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