 /*
4240 * Returns true if @child can be reached recursively from @bs
4241 */
4242 static bool bdrv_recurse_has_child(BlockDriverState *bs,
4243 BlockDriverState *child)
4244 {
4245 BdrvChild *c;
4246
4247 if (bs == child) {
4248 return true;
4249 }
4250
4251 QLIST_FOREACH(c, &bs->children, next) {
4252 if (bdrv_recurse_has_child(c->bs, child)) {
4253 return true;
4254 }
4255 }
4256
4257 return false;
4258 }
4259
4260 /*
4261 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4262 * reopen of multiple devices.
4263 *
4264 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4265 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4266 * be created and initialized. This newly created BlockReopenQueue should be
4267 * passed back in for subsequent calls that are intended to be of the same
4268 * atomic 'set'.
4269 *
4270 * bs is the BlockDriverState to add to the reopen queue.
4271 *
4272 * options contains the changed options for the associated bs
4273 * (the BlockReopenQueue takes ownership)
4274 *
4275 * flags contains the open flags for the associated bs
4276 *
4277 * returns a pointer to bs_queue, which is either the newly allocated
4278 * bs_queue, or the existing bs_queue being used.
4279 *
4280 * bs is drained here and undrained by bdrv_reopen_queue_free().
4281 *
4282 * To be called with bs->aio_context locked.
4283 */
4284 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
4285 BlockDriverState *bs,
4286 QDict *options,
4287 const BdrvChildClass *klass,
4288 BdrvChildRole role,
4289 bool parent_is_format,
4290 QDict *parent_options,
4291 int parent_flags,
4292 bool keep_old_opts)
4293 {
4294 assert(bs != NULL);
4295
4296 BlockReopenQueueEntry *bs_entry;
4297 BdrvChild *child;
4298 QDict *old_options, *explicit_options, *options_copy;
4299 int flags;
4300 QemuOpts *opts;
4301
4302 GLOBAL_STATE_CODE();
4303
4304 bdrv_drained_begin(bs);
4305
4306 if (bs_queue == NULL) {
4307 bs_queue = g_new0(BlockReopenQueue, 1);
4308 QTAILQ_INIT(bs_queue);
4309 }
4310
4311 if (!options) {
4312 options = qdict_new();
4313 }
4314
4315 /* Check if this BlockDriverState is already in the queue */
4316 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4317 if (bs == bs_entry->state.bs) {
4318 break;
4319 }
4320 }
4321
4322 /*
4323 * Precedence of options:
4324 * 1. Explicitly passed in options (highest)
4325 * 2. Retained from explicitly set options of bs
4326 * 3. Inherited from parent node
4327 * 4. Retained from effective options of bs
4328 */
4329
4330 /* Old explicitly set values (don't overwrite by inherited value) */
4331 if (bs_entry || keep_old_opts) {
4332 old_options = qdict_clone_shallow(bs_entry ?
4333 bs_entry->state.explicit_options :
4334 bs->explicit_options);
4335 bdrv_join_options(bs, options, old_options);
4336 qobject_unref(old_options);
4337 }
4338
4339 explicit_options = qdict_clone_shallow(options);
4340
4341 /* Inherit from parent node */
4342 if (parent_options) {
4343 flags = 0;
4344 klass->inherit_options(role, parent_is_format, &flags, options,
4345 parent_flags, parent_options);
4346 } else {
4347 flags = bdrv_get_flags(bs);
4348 }
4349
4350 if (keep_old_opts) {
4351 /* Old values are used for options that aren't set yet */
4352 old_options = qdict_clone_shallow(bs->options);
4353 bdrv_join_options(bs, options, old_options);
4354 qobject_unref(old_options);
4355 }
4356
4357 /* We have the final set of options so let's update the flags */
4358 options_copy = qdict_clone_shallow(options);
4359 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4360 qemu_opts_absorb_qdict(opts, options_copy, NULL);
4361 update_flags_from_options(&flags, opts);
4362 qemu_opts_del(opts);
4363 qobject_unref(options_copy);
4364
4365 /* bdrv_open_inherit() sets and clears some additional flags internally */
4366 flags &= ~BDRV_O_PROTOCOL;
4367 if (flags & BDRV_O_RDWR) {
4368 flags |= BDRV_O_ALLOW_RDWR;
4369 }
4370
4371 if (!bs_entry) {
4372 bs_entry = g_new0(BlockReopenQueueEntry, 1);
4373 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
4374 } else {
4375 qobject_unref(bs_entry->state.options);
4376 qobject_unref(bs_entry->state.explicit_options);
4377 }
4378
4379 bs_entry->state.bs = bs;
4380 bs_entry->state.options = options;
4381 bs_entry->state.explicit_options = explicit_options;
4382 bs_entry->state.flags = flags;
4383
4384 /*
4385 * If keep_old_opts is false then it means that unspecified
4386 * options must be reset to their original value. We don't allow
4387 * resetting 'backing' but we need to know if the option is
4388 * missing in order to decide if we have to return an error.
4389 */
4390 if (!keep_old_opts) {
4391 bs_entry->state.backing_missing =
4392 !qdict_haskey(options, "backing") &&
4393 !qdict_haskey(options, "backing.driver");
4394 }
4395
4396 QLIST_FOREACH(child, &bs->children, next) {
4397 QDict *new_child_options = NULL;
4398 bool child_keep_old = keep_old_opts;
4399
4400 /* reopen can only change the options of block devices that were
4401 * implicitly created and inherited options. For other (referenced)
4402 * block devices, a syntax like "backing.foo" results in an error. */
4403 if (child->bs->inherits_from != bs) {
4404 continue;
4405 }
4406
4407 /* Check if the options contain a child reference */
4408 if (qdict_haskey(options, child->name)) {
4409 const char *childref = qdict_get_try_str(options, child->name);
4410 /*
4411 * The current child must not be reopened if the child
4412 * reference is null or points to a different node.
4413 */
4414 if (g_strcmp0(childref, child->bs->node_name)) {
4415 continue;
4416 }
4417 /*
4418 * If the child reference points to the current child then
4419 * reopen it with its existing set of options (note that
4420 * it can still inherit new options from the parent).
4421 */
4422 child_keep_old = true;
4423 } else {
4424 /* Extract child options ("child-name.*") */
4425 char *child_key_dot = g_strdup_printf("%s.", child->name);
4426 qdict_extract_subqdict(explicit_options, NULL, child_key_dot);
4427 qdict_extract_subqdict(options, &new_child_options, child_key_dot);
4428 g_free(child_key_dot);
4429 }
4430
4431 bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options,
4432 child->klass, child->role, bs->drv->is_format,
4433 options, flags, child_keep_old);
4434 }
4435
4436 return bs_queue;
4437 }
4438
4439 /* To be called with bs->aio_context locked */
4440 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
4441 BlockDriverState *bs,
4442 QDict *options, bool keep_old_opts)
4443 {
4444 GLOBAL_STATE_CODE();
4445
4446 return bdrv_reopen_queue_child(bs_queue, bs, options, NULL, 0, false,
4447 NULL, 0, keep_old_opts);
4448 }
4449
4450 void bdrv_reopen_queue_free(BlockReopenQueue *bs_queue)
4451 {
4452 GLOBAL_STATE_CODE();
4453 if (bs_queue) {
4454 BlockReopenQueueEntry *bs_entry, *next;
4455 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4456 AioContext *ctx = bdrv_get_aio_context(bs_entry->state.bs);
4457
4458 aio_context_acquire(ctx);
4459 bdrv_drained_end(bs_entry->state.bs);
4460 aio_context_release(ctx);
4461
4462 qobject_unref(bs_entry->state.explicit_options);
4463 qobject_unref(bs_entry->state.options);
4464 g_free(bs_entry);
4465 }
4466 g_free(bs_queue);
4467 }
4468 }
4469
4470 /*
4471 * Reopen multiple BlockDriverStates atomically & transactionally.
4472 *
4473 * The queue passed in (bs_queue) must have been built up previous
4474 * via bdrv_reopen_queue().
4475 *
4476 * Reopens all BDS specified in the queue, with the appropriate
4477 * flags. All devices are prepared for reopen, and failure of any
4478 * device will cause all device changes to be abandoned, and intermediate
4479 * data cleaned up.
4480 *
4481 * If all devices prepare successfully, then the changes are committed
4482 * to all devices.
4483 *
4484 * All affected nodes must be drained between bdrv_reopen_queue() and
4485 * bdrv_reopen_multiple().
4486 *
4487 * To be called from the main thread, with all other AioContexts unlocked.
4488 */
4489 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
4490 {
4491 int ret = -1;
4492 BlockReopenQueueEntry *bs_entry, *next;
4493 AioContext *ctx;
4494 Transaction *tran = tran_new();
4495 g_autoptr(GSList) refresh_list = NULL;
4496
4497 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4498 assert(bs_queue != NULL);
4499 GLOBAL_STATE_CODE();
4500
4501 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4502 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4503 aio_context_acquire(ctx);
4504 ret = bdrv_flush(bs_entry->state.bs);
4505 aio_context_release(ctx);
4506 if (ret < 0) {
4507 error_setg_errno(errp, -ret, "Error flushing drive");
4508 goto abort;
4509 }
4510 }
4511
4512 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4513 assert(bs_entry->state.bs->quiesce_counter > 0);
4514 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4515 aio_context_acquire(ctx);
4516 ret = bdrv_reopen_prepare(&bs_entry->state, bs_queue, tran, errp);
4517 aio_context_release(ctx);
4518 if (ret < 0) {
4519 goto abort;
4520 }
4521 bs_entry->prepared = true;
4522 }
4523
4524 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
4525 BDRVReopenState *state = &bs_entry->state;
4526
4527 refresh_list = g_slist_prepend(refresh_list, state->bs);
4528 if (state->old_backing_bs) {
4529 refresh_list = g_slist_prepend(refresh_list, state->old_backing_bs);
4530 }
4531 if (state->old_file_bs) {
4532 refresh_list = g_slist_prepend(refresh_list, state->old_file_bs);
4533 }
4534 }
4535
4536 /*
4537 * Note that file-posix driver rely on permission update done during reopen
4538 * (even if no permission changed), because it wants "new" permissions for
4539 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4540 * in raw_reopen_prepare() which is called with "old" permissions.
4541 */
4542 ret = bdrv_list_refresh_perms(refresh_list, bs_queue, tran, errp);
4543 if (ret < 0) {
4544 goto abort;
4545 }
4546
4547 /*
4548 * If we reach this point, we have success and just need to apply the
4549 * changes.
4550 *
4551 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4552 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4553 * children are usually goes after parents in reopen-queue, so go from last
4554 * to first element.
4555 */
4556 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4557 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4558 aio_context_acquire(ctx);
4559 bdrv_reopen_commit(&bs_entry->state);
4560 aio_context_release(ctx);
4561 }
4562
4563 tran_commit(tran);
4564
4565 QTAILQ_FOREACH_REVERSE(bs_entry, bs_queue, entry) {
4566 BlockDriverState *bs = bs_entry->state.bs;
4567
4568 if (bs->drv->bdrv_reopen_commit_post) {
4569 ctx = bdrv_get_aio_context(bs);
4570 aio_context_acquire(ctx);
4571 bs->drv->bdrv_reopen_commit_post(&bs_entry->state);
4572 aio_context_release(ctx);
4573 }
4574 }
4575
4576 ret = 0;
4577 goto cleanup;
4578
4579 abort:
4580 tran_abort(tran);
4581 QTAILQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
4582 if (bs_entry->prepared) {
4583 ctx = bdrv_get_aio_context(bs_entry->state.bs);
4584 aio_context_acquire(ctx);
4585 bdrv_reopen_abort(&bs_entry->state);
4586 aio_context_release(ctx);
4587 }
4588 }
4589
4590 cleanup:
4591 bdrv_reopen_queue_free(bs_queue);
4592
4593 return ret;
4594 }
4595
4596 int bdrv_reopen(BlockDriverState *bs, QDict *opts, bool keep_old_opts,
4597 Error **errp)
4598 {
4599 AioContext *ctx = bdrv_get_aio_context(bs);
4600 BlockReopenQueue *queue;
4601 int ret;
4602
4603 GLOBAL_STATE_CODE();
4604
4605 queue = bdrv_reopen_queue(NULL, bs, opts, keep_old_opts);
4606
4607 if (ctx != qemu_get_aio_context()) {
4608 aio_context_release(ctx);
4609 }
4610 ret = bdrv_reopen_multiple(queue, errp);
4611
4612 if (ctx != qemu_get_aio_context()) {
4613 aio_context_acquire(ctx);
4614 }
4615
4616 return ret;
4617 }
4618
4619 int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
4620 Error **errp)
4621 {
4622 QDict *opts = qdict_new();
4623
4624 GLOBAL_STATE_CODE();
4625
4626 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, read_only);
4627
4628 return bdrv_reopen(bs, opts, true, errp);
4629 }
4630
4631 /*
4632 * Take a BDRVReopenState and check if the value of 'backing' in the
4633 * reopen_state->options QDict is valid or not.
4634 *
4635 * If 'backing' is missing from the QDict then return 0.
4636 *
4637 * If 'backing' contains the node name of the backing file of
4638 * reopen_state->bs then return 0.
4639 *
4640 * If 'backing' contains a different node name (or is null) then check
4641 * whether the current backing file can be replaced with the new one.
4642 * If that's the case then reopen_state->replace_backing_bs is set to
4643 * true and reopen_state->new_backing_bs contains a pointer to the new
4644 * backing BlockDriverState (or NULL).
4645 *
4646 * Return 0 on success, otherwise return < 0 and set @errp.
4647 *
4648 * The caller must hold the AioContext lock of @reopen_state->bs.
4649 * @reopen_state->bs can move to a different AioContext in this function.
4650 * Callers must make sure that their AioContext locking is still correct after
4651 * this.
4652 */
4653 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
4654 bool is_backing, Transaction *tran,
4655 Error **errp)
4656 {
4657 BlockDriverState *bs = reopen_state->bs;
4658 BlockDriverState *new_child_bs;
4659 BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
4660 child_bs(bs->file);
4661 const char *child_name = is_backing ? "backing" : "file";
4662 QObject *value;
4663 const char *str;
4664 AioContext *ctx, *old_ctx;
4665 int ret;
4666
4667 GLOBAL_STATE_CODE();
4668
4669 value = qdict_get(reopen_state->options, child_name);
4670 if (value == NULL) {
4671 return 0;
4672 }
4673
4674 switch (qobject_type(value)) {
4675 case QTYPE_QNULL:
4676 assert(is_backing); /* The 'file' option does not allow a null value */
4677 new_child_bs = NULL;
4678 break;
4679 case QTYPE_QSTRING:
4680 str = qstring_get_str(qobject_to(QString, value));
4681 new_child_bs = bdrv_lookup_bs(NULL, str, errp);
4682 if (new_child_bs == NULL) {
4683 return -EINVAL;
4684 } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
4685 error_setg(errp, "Making '%s' a %s child of '%s' would create a "
4686 "cycle", str, child_name, bs->node_name);
4687 return -EINVAL;
4688 }
4689 break;
4690 default:
4691 /*
4692 * The options QDict has been flattened, so 'backing' and 'file'
4693 * do not allow any other data type here.
4694 */
4695 g_assert_not_reached();
4696 }
4697
4698 if (old_child_bs == new_child_bs) {
4699 return 0;
4700 }
4701
4702 if (old_child_bs) {
4703 if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
4704 return 0;
4705 }
4706
4707 if (old_child_bs->implicit) {
4708 error_setg(errp, "Cannot replace implicit %s child of %s",
4709 child_name, bs->node_name);
4710 return -EPERM;
4711 }
4712 }
4713
4714 if (bs->drv->is_filter && !old_child_bs) {
4715 /*
4716 * Filters always have a file or a backing child, so we are trying to
4717 * change wrong child
4718 */
4719 error_setg(errp, "'%s' is a %s filter node that does not support a "
4720 "%s child", bs->node_name, bs->drv->format_name, child_name);
4721 return -EINVAL;
4722 }
4723
4724 if (is_backing) {
4725 reopen_state->old_backing_bs = old_child_bs;
4726 } else {
4727 reopen_state->old_file_bs = old_child_bs;
4728 }
4729
4730 old_ctx = bdrv_get_aio_context(bs);
4731 ctx = bdrv_get_aio_context(new_child_bs);
4732 if (old_ctx != ctx) {
4733 aio_context_release(old_ctx);
4734 aio_context_acquire(ctx);
4735 }
4736
4737 ret = bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
4738 tran, errp);
4739
4740 if (old_ctx != ctx) {
4741 aio_context_release(ctx);
4742 aio_context_acquire(old_ctx);
4743 }
4744
4745 return ret;
4746 }
4747
4748 /*
4749 * Prepares a BlockDriverState for reopen. All changes are staged in the
4750 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4751 * the block driver layer .bdrv_reopen_prepare()
4752 *
4753 * bs is the BlockDriverState to reopen
4754 * flags are the new open flags
4755 * queue is the reopen queue
4756 *
4757 * Returns 0 on success, non-zero on error. On error errp will be set
4758 * as well.
4759 *
4760 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4761 * It is the responsibility of the caller to then call the abort() or
4762 * commit() for any other BDS that have been left in a prepare() state
4763 *
4764 * The caller must hold the AioContext lock of @reopen_state->bs.
4765 */
4766 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
4767 BlockReopenQueue *queue,
4768 Transaction *change_child_tran, Error **errp)
4769 {
4770 int ret = -1;
4771 int old_flags;
4772 Error *local_err = NULL;
4773 BlockDriver *drv;
4774 QemuOpts *opts;
4775 QDict *orig_reopen_opts;
4776 char *discard = NULL;
4777 bool read_only;
4778 bool drv_prepared = false;
4779
4780 assert(reopen_state != NULL);
4781 assert(reopen_state->bs->drv != NULL);
4782 GLOBAL_STATE_CODE();
4783 drv = reopen_state->bs->drv;
4784
4785 /* This function and each driver's bdrv_reopen_prepare() remove
4786 * entries from reopen_state->options as they are processed, so
4787 * we need to make a copy of the original QDict. */
4788 orig_reopen_opts = qdict_clone_shallow(reopen_state->options);
4789
4790 /* Process generic block layer options */
4791 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
4792 if (!qemu_opts_absorb_qdict(opts, reopen_state->options, errp)) {
4793 ret = -EINVAL;
4794 goto error;
4795 }
4796
4797 /* This was already called in bdrv_reopen_queue_child() so the flags
4798 * are up-to-date. This time we simply want to remove the options from
4799 * QemuOpts in order to indicate that they have been processed. */
4800 old_flags = reopen_state->flags;
4801 update_flags_from_options(&reopen_state->flags, opts);
4802 assert(old_flags == reopen_state->flags);
4803
4804 discard = qemu_opt_get_del(opts, BDRV_OPT_DISCARD);
4805 if (discard != NULL) {
4806 if (bdrv_parse_discard_flags(discard, &reopen_state->flags) != 0) {
4807 error_setg(errp, "Invalid discard option");
4808 ret = -EINVAL;
4809 goto error;
4810 }
4811 }
4812
4813 reopen_state->detect_zeroes =
4814 bdrv_parse_detect_zeroes(opts, reopen_state->flags, &local_err);
4815 if (local_err) {
4816 error_propagate(errp, local_err);
4817 ret = -EINVAL;
4818 goto error;
4819 }
4820
4821 /* All other options (including node-name and driver) must be unchanged.
4822 * Put them back into the QDict, so that they are checked at the end
4823 * of this function. */
4824 qemu_opts_to_qdict(opts, reopen_state->options);
4825
4826 /* If we are to stay read-only, do not allow permission change
4827 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4828 * not set, or if the BDS still has copy_on_read enabled */
4829 read_only = !(reopen_state->flags & BDRV_O_RDWR);
4830 ret = bdrv_can_set_read_only(reopen_state->bs, read_only, true, &local_err);
4831 if (local_err) {
4832 error_propagate(errp, local_err);
4833 goto error;
4834 }
4835
4836 if (drv->bdrv_reopen_prepare) {
4837 /*
4838 * If a driver-specific option is missing, it means that we
4839 * should reset it to its default value.
4840 * But not all options allow that, so we need to check it first.
4841 */
4842 ret = bdrv_reset_options_allowed(reopen_state->bs,
4843 reopen_state->options, errp);
4844 if (ret) {
4845 goto error;
4846 }
4847
4848 ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
4849 if (ret) {
4850 if (local_err != NULL) {
4851 error_propagate(errp, local_err);
4852 } else {
4853 bdrv_refresh_filename(reopen_state->bs);
4854 error_setg(errp, "failed while preparing to reopen image '%s'",
4855 reopen_state->bs->filename);
4856 }
4857 goto error;
4858 }
4859 } else {
4860 /* It is currently mandatory to have a bdrv_reopen_prepare()
4861 * handler for each supported drv. */
4862 error_setg(errp, "Block format '%s' used by node '%s' "
4863 "does not support reopening files", drv->format_name,
4864 bdrv_get_device_or_node_name(reopen_state->bs));
4865 ret = -1;
4866 goto error;
4867 }
4868
4869 drv_prepared = true;
4870
4871 /*
4872 * We must provide the 'backing' option if the BDS has a backing
4873 * file or if the image file has a backing file name as part of
4874 * its metadata. Otherwise the 'backing' option can be omitted.
4875 */
4876 if (drv->supports_backing && reopen_state->backing_missing &&
4877 (reopen_state->bs->backing || reopen_state->bs->backing_file[0])) {
4878 error_setg(errp, "backing is missing for '%s'",
4879 reopen_state->bs->node_name);
4880 ret = -EINVAL;
4881 goto error;
4882 }
4883
4884 /*
4885 * Allow changing the 'backing' option. The new value can be
4886 * either a reference to an existing node (using its node name)
4887 * or NULL to simply detach the current backing file.
4888 */
4889 ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
4890 change_child_tran, errp);
4891 if (ret < 0) {
4892 goto error;
4893 }
4894 qdict_del(reopen_state->options, "backing");
4895
4896 /* Allow changing the 'file' option. In this case NULL is not allowed */
4897 ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
4898 change_child_tran, errp);
4899 if (ret < 0) {
4900 goto error;
4901 }
4902 qdict_del(reopen_state->options, "file");
4903
4904 /* Options that are not handled are only okay if they are unchanged
4905 * compared to the old state. It is expected that some options are only
4906 * used for the initial open, but not reopen (e.g. filename) */
4907 if (qdict_size(reopen_state->options)) {
4908 const QDictEntry *entry = qdict_first(reopen_state->options);
4909
4910 do {
4911 QObject *new = entry->value;
4912 QObject *old = qdict_get(reopen_state->bs->options, entry->key);
4913
4914 /* Allow child references (child_name=node_name) as long as they
4915 * point to the current child (i.e. everything stays the same). */
4916 if (qobject_type(new) == QTYPE_QSTRING) {
4917 BdrvChild *child;
4918 QLIST_FOREACH(child, &reopen_state->bs->children, next) {
4919 if (!strcmp(child->name, entry->key)) {
4920 break;
4921 }
4922 }
4923
4924 if (child) {
4925 if (!strcmp(child->bs->node_name,
4926 qstring_get_str(qobject_to(QString, new)))) {
4927 continue; /* Found child with this name, skip option */
4928 }
4929 }
4930 }
4931
4932 /*
4933 * TODO: When using -drive to specify blockdev options, all values
4934 * will be strings; however, when using -blockdev, blockdev-add or
4935 * filenames using the json:{} pseudo-protocol, they will be
4936 * correctly typed.
4937 * In contrast, reopening options are (currently) always strings
4938 * (because you can only specify them through qemu-io; all other
4939 * callers do not specify any options).
4940 * Therefore, when using anything other than -drive to create a BDS,
4941 * this cannot detect non-string options as unchanged, because
4942 * qobject_is_equal() always returns false for objects of different
4943 * type. In the future, this should be remedied by correctly typing
4944 * all options. For now, this is not too big of an issue because
4945 * the user can simply omit options which cannot be changed anyway,
4946 * so they will stay unchanged.
4947 */
4948 if (!qobject_is_equal(new, old)) {
4949 error_setg(errp, "Cannot change the option '%s'", entry->key);
4950 ret = -EINVAL;
4951 goto error;
4952 }
4953 } while ((entry = qdict_next(reopen_state->options, entry)));
4954 }
4955
4956 ret = 0;
4957
4958 /* Restore the original reopen_state->options QDict */
4959 qobject_unref(reopen_state->options);
4960 reopen_state->options = qobject_ref(orig_reopen_opts);
4961
4962 error:
4963 if (ret < 0 && drv_prepared) {
4964 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4965 * call drv->bdrv_reopen_abort() before signaling an error
4966 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4967 * when the respective bdrv_reopen_prepare() has failed) */
4968 if (drv->bdrv_reopen_abort) {
4969 drv->bdrv_reopen_abort(reopen_state);
4970 }
4971 }
4972 qemu_opts_del(opts);
4973 qobject_unref(orig_reopen_opts);
4974 g_free(discard);
4975 return ret;
4976 }
4977
4978 /*
4979 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4980 * makes them final by swapping the staging BlockDriverState contents into
4981 * the active BlockDriverState contents.
4982 */
4983 static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
4984 {
4985 BlockDriver *drv;
4986 BlockDriverState *bs;
4987 BdrvChild *child;
4988
4989 assert(reopen_state != NULL);
4990 bs = reopen_state->bs;
4991 drv = bs->drv;
4992 assert(drv != NULL);
4993 GLOBAL_STATE_CODE();
4994
4995 /* If there are any driver level actions to take */
4996 if (drv->bdrv_reopen_commit) {
4997 drv->bdrv_reopen_commit(reopen_state);
4998 }
4999
5000 /* set BDS specific flags now */
5001 qobject_unref(bs->explicit_options);
5002 qobject_unref(bs->options);
5003 qobject_ref(reopen_state->explicit_options);
5004 qobject_ref(reopen_state->options);
5005
5006 bs->explicit_options = reopen_state->explicit_options;
5007 bs->options = reopen_state->options;
5008 bs->open_flags = reopen_state->flags;
5009 bs->detect_zeroes = reopen_state->detect_zeroes;
5010
5011 /* Remove child references from bs->options and bs->explicit_options.
5012 * Child options were already removed in bdrv_reopen_queue_child() */
5013 QLIST_FOREACH(child, &bs->children, next) {
5014 qdict_del(bs->explicit_options, child->name);
5015 qdict_del(bs->options, child->name);
5016 }
5017 /* backing is probably removed, so it's not handled by previous loop */
5018 qdict_del(bs->explicit_options, "backing");
5019 qdict_del(bs->options, "backing");
5020
5021 bdrv_graph_rdlock_main_loop();
5022 bdrv_refresh_limits(bs, NULL, NULL);
5023 bdrv_graph_rdunlock_main_loop();
5024 bdrv_refresh_total_sectors(bs, bs->total_sectors);
5025 }
5026
5027 /*
5028 * Abort the reopen, and delete and free the staged changes in
5029 * reopen_state
5030 */
5031 static void bdrv_reopen_abort(BDRVReopenState *reopen_state)
5032 {
5033 BlockDriver *drv;
5034
5035 assert(reopen_state != NULL);
5036 drv = reopen_state->bs->drv;
5037 assert(drv != NULL);
5038 GLOBAL_STATE_CODE();
5039
5040 if (drv->bdrv_reopen_abort) {
5041 drv->bdrv_reopen_abort(reopen_state);
5042 }
5043 }
5044
5045
5046 static void bdrv_close(BlockDriverState *bs)
5047 {
5048 BdrvAioNotifier *ban, *ban_next;
5049 BdrvChild *child, *next;
5050
5051 GLOBAL_STATE_CODE();
5052 assert(!bs->refcnt);
5053
5054 bdrv_drained_begin(bs); /* complete I/O */
5055 bdrv_flush(bs);
5056 bdrv_drain(bs); /* in case flush left pending I/O */
5057
5058 if (bs->drv) {
5059 if (bs->drv->bdrv_close) {
5060 /* Must unfreeze all children, so bdrv_unref_child() works */
5061 bs->drv->bdrv_close(bs);
5062 }
5063 bs->drv = NULL;
5064 }
5065
5066 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
5067 bdrv_unref_child(bs, child);
5068 }
5069
5070 assert(!bs->backing);
5071 assert(!bs->file);
5072 g_free(bs->opaque);
5073 bs->opaque = NULL;
5074 qatomic_set(&bs->copy_on_read, 0);
5075 bs->backing_file[0] = '\0';
5076 bs->backing_format[0] = '\0';
5077 bs->total_sectors = 0;
5078 bs->encrypted = false;
5079 bs->sg = false;
5080 qobject_unref(bs->options);
5081 qobject_unref(bs->explicit_options);
5082 bs->options = NULL;
5083 bs->explicit_options = NULL;
5084 qobject_unref(bs->full_open_options);
5085 bs->full_open_options = NULL;
5086 g_free(bs->block_status_cache);
5087 bs->block_status_cache = NULL;
5088
5089 bdrv_release_named_dirty_bitmaps(bs);
5090 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
5091
5092 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
5093 g_free(ban);
5094 }
5095 QLIST_INIT(&bs->aio_notifiers);
5096 bdrv_drained_end(bs);
5097
5098 /*
5099 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
5100 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5101 * gets called.
5102 */
5103 if (bs->quiesce_counter) {
5104 bdrv_drain_all_end_quiesce(bs);
5105 }
5106 }
5107
5108 void bdrv_close_all(void)
5109 {
5110 GLOBAL_STATE_CODE();
5111 assert(job_next(NULL) == NULL);
5112
5113 /* Drop references from requests still in flight, such as canceled block
5114 * jobs whose AIO context has not been polled yet */
5115 bdrv_drain_all();
5116
5117 blk_remove_all_bs();
5118 blockdev_close_all_bdrv_states();
5119
5120 assert(QTAILQ_EMPTY(&all_bdrv_states));
5121 }
5122
5123 static bool should_update_child(BdrvChild *c, BlockDriverState *to)
5124 {
5125 GQueue *queue;
5126 GHashTable *found;
5127 bool ret;
5128
5129 if (c->klass->stay_at_node) {
5130 return false;
5131 }
5132
5133 /* If the child @c belongs to the BDS @to, replacing the current
5134 * c->bs by @to would mean to create a loop.
5135 *
5136 * Such a case occurs when appending a BDS to a backing chain.
5137 * For instance, imagine the following chain:
5138 *
5139 * guest device -> node A -> further backing chain...
5140 *
5141 * Now we create a new BDS B which we want to put on top of this
5142 * chain, so we first attach A as its backing node:
5143 *
5144 * node B
5145 * |
5146 * v
5147 * guest device -> node A -> further backing chain...
5148 *
5149 * Finally we want to replace A by B. When doing that, we want to
5150 * replace all pointers to A by pointers to B -- except for the
5151 * pointer from B because (1) that would create a loop, and (2)
5152 * that pointer should simply stay intact:
5153 *
5154 * guest device -> node B
5155 * |
5156 * v
5157 * node A -> further backing chain...
5158 *
5159 * In general, when replacing a node A (c->bs) by a node B (@to),
5160 * if A is a child of B, that means we cannot replace A by B there
5161 * because that would create a loop. Silently detaching A from B
5162 * is also not really an option. So overall just leaving A in
5163 * place there is the most sensible choice.
5164 *
5165 * We would also create a loop in any cases where @c is only
5166 * indirectly referenced by @to. Prevent this by returning false
5167 * if @c is found (by breadth-first search) anywhere in the whole
5168 * subtree of @to.
5169 */
5170
5171 ret = true;
5172 found = g_hash_table_new(NULL, NULL);
5173 g_hash_table_add(found, to);
5174 queue = g_queue_new();
5175 g_queue_push_tail(queue, to);
5176
5177 while (!g_queue_is_empty(queue)) {
5178 BlockDriverState *v = g_queue_pop_head(queue);
5179 BdrvChild *c2;
5180
5181 QLIST_FOREACH(c2, &v->children, next) {
5182 if (c2 == c) {
5183 ret = false;
5184 break;
5185 }
5186
5187 if (g_hash_table_contains(found, c2->bs)) {
5188 continue;
5189 }
5190
5191 g_queue_push_tail(queue, c2->bs);
5192 g_hash_table_add(found, c2->bs);
5193 }
5194 }
5195
5196 g_queue_free(queue);
5197 g_hash_table_destroy(found);
5198
5199 return ret;
5200 }
5201
5202 static void bdrv_remove_child_commit(void *opaque)
5203 {
5204 GLOBAL_STATE_CODE();
5205 bdrv_child_free(opaque);
5206 }
5207
5208 static TransactionActionDrv bdrv_remove_child_drv = {
5209 .commit = bdrv_remove_child_commit,
5210 };
5211
5212 /* Function doesn't update permissions, caller is responsible for this. */
5213 static void bdrv_remove_child(BdrvChild *child, Transaction *tran)
5214 {
5215 if (!child) {
5216 return;
5217 }
5218
5219 if (child->bs) {
5220 BlockDriverState *bs = child->bs;
5221 bdrv_drained_begin(bs);
5222 bdrv_replace_child_tran(child, NULL, tran);
5223 bdrv_drained_end(bs);
5224 }
5225
5226 tran_add(tran, &bdrv_remove_child_drv, child);
5227 }
5228
5229 static void undrain_on_clean_cb(void *opaque)
5230 {
5231 bdrv_drained_end(opaque);
5232 }
5233
5234 static TransactionActionDrv undrain_on_clean = {
5235 .clean = undrain_on_clean_cb,
5236 };
5237
5238 static int bdrv_replace_node_noperm(BlockDriverState *from,
5239 BlockDriverState *to,
5240 bool auto_skip, Transaction *tran,
5241 Error **errp)
5242 {
5243 BdrvChild *c, *next;
5244
5245 GLOBAL_STATE_CODE();
5246
5247 bdrv_drained_begin(from);
5248 bdrv_drained_begin(to);
5249 tran_add(tran, &undrain_on_clean, from);
5250 tran_add(tran, &undrain_on_clean, to);
5251
5252 QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
5253 assert(c->bs == from);
5254 if (!should_update_child(c, to)) {
5255 if (auto_skip) {
5256 continue;
5257 }
5258 error_setg(errp, "Should not change '%s' link to '%s'",
5259 c->name, from->node_name);
5260 return -EINVAL;
5261 }
5262 if (c->frozen) {
5263 error_setg(errp, "Cannot change '%s' link to '%s'",
5264 c->name, from->node_name);
5265 return -EPERM;
5266 }
5267 bdrv_replace_child_tran(c, to, tran);
5268 }
5269
5270 return 0;
5271 }
5272
5273 /*
5274 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5275 * if it creates a parent-child relation loop or if parent is block-job.
5276 *
5277 * With auto_skip=false the error is returned if from has a parent which should
5278 * not be updated.
5279 *
5280 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5281 * case backing link of the cow-parent of @to is removed.
5282 */
5283 static int bdrv_replace_node_common(BlockDriverState *from,
5284 BlockDriverState *to,
5285 bool auto_skip, bool detach_subchain,
5286 Error **errp)
5287 {
5288 Transaction *tran = tran_new();
5289 g_autoptr(GSList) refresh_list = NULL;
5290 BlockDriverState *to_cow_parent = NULL;
5291 int ret;
5292
5293 GLOBAL_STATE_CODE();
5294
5295 if (detach_subchain) {
5296 assert(bdrv_chain_contains(from, to));
5297 assert(from != to);
5298 for (to_cow_parent = from;
5299 bdrv_filter_or_cow_bs(to_cow_parent) != to;
5300 to_cow_parent = bdrv_filter_or_cow_bs(to_cow_parent))
5301 {
5302 ;
5303 }
5304 }
5305
5306 /* Make sure that @from doesn't go away until we have successfully attached
5307 * all of its parents to @to. */
5308 bdrv_ref(from);
5309
5310 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5311 assert(bdrv_get_aio_context(from) == bdrv_get_aio_context(to));
5312 bdrv_drained_begin(from);
5313
5314 /*
5315 * Do the replacement without permission update.
5316 * Replacement may influence the permissions, we should calculate new
5317 * permissions based on new graph. If we fail, we'll roll-back the
5318 * replacement.
5319 */
5320 ret = bdrv_replace_node_noperm(from, to, auto_skip, tran, errp);
5321 if (ret < 0) {
5322 goto out;
5323 }
5324
5325 if (detach_subchain) {
5326 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent), tran);
5327 }
5328
5329 refresh_list = g_slist_prepend(refresh_list, to);
5330 refresh_list = g_slist_prepend(refresh_list, from);
5331
5332 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5333 if (ret < 0) {
5334 goto out;
5335 }
5336
5337 ret = 0;
5338
5339 out:
5340 tran_finalize(tran, ret);
5341
5342 bdrv_drained_end(from);
5343 bdrv_unref(from);
5344
5345 return ret;
5346 }
5347
5348 int bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
5349 Error **errp)
5350 {
5351 GLOBAL_STATE_CODE();
5352
5353 return bdrv_replace_node_common(from, to, true, false, errp);
5354 }
5355
5356 int bdrv_drop_filter(BlockDriverState *bs, Error **errp)
5357 {
5358 GLOBAL_STATE_CODE();
5359
5360 return bdrv_replace_node_common(bs, bdrv_filter_or_cow_bs(bs), true, true,
5361 errp);
5362 }
5363
5364 /*
5365 * Add new bs contents at the top of an image chain while the chain is
5366 * live, while keeping required fields on the top layer.
5367 *
5368 * This will modify the BlockDriverState fields, and swap contents
5369 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5370 *
5371 * bs_new must not be attached to a BlockBackend and must not have backing
5372 * child.
5373 *
5374 * This function does not create any image files.
5375 *
5376 * The caller must hold the AioContext lock for @bs_top.
5377 */
5378 int bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
5379 Error **errp)
5380 {
5381 int ret;
5382 BdrvChild *child;
5383 Transaction *tran = tran_new();
5384 AioContext *old_context, *new_context = NULL;
5385
5386 GLOBAL_STATE_CODE();
5387
5388 assert(!bs_new->backing);
5389
5390 old_context = bdrv_get_aio_context(bs_top);
5391
5392 child = bdrv_attach_child_noperm(bs_new, bs_top, "backing",
5393 &child_of_bds, bdrv_backing_role(bs_new),
5394 tran, errp);
5395 if (!child) {
5396 ret = -EINVAL;
5397 goto out;
5398 }
5399
5400 /*
5401 * bdrv_attach_child_noperm could change the AioContext of bs_top.
5402 * bdrv_replace_node_noperm calls bdrv_drained_begin, so let's temporarily
5403 * hold the new AioContext, since bdrv_drained_begin calls BDRV_POLL_WHILE
5404 * that assumes the new lock is taken.
5405 */
5406 new_context = bdrv_get_aio_context(bs_top);
5407
5408 if (old_context != new_context) {
5409 aio_context_release(old_context);
5410 aio_context_acquire(new_context);
5411 }
5412
5413 ret = bdrv_replace_node_noperm(bs_top, bs_new, true, tran, errp);
5414 if (ret < 0) {
5415 goto out;
5416 }
5417
5418 ret = bdrv_refresh_perms(bs_new, tran, errp);
5419 out:
5420 tran_finalize(tran, ret);
5421
5422 bdrv_graph_rdlock_main_loop();
5423 bdrv_refresh_limits(bs_top, NULL, NULL);
5424 bdrv_graph_rdunlock_main_loop();
5425
5426 if (new_context && old_context != new_context) {
5427 aio_context_release(new_context);
5428 aio_context_acquire(old_context);
5429 }
5430
5431 return ret;
5432 }
5433
5434 /* Not for empty child */
5435 int bdrv_replace_child_bs(BdrvChild *child, BlockDriverState *new_bs,
5436 Error **errp)
5437 {
5438 int ret;
5439 Transaction *tran = tran_new();
5440 g_autoptr(GSList) refresh_list = NULL;
5441 BlockDriverState *old_bs = child->bs;
5442
5443 GLOBAL_STATE_CODE();
5444
5445 bdrv_ref(old_bs);
5446 bdrv_drained_begin(old_bs);
5447 bdrv_drained_begin(new_bs);
5448
5449 bdrv_replace_child_tran(child, new_bs, tran);
5450
5451 refresh_list = g_slist_prepend(refresh_list, old_bs);
5452 refresh_list = g_slist_prepend(refresh_list, new_bs);
5453
5454 ret = bdrv_list_refresh_perms(refresh_list, NULL, tran, errp);
5455
5456 tran_finalize(tran, ret);
5457
5458 bdrv_drained_end(old_bs);
5459 bdrv_drained_end(new_bs);
5460 bdrv_unref(old_bs);
5461
5462 return ret;
5463 }
5464
5465 static void bdrv_delete(BlockDriverState *bs)
5466 {
5467 assert(bdrv_op_blocker_is_empty(bs));
5468 assert(!bs->refcnt);
5469 GLOBAL_STATE_CODE();
5470
5471 /* remove from list, if necessary */
5472 if (bs->node_name[0] != '\0') {
5473 QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
5474 }
5475 QTAILQ_REMOVE(&all_bdrv_states, bs, bs_list);
5476
5477 bdrv_close(bs);
5478
5479 g_free(bs);
5480 }
5481
5482
5483 /*
5484 * Replace @bs by newly created block node.
5485 *
5486 * @options is a QDict of options to pass to the block drivers, or NULL for an
5487 * empty set of options. The reference to the QDict belongs to the block layer
5488 * after the call (even on failure), so if the caller intends to reuse the
5489 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5490 *
5491 * The caller holds the AioContext lock for @bs. It must make sure that @bs
5492 * stays in the same AioContext, i.e. @options must not refer to nodes in a
5493 * different AioContext.
5494 */
5495 BlockDriverState *bdrv_insert_node(BlockDriverState *bs, QDict *options,
5496 int flags, Error **errp)
5497 {
5498 ERRP_GUARD();
5499 int ret;
5500 AioContext *ctx = bdrv_get_aio_context(bs);
5501 BlockDriverState *new_node_bs = NULL;
5502 const char *drvname, *node_name;
5503 BlockDriver *drv;
5504
5505 drvname = qdict_get_try_str(options, "driver");
5506 if (!drvname) {
5507 error_setg(errp, "driver is not specified");
5508 goto fail;
5509 }
5510
5511 drv = bdrv_find_format(drvname);
5512 if (!drv) {
5513 error_setg(errp, "Unknown driver: '%s'", drvname);
5514 goto fail;
5515 }
5516
5517 node_name = qdict_get_try_str(options, "node-name");
5518
5519 GLOBAL_STATE_CODE();
5520
5521 aio_context_release(ctx);
5522 aio_context_acquire(qemu_get_aio_context());
5523 new_node_bs = bdrv_new_open_driver_opts(drv, node_name, options, flags,
5524 errp);
5525 aio_context_release(qemu_get_aio_context());
5526 aio_context_acquire(ctx);
5527 assert(bdrv_get_aio_context(bs) == ctx);
5528
5529 options = NULL; /* bdrv_new_open_driver() eats options */
5530 if (!new_node_bs) {
5531 error_prepend(errp, "Could not create node: ");
5532 goto fail;
5533 }
5534
5535 bdrv_drained_begin(bs);
5536 ret = bdrv_replace_node(bs, new_node_bs, errp);
5537 bdrv_drained_end(bs);
5538
5539 if (ret < 0) {
5540 error_prepend(errp, "Could not replace node: ");
5541 goto fail;
5542 }
5543
5544 return new_node_bs;
5545
5546 fail:
5547 qobject_unref(options);
5548 bdrv_unref(new_node_bs);
5549 return NULL;
5550 }
5551
5552 /*
5553 * Run consistency checks on an image
5554 *
5555 * Returns 0 if the check could be completed (it doesn't mean that the image is
5556 * free of errors) or -errno when an internal error occurred. The results of the
5557 * check are stored in res.
5558 */
5559 int coroutine_fn bdrv_co_check(BlockDriverState *bs,
5560 BdrvCheckResult *res, BdrvCheckMode fix)
5561 {
5562 IO_CODE();
5563 assert_bdrv_graph_readable();
5564 if (bs->drv == NULL) {
5565 return -ENOMEDIUM;
5566 }
5567 if (bs->drv->bdrv_co_check == NULL) {
5568 return -ENOTSUP;
5569 }
5570
5571 memset(res, 0, sizeof(*res));
5572 return bs->drv->bdrv_co_check(bs, res, fix);
5573 }
5574
5575 /*
5576 * Return values:
5577 * 0 - success
5578 * -EINVAL - backing format specified, but no file
5579 * -ENOSPC - can't update the backing file because no space is left in the
5580 * image file header
5581 * -ENOTSUP - format driver doesn't support changing the backing file
5582 */
5583 int bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
5584 const char *backing_fmt, bool require)
5585 {
5586 BlockDriver *drv = bs->drv;
5587 int ret;
5588
5589 GLOBAL_STATE_CODE();
5590
5591 if (!drv) {
5592 return -ENOMEDIUM;
5593 }
5594
5595 /* Backing file format doesn't make sense without a backing file */
5596 if (backing_fmt && !backing_file) {
5597 return -EINVAL;
5598 }
5599
5600 if (require && backing_file && !backing_fmt) {
5601 return -EINVAL;
5602 }
5603
5604 if (drv->bdrv_change_backing_file != NULL) {
5605 ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
5606 } else {
5607 ret = -ENOTSUP;
5608 }
5609
5610 if (ret == 0) {
5611 pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
5612 pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
5613 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
5614 backing_file ?: "");
5615 }
5616 return ret;
5617 }
5618
5619 /*
5620 * Finds the first non-filter node above bs in the chain between
5621 * active and bs. The returned node is either an immediate parent of
5622 * bs, or there are only filter nodes between the two.
5623 *
5624 * Returns NULL if bs is not found in active's image chain,
5625 * or if active == bs.
5626 *
5627 * Returns the bottommost base image if bs == NULL.
5628 */
5629 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
5630 BlockDriverState *bs)
5631 {
5632
5633 GLOBAL_STATE_CODE();
5634
5635 bs = bdrv_skip_filters(bs);
5636 active = bdrv_skip_filters(active);
5637
5638 while (active) {
5639 BlockDriverState *next = bdrv_backing_chain_next(active);
5640 if (bs == next) {
5641 return active;
5642 }
5643 active = next;
5644 }
5645
5646 return NULL;
5647 }
5648
5649 /* Given a BDS, searches for the base layer. */
5650 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
5651 {
5652 GLOBAL_STATE_CODE();
5653
5654 return bdrv_find_overlay(bs, NULL);
5655 }
5656
5657 /*
5658 * Return true if at least one of the COW (backing) and filter links
5659 * between @bs and @base is frozen. @errp is set if that's the case.
5660 * @base must be reachable from @bs, or NULL.
5661 */
5662 bool bdrv_is_backing_chain_frozen(BlockDriverState *bs, BlockDriverState *base,
5663 Error **errp)
5664 {
5665 BlockDriverState *i;
5666 BdrvChild *child;
5667
5668 GLOBAL_STATE_CODE();
5669
5670 for (i = bs; i != base; i = child_bs(child)) {
5671 child = bdrv_filter_or_cow_child(i);
5672
5673 if (child && child->frozen) {
5674 error_setg(errp, "Cannot change '%s' link from '%s' to '%s'",
5675 child->name, i->node_name, child->bs->node_name);
5676 return true;
5677 }
5678 }
5679
5680 return false;
5681 }
5682
5683 /*
5684 * Freeze all COW (backing) and filter links between @bs and @base.
5685 * If any of the links is already frozen the operation is aborted and
5686 * none of the links are modified.
5687 * @base must be reachable from @bs, or NULL.
5688 * Returns 0 on success. On failure returns < 0 and sets @errp.
5689 */
5690 int bdrv_freeze_backing_chain(BlockDriverState *bs, BlockDriverState *base,
5691 Error **errp)
5692 {
5693 BlockDriverState *i;
5694 BdrvChild *child;
5695
5696 GLOBAL_STATE_CODE();
5697
5698 if (bdrv_is_backing_chain_frozen(bs, base, errp)) {
5699 return -EPERM;
5700 }
5701
5702 for (i = bs; i != base; i = child_bs(child)) {
5703 child = bdrv_filter_or_cow_child(i);
5704 if (child && child->bs->never_freeze) {
5705 error_setg(errp, "Cannot freeze '%s' link to '%s'",
5706 child->name, child->bs->node_name);
5707 return -EPERM;
5708 }
5709 }
5710
5711 for (i = bs; i != base; i = child_bs(child)) {
5712 child = bdrv_filter_or_cow_child(i);
5713 if (child) {
5714 child->frozen = true;
5715 }
5716 }
5717
5718 return 0;
5719 }
5720
5721 /*
5722 * Unfreeze all COW (backing) and filter links between @bs and @base.
5723 * The caller must ensure that all links are frozen before using this
5724 * function.
5725 * @base must be reachable from @bs, or NULL.
5726 */
5727 void bdrv_unfreeze_backing_chain(BlockDriverState *bs, BlockDriverState *base)
5728 {
5729 BlockDriverState *i;
5730 BdrvChild *child;
5731
5732 GLOBAL_STATE_CODE();
5733
5734 for (i = bs; i != base; i = child_bs(child)) {
5735 child = bdrv_filter_or_cow_child(i);
5736 if (child) {
5737 assert(child->frozen);
5738 child->frozen = false;
5739 }
5740 }
5741 }
5742
5743 /*
5744 * Drops images above 'base' up to and including 'top', and sets the image
5745 * above 'top' to have base as its backing file.
5746 *
5747 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5748 * information in 'bs' can be properly updated.
5749 *
5750 * E.g., this will convert the following chain:
5751 * bottom <- base <- intermediate <- top <- active
5752 *
5753 * to
5754 *
5755 * bottom <- base <- active
5756 *
5757 * It is allowed for bottom==base, in which case it converts:
5758 *
5759 * base <- intermediate <- top <- active
5760 *
5761 * to
5762 *
5763 * base <- active
5764 *
5765 * If backing_file_str is non-NULL, it will be used when modifying top's
5766 * overlay image metadata.
5767 *
5768 * Error conditions:
5769 * if active == top, that is considered an error
5770 *
5771 */
5772 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
5773 const char *backing_file_str)
5774 {
5775 BlockDriverState *explicit_top = top;
5776 bool update_inherits_from;
5777 BdrvChild *c;
5778 Error *local_err = NULL;
5779 int ret = -EIO;
5780 g_autoptr(GSList) updated_children = NULL;
5781 GSList *p;
5782
5783 GLOBAL_STATE_CODE();
5784
5785 bdrv_ref(top);
5786 bdrv_drained_begin(base);
5787
5788 if (!top->drv || !base->drv) {
5789 goto exit;
5790 }
5791
5792 /* Make sure that base is in the backing chain of top */
5793 if (!bdrv_chain_contains(top, base)) {
5794 goto exit;
5795 }
5796
5797 /* If 'base' recursively inherits from 'top' then we should set
5798 * base->inherits_from to top->inherits_from after 'top' and all
5799 * other intermediate nodes have been dropped.
5800 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5801 * it because no one inherits from it. We use explicit_top for that. */
5802 explicit_top = bdrv_skip_implicit_filters(explicit_top);
5803 update_inherits_from = bdrv_inherits_from_recursive(base, explicit_top);
5804
5805 /* success - we can delete the intermediate states, and link top->base */
5806 if (!backing_file_str) {
5807 bdrv_refresh_filename(base);
5808 backing_file_str = base->filename;
5809 }
5810
5811 QLIST_FOREACH(c, &top->parents, next_parent) {
5812 updated_children = g_slist_prepend(updated_children, c);
5813 }
5814
5815 /*
5816 * It seems correct to pass detach_subchain=true here, but it triggers
5817 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5818 * another drained section, which modify the graph (for example, removing
5819 * the child, which we keep in updated_children list). So, it's a TODO.
5820 *
5821 * Note, bug triggered if pass detach_subchain=true here and run
5822 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5823 * That's a FIXME.
5824 */
5825 bdrv_replace_node_common(top, base, false, false, &local_err);
5826 if (local_err) {
5827 error_report_err(local_err);
5828 goto exit;
5829 }
5830
5831 for (p = updated_children; p; p = p->next) {
5832 c = p->data;
5833
5834 if (c->klass->update_filename) {
5835 ret = c->klass->update_filename(c, base, backing_file_str,
5836 &local_err);
5837 if (ret < 0) {
5838 /*
5839 * TODO: Actually, we want to rollback all previous iterations
5840 * of this loop, and (which is almost impossible) previous
5841 * bdrv_replace_node()...
5842 *
5843 * Note, that c->klass->update_filename may lead to permission
5844 * update, so it's a bad idea to call it inside permission
5845 * update transaction of bdrv_replace_node.
5846 */
5847 error_report_err(local_err);
5848 goto exit;
5849 }
5850 }
5851 }
5852
5853 if (update_inherits_from) {
5854 base->inherits_from = explicit_top->inherits_from;
5855 }
5856
5857 ret = 0;
5858 exit:
5859 bdrv_drained_end(base);
5860 bdrv_unref(top);
5861 return ret;
5862 }
5863
5864 /**
5865 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
5866 * sums the size of all data-bearing children. (This excludes backing
5867 * children.)
5868 */
5869 static int64_t coroutine_fn GRAPH_RDLOCK
5870 bdrv_sum_allocated_file_size(BlockDriverState *bs)
5871 {
5872 BdrvChild *child;
5873 int64_t child_size, sum = 0;
5874
5875 QLIST_FOREACH(child, &bs->children, next) {
5876 if (child->role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
5877 BDRV_CHILD_FILTERED))
5878 {
5879 child_size = bdrv_co_get_allocated_file_size(child->bs);
5880 if (child_size < 0) {
5881 return child_size;
5882 }
5883 sum += child_size;
5884 }
5885 }
5886
5887 return sum;
5888 }
5889
5890 /**
5891 * Length of a allocated file in bytes. Sparse files are counted by actual
5892 * allocated space. Return < 0 if error or unknown.
5893 */
5894 int64_t coroutine_fn bdrv_co_get_allocated_file_size(BlockDriverState *bs)
5895 {
5896 BlockDriver *drv = bs->drv;
5897 IO_CODE();
5898 assert_bdrv_graph_readable();
5899
5900 if (!drv) {
5901 return -ENOMEDIUM;
5902 }
5903 if (drv->bdrv_co_get_allocated_file_size) {
5904 return drv->bdrv_co_get_allocated_file_size(bs);
5905 }
5906
5907 if (drv->bdrv_file_open) {
5908 /*
5909 * Protocol drivers default to -ENOTSUP (most of their data is
5910 * not stored in any of their children (if they even have any),
5911 * so there is no generic way to figure it out).
5912 */
5913 return -ENOTSUP;
5914 } else if (drv->is_filter) {
5915 /* Filter drivers default to the size of their filtered child */
5916 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs));
5917 } else {
5918 /* Other drivers default to summing their children's sizes */
5919 return bdrv_sum_allocated_file_size(bs);
5920 }
5921 }
5922
5923 /*
5924 * bdrv_measure:
5925 * @drv: Format driver
5926 * @opts: Creation options for new image
5927 * @in_bs: Existing image containing data for new image (may be NULL)
5928 * @errp: Error object
5929 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5930 * or NULL on error
5931 *
5932 * Calculate file size required to create a new image.
5933 *
5934 * If @in_bs is given then space for allocated clusters and zero clusters
5935 * from that image are included in the calculation. If @opts contains a
5936 * backing file that is shared by @in_bs then backing clusters may be omitted
5937 * from the calculation.
5938 *
5939 * If @in_bs is NULL then the calculation includes no allocated clusters
5940 * unless a preallocation option is given in @opts.
5941 *
5942 * Note that @in_bs may use a different BlockDriver from @drv.
5943 *
5944 * If an error occurs the @errp pointer is set.
5945 */
5946 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
5947 BlockDriverState *in_bs, Error **errp)
5948 {
5949 IO_CODE();
5950 if (!drv->bdrv_measure) {
5951 error_setg(errp, "Block driver '%s' does not support size measurement",
5952 drv->format_name);
5953 return NULL;
5954 }
5955
5956 return drv->bdrv_measure(opts, in_bs, errp);
5957 }
5958
5959 /**
5960 * Return number of sectors on success, -errno on error.
5961 */
5962 int64_t coroutine_fn bdrv_co_nb_sectors(BlockDriverState *bs)
5963 {
5964 BlockDriver *drv = bs->drv;
5965 IO_CODE();
5966 assert_bdrv_graph_readable();
5967
5968 if (!drv)
5969 return -ENOMEDIUM;
5970
5971 if (bs->bl.has_variable_length) {
5972 int ret = bdrv_co_refresh_total_sectors(bs, bs->total_sectors);
5973 if (ret < 0) {
5974 return ret;
5975 }
5976 }
5977 return bs->total_sectors;
5978 }
5979
5980 /*
5981 * This wrapper is written by hand because this function is in the hot I/O path,
5982 * via blk_get_geometry.
5983 */
5984 int64_t coroutine_mixed_fn bdrv_nb_sectors(BlockDriverState *bs)
5985 {
5986 BlockDriver *drv = bs->drv;
5987 IO_CODE();
5988
5989 if (!drv)
5990 return -ENOMEDIUM;
5991
5992 if (bs->bl.has_variable_length) {
5993 int ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
5994 if (ret < 0) {
5995 return ret;
5996 }
5997 }
5998
5999 return bs->total_sectors;
6000 }
6001
6002 /**
6003 * Return length in bytes on success, -errno on error.
6004 * The length is always a multiple of BDRV_SECTOR_SIZE.
6005 */
6006 int64_t coroutine_fn bdrv_co_getlength(BlockDriverState *bs)
6007 {
6008 int64_t ret;
6009 IO_CODE();
6010 assert_bdrv_graph_readable();
6011
6012 ret = bdrv_co_nb_sectors(bs);
6013 if (ret < 0) {
6014 return ret;
6015 }
6016 if (ret > INT64_MAX / BDRV_SECTOR_SIZE) {
6017 return -EFBIG;
6018 }
6019 return ret * BDRV_SECTOR_SIZE;
6020 }
6021
6022 bool bdrv_is_sg(BlockDriverState *bs)
6023 {
6024 IO_CODE();
6025 return bs->sg;
6026 }
6027
6028 /**
6029 * Return whether the given node supports compressed writes.
6030 */
6031 bool bdrv_supports_compressed_writes(BlockDriverState *bs)
6032 {
6033 BlockDriverState *filtered;
6034 IO_CODE();
6035
6036 if (!bs->drv || !block_driver_can_compress(bs->drv)) {
6037 return false;
6038 }
6039
6040 filtered = bdrv_filter_bs(bs);
6041 if (filtered) {
6042 /*
6043 * Filters can only forward compressed writes, so we have to
6044 * check the child.
6045 */
6046 return bdrv_supports_compressed_writes(filtered);
6047 }
6048
6049 return true;
6050 }
6051
6052 const char *bdrv_get_format_name(BlockDriverState *bs)
6053 {
6054 IO_CODE();
6055 return bs->drv ? bs->drv->format_name : NULL;
6056 }
6057
6058 static int qsort_strcmp(const void *a, const void *b)
6059 {
6060 return strcmp(*(char *const *)a, *(char *const *)b);
6061 }
6062
6063 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
6064 void *opaque, bool read_only)
6065 {
6066 BlockDriver *drv;
6067 int count = 0;
6068 int i;
6069 const char **formats = NULL;
6070
6071 GLOBAL_STATE_CODE();
6072
6073 QLIST_FOREACH(drv, &bdrv_drivers, list) {
6074 if (drv->format_name) {
6075 bool found = false;
6076 int i = count;
6077
6078 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, read_only)) {
6079 continue;
6080 }
6081
6082 while (formats && i && !found) {
6083 found = !strcmp(formats[--i], drv->format_name);
6084 }
6085
6086 if (!found) {
6087 formats = g_renew(const char *, formats, count + 1);
6088 formats[count++] = drv->format_name;
6089 }
6090 }
6091 }
6092
6093 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); i++) {
6094 const char *format_name = block_driver_modules[i].format_name;
6095
6096 if (format_name) {
6097 bool found = false;
6098 int j = count;
6099
6100 if (use_bdrv_whitelist &&
6101 !bdrv_format_is_whitelisted(format_name, read_only)) {
6102 continue;
6103 }
6104
6105 while (formats && j && !found) {
6106 found = !strcmp(formats[--j], format_name);
6107 }
6108
6109 if (!found) {
6110 formats = g_renew(const char *, formats, count + 1);
6111 formats[count++] = format_name;
6112 }
6113 }
6114 }
6115
6116 qsort(formats, count, sizeof(formats[0]), qsort_strcmp);
6117
6118 for (i = 0; i < count; i++) {
6119 it(opaque, formats[i]);
6120 }
6121
6122 g_free(formats);
6123 }
6124
6125 /* This function is to find a node in the bs graph */
6126 BlockDriverState *bdrv_find_node(const char *node_name)
6127 {
6128 BlockDriverState *bs;
6129
6130 assert(node_name);
6131 GLOBAL_STATE_CODE();
6132
6133 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6134 if (!strcmp(node_name, bs->node_name)) {
6135 return bs;
6136 }
6137 }
6138 return NULL;
6139 }
6140
6141 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6142 BlockDeviceInfoList *bdrv_named_nodes_list(bool flat,
6143 Error **errp)
6144 {
6145 BlockDeviceInfoList *list;
6146 BlockDriverState *bs;
6147
6148 GLOBAL_STATE_CODE();
6149
6150 list = NULL;
6151 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6152 BlockDeviceInfo *info = bdrv_block_device_info(NULL, bs, flat, errp);
6153 if (!info) {
6154 qapi_free_BlockDeviceInfoList(list);
6155 return NULL;
6156 }
6157 QAPI_LIST_PREPEND(list, info);
6158 }
6159
6160 return list;
6161 }
6162
6163 typedef struct XDbgBlockGraphConstructor {
6164 XDbgBlockGraph *graph;
6165 GHashTable *graph_nodes;
6166 } XDbgBlockGraphConstructor;
6167
6168 static XDbgBlockGraphConstructor *xdbg_graph_new(void)
6169 {
6170 XDbgBlockGraphConstructor *gr = g_new(XDbgBlockGraphConstructor, 1);
6171
6172 gr->graph = g_new0(XDbgBlockGraph, 1);
6173 gr->graph_nodes = g_hash_table_new(NULL, NULL);
6174
6175 return gr;
6176 }
6177
6178 static XDbgBlockGraph *xdbg_graph_finalize(XDbgBlockGraphConstructor *gr)
6179 {
6180 XDbgBlockGraph *graph = gr->graph;
6181
6182 g_hash_table_destroy(gr->graph_nodes);
6183 g_free(gr);
6184
6185 return graph;
6186 }
6187
6188 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor *gr, void *node)
6189 {
6190 uintptr_t ret = (uintptr_t)g_hash_table_lookup(gr->graph_nodes, node);
6191
6192 if (ret != 0) {
6193 return ret;
6194 }
6195
6196 /*
6197 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6198 * answer of g_hash_table_lookup.
6199 */
6200 ret = g_hash_table_size(gr->graph_nodes) + 1;
6201 g_hash_table_insert(gr->graph_nodes, node, (void *)ret);
6202
6203 return ret;
6204 }
6205
6206 static void xdbg_graph_add_node(XDbgBlockGraphConstructor *gr, void *node,
6207 XDbgBlockGraphNodeType type, const char *name)
6208 {
6209 XDbgBlockGraphNode *n;
6210
6211 n = g_new0(XDbgBlockGraphNode, 1);
6212
6213 n->id = xdbg_graph_node_num(gr, node);
6214 n->type = type;
6215 n->name = g_strdup(name);
6216
6217 QAPI_LIST_PREPEND(gr->graph->nodes, n);
6218 }
6219
6220 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor *gr, void *parent,
6221 const BdrvChild *child)
6222 {
6223 BlockPermission qapi_perm;
6224 XDbgBlockGraphEdge *edge;
6225 GLOBAL_STATE_CODE();
6226
6227 edge = g_new0(XDbgBlockGraphEdge, 1);
6228
6229 edge->parent = xdbg_graph_node_num(gr, parent);
6230 edge->child = xdbg_graph_node_num(gr, child->bs);
6231 edge->name = g_strdup(child->name);
6232
6233 for (qapi_perm = 0; qapi_perm < BLOCK_PERMISSION__MAX; qapi_perm++) {
6234 uint64_t flag = bdrv_qapi_perm_to_blk_perm(qapi_perm);
6235
6236 if (flag & child->perm) {
6237 QAPI_LIST_PREPEND(edge->perm, qapi_perm);
6238 }
6239 if (flag & child->shared_perm) {
6240 QAPI_LIST_PREPEND(edge->shared_perm, qapi_perm);
6241 }
6242 }
6243
6244 QAPI_LIST_PREPEND(gr->graph->edges, edge);
6245 }
6246
6247
6248 XDbgBlockGraph *bdrv_get_xdbg_block_graph(Error **errp)
6249 {
6250 BlockBackend *blk;
6251 BlockJob *job;
6252 BlockDriverState *bs;
6253 BdrvChild *child;
6254 XDbgBlockGraphConstructor *gr = xdbg_graph_new();
6255
6256 GLOBAL_STATE_CODE();
6257
6258 for (blk = blk_all_next(NULL); blk; blk = blk_all_next(blk)) {
6259 char *allocated_name = NULL;
6260 const char *name = blk_name(blk);
6261
6262 if (!*name) {
6263 name = allocated_name = blk_get_attached_dev_id(blk);
6264 }
6265 xdbg_graph_add_node(gr, blk, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND,
6266 name);
6267 g_free(allocated_name);
6268 if (blk_root(blk)) {
6269 xdbg_graph_add_edge(gr, blk, blk_root(blk));
6270 }
6271 }
6272
6273 WITH_JOB_LOCK_GUARD() {
6274 for (job = block_job_next_locked(NULL); job;
6275 job = block_job_next_locked(job)) {
6276 GSList *el;
6277
6278 xdbg_graph_add_node(gr, job, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB,
6279 job->job.id);
6280 for (el = job->nodes; el; el = el->next) {
6281 xdbg_graph_add_edge(gr, job, (BdrvChild *)el->data);
6282 }
6283 }
6284 }
6285
6286 QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
6287 xdbg_graph_add_node(gr, bs, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER,
6288 bs->node_name);
6289 QLIST_FOREACH(child, &bs->children, next) {
6290 xdbg_graph_add_edge(gr, bs, child);
6291 }
6292 }
6293
6294 return xdbg_graph_finalize(gr);
6295 }
6296
6297 BlockDriverState *bdrv_lookup_bs(const char *device,
6298 const char *node_name,
6299 Error **errp)
6300 {
6301 BlockBackend *blk;
6302 BlockDriverState *bs;
6303
6304 GLOBAL_STATE_CODE();
6305
6306 if (device) {
6307 blk = blk_by_name(device);
6308
6309 if (blk) {
6310 bs = blk_bs(blk);
6311 if (!bs) {
6312 error_setg(errp, "Device '%s' has no medium", device);
6313 }
6314
6315 return bs;
6316 }
6317 }
6318
6319 if (node_name) {
6320 bs = bdrv_find_node(node_name);
6321
6322 if (bs) {
6323 return bs;
6324 }
6325 }
6326
6327 error_setg(errp, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6328 device ? device : "",
6329 node_name ? node_name : "");
6330 return NULL;
6331 }
6332
6333 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6334 * return false. If either argument is NULL, return false. */
6335 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base)
6336 {
6337
6338 GLOBAL_STATE_CODE();
6339
6340 while (top && top != base) {
6341 top = bdrv_filter_or_cow_bs(top);
6342 }
6343
6344 return top != NULL;
6345 }
6346
6347 BlockDriverState *bdrv_next_node(BlockDriverState *bs)
6348 {
6349 GLOBAL_STATE_CODE();
6350 if (!bs) {
6351 return QTAILQ_FIRST(&graph_bdrv_states);
6352 }
6353 return QTAILQ_NEXT(bs, node_list);
6354 }
6355
6356 BlockDriverState *bdrv_next_all_states(BlockDriverState *bs)
6357 {
6358 GLOBAL_STATE_CODE();
6359 if (!bs) {
6360 return QTAILQ_FIRST(&all_bdrv_states);
6361 }
6362 return QTAILQ_NEXT(bs, bs_list);
6363 }
6364
6365 const char *bdrv_get_node_name(const BlockDriverState *bs)
6366 {
6367 IO_CODE();
6368 return bs->node_name;
6369 }
6370
6371 const char *bdrv_get_parent_name(const BlockDriverState *bs)
6372 {
6373 BdrvChild *c;
6374 const char *name;
6375 IO_CODE();
6376
6377 /* If multiple parents have a name, just pick the first one. */
6378 QLIST_FOREACH(c, &bs->parents, next_parent) {
6379 if (c->klass->get_name) {
6380 name = c->klass->get_name(c);
6381 if (name && *name) {
6382 return name;
6383 }
6384 }
6385 }
6386
6387 return NULL;
6388 }
6389
6390 /* TODO check what callers really want: bs->node_name or blk_name() */
6391 const char *bdrv_get_device_name(const BlockDriverState *bs)
6392 {
6393 IO_CODE();
6394 return bdrv_get_parent_name(bs) ?: "";
6395 }
6396
6397 /* This can be used to identify nodes that might not have a device
6398 * name associated. Since node and device names live in the same
6399 * namespace, the result is unambiguous. The exception is if both are
6400 * absent, then this returns an empty (non-null) string. */
6401 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs)
6402 {
6403 IO_CODE();
6404 return bdrv_get_parent_name(bs) ?: bs->node_name;
6405 }
6406
6407 int bdrv_get_flags(BlockDriverState *bs)
6408 {
6409 IO_CODE();
6410 return bs->open_flags;
6411 }
6412
6413 int bdrv_has_zero_init_1(BlockDriverState *bs)
6414 {
6415 GLOBAL_STATE_CODE();
6416 return 1;
6417 }
6418
6419 int bdrv_has_zero_init(BlockDriverState *bs)
6420 {
6421 BlockDriverState *filtered;
6422 GLOBAL_STATE_CODE();
6423
6424 if (!bs->drv) {
6425 return 0;
6426 }
6427
6428 /* If BS is a copy on write image, it is initialized to
6429 the contents of the base image, which may not be zeroes. */
6430 if (bdrv_cow_child(bs)) {
6431 return 0;
6432 }
6433 if (bs->drv->bdrv_has_zero_init) {
6434 return bs->drv->bdrv_has_zero_init(bs);
6435 }
6436
6437 filtered = bdrv_filter_bs(bs);
6438 if (filtered) {
6439 return bdrv_has_zero_init(filtered);
6440 }
6441
6442 /* safe default */
6443 return 0;
6444 }
6445
6446 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
6447 {
6448 IO_CODE();
6449 if (!(bs->open_flags & BDRV_O_UNMAP)) {
6450 return false;
6451 }
6452
6453 return bs->supported_zero_flags & BDRV_REQ_MAY_UNMAP;
6454 }
6455
6456 void bdrv_get_backing_filename(BlockDriverState *bs,
6457 char *filename, int filename_size)
6458 {
6459 IO_CODE();
6460 pstrcpy(filename, filename_size, bs->backing_file);
6461 }
6462
6463 int coroutine_fn bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
6464 {
6465 int ret;
6466 BlockDriver *drv = bs->drv;
6467 IO_CODE();
6468 assert_bdrv_graph_readable();
6469
6470 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6471 if (!drv) {
6472 return -ENOMEDIUM;
6473 }
6474 if (!drv->bdrv_co_get_info) {
6475 BlockDriverState *filtered = bdrv_filter_bs(bs);
6476 if (filtered) {
6477 return bdrv_co_get_info(filtered, bdi);
6478 }
6479 return -ENOTSUP;
6480 }
6481 memset(bdi, 0, sizeof(*bdi));
6482 ret = drv->bdrv_co_get_info(bs, bdi);
6483 if (ret < 0) {
6484 return ret;
6485 }
6486
6487 if (bdi->cluster_size > BDRV_MAX_ALIGNMENT) {
6488 return -EINVAL;
6489 }
6490
6491 return 0;
6492 }
6493
6494 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs,
6495 Error **errp)
6496 {
6497 BlockDriver *drv = bs->drv;
6498 IO_CODE();
6499 if (drv && drv->bdrv_get_specific_info) {
6500 return drv->bdrv_get_specific_info(bs, errp);
6501 }
6502 return NULL;
6503 }
6504
6505 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs)
6506 {
6507 BlockDriver *drv = bs->drv;
6508 IO_CODE();
6509 if (!drv || !drv->bdrv_get_specific_stats) {
6510 return NULL;
6511 }
6512 return drv->bdrv_get_specific_stats(bs);
6513 }
6514
6515 void coroutine_fn bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event)
6516 {
6517 IO_CODE();
6518 assert_bdrv_graph_readable();
6519
6520 if (!bs || !bs->drv || !bs->drv->bdrv_co_debug_event) {
6521 return;
6522 }
6523
6524 bs->drv->bdrv_co_debug_event(bs, event);
6525 }
6526
6527 static BlockDriverState *bdrv_find_debug_node(BlockDriverState *bs)
6528 {
6529 GLOBAL_STATE_CODE();
6530 while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
6531 bs = bdrv_primary_bs(bs);
6532 }
6533
6534 if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
6535 assert(bs->drv->bdrv_debug_remove_breakpoint);
6536 return bs;
6537 }
6538
6539 return NULL;
6540 }
6541
6542 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
6543 const char *tag)
6544 {
6545 GLOBAL_STATE_CODE();
6546 bs = bdrv_find_debug_node(bs);
6547 if (bs) {
6548 return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
6549 }
6550
6551 return -ENOTSUP;
6552 }
6553
6554 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
6555 {
6556 GLOBAL_STATE_CODE();
6557 bs = bdrv_find_debug_node(bs);
6558 if (bs) {
6559 return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
6560 }
6561
6562 return -ENOTSUP;
6563 }
6564
6565 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
6566 {
6567 GLOBAL_STATE_CODE();
6568 while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
6569 bs = bdrv_primary_bs(bs);
6570 }
6571
6572 if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
6573 return bs->drv->bdrv_debug_resume(bs, tag);
6574 }
6575
6576 return -ENOTSUP;
6577 }
6578
6579 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
6580 {
6581 GLOBAL_STATE_CODE();
6582 while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
6583 bs = bdrv_primary_bs(bs);
6584 }
6585
6586 if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
6587 return bs->drv->bdrv_debug_is_suspended(bs, tag);
6588 }
6589
6590 return false;
6591 }
6592
6593 /* backing_file can either be relative, or absolute, or a protocol. If it is
6594 * relative, it must be relative to the chain. So, passing in bs->filename
6595 * from a BDS as backing_file should not be done, as that may be relative to
6596 * the CWD rather than the chain. */
6597 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
6598 const char *backing_file)
6599 {
6600 char *filename_full = NULL;
6601 char *backing_file_full = NULL;
6602 char *filename_tmp = NULL;
6603 int is_protocol = 0;
6604 bool filenames_refreshed = false;
6605 BlockDriverState *curr_bs = NULL;
6606 BlockDriverState *retval = NULL;
6607 BlockDriverState *bs_below;
6608
6609 GLOBAL_STATE_CODE();
6610
6611 if (!bs || !bs->drv || !backing_file) {
6612 return NULL;
6613 }
6614
6615 filename_full = g_malloc(PATH_MAX);
6616 backing_file_full = g_malloc(PATH_MAX);
6617
6618 is_protocol = path_has_protocol(backing_file);
6619
6620 /*
6621 * Being largely a legacy function, skip any filters here
6622 * (because filters do not have normal filenames, so they cannot
6623 * match anyway; and allowing json:{} filenames is a bit out of
6624 * scope).
6625 */
6626 for (curr_bs = bdrv_skip_filters(bs);
6627 bdrv_cow_child(curr_bs) != NULL;
6628 curr_bs = bs_below)
6629 {
6630 bs_below = bdrv_backing_chain_next(curr_bs);
6631
6632 if (bdrv_backing_overridden(curr_bs)) {
6633 /*
6634 * If the backing file was overridden, we can only compare
6635 * directly against the backing node's filename.
6636 */
6637
6638 if (!filenames_refreshed) {
6639 /*
6640 * This will automatically refresh all of the
6641 * filenames in the rest of the backing chain, so we
6642 * only need to do this once.
6643 */
6644 bdrv_refresh_filename(bs_below);
6645 filenames_refreshed = true;
6646 }
6647
6648 if (strcmp(backing_file, bs_below->filename) == 0) {
6649 retval = bs_below;
6650 break;
6651 }
6652 } else if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
6653 /*
6654 * If either of the filename paths is actually a protocol, then
6655 * compare unmodified paths; otherwise make paths relative.
6656 */
6657 char *backing_file_full_ret;
6658
6659 if (strcmp(backing_file, curr_bs->backing_file) == 0) {
6660 retval = bs_below;
6661 break;
6662 }
6663 /* Also check against the full backing filename for the image */
6664 backing_file_full_ret = bdrv_get_full_backing_filename(curr_bs,
6665 NULL);
6666 if (backing_file_full_ret) {
6667 bool equal = strcmp(backing_file, backing_file_full_ret) == 0;
6668 g_free(backing_file_full_ret);
6669 if (equal) {
6670 retval = bs_below;
6671 break;
6672 }
6673 }
6674 } else {
6675 /* If not an absolute filename path, make it relative to the current
6676 * image's filename path */
6677 filename_tmp = bdrv_make_absolute_filename(curr_bs, backing_file,
6678 NULL);
6679 /* We are going to compare canonicalized absolute pathnames */
6680 if (!filename_tmp || !realpath(filename_tmp, filename_full)) {
6681 g_free(filename_tmp);
6682 continue;
6683 }
6684 g_free(filename_tmp);
6685
6686 /* We need to make sure the backing filename we are comparing against
6687 * is relative to the current image filename (or absolute) */
6688 filename_tmp = bdrv_get_full_backing_filename(curr_bs, NULL);
6689 if (!filename_tmp || !realpath(filename_tmp, backing_file_full)) {
6690 g_free(filename_tmp);
6691 continue;
6692 }
6693 g_free(filename_tmp);
6694
6695 if (strcmp(backing_file_full, filename_full) == 0) {
6696 retval = bs_below;
6697 break;
6698 }
6699 }
6700 }
6701
6702 g_free(filename_full);
6703 g_free(backing_file_full);
6704 return retval;
6705 }
6706
6707 void bdrv_init(void)
6708 {
6709 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6710 use_bdrv_whitelist = 1;
6711 #endif
6712 module_call_init(MODULE_INIT_BLOCK);
6713 }
6714
6715 void bdrv_init_with_whitelist(void)
6716 {
6717 use_bdrv_whitelist = 1;
6718 bdrv_init();
6719 }
6720
6721 int bdrv_activate(BlockDriverState *bs, Error **errp)
6722 {
6723 BdrvChild *child, *parent;
6724 Error *local_err = NULL;
6725 int ret;
6726 BdrvDirtyBitmap *bm;
6727
6728 GLOBAL_STATE_CODE();
6729
6730 if (!bs->drv) {
6731 return -ENOMEDIUM;
6732 }
6733
6734 QLIST_FOREACH(child, &bs->children, next) {
6735 bdrv_activate(child->bs, &local_err);
6736 if (local_err) {
6737 error_propagate(errp, local_err);
6738 return -EINVAL;
6739 }
6740 }
6741
6742 /*
6743 * Update permissions, they may differ for inactive nodes.
6744 *
6745 * Note that the required permissions of inactive images are always a
6746 * subset of the permissions required after activating the image. This
6747 * allows us to just get the permissions upfront without restricting
6748 * bdrv_co_invalidate_cache().
6749 *
6750 * It also means that in error cases, we don't have to try and revert to
6751 * the old permissions (which is an operation that could fail, too). We can
6752 * just keep the extended permissions for the next time that an activation
6753 * of the image is tried.
6754 */
6755 if (bs->open_flags & BDRV_O_INACTIVE) {
6756 bs->open_flags &= ~BDRV_O_INACTIVE;
6757 ret = bdrv_refresh_perms(bs, NULL, errp);
6758 if (ret < 0) {
6759 bs->open_flags |= BDRV_O_INACTIVE;
6760 return ret;
6761 }
6762
6763 ret = bdrv_invalidate_cache(bs, errp);
6764 if (ret < 0) {
6765 bs->open_flags |= BDRV_O_INACTIVE;
6766 return ret;
6767 }
6768
6769 FOR_EACH_DIRTY_BITMAP(bs, bm) {
6770 bdrv_dirty_bitmap_skip_store(bm, false);
6771 }
6772
6773 ret = bdrv_refresh_total_sectors(bs, bs->total_sectors);
6774 if (ret < 0) {
6775 bs->open_flags |= BDRV_O_INACTIVE;
6776 error_setg_errno(errp, -ret, "Could not refresh total sector count");
6777 return ret;
6778 }
6779 }
6780
6781 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6782 if (parent->klass->activate) {
6783 parent->klass->activate(parent, &local_err);
6784 if (local_err) {
6785 bs->open_flags |= BDRV_O_INACTIVE;
6786 error_propagate(errp, local_err);
6787 return -EINVAL;
6788 }
6789 }
6790 }
6791
6792 return 0;
6793 }
6794
6795 int coroutine_fn bdrv_co_invalidate_cache(BlockDriverState *bs, Error **errp)
6796 {
6797 Error *local_err = NULL;
6798 IO_CODE();
6799
6800 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6801 assert_bdrv_graph_readable();
6802
6803 if (bs->drv->bdrv_co_invalidate_cache) {
6804 bs->drv->bdrv_co_invalidate_cache(bs, &local_err);
6805 if (local_err) {
6806 error_propagate(errp, local_err);
6807 return -EINVAL;
6808 }
6809 }
6810
6811 return 0;
6812 }
6813
6814 void bdrv_activate_all(Error **errp)
6815 {
6816 BlockDriverState *bs;
6817 BdrvNextIterator it;
6818
6819 GLOBAL_STATE_CODE();
6820
6821 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6822 AioContext *aio_context = bdrv_get_aio_context(bs);
6823 int ret;
6824
6825 aio_context_acquire(aio_context);
6826 ret = bdrv_activate(bs, errp);
6827 aio_context_release(aio_context);
6828 if (ret < 0) {
6829 bdrv_next_cleanup(&it);
6830 return;
6831 }
6832 }
6833 }
6834
6835 static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
6836 {
6837 BdrvChild *parent;
6838 GLOBAL_STATE_CODE();
6839
6840 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6841 if (parent->klass->parent_is_bds) {
6842 BlockDriverState *parent_bs = parent->opaque;
6843 if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
6844 return true;
6845 }
6846 }
6847 }
6848
6849 return false;
6850 }
6851
6852 static int bdrv_inactivate_recurse(BlockDriverState *bs)
6853 {
6854 BdrvChild *child, *parent;
6855 int ret;
6856 uint64_t cumulative_perms, cumulative_shared_perms;
6857
6858 GLOBAL_STATE_CODE();
6859
6860 if (!bs->drv) {
6861 return -ENOMEDIUM;
6862 }
6863
6864 /* Make sure that we don't inactivate a child before its parent.
6865 * It will be covered by recursion from the yet active parent. */
6866 if (bdrv_has_bds_parent(bs, true)) {
6867 return 0;
6868 }
6869
6870 assert(!(bs->open_flags & BDRV_O_INACTIVE));
6871
6872 /* Inactivate this node */
6873 if (bs->drv->bdrv_inactivate) {
6874 ret = bs->drv->bdrv_inactivate(bs);
6875 if (ret < 0) {
6876 return ret;
6877 }
6878 }
6879
6880 QLIST_FOREACH(parent, &bs->parents, next_parent) {
6881 if (parent->klass->inactivate) {
6882 ret = parent->klass->inactivate(parent);
6883 if (ret < 0) {
6884 return ret;
6885 }
6886 }
6887 }
6888
6889 bdrv_get_cumulative_perm(bs, &cumulative_perms,
6890 &cumulative_shared_perms);
6891 if (cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
6892 /* Our inactive parents still need write access. Inactivation failed. */
6893 return -EPERM;
6894 }
6895
6896 bs->open_flags |= BDRV_O_INACTIVE;
6897
6898 /*
6899 * Update permissions, they may differ for inactive nodes.
6900 * We only tried to loosen restrictions, so errors are not fatal, ignore
6901 * them.
6902 */
6903 bdrv_refresh_perms(bs, NULL, NULL);
6904
6905 /* Recursively inactivate children */
6906 QLIST_FOREACH(child, &bs->children, next) {
6907 ret = bdrv_inactivate_recurse(child->bs);
6908 if (ret < 0) {
6909 return ret;
6910 }
6911 }
6912
6913 return 0;
6914 }
6915
6916 int bdrv_inactivate_all(void)
6917 {
6918 BlockDriverState *bs = NULL;
6919 BdrvNextIterator it;
6920 int ret = 0;
6921 GSList *aio_ctxs = NULL, *ctx;
6922
6923 GLOBAL_STATE_CODE();
6924
6925 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6926 AioContext *aio_context = bdrv_get_aio_context(bs);
6927
6928 if (!g_slist_find(aio_ctxs, aio_context)) {
6929 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
6930 aio_context_acquire(aio_context);
6931 }
6932 }
6933
6934 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
6935 /* Nodes with BDS parents are covered by recursion from the last
6936 * parent that gets inactivated. Don't inactivate them a second
6937 * time if that has already happened. */
6938 if (bdrv_has_bds_parent(bs, false)) {
6939 continue;
6940 }
6941 ret = bdrv_inactivate_recurse(bs);
6942 if (ret < 0) {
6943 bdrv_next_cleanup(&it);
6944 goto out;
6945 }
6946 }
6947
6948 out:
6949 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
6950 AioContext *aio_context = ctx->data;
6951 aio_context_release(aio_context);
6952 }
6953 g_slist_free(aio_ctxs);
6954
6955 return ret;
6956 }
6957
6958 /**************************************************************/
6959 /* removable device support */
6960
6961 /**
6962 * Return TRUE if the media is present
6963 */
6964 bool coroutine_fn bdrv_co_is_inserted(BlockDriverState *bs)
6965 {
6966 BlockDriver *drv = bs->drv;
6967 BdrvChild *child;
6968 IO_CODE();
6969 assert_bdrv_graph_readable();
6970
6971 if (!drv) {
6972 return false;
6973 }
6974 if (drv->bdrv_co_is_inserted) {
6975 return drv->bdrv_co_is_inserted(bs);
6976 }
6977 QLIST_FOREACH(child, &bs->children, next) {
6978 if (!bdrv_co_is_inserted(child->bs)) {
6979 return false;
6980 }
6981 }
6982 return true;
6983 }
6984
6985 /**
6986 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6987 */
6988 void coroutine_fn bdrv_co_eject(BlockDriverState *bs, bool eject_flag)
6989 {
6990 BlockDriver *drv = bs->drv;
6991 IO_CODE();
6992 assert_bdrv_graph_readable();
6993
6994 if (drv && drv->bdrv_co_eject) {
6995 drv->bdrv_co_eject(bs, eject_flag);
6996 }
6997 }
6998
6999 /**
7000 * Lock or unlock the media (if it is locked, the user won't be able
7001 * to eject it manually).
7002 */
7003 void coroutine_fn bdrv_co_lock_medium(BlockDriverState *bs, bool locked)
7004 {
7005 BlockDriver *drv = bs->drv;
7006 IO_CODE();
7007 assert_bdrv_graph_readable();
7008 trace_bdrv_lock_medium(bs, locked);
7009
7010 if (drv && drv->bdrv_co_lock_medium) {
7011 drv->bdrv_co_lock_medium(bs, locked);
7012 }
7013 }
7014
7015 /* Get a reference to bs */
7016 void bdrv_ref(BlockDriverState *bs)
7017 {
7018 GLOBAL_STATE_CODE();
7019 bs->refcnt++;
7020 }
7021
7022 /* Release a previously grabbed reference to bs.
7023 * If after releasing, reference count is zero, the BlockDriverState is
7024 * deleted. */
7025 void bdrv_unref(BlockDriverState *bs)
7026 {
7027 GLOBAL_STATE_CODE();
7028 if (!bs) {
7029 return;
7030 }
7031 assert(bs->refcnt > 0);
7032 if (--bs->refcnt == 0) {
7033 bdrv_delete(bs);
7034 }
7035 }
7036
7037 struct BdrvOpBlocker {
7038 Error *reason;
7039 QLIST_ENTRY(BdrvOpBlocker) list;
7040 };
7041
7042 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
7043 {
7044 BdrvOpBlocker *blocker;
7045 GLOBAL_STATE_CODE();
7046 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7047 if (!QLIST_EMPTY(&bs->op_blockers[op])) {
7048 blocker = QLIST_FIRST(&bs->op_blockers[op]);
7049 error_propagate_prepend(errp, error_copy(blocker->reason),
7050 "Node '%s' is busy: ",
7051 bdrv_get_device_or_node_name(bs));
7052 return true;
7053 }
7054 return false;
7055 }
7056
7057 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
7058 {
7059 BdrvOpBlocker *blocker;
7060 GLOBAL_STATE_CODE();
7061 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7062
7063 blocker = g_new0(BdrvOpBlocker, 1);
7064 blocker->reason = reason;
7065 QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
7066 }
7067
7068 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
7069 {
7070 BdrvOpBlocker *blocker, *next;
7071 GLOBAL_STATE_CODE();
7072 assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
7073 QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
7074 if (blocker->reason == reason) {
7075 QLIST_REMOVE(blocker, list);
7076 g_free(blocker);
7077 }
7078 }
7079 }
7080
7081 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
7082 {
7083 int i;
7084 GLOBAL_STATE_CODE();
7085 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7086 bdrv_op_block(bs, i, reason);
7087 }
7088 }
7089
7090 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
7091 {
7092 int i;
7093 GLOBAL_STATE_CODE();
7094 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7095 bdrv_op_unblock(bs, i, reason);
7096 }
7097 }
7098
7099 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
7100 {
7101 int i;
7102 GLOBAL_STATE_CODE();
7103 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
7104 if (!QLIST_EMPTY(&bs->op_blockers[i])) {
7105 return false;
7106 }
7107 }
7108 return true;
7109 }
7110
7111 /*
7112 * Must not be called while holding the lock of an AioContext other than the
7113 * current one.
7114 */
7115 void bdrv_img_create(const char *filename, const char *fmt,
7116 const char *base_filename, const char *base_fmt,
7117 char *options, uint64_t img_size, int flags, bool quiet,
7118 Error **errp)
7119 {
7120 QemuOptsList *create_opts = NULL;
7121 QemuOpts *opts = NULL;
7122 const char *backing_fmt, *backing_file;
7123 int64_t size;
7124 BlockDriver *drv, *proto_drv;
7125 Error *local_err = NULL;
7126 int ret = 0;
7127
7128 GLOBAL_STATE_CODE();
7129
7130 /* Find driver and parse its options */
7131 drv = bdrv_find_format(fmt);
7132 if (!drv) {
7133 error_setg(errp, "Unknown file format '%s'", fmt);
7134 return;
7135 }
7136
7137 proto_drv = bdrv_find_protocol(filename, true, errp);
7138 if (!proto_drv) {
7139 return;
7140 }
7141
7142 if (!drv->create_opts) {
7143 error_setg(errp, "Format driver '%s' does not support image creation",
7144 drv->format_name);
7145 return;
7146 }
7147
7148 if (!proto_drv->create_opts) {
7149 error_setg(errp, "Protocol driver '%s' does not support image creation",
7150 proto_drv->format_name);
7151 return;
7152 }
7153
7154 aio_context_acquire(qemu_get_aio_context());
7155
7156 /* Create parameter list */
7157 create_opts = qemu_opts_append(create_opts, drv->create_opts);
7158 create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
7159
7160 opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
7161
7162 /* Parse -o options */
7163 if (options) {
7164 if (!qemu_opts_do_parse(opts, options, NULL, errp)) {
7165 goto out;
7166 }
7167 }
7168
7169 if (!qemu_opt_get(opts, BLOCK_OPT_SIZE)) {
7170 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
7171 } else if (img_size != UINT64_C(-1)) {
7172 error_setg(errp, "The image size must be specified only once");
7173 goto out;
7174 }
7175
7176 if (base_filename) {
7177 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
7178 NULL)) {
7179 error_setg(errp, "Backing file not supported for file format '%s'",
7180 fmt);
7181 goto out;
7182 }
7183 }
7184
7185 if (base_fmt) {
7186 if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
7187 error_setg(errp, "Backing file format not supported for file "
7188 "format '%s'", fmt);
7189 goto out;
7190 }
7191 }
7192
7193 backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
7194 if (backing_file) {
7195 if (!strcmp(filename, backing_file)) {
7196 error_setg(errp, "Error: Trying to create an image with the "
7197 "same filename as the backing file");
7198 goto out;
7199 }
7200 if (backing_file[0] == '\0') {
7201 error_setg(errp, "Expected backing file name, got empty string");
7202 goto out;
7203 }
7204 }
7205
7206 backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
7207
7208 /* The size for the image must always be specified, unless we have a backing
7209 * file and we have not been forbidden from opening it. */
7210 size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, img_size);
7211 if (backing_file && !(flags & BDRV_O_NO_BACKING)) {
7212 BlockDriverState *bs;
7213 char *full_backing;
7214 int back_flags;
7215 QDict *backing_options = NULL;
7216
7217 full_backing =
7218 bdrv_get_full_backing_filename_from_filename(filename, backing_file,
7219 &local_err);
7220 if (local_err) {
7221 goto out;
7222 }
7223 assert(full_backing);
7224
7225 /*
7226 * No need to do I/O here, which allows us to open encrypted
7227 * backing images without needing the secret
7228 */
7229 back_flags = flags;
7230 back_flags &= ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
7231 back_flags |= BDRV_O_NO_IO;
7232
7233 backing_options = qdict_new();
7234 if (backing_fmt) {
7235 qdict_put_str(backing_options, "driver", backing_fmt);
7236 }
7237 qdict_put_bool(backing_options, BDRV_OPT_FORCE_SHARE, true);
7238
7239 bs = bdrv_open(full_backing, NULL, backing_options, back_flags,
7240 &local_err);
7241 g_free(full_backing);
7242 if (!bs) {
7243 error_append_hint(&local_err, "Could not open backing image.\n");
7244 goto out;
7245 } else {
7246 if (!backing_fmt) {
7247 error_setg(&local_err,
7248 "Backing file specified without backing format");
7249 error_append_hint(&local_err, "Detected format of %s.\n",
7250 bs->drv->format_name);
7251 goto out;
7252 }
7253 if (size == -1) {
7254 /* Opened BS, have no size */
7255 size = bdrv_getlength(bs);
7256 if (size < 0) {
7257 error_setg_errno(errp, -size, "Could not get size of '%s'",
7258 backing_file);
7259 bdrv_unref(bs);
7260 goto out;
7261 }
7262 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size, &error_abort);
7263 }
7264 bdrv_unref(bs);
7265 }
7266 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7267 } else if (backing_file && !backing_fmt) {
7268 error_setg(&local_err,
7269 "Backing file specified without backing format");
7270 goto out;
7271 }
7272
7273 if (size == -1) {
7274 error_setg(errp, "Image creation needs a size parameter");
7275 goto out;
7276 }
7277
7278 if (!quiet) {
7279 printf("Formatting '%s', fmt=%s ", filename, fmt);
7280 qemu_opts_print(opts, " ");
7281 puts("");
7282 fflush(stdout);
7283 }
7284
7285 ret = bdrv_create(drv, filename, opts, &local_err);
7286
7287 if (ret == -EFBIG) {
7288 /* This is generally a better message than whatever the driver would
7289 * deliver (especially because of the cluster_size_hint), since that
7290 * is most probably not much different from "image too large". */
7291 const char *cluster_size_hint = "";
7292 if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
7293 cluster_size_hint = " (try using a larger cluster size)";
7294 }
7295 error_setg(errp, "The image size is too large for file format '%s'"
7296 "%s", fmt, cluster_size_hint);
7297 error_free(local_err);
7298 local_err = NULL;
7299 }
7300
7301 out:
7302 qemu_opts_del(opts);
7303 qemu_opts_free(create_opts);
7304 error_propagate(errp, local_err);
7305 aio_context_release(qemu_get_aio_context());
7306 }
7307
7308 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
7309 {
7310 IO_CODE();
7311 return bs ? bs->aio_context : qemu_get_aio_context();
7312 }
7313
7314 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs)
7315 {
7316 Coroutine *self = qemu_coroutine_self();
7317 AioContext *old_ctx = qemu_coroutine_get_aio_context(self);
7318 AioContext *new_ctx;
7319 IO_CODE();
7320
7321 /*
7322 * Increase bs->in_flight to ensure that this operation is completed before
7323 * moving the node to a different AioContext. Read new_ctx only afterwards.
7324 */
7325 bdrv_inc_in_flight(bs);
7326
7327 new_ctx = bdrv_get_aio_context(bs);
7328 aio_co_reschedule_self(new_ctx);
7329 return old_ctx;
7330 }
7331
7332 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx)
7333 {
7334 IO_CODE();
7335 aio_co_reschedule_self(old_ctx);
7336 bdrv_dec_in_flight(bs);
7337 }
7338
7339 void coroutine_fn bdrv_co_lock(BlockDriverState *bs)
7340 {
7341 AioContext *ctx = bdrv_get_aio_context(bs);
7342
7343 /* In the main thread, bs->aio_context won't change concurrently */
7344 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7345
7346 /*
7347 * We're in coroutine context, so we already hold the lock of the main
7348 * loop AioContext. Don't lock it twice to avoid deadlocks.
7349 */
7350 assert(qemu_in_coroutine());
7351 if (ctx != qemu_get_aio_context()) {
7352 aio_context_acquire(ctx);
7353 }
7354 }
7355
7356 void coroutine_fn bdrv_co_unlock(BlockDriverState *bs)
7357 {
7358 AioContext *ctx = bdrv_get_aio_context(bs);
7359
7360 assert(qemu_in_coroutine());
7361 if (ctx != qemu_get_aio_context()) {
7362 aio_context_release(ctx);
7363 }
7364 }
7365
7366 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier *ban)
7367 {
7368 GLOBAL_STATE_CODE();
7369 QLIST_REMOVE(ban, list);
7370 g_free(ban);
7371 }
7372
7373 static void bdrv_detach_aio_context(BlockDriverState *bs)
7374 {
7375 BdrvAioNotifier *baf, *baf_tmp;
7376
7377 assert(!bs->walking_aio_notifiers);
7378 GLOBAL_STATE_CODE();
7379 bs->walking_aio_notifiers = true;
7380 QLIST_FOREACH_SAFE(baf, &bs->aio_notifiers, list, baf_tmp) {
7381 if (baf->deleted) {
7382 bdrv_do_remove_aio_context_notifier(baf);
7383 } else {
7384 baf->detach_aio_context(baf->opaque);
7385 }
7386 }
7387 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7388 * remove remaining aio notifiers if we aren't called again.
7389 */
7390 bs->walking_aio_notifiers = false;
7391
7392 if (bs->drv && bs->drv->bdrv_detach_aio_context) {
7393 bs->drv->bdrv_detach_aio_context(bs);
7394 }
7395
7396 bs->aio_context = NULL;
7397 }
7398
7399 static void bdrv_attach_aio_context(BlockDriverState *bs,
7400 AioContext *new_context)
7401 {
7402 BdrvAioNotifier *ban, *ban_tmp;
7403 GLOBAL_STATE_CODE();
7404
7405 bs->aio_context = new_context;
7406
7407 if (bs->drv && bs->drv->bdrv_attach_aio_context) {
7408 bs->drv->bdrv_attach_aio_context(bs, new_context);
7409 }
7410
7411 assert(!bs->walking_aio_notifiers);
7412 bs->walking_aio_notifiers = true;
7413 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_tmp) {
7414 if (ban->deleted) {
7415 bdrv_do_remove_aio_context_notifier(ban);
7416 } else {
7417 ban->attached_aio_context(new_context, ban->opaque);
7418 }
7419 }
7420 bs->walking_aio_notifiers = false;
7421 }
7422
7423 typedef struct BdrvStateSetAioContext {
7424 AioContext *new_ctx;
7425 BlockDriverState *bs;
7426 } BdrvStateSetAioContext;
7427
7428 static bool bdrv_parent_change_aio_context(BdrvChild *c, AioContext *ctx,
7429 GHashTable *visited,
7430 Transaction *tran,
7431 Error **errp)
7432 {
7433 GLOBAL_STATE_CODE();
7434 if (g_hash_table_contains(visited, c)) {
7435 return true;
7436 }
7437 g_hash_table_add(visited, c);
7438
7439 /*
7440 * A BdrvChildClass that doesn't handle AioContext changes cannot
7441 * tolerate any AioContext changes
7442 */
7443 if (!c->klass->change_aio_ctx) {
7444 char *user = bdrv_child_user_desc(c);
7445 error_setg(errp, "Changing iothreads is not supported by %s", user);
7446 g_free(user);
7447 return false;
7448 }
7449 if (!c->klass->change_aio_ctx(c, ctx, visited, tran, errp)) {
7450 assert(!errp || *errp);
7451 return false;
7452 }
7453 return true;
7454 }
7455
7456 bool bdrv_child_change_aio_context(BdrvChild *c, AioContext *ctx,
7457 GHashTable *visited, Transaction *tran,
7458 Error **errp)
7459 {
7460 GLOBAL_STATE_CODE();
7461 if (g_hash_table_contains(visited, c)) {
7462 return true;
7463 }
7464 g_hash_table_add(visited, c);
7465 return bdrv_change_aio_context(c->bs, ctx, visited, tran, errp);
7466 }
7467
7468 static void bdrv_set_aio_context_clean(void *opaque)
7469 {
7470 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7471 BlockDriverState *bs = (BlockDriverState *) state->bs;
7472
7473 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7474 bdrv_drained_end(bs);
7475
7476 g_free(state);
7477 }
7478
7479 static void bdrv_set_aio_context_commit(void *opaque)
7480 {
7481 BdrvStateSetAioContext *state = (BdrvStateSetAioContext *) opaque;
7482 BlockDriverState *bs = (BlockDriverState *) state->bs;
7483 AioContext *new_context = state->new_ctx;
7484 AioContext *old_context = bdrv_get_aio_context(bs);
7485
7486 /*
7487 * Take the old AioContex when detaching it from bs.
7488 * At this point, new_context lock is already acquired, and we are now
7489 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7490 * does not call AIO_POLL_WHILE().
7491 */
7492 if (old_context != qemu_get_aio_context()) {
7493 aio_context_acquire(old_context);
7494 }
7495 bdrv_detach_aio_context(bs);
7496 if (old_context != qemu_get_aio_context()) {
7497 aio_context_release(old_context);
7498 }
7499 bdrv_attach_aio_context(bs, new_context);
7500 }
7501
7502 static TransactionActionDrv set_aio_context = {
7503 .commit = bdrv_set_aio_context_commit,
7504 .clean = bdrv_set_aio_context_clean,
7505 };
7506
7507 /*
7508 * Changes the AioContext used for fd handlers, timers, and BHs by this
7509 * BlockDriverState and all its children and parents.
7510 *
7511 * Must be called from the main AioContext.
7512 *
7513 * The caller must own the AioContext lock for the old AioContext of bs, but it
7514 * must not own the AioContext lock for new_context (unless new_context is the
7515 * same as the current context of bs).
7516 *
7517 * @visited will accumulate all visited BdrvChild objects. The caller is
7518 * responsible for freeing the list afterwards.
7519 */
7520 static bool bdrv_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7521 GHashTable *visited, Transaction *tran,
7522 Error **errp)
7523 {
7524 BdrvChild *c;
7525 BdrvStateSetAioContext *state;
7526
7527 GLOBAL_STATE_CODE();
7528
7529 if (bdrv_get_aio_context(bs) == ctx) {
7530 return true;
7531 }
7532
7533 QLIST_FOREACH(c, &bs->parents, next_parent) {
7534 if (!bdrv_parent_change_aio_context(c, ctx, visited, tran, errp)) {
7535 return false;
7536 }
7537 }
7538
7539 QLIST_FOREACH(c, &bs->children, next) {
7540 if (!bdrv_child_change_aio_context(c, ctx, visited, tran, errp)) {
7541 return false;
7542 }
7543 }
7544
7545 state = g_new(BdrvStateSetAioContext, 1);
7546 *state = (BdrvStateSetAioContext) {
7547 .new_ctx = ctx,
7548 .bs = bs,
7549 };
7550
7551 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7552 bdrv_drained_begin(bs);
7553
7554 tran_add(tran, &set_aio_context, state);
7555
7556 return true;
7557 }
7558
7559 /*
7560 * Change bs's and recursively all of its parents' and children's AioContext
7561 * to the given new context, returning an error if that isn't possible.
7562 *
7563 * If ignore_child is not NULL, that child (and its subgraph) will not
7564 * be touched.
7565 *
7566 * This function still requires the caller to take the bs current
7567 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7568 * assumes the lock is always held if bs is in another AioContext.
7569 * For the same reason, it temporarily also holds the new AioContext, since
7570 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7571 * Therefore the new AioContext lock must not be taken by the caller.
7572 */
7573 int bdrv_try_change_aio_context(BlockDriverState *bs, AioContext *ctx,
7574 BdrvChild *ignore_child, Error **errp)
7575 {
7576 Transaction *tran;
7577 GHashTable *visited;
7578 int ret;
7579 AioContext *old_context = bdrv_get_aio_context(bs);
7580 GLOBAL_STATE_CODE();
7581
7582 /*
7583 * Recursion phase: go through all nodes of the graph.
7584 * Take care of checking that all nodes support changing AioContext
7585 * and drain them, builing a linear list of callbacks to run if everything
7586 * is successful (the transaction itself).
7587 */
7588 tran = tran_new();
7589 visited = g_hash_table_new(NULL, NULL);
7590 if (ignore_child) {
7591 g_hash_table_add(visited, ignore_child);
7592 }
7593 ret = bdrv_change_aio_context(bs, ctx, visited, tran, errp);
7594 g_hash_table_destroy(visited);
7595
7596 /*
7597 * Linear phase: go through all callbacks collected in the transaction.
7598 * Run all callbacks collected in the recursion to switch all nodes
7599 * AioContext lock (transaction commit), or undo all changes done in the
7600 * recursion (transaction abort).
7601 */
7602
7603 if (!ret) {
7604 /* Just run clean() callbacks. No AioContext changed. */
7605 tran_abort(tran);
7606 return -EPERM;
7607 }
7608
7609 /*
7610 * Release old AioContext, it won't be needed anymore, as all
7611 * bdrv_drained_begin() have been called already.
7612 */
7613 if (qemu_get_aio_context() != old_context) {
7614 aio_context_release(old_context);
7615 }
7616
7617 /*
7618 * Acquire new AioContext since bdrv_drained_end() is going to be called
7619 * after we switched all nodes in the new AioContext, and the function
7620 * assumes that the lock of the bs is always taken.
7621 */
7622 if (qemu_get_aio_context() != ctx) {
7623 aio_context_acquire(ctx);
7624 }
7625
7626 tran_commit(tran);
7627
7628 if (qemu_get_aio_context() != ctx) {
7629 aio_context_release(ctx);
7630 }
7631
7632 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7633 if (qemu_get_aio_context() != old_context) {
7634 aio_context_acquire(old_context);
7635 }
7636
7637 return 0;
7638 }
7639
7640 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
7641 void (*attached_aio_context)(AioContext *new_context, void *opaque),
7642 void (*detach_aio_context)(void *opaque), void *opaque)
7643 {
7644 BdrvAioNotifier *ban = g_new(BdrvAioNotifier, 1);
7645 *ban = (BdrvAioNotifier){
7646 .attached_aio_context = attached_aio_context,
7647 .detach_aio_context = detach_aio_context,
7648 .opaque = opaque
7649 };
7650 GLOBAL_STATE_CODE();
7651
7652 QLIST_INSERT_HEAD(&bs->aio_notifiers, ban, list);
7653 }
7654
7655 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
7656 void (*attached_aio_context)(AioContext *,
7657 void *),
7658 void (*detach_aio_context)(void *),
7659 void *opaque)
7660 {
7661 BdrvAioNotifier *ban, *ban_next;
7662 GLOBAL_STATE_CODE();
7663
7664 QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
7665 if (ban->attached_aio_context == attached_aio_context &&
7666 ban->detach_aio_context == detach_aio_context &&
7667 ban->opaque == opaque &&
7668 ban->deleted == false)
7669 {
7670 if (bs->walking_aio_notifiers) {
7671 ban->deleted = true;
7672 } else {
7673 bdrv_do_remove_aio_context_notifier(ban);
7674 }
7675 return;
7676 }
7677 }
7678
7679 abort();
7680 }
7681
7682 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
7683 BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
7684 bool force,
7685 Error **errp)
7686 {
7687 GLOBAL_STATE_CODE();
7688 if (!bs->drv) {
7689 error_setg(errp, "Node is ejected");
7690 return -ENOMEDIUM;
7691 }
7692 if (!bs->drv->bdrv_amend_options) {
7693 error_setg(errp, "Block driver '%s' does not support option amendment",
7694 bs->drv->format_name);
7695 return -ENOTSUP;
7696 }
7697 return bs->drv->bdrv_amend_options(bs, opts, status_cb,
7698 cb_opaque, force, errp);
7699 }
7700
7701 /*
7702 * This function checks whether the given @to_replace is allowed to be
7703 * replaced by a node that always shows the same data as @bs. This is
7704 * used for example to verify whether the mirror job can replace
7705 * @to_replace by the target mirrored from @bs.
7706 * To be replaceable, @bs and @to_replace may either be guaranteed to
7707 * always show the same data (because they are only connected through
7708 * filters), or some driver may allow replacing one of its children
7709 * because it can guarantee that this child's data is not visible at
7710 * all (for example, for dissenting quorum children that have no other
7711 * parents).
7712 */
7713 bool bdrv_recurse_can_replace(BlockDriverState *bs,
7714 BlockDriverState *to_replace)
7715 {
7716 BlockDriverState *filtered;
7717
7718 GLOBAL_STATE_CODE();
7719
7720 if (!bs || !bs->drv) {
7721 return false;
7722 }
7723
7724 if (bs == to_replace) {
7725 return true;
7726 }
7727
7728 /* See what the driver can do */
7729 if (bs->drv->bdrv_recurse_can_replace) {
7730 return bs->drv->bdrv_recurse_can_replace(bs, to_replace);
7731 }
7732
7733 /* For filters without an own implementation, we can recurse on our own */
7734 filtered = bdrv_filter_bs(bs);
7735 if (filtered) {
7736 return bdrv_recurse_can_replace(filtered, to_replace);
7737 }
7738
7739 /* Safe default */
7740 return false;
7741 }
7742
7743 /*
7744 * Check whether the given @node_name can be replaced by a node that
7745 * has the same data as @parent_bs. If so, return @node_name's BDS;
7746 * NULL otherwise.
7747 *
7748 * @node_name must be a (recursive) *child of @parent_bs (or this
7749 * function will return NULL).
7750 *
7751 * The result (whether the node can be replaced or not) is only valid
7752 * for as long as no graph or permission changes occur.
7753 */
7754 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
7755 const char *node_name, Error **errp)
7756 {
7757 BlockDriverState *to_replace_bs = bdrv_find_node(node_name);
7758 AioContext *aio_context;
7759
7760 GLOBAL_STATE_CODE();
7761
7762 if (!to_replace_bs) {
7763 error_setg(errp, "Failed to find node with node-name='%s'", node_name);
7764 return NULL;
7765 }
7766
7767 aio_context = bdrv_get_aio_context(to_replace_bs);
7768 aio_context_acquire(aio_context);
7769
7770 if (bdrv_op_is_blocked(to_replace_bs, BLOCK_OP_TYPE_REPLACE, errp)) {
7771 to_replace_bs = NULL;
7772 goto out;
7773 }
7774
7775 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7776 * most non filter in order to prevent data corruption.
7777 * Another benefit is that this tests exclude backing files which are
7778 * blocked by the backing blockers.
7779 */
7780 if (!bdrv_recurse_can_replace(parent_bs, to_replace_bs)) {
7781 error_setg(errp, "Cannot replace '%s' by a node mirrored from '%s', "
7782 "because it cannot be guaranteed that doing so would not "
7783 "lead to an abrupt change of visible data",
7784 node_name, parent_bs->node_name);
7785 to_replace_bs = NULL;
7786 goto out;
7787 }
7788
7789 out:
7790 aio_context_release(aio_context);
7791 return to_replace_bs;
7792 }
7793
7794 /**
7795 * Iterates through the list of runtime option keys that are said to
7796 * be "strong" for a BDS. An option is called "strong" if it changes
7797 * a BDS's data. For example, the null block driver's "size" and
7798 * "read-zeroes" options are strong, but its "latency-ns" option is
7799 * not.
7800 *
7801 * If a key returned by this function ends with a dot, all options
7802 * starting with that prefix are strong.
7803 */
7804 static const char *const *strong_options(BlockDriverState *bs,
7805 const char *const *curopt)
7806 {
7807 static const char *const global_options[] = {
7808 "driver", "filename", NULL
7809 };
7810
7811 if (!curopt) {
7812 return &global_options[0];
7813 }
7814
7815 curopt++;
7816 if (curopt == &global_options[ARRAY_SIZE(global_options) - 1] && bs->drv) {
7817 curopt = bs->drv->strong_runtime_opts;
7818 }
7819
7820 return (curopt && *curopt) ? curopt : NULL;
7821 }
7822
7823 /**
7824 * Copies all strong runtime options from bs->options to the given
7825 * QDict. The set of strong option keys is determined by invoking
7826 * strong_options().
7827 *
7828 * Returns true iff any strong option was present in bs->options (and
7829 * thus copied to the target QDict) with the exception of "filename"
7830 * and "driver". The caller is expected to use this value to decide
7831 * whether the existence of strong options prevents the generation of
7832 * a plain filename.
7833 */
7834 static bool append_strong_runtime_options(QDict *d, BlockDriverState *bs)
7835 {
7836 bool found_any = false;
7837 const char *const *option_name = NULL;
7838
7839 if (!bs->drv) {
7840 return false;
7841 }
7842
7843 while ((option_name = strong_options(bs, option_name))) {
7844 bool option_given = false;
7845
7846 assert(strlen(*option_name) > 0);
7847 if ((*option_name)[strlen(*option_name) - 1] != '.') {
7848 QObject *entry = qdict_get(bs->options, *option_name);
7849 if (!entry) {
7850 continue;
7851 }
7852
7853 qdict_put_obj(d, *option_name, qobject_ref(entry));
7854 option_given = true;
7855 } else {
7856 const QDictEntry *entry;
7857 for (entry = qdict_first(bs->options); entry;
7858 entry = qdict_next(bs->options, entry))
7859 {
7860 if (strstart(qdict_entry_key(entry), *option_name, NULL)) {
7861 qdict_put_obj(d, qdict_entry_key(entry),
7862 qobject_ref(qdict_entry_value(entry)));
7863 option_given = true;
7864 }
7865 }
7866 }
7867
7868 /* While "driver" and "filename" need to be included in a JSON filename,
7869 * their existence does not prohibit generation of a plain filename. */
7870 if (!found_any && option_given &&
7871 strcmp(*option_name, "driver") && strcmp(*option_name, "filename"))
7872 {
7873 found_any = true;
7874 }
7875 }
7876
7877 if (!qdict_haskey(d, "driver")) {
7878 /* Drivers created with bdrv_new_open_driver() may not have a
7879 * @driver option. Add it here. */
7880 qdict_put_str(d, "driver", bs->drv->format_name);
7881 }
7882
7883 return found_any;
7884 }
7885
7886 /* Note: This function may return false positives; it may return true
7887 * even if opening the backing file specified by bs's image header
7888 * would result in exactly bs->backing. */
7889 static bool bdrv_backing_overridden(BlockDriverState *bs)
7890 {
7891 GLOBAL_STATE_CODE();
7892 if (bs->backing) {
7893 return strcmp(bs->auto_backing_file,
7894 bs->backing->bs->filename);
7895 } else {
7896 /* No backing BDS, so if the image header reports any backing
7897 * file, it must have been suppressed */
7898 return bs->auto_backing_file[0] != '\0';
7899 }
7900 }
7901
7902 /* Updates the following BDS fields:
7903 * - exact_filename: A filename which may be used for opening a block device
7904 * which (mostly) equals the given BDS (even without any
7905 * other options; so reading and writing must return the same
7906 * results, but caching etc. may be different)
7907 * - full_open_options: Options which, when given when opening a block device
7908 * (without a filename), result in a BDS (mostly)
7909 * equalling the given one
7910 * - filename: If exact_filename is set, it is copied here. Otherwise,
7911 * full_open_options is converted to a JSON object, prefixed with
7912 * "json:" (for use through the JSON pseudo protocol) and put here.
7913 */
7914 void bdrv_refresh_filename(BlockDriverState *bs)
7915 {
7916 BlockDriver *drv = bs->drv;
7917 BdrvChild *child;
7918 BlockDriverState *primary_child_bs;
7919 QDict *opts;
7920 bool backing_overridden;
7921 bool generate_json_filename; /* Whether our default implementation should
7922 fill exact_filename (false) or not (true) */
7923
7924 GLOBAL_STATE_CODE();
7925
7926 if (!drv) {
7927 return;
7928 }
7929
7930 /* This BDS's file name may depend on any of its children's file names, so
7931 * refresh those first */
7932 QLIST_FOREACH(child, &bs->children, next) {
7933 bdrv_refresh_filename(child->bs);
7934 }
7935
7936 if (bs->implicit) {
7937 /* For implicit nodes, just copy everything from the single child */
7938 child = QLIST_FIRST(&bs->children);
7939 assert(QLIST_NEXT(child, next) == NULL);
7940
7941 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
7942 child->bs->exact_filename);
7943 pstrcpy(bs->filename, sizeof(bs->filename), child->bs->filename);
7944
7945 qobject_unref(bs->full_open_options);
7946 bs->full_open_options = qobject_ref(child->bs->full_open_options);
7947
7948 return;
7949 }
7950
7951 backing_overridden = bdrv_backing_overridden(bs);
7952
7953 if (bs->open_flags & BDRV_O_NO_IO) {
7954 /* Without I/O, the backing file does not change anything.
7955 * Therefore, in such a case (primarily qemu-img), we can
7956 * pretend the backing file has not been overridden even if
7957 * it technically has been. */
7958 backing_overridden = false;
7959 }
7960
7961 /* Gather the options QDict */
7962 opts = qdict_new();
7963 generate_json_filename = append_strong_runtime_options(opts, bs);
7964 generate_json_filename |= backing_overridden;
7965
7966 if (drv->bdrv_gather_child_options) {
7967 /* Some block drivers may not want to present all of their children's
7968 * options, or name them differently from BdrvChild.name */
7969 drv->bdrv_gather_child_options(bs, opts, backing_overridden);
7970 } else {
7971 QLIST_FOREACH(child, &bs->children, next) {
7972 if (child == bs->backing && !backing_overridden) {
7973 /* We can skip the backing BDS if it has not been overridden */
7974 continue;
7975 }
7976
7977 qdict_put(opts, child->name,
7978 qobject_ref(child->bs->full_open_options));
7979 }
7980
7981 if (backing_overridden && !bs->backing) {
7982 /* Force no backing file */
7983 qdict_put_null(opts, "backing");
7984 }
7985 }
7986
7987 qobject_unref(bs->full_open_options);
7988 bs->full_open_options = opts;
7989
7990 primary_child_bs = bdrv_primary_bs(bs);
7991
7992 if (drv->bdrv_refresh_filename) {
7993 /* Obsolete information is of no use here, so drop the old file name
7994 * information before refreshing it */
7995 bs->exact_filename[0] = '\0';
7996
7997 drv->bdrv_refresh_filename(bs);
7998 } else if (primary_child_bs) {
7999 /*
8000 * Try to reconstruct valid information from the underlying
8001 * file -- this only works for format nodes (filter nodes
8002 * cannot be probed and as such must be selected by the user
8003 * either through an options dict, or through a special
8004 * filename which the filter driver must construct in its
8005 * .bdrv_refresh_filename() implementation).
8006 */
8007
8008 bs->exact_filename[0] = '\0';
8009
8010 /*
8011 * We can use the underlying file's filename if:
8012 * - it has a filename,
8013 * - the current BDS is not a filter,
8014 * - the file is a protocol BDS, and
8015 * - opening that file (as this BDS's format) will automatically create
8016 * the BDS tree we have right now, that is:
8017 * - the user did not significantly change this BDS's behavior with
8018 * some explicit (strong) options
8019 * - no non-file child of this BDS has been overridden by the user
8020 * Both of these conditions are represented by generate_json_filename.
8021 */
8022 if (primary_child_bs->exact_filename[0] &&
8023 primary_child_bs->drv->bdrv_file_open &&
8024 !drv->is_filter && !generate_json_filename)
8025 {
8026 strcpy(bs->exact_filename, primary_child_bs->exact_filename);
8027 }
8028 }
8029
8030 if (bs->exact_filename[0]) {
8031 pstrcpy(bs->filename, sizeof(bs->filename), bs->exact_filename);
8032 } else {
8033 GString *json = qobject_to_json(QOBJECT(bs->full_open_options));
8034 if (snprintf(bs->filename, sizeof(bs->filename), "json:%s",
8035 json->str) >= sizeof(bs->filename)) {
8036 /* Give user a hint if we truncated things. */
8037 strcpy(bs->filename + sizeof(bs->filename) - 4, "...");
8038 }
8039 g_string_free(json, true);
8040 }
8041 }
8042
8043 char *bdrv_dirname(BlockDriverState *bs, Error **errp)
8044 {
8045 BlockDriver *drv = bs->drv;
8046 BlockDriverState *child_bs;
8047
8048 GLOBAL_STATE_CODE();
8049
8050 if (!drv) {
8051 error_setg(errp, "Node '%s' is ejected", bs->node_name);
8052 return NULL;
8053 }
8054
8055 if (drv->bdrv_dirname) {
8056 return drv->bdrv_dirname(bs, errp);
8057 }
8058
8059 child_bs = bdrv_primary_bs(bs);
8060 if (child_bs) {
8061 return bdrv_dirname(child_bs, errp);
8062 }
8063
8064 bdrv_refresh_filename(bs);
8065 if (bs->exact_filename[0] != '\0') {
8066 return path_combine(bs->exact_filename, "");
8067 }
8068
8069 error_setg(errp, "Cannot generate a base directory for %s nodes",
8070 drv->format_name);
8071 return NULL;
8072 }
8073
8074 /*
8075 * Hot add/remove a BDS's child. So the user can take a child offline when
8076 * it is broken and take a new child online
8077 */
8078 void bdrv_add_child(BlockDriverState *parent_bs, BlockDriverState *child_bs,
8079 Error **errp)
8080 {
8081 GLOBAL_STATE_CODE();
8082 if (!parent_bs->drv || !parent_bs->drv->bdrv_add_child) {
8083 error_setg(errp, "The node %s does not support adding a child",
8084 bdrv_get_device_or_node_name(parent_bs));
8085 return;
8086 }
8087
8088 /*
8089 * Non-zoned block drivers do not follow zoned storage constraints
8090 * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned
8091 * drivers in a graph.
8092 */
8093 if (!parent_bs->drv->supports_zoned_children &&
8094 child_bs->bl.zoned == BLK_Z_HM) {
8095 /*
8096 * The host-aware model allows zoned storage constraints and random
8097 * write. Allow mixing host-aware and non-zoned drivers. Using
8098 * host-aware device as a regular device.
8099 */
8100 error_setg(errp, "Cannot add a %s child to a %s parent",
8101 child_bs->bl.zoned == BLK_Z_HM ? "zoned" : "non-zoned",
8102 parent_bs->drv->supports_zoned_children ?
8103 "support zoned children" : "not support zoned children");
8104 return;
8105 }
8106
8107 if (!QLIST_EMPTY(&child_bs->parents)) {
8108 error_setg(errp, "The node %s already has a parent",
8109 child_bs->node_name);
8110 return;
8111 }
8112
8113 parent_bs->drv->bdrv_add_child(parent_bs, child_bs, errp);
8114 }
8115
8116 void bdrv_del_child(BlockDriverState *parent_bs, BdrvChild *child, Error **errp)
8117 {
8118 BdrvChild *tmp;
8119
8120 GLOBAL_STATE_CODE();
8121 if (!parent_bs->drv || !parent_bs->drv->bdrv_del_child) {
8122 error_setg(errp, "The node %s does not support removing a child",
8123 bdrv_get_device_or_node_name(parent_bs));
8124 return;
8125 }
8126
8127 QLIST_FOREACH(tmp, &parent_bs->children, next) {
8128 if (tmp == child) {
8129 break;
8130 }
8131 }
8132
8133 if (!tmp) {
8134 error_setg(errp, "The node %s does not have a child named %s",
8135 bdrv_get_device_or_node_name(parent_bs),
8136 bdrv_get_device_or_node_name(child->bs));
8137 return;
8138 }
8139
8140 parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
8141 }
8142
8143 int bdrv_make_empty(BdrvChild *c, Error **errp)
8144 {
8145 BlockDriver *drv = c->bs->drv;
8146 int ret;
8147
8148 GLOBAL_STATE_CODE();
8149 assert(c->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED));
8150
8151 if (!drv->bdrv_make_empty) {
8152 error_setg(errp, "%s does not support emptying nodes",
8153 drv->format_name);
8154 return -ENOTSUP;
8155 }
8156
8157 ret = drv->bdrv_make_empty(c->bs);
8158 if (ret < 0) {
8159 error_setg_errno(errp, -ret, "Failed to empty %s",
8160 c->bs->filename);
8161 return ret;
8162 }
8163
8164 return 0;
8165 }
8166
8167 /*
8168 * Return the child that @bs acts as an overlay for, and from which data may be
8169 * copied in COW or COR operations. Usually this is the backing file.
8170 */
8171 BdrvChild *bdrv_cow_child(BlockDriverState *bs)
8172 {
8173 IO_CODE();
8174
8175 if (!bs || !bs->drv) {
8176 return NULL;
8177 }
8178
8179 if (bs->drv->is_filter) {
8180 return NULL;
8181 }
8182
8183 if (!bs->backing) {
8184 return NULL;
8185 }
8186
8187 assert(bs->backing->role & BDRV_CHILD_COW);
8188 return bs->backing;
8189 }
8190
8191 /*
8192 * If @bs acts as a filter for exactly one of its children, return
8193 * that child.
8194 */
8195 BdrvChild *bdrv_filter_child(BlockDriverState *bs)
8196 {
8197 BdrvChild *c;
8198 IO_CODE();
8199
8200 if (!bs || !bs->drv) {
8201 return NULL;
8202 }
8203
8204 if (!bs->drv->is_filter) {
8205 return NULL;
8206 }
8207
8208 /* Only one of @backing or @file may be used */
8209 assert(!(bs->backing && bs->file));
8210
8211 c = bs->backing ?: bs->file;
8212 if (!c) {
8213 return NULL;
8214 }
8215
8216 assert(c->role & BDRV_CHILD_FILTERED);
8217 return c;
8218 }
8219
8220 /*
8221 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8222 * whichever is non-NULL.
8223 *
8224 * Return NULL if both are NULL.
8225 */
8226 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs)
8227 {
8228 BdrvChild *cow_child = bdrv_cow_child(bs);
8229 BdrvChild *filter_child = bdrv_filter_child(bs);
8230 IO_CODE();
8231
8232 /* Filter nodes cannot have COW backing files */
8233 assert(!(cow_child && filter_child));
8234
8235 return cow_child ?: filter_child;
8236 }
8237
8238 /*
8239 * Return the primary child of this node: For filters, that is the
8240 * filtered child. For other nodes, that is usually the child storing
8241 * metadata.
8242 * (A generally more helpful description is that this is (usually) the
8243 * child that has the same filename as @bs.)
8244 *
8245 * Drivers do not necessarily have a primary child; for example quorum
8246 * does not.
8247 */
8248 BdrvChild *bdrv_primary_child(BlockDriverState *bs)
8249 {
8250 BdrvChild *c, *found = NULL;
8251 IO_CODE();
8252
8253 QLIST_FOREACH(c, &bs->children, next) {
8254 if (c->role & BDRV_CHILD_PRIMARY) {
8255 assert(!found);
8256 found = c;
8257 }
8258 }
8259
8260 return found;
8261 }
8262
8263 static BlockDriverState *bdrv_do_skip_filters(BlockDriverState *bs,
8264 bool stop_on_explicit_filter)
8265 {
8266 BdrvChild *c;
8267
8268 if (!bs) {
8269 return NULL;
8270 }
8271
8272 while (!(stop_on_explicit_filter && !bs->implicit)) {
8273 c = bdrv_filter_child(bs);
8274 if (!c) {
8275 /*
8276 * A filter that is embedded in a working block graph must
8277 * have a child. Assert this here so this function does
8278 * not return a filter node that is not expected by the
8279 * caller.
8280 */
8281 assert(!bs->drv || !bs->drv->is_filter);
8282 break;
8283 }
8284 bs = c->bs;
8285 }
8286 /*
8287 * Note that this treats nodes with bs->drv == NULL as not being
8288 * filters (bs->drv == NULL should be replaced by something else
8289 * anyway).
8290 * The advantage of this behavior is that this function will thus
8291 * always return a non-NULL value (given a non-NULL @bs).
8292 */
8293
8294 return bs;
8295 }
8296
8297 /*
8298 * Return the first BDS that has not been added implicitly or that
8299 * does not have a filtered child down the chain starting from @bs
8300 * (including @bs itself).
8301 */
8302 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs)
8303 {
8304 GLOBAL_STATE_CODE();
8305 return bdrv_do_skip_filters(bs, true);
8306 }
8307
8308 /*
8309 * Return the first BDS that does not have a filtered child down the
8310 * chain starting from @bs (including @bs itself).
8311 */
8312 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs)
8313 {
8314 IO_CODE();
8315 return bdrv_do_skip_filters(bs, false);
8316 }
8317
8318 /*
8319 * For a backing chain, return the first non-filter backing image of
8320 * the first non-filter image.
8321 */
8322 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs)
8323 {
8324 IO_CODE();
8325 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs)));
8326 }
8327
8328 /**
8329 * Check whether [offset, offset + bytes) overlaps with the cached
8330 * block-status data region.
8331 *
8332 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8333 * which is what bdrv_bsc_is_data()'s interface needs.
8334 * Otherwise, *pnum is not touched.
8335 */
8336 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState *bs,
8337 int64_t offset, int64_t bytes,
8338 int64_t *pnum)
8339 {
8340 BdrvBlockStatusCache *bsc = qatomic_rcu_read(&bs->block_status_cache);
8341 bool overlaps;
8342
8343 overlaps =
8344 qatomic_read(&bsc->valid) &&
8345 ranges_overlap(offset, bytes, bsc->data_start,
8346 bsc->data_end - bsc->data_start);
8347
8348 if (overlaps && pnum) {
8349 *pnum = bsc->data_end - offset;
8350 }
8351
8352 return overlaps;
8353 }
8354
8355 /**
8356 * See block_int.h for this function's documentation.
8357 */
8358 bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum)
8359 {
8360 IO_CODE();
8361 RCU_READ_LOCK_GUARD();
8362 return bdrv_bsc_range_overlaps_locked(bs, offset, 1, pnum);
8363 }
8364
8365 /**
8366 * See block_int.h for this function's documentation.
8367 */
8368 void bdrv_bsc_invalidate_range(BlockDriverState *bs,
8369 int64_t offset, int64_t bytes)
8370 {
8371 IO_CODE();
8372 RCU_READ_LOCK_GUARD();
8373
8374 if (bdrv_bsc_range_overlaps_locked(bs, offset, bytes, NULL)) {
8375 qatomic_set(&bs->block_status_cache->valid, false);
8376 }
8377 }
8378
8379 /**
8380 * See block_int.h for this function's documentation.
8381 */
8382 void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes)
8383 {
8384 BdrvBlockStatusCache *new_bsc = g_new(BdrvBlockStatusCache, 1);
8385 BdrvBlockStatusCache *old_bsc;
8386 IO_CODE();
8387
8388 *new_bsc = (BdrvBlockStatusCache) {
8389 .valid = true,
8390 .data_start = offset,
8391 .data_end = offset + bytes,
8392 };
8393
8394 QEMU_LOCK_GUARD(&bs->bsc_modify_lock);
8395
8396 old_bsc = qatomic_rcu_read(&bs->block_status_cache);
8397 qatomic_rcu_set(&bs->block_status_cache, new_bsc);
8398 if (old_bsc) {
8399 g_free_rcu(old_bsc, rcu);
8400 }
8401 }