vl.c: fix migration failure for 3.1 and older machine types
[qemu.git] / block.c
1 /*
2 * QEMU System Emulator block driver
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "module_block.h"
33 #include "qemu/main-loop.h"
34 #include "qemu/module.h"
35 #include "qapi/error.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qjson.h"
38 #include "qapi/qmp/qnull.h"
39 #include "qapi/qmp/qstring.h"
40 #include "qapi/qobject-output-visitor.h"
41 #include "qapi/qapi-visit-block-core.h"
42 #include "sysemu/block-backend.h"
43 #include "sysemu/sysemu.h"
44 #include "qemu/notify.h"
45 #include "qemu/option.h"
46 #include "qemu/coroutine.h"
47 #include "block/qapi.h"
48 #include "qemu/timer.h"
49 #include "qemu/cutils.h"
50 #include "qemu/id.h"
51
52 #ifdef CONFIG_BSD
53 #include <sys/ioctl.h>
54 #include <sys/queue.h>
55 #ifndef __DragonFly__
56 #include <sys/disk.h>
57 #endif
58 #endif
59
60 #ifdef _WIN32
61 #include <windows.h>
62 #endif
63
64 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
65
66 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
67 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
68
69 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
70 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
71
72 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
73 QLIST_HEAD_INITIALIZER(bdrv_drivers);
74
75 static BlockDriverState *bdrv_open_inherit(const char *filename,
76 const char *reference,
77 QDict *options, int flags,
78 BlockDriverState *parent,
79 const BdrvChildRole *child_role,
80 Error **errp);
81
82 /* If non-zero, use only whitelisted block drivers */
83 static int use_bdrv_whitelist;
84
85 #ifdef _WIN32
86 static int is_windows_drive_prefix(const char *filename)
87 {
88 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
89 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
90 filename[1] == ':');
91 }
92
93 int is_windows_drive(const char *filename)
94 {
95 if (is_windows_drive_prefix(filename) &&
96 filename[2] == '\0')
97 return 1;
98 if (strstart(filename, "\\\\.\\", NULL) ||
99 strstart(filename, "//./", NULL))
100 return 1;
101 return 0;
102 }
103 #endif
104
105 size_t bdrv_opt_mem_align(BlockDriverState *bs)
106 {
107 if (!bs || !bs->drv) {
108 /* page size or 4k (hdd sector size) should be on the safe side */
109 return MAX(4096, qemu_real_host_page_size);
110 }
111
112 return bs->bl.opt_mem_alignment;
113 }
114
115 size_t bdrv_min_mem_align(BlockDriverState *bs)
116 {
117 if (!bs || !bs->drv) {
118 /* page size or 4k (hdd sector size) should be on the safe side */
119 return MAX(4096, qemu_real_host_page_size);
120 }
121
122 return bs->bl.min_mem_alignment;
123 }
124
125 /* check if the path starts with "<protocol>:" */
126 int path_has_protocol(const char *path)
127 {
128 const char *p;
129
130 #ifdef _WIN32
131 if (is_windows_drive(path) ||
132 is_windows_drive_prefix(path)) {
133 return 0;
134 }
135 p = path + strcspn(path, ":/\\");
136 #else
137 p = path + strcspn(path, ":/");
138 #endif
139
140 return *p == ':';
141 }
142
143 int path_is_absolute(const char *path)
144 {
145 #ifdef _WIN32
146 /* specific case for names like: "\\.\d:" */
147 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
148 return 1;
149 }
150 return (*path == '/' || *path == '\\');
151 #else
152 return (*path == '/');
153 #endif
154 }
155
156 /* if filename is absolute, just return its duplicate. Otherwise, build a
157 path to it by considering it is relative to base_path. URL are
158 supported. */
159 char *path_combine(const char *base_path, const char *filename)
160 {
161 const char *protocol_stripped = NULL;
162 const char *p, *p1;
163 char *result;
164 int len;
165
166 if (path_is_absolute(filename)) {
167 return g_strdup(filename);
168 }
169
170 if (path_has_protocol(base_path)) {
171 protocol_stripped = strchr(base_path, ':');
172 if (protocol_stripped) {
173 protocol_stripped++;
174 }
175 }
176 p = protocol_stripped ?: base_path;
177
178 p1 = strrchr(base_path, '/');
179 #ifdef _WIN32
180 {
181 const char *p2;
182 p2 = strrchr(base_path, '\\');
183 if (!p1 || p2 > p1) {
184 p1 = p2;
185 }
186 }
187 #endif
188 if (p1) {
189 p1++;
190 } else {
191 p1 = base_path;
192 }
193 if (p1 > p) {
194 p = p1;
195 }
196 len = p - base_path;
197
198 result = g_malloc(len + strlen(filename) + 1);
199 memcpy(result, base_path, len);
200 strcpy(result + len, filename);
201
202 return result;
203 }
204
205 /*
206 * Helper function for bdrv_parse_filename() implementations to remove optional
207 * protocol prefixes (especially "file:") from a filename and for putting the
208 * stripped filename into the options QDict if there is such a prefix.
209 */
210 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
211 QDict *options)
212 {
213 if (strstart(filename, prefix, &filename)) {
214 /* Stripping the explicit protocol prefix may result in a protocol
215 * prefix being (wrongly) detected (if the filename contains a colon) */
216 if (path_has_protocol(filename)) {
217 QString *fat_filename;
218
219 /* This means there is some colon before the first slash; therefore,
220 * this cannot be an absolute path */
221 assert(!path_is_absolute(filename));
222
223 /* And we can thus fix the protocol detection issue by prefixing it
224 * by "./" */
225 fat_filename = qstring_from_str("./");
226 qstring_append(fat_filename, filename);
227
228 assert(!path_has_protocol(qstring_get_str(fat_filename)));
229
230 qdict_put(options, "filename", fat_filename);
231 } else {
232 /* If no protocol prefix was detected, we can use the shortened
233 * filename as-is */
234 qdict_put_str(options, "filename", filename);
235 }
236 }
237 }
238
239
240 /* Returns whether the image file is opened as read-only. Note that this can
241 * return false and writing to the image file is still not possible because the
242 * image is inactivated. */
243 bool bdrv_is_read_only(BlockDriverState *bs)
244 {
245 return bs->read_only;
246 }
247
248 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
249 bool ignore_allow_rdw, Error **errp)
250 {
251 /* Do not set read_only if copy_on_read is enabled */
252 if (bs->copy_on_read && read_only) {
253 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
254 bdrv_get_device_or_node_name(bs));
255 return -EINVAL;
256 }
257
258 /* Do not clear read_only if it is prohibited */
259 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
260 !ignore_allow_rdw)
261 {
262 error_setg(errp, "Node '%s' is read only",
263 bdrv_get_device_or_node_name(bs));
264 return -EPERM;
265 }
266
267 return 0;
268 }
269
270 /*
271 * Called by a driver that can only provide a read-only image.
272 *
273 * Returns 0 if the node is already read-only or it could switch the node to
274 * read-only because BDRV_O_AUTO_RDONLY is set.
275 *
276 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
277 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
278 * is not NULL, it is used as the error message for the Error object.
279 */
280 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
281 Error **errp)
282 {
283 int ret = 0;
284
285 if (!(bs->open_flags & BDRV_O_RDWR)) {
286 return 0;
287 }
288 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
289 goto fail;
290 }
291
292 ret = bdrv_can_set_read_only(bs, true, false, NULL);
293 if (ret < 0) {
294 goto fail;
295 }
296
297 bs->read_only = true;
298 bs->open_flags &= ~BDRV_O_RDWR;
299
300 return 0;
301
302 fail:
303 error_setg(errp, "%s", errmsg ?: "Image is read-only");
304 return -EACCES;
305 }
306
307 /*
308 * If @backing is empty, this function returns NULL without setting
309 * @errp. In all other cases, NULL will only be returned with @errp
310 * set.
311 *
312 * Therefore, a return value of NULL without @errp set means that
313 * there is no backing file; if @errp is set, there is one but its
314 * absolute filename cannot be generated.
315 */
316 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
317 const char *backing,
318 Error **errp)
319 {
320 if (backing[0] == '\0') {
321 return NULL;
322 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
323 return g_strdup(backing);
324 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
325 error_setg(errp, "Cannot use relative backing file names for '%s'",
326 backed);
327 return NULL;
328 } else {
329 return path_combine(backed, backing);
330 }
331 }
332
333 /*
334 * If @filename is empty or NULL, this function returns NULL without
335 * setting @errp. In all other cases, NULL will only be returned with
336 * @errp set.
337 */
338 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
339 const char *filename, Error **errp)
340 {
341 char *dir, *full_name;
342
343 if (!filename || filename[0] == '\0') {
344 return NULL;
345 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
346 return g_strdup(filename);
347 }
348
349 dir = bdrv_dirname(relative_to, errp);
350 if (!dir) {
351 return NULL;
352 }
353
354 full_name = g_strconcat(dir, filename, NULL);
355 g_free(dir);
356 return full_name;
357 }
358
359 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
360 {
361 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
362 }
363
364 void bdrv_register(BlockDriver *bdrv)
365 {
366 assert(bdrv->format_name);
367 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
368 }
369
370 BlockDriverState *bdrv_new(void)
371 {
372 BlockDriverState *bs;
373 int i;
374
375 bs = g_new0(BlockDriverState, 1);
376 QLIST_INIT(&bs->dirty_bitmaps);
377 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
378 QLIST_INIT(&bs->op_blockers[i]);
379 }
380 notifier_with_return_list_init(&bs->before_write_notifiers);
381 qemu_co_mutex_init(&bs->reqs_lock);
382 qemu_mutex_init(&bs->dirty_bitmap_mutex);
383 bs->refcnt = 1;
384 bs->aio_context = qemu_get_aio_context();
385
386 qemu_co_queue_init(&bs->flush_queue);
387
388 for (i = 0; i < bdrv_drain_all_count; i++) {
389 bdrv_drained_begin(bs);
390 }
391
392 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
393
394 return bs;
395 }
396
397 static BlockDriver *bdrv_do_find_format(const char *format_name)
398 {
399 BlockDriver *drv1;
400
401 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
402 if (!strcmp(drv1->format_name, format_name)) {
403 return drv1;
404 }
405 }
406
407 return NULL;
408 }
409
410 BlockDriver *bdrv_find_format(const char *format_name)
411 {
412 BlockDriver *drv1;
413 int i;
414
415 drv1 = bdrv_do_find_format(format_name);
416 if (drv1) {
417 return drv1;
418 }
419
420 /* The driver isn't registered, maybe we need to load a module */
421 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
422 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
423 block_module_load_one(block_driver_modules[i].library_name);
424 break;
425 }
426 }
427
428 return bdrv_do_find_format(format_name);
429 }
430
431 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
432 {
433 static const char *whitelist_rw[] = {
434 CONFIG_BDRV_RW_WHITELIST
435 };
436 static const char *whitelist_ro[] = {
437 CONFIG_BDRV_RO_WHITELIST
438 };
439 const char **p;
440
441 if (!whitelist_rw[0] && !whitelist_ro[0]) {
442 return 1; /* no whitelist, anything goes */
443 }
444
445 for (p = whitelist_rw; *p; p++) {
446 if (!strcmp(format_name, *p)) {
447 return 1;
448 }
449 }
450 if (read_only) {
451 for (p = whitelist_ro; *p; p++) {
452 if (!strcmp(format_name, *p)) {
453 return 1;
454 }
455 }
456 }
457 return 0;
458 }
459
460 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
461 {
462 return bdrv_format_is_whitelisted(drv->format_name, read_only);
463 }
464
465 bool bdrv_uses_whitelist(void)
466 {
467 return use_bdrv_whitelist;
468 }
469
470 typedef struct CreateCo {
471 BlockDriver *drv;
472 char *filename;
473 QemuOpts *opts;
474 int ret;
475 Error *err;
476 } CreateCo;
477
478 static void coroutine_fn bdrv_create_co_entry(void *opaque)
479 {
480 Error *local_err = NULL;
481 int ret;
482
483 CreateCo *cco = opaque;
484 assert(cco->drv);
485
486 ret = cco->drv->bdrv_co_create_opts(cco->filename, cco->opts, &local_err);
487 error_propagate(&cco->err, local_err);
488 cco->ret = ret;
489 }
490
491 int bdrv_create(BlockDriver *drv, const char* filename,
492 QemuOpts *opts, Error **errp)
493 {
494 int ret;
495
496 Coroutine *co;
497 CreateCo cco = {
498 .drv = drv,
499 .filename = g_strdup(filename),
500 .opts = opts,
501 .ret = NOT_DONE,
502 .err = NULL,
503 };
504
505 if (!drv->bdrv_co_create_opts) {
506 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
507 ret = -ENOTSUP;
508 goto out;
509 }
510
511 if (qemu_in_coroutine()) {
512 /* Fast-path if already in coroutine context */
513 bdrv_create_co_entry(&cco);
514 } else {
515 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
516 qemu_coroutine_enter(co);
517 while (cco.ret == NOT_DONE) {
518 aio_poll(qemu_get_aio_context(), true);
519 }
520 }
521
522 ret = cco.ret;
523 if (ret < 0) {
524 if (cco.err) {
525 error_propagate(errp, cco.err);
526 } else {
527 error_setg_errno(errp, -ret, "Could not create image");
528 }
529 }
530
531 out:
532 g_free(cco.filename);
533 return ret;
534 }
535
536 /**
537 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
538 * least the given @minimum_size.
539 *
540 * On success, return @blk's actual length.
541 * Otherwise, return -errno.
542 */
543 static int64_t create_file_fallback_truncate(BlockBackend *blk,
544 int64_t minimum_size, Error **errp)
545 {
546 Error *local_err = NULL;
547 int64_t size;
548 int ret;
549
550 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, &local_err);
551 if (ret < 0 && ret != -ENOTSUP) {
552 error_propagate(errp, local_err);
553 return ret;
554 }
555
556 size = blk_getlength(blk);
557 if (size < 0) {
558 error_free(local_err);
559 error_setg_errno(errp, -size,
560 "Failed to inquire the new image file's length");
561 return size;
562 }
563
564 if (size < minimum_size) {
565 /* Need to grow the image, but we failed to do that */
566 error_propagate(errp, local_err);
567 return -ENOTSUP;
568 }
569
570 error_free(local_err);
571 local_err = NULL;
572
573 return size;
574 }
575
576 /**
577 * Helper function for bdrv_create_file_fallback(): Zero the first
578 * sector to remove any potentially pre-existing image header.
579 */
580 static int create_file_fallback_zero_first_sector(BlockBackend *blk,
581 int64_t current_size,
582 Error **errp)
583 {
584 int64_t bytes_to_clear;
585 int ret;
586
587 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
588 if (bytes_to_clear) {
589 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
590 if (ret < 0) {
591 error_setg_errno(errp, -ret,
592 "Failed to clear the new image's first sector");
593 return ret;
594 }
595 }
596
597 return 0;
598 }
599
600 static int bdrv_create_file_fallback(const char *filename, BlockDriver *drv,
601 QemuOpts *opts, Error **errp)
602 {
603 BlockBackend *blk;
604 QDict *options;
605 int64_t size = 0;
606 char *buf = NULL;
607 PreallocMode prealloc;
608 Error *local_err = NULL;
609 int ret;
610
611 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
612 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
613 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
614 PREALLOC_MODE_OFF, &local_err);
615 g_free(buf);
616 if (local_err) {
617 error_propagate(errp, local_err);
618 return -EINVAL;
619 }
620
621 if (prealloc != PREALLOC_MODE_OFF) {
622 error_setg(errp, "Unsupported preallocation mode '%s'",
623 PreallocMode_str(prealloc));
624 return -ENOTSUP;
625 }
626
627 options = qdict_new();
628 qdict_put_str(options, "driver", drv->format_name);
629
630 blk = blk_new_open(filename, NULL, options,
631 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
632 if (!blk) {
633 error_prepend(errp, "Protocol driver '%s' does not support image "
634 "creation, and opening the image failed: ",
635 drv->format_name);
636 return -EINVAL;
637 }
638
639 size = create_file_fallback_truncate(blk, size, errp);
640 if (size < 0) {
641 ret = size;
642 goto out;
643 }
644
645 ret = create_file_fallback_zero_first_sector(blk, size, errp);
646 if (ret < 0) {
647 goto out;
648 }
649
650 ret = 0;
651 out:
652 blk_unref(blk);
653 return ret;
654 }
655
656 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
657 {
658 BlockDriver *drv;
659
660 drv = bdrv_find_protocol(filename, true, errp);
661 if (drv == NULL) {
662 return -ENOENT;
663 }
664
665 if (drv->bdrv_co_create_opts) {
666 return bdrv_create(drv, filename, opts, errp);
667 } else {
668 return bdrv_create_file_fallback(filename, drv, opts, errp);
669 }
670 }
671
672 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
673 {
674 Error *local_err = NULL;
675 int ret;
676
677 assert(bs != NULL);
678
679 if (!bs->drv) {
680 error_setg(errp, "Block node '%s' is not opened", bs->filename);
681 return -ENOMEDIUM;
682 }
683
684 if (!bs->drv->bdrv_co_delete_file) {
685 error_setg(errp, "Driver '%s' does not support image deletion",
686 bs->drv->format_name);
687 return -ENOTSUP;
688 }
689
690 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
691 if (ret < 0) {
692 error_propagate(errp, local_err);
693 }
694
695 return ret;
696 }
697
698 /**
699 * Try to get @bs's logical and physical block size.
700 * On success, store them in @bsz struct and return 0.
701 * On failure return -errno.
702 * @bs must not be empty.
703 */
704 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
705 {
706 BlockDriver *drv = bs->drv;
707
708 if (drv && drv->bdrv_probe_blocksizes) {
709 return drv->bdrv_probe_blocksizes(bs, bsz);
710 } else if (drv && drv->is_filter && bs->file) {
711 return bdrv_probe_blocksizes(bs->file->bs, bsz);
712 }
713
714 return -ENOTSUP;
715 }
716
717 /**
718 * Try to get @bs's geometry (cyls, heads, sectors).
719 * On success, store them in @geo struct and return 0.
720 * On failure return -errno.
721 * @bs must not be empty.
722 */
723 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
724 {
725 BlockDriver *drv = bs->drv;
726
727 if (drv && drv->bdrv_probe_geometry) {
728 return drv->bdrv_probe_geometry(bs, geo);
729 } else if (drv && drv->is_filter && bs->file) {
730 return bdrv_probe_geometry(bs->file->bs, geo);
731 }
732
733 return -ENOTSUP;
734 }
735
736 /*
737 * Create a uniquely-named empty temporary file.
738 * Return 0 upon success, otherwise a negative errno value.
739 */
740 int get_tmp_filename(char *filename, int size)
741 {
742 #ifdef _WIN32
743 char temp_dir[MAX_PATH];
744 /* GetTempFileName requires that its output buffer (4th param)
745 have length MAX_PATH or greater. */
746 assert(size >= MAX_PATH);
747 return (GetTempPath(MAX_PATH, temp_dir)
748 && GetTempFileName(temp_dir, "qem", 0, filename)
749 ? 0 : -GetLastError());
750 #else
751 int fd;
752 const char *tmpdir;
753 tmpdir = getenv("TMPDIR");
754 if (!tmpdir) {
755 tmpdir = "/var/tmp";
756 }
757 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
758 return -EOVERFLOW;
759 }
760 fd = mkstemp(filename);
761 if (fd < 0) {
762 return -errno;
763 }
764 if (close(fd) != 0) {
765 unlink(filename);
766 return -errno;
767 }
768 return 0;
769 #endif
770 }
771
772 /*
773 * Detect host devices. By convention, /dev/cdrom[N] is always
774 * recognized as a host CDROM.
775 */
776 static BlockDriver *find_hdev_driver(const char *filename)
777 {
778 int score_max = 0, score;
779 BlockDriver *drv = NULL, *d;
780
781 QLIST_FOREACH(d, &bdrv_drivers, list) {
782 if (d->bdrv_probe_device) {
783 score = d->bdrv_probe_device(filename);
784 if (score > score_max) {
785 score_max = score;
786 drv = d;
787 }
788 }
789 }
790
791 return drv;
792 }
793
794 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
795 {
796 BlockDriver *drv1;
797
798 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
799 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
800 return drv1;
801 }
802 }
803
804 return NULL;
805 }
806
807 BlockDriver *bdrv_find_protocol(const char *filename,
808 bool allow_protocol_prefix,
809 Error **errp)
810 {
811 BlockDriver *drv1;
812 char protocol[128];
813 int len;
814 const char *p;
815 int i;
816
817 /* TODO Drivers without bdrv_file_open must be specified explicitly */
818
819 /*
820 * XXX(hch): we really should not let host device detection
821 * override an explicit protocol specification, but moving this
822 * later breaks access to device names with colons in them.
823 * Thanks to the brain-dead persistent naming schemes on udev-
824 * based Linux systems those actually are quite common.
825 */
826 drv1 = find_hdev_driver(filename);
827 if (drv1) {
828 return drv1;
829 }
830
831 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
832 return &bdrv_file;
833 }
834
835 p = strchr(filename, ':');
836 assert(p != NULL);
837 len = p - filename;
838 if (len > sizeof(protocol) - 1)
839 len = sizeof(protocol) - 1;
840 memcpy(protocol, filename, len);
841 protocol[len] = '\0';
842
843 drv1 = bdrv_do_find_protocol(protocol);
844 if (drv1) {
845 return drv1;
846 }
847
848 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
849 if (block_driver_modules[i].protocol_name &&
850 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
851 block_module_load_one(block_driver_modules[i].library_name);
852 break;
853 }
854 }
855
856 drv1 = bdrv_do_find_protocol(protocol);
857 if (!drv1) {
858 error_setg(errp, "Unknown protocol '%s'", protocol);
859 }
860 return drv1;
861 }
862
863 /*
864 * Guess image format by probing its contents.
865 * This is not a good idea when your image is raw (CVE-2008-2004), but
866 * we do it anyway for backward compatibility.
867 *
868 * @buf contains the image's first @buf_size bytes.
869 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
870 * but can be smaller if the image file is smaller)
871 * @filename is its filename.
872 *
873 * For all block drivers, call the bdrv_probe() method to get its
874 * probing score.
875 * Return the first block driver with the highest probing score.
876 */
877 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
878 const char *filename)
879 {
880 int score_max = 0, score;
881 BlockDriver *drv = NULL, *d;
882
883 QLIST_FOREACH(d, &bdrv_drivers, list) {
884 if (d->bdrv_probe) {
885 score = d->bdrv_probe(buf, buf_size, filename);
886 if (score > score_max) {
887 score_max = score;
888 drv = d;
889 }
890 }
891 }
892
893 return drv;
894 }
895
896 static int find_image_format(BlockBackend *file, const char *filename,
897 BlockDriver **pdrv, Error **errp)
898 {
899 BlockDriver *drv;
900 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
901 int ret = 0;
902
903 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
904 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
905 *pdrv = &bdrv_raw;
906 return ret;
907 }
908
909 ret = blk_pread(file, 0, buf, sizeof(buf));
910 if (ret < 0) {
911 error_setg_errno(errp, -ret, "Could not read image for determining its "
912 "format");
913 *pdrv = NULL;
914 return ret;
915 }
916
917 drv = bdrv_probe_all(buf, ret, filename);
918 if (!drv) {
919 error_setg(errp, "Could not determine image format: No compatible "
920 "driver found");
921 ret = -ENOENT;
922 }
923 *pdrv = drv;
924 return ret;
925 }
926
927 /**
928 * Set the current 'total_sectors' value
929 * Return 0 on success, -errno on error.
930 */
931 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
932 {
933 BlockDriver *drv = bs->drv;
934
935 if (!drv) {
936 return -ENOMEDIUM;
937 }
938
939 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
940 if (bdrv_is_sg(bs))
941 return 0;
942
943 /* query actual device if possible, otherwise just trust the hint */
944 if (drv->bdrv_getlength) {
945 int64_t length = drv->bdrv_getlength(bs);
946 if (length < 0) {
947 return length;
948 }
949 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
950 }
951
952 bs->total_sectors = hint;
953 return 0;
954 }
955
956 /**
957 * Combines a QDict of new block driver @options with any missing options taken
958 * from @old_options, so that leaving out an option defaults to its old value.
959 */
960 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
961 QDict *old_options)
962 {
963 if (bs->drv && bs->drv->bdrv_join_options) {
964 bs->drv->bdrv_join_options(options, old_options);
965 } else {
966 qdict_join(options, old_options, false);
967 }
968 }
969
970 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
971 int open_flags,
972 Error **errp)
973 {
974 Error *local_err = NULL;
975 char *value = qemu_opt_get_del(opts, "detect-zeroes");
976 BlockdevDetectZeroesOptions detect_zeroes =
977 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
978 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
979 g_free(value);
980 if (local_err) {
981 error_propagate(errp, local_err);
982 return detect_zeroes;
983 }
984
985 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
986 !(open_flags & BDRV_O_UNMAP))
987 {
988 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
989 "without setting discard operation to unmap");
990 }
991
992 return detect_zeroes;
993 }
994
995 /**
996 * Set open flags for aio engine
997 *
998 * Return 0 on success, -1 if the engine specified is invalid
999 */
1000 int bdrv_parse_aio(const char *mode, int *flags)
1001 {
1002 if (!strcmp(mode, "threads")) {
1003 /* do nothing, default */
1004 } else if (!strcmp(mode, "native")) {
1005 *flags |= BDRV_O_NATIVE_AIO;
1006 #ifdef CONFIG_LINUX_IO_URING
1007 } else if (!strcmp(mode, "io_uring")) {
1008 *flags |= BDRV_O_IO_URING;
1009 #endif
1010 } else {
1011 return -1;
1012 }
1013
1014 return 0;
1015 }
1016
1017 /**
1018 * Set open flags for a given discard mode
1019 *
1020 * Return 0 on success, -1 if the discard mode was invalid.
1021 */
1022 int bdrv_parse_discard_flags(const char *mode, int *flags)
1023 {
1024 *flags &= ~BDRV_O_UNMAP;
1025
1026 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1027 /* do nothing */
1028 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1029 *flags |= BDRV_O_UNMAP;
1030 } else {
1031 return -1;
1032 }
1033
1034 return 0;
1035 }
1036
1037 /**
1038 * Set open flags for a given cache mode
1039 *
1040 * Return 0 on success, -1 if the cache mode was invalid.
1041 */
1042 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1043 {
1044 *flags &= ~BDRV_O_CACHE_MASK;
1045
1046 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1047 *writethrough = false;
1048 *flags |= BDRV_O_NOCACHE;
1049 } else if (!strcmp(mode, "directsync")) {
1050 *writethrough = true;
1051 *flags |= BDRV_O_NOCACHE;
1052 } else if (!strcmp(mode, "writeback")) {
1053 *writethrough = false;
1054 } else if (!strcmp(mode, "unsafe")) {
1055 *writethrough = false;
1056 *flags |= BDRV_O_NO_FLUSH;
1057 } else if (!strcmp(mode, "writethrough")) {
1058 *writethrough = true;
1059 } else {
1060 return -1;
1061 }
1062
1063 return 0;
1064 }
1065
1066 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1067 {
1068 BlockDriverState *parent = c->opaque;
1069 return g_strdup(bdrv_get_device_or_node_name(parent));
1070 }
1071
1072 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1073 {
1074 BlockDriverState *bs = child->opaque;
1075 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1076 }
1077
1078 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1079 {
1080 BlockDriverState *bs = child->opaque;
1081 return bdrv_drain_poll(bs, false, NULL, false);
1082 }
1083
1084 static void bdrv_child_cb_drained_end(BdrvChild *child,
1085 int *drained_end_counter)
1086 {
1087 BlockDriverState *bs = child->opaque;
1088 bdrv_drained_end_no_poll(bs, drained_end_counter);
1089 }
1090
1091 static void bdrv_child_cb_attach(BdrvChild *child)
1092 {
1093 BlockDriverState *bs = child->opaque;
1094 bdrv_apply_subtree_drain(child, bs);
1095 }
1096
1097 static void bdrv_child_cb_detach(BdrvChild *child)
1098 {
1099 BlockDriverState *bs = child->opaque;
1100 bdrv_unapply_subtree_drain(child, bs);
1101 }
1102
1103 static int bdrv_child_cb_inactivate(BdrvChild *child)
1104 {
1105 BlockDriverState *bs = child->opaque;
1106 assert(bs->open_flags & BDRV_O_INACTIVE);
1107 return 0;
1108 }
1109
1110 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1111 GSList **ignore, Error **errp)
1112 {
1113 BlockDriverState *bs = child->opaque;
1114 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1115 }
1116
1117 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1118 GSList **ignore)
1119 {
1120 BlockDriverState *bs = child->opaque;
1121 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1122 }
1123
1124 /*
1125 * Returns the options and flags that a temporary snapshot should get, based on
1126 * the originally requested flags (the originally requested image will have
1127 * flags like a backing file)
1128 */
1129 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1130 int parent_flags, QDict *parent_options)
1131 {
1132 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1133
1134 /* For temporary files, unconditional cache=unsafe is fine */
1135 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1136 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1137
1138 /* Copy the read-only and discard options from the parent */
1139 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1140 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1141
1142 /* aio=native doesn't work for cache.direct=off, so disable it for the
1143 * temporary snapshot */
1144 *child_flags &= ~BDRV_O_NATIVE_AIO;
1145 }
1146
1147 /*
1148 * Returns the options and flags that bs->file should get if a protocol driver
1149 * is expected, based on the given options and flags for the parent BDS
1150 */
1151 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
1152 int parent_flags, QDict *parent_options)
1153 {
1154 int flags = parent_flags;
1155
1156 /* Enable protocol handling, disable format probing for bs->file */
1157 flags |= BDRV_O_PROTOCOL;
1158
1159 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
1160 * the parent. */
1161 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1162 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1163 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1164
1165 /* Inherit the read-only option from the parent if it's not set */
1166 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1167 qdict_copy_default(child_options, parent_options, BDRV_OPT_AUTO_READ_ONLY);
1168
1169 /* Our block drivers take care to send flushes and respect unmap policy,
1170 * so we can default to enable both on lower layers regardless of the
1171 * corresponding parent options. */
1172 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1173
1174 /* Clear flags that only apply to the top layer */
1175 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
1176 BDRV_O_NO_IO);
1177
1178 *child_flags = flags;
1179 }
1180
1181 const BdrvChildRole child_file = {
1182 .parent_is_bds = true,
1183 .get_parent_desc = bdrv_child_get_parent_desc,
1184 .inherit_options = bdrv_inherited_options,
1185 .drained_begin = bdrv_child_cb_drained_begin,
1186 .drained_poll = bdrv_child_cb_drained_poll,
1187 .drained_end = bdrv_child_cb_drained_end,
1188 .attach = bdrv_child_cb_attach,
1189 .detach = bdrv_child_cb_detach,
1190 .inactivate = bdrv_child_cb_inactivate,
1191 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1192 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1193 };
1194
1195 /*
1196 * Returns the options and flags that bs->file should get if the use of formats
1197 * (and not only protocols) is permitted for it, based on the given options and
1198 * flags for the parent BDS
1199 */
1200 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
1201 int parent_flags, QDict *parent_options)
1202 {
1203 child_file.inherit_options(child_flags, child_options,
1204 parent_flags, parent_options);
1205
1206 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
1207 }
1208
1209 const BdrvChildRole child_format = {
1210 .parent_is_bds = true,
1211 .get_parent_desc = bdrv_child_get_parent_desc,
1212 .inherit_options = bdrv_inherited_fmt_options,
1213 .drained_begin = bdrv_child_cb_drained_begin,
1214 .drained_poll = bdrv_child_cb_drained_poll,
1215 .drained_end = bdrv_child_cb_drained_end,
1216 .attach = bdrv_child_cb_attach,
1217 .detach = bdrv_child_cb_detach,
1218 .inactivate = bdrv_child_cb_inactivate,
1219 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1220 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1221 };
1222
1223 static void bdrv_backing_attach(BdrvChild *c)
1224 {
1225 BlockDriverState *parent = c->opaque;
1226 BlockDriverState *backing_hd = c->bs;
1227
1228 assert(!parent->backing_blocker);
1229 error_setg(&parent->backing_blocker,
1230 "node is used as backing hd of '%s'",
1231 bdrv_get_device_or_node_name(parent));
1232
1233 bdrv_refresh_filename(backing_hd);
1234
1235 parent->open_flags &= ~BDRV_O_NO_BACKING;
1236 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
1237 backing_hd->filename);
1238 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
1239 backing_hd->drv ? backing_hd->drv->format_name : "");
1240
1241 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1242 /* Otherwise we won't be able to commit or stream */
1243 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1244 parent->backing_blocker);
1245 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1246 parent->backing_blocker);
1247 /*
1248 * We do backup in 3 ways:
1249 * 1. drive backup
1250 * The target bs is new opened, and the source is top BDS
1251 * 2. blockdev backup
1252 * Both the source and the target are top BDSes.
1253 * 3. internal backup(used for block replication)
1254 * Both the source and the target are backing file
1255 *
1256 * In case 1 and 2, neither the source nor the target is the backing file.
1257 * In case 3, we will block the top BDS, so there is only one block job
1258 * for the top BDS and its backing chain.
1259 */
1260 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1261 parent->backing_blocker);
1262 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1263 parent->backing_blocker);
1264
1265 bdrv_child_cb_attach(c);
1266 }
1267
1268 static void bdrv_backing_detach(BdrvChild *c)
1269 {
1270 BlockDriverState *parent = c->opaque;
1271
1272 assert(parent->backing_blocker);
1273 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1274 error_free(parent->backing_blocker);
1275 parent->backing_blocker = NULL;
1276
1277 bdrv_child_cb_detach(c);
1278 }
1279
1280 /*
1281 * Returns the options and flags that bs->backing should get, based on the
1282 * given options and flags for the parent BDS
1283 */
1284 static void bdrv_backing_options(int *child_flags, QDict *child_options,
1285 int parent_flags, QDict *parent_options)
1286 {
1287 int flags = parent_flags;
1288
1289 /* The cache mode is inherited unmodified for backing files; except WCE,
1290 * which is only applied on the top level (BlockBackend) */
1291 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1292 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1293 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1294
1295 /* backing files always opened read-only */
1296 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1297 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1298 flags &= ~BDRV_O_COPY_ON_READ;
1299
1300 /* snapshot=on is handled on the top layer */
1301 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
1302
1303 *child_flags = flags;
1304 }
1305
1306 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1307 const char *filename, Error **errp)
1308 {
1309 BlockDriverState *parent = c->opaque;
1310 bool read_only = bdrv_is_read_only(parent);
1311 int ret;
1312
1313 if (read_only) {
1314 ret = bdrv_reopen_set_read_only(parent, false, errp);
1315 if (ret < 0) {
1316 return ret;
1317 }
1318 }
1319
1320 ret = bdrv_change_backing_file(parent, filename,
1321 base->drv ? base->drv->format_name : "");
1322 if (ret < 0) {
1323 error_setg_errno(errp, -ret, "Could not update backing file link");
1324 }
1325
1326 if (read_only) {
1327 bdrv_reopen_set_read_only(parent, true, NULL);
1328 }
1329
1330 return ret;
1331 }
1332
1333 const BdrvChildRole child_backing = {
1334 .parent_is_bds = true,
1335 .get_parent_desc = bdrv_child_get_parent_desc,
1336 .attach = bdrv_backing_attach,
1337 .detach = bdrv_backing_detach,
1338 .inherit_options = bdrv_backing_options,
1339 .drained_begin = bdrv_child_cb_drained_begin,
1340 .drained_poll = bdrv_child_cb_drained_poll,
1341 .drained_end = bdrv_child_cb_drained_end,
1342 .inactivate = bdrv_child_cb_inactivate,
1343 .update_filename = bdrv_backing_update_filename,
1344 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1345 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1346 };
1347
1348 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1349 {
1350 int open_flags = flags;
1351
1352 /*
1353 * Clear flags that are internal to the block layer before opening the
1354 * image.
1355 */
1356 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1357
1358 return open_flags;
1359 }
1360
1361 static void update_flags_from_options(int *flags, QemuOpts *opts)
1362 {
1363 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1364
1365 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1366 *flags |= BDRV_O_NO_FLUSH;
1367 }
1368
1369 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1370 *flags |= BDRV_O_NOCACHE;
1371 }
1372
1373 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1374 *flags |= BDRV_O_RDWR;
1375 }
1376
1377 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1378 *flags |= BDRV_O_AUTO_RDONLY;
1379 }
1380 }
1381
1382 static void update_options_from_flags(QDict *options, int flags)
1383 {
1384 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1385 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1386 }
1387 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1388 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1389 flags & BDRV_O_NO_FLUSH);
1390 }
1391 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1392 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1393 }
1394 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1395 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1396 flags & BDRV_O_AUTO_RDONLY);
1397 }
1398 }
1399
1400 static void bdrv_assign_node_name(BlockDriverState *bs,
1401 const char *node_name,
1402 Error **errp)
1403 {
1404 char *gen_node_name = NULL;
1405
1406 if (!node_name) {
1407 node_name = gen_node_name = id_generate(ID_BLOCK);
1408 } else if (!id_wellformed(node_name)) {
1409 /*
1410 * Check for empty string or invalid characters, but not if it is
1411 * generated (generated names use characters not available to the user)
1412 */
1413 error_setg(errp, "Invalid node name");
1414 return;
1415 }
1416
1417 /* takes care of avoiding namespaces collisions */
1418 if (blk_by_name(node_name)) {
1419 error_setg(errp, "node-name=%s is conflicting with a device id",
1420 node_name);
1421 goto out;
1422 }
1423
1424 /* takes care of avoiding duplicates node names */
1425 if (bdrv_find_node(node_name)) {
1426 error_setg(errp, "Duplicate node name");
1427 goto out;
1428 }
1429
1430 /* Make sure that the node name isn't truncated */
1431 if (strlen(node_name) >= sizeof(bs->node_name)) {
1432 error_setg(errp, "Node name too long");
1433 goto out;
1434 }
1435
1436 /* copy node name into the bs and insert it into the graph list */
1437 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1438 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1439 out:
1440 g_free(gen_node_name);
1441 }
1442
1443 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1444 const char *node_name, QDict *options,
1445 int open_flags, Error **errp)
1446 {
1447 Error *local_err = NULL;
1448 int i, ret;
1449
1450 bdrv_assign_node_name(bs, node_name, &local_err);
1451 if (local_err) {
1452 error_propagate(errp, local_err);
1453 return -EINVAL;
1454 }
1455
1456 bs->drv = drv;
1457 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1458 bs->opaque = g_malloc0(drv->instance_size);
1459
1460 if (drv->bdrv_file_open) {
1461 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1462 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1463 } else if (drv->bdrv_open) {
1464 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1465 } else {
1466 ret = 0;
1467 }
1468
1469 if (ret < 0) {
1470 if (local_err) {
1471 error_propagate(errp, local_err);
1472 } else if (bs->filename[0]) {
1473 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1474 } else {
1475 error_setg_errno(errp, -ret, "Could not open image");
1476 }
1477 goto open_failed;
1478 }
1479
1480 ret = refresh_total_sectors(bs, bs->total_sectors);
1481 if (ret < 0) {
1482 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1483 return ret;
1484 }
1485
1486 bdrv_refresh_limits(bs, &local_err);
1487 if (local_err) {
1488 error_propagate(errp, local_err);
1489 return -EINVAL;
1490 }
1491
1492 assert(bdrv_opt_mem_align(bs) != 0);
1493 assert(bdrv_min_mem_align(bs) != 0);
1494 assert(is_power_of_2(bs->bl.request_alignment));
1495
1496 for (i = 0; i < bs->quiesce_counter; i++) {
1497 if (drv->bdrv_co_drain_begin) {
1498 drv->bdrv_co_drain_begin(bs);
1499 }
1500 }
1501
1502 return 0;
1503 open_failed:
1504 bs->drv = NULL;
1505 if (bs->file != NULL) {
1506 bdrv_unref_child(bs, bs->file);
1507 bs->file = NULL;
1508 }
1509 g_free(bs->opaque);
1510 bs->opaque = NULL;
1511 return ret;
1512 }
1513
1514 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1515 int flags, Error **errp)
1516 {
1517 BlockDriverState *bs;
1518 int ret;
1519
1520 bs = bdrv_new();
1521 bs->open_flags = flags;
1522 bs->explicit_options = qdict_new();
1523 bs->options = qdict_new();
1524 bs->opaque = NULL;
1525
1526 update_options_from_flags(bs->options, flags);
1527
1528 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1529 if (ret < 0) {
1530 qobject_unref(bs->explicit_options);
1531 bs->explicit_options = NULL;
1532 qobject_unref(bs->options);
1533 bs->options = NULL;
1534 bdrv_unref(bs);
1535 return NULL;
1536 }
1537
1538 return bs;
1539 }
1540
1541 QemuOptsList bdrv_runtime_opts = {
1542 .name = "bdrv_common",
1543 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1544 .desc = {
1545 {
1546 .name = "node-name",
1547 .type = QEMU_OPT_STRING,
1548 .help = "Node name of the block device node",
1549 },
1550 {
1551 .name = "driver",
1552 .type = QEMU_OPT_STRING,
1553 .help = "Block driver to use for the node",
1554 },
1555 {
1556 .name = BDRV_OPT_CACHE_DIRECT,
1557 .type = QEMU_OPT_BOOL,
1558 .help = "Bypass software writeback cache on the host",
1559 },
1560 {
1561 .name = BDRV_OPT_CACHE_NO_FLUSH,
1562 .type = QEMU_OPT_BOOL,
1563 .help = "Ignore flush requests",
1564 },
1565 {
1566 .name = BDRV_OPT_READ_ONLY,
1567 .type = QEMU_OPT_BOOL,
1568 .help = "Node is opened in read-only mode",
1569 },
1570 {
1571 .name = BDRV_OPT_AUTO_READ_ONLY,
1572 .type = QEMU_OPT_BOOL,
1573 .help = "Node can become read-only if opening read-write fails",
1574 },
1575 {
1576 .name = "detect-zeroes",
1577 .type = QEMU_OPT_STRING,
1578 .help = "try to optimize zero writes (off, on, unmap)",
1579 },
1580 {
1581 .name = BDRV_OPT_DISCARD,
1582 .type = QEMU_OPT_STRING,
1583 .help = "discard operation (ignore/off, unmap/on)",
1584 },
1585 {
1586 .name = BDRV_OPT_FORCE_SHARE,
1587 .type = QEMU_OPT_BOOL,
1588 .help = "always accept other writers (default: off)",
1589 },
1590 { /* end of list */ }
1591 },
1592 };
1593
1594 static QemuOptsList fallback_create_opts = {
1595 .name = "fallback-create-opts",
1596 .head = QTAILQ_HEAD_INITIALIZER(fallback_create_opts.head),
1597 .desc = {
1598 {
1599 .name = BLOCK_OPT_SIZE,
1600 .type = QEMU_OPT_SIZE,
1601 .help = "Virtual disk size"
1602 },
1603 {
1604 .name = BLOCK_OPT_PREALLOC,
1605 .type = QEMU_OPT_STRING,
1606 .help = "Preallocation mode (allowed values: off)"
1607 },
1608 { /* end of list */ }
1609 }
1610 };
1611
1612 /*
1613 * Common part for opening disk images and files
1614 *
1615 * Removes all processed options from *options.
1616 */
1617 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1618 QDict *options, Error **errp)
1619 {
1620 int ret, open_flags;
1621 const char *filename;
1622 const char *driver_name = NULL;
1623 const char *node_name = NULL;
1624 const char *discard;
1625 QemuOpts *opts;
1626 BlockDriver *drv;
1627 Error *local_err = NULL;
1628
1629 assert(bs->file == NULL);
1630 assert(options != NULL && bs->options != options);
1631
1632 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1633 qemu_opts_absorb_qdict(opts, options, &local_err);
1634 if (local_err) {
1635 error_propagate(errp, local_err);
1636 ret = -EINVAL;
1637 goto fail_opts;
1638 }
1639
1640 update_flags_from_options(&bs->open_flags, opts);
1641
1642 driver_name = qemu_opt_get(opts, "driver");
1643 drv = bdrv_find_format(driver_name);
1644 assert(drv != NULL);
1645
1646 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1647
1648 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1649 error_setg(errp,
1650 BDRV_OPT_FORCE_SHARE
1651 "=on can only be used with read-only images");
1652 ret = -EINVAL;
1653 goto fail_opts;
1654 }
1655
1656 if (file != NULL) {
1657 bdrv_refresh_filename(blk_bs(file));
1658 filename = blk_bs(file)->filename;
1659 } else {
1660 /*
1661 * Caution: while qdict_get_try_str() is fine, getting
1662 * non-string types would require more care. When @options
1663 * come from -blockdev or blockdev_add, its members are typed
1664 * according to the QAPI schema, but when they come from
1665 * -drive, they're all QString.
1666 */
1667 filename = qdict_get_try_str(options, "filename");
1668 }
1669
1670 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1671 error_setg(errp, "The '%s' block driver requires a file name",
1672 drv->format_name);
1673 ret = -EINVAL;
1674 goto fail_opts;
1675 }
1676
1677 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1678 drv->format_name);
1679
1680 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1681
1682 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1683 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1684 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1685 } else {
1686 ret = -ENOTSUP;
1687 }
1688 if (ret < 0) {
1689 error_setg(errp,
1690 !bs->read_only && bdrv_is_whitelisted(drv, true)
1691 ? "Driver '%s' can only be used for read-only devices"
1692 : "Driver '%s' is not whitelisted",
1693 drv->format_name);
1694 goto fail_opts;
1695 }
1696 }
1697
1698 /* bdrv_new() and bdrv_close() make it so */
1699 assert(atomic_read(&bs->copy_on_read) == 0);
1700
1701 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1702 if (!bs->read_only) {
1703 bdrv_enable_copy_on_read(bs);
1704 } else {
1705 error_setg(errp, "Can't use copy-on-read on read-only device");
1706 ret = -EINVAL;
1707 goto fail_opts;
1708 }
1709 }
1710
1711 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1712 if (discard != NULL) {
1713 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1714 error_setg(errp, "Invalid discard option");
1715 ret = -EINVAL;
1716 goto fail_opts;
1717 }
1718 }
1719
1720 bs->detect_zeroes =
1721 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1722 if (local_err) {
1723 error_propagate(errp, local_err);
1724 ret = -EINVAL;
1725 goto fail_opts;
1726 }
1727
1728 if (filename != NULL) {
1729 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1730 } else {
1731 bs->filename[0] = '\0';
1732 }
1733 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1734
1735 /* Open the image, either directly or using a protocol */
1736 open_flags = bdrv_open_flags(bs, bs->open_flags);
1737 node_name = qemu_opt_get(opts, "node-name");
1738
1739 assert(!drv->bdrv_file_open || file == NULL);
1740 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1741 if (ret < 0) {
1742 goto fail_opts;
1743 }
1744
1745 qemu_opts_del(opts);
1746 return 0;
1747
1748 fail_opts:
1749 qemu_opts_del(opts);
1750 return ret;
1751 }
1752
1753 static QDict *parse_json_filename(const char *filename, Error **errp)
1754 {
1755 QObject *options_obj;
1756 QDict *options;
1757 int ret;
1758
1759 ret = strstart(filename, "json:", &filename);
1760 assert(ret);
1761
1762 options_obj = qobject_from_json(filename, errp);
1763 if (!options_obj) {
1764 error_prepend(errp, "Could not parse the JSON options: ");
1765 return NULL;
1766 }
1767
1768 options = qobject_to(QDict, options_obj);
1769 if (!options) {
1770 qobject_unref(options_obj);
1771 error_setg(errp, "Invalid JSON object given");
1772 return NULL;
1773 }
1774
1775 qdict_flatten(options);
1776
1777 return options;
1778 }
1779
1780 static void parse_json_protocol(QDict *options, const char **pfilename,
1781 Error **errp)
1782 {
1783 QDict *json_options;
1784 Error *local_err = NULL;
1785
1786 /* Parse json: pseudo-protocol */
1787 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1788 return;
1789 }
1790
1791 json_options = parse_json_filename(*pfilename, &local_err);
1792 if (local_err) {
1793 error_propagate(errp, local_err);
1794 return;
1795 }
1796
1797 /* Options given in the filename have lower priority than options
1798 * specified directly */
1799 qdict_join(options, json_options, false);
1800 qobject_unref(json_options);
1801 *pfilename = NULL;
1802 }
1803
1804 /*
1805 * Fills in default options for opening images and converts the legacy
1806 * filename/flags pair to option QDict entries.
1807 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1808 * block driver has been specified explicitly.
1809 */
1810 static int bdrv_fill_options(QDict **options, const char *filename,
1811 int *flags, Error **errp)
1812 {
1813 const char *drvname;
1814 bool protocol = *flags & BDRV_O_PROTOCOL;
1815 bool parse_filename = false;
1816 BlockDriver *drv = NULL;
1817 Error *local_err = NULL;
1818
1819 /*
1820 * Caution: while qdict_get_try_str() is fine, getting non-string
1821 * types would require more care. When @options come from
1822 * -blockdev or blockdev_add, its members are typed according to
1823 * the QAPI schema, but when they come from -drive, they're all
1824 * QString.
1825 */
1826 drvname = qdict_get_try_str(*options, "driver");
1827 if (drvname) {
1828 drv = bdrv_find_format(drvname);
1829 if (!drv) {
1830 error_setg(errp, "Unknown driver '%s'", drvname);
1831 return -ENOENT;
1832 }
1833 /* If the user has explicitly specified the driver, this choice should
1834 * override the BDRV_O_PROTOCOL flag */
1835 protocol = drv->bdrv_file_open;
1836 }
1837
1838 if (protocol) {
1839 *flags |= BDRV_O_PROTOCOL;
1840 } else {
1841 *flags &= ~BDRV_O_PROTOCOL;
1842 }
1843
1844 /* Translate cache options from flags into options */
1845 update_options_from_flags(*options, *flags);
1846
1847 /* Fetch the file name from the options QDict if necessary */
1848 if (protocol && filename) {
1849 if (!qdict_haskey(*options, "filename")) {
1850 qdict_put_str(*options, "filename", filename);
1851 parse_filename = true;
1852 } else {
1853 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1854 "the same time");
1855 return -EINVAL;
1856 }
1857 }
1858
1859 /* Find the right block driver */
1860 /* See cautionary note on accessing @options above */
1861 filename = qdict_get_try_str(*options, "filename");
1862
1863 if (!drvname && protocol) {
1864 if (filename) {
1865 drv = bdrv_find_protocol(filename, parse_filename, errp);
1866 if (!drv) {
1867 return -EINVAL;
1868 }
1869
1870 drvname = drv->format_name;
1871 qdict_put_str(*options, "driver", drvname);
1872 } else {
1873 error_setg(errp, "Must specify either driver or file");
1874 return -EINVAL;
1875 }
1876 }
1877
1878 assert(drv || !protocol);
1879
1880 /* Driver-specific filename parsing */
1881 if (drv && drv->bdrv_parse_filename && parse_filename) {
1882 drv->bdrv_parse_filename(filename, *options, &local_err);
1883 if (local_err) {
1884 error_propagate(errp, local_err);
1885 return -EINVAL;
1886 }
1887
1888 if (!drv->bdrv_needs_filename) {
1889 qdict_del(*options, "filename");
1890 }
1891 }
1892
1893 return 0;
1894 }
1895
1896 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1897 uint64_t perm, uint64_t shared,
1898 GSList *ignore_children,
1899 bool *tighten_restrictions, Error **errp);
1900 static void bdrv_child_abort_perm_update(BdrvChild *c);
1901 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1902
1903 typedef struct BlockReopenQueueEntry {
1904 bool prepared;
1905 bool perms_checked;
1906 BDRVReopenState state;
1907 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
1908 } BlockReopenQueueEntry;
1909
1910 /*
1911 * Return the flags that @bs will have after the reopens in @q have
1912 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1913 * return the current flags.
1914 */
1915 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1916 {
1917 BlockReopenQueueEntry *entry;
1918
1919 if (q != NULL) {
1920 QTAILQ_FOREACH(entry, q, entry) {
1921 if (entry->state.bs == bs) {
1922 return entry->state.flags;
1923 }
1924 }
1925 }
1926
1927 return bs->open_flags;
1928 }
1929
1930 /* Returns whether the image file can be written to after the reopen queue @q
1931 * has been successfully applied, or right now if @q is NULL. */
1932 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1933 BlockReopenQueue *q)
1934 {
1935 int flags = bdrv_reopen_get_flags(q, bs);
1936
1937 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1938 }
1939
1940 /*
1941 * Return whether the BDS can be written to. This is not necessarily
1942 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1943 * be written to but do not count as read-only images.
1944 */
1945 bool bdrv_is_writable(BlockDriverState *bs)
1946 {
1947 return bdrv_is_writable_after_reopen(bs, NULL);
1948 }
1949
1950 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1951 BdrvChild *c, const BdrvChildRole *role,
1952 BlockReopenQueue *reopen_queue,
1953 uint64_t parent_perm, uint64_t parent_shared,
1954 uint64_t *nperm, uint64_t *nshared)
1955 {
1956 assert(bs->drv && bs->drv->bdrv_child_perm);
1957 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1958 parent_perm, parent_shared,
1959 nperm, nshared);
1960 /* TODO Take force_share from reopen_queue */
1961 if (child_bs && child_bs->force_share) {
1962 *nshared = BLK_PERM_ALL;
1963 }
1964 }
1965
1966 /*
1967 * Check whether permissions on this node can be changed in a way that
1968 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1969 * permissions of all its parents. This involves checking whether all necessary
1970 * permission changes to child nodes can be performed.
1971 *
1972 * Will set *tighten_restrictions to true if and only if new permissions have to
1973 * be taken or currently shared permissions are to be unshared. Otherwise,
1974 * errors are not fatal as long as the caller accepts that the restrictions
1975 * remain tighter than they need to be. The caller still has to abort the
1976 * transaction.
1977 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1978 * encounter fatal errors even though no restrictions are to be tightened. For
1979 * example, changing a node from RW to RO will fail if the WRITE permission is
1980 * to be kept.
1981 *
1982 * A call to this function must always be followed by a call to bdrv_set_perm()
1983 * or bdrv_abort_perm_update().
1984 */
1985 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1986 uint64_t cumulative_perms,
1987 uint64_t cumulative_shared_perms,
1988 GSList *ignore_children,
1989 bool *tighten_restrictions, Error **errp)
1990 {
1991 BlockDriver *drv = bs->drv;
1992 BdrvChild *c;
1993 int ret;
1994
1995 assert(!q || !tighten_restrictions);
1996
1997 if (tighten_restrictions) {
1998 uint64_t current_perms, current_shared;
1999 uint64_t added_perms, removed_shared_perms;
2000
2001 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
2002
2003 added_perms = cumulative_perms & ~current_perms;
2004 removed_shared_perms = current_shared & ~cumulative_shared_perms;
2005
2006 *tighten_restrictions = added_perms || removed_shared_perms;
2007 }
2008
2009 /* Write permissions never work with read-only images */
2010 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2011 !bdrv_is_writable_after_reopen(bs, q))
2012 {
2013 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2014 error_setg(errp, "Block node is read-only");
2015 } else {
2016 uint64_t current_perms, current_shared;
2017 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
2018 if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
2019 error_setg(errp, "Cannot make block node read-only, there is "
2020 "a writer on it");
2021 } else {
2022 error_setg(errp, "Cannot make block node read-only and create "
2023 "a writer on it");
2024 }
2025 }
2026
2027 return -EPERM;
2028 }
2029
2030 /* Check this node */
2031 if (!drv) {
2032 return 0;
2033 }
2034
2035 if (drv->bdrv_check_perm) {
2036 return drv->bdrv_check_perm(bs, cumulative_perms,
2037 cumulative_shared_perms, errp);
2038 }
2039
2040 /* Drivers that never have children can omit .bdrv_child_perm() */
2041 if (!drv->bdrv_child_perm) {
2042 assert(QLIST_EMPTY(&bs->children));
2043 return 0;
2044 }
2045
2046 /* Check all children */
2047 QLIST_FOREACH(c, &bs->children, next) {
2048 uint64_t cur_perm, cur_shared;
2049 bool child_tighten_restr;
2050
2051 bdrv_child_perm(bs, c->bs, c, c->role, q,
2052 cumulative_perms, cumulative_shared_perms,
2053 &cur_perm, &cur_shared);
2054 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children,
2055 tighten_restrictions ? &child_tighten_restr
2056 : NULL,
2057 errp);
2058 if (tighten_restrictions) {
2059 *tighten_restrictions |= child_tighten_restr;
2060 }
2061 if (ret < 0) {
2062 return ret;
2063 }
2064 }
2065
2066 return 0;
2067 }
2068
2069 /*
2070 * Notifies drivers that after a previous bdrv_check_perm() call, the
2071 * permission update is not performed and any preparations made for it (e.g.
2072 * taken file locks) need to be undone.
2073 *
2074 * This function recursively notifies all child nodes.
2075 */
2076 static void bdrv_abort_perm_update(BlockDriverState *bs)
2077 {
2078 BlockDriver *drv = bs->drv;
2079 BdrvChild *c;
2080
2081 if (!drv) {
2082 return;
2083 }
2084
2085 if (drv->bdrv_abort_perm_update) {
2086 drv->bdrv_abort_perm_update(bs);
2087 }
2088
2089 QLIST_FOREACH(c, &bs->children, next) {
2090 bdrv_child_abort_perm_update(c);
2091 }
2092 }
2093
2094 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
2095 uint64_t cumulative_shared_perms)
2096 {
2097 BlockDriver *drv = bs->drv;
2098 BdrvChild *c;
2099
2100 if (!drv) {
2101 return;
2102 }
2103
2104 /* Update this node */
2105 if (drv->bdrv_set_perm) {
2106 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2107 }
2108
2109 /* Drivers that never have children can omit .bdrv_child_perm() */
2110 if (!drv->bdrv_child_perm) {
2111 assert(QLIST_EMPTY(&bs->children));
2112 return;
2113 }
2114
2115 /* Update all children */
2116 QLIST_FOREACH(c, &bs->children, next) {
2117 uint64_t cur_perm, cur_shared;
2118 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2119 cumulative_perms, cumulative_shared_perms,
2120 &cur_perm, &cur_shared);
2121 bdrv_child_set_perm(c, cur_perm, cur_shared);
2122 }
2123 }
2124
2125 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2126 uint64_t *shared_perm)
2127 {
2128 BdrvChild *c;
2129 uint64_t cumulative_perms = 0;
2130 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2131
2132 QLIST_FOREACH(c, &bs->parents, next_parent) {
2133 cumulative_perms |= c->perm;
2134 cumulative_shared_perms &= c->shared_perm;
2135 }
2136
2137 *perm = cumulative_perms;
2138 *shared_perm = cumulative_shared_perms;
2139 }
2140
2141 static char *bdrv_child_user_desc(BdrvChild *c)
2142 {
2143 if (c->role->get_parent_desc) {
2144 return c->role->get_parent_desc(c);
2145 }
2146
2147 return g_strdup("another user");
2148 }
2149
2150 char *bdrv_perm_names(uint64_t perm)
2151 {
2152 struct perm_name {
2153 uint64_t perm;
2154 const char *name;
2155 } permissions[] = {
2156 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2157 { BLK_PERM_WRITE, "write" },
2158 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2159 { BLK_PERM_RESIZE, "resize" },
2160 { BLK_PERM_GRAPH_MOD, "change children" },
2161 { 0, NULL }
2162 };
2163
2164 GString *result = g_string_sized_new(30);
2165 struct perm_name *p;
2166
2167 for (p = permissions; p->name; p++) {
2168 if (perm & p->perm) {
2169 if (result->len > 0) {
2170 g_string_append(result, ", ");
2171 }
2172 g_string_append(result, p->name);
2173 }
2174 }
2175
2176 return g_string_free(result, FALSE);
2177 }
2178
2179 /*
2180 * Checks whether a new reference to @bs can be added if the new user requires
2181 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2182 * set, the BdrvChild objects in this list are ignored in the calculations;
2183 * this allows checking permission updates for an existing reference.
2184 *
2185 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2186 *
2187 * Needs to be followed by a call to either bdrv_set_perm() or
2188 * bdrv_abort_perm_update(). */
2189 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
2190 uint64_t new_used_perm,
2191 uint64_t new_shared_perm,
2192 GSList *ignore_children,
2193 bool *tighten_restrictions,
2194 Error **errp)
2195 {
2196 BdrvChild *c;
2197 uint64_t cumulative_perms = new_used_perm;
2198 uint64_t cumulative_shared_perms = new_shared_perm;
2199
2200 assert(!q || !tighten_restrictions);
2201
2202 /* There is no reason why anyone couldn't tolerate write_unchanged */
2203 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
2204
2205 QLIST_FOREACH(c, &bs->parents, next_parent) {
2206 if (g_slist_find(ignore_children, c)) {
2207 continue;
2208 }
2209
2210 if ((new_used_perm & c->shared_perm) != new_used_perm) {
2211 char *user = bdrv_child_user_desc(c);
2212 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
2213
2214 if (tighten_restrictions) {
2215 *tighten_restrictions = true;
2216 }
2217
2218 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
2219 "allow '%s' on %s",
2220 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2221 g_free(user);
2222 g_free(perm_names);
2223 return -EPERM;
2224 }
2225
2226 if ((c->perm & new_shared_perm) != c->perm) {
2227 char *user = bdrv_child_user_desc(c);
2228 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
2229
2230 if (tighten_restrictions) {
2231 *tighten_restrictions = true;
2232 }
2233
2234 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
2235 "'%s' on %s",
2236 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2237 g_free(user);
2238 g_free(perm_names);
2239 return -EPERM;
2240 }
2241
2242 cumulative_perms |= c->perm;
2243 cumulative_shared_perms &= c->shared_perm;
2244 }
2245
2246 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
2247 ignore_children, tighten_restrictions, errp);
2248 }
2249
2250 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2251 * bdrv_child_abort_perm_update(). */
2252 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
2253 uint64_t perm, uint64_t shared,
2254 GSList *ignore_children,
2255 bool *tighten_restrictions, Error **errp)
2256 {
2257 int ret;
2258
2259 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
2260 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children,
2261 tighten_restrictions, errp);
2262 g_slist_free(ignore_children);
2263
2264 if (ret < 0) {
2265 return ret;
2266 }
2267
2268 if (!c->has_backup_perm) {
2269 c->has_backup_perm = true;
2270 c->backup_perm = c->perm;
2271 c->backup_shared_perm = c->shared_perm;
2272 }
2273 /*
2274 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2275 * same child twice during check_perm procedure
2276 */
2277
2278 c->perm = perm;
2279 c->shared_perm = shared;
2280
2281 return 0;
2282 }
2283
2284 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
2285 {
2286 uint64_t cumulative_perms, cumulative_shared_perms;
2287
2288 c->has_backup_perm = false;
2289
2290 c->perm = perm;
2291 c->shared_perm = shared;
2292
2293 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
2294 &cumulative_shared_perms);
2295 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
2296 }
2297
2298 static void bdrv_child_abort_perm_update(BdrvChild *c)
2299 {
2300 if (c->has_backup_perm) {
2301 c->perm = c->backup_perm;
2302 c->shared_perm = c->backup_shared_perm;
2303 c->has_backup_perm = false;
2304 }
2305
2306 bdrv_abort_perm_update(c->bs);
2307 }
2308
2309 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2310 Error **errp)
2311 {
2312 Error *local_err = NULL;
2313 int ret;
2314 bool tighten_restrictions;
2315
2316 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL,
2317 &tighten_restrictions, &local_err);
2318 if (ret < 0) {
2319 bdrv_child_abort_perm_update(c);
2320 if (tighten_restrictions) {
2321 error_propagate(errp, local_err);
2322 } else {
2323 /*
2324 * Our caller may intend to only loosen restrictions and
2325 * does not expect this function to fail. Errors are not
2326 * fatal in such a case, so we can just hide them from our
2327 * caller.
2328 */
2329 error_free(local_err);
2330 ret = 0;
2331 }
2332 return ret;
2333 }
2334
2335 bdrv_child_set_perm(c, perm, shared);
2336
2337 return 0;
2338 }
2339
2340 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2341 {
2342 uint64_t parent_perms, parent_shared;
2343 uint64_t perms, shared;
2344
2345 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2346 bdrv_child_perm(bs, c->bs, c, c->role, NULL, parent_perms, parent_shared,
2347 &perms, &shared);
2348
2349 return bdrv_child_try_set_perm(c, perms, shared, errp);
2350 }
2351
2352 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2353 const BdrvChildRole *role,
2354 BlockReopenQueue *reopen_queue,
2355 uint64_t perm, uint64_t shared,
2356 uint64_t *nperm, uint64_t *nshared)
2357 {
2358 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2359 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2360 }
2361
2362 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
2363 const BdrvChildRole *role,
2364 BlockReopenQueue *reopen_queue,
2365 uint64_t perm, uint64_t shared,
2366 uint64_t *nperm, uint64_t *nshared)
2367 {
2368 bool backing = (role == &child_backing);
2369 assert(role == &child_backing || role == &child_file);
2370
2371 if (!backing) {
2372 int flags = bdrv_reopen_get_flags(reopen_queue, bs);
2373
2374 /* Apart from the modifications below, the same permissions are
2375 * forwarded and left alone as for filters */
2376 bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
2377 &perm, &shared);
2378
2379 /* Format drivers may touch metadata even if the guest doesn't write */
2380 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2381 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2382 }
2383
2384 /* bs->file always needs to be consistent because of the metadata. We
2385 * can never allow other users to resize or write to it. */
2386 if (!(flags & BDRV_O_NO_IO)) {
2387 perm |= BLK_PERM_CONSISTENT_READ;
2388 }
2389 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2390 } else {
2391 /* We want consistent read from backing files if the parent needs it.
2392 * No other operations are performed on backing files. */
2393 perm &= BLK_PERM_CONSISTENT_READ;
2394
2395 /* If the parent can deal with changing data, we're okay with a
2396 * writable and resizable backing file. */
2397 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2398 if (shared & BLK_PERM_WRITE) {
2399 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2400 } else {
2401 shared = 0;
2402 }
2403
2404 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2405 BLK_PERM_WRITE_UNCHANGED;
2406 }
2407
2408 if (bs->open_flags & BDRV_O_INACTIVE) {
2409 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2410 }
2411
2412 *nperm = perm;
2413 *nshared = shared;
2414 }
2415
2416 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2417 {
2418 static const uint64_t permissions[] = {
2419 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2420 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2421 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2422 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2423 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2424 };
2425
2426 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2427 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2428
2429 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2430
2431 return permissions[qapi_perm];
2432 }
2433
2434 static void bdrv_replace_child_noperm(BdrvChild *child,
2435 BlockDriverState *new_bs)
2436 {
2437 BlockDriverState *old_bs = child->bs;
2438 int new_bs_quiesce_counter;
2439 int drain_saldo;
2440
2441 assert(!child->frozen);
2442
2443 if (old_bs && new_bs) {
2444 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2445 }
2446
2447 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2448 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2449
2450 /*
2451 * If the new child node is drained but the old one was not, flush
2452 * all outstanding requests to the old child node.
2453 */
2454 while (drain_saldo > 0 && child->role->drained_begin) {
2455 bdrv_parent_drained_begin_single(child, true);
2456 drain_saldo--;
2457 }
2458
2459 if (old_bs) {
2460 /* Detach first so that the recursive drain sections coming from @child
2461 * are already gone and we only end the drain sections that came from
2462 * elsewhere. */
2463 if (child->role->detach) {
2464 child->role->detach(child);
2465 }
2466 QLIST_REMOVE(child, next_parent);
2467 }
2468
2469 child->bs = new_bs;
2470
2471 if (new_bs) {
2472 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2473
2474 /*
2475 * Detaching the old node may have led to the new node's
2476 * quiesce_counter having been decreased. Not a problem, we
2477 * just need to recognize this here and then invoke
2478 * drained_end appropriately more often.
2479 */
2480 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2481 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2482
2483 /* Attach only after starting new drained sections, so that recursive
2484 * drain sections coming from @child don't get an extra .drained_begin
2485 * callback. */
2486 if (child->role->attach) {
2487 child->role->attach(child);
2488 }
2489 }
2490
2491 /*
2492 * If the old child node was drained but the new one is not, allow
2493 * requests to come in only after the new node has been attached.
2494 */
2495 while (drain_saldo < 0 && child->role->drained_end) {
2496 bdrv_parent_drained_end_single(child);
2497 drain_saldo++;
2498 }
2499 }
2500
2501 /*
2502 * Updates @child to change its reference to point to @new_bs, including
2503 * checking and applying the necessary permisson updates both to the old node
2504 * and to @new_bs.
2505 *
2506 * NULL is passed as @new_bs for removing the reference before freeing @child.
2507 *
2508 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2509 * function uses bdrv_set_perm() to update the permissions according to the new
2510 * reference that @new_bs gets.
2511 */
2512 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2513 {
2514 BlockDriverState *old_bs = child->bs;
2515 uint64_t perm, shared_perm;
2516
2517 bdrv_replace_child_noperm(child, new_bs);
2518
2519 /*
2520 * Start with the new node's permissions. If @new_bs is a (direct
2521 * or indirect) child of @old_bs, we must complete the permission
2522 * update on @new_bs before we loosen the restrictions on @old_bs.
2523 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2524 * updating the permissions of @new_bs, and thus not purely loosen
2525 * restrictions.
2526 */
2527 if (new_bs) {
2528 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2529 bdrv_set_perm(new_bs, perm, shared_perm);
2530 }
2531
2532 if (old_bs) {
2533 /* Update permissions for old node. This is guaranteed to succeed
2534 * because we're just taking a parent away, so we're loosening
2535 * restrictions. */
2536 bool tighten_restrictions;
2537 int ret;
2538
2539 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2540 ret = bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL,
2541 &tighten_restrictions, NULL);
2542 assert(tighten_restrictions == false);
2543 if (ret < 0) {
2544 /* We only tried to loosen restrictions, so errors are not fatal */
2545 bdrv_abort_perm_update(old_bs);
2546 } else {
2547 bdrv_set_perm(old_bs, perm, shared_perm);
2548 }
2549
2550 /* When the parent requiring a non-default AioContext is removed, the
2551 * node moves back to the main AioContext */
2552 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2553 }
2554 }
2555
2556 /*
2557 * This function steals the reference to child_bs from the caller.
2558 * That reference is later dropped by bdrv_root_unref_child().
2559 *
2560 * On failure NULL is returned, errp is set and the reference to
2561 * child_bs is also dropped.
2562 *
2563 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2564 * (unless @child_bs is already in @ctx).
2565 */
2566 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2567 const char *child_name,
2568 const BdrvChildRole *child_role,
2569 AioContext *ctx,
2570 uint64_t perm, uint64_t shared_perm,
2571 void *opaque, Error **errp)
2572 {
2573 BdrvChild *child;
2574 Error *local_err = NULL;
2575 int ret;
2576
2577 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, NULL,
2578 errp);
2579 if (ret < 0) {
2580 bdrv_abort_perm_update(child_bs);
2581 bdrv_unref(child_bs);
2582 return NULL;
2583 }
2584
2585 child = g_new(BdrvChild, 1);
2586 *child = (BdrvChild) {
2587 .bs = NULL,
2588 .name = g_strdup(child_name),
2589 .role = child_role,
2590 .perm = perm,
2591 .shared_perm = shared_perm,
2592 .opaque = opaque,
2593 };
2594
2595 /* If the AioContexts don't match, first try to move the subtree of
2596 * child_bs into the AioContext of the new parent. If this doesn't work,
2597 * try moving the parent into the AioContext of child_bs instead. */
2598 if (bdrv_get_aio_context(child_bs) != ctx) {
2599 ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err);
2600 if (ret < 0 && child_role->can_set_aio_ctx) {
2601 GSList *ignore = g_slist_prepend(NULL, child);
2602 ctx = bdrv_get_aio_context(child_bs);
2603 if (child_role->can_set_aio_ctx(child, ctx, &ignore, NULL)) {
2604 error_free(local_err);
2605 ret = 0;
2606 g_slist_free(ignore);
2607 ignore = g_slist_prepend(NULL, child);
2608 child_role->set_aio_ctx(child, ctx, &ignore);
2609 }
2610 g_slist_free(ignore);
2611 }
2612 if (ret < 0) {
2613 error_propagate(errp, local_err);
2614 g_free(child);
2615 bdrv_abort_perm_update(child_bs);
2616 return NULL;
2617 }
2618 }
2619
2620 /* This performs the matching bdrv_set_perm() for the above check. */
2621 bdrv_replace_child(child, child_bs);
2622
2623 return child;
2624 }
2625
2626 /*
2627 * This function transfers the reference to child_bs from the caller
2628 * to parent_bs. That reference is later dropped by parent_bs on
2629 * bdrv_close() or if someone calls bdrv_unref_child().
2630 *
2631 * On failure NULL is returned, errp is set and the reference to
2632 * child_bs is also dropped.
2633 *
2634 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2635 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2636 */
2637 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2638 BlockDriverState *child_bs,
2639 const char *child_name,
2640 const BdrvChildRole *child_role,
2641 Error **errp)
2642 {
2643 BdrvChild *child;
2644 uint64_t perm, shared_perm;
2645
2646 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2647
2648 assert(parent_bs->drv);
2649 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2650 perm, shared_perm, &perm, &shared_perm);
2651
2652 child = bdrv_root_attach_child(child_bs, child_name, child_role,
2653 bdrv_get_aio_context(parent_bs),
2654 perm, shared_perm, parent_bs, errp);
2655 if (child == NULL) {
2656 return NULL;
2657 }
2658
2659 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2660 return child;
2661 }
2662
2663 static void bdrv_detach_child(BdrvChild *child)
2664 {
2665 QLIST_SAFE_REMOVE(child, next);
2666
2667 bdrv_replace_child(child, NULL);
2668
2669 g_free(child->name);
2670 g_free(child);
2671 }
2672
2673 void bdrv_root_unref_child(BdrvChild *child)
2674 {
2675 BlockDriverState *child_bs;
2676
2677 child_bs = child->bs;
2678 bdrv_detach_child(child);
2679 bdrv_unref(child_bs);
2680 }
2681
2682 /**
2683 * Clear all inherits_from pointers from children and grandchildren of
2684 * @root that point to @root, where necessary.
2685 */
2686 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child)
2687 {
2688 BdrvChild *c;
2689
2690 if (child->bs->inherits_from == root) {
2691 /*
2692 * Remove inherits_from only when the last reference between root and
2693 * child->bs goes away.
2694 */
2695 QLIST_FOREACH(c, &root->children, next) {
2696 if (c != child && c->bs == child->bs) {
2697 break;
2698 }
2699 }
2700 if (c == NULL) {
2701 child->bs->inherits_from = NULL;
2702 }
2703 }
2704
2705 QLIST_FOREACH(c, &child->bs->children, next) {
2706 bdrv_unset_inherits_from(root, c);
2707 }
2708 }
2709
2710 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2711 {
2712 if (child == NULL) {
2713 return;
2714 }
2715
2716 bdrv_unset_inherits_from(parent, child);
2717 bdrv_root_unref_child(child);
2718 }
2719
2720
2721 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2722 {
2723 BdrvChild *c;
2724 QLIST_FOREACH(c, &bs->parents, next_parent) {
2725 if (c->role->change_media) {
2726 c->role->change_media(c, load);
2727 }
2728 }
2729 }
2730
2731 /* Return true if you can reach parent going through child->inherits_from
2732 * recursively. If parent or child are NULL, return false */
2733 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2734 BlockDriverState *parent)
2735 {
2736 while (child && child != parent) {
2737 child = child->inherits_from;
2738 }
2739
2740 return child != NULL;
2741 }
2742
2743 /*
2744 * Sets the backing file link of a BDS. A new reference is created; callers
2745 * which don't need their own reference any more must call bdrv_unref().
2746 */
2747 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2748 Error **errp)
2749 {
2750 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2751 bdrv_inherits_from_recursive(backing_hd, bs);
2752
2753 if (bdrv_is_backing_chain_frozen(bs, backing_bs(bs), errp)) {
2754 return;
2755 }
2756
2757 if (backing_hd) {
2758 bdrv_ref(backing_hd);
2759 }
2760
2761 if (bs->backing) {
2762 bdrv_unref_child(bs, bs->backing);
2763 bs->backing = NULL;
2764 }
2765
2766 if (!backing_hd) {
2767 goto out;
2768 }
2769
2770 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2771 errp);
2772 /* If backing_hd was already part of bs's backing chain, and
2773 * inherits_from pointed recursively to bs then let's update it to
2774 * point directly to bs (else it will become NULL). */
2775 if (bs->backing && update_inherits_from) {
2776 backing_hd->inherits_from = bs;
2777 }
2778
2779 out:
2780 bdrv_refresh_limits(bs, NULL);
2781 }
2782
2783 /*
2784 * Opens the backing file for a BlockDriverState if not yet open
2785 *
2786 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2787 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2788 * itself, all options starting with "${bdref_key}." are considered part of the
2789 * BlockdevRef.
2790 *
2791 * TODO Can this be unified with bdrv_open_image()?
2792 */
2793 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2794 const char *bdref_key, Error **errp)
2795 {
2796 char *backing_filename = NULL;
2797 char *bdref_key_dot;
2798 const char *reference = NULL;
2799 int ret = 0;
2800 bool implicit_backing = false;
2801 BlockDriverState *backing_hd;
2802 QDict *options;
2803 QDict *tmp_parent_options = NULL;
2804 Error *local_err = NULL;
2805
2806 if (bs->backing != NULL) {
2807 goto free_exit;
2808 }
2809
2810 /* NULL means an empty set of options */
2811 if (parent_options == NULL) {
2812 tmp_parent_options = qdict_new();
2813 parent_options = tmp_parent_options;
2814 }
2815
2816 bs->open_flags &= ~BDRV_O_NO_BACKING;
2817
2818 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2819 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2820 g_free(bdref_key_dot);
2821
2822 /*
2823 * Caution: while qdict_get_try_str() is fine, getting non-string
2824 * types would require more care. When @parent_options come from
2825 * -blockdev or blockdev_add, its members are typed according to
2826 * the QAPI schema, but when they come from -drive, they're all
2827 * QString.
2828 */
2829 reference = qdict_get_try_str(parent_options, bdref_key);
2830 if (reference || qdict_haskey(options, "file.filename")) {
2831 /* keep backing_filename NULL */
2832 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2833 qobject_unref(options);
2834 goto free_exit;
2835 } else {
2836 if (qdict_size(options) == 0) {
2837 /* If the user specifies options that do not modify the
2838 * backing file's behavior, we might still consider it the
2839 * implicit backing file. But it's easier this way, and
2840 * just specifying some of the backing BDS's options is
2841 * only possible with -drive anyway (otherwise the QAPI
2842 * schema forces the user to specify everything). */
2843 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2844 }
2845
2846 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2847 if (local_err) {
2848 ret = -EINVAL;
2849 error_propagate(errp, local_err);
2850 qobject_unref(options);
2851 goto free_exit;
2852 }
2853 }
2854
2855 if (!bs->drv || !bs->drv->supports_backing) {
2856 ret = -EINVAL;
2857 error_setg(errp, "Driver doesn't support backing files");
2858 qobject_unref(options);
2859 goto free_exit;
2860 }
2861
2862 if (!reference &&
2863 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2864 qdict_put_str(options, "driver", bs->backing_format);
2865 }
2866
2867 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2868 &child_backing, errp);
2869 if (!backing_hd) {
2870 bs->open_flags |= BDRV_O_NO_BACKING;
2871 error_prepend(errp, "Could not open backing file: ");
2872 ret = -EINVAL;
2873 goto free_exit;
2874 }
2875
2876 if (implicit_backing) {
2877 bdrv_refresh_filename(backing_hd);
2878 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2879 backing_hd->filename);
2880 }
2881
2882 /* Hook up the backing file link; drop our reference, bs owns the
2883 * backing_hd reference now */
2884 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2885 bdrv_unref(backing_hd);
2886 if (local_err) {
2887 error_propagate(errp, local_err);
2888 ret = -EINVAL;
2889 goto free_exit;
2890 }
2891
2892 qdict_del(parent_options, bdref_key);
2893
2894 free_exit:
2895 g_free(backing_filename);
2896 qobject_unref(tmp_parent_options);
2897 return ret;
2898 }
2899
2900 static BlockDriverState *
2901 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2902 BlockDriverState *parent, const BdrvChildRole *child_role,
2903 bool allow_none, Error **errp)
2904 {
2905 BlockDriverState *bs = NULL;
2906 QDict *image_options;
2907 char *bdref_key_dot;
2908 const char *reference;
2909
2910 assert(child_role != NULL);
2911
2912 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2913 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2914 g_free(bdref_key_dot);
2915
2916 /*
2917 * Caution: while qdict_get_try_str() is fine, getting non-string
2918 * types would require more care. When @options come from
2919 * -blockdev or blockdev_add, its members are typed according to
2920 * the QAPI schema, but when they come from -drive, they're all
2921 * QString.
2922 */
2923 reference = qdict_get_try_str(options, bdref_key);
2924 if (!filename && !reference && !qdict_size(image_options)) {
2925 if (!allow_none) {
2926 error_setg(errp, "A block device must be specified for \"%s\"",
2927 bdref_key);
2928 }
2929 qobject_unref(image_options);
2930 goto done;
2931 }
2932
2933 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2934 parent, child_role, errp);
2935 if (!bs) {
2936 goto done;
2937 }
2938
2939 done:
2940 qdict_del(options, bdref_key);
2941 return bs;
2942 }
2943
2944 /*
2945 * Opens a disk image whose options are given as BlockdevRef in another block
2946 * device's options.
2947 *
2948 * If allow_none is true, no image will be opened if filename is false and no
2949 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2950 *
2951 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2952 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2953 * itself, all options starting with "${bdref_key}." are considered part of the
2954 * BlockdevRef.
2955 *
2956 * The BlockdevRef will be removed from the options QDict.
2957 */
2958 BdrvChild *bdrv_open_child(const char *filename,
2959 QDict *options, const char *bdref_key,
2960 BlockDriverState *parent,
2961 const BdrvChildRole *child_role,
2962 bool allow_none, Error **errp)
2963 {
2964 BlockDriverState *bs;
2965
2966 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2967 allow_none, errp);
2968 if (bs == NULL) {
2969 return NULL;
2970 }
2971
2972 return bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2973 }
2974
2975 /* TODO Future callers may need to specify parent/child_role in order for
2976 * option inheritance to work. Existing callers use it for the root node. */
2977 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
2978 {
2979 BlockDriverState *bs = NULL;
2980 Error *local_err = NULL;
2981 QObject *obj = NULL;
2982 QDict *qdict = NULL;
2983 const char *reference = NULL;
2984 Visitor *v = NULL;
2985
2986 if (ref->type == QTYPE_QSTRING) {
2987 reference = ref->u.reference;
2988 } else {
2989 BlockdevOptions *options = &ref->u.definition;
2990 assert(ref->type == QTYPE_QDICT);
2991
2992 v = qobject_output_visitor_new(&obj);
2993 visit_type_BlockdevOptions(v, NULL, &options, &local_err);
2994 if (local_err) {
2995 error_propagate(errp, local_err);
2996 goto fail;
2997 }
2998 visit_complete(v, &obj);
2999
3000 qdict = qobject_to(QDict, obj);
3001 qdict_flatten(qdict);
3002
3003 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3004 * compatibility with other callers) rather than what we want as the
3005 * real defaults. Apply the defaults here instead. */
3006 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3007 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3008 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3009 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3010
3011 }
3012
3013 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, errp);
3014 obj = NULL;
3015
3016 fail:
3017 qobject_unref(obj);
3018 visit_free(v);
3019 return bs;
3020 }
3021
3022 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3023 int flags,
3024 QDict *snapshot_options,
3025 Error **errp)
3026 {
3027 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3028 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3029 int64_t total_size;
3030 QemuOpts *opts = NULL;
3031 BlockDriverState *bs_snapshot = NULL;
3032 Error *local_err = NULL;
3033 int ret;
3034
3035 /* if snapshot, we create a temporary backing file and open it
3036 instead of opening 'filename' directly */
3037
3038 /* Get the required size from the image */
3039 total_size = bdrv_getlength(bs);
3040 if (total_size < 0) {
3041 error_setg_errno(errp, -total_size, "Could not get image size");
3042 goto out;
3043 }
3044
3045 /* Create the temporary image */
3046 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3047 if (ret < 0) {
3048 error_setg_errno(errp, -ret, "Could not get temporary filename");
3049 goto out;
3050 }
3051
3052 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3053 &error_abort);
3054 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3055 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3056 qemu_opts_del(opts);
3057 if (ret < 0) {
3058 error_prepend(errp, "Could not create temporary overlay '%s': ",
3059 tmp_filename);
3060 goto out;
3061 }
3062
3063 /* Prepare options QDict for the temporary file */
3064 qdict_put_str(snapshot_options, "file.driver", "file");
3065 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3066 qdict_put_str(snapshot_options, "driver", "qcow2");
3067
3068 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3069 snapshot_options = NULL;
3070 if (!bs_snapshot) {
3071 goto out;
3072 }
3073
3074 /* bdrv_append() consumes a strong reference to bs_snapshot
3075 * (i.e. it will call bdrv_unref() on it) even on error, so in
3076 * order to be able to return one, we have to increase
3077 * bs_snapshot's refcount here */
3078 bdrv_ref(bs_snapshot);
3079 bdrv_append(bs_snapshot, bs, &local_err);
3080 if (local_err) {
3081 error_propagate(errp, local_err);
3082 bs_snapshot = NULL;
3083 goto out;
3084 }
3085
3086 out:
3087 qobject_unref(snapshot_options);
3088 g_free(tmp_filename);
3089 return bs_snapshot;
3090 }
3091
3092 /*
3093 * Opens a disk image (raw, qcow2, vmdk, ...)
3094 *
3095 * options is a QDict of options to pass to the block drivers, or NULL for an
3096 * empty set of options. The reference to the QDict belongs to the block layer
3097 * after the call (even on failure), so if the caller intends to reuse the
3098 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3099 *
3100 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3101 * If it is not NULL, the referenced BDS will be reused.
3102 *
3103 * The reference parameter may be used to specify an existing block device which
3104 * should be opened. If specified, neither options nor a filename may be given,
3105 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3106 */
3107 static BlockDriverState *bdrv_open_inherit(const char *filename,
3108 const char *reference,
3109 QDict *options, int flags,
3110 BlockDriverState *parent,
3111 const BdrvChildRole *child_role,
3112 Error **errp)
3113 {
3114 int ret;
3115 BlockBackend *file = NULL;
3116 BlockDriverState *bs;
3117 BlockDriver *drv = NULL;
3118 BdrvChild *child;
3119 const char *drvname;
3120 const char *backing;
3121 Error *local_err = NULL;
3122 QDict *snapshot_options = NULL;
3123 int snapshot_flags = 0;
3124
3125 assert(!child_role || !flags);
3126 assert(!child_role == !parent);
3127
3128 if (reference) {
3129 bool options_non_empty = options ? qdict_size(options) : false;
3130 qobject_unref(options);
3131
3132 if (filename || options_non_empty) {
3133 error_setg(errp, "Cannot reference an existing block device with "
3134 "additional options or a new filename");
3135 return NULL;
3136 }
3137
3138 bs = bdrv_lookup_bs(reference, reference, errp);
3139 if (!bs) {
3140 return NULL;
3141 }
3142
3143 bdrv_ref(bs);
3144 return bs;
3145 }
3146
3147 bs = bdrv_new();
3148
3149 /* NULL means an empty set of options */
3150 if (options == NULL) {
3151 options = qdict_new();
3152 }
3153
3154 /* json: syntax counts as explicit options, as if in the QDict */
3155 parse_json_protocol(options, &filename, &local_err);
3156 if (local_err) {
3157 goto fail;
3158 }
3159
3160 bs->explicit_options = qdict_clone_shallow(options);
3161
3162 if (child_role) {
3163 bs->inherits_from = parent;
3164 child_role->inherit_options(&flags, options,
3165 parent->open_flags, parent->options);
3166 }
3167
3168 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3169 if (local_err) {
3170 goto fail;
3171 }
3172
3173 /*
3174 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3175 * Caution: getting a boolean member of @options requires care.
3176 * When @options come from -blockdev or blockdev_add, members are
3177 * typed according to the QAPI schema, but when they come from
3178 * -drive, they're all QString.
3179 */
3180 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3181 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3182 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3183 } else {
3184 flags &= ~BDRV_O_RDWR;
3185 }
3186
3187 if (flags & BDRV_O_SNAPSHOT) {
3188 snapshot_options = qdict_new();
3189 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3190 flags, options);
3191 /* Let bdrv_backing_options() override "read-only" */
3192 qdict_del(options, BDRV_OPT_READ_ONLY);
3193 bdrv_backing_options(&flags, options, flags, options);
3194 }
3195
3196 bs->open_flags = flags;
3197 bs->options = options;
3198 options = qdict_clone_shallow(options);
3199
3200 /* Find the right image format driver */
3201 /* See cautionary note on accessing @options above */
3202 drvname = qdict_get_try_str(options, "driver");
3203 if (drvname) {
3204 drv = bdrv_find_format(drvname);
3205 if (!drv) {
3206 error_setg(errp, "Unknown driver: '%s'", drvname);
3207 goto fail;
3208 }
3209 }
3210
3211 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3212
3213 /* See cautionary note on accessing @options above */
3214 backing = qdict_get_try_str(options, "backing");
3215 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3216 (backing && *backing == '\0'))
3217 {
3218 if (backing) {
3219 warn_report("Use of \"backing\": \"\" is deprecated; "
3220 "use \"backing\": null instead");
3221 }
3222 flags |= BDRV_O_NO_BACKING;
3223 qdict_del(bs->explicit_options, "backing");
3224 qdict_del(bs->options, "backing");
3225 qdict_del(options, "backing");
3226 }
3227
3228 /* Open image file without format layer. This BlockBackend is only used for
3229 * probing, the block drivers will do their own bdrv_open_child() for the
3230 * same BDS, which is why we put the node name back into options. */
3231 if ((flags & BDRV_O_PROTOCOL) == 0) {
3232 BlockDriverState *file_bs;
3233
3234 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3235 &child_file, true, &local_err);
3236 if (local_err) {
3237 goto fail;
3238 }
3239 if (file_bs != NULL) {
3240 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3241 * looking at the header to guess the image format. This works even
3242 * in cases where a guest would not see a consistent state. */
3243 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3244 blk_insert_bs(file, file_bs, &local_err);
3245 bdrv_unref(file_bs);
3246 if (local_err) {
3247 goto fail;
3248 }
3249
3250 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3251 }
3252 }
3253
3254 /* Image format probing */
3255 bs->probed = !drv;
3256 if (!drv && file) {
3257 ret = find_image_format(file, filename, &drv, &local_err);
3258 if (ret < 0) {
3259 goto fail;
3260 }
3261 /*
3262 * This option update would logically belong in bdrv_fill_options(),
3263 * but we first need to open bs->file for the probing to work, while
3264 * opening bs->file already requires the (mostly) final set of options
3265 * so that cache mode etc. can be inherited.
3266 *
3267 * Adding the driver later is somewhat ugly, but it's not an option
3268 * that would ever be inherited, so it's correct. We just need to make
3269 * sure to update both bs->options (which has the full effective
3270 * options for bs) and options (which has file.* already removed).
3271 */
3272 qdict_put_str(bs->options, "driver", drv->format_name);
3273 qdict_put_str(options, "driver", drv->format_name);
3274 } else if (!drv) {
3275 error_setg(errp, "Must specify either driver or file");
3276 goto fail;
3277 }
3278
3279 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3280 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3281 /* file must be NULL if a protocol BDS is about to be created
3282 * (the inverse results in an error message from bdrv_open_common()) */
3283 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3284
3285 /* Open the image */
3286 ret = bdrv_open_common(bs, file, options, &local_err);
3287 if (ret < 0) {
3288 goto fail;
3289 }
3290
3291 if (file) {
3292 blk_unref(file);
3293 file = NULL;
3294 }
3295
3296 /* If there is a backing file, use it */
3297 if ((flags & BDRV_O_NO_BACKING) == 0) {
3298 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3299 if (ret < 0) {
3300 goto close_and_fail;
3301 }
3302 }
3303
3304 /* Remove all children options and references
3305 * from bs->options and bs->explicit_options */
3306 QLIST_FOREACH(child, &bs->children, next) {
3307 char *child_key_dot;
3308 child_key_dot = g_strdup_printf("%s.", child->name);
3309 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3310 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3311 qdict_del(bs->explicit_options, child->name);
3312 qdict_del(bs->options, child->name);
3313 g_free(child_key_dot);
3314 }
3315
3316 /* Check if any unknown options were used */
3317 if (qdict_size(options) != 0) {
3318 const QDictEntry *entry = qdict_first(options);
3319 if (flags & BDRV_O_PROTOCOL) {
3320 error_setg(errp, "Block protocol '%s' doesn't support the option "
3321 "'%s'", drv->format_name, entry->key);
3322 } else {
3323 error_setg(errp,
3324 "Block format '%s' does not support the option '%s'",
3325 drv->format_name, entry->key);
3326 }
3327
3328 goto close_and_fail;
3329 }
3330
3331 bdrv_parent_cb_change_media(bs, true);
3332
3333 qobject_unref(options);
3334 options = NULL;
3335
3336 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3337 * temporary snapshot afterwards. */
3338 if (snapshot_flags) {
3339 BlockDriverState *snapshot_bs;
3340 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3341 snapshot_options, &local_err);
3342 snapshot_options = NULL;
3343 if (local_err) {
3344 goto close_and_fail;
3345 }
3346 /* We are not going to return bs but the overlay on top of it
3347 * (snapshot_bs); thus, we have to drop the strong reference to bs
3348 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3349 * though, because the overlay still has a reference to it. */
3350 bdrv_unref(bs);
3351 bs = snapshot_bs;
3352 }
3353
3354 return bs;
3355
3356 fail:
3357 blk_unref(file);
3358 qobject_unref(snapshot_options);
3359 qobject_unref(bs->explicit_options);
3360 qobject_unref(bs->options);
3361 qobject_unref(options);
3362 bs->options = NULL;
3363 bs->explicit_options = NULL;
3364 bdrv_unref(bs);
3365 error_propagate(errp, local_err);
3366 return NULL;
3367
3368 close_and_fail:
3369 bdrv_unref(bs);
3370 qobject_unref(snapshot_options);
3371 qobject_unref(options);
3372 error_propagate(errp, local_err);
3373 return NULL;
3374 }
3375
3376 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3377 QDict *options, int flags, Error **errp)
3378 {
3379 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3380 NULL, errp);
3381 }
3382
3383 /* Return true if the NULL-terminated @list contains @str */
3384 static bool is_str_in_list(const char *str, const char *const *list)
3385 {
3386 if (str && list) {
3387 int i;
3388 for (i = 0; list[i] != NULL; i++) {
3389 if (!strcmp(str, list[i])) {
3390 return true;
3391 }
3392 }
3393 }
3394 return false;
3395 }
3396
3397 /*
3398 * Check that every option set in @bs->options is also set in
3399 * @new_opts.
3400 *
3401 * Options listed in the common_options list and in
3402 * @bs->drv->mutable_opts are skipped.
3403 *
3404 * Return 0 on success, otherwise return -EINVAL and set @errp.
3405 */
3406 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3407 const QDict *new_opts, Error **errp)
3408 {
3409 const QDictEntry *e;
3410 /* These options are common to all block drivers and are handled
3411 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3412 const char *const common_options[] = {
3413 "node-name", "discard", "cache.direct", "cache.no-flush",
3414 "read-only", "auto-read-only", "detect-zeroes", NULL
3415 };
3416
3417 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3418 if (!qdict_haskey(new_opts, e->key) &&
3419 !is_str_in_list(e->key, common_options) &&
3420 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3421 error_setg(errp, "Option '%s' cannot be reset "
3422 "to its default value", e->key);
3423 return -EINVAL;
3424 }
3425 }
3426
3427 return 0;
3428 }
3429
3430 /*
3431 * Returns true if @child can be reached recursively from @bs
3432 */
3433 static bool bdrv_recurse_has_child(BlockDriverState *bs,
3434 BlockDriverState *child)
3435 {
3436 BdrvChild *c;
3437
3438 if (bs == child) {
3439 return true;
3440 }
3441
3442 QLIST_FOREACH(c, &bs->children, next) {
3443 if (bdrv_recurse_has_child(c->bs, child)) {
3444 return true;
3445 }
3446 }
3447
3448 return false;
3449 }
3450
3451 /*
3452 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3453 * reopen of multiple devices.
3454 *
3455 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3456 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3457 * be created and initialized. This newly created BlockReopenQueue should be
3458 * passed back in for subsequent calls that are intended to be of the same
3459 * atomic 'set'.
3460 *
3461 * bs is the BlockDriverState to add to the reopen queue.
3462 *
3463 * options contains the changed options for the associated bs
3464 * (the BlockReopenQueue takes ownership)
3465 *
3466 * flags contains the open flags for the associated bs
3467 *
3468 * returns a pointer to bs_queue, which is either the newly allocated
3469 * bs_queue, or the existing bs_queue being used.
3470 *
3471 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3472 */
3473 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
3474 BlockDriverState *bs,
3475 QDict *options,
3476 const BdrvChildRole *role,
3477 QDict *parent_options,
3478 int parent_flags,
3479 bool keep_old_opts)
3480 {
3481 assert(bs != NULL);
3482
3483 BlockReopenQueueEntry *bs_entry;
3484 BdrvChild *child;
3485 QDict *old_options, *explicit_options, *options_copy;
3486 int flags;
3487 QemuOpts *opts;
3488
3489 /* Make sure that the caller remembered to use a drained section. This is
3490 * important to avoid graph changes between the recursive queuing here and
3491 * bdrv_reopen_multiple(). */
3492 assert(bs->quiesce_counter > 0);
3493
3494 if (bs_queue == NULL) {
3495 bs_queue = g_new0(BlockReopenQueue, 1);
3496 QTAILQ_INIT(bs_queue);
3497 }
3498
3499 if (!options) {
3500 options = qdict_new();
3501 }
3502
3503 /* Check if this BlockDriverState is already in the queue */
3504 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3505 if (bs == bs_entry->state.bs) {
3506 break;
3507 }
3508 }
3509
3510 /*
3511 * Precedence of options:
3512 * 1. Explicitly passed in options (highest)
3513 * 2. Retained from explicitly set options of bs
3514 * 3. Inherited from parent node
3515 * 4. Retained from effective options of bs
3516 */
3517
3518 /* Old explicitly set values (don't overwrite by inherited value) */
3519 if (bs_entry || keep_old_opts) {
3520 old_options = qdict_clone_shallow(bs_entry ?
3521 bs_entry->state.explicit_options :
3522 bs->explicit_options);
3523 bdrv_join_options(bs, options, old_options);
3524 qobject_unref(old_options);
3525 }
3526
3527 explicit_options = qdict_clone_shallow(options);
3528
3529 /* Inherit from parent node */
3530 if (parent_options) {
3531 flags = 0;
3532 role->inherit_options(&flags, options, parent_flags, parent_options);
3533 } else {
3534 flags = bdrv_get_flags(bs);
3535 }
3536
3537 if (keep_old_opts) {
3538 /* Old values are used for options that aren't set yet */
3539 old_options = qdict_clone_shallow(bs->options);
3540 bdrv_join_options(bs, options, old_options);
3541 qobject_unref(old_options);
3542 }
3543
3544 /* We have the final set of options so let's update the flags */
3545 options_copy = qdict_clone_shallow(options);
3546 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3547 qemu_opts_absorb_qdict(opts, options_copy, NULL);
3548 update_flags_from_options(&flags, opts);
3549 qemu_opts_del(opts);
3550 qobject_unref(options_copy);
3551
3552 /* bdrv_open_inherit() sets and clears some additional flags internally */
3553 flags &= ~BDRV_O_PROTOCOL;
3554 if (flags & BDRV_O_RDWR