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