fuzz: use qemu_get_exec_dir
[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 "block/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 BdrvChildClass *child_class,
80 BdrvChildRole child_role,
81 Error **errp);
82
83 /* If non-zero, use only whitelisted block drivers */
84 static int use_bdrv_whitelist;
85
86 #ifdef _WIN32
87 static int is_windows_drive_prefix(const char *filename)
88 {
89 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
90 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
91 filename[1] == ':');
92 }
93
94 int is_windows_drive(const char *filename)
95 {
96 if (is_windows_drive_prefix(filename) &&
97 filename[2] == '\0')
98 return 1;
99 if (strstart(filename, "\\\\.\\", NULL) ||
100 strstart(filename, "//./", NULL))
101 return 1;
102 return 0;
103 }
104 #endif
105
106 size_t bdrv_opt_mem_align(BlockDriverState *bs)
107 {
108 if (!bs || !bs->drv) {
109 /* page size or 4k (hdd sector size) should be on the safe side */
110 return MAX(4096, qemu_real_host_page_size);
111 }
112
113 return bs->bl.opt_mem_alignment;
114 }
115
116 size_t bdrv_min_mem_align(BlockDriverState *bs)
117 {
118 if (!bs || !bs->drv) {
119 /* page size or 4k (hdd sector size) should be on the safe side */
120 return MAX(4096, qemu_real_host_page_size);
121 }
122
123 return bs->bl.min_mem_alignment;
124 }
125
126 /* check if the path starts with "<protocol>:" */
127 int path_has_protocol(const char *path)
128 {
129 const char *p;
130
131 #ifdef _WIN32
132 if (is_windows_drive(path) ||
133 is_windows_drive_prefix(path)) {
134 return 0;
135 }
136 p = path + strcspn(path, ":/\\");
137 #else
138 p = path + strcspn(path, ":/");
139 #endif
140
141 return *p == ':';
142 }
143
144 int path_is_absolute(const char *path)
145 {
146 #ifdef _WIN32
147 /* specific case for names like: "\\.\d:" */
148 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
149 return 1;
150 }
151 return (*path == '/' || *path == '\\');
152 #else
153 return (*path == '/');
154 #endif
155 }
156
157 /* if filename is absolute, just return its duplicate. Otherwise, build a
158 path to it by considering it is relative to base_path. URL are
159 supported. */
160 char *path_combine(const char *base_path, const char *filename)
161 {
162 const char *protocol_stripped = NULL;
163 const char *p, *p1;
164 char *result;
165 int len;
166
167 if (path_is_absolute(filename)) {
168 return g_strdup(filename);
169 }
170
171 if (path_has_protocol(base_path)) {
172 protocol_stripped = strchr(base_path, ':');
173 if (protocol_stripped) {
174 protocol_stripped++;
175 }
176 }
177 p = protocol_stripped ?: base_path;
178
179 p1 = strrchr(base_path, '/');
180 #ifdef _WIN32
181 {
182 const char *p2;
183 p2 = strrchr(base_path, '\\');
184 if (!p1 || p2 > p1) {
185 p1 = p2;
186 }
187 }
188 #endif
189 if (p1) {
190 p1++;
191 } else {
192 p1 = base_path;
193 }
194 if (p1 > p) {
195 p = p1;
196 }
197 len = p - base_path;
198
199 result = g_malloc(len + strlen(filename) + 1);
200 memcpy(result, base_path, len);
201 strcpy(result + len, filename);
202
203 return result;
204 }
205
206 /*
207 * Helper function for bdrv_parse_filename() implementations to remove optional
208 * protocol prefixes (especially "file:") from a filename and for putting the
209 * stripped filename into the options QDict if there is such a prefix.
210 */
211 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
212 QDict *options)
213 {
214 if (strstart(filename, prefix, &filename)) {
215 /* Stripping the explicit protocol prefix may result in a protocol
216 * prefix being (wrongly) detected (if the filename contains a colon) */
217 if (path_has_protocol(filename)) {
218 QString *fat_filename;
219
220 /* This means there is some colon before the first slash; therefore,
221 * this cannot be an absolute path */
222 assert(!path_is_absolute(filename));
223
224 /* And we can thus fix the protocol detection issue by prefixing it
225 * by "./" */
226 fat_filename = qstring_from_str("./");
227 qstring_append(fat_filename, filename);
228
229 assert(!path_has_protocol(qstring_get_str(fat_filename)));
230
231 qdict_put(options, "filename", fat_filename);
232 } else {
233 /* If no protocol prefix was detected, we can use the shortened
234 * filename as-is */
235 qdict_put_str(options, "filename", filename);
236 }
237 }
238 }
239
240
241 /* Returns whether the image file is opened as read-only. Note that this can
242 * return false and writing to the image file is still not possible because the
243 * image is inactivated. */
244 bool bdrv_is_read_only(BlockDriverState *bs)
245 {
246 return bs->read_only;
247 }
248
249 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
250 bool ignore_allow_rdw, Error **errp)
251 {
252 /* Do not set read_only if copy_on_read is enabled */
253 if (bs->copy_on_read && read_only) {
254 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
255 bdrv_get_device_or_node_name(bs));
256 return -EINVAL;
257 }
258
259 /* Do not clear read_only if it is prohibited */
260 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
261 !ignore_allow_rdw)
262 {
263 error_setg(errp, "Node '%s' is read only",
264 bdrv_get_device_or_node_name(bs));
265 return -EPERM;
266 }
267
268 return 0;
269 }
270
271 /*
272 * Called by a driver that can only provide a read-only image.
273 *
274 * Returns 0 if the node is already read-only or it could switch the node to
275 * read-only because BDRV_O_AUTO_RDONLY is set.
276 *
277 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
278 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
279 * is not NULL, it is used as the error message for the Error object.
280 */
281 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
282 Error **errp)
283 {
284 int ret = 0;
285
286 if (!(bs->open_flags & BDRV_O_RDWR)) {
287 return 0;
288 }
289 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
290 goto fail;
291 }
292
293 ret = bdrv_can_set_read_only(bs, true, false, NULL);
294 if (ret < 0) {
295 goto fail;
296 }
297
298 bs->read_only = true;
299 bs->open_flags &= ~BDRV_O_RDWR;
300
301 return 0;
302
303 fail:
304 error_setg(errp, "%s", errmsg ?: "Image is read-only");
305 return -EACCES;
306 }
307
308 /*
309 * If @backing is empty, this function returns NULL without setting
310 * @errp. In all other cases, NULL will only be returned with @errp
311 * set.
312 *
313 * Therefore, a return value of NULL without @errp set means that
314 * there is no backing file; if @errp is set, there is one but its
315 * absolute filename cannot be generated.
316 */
317 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
318 const char *backing,
319 Error **errp)
320 {
321 if (backing[0] == '\0') {
322 return NULL;
323 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
324 return g_strdup(backing);
325 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
326 error_setg(errp, "Cannot use relative backing file names for '%s'",
327 backed);
328 return NULL;
329 } else {
330 return path_combine(backed, backing);
331 }
332 }
333
334 /*
335 * If @filename is empty or NULL, this function returns NULL without
336 * setting @errp. In all other cases, NULL will only be returned with
337 * @errp set.
338 */
339 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
340 const char *filename, Error **errp)
341 {
342 char *dir, *full_name;
343
344 if (!filename || filename[0] == '\0') {
345 return NULL;
346 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
347 return g_strdup(filename);
348 }
349
350 dir = bdrv_dirname(relative_to, errp);
351 if (!dir) {
352 return NULL;
353 }
354
355 full_name = g_strconcat(dir, filename, NULL);
356 g_free(dir);
357 return full_name;
358 }
359
360 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
361 {
362 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
363 }
364
365 void bdrv_register(BlockDriver *bdrv)
366 {
367 assert(bdrv->format_name);
368 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
369 }
370
371 BlockDriverState *bdrv_new(void)
372 {
373 BlockDriverState *bs;
374 int i;
375
376 bs = g_new0(BlockDriverState, 1);
377 QLIST_INIT(&bs->dirty_bitmaps);
378 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
379 QLIST_INIT(&bs->op_blockers[i]);
380 }
381 notifier_with_return_list_init(&bs->before_write_notifiers);
382 qemu_co_mutex_init(&bs->reqs_lock);
383 qemu_mutex_init(&bs->dirty_bitmap_mutex);
384 bs->refcnt = 1;
385 bs->aio_context = qemu_get_aio_context();
386
387 qemu_co_queue_init(&bs->flush_queue);
388
389 for (i = 0; i < bdrv_drain_all_count; i++) {
390 bdrv_drained_begin(bs);
391 }
392
393 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
394
395 return bs;
396 }
397
398 static BlockDriver *bdrv_do_find_format(const char *format_name)
399 {
400 BlockDriver *drv1;
401
402 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
403 if (!strcmp(drv1->format_name, format_name)) {
404 return drv1;
405 }
406 }
407
408 return NULL;
409 }
410
411 BlockDriver *bdrv_find_format(const char *format_name)
412 {
413 BlockDriver *drv1;
414 int i;
415
416 drv1 = bdrv_do_find_format(format_name);
417 if (drv1) {
418 return drv1;
419 }
420
421 /* The driver isn't registered, maybe we need to load a module */
422 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
423 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
424 block_module_load_one(block_driver_modules[i].library_name);
425 break;
426 }
427 }
428
429 return bdrv_do_find_format(format_name);
430 }
431
432 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
433 {
434 static const char *whitelist_rw[] = {
435 CONFIG_BDRV_RW_WHITELIST
436 NULL
437 };
438 static const char *whitelist_ro[] = {
439 CONFIG_BDRV_RO_WHITELIST
440 NULL
441 };
442 const char **p;
443
444 if (!whitelist_rw[0] && !whitelist_ro[0]) {
445 return 1; /* no whitelist, anything goes */
446 }
447
448 for (p = whitelist_rw; *p; p++) {
449 if (!strcmp(format_name, *p)) {
450 return 1;
451 }
452 }
453 if (read_only) {
454 for (p = whitelist_ro; *p; p++) {
455 if (!strcmp(format_name, *p)) {
456 return 1;
457 }
458 }
459 }
460 return 0;
461 }
462
463 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
464 {
465 return bdrv_format_is_whitelisted(drv->format_name, read_only);
466 }
467
468 bool bdrv_uses_whitelist(void)
469 {
470 return use_bdrv_whitelist;
471 }
472
473 typedef struct CreateCo {
474 BlockDriver *drv;
475 char *filename;
476 QemuOpts *opts;
477 int ret;
478 Error *err;
479 } CreateCo;
480
481 static void coroutine_fn bdrv_create_co_entry(void *opaque)
482 {
483 Error *local_err = NULL;
484 int ret;
485
486 CreateCo *cco = opaque;
487 assert(cco->drv);
488
489 ret = cco->drv->bdrv_co_create_opts(cco->drv,
490 cco->filename, cco->opts, &local_err);
491 error_propagate(&cco->err, local_err);
492 cco->ret = ret;
493 }
494
495 int bdrv_create(BlockDriver *drv, const char* filename,
496 QemuOpts *opts, Error **errp)
497 {
498 int ret;
499
500 Coroutine *co;
501 CreateCo cco = {
502 .drv = drv,
503 .filename = g_strdup(filename),
504 .opts = opts,
505 .ret = NOT_DONE,
506 .err = NULL,
507 };
508
509 if (!drv->bdrv_co_create_opts) {
510 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
511 ret = -ENOTSUP;
512 goto out;
513 }
514
515 if (qemu_in_coroutine()) {
516 /* Fast-path if already in coroutine context */
517 bdrv_create_co_entry(&cco);
518 } else {
519 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
520 qemu_coroutine_enter(co);
521 while (cco.ret == NOT_DONE) {
522 aio_poll(qemu_get_aio_context(), true);
523 }
524 }
525
526 ret = cco.ret;
527 if (ret < 0) {
528 if (cco.err) {
529 error_propagate(errp, cco.err);
530 } else {
531 error_setg_errno(errp, -ret, "Could not create image");
532 }
533 }
534
535 out:
536 g_free(cco.filename);
537 return ret;
538 }
539
540 /**
541 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
542 * least the given @minimum_size.
543 *
544 * On success, return @blk's actual length.
545 * Otherwise, return -errno.
546 */
547 static int64_t create_file_fallback_truncate(BlockBackend *blk,
548 int64_t minimum_size, Error **errp)
549 {
550 Error *local_err = NULL;
551 int64_t size;
552 int ret;
553
554 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
555 &local_err);
556 if (ret < 0 && ret != -ENOTSUP) {
557 error_propagate(errp, local_err);
558 return ret;
559 }
560
561 size = blk_getlength(blk);
562 if (size < 0) {
563 error_free(local_err);
564 error_setg_errno(errp, -size,
565 "Failed to inquire the new image file's length");
566 return size;
567 }
568
569 if (size < minimum_size) {
570 /* Need to grow the image, but we failed to do that */
571 error_propagate(errp, local_err);
572 return -ENOTSUP;
573 }
574
575 error_free(local_err);
576 local_err = NULL;
577
578 return size;
579 }
580
581 /**
582 * Helper function for bdrv_create_file_fallback(): Zero the first
583 * sector to remove any potentially pre-existing image header.
584 */
585 static int create_file_fallback_zero_first_sector(BlockBackend *blk,
586 int64_t current_size,
587 Error **errp)
588 {
589 int64_t bytes_to_clear;
590 int ret;
591
592 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
593 if (bytes_to_clear) {
594 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
595 if (ret < 0) {
596 error_setg_errno(errp, -ret,
597 "Failed to clear the new image's first sector");
598 return ret;
599 }
600 }
601
602 return 0;
603 }
604
605 /**
606 * Simple implementation of bdrv_co_create_opts for protocol drivers
607 * which only support creation via opening a file
608 * (usually existing raw storage device)
609 */
610 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
611 const char *filename,
612 QemuOpts *opts,
613 Error **errp)
614 {
615 BlockBackend *blk;
616 QDict *options;
617 int64_t size = 0;
618 char *buf = NULL;
619 PreallocMode prealloc;
620 Error *local_err = NULL;
621 int ret;
622
623 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
624 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
625 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
626 PREALLOC_MODE_OFF, &local_err);
627 g_free(buf);
628 if (local_err) {
629 error_propagate(errp, local_err);
630 return -EINVAL;
631 }
632
633 if (prealloc != PREALLOC_MODE_OFF) {
634 error_setg(errp, "Unsupported preallocation mode '%s'",
635 PreallocMode_str(prealloc));
636 return -ENOTSUP;
637 }
638
639 options = qdict_new();
640 qdict_put_str(options, "driver", drv->format_name);
641
642 blk = blk_new_open(filename, NULL, options,
643 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
644 if (!blk) {
645 error_prepend(errp, "Protocol driver '%s' does not support image "
646 "creation, and opening the image failed: ",
647 drv->format_name);
648 return -EINVAL;
649 }
650
651 size = create_file_fallback_truncate(blk, size, errp);
652 if (size < 0) {
653 ret = size;
654 goto out;
655 }
656
657 ret = create_file_fallback_zero_first_sector(blk, size, errp);
658 if (ret < 0) {
659 goto out;
660 }
661
662 ret = 0;
663 out:
664 blk_unref(blk);
665 return ret;
666 }
667
668 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
669 {
670 BlockDriver *drv;
671
672 drv = bdrv_find_protocol(filename, true, errp);
673 if (drv == NULL) {
674 return -ENOENT;
675 }
676
677 return bdrv_create(drv, filename, opts, errp);
678 }
679
680 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
681 {
682 Error *local_err = NULL;
683 int ret;
684
685 assert(bs != NULL);
686
687 if (!bs->drv) {
688 error_setg(errp, "Block node '%s' is not opened", bs->filename);
689 return -ENOMEDIUM;
690 }
691
692 if (!bs->drv->bdrv_co_delete_file) {
693 error_setg(errp, "Driver '%s' does not support image deletion",
694 bs->drv->format_name);
695 return -ENOTSUP;
696 }
697
698 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
699 if (ret < 0) {
700 error_propagate(errp, local_err);
701 }
702
703 return ret;
704 }
705
706 /**
707 * Try to get @bs's logical and physical block size.
708 * On success, store them in @bsz struct and return 0.
709 * On failure return -errno.
710 * @bs must not be empty.
711 */
712 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
713 {
714 BlockDriver *drv = bs->drv;
715 BlockDriverState *filtered = bdrv_filter_bs(bs);
716
717 if (drv && drv->bdrv_probe_blocksizes) {
718 return drv->bdrv_probe_blocksizes(bs, bsz);
719 } else if (filtered) {
720 return bdrv_probe_blocksizes(filtered, bsz);
721 }
722
723 return -ENOTSUP;
724 }
725
726 /**
727 * Try to get @bs's geometry (cyls, heads, sectors).
728 * On success, store them in @geo struct and return 0.
729 * On failure return -errno.
730 * @bs must not be empty.
731 */
732 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
733 {
734 BlockDriver *drv = bs->drv;
735 BlockDriverState *filtered = bdrv_filter_bs(bs);
736
737 if (drv && drv->bdrv_probe_geometry) {
738 return drv->bdrv_probe_geometry(bs, geo);
739 } else if (filtered) {
740 return bdrv_probe_geometry(filtered, geo);
741 }
742
743 return -ENOTSUP;
744 }
745
746 /*
747 * Create a uniquely-named empty temporary file.
748 * Return 0 upon success, otherwise a negative errno value.
749 */
750 int get_tmp_filename(char *filename, int size)
751 {
752 #ifdef _WIN32
753 char temp_dir[MAX_PATH];
754 /* GetTempFileName requires that its output buffer (4th param)
755 have length MAX_PATH or greater. */
756 assert(size >= MAX_PATH);
757 return (GetTempPath(MAX_PATH, temp_dir)
758 && GetTempFileName(temp_dir, "qem", 0, filename)
759 ? 0 : -GetLastError());
760 #else
761 int fd;
762 const char *tmpdir;
763 tmpdir = getenv("TMPDIR");
764 if (!tmpdir) {
765 tmpdir = "/var/tmp";
766 }
767 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
768 return -EOVERFLOW;
769 }
770 fd = mkstemp(filename);
771 if (fd < 0) {
772 return -errno;
773 }
774 if (close(fd) != 0) {
775 unlink(filename);
776 return -errno;
777 }
778 return 0;
779 #endif
780 }
781
782 /*
783 * Detect host devices. By convention, /dev/cdrom[N] is always
784 * recognized as a host CDROM.
785 */
786 static BlockDriver *find_hdev_driver(const char *filename)
787 {
788 int score_max = 0, score;
789 BlockDriver *drv = NULL, *d;
790
791 QLIST_FOREACH(d, &bdrv_drivers, list) {
792 if (d->bdrv_probe_device) {
793 score = d->bdrv_probe_device(filename);
794 if (score > score_max) {
795 score_max = score;
796 drv = d;
797 }
798 }
799 }
800
801 return drv;
802 }
803
804 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
805 {
806 BlockDriver *drv1;
807
808 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
809 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
810 return drv1;
811 }
812 }
813
814 return NULL;
815 }
816
817 BlockDriver *bdrv_find_protocol(const char *filename,
818 bool allow_protocol_prefix,
819 Error **errp)
820 {
821 BlockDriver *drv1;
822 char protocol[128];
823 int len;
824 const char *p;
825 int i;
826
827 /* TODO Drivers without bdrv_file_open must be specified explicitly */
828
829 /*
830 * XXX(hch): we really should not let host device detection
831 * override an explicit protocol specification, but moving this
832 * later breaks access to device names with colons in them.
833 * Thanks to the brain-dead persistent naming schemes on udev-
834 * based Linux systems those actually are quite common.
835 */
836 drv1 = find_hdev_driver(filename);
837 if (drv1) {
838 return drv1;
839 }
840
841 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
842 return &bdrv_file;
843 }
844
845 p = strchr(filename, ':');
846 assert(p != NULL);
847 len = p - filename;
848 if (len > sizeof(protocol) - 1)
849 len = sizeof(protocol) - 1;
850 memcpy(protocol, filename, len);
851 protocol[len] = '\0';
852
853 drv1 = bdrv_do_find_protocol(protocol);
854 if (drv1) {
855 return drv1;
856 }
857
858 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
859 if (block_driver_modules[i].protocol_name &&
860 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
861 block_module_load_one(block_driver_modules[i].library_name);
862 break;
863 }
864 }
865
866 drv1 = bdrv_do_find_protocol(protocol);
867 if (!drv1) {
868 error_setg(errp, "Unknown protocol '%s'", protocol);
869 }
870 return drv1;
871 }
872
873 /*
874 * Guess image format by probing its contents.
875 * This is not a good idea when your image is raw (CVE-2008-2004), but
876 * we do it anyway for backward compatibility.
877 *
878 * @buf contains the image's first @buf_size bytes.
879 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
880 * but can be smaller if the image file is smaller)
881 * @filename is its filename.
882 *
883 * For all block drivers, call the bdrv_probe() method to get its
884 * probing score.
885 * Return the first block driver with the highest probing score.
886 */
887 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
888 const char *filename)
889 {
890 int score_max = 0, score;
891 BlockDriver *drv = NULL, *d;
892
893 QLIST_FOREACH(d, &bdrv_drivers, list) {
894 if (d->bdrv_probe) {
895 score = d->bdrv_probe(buf, buf_size, filename);
896 if (score > score_max) {
897 score_max = score;
898 drv = d;
899 }
900 }
901 }
902
903 return drv;
904 }
905
906 static int find_image_format(BlockBackend *file, const char *filename,
907 BlockDriver **pdrv, Error **errp)
908 {
909 BlockDriver *drv;
910 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
911 int ret = 0;
912
913 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
914 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
915 *pdrv = &bdrv_raw;
916 return ret;
917 }
918
919 ret = blk_pread(file, 0, buf, sizeof(buf));
920 if (ret < 0) {
921 error_setg_errno(errp, -ret, "Could not read image for determining its "
922 "format");
923 *pdrv = NULL;
924 return ret;
925 }
926
927 drv = bdrv_probe_all(buf, ret, filename);
928 if (!drv) {
929 error_setg(errp, "Could not determine image format: No compatible "
930 "driver found");
931 ret = -ENOENT;
932 }
933 *pdrv = drv;
934 return ret;
935 }
936
937 /**
938 * Set the current 'total_sectors' value
939 * Return 0 on success, -errno on error.
940 */
941 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
942 {
943 BlockDriver *drv = bs->drv;
944
945 if (!drv) {
946 return -ENOMEDIUM;
947 }
948
949 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
950 if (bdrv_is_sg(bs))
951 return 0;
952
953 /* query actual device if possible, otherwise just trust the hint */
954 if (drv->bdrv_getlength) {
955 int64_t length = drv->bdrv_getlength(bs);
956 if (length < 0) {
957 return length;
958 }
959 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
960 }
961
962 bs->total_sectors = hint;
963 return 0;
964 }
965
966 /**
967 * Combines a QDict of new block driver @options with any missing options taken
968 * from @old_options, so that leaving out an option defaults to its old value.
969 */
970 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
971 QDict *old_options)
972 {
973 if (bs->drv && bs->drv->bdrv_join_options) {
974 bs->drv->bdrv_join_options(options, old_options);
975 } else {
976 qdict_join(options, old_options, false);
977 }
978 }
979
980 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
981 int open_flags,
982 Error **errp)
983 {
984 Error *local_err = NULL;
985 char *value = qemu_opt_get_del(opts, "detect-zeroes");
986 BlockdevDetectZeroesOptions detect_zeroes =
987 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
988 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
989 g_free(value);
990 if (local_err) {
991 error_propagate(errp, local_err);
992 return detect_zeroes;
993 }
994
995 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
996 !(open_flags & BDRV_O_UNMAP))
997 {
998 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
999 "without setting discard operation to unmap");
1000 }
1001
1002 return detect_zeroes;
1003 }
1004
1005 /**
1006 * Set open flags for aio engine
1007 *
1008 * Return 0 on success, -1 if the engine specified is invalid
1009 */
1010 int bdrv_parse_aio(const char *mode, int *flags)
1011 {
1012 if (!strcmp(mode, "threads")) {
1013 /* do nothing, default */
1014 } else if (!strcmp(mode, "native")) {
1015 *flags |= BDRV_O_NATIVE_AIO;
1016 #ifdef CONFIG_LINUX_IO_URING
1017 } else if (!strcmp(mode, "io_uring")) {
1018 *flags |= BDRV_O_IO_URING;
1019 #endif
1020 } else {
1021 return -1;
1022 }
1023
1024 return 0;
1025 }
1026
1027 /**
1028 * Set open flags for a given discard mode
1029 *
1030 * Return 0 on success, -1 if the discard mode was invalid.
1031 */
1032 int bdrv_parse_discard_flags(const char *mode, int *flags)
1033 {
1034 *flags &= ~BDRV_O_UNMAP;
1035
1036 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1037 /* do nothing */
1038 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1039 *flags |= BDRV_O_UNMAP;
1040 } else {
1041 return -1;
1042 }
1043
1044 return 0;
1045 }
1046
1047 /**
1048 * Set open flags for a given cache mode
1049 *
1050 * Return 0 on success, -1 if the cache mode was invalid.
1051 */
1052 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1053 {
1054 *flags &= ~BDRV_O_CACHE_MASK;
1055
1056 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1057 *writethrough = false;
1058 *flags |= BDRV_O_NOCACHE;
1059 } else if (!strcmp(mode, "directsync")) {
1060 *writethrough = true;
1061 *flags |= BDRV_O_NOCACHE;
1062 } else if (!strcmp(mode, "writeback")) {
1063 *writethrough = false;
1064 } else if (!strcmp(mode, "unsafe")) {
1065 *writethrough = false;
1066 *flags |= BDRV_O_NO_FLUSH;
1067 } else if (!strcmp(mode, "writethrough")) {
1068 *writethrough = true;
1069 } else {
1070 return -1;
1071 }
1072
1073 return 0;
1074 }
1075
1076 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1077 {
1078 BlockDriverState *parent = c->opaque;
1079 return g_strdup(bdrv_get_device_or_node_name(parent));
1080 }
1081
1082 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1083 {
1084 BlockDriverState *bs = child->opaque;
1085 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1086 }
1087
1088 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1089 {
1090 BlockDriverState *bs = child->opaque;
1091 return bdrv_drain_poll(bs, false, NULL, false);
1092 }
1093
1094 static void bdrv_child_cb_drained_end(BdrvChild *child,
1095 int *drained_end_counter)
1096 {
1097 BlockDriverState *bs = child->opaque;
1098 bdrv_drained_end_no_poll(bs, drained_end_counter);
1099 }
1100
1101 static int bdrv_child_cb_inactivate(BdrvChild *child)
1102 {
1103 BlockDriverState *bs = child->opaque;
1104 assert(bs->open_flags & BDRV_O_INACTIVE);
1105 return 0;
1106 }
1107
1108 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1109 GSList **ignore, Error **errp)
1110 {
1111 BlockDriverState *bs = child->opaque;
1112 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1113 }
1114
1115 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1116 GSList **ignore)
1117 {
1118 BlockDriverState *bs = child->opaque;
1119 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1120 }
1121
1122 /*
1123 * Returns the options and flags that a temporary snapshot should get, based on
1124 * the originally requested flags (the originally requested image will have
1125 * flags like a backing file)
1126 */
1127 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1128 int parent_flags, QDict *parent_options)
1129 {
1130 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1131
1132 /* For temporary files, unconditional cache=unsafe is fine */
1133 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1134 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1135
1136 /* Copy the read-only and discard options from the parent */
1137 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1138 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1139
1140 /* aio=native doesn't work for cache.direct=off, so disable it for the
1141 * temporary snapshot */
1142 *child_flags &= ~BDRV_O_NATIVE_AIO;
1143 }
1144
1145 static void bdrv_backing_attach(BdrvChild *c)
1146 {
1147 BlockDriverState *parent = c->opaque;
1148 BlockDriverState *backing_hd = c->bs;
1149
1150 assert(!parent->backing_blocker);
1151 error_setg(&parent->backing_blocker,
1152 "node is used as backing hd of '%s'",
1153 bdrv_get_device_or_node_name(parent));
1154
1155 bdrv_refresh_filename(backing_hd);
1156
1157 parent->open_flags &= ~BDRV_O_NO_BACKING;
1158
1159 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1160 /* Otherwise we won't be able to commit or stream */
1161 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1162 parent->backing_blocker);
1163 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1164 parent->backing_blocker);
1165 /*
1166 * We do backup in 3 ways:
1167 * 1. drive backup
1168 * The target bs is new opened, and the source is top BDS
1169 * 2. blockdev backup
1170 * Both the source and the target are top BDSes.
1171 * 3. internal backup(used for block replication)
1172 * Both the source and the target are backing file
1173 *
1174 * In case 1 and 2, neither the source nor the target is the backing file.
1175 * In case 3, we will block the top BDS, so there is only one block job
1176 * for the top BDS and its backing chain.
1177 */
1178 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1179 parent->backing_blocker);
1180 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1181 parent->backing_blocker);
1182 }
1183
1184 static void bdrv_backing_detach(BdrvChild *c)
1185 {
1186 BlockDriverState *parent = c->opaque;
1187
1188 assert(parent->backing_blocker);
1189 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1190 error_free(parent->backing_blocker);
1191 parent->backing_blocker = NULL;
1192 }
1193
1194 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1195 const char *filename, Error **errp)
1196 {
1197 BlockDriverState *parent = c->opaque;
1198 bool read_only = bdrv_is_read_only(parent);
1199 int ret;
1200
1201 if (read_only) {
1202 ret = bdrv_reopen_set_read_only(parent, false, errp);
1203 if (ret < 0) {
1204 return ret;
1205 }
1206 }
1207
1208 ret = bdrv_change_backing_file(parent, filename,
1209 base->drv ? base->drv->format_name : "",
1210 false);
1211 if (ret < 0) {
1212 error_setg_errno(errp, -ret, "Could not update backing file link");
1213 }
1214
1215 if (read_only) {
1216 bdrv_reopen_set_read_only(parent, true, NULL);
1217 }
1218
1219 return ret;
1220 }
1221
1222 /*
1223 * Returns the options and flags that a generic child of a BDS should
1224 * get, based on the given options and flags for the parent BDS.
1225 */
1226 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1227 int *child_flags, QDict *child_options,
1228 int parent_flags, QDict *parent_options)
1229 {
1230 int flags = parent_flags;
1231
1232 /*
1233 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1234 * Generally, the question to answer is: Should this child be
1235 * format-probed by default?
1236 */
1237
1238 /*
1239 * Pure and non-filtered data children of non-format nodes should
1240 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1241 * set). This only affects a very limited set of drivers (namely
1242 * quorum and blkverify when this comment was written).
1243 * Force-clear BDRV_O_PROTOCOL then.
1244 */
1245 if (!parent_is_format &&
1246 (role & BDRV_CHILD_DATA) &&
1247 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1248 {
1249 flags &= ~BDRV_O_PROTOCOL;
1250 }
1251
1252 /*
1253 * All children of format nodes (except for COW children) and all
1254 * metadata children in general should never be format-probed.
1255 * Force-set BDRV_O_PROTOCOL then.
1256 */
1257 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1258 (role & BDRV_CHILD_METADATA))
1259 {
1260 flags |= BDRV_O_PROTOCOL;
1261 }
1262
1263 /*
1264 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1265 * the parent.
1266 */
1267 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1268 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1269 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1270
1271 if (role & BDRV_CHILD_COW) {
1272 /* backing files are opened read-only by default */
1273 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1274 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1275 } else {
1276 /* Inherit the read-only option from the parent if it's not set */
1277 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1278 qdict_copy_default(child_options, parent_options,
1279 BDRV_OPT_AUTO_READ_ONLY);
1280 }
1281
1282 /*
1283 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1284 * can default to enable it on lower layers regardless of the
1285 * parent option.
1286 */
1287 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1288
1289 /* Clear flags that only apply to the top layer */
1290 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1291
1292 if (role & BDRV_CHILD_METADATA) {
1293 flags &= ~BDRV_O_NO_IO;
1294 }
1295 if (role & BDRV_CHILD_COW) {
1296 flags &= ~BDRV_O_TEMPORARY;
1297 }
1298
1299 *child_flags = flags;
1300 }
1301
1302 static void bdrv_child_cb_attach(BdrvChild *child)
1303 {
1304 BlockDriverState *bs = child->opaque;
1305
1306 if (child->role & BDRV_CHILD_COW) {
1307 bdrv_backing_attach(child);
1308 }
1309
1310 bdrv_apply_subtree_drain(child, bs);
1311 }
1312
1313 static void bdrv_child_cb_detach(BdrvChild *child)
1314 {
1315 BlockDriverState *bs = child->opaque;
1316
1317 if (child->role & BDRV_CHILD_COW) {
1318 bdrv_backing_detach(child);
1319 }
1320
1321 bdrv_unapply_subtree_drain(child, bs);
1322 }
1323
1324 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1325 const char *filename, Error **errp)
1326 {
1327 if (c->role & BDRV_CHILD_COW) {
1328 return bdrv_backing_update_filename(c, base, filename, errp);
1329 }
1330 return 0;
1331 }
1332
1333 const BdrvChildClass child_of_bds = {
1334 .parent_is_bds = true,
1335 .get_parent_desc = bdrv_child_get_parent_desc,
1336 .inherit_options = bdrv_inherited_options,
1337 .drained_begin = bdrv_child_cb_drained_begin,
1338 .drained_poll = bdrv_child_cb_drained_poll,
1339 .drained_end = bdrv_child_cb_drained_end,
1340 .attach = bdrv_child_cb_attach,
1341 .detach = bdrv_child_cb_detach,
1342 .inactivate = bdrv_child_cb_inactivate,
1343 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1344 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1345 .update_filename = bdrv_child_cb_update_filename,
1346 };
1347
1348 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1349 {
1350 int open_flags = flags;
1351
1352 /*
1353 * Clear flags that are internal to the block layer before opening the
1354 * image.
1355 */
1356 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1357
1358 return open_flags;
1359 }
1360
1361 static void update_flags_from_options(int *flags, QemuOpts *opts)
1362 {
1363 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1364
1365 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1366 *flags |= BDRV_O_NO_FLUSH;
1367 }
1368
1369 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1370 *flags |= BDRV_O_NOCACHE;
1371 }
1372
1373 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1374 *flags |= BDRV_O_RDWR;
1375 }
1376
1377 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1378 *flags |= BDRV_O_AUTO_RDONLY;
1379 }
1380 }
1381
1382 static void update_options_from_flags(QDict *options, int flags)
1383 {
1384 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1385 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1386 }
1387 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1388 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1389 flags & BDRV_O_NO_FLUSH);
1390 }
1391 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1392 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1393 }
1394 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1395 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1396 flags & BDRV_O_AUTO_RDONLY);
1397 }
1398 }
1399
1400 static void bdrv_assign_node_name(BlockDriverState *bs,
1401 const char *node_name,
1402 Error **errp)
1403 {
1404 char *gen_node_name = NULL;
1405
1406 if (!node_name) {
1407 node_name = gen_node_name = id_generate(ID_BLOCK);
1408 } else if (!id_wellformed(node_name)) {
1409 /*
1410 * Check for empty string or invalid characters, but not if it is
1411 * generated (generated names use characters not available to the user)
1412 */
1413 error_setg(errp, "Invalid node name");
1414 return;
1415 }
1416
1417 /* takes care of avoiding namespaces collisions */
1418 if (blk_by_name(node_name)) {
1419 error_setg(errp, "node-name=%s is conflicting with a device id",
1420 node_name);
1421 goto out;
1422 }
1423
1424 /* takes care of avoiding duplicates node names */
1425 if (bdrv_find_node(node_name)) {
1426 error_setg(errp, "Duplicate node name");
1427 goto out;
1428 }
1429
1430 /* Make sure that the node name isn't truncated */
1431 if (strlen(node_name) >= sizeof(bs->node_name)) {
1432 error_setg(errp, "Node name too long");
1433 goto out;
1434 }
1435
1436 /* copy node name into the bs and insert it into the graph list */
1437 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1438 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1439 out:
1440 g_free(gen_node_name);
1441 }
1442
1443 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1444 const char *node_name, QDict *options,
1445 int open_flags, Error **errp)
1446 {
1447 Error *local_err = NULL;
1448 int i, ret;
1449
1450 bdrv_assign_node_name(bs, node_name, &local_err);
1451 if (local_err) {
1452 error_propagate(errp, local_err);
1453 return -EINVAL;
1454 }
1455
1456 bs->drv = drv;
1457 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1458 bs->opaque = g_malloc0(drv->instance_size);
1459
1460 if (drv->bdrv_file_open) {
1461 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1462 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1463 } else if (drv->bdrv_open) {
1464 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1465 } else {
1466 ret = 0;
1467 }
1468
1469 if (ret < 0) {
1470 if (local_err) {
1471 error_propagate(errp, local_err);
1472 } else if (bs->filename[0]) {
1473 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1474 } else {
1475 error_setg_errno(errp, -ret, "Could not open image");
1476 }
1477 goto open_failed;
1478 }
1479
1480 ret = refresh_total_sectors(bs, bs->total_sectors);
1481 if (ret < 0) {
1482 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1483 return ret;
1484 }
1485
1486 bdrv_refresh_limits(bs, &local_err);
1487 if (local_err) {
1488 error_propagate(errp, local_err);
1489 return -EINVAL;
1490 }
1491
1492 assert(bdrv_opt_mem_align(bs) != 0);
1493 assert(bdrv_min_mem_align(bs) != 0);
1494 assert(is_power_of_2(bs->bl.request_alignment));
1495
1496 for (i = 0; i < bs->quiesce_counter; i++) {
1497 if (drv->bdrv_co_drain_begin) {
1498 drv->bdrv_co_drain_begin(bs);
1499 }
1500 }
1501
1502 return 0;
1503 open_failed:
1504 bs->drv = NULL;
1505 if (bs->file != NULL) {
1506 bdrv_unref_child(bs, bs->file);
1507 bs->file = NULL;
1508 }
1509 g_free(bs->opaque);
1510 bs->opaque = NULL;
1511 return ret;
1512 }
1513
1514 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1515 int flags, Error **errp)
1516 {
1517 BlockDriverState *bs;
1518 int ret;
1519
1520 bs = bdrv_new();
1521 bs->open_flags = flags;
1522 bs->explicit_options = qdict_new();
1523 bs->options = qdict_new();
1524 bs->opaque = NULL;
1525
1526 update_options_from_flags(bs->options, flags);
1527
1528 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1529 if (ret < 0) {
1530 qobject_unref(bs->explicit_options);
1531 bs->explicit_options = NULL;
1532 qobject_unref(bs->options);
1533 bs->options = NULL;
1534 bdrv_unref(bs);
1535 return NULL;
1536 }
1537
1538 return bs;
1539 }
1540
1541 QemuOptsList bdrv_runtime_opts = {
1542 .name = "bdrv_common",
1543 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1544 .desc = {
1545 {
1546 .name = "node-name",
1547 .type = QEMU_OPT_STRING,
1548 .help = "Node name of the block device node",
1549 },
1550 {
1551 .name = "driver",
1552 .type = QEMU_OPT_STRING,
1553 .help = "Block driver to use for the node",
1554 },
1555 {
1556 .name = BDRV_OPT_CACHE_DIRECT,
1557 .type = QEMU_OPT_BOOL,
1558 .help = "Bypass software writeback cache on the host",
1559 },
1560 {
1561 .name = BDRV_OPT_CACHE_NO_FLUSH,
1562 .type = QEMU_OPT_BOOL,
1563 .help = "Ignore flush requests",
1564 },
1565 {
1566 .name = BDRV_OPT_READ_ONLY,
1567 .type = QEMU_OPT_BOOL,
1568 .help = "Node is opened in read-only mode",
1569 },
1570 {
1571 .name = BDRV_OPT_AUTO_READ_ONLY,
1572 .type = QEMU_OPT_BOOL,
1573 .help = "Node can become read-only if opening read-write fails",
1574 },
1575 {
1576 .name = "detect-zeroes",
1577 .type = QEMU_OPT_STRING,
1578 .help = "try to optimize zero writes (off, on, unmap)",
1579 },
1580 {
1581 .name = BDRV_OPT_DISCARD,
1582 .type = QEMU_OPT_STRING,
1583 .help = "discard operation (ignore/off, unmap/on)",
1584 },
1585 {
1586 .name = BDRV_OPT_FORCE_SHARE,
1587 .type = QEMU_OPT_BOOL,
1588 .help = "always accept other writers (default: off)",
1589 },
1590 { /* end of list */ }
1591 },
1592 };
1593
1594 QemuOptsList bdrv_create_opts_simple = {
1595 .name = "simple-create-opts",
1596 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1597 .desc = {
1598 {
1599 .name = BLOCK_OPT_SIZE,
1600 .type = QEMU_OPT_SIZE,
1601 .help = "Virtual disk size"
1602 },
1603 {
1604 .name = BLOCK_OPT_PREALLOC,
1605 .type = QEMU_OPT_STRING,
1606 .help = "Preallocation mode (allowed values: off)"
1607 },
1608 { /* end of list */ }
1609 }
1610 };
1611
1612 /*
1613 * Common part for opening disk images and files
1614 *
1615 * Removes all processed options from *options.
1616 */
1617 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1618 QDict *options, Error **errp)
1619 {
1620 int ret, open_flags;
1621 const char *filename;
1622 const char *driver_name = NULL;
1623 const char *node_name = NULL;
1624 const char *discard;
1625 QemuOpts *opts;
1626 BlockDriver *drv;
1627 Error *local_err = NULL;
1628
1629 assert(bs->file == NULL);
1630 assert(options != NULL && bs->options != options);
1631
1632 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1633 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1634 ret = -EINVAL;
1635 goto fail_opts;
1636 }
1637
1638 update_flags_from_options(&bs->open_flags, opts);
1639
1640 driver_name = qemu_opt_get(opts, "driver");
1641 drv = bdrv_find_format(driver_name);
1642 assert(drv != NULL);
1643
1644 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1645
1646 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1647 error_setg(errp,
1648 BDRV_OPT_FORCE_SHARE
1649 "=on can only be used with read-only images");
1650 ret = -EINVAL;
1651 goto fail_opts;
1652 }
1653
1654 if (file != NULL) {
1655 bdrv_refresh_filename(blk_bs(file));
1656 filename = blk_bs(file)->filename;
1657 } else {
1658 /*
1659 * Caution: while qdict_get_try_str() is fine, getting
1660 * non-string types would require more care. When @options
1661 * come from -blockdev or blockdev_add, its members are typed
1662 * according to the QAPI schema, but when they come from
1663 * -drive, they're all QString.
1664 */
1665 filename = qdict_get_try_str(options, "filename");
1666 }
1667
1668 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1669 error_setg(errp, "The '%s' block driver requires a file name",
1670 drv->format_name);
1671 ret = -EINVAL;
1672 goto fail_opts;
1673 }
1674
1675 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1676 drv->format_name);
1677
1678 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1679
1680 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1681 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1682 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1683 } else {
1684 ret = -ENOTSUP;
1685 }
1686 if (ret < 0) {
1687 error_setg(errp,
1688 !bs->read_only && bdrv_is_whitelisted(drv, true)
1689 ? "Driver '%s' can only be used for read-only devices"
1690 : "Driver '%s' is not whitelisted",
1691 drv->format_name);
1692 goto fail_opts;
1693 }
1694 }
1695
1696 /* bdrv_new() and bdrv_close() make it so */
1697 assert(qatomic_read(&bs->copy_on_read) == 0);
1698
1699 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1700 if (!bs->read_only) {
1701 bdrv_enable_copy_on_read(bs);
1702 } else {
1703 error_setg(errp, "Can't use copy-on-read on read-only device");
1704 ret = -EINVAL;
1705 goto fail_opts;
1706 }
1707 }
1708
1709 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1710 if (discard != NULL) {
1711 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1712 error_setg(errp, "Invalid discard option");
1713 ret = -EINVAL;
1714 goto fail_opts;
1715 }
1716 }
1717
1718 bs->detect_zeroes =
1719 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1720 if (local_err) {
1721 error_propagate(errp, local_err);
1722 ret = -EINVAL;
1723 goto fail_opts;
1724 }
1725
1726 if (filename != NULL) {
1727 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1728 } else {
1729 bs->filename[0] = '\0';
1730 }
1731 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1732
1733 /* Open the image, either directly or using a protocol */
1734 open_flags = bdrv_open_flags(bs, bs->open_flags);
1735 node_name = qemu_opt_get(opts, "node-name");
1736
1737 assert(!drv->bdrv_file_open || file == NULL);
1738 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1739 if (ret < 0) {
1740 goto fail_opts;
1741 }
1742
1743 qemu_opts_del(opts);
1744 return 0;
1745
1746 fail_opts:
1747 qemu_opts_del(opts);
1748 return ret;
1749 }
1750
1751 static QDict *parse_json_filename(const char *filename, Error **errp)
1752 {
1753 QObject *options_obj;
1754 QDict *options;
1755 int ret;
1756
1757 ret = strstart(filename, "json:", &filename);
1758 assert(ret);
1759
1760 options_obj = qobject_from_json(filename, errp);
1761 if (!options_obj) {
1762 error_prepend(errp, "Could not parse the JSON options: ");
1763 return NULL;
1764 }
1765
1766 options = qobject_to(QDict, options_obj);
1767 if (!options) {
1768 qobject_unref(options_obj);
1769 error_setg(errp, "Invalid JSON object given");
1770 return NULL;
1771 }
1772
1773 qdict_flatten(options);
1774
1775 return options;
1776 }
1777
1778 static void parse_json_protocol(QDict *options, const char **pfilename,
1779 Error **errp)
1780 {
1781 QDict *json_options;
1782 Error *local_err = NULL;
1783
1784 /* Parse json: pseudo-protocol */
1785 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1786 return;
1787 }
1788
1789 json_options = parse_json_filename(*pfilename, &local_err);
1790 if (local_err) {
1791 error_propagate(errp, local_err);
1792 return;
1793 }
1794
1795 /* Options given in the filename have lower priority than options
1796 * specified directly */
1797 qdict_join(options, json_options, false);
1798 qobject_unref(json_options);
1799 *pfilename = NULL;
1800 }
1801
1802 /*
1803 * Fills in default options for opening images and converts the legacy
1804 * filename/flags pair to option QDict entries.
1805 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1806 * block driver has been specified explicitly.
1807 */
1808 static int bdrv_fill_options(QDict **options, const char *filename,
1809 int *flags, Error **errp)
1810 {
1811 const char *drvname;
1812 bool protocol = *flags & BDRV_O_PROTOCOL;
1813 bool parse_filename = false;
1814 BlockDriver *drv = NULL;
1815 Error *local_err = NULL;
1816
1817 /*
1818 * Caution: while qdict_get_try_str() is fine, getting non-string
1819 * types would require more care. When @options come from
1820 * -blockdev or blockdev_add, its members are typed according to
1821 * the QAPI schema, but when they come from -drive, they're all
1822 * QString.
1823 */
1824 drvname = qdict_get_try_str(*options, "driver");
1825 if (drvname) {
1826 drv = bdrv_find_format(drvname);
1827 if (!drv) {
1828 error_setg(errp, "Unknown driver '%s'", drvname);
1829 return -ENOENT;
1830 }
1831 /* If the user has explicitly specified the driver, this choice should
1832 * override the BDRV_O_PROTOCOL flag */
1833 protocol = drv->bdrv_file_open;
1834 }
1835
1836 if (protocol) {
1837 *flags |= BDRV_O_PROTOCOL;
1838 } else {
1839 *flags &= ~BDRV_O_PROTOCOL;
1840 }
1841
1842 /* Translate cache options from flags into options */
1843 update_options_from_flags(*options, *flags);
1844
1845 /* Fetch the file name from the options QDict if necessary */
1846 if (protocol && filename) {
1847 if (!qdict_haskey(*options, "filename")) {
1848 qdict_put_str(*options, "filename", filename);
1849 parse_filename = true;
1850 } else {
1851 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1852 "the same time");
1853 return -EINVAL;
1854 }
1855 }
1856
1857 /* Find the right block driver */
1858 /* See cautionary note on accessing @options above */
1859 filename = qdict_get_try_str(*options, "filename");
1860
1861 if (!drvname && protocol) {
1862 if (filename) {
1863 drv = bdrv_find_protocol(filename, parse_filename, errp);
1864 if (!drv) {
1865 return -EINVAL;
1866 }
1867
1868 drvname = drv->format_name;
1869 qdict_put_str(*options, "driver", drvname);
1870 } else {
1871 error_setg(errp, "Must specify either driver or file");
1872 return -EINVAL;
1873 }
1874 }
1875
1876 assert(drv || !protocol);
1877
1878 /* Driver-specific filename parsing */
1879 if (drv && drv->bdrv_parse_filename && parse_filename) {
1880 drv->bdrv_parse_filename(filename, *options, &local_err);
1881 if (local_err) {
1882 error_propagate(errp, local_err);
1883 return -EINVAL;
1884 }
1885
1886 if (!drv->bdrv_needs_filename) {
1887 qdict_del(*options, "filename");
1888 }
1889 }
1890
1891 return 0;
1892 }
1893
1894 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1895 uint64_t perm, uint64_t shared,
1896 GSList *ignore_children,
1897 bool *tighten_restrictions, Error **errp);
1898 static void bdrv_child_abort_perm_update(BdrvChild *c);
1899 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1900
1901 typedef struct BlockReopenQueueEntry {
1902 bool prepared;
1903 bool perms_checked;
1904 BDRVReopenState state;
1905 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
1906 } BlockReopenQueueEntry;
1907
1908 /*
1909 * Return the flags that @bs will have after the reopens in @q have
1910 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1911 * return the current flags.
1912 */
1913 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1914 {
1915 BlockReopenQueueEntry *entry;
1916
1917 if (q != NULL) {
1918 QTAILQ_FOREACH(entry, q, entry) {
1919 if (entry->state.bs == bs) {
1920 return entry->state.flags;
1921 }
1922 }
1923 }
1924
1925 return bs->open_flags;
1926 }
1927
1928 /* Returns whether the image file can be written to after the reopen queue @q
1929 * has been successfully applied, or right now if @q is NULL. */
1930 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1931 BlockReopenQueue *q)
1932 {
1933 int flags = bdrv_reopen_get_flags(q, bs);
1934
1935 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1936 }
1937
1938 /*
1939 * Return whether the BDS can be written to. This is not necessarily
1940 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1941 * be written to but do not count as read-only images.
1942 */
1943 bool bdrv_is_writable(BlockDriverState *bs)
1944 {
1945 return bdrv_is_writable_after_reopen(bs, NULL);
1946 }
1947
1948 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1949 BdrvChild *c, BdrvChildRole role,
1950 BlockReopenQueue *reopen_queue,
1951 uint64_t parent_perm, uint64_t parent_shared,
1952 uint64_t *nperm, uint64_t *nshared)
1953 {
1954 assert(bs->drv && bs->drv->bdrv_child_perm);
1955 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1956 parent_perm, parent_shared,
1957 nperm, nshared);
1958 /* TODO Take force_share from reopen_queue */
1959 if (child_bs && child_bs->force_share) {
1960 *nshared = BLK_PERM_ALL;
1961 }
1962 }
1963
1964 /*
1965 * Check whether permissions on this node can be changed in a way that
1966 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1967 * permissions of all its parents. This involves checking whether all necessary
1968 * permission changes to child nodes can be performed.
1969 *
1970 * Will set *tighten_restrictions to true if and only if new permissions have to
1971 * be taken or currently shared permissions are to be unshared. Otherwise,
1972 * errors are not fatal as long as the caller accepts that the restrictions
1973 * remain tighter than they need to be. The caller still has to abort the
1974 * transaction.
1975 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1976 * encounter fatal errors even though no restrictions are to be tightened. For
1977 * example, changing a node from RW to RO will fail if the WRITE permission is
1978 * to be kept.
1979 *
1980 * A call to this function must always be followed by a call to bdrv_set_perm()
1981 * or bdrv_abort_perm_update().
1982 */
1983 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1984 uint64_t cumulative_perms,
1985 uint64_t cumulative_shared_perms,
1986 GSList *ignore_children,
1987 bool *tighten_restrictions, Error **errp)
1988 {
1989 BlockDriver *drv = bs->drv;
1990 BdrvChild *c;
1991 int ret;
1992
1993 assert(!q || !tighten_restrictions);
1994
1995 if (tighten_restrictions) {
1996 uint64_t current_perms, current_shared;
1997 uint64_t added_perms, removed_shared_perms;
1998
1999 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
2000
2001 added_perms = cumulative_perms & ~current_perms;
2002 removed_shared_perms = current_shared & ~cumulative_shared_perms;
2003
2004 *tighten_restrictions = added_perms || removed_shared_perms;
2005 }
2006
2007 /* Write permissions never work with read-only images */
2008 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2009 !bdrv_is_writable_after_reopen(bs, q))
2010 {
2011 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2012 error_setg(errp, "Block node is read-only");
2013 } else {
2014 uint64_t current_perms, current_shared;
2015 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
2016 if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
2017 error_setg(errp, "Cannot make block node read-only, there is "
2018 "a writer on it");
2019 } else {
2020 error_setg(errp, "Cannot make block node read-only and create "
2021 "a writer on it");
2022 }
2023 }
2024
2025 return -EPERM;
2026 }
2027
2028 /*
2029 * Unaligned requests will automatically be aligned to bl.request_alignment
2030 * and without RESIZE we can't extend requests to write to space beyond the
2031 * end of the image, so it's required that the image size is aligned.
2032 */
2033 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2034 !(cumulative_perms & BLK_PERM_RESIZE))
2035 {
2036 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2037 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2038 "Image size is not a multiple of request "
2039 "alignment");
2040 return -EPERM;
2041 }
2042 }
2043
2044 /* Check this node */
2045 if (!drv) {
2046 return 0;
2047 }
2048
2049 if (drv->bdrv_check_perm) {
2050 return drv->bdrv_check_perm(bs, cumulative_perms,
2051 cumulative_shared_perms, errp);
2052 }
2053
2054 /* Drivers that never have children can omit .bdrv_child_perm() */
2055 if (!drv->bdrv_child_perm) {
2056 assert(QLIST_EMPTY(&bs->children));
2057 return 0;
2058 }
2059
2060 /* Check all children */
2061 QLIST_FOREACH(c, &bs->children, next) {
2062 uint64_t cur_perm, cur_shared;
2063 bool child_tighten_restr;
2064
2065 bdrv_child_perm(bs, c->bs, c, c->role, q,
2066 cumulative_perms, cumulative_shared_perms,
2067 &cur_perm, &cur_shared);
2068 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children,
2069 tighten_restrictions ? &child_tighten_restr
2070 : NULL,
2071 errp);
2072 if (tighten_restrictions) {
2073 *tighten_restrictions |= child_tighten_restr;
2074 }
2075 if (ret < 0) {
2076 return ret;
2077 }
2078 }
2079
2080 return 0;
2081 }
2082
2083 /*
2084 * Notifies drivers that after a previous bdrv_check_perm() call, the
2085 * permission update is not performed and any preparations made for it (e.g.
2086 * taken file locks) need to be undone.
2087 *
2088 * This function recursively notifies all child nodes.
2089 */
2090 static void bdrv_abort_perm_update(BlockDriverState *bs)
2091 {
2092 BlockDriver *drv = bs->drv;
2093 BdrvChild *c;
2094
2095 if (!drv) {
2096 return;
2097 }
2098
2099 if (drv->bdrv_abort_perm_update) {
2100 drv->bdrv_abort_perm_update(bs);
2101 }
2102
2103 QLIST_FOREACH(c, &bs->children, next) {
2104 bdrv_child_abort_perm_update(c);
2105 }
2106 }
2107
2108 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
2109 uint64_t cumulative_shared_perms)
2110 {
2111 BlockDriver *drv = bs->drv;
2112 BdrvChild *c;
2113
2114 if (!drv) {
2115 return;
2116 }
2117
2118 /* Update this node */
2119 if (drv->bdrv_set_perm) {
2120 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2121 }
2122
2123 /* Drivers that never have children can omit .bdrv_child_perm() */
2124 if (!drv->bdrv_child_perm) {
2125 assert(QLIST_EMPTY(&bs->children));
2126 return;
2127 }
2128
2129 /* Update all children */
2130 QLIST_FOREACH(c, &bs->children, next) {
2131 uint64_t cur_perm, cur_shared;
2132 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2133 cumulative_perms, cumulative_shared_perms,
2134 &cur_perm, &cur_shared);
2135 bdrv_child_set_perm(c, cur_perm, cur_shared);
2136 }
2137 }
2138
2139 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2140 uint64_t *shared_perm)
2141 {
2142 BdrvChild *c;
2143 uint64_t cumulative_perms = 0;
2144 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2145
2146 QLIST_FOREACH(c, &bs->parents, next_parent) {
2147 cumulative_perms |= c->perm;
2148 cumulative_shared_perms &= c->shared_perm;
2149 }
2150
2151 *perm = cumulative_perms;
2152 *shared_perm = cumulative_shared_perms;
2153 }
2154
2155 static char *bdrv_child_user_desc(BdrvChild *c)
2156 {
2157 if (c->klass->get_parent_desc) {
2158 return c->klass->get_parent_desc(c);
2159 }
2160
2161 return g_strdup("another user");
2162 }
2163
2164 char *bdrv_perm_names(uint64_t perm)
2165 {
2166 struct perm_name {
2167 uint64_t perm;
2168 const char *name;
2169 } permissions[] = {
2170 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2171 { BLK_PERM_WRITE, "write" },
2172 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2173 { BLK_PERM_RESIZE, "resize" },
2174 { BLK_PERM_GRAPH_MOD, "change children" },
2175 { 0, NULL }
2176 };
2177
2178 GString *result = g_string_sized_new(30);
2179 struct perm_name *p;
2180
2181 for (p = permissions; p->name; p++) {
2182 if (perm & p->perm) {
2183 if (result->len > 0) {
2184 g_string_append(result, ", ");
2185 }
2186 g_string_append(result, p->name);
2187 }
2188 }
2189
2190 return g_string_free(result, FALSE);
2191 }
2192
2193 /*
2194 * Checks whether a new reference to @bs can be added if the new user requires
2195 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2196 * set, the BdrvChild objects in this list are ignored in the calculations;
2197 * this allows checking permission updates for an existing reference.
2198 *
2199 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2200 *
2201 * Needs to be followed by a call to either bdrv_set_perm() or
2202 * bdrv_abort_perm_update(). */
2203 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
2204 uint64_t new_used_perm,
2205 uint64_t new_shared_perm,
2206 GSList *ignore_children,
2207 bool *tighten_restrictions,
2208 Error **errp)
2209 {
2210 BdrvChild *c;
2211 uint64_t cumulative_perms = new_used_perm;
2212 uint64_t cumulative_shared_perms = new_shared_perm;
2213
2214 assert(!q || !tighten_restrictions);
2215
2216 /* There is no reason why anyone couldn't tolerate write_unchanged */
2217 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
2218
2219 QLIST_FOREACH(c, &bs->parents, next_parent) {
2220 if (g_slist_find(ignore_children, c)) {
2221 continue;
2222 }
2223
2224 if ((new_used_perm & c->shared_perm) != new_used_perm) {
2225 char *user = bdrv_child_user_desc(c);
2226 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
2227
2228 if (tighten_restrictions) {
2229 *tighten_restrictions = true;
2230 }
2231
2232 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
2233 "allow '%s' on %s",
2234 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2235 g_free(user);
2236 g_free(perm_names);
2237 return -EPERM;
2238 }
2239
2240 if ((c->perm & new_shared_perm) != c->perm) {
2241 char *user = bdrv_child_user_desc(c);
2242 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
2243
2244 if (tighten_restrictions) {
2245 *tighten_restrictions = true;
2246 }
2247
2248 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
2249 "'%s' on %s",
2250 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2251 g_free(user);
2252 g_free(perm_names);
2253 return -EPERM;
2254 }
2255
2256 cumulative_perms |= c->perm;
2257 cumulative_shared_perms &= c->shared_perm;
2258 }
2259
2260 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
2261 ignore_children, tighten_restrictions, errp);
2262 }
2263
2264 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2265 * bdrv_child_abort_perm_update(). */
2266 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
2267 uint64_t perm, uint64_t shared,
2268 GSList *ignore_children,
2269 bool *tighten_restrictions, Error **errp)
2270 {
2271 int ret;
2272
2273 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
2274 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children,
2275 tighten_restrictions, errp);
2276 g_slist_free(ignore_children);
2277
2278 if (ret < 0) {
2279 return ret;
2280 }
2281
2282 if (!c->has_backup_perm) {
2283 c->has_backup_perm = true;
2284 c->backup_perm = c->perm;
2285 c->backup_shared_perm = c->shared_perm;
2286 }
2287 /*
2288 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2289 * same child twice during check_perm procedure
2290 */
2291
2292 c->perm = perm;
2293 c->shared_perm = shared;
2294
2295 return 0;
2296 }
2297
2298 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
2299 {
2300 uint64_t cumulative_perms, cumulative_shared_perms;
2301
2302 c->has_backup_perm = false;
2303
2304 c->perm = perm;
2305 c->shared_perm = shared;
2306
2307 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
2308 &cumulative_shared_perms);
2309 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
2310 }
2311
2312 static void bdrv_child_abort_perm_update(BdrvChild *c)
2313 {
2314 if (c->has_backup_perm) {
2315 c->perm = c->backup_perm;
2316 c->shared_perm = c->backup_shared_perm;
2317 c->has_backup_perm = false;
2318 }
2319
2320 bdrv_abort_perm_update(c->bs);
2321 }
2322
2323 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2324 Error **errp)
2325 {
2326 Error *local_err = NULL;
2327 int ret;
2328 bool tighten_restrictions;
2329
2330 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL,
2331 &tighten_restrictions, &local_err);
2332 if (ret < 0) {
2333 bdrv_child_abort_perm_update(c);
2334 if (tighten_restrictions) {
2335 error_propagate(errp, local_err);
2336 } else {
2337 /*
2338 * Our caller may intend to only loosen restrictions and
2339 * does not expect this function to fail. Errors are not
2340 * fatal in such a case, so we can just hide them from our
2341 * caller.
2342 */
2343 error_free(local_err);
2344 ret = 0;
2345 }
2346 return ret;
2347 }
2348
2349 bdrv_child_set_perm(c, perm, shared);
2350
2351 return 0;
2352 }
2353
2354 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2355 {
2356 uint64_t parent_perms, parent_shared;
2357 uint64_t perms, shared;
2358
2359 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2360 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2361 parent_perms, parent_shared, &perms, &shared);
2362
2363 return bdrv_child_try_set_perm(c, perms, shared, errp);
2364 }
2365
2366 /*
2367 * Default implementation for .bdrv_child_perm() for block filters:
2368 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2369 * filtered child.
2370 */
2371 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2372 BdrvChildRole role,
2373 BlockReopenQueue *reopen_queue,
2374 uint64_t perm, uint64_t shared,
2375 uint64_t *nperm, uint64_t *nshared)
2376 {
2377 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2378 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2379 }
2380
2381 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2382 BdrvChildRole role,
2383 BlockReopenQueue *reopen_queue,
2384 uint64_t perm, uint64_t shared,
2385 uint64_t *nperm, uint64_t *nshared)
2386 {
2387 assert(role & BDRV_CHILD_COW);
2388
2389 /*
2390 * We want consistent read from backing files if the parent needs it.
2391 * No other operations are performed on backing files.
2392 */
2393 perm &= BLK_PERM_CONSISTENT_READ;
2394
2395 /*
2396 * If the parent can deal with changing data, we're okay with a
2397 * writable and resizable backing file.
2398 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2399 */
2400 if (shared & BLK_PERM_WRITE) {
2401 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2402 } else {
2403 shared = 0;
2404 }
2405
2406 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2407 BLK_PERM_WRITE_UNCHANGED;
2408
2409 if (bs->open_flags & BDRV_O_INACTIVE) {
2410 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2411 }
2412
2413 *nperm = perm;
2414 *nshared = shared;
2415 }
2416
2417 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2418 BdrvChildRole role,
2419 BlockReopenQueue *reopen_queue,
2420 uint64_t perm, uint64_t shared,
2421 uint64_t *nperm, uint64_t *nshared)
2422 {
2423 int flags;
2424
2425 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2426
2427 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2428
2429 /*
2430 * Apart from the modifications below, the same permissions are
2431 * forwarded and left alone as for filters
2432 */
2433 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2434 perm, shared, &perm, &shared);
2435
2436 if (role & BDRV_CHILD_METADATA) {
2437 /* Format drivers may touch metadata even if the guest doesn't write */
2438 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2439 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2440 }
2441
2442 /*
2443 * bs->file always needs to be consistent because of the
2444 * metadata. We can never allow other users to resize or write
2445 * to it.
2446 */
2447 if (!(flags & BDRV_O_NO_IO)) {
2448 perm |= BLK_PERM_CONSISTENT_READ;
2449 }
2450 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2451 }
2452
2453 if (role & BDRV_CHILD_DATA) {
2454 /*
2455 * Technically, everything in this block is a subset of the
2456 * BDRV_CHILD_METADATA path taken above, and so this could
2457 * be an "else if" branch. However, that is not obvious, and
2458 * this function is not performance critical, therefore we let
2459 * this be an independent "if".
2460 */
2461
2462 /*
2463 * We cannot allow other users to resize the file because the
2464 * format driver might have some assumptions about the size
2465 * (e.g. because it is stored in metadata, or because the file
2466 * is split into fixed-size data files).
2467 */
2468 shared &= ~BLK_PERM_RESIZE;
2469
2470 /*
2471 * WRITE_UNCHANGED often cannot be performed as such on the
2472 * data file. For example, the qcow2 driver may still need to
2473 * write copied clusters on copy-on-read.
2474 */
2475 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2476 perm |= BLK_PERM_WRITE;
2477 }
2478
2479 /*
2480 * If the data file is written to, the format driver may
2481 * expect to be able to resize it by writing beyond the EOF.
2482 */
2483 if (perm & BLK_PERM_WRITE) {
2484 perm |= BLK_PERM_RESIZE;
2485 }
2486 }
2487
2488 if (bs->open_flags & BDRV_O_INACTIVE) {
2489 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2490 }
2491
2492 *nperm = perm;
2493 *nshared = shared;
2494 }
2495
2496 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2497 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2498 uint64_t perm, uint64_t shared,
2499 uint64_t *nperm, uint64_t *nshared)
2500 {
2501 if (role & BDRV_CHILD_FILTERED) {
2502 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2503 BDRV_CHILD_COW)));
2504 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2505 perm, shared, nperm, nshared);
2506 } else if (role & BDRV_CHILD_COW) {
2507 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2508 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2509 perm, shared, nperm, nshared);
2510 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2511 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2512 perm, shared, nperm, nshared);
2513 } else {
2514 g_assert_not_reached();
2515 }
2516 }
2517
2518 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2519 {
2520 static const uint64_t permissions[] = {
2521 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2522 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2523 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2524 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2525 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2526 };
2527
2528 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2529 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2530
2531 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2532
2533 return permissions[qapi_perm];
2534 }
2535
2536 static void bdrv_replace_child_noperm(BdrvChild *child,
2537 BlockDriverState *new_bs)
2538 {
2539 BlockDriverState *old_bs = child->bs;
2540 int new_bs_quiesce_counter;
2541 int drain_saldo;
2542
2543 assert(!child->frozen);
2544
2545 if (old_bs && new_bs) {
2546 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2547 }
2548
2549 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2550 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2551
2552 /*
2553 * If the new child node is drained but the old one was not, flush
2554 * all outstanding requests to the old child node.
2555 */
2556 while (drain_saldo > 0 && child->klass->drained_begin) {
2557 bdrv_parent_drained_begin_single(child, true);
2558 drain_saldo--;
2559 }
2560
2561 if (old_bs) {
2562 /* Detach first so that the recursive drain sections coming from @child
2563 * are already gone and we only end the drain sections that came from
2564 * elsewhere. */
2565 if (child->klass->detach) {
2566 child->klass->detach(child);
2567 }
2568 QLIST_REMOVE(child, next_parent);
2569 }
2570
2571 child->bs = new_bs;
2572
2573 if (new_bs) {
2574 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2575
2576 /*
2577 * Detaching the old node may have led to the new node's
2578 * quiesce_counter having been decreased. Not a problem, we
2579 * just need to recognize this here and then invoke
2580 * drained_end appropriately more often.
2581 */
2582 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2583 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2584
2585 /* Attach only after starting new drained sections, so that recursive
2586 * drain sections coming from @child don't get an extra .drained_begin
2587 * callback. */
2588 if (child->klass->attach) {
2589 child->klass->attach(child);
2590 }
2591 }
2592
2593 /*
2594 * If the old child node was drained but the new one is not, allow
2595 * requests to come in only after the new node has been attached.
2596 */
2597 while (drain_saldo < 0 && child->klass->drained_end) {
2598 bdrv_parent_drained_end_single(child);
2599 drain_saldo++;
2600 }
2601 }
2602
2603 /*
2604 * Updates @child to change its reference to point to @new_bs, including
2605 * checking and applying the necessary permission updates both to the old node
2606 * and to @new_bs.
2607 *
2608 * NULL is passed as @new_bs for removing the reference before freeing @child.
2609 *
2610 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2611 * function uses bdrv_set_perm() to update the permissions according to the new
2612 * reference that @new_bs gets.
2613 *
2614 * Callers must ensure that child->frozen is false.
2615 */
2616 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2617 {
2618 BlockDriverState *old_bs = child->bs;
2619 uint64_t perm, shared_perm;
2620
2621 /* Asserts that child->frozen == false */
2622 bdrv_replace_child_noperm(child, new_bs);
2623
2624 /*
2625 * Start with the new node's permissions. If @new_bs is a (direct
2626 * or indirect) child of @old_bs, we must complete the permission
2627 * update on @new_bs before we loosen the restrictions on @old_bs.
2628 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2629 * updating the permissions of @new_bs, and thus not purely loosen
2630 * restrictions.
2631 */
2632 if (new_bs) {
2633 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2634 bdrv_set_perm(new_bs, perm, shared_perm);
2635 }
2636
2637 if (old_bs) {
2638 /* Update permissions for old node. This is guaranteed to succeed
2639 * because we're just taking a parent away, so we're loosening
2640 * restrictions. */
2641 bool tighten_restrictions;
2642 int ret;
2643
2644 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2645 ret = bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL,
2646 &tighten_restrictions, NULL);
2647 assert(tighten_restrictions == false);
2648 if (ret < 0) {
2649 /* We only tried to loosen restrictions, so errors are not fatal */
2650 bdrv_abort_perm_update(old_bs);
2651 } else {
2652 bdrv_set_perm(old_bs, perm, shared_perm);
2653 }
2654
2655 /* When the parent requiring a non-default AioContext is removed, the
2656 * node moves back to the main AioContext */
2657 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2658 }
2659 }
2660
2661 /*
2662 * This function steals the reference to child_bs from the caller.
2663 * That reference is later dropped by bdrv_root_unref_child().
2664 *
2665 * On failure NULL is returned, errp is set and the reference to
2666 * child_bs is also dropped.
2667 *
2668 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2669 * (unless @child_bs is already in @ctx).
2670 */
2671 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2672 const char *child_name,
2673 const BdrvChildClass *child_class,
2674 BdrvChildRole child_role,
2675 AioContext *ctx,
2676 uint64_t perm, uint64_t shared_perm,
2677 void *opaque, Error **errp)
2678 {
2679 BdrvChild *child;
2680 Error *local_err = NULL;
2681 int ret;
2682
2683 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, NULL,
2684 errp);
2685 if (ret < 0) {
2686 bdrv_abort_perm_update(child_bs);
2687 bdrv_unref(child_bs);
2688 return NULL;
2689 }
2690
2691 child = g_new(BdrvChild, 1);
2692 *child = (BdrvChild) {
2693 .bs = NULL,
2694 .name = g_strdup(child_name),
2695 .klass = child_class,
2696 .role = child_role,
2697 .perm = perm,
2698 .shared_perm = shared_perm,
2699 .opaque = opaque,
2700 };
2701
2702 /* If the AioContexts don't match, first try to move the subtree of
2703 * child_bs into the AioContext of the new parent. If this doesn't work,
2704 * try moving the parent into the AioContext of child_bs instead. */
2705 if (bdrv_get_aio_context(child_bs) != ctx) {
2706 ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err);
2707 if (ret < 0 && child_class->can_set_aio_ctx) {
2708 GSList *ignore = g_slist_prepend(NULL, child);
2709 ctx = bdrv_get_aio_context(child_bs);
2710 if (child_class->can_set_aio_ctx(child, ctx, &ignore, NULL)) {
2711 error_free(local_err);
2712 ret = 0;
2713 g_slist_free(ignore);
2714 ignore = g_slist_prepend(NULL, child);
2715 child_class->set_aio_ctx(child, ctx, &ignore);
2716 }
2717 g_slist_free(ignore);
2718 }
2719 if (ret < 0) {
2720 error_propagate(errp, local_err);
2721 g_free(child);
2722 bdrv_abort_perm_update(child_bs);
2723 bdrv_unref(child_bs);
2724 return NULL;
2725 }
2726 }
2727
2728 /* This performs the matching bdrv_set_perm() for the above check. */
2729 bdrv_replace_child(child, child_bs);
2730
2731 return child;
2732 }
2733
2734 /*
2735 * This function transfers the reference to child_bs from the caller
2736 * to parent_bs. That reference is later dropped by parent_bs on
2737 * bdrv_close() or if someone calls bdrv_unref_child().
2738 *
2739 * On failure NULL is returned, errp is set and the reference to
2740 * child_bs is also dropped.
2741 *
2742 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2743 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2744 */
2745 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2746 BlockDriverState *child_bs,
2747 const char *child_name,
2748 const BdrvChildClass *child_class,
2749 BdrvChildRole child_role,
2750 Error **errp)
2751 {
2752 BdrvChild *child;
2753 uint64_t perm, shared_perm;
2754
2755 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2756
2757 assert(parent_bs->drv);
2758 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2759 perm, shared_perm, &perm, &shared_perm);
2760
2761 child = bdrv_root_attach_child(child_bs, child_name, child_class,
2762 child_role, bdrv_get_aio_context(parent_bs),
2763 perm, shared_perm, parent_bs, errp);
2764 if (child == NULL) {
2765 return NULL;
2766 }
2767
2768 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2769 return child;
2770 }
2771
2772 static void bdrv_detach_child(BdrvChild *child)
2773 {
2774 QLIST_SAFE_REMOVE(child, next);
2775
2776 bdrv_replace_child(child, NULL);
2777
2778 g_free(child->name);
2779 g_free(child);
2780 }
2781
2782 /* Callers must ensure that child->frozen is false. */
2783 void bdrv_root_unref_child(BdrvChild *child)
2784 {
2785 BlockDriverState *child_bs;
2786
2787 child_bs = child->bs;
2788 bdrv_detach_child(child);
2789 bdrv_unref(child_bs);
2790 }
2791
2792 /**
2793 * Clear all inherits_from pointers from children and grandchildren of
2794 * @root that point to @root, where necessary.
2795 */
2796 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child)
2797 {
2798 BdrvChild *c;
2799
2800 if (child->bs->inherits_from == root) {
2801 /*
2802 * Remove inherits_from only when the last reference between root and
2803 * child->bs goes away.
2804 */
2805 QLIST_FOREACH(c, &root->children, next) {
2806 if (c != child && c->bs == child->bs) {
2807 break;
2808 }
2809 }
2810 if (c == NULL) {
2811 child->bs->inherits_from = NULL;
2812 }
2813 }
2814
2815 QLIST_FOREACH(c, &child->bs->children, next) {
2816 bdrv_unset_inherits_from(root, c);
2817 }
2818 }
2819
2820 /* Callers must ensure that child->frozen is false. */
2821 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2822 {
2823 if (child == NULL) {
2824 return;
2825 }
2826
2827 bdrv_unset_inherits_from(parent, child);
2828 bdrv_root_unref_child(child);
2829 }
2830
2831
2832 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2833 {
2834 BdrvChild *c;
2835 QLIST_FOREACH(c, &bs->parents, next_parent) {
2836 if (c->klass->change_media) {
2837 c->klass->change_media(c, load);
2838 }
2839 }
2840 }
2841
2842 /* Return true if you can reach parent going through child->inherits_from
2843 * recursively. If parent or child are NULL, return false */
2844 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2845 BlockDriverState *parent)
2846 {
2847 while (child && child != parent) {
2848 child = child->inherits_from;
2849 }
2850
2851 return child != NULL;
2852 }
2853
2854 /*
2855 * Return the BdrvChildRole for @bs's backing child. bs->backing is
2856 * mostly used for COW backing children (role = COW), but also for
2857 * filtered children (role = FILTERED | PRIMARY).
2858 */
2859 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
2860 {
2861 if (bs->drv && bs->drv->is_filter) {
2862 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
2863 } else {
2864 return BDRV_CHILD_COW;
2865 }
2866 }
2867
2868 /*
2869 * Sets the bs->backing link of a BDS. A new reference is created; callers
2870 * which don't need their own reference any more must call bdrv_unref().
2871 */
2872 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2873 Error **errp)
2874 {
2875 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2876 bdrv_inherits_from_recursive(backing_hd, bs);
2877
2878 if (bdrv_is_backing_chain_frozen(bs, child_bs(bs->backing), errp)) {
2879 return;
2880 }
2881
2882 if (backing_hd) {
2883 bdrv_ref(backing_hd);
2884 }
2885
2886 if (bs->backing) {
2887 /* Cannot be frozen, we checked that above */
2888 bdrv_unref_child(bs, bs->backing);
2889 bs->backing = NULL;
2890 }
2891
2892 if (!backing_hd) {
2893 goto out;
2894 }
2895
2896 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_of_bds,
2897 bdrv_backing_role(bs), errp);
2898 /* If backing_hd was already part of bs's backing chain, and
2899 * inherits_from pointed recursively to bs then let's update it to
2900 * point directly to bs (else it will become NULL). */
2901 if (bs->backing && update_inherits_from) {
2902 backing_hd->inherits_from = bs;
2903 }
2904
2905 out:
2906 bdrv_refresh_limits(bs, NULL);
2907 }
2908
2909 /*
2910 * Opens the backing file for a BlockDriverState if not yet open
2911 *
2912 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2913 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2914 * itself, all options starting with "${bdref_key}." are considered part of the
2915 * BlockdevRef.
2916 *
2917 * TODO Can this be unified with bdrv_open_image()?
2918 */
2919 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2920 const char *bdref_key, Error **errp)
2921 {
2922 char *backing_filename = NULL;
2923 char *bdref_key_dot;
2924 const char *reference = NULL;
2925 int ret = 0;
2926 bool implicit_backing = false;
2927 BlockDriverState *backing_hd;
2928 QDict *options;
2929 QDict *tmp_parent_options = NULL;
2930 Error *local_err = NULL;
2931
2932 if (bs->backing != NULL) {
2933 goto free_exit;
2934 }
2935
2936 /* NULL means an empty set of options */
2937 if (parent_options == NULL) {
2938 tmp_parent_options = qdict_new();
2939 parent_options = tmp_parent_options;
2940 }
2941
2942 bs->open_flags &= ~BDRV_O_NO_BACKING;
2943
2944 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2945 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2946 g_free(bdref_key_dot);
2947
2948 /*
2949 * Caution: while qdict_get_try_str() is fine, getting non-string
2950 * types would require more care. When @parent_options come from
2951 * -blockdev or blockdev_add, its members are typed according to
2952 * the QAPI schema, but when they come from -drive, they're all
2953 * QString.
2954 */
2955 reference = qdict_get_try_str(parent_options, bdref_key);
2956 if (reference || qdict_haskey(options, "file.filename")) {
2957 /* keep backing_filename NULL */
2958 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2959 qobject_unref(options);
2960 goto free_exit;
2961 } else {
2962 if (qdict_size(options) == 0) {
2963 /* If the user specifies options that do not modify the
2964 * backing file's behavior, we might still consider it the
2965 * implicit backing file. But it's easier this way, and
2966 * just specifying some of the backing BDS's options is
2967 * only possible with -drive anyway (otherwise the QAPI
2968 * schema forces the user to specify everything). */
2969 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2970 }
2971
2972 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2973 if (local_err) {
2974 ret = -EINVAL;
2975 error_propagate(errp, local_err);
2976 qobject_unref(options);
2977 goto free_exit;
2978 }
2979 }
2980
2981 if (!bs->drv || !bs->drv->supports_backing) {
2982 ret = -EINVAL;
2983 error_setg(errp, "Driver doesn't support backing files");
2984 qobject_unref(options);
2985 goto free_exit;
2986 }
2987
2988 if (!reference &&
2989 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2990 qdict_put_str(options, "driver", bs->backing_format);
2991 }
2992
2993 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2994 &child_of_bds, bdrv_backing_role(bs), errp);
2995 if (!backing_hd) {
2996 bs->open_flags |= BDRV_O_NO_BACKING;
2997 error_prepend(errp, "Could not open backing file: ");
2998 ret = -EINVAL;
2999 goto free_exit;
3000 }
3001
3002 if (implicit_backing) {
3003 bdrv_refresh_filename(backing_hd);
3004 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3005 backing_hd->filename);
3006 }
3007
3008 /* Hook up the backing file link; drop our reference, bs owns the
3009 * backing_hd reference now */
3010 bdrv_set_backing_hd(bs, backing_hd, &local_err);
3011 bdrv_unref(backing_hd);
3012 if (local_err) {
3013 error_propagate(errp, local_err);
3014 ret = -EINVAL;
3015 goto free_exit;
3016 }
3017
3018 qdict_del(parent_options, bdref_key);
3019
3020 free_exit:
3021 g_free(backing_filename);
3022 qobject_unref(tmp_parent_options);
3023 return ret;
3024 }
3025
3026 static BlockDriverState *
3027 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3028 BlockDriverState *parent, const BdrvChildClass *child_class,
3029 BdrvChildRole child_role, bool allow_none, Error **errp)
3030 {
3031 BlockDriverState *bs = NULL;
3032 QDict *image_options;
3033 char *bdref_key_dot;
3034 const char *reference;
3035
3036 assert(child_class != NULL);
3037
3038 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3039 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3040 g_free(bdref_key_dot);
3041
3042 /*
3043 * Caution: while qdict_get_try_str() is fine, getting non-string
3044 * types would require more care. When @options come from
3045 * -blockdev or blockdev_add, its members are typed according to
3046 * the QAPI schema, but when they come from -drive, they're all
3047 * QString.
3048 */
3049 reference = qdict_get_try_str(options, bdref_key);
3050 if (!filename && !reference && !qdict_size(image_options)) {
3051 if (!allow_none) {
3052 error_setg(errp, "A block device must be specified for \"%s\"",
3053 bdref_key);
3054 }
3055 qobject_unref(image_options);
3056 goto done;
3057 }
3058
3059 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3060 parent, child_class, child_role, errp);
3061 if (!bs) {
3062 goto done;
3063 }
3064
3065 done:
3066 qdict_del(options, bdref_key);
3067 return bs;
3068 }
3069
3070 /*
3071 * Opens a disk image whose options are given as BlockdevRef in another block
3072 * device's options.
3073 *
3074 * If allow_none is true, no image will be opened if filename is false and no
3075 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3076 *
3077 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3078 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3079 * itself, all options starting with "${bdref_key}." are considered part of the
3080 * BlockdevRef.
3081 *
3082 * The BlockdevRef will be removed from the options QDict.
3083 */
3084 BdrvChild *bdrv_open_child(const char *filename,
3085 QDict *options, const char *bdref_key,
3086 BlockDriverState *parent,
3087 const BdrvChildClass *child_class,
3088 BdrvChildRole child_role,
3089 bool allow_none, Error **errp)
3090 {
3091 BlockDriverState *bs;
3092
3093 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3094 child_role, allow_none, errp);
3095 if (bs == NULL) {
3096 return NULL;
3097 }
3098
3099 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3100 errp);
3101 }
3102
3103 /*
3104 * TODO Future callers may need to specify parent/child_class in order for
3105 * option inheritance to work. Existing callers use it for the root node.
3106 */
3107 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3108 {
3109 BlockDriverState *bs = NULL;
3110 QObject *obj = NULL;
3111 QDict *qdict = NULL;
3112 const char *reference = NULL;
3113 Visitor *v = NULL;
3114
3115 if (ref->type == QTYPE_QSTRING) {
3116 reference = ref->u.reference;
3117 } else {
3118 BlockdevOptions *options = &ref->u.definition;
3119 assert(ref->type == QTYPE_QDICT);
3120
3121 v = qobject_output_visitor_new(&obj);
3122 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3123 visit_complete(v, &obj);
3124
3125 qdict = qobject_to(QDict, obj);
3126 qdict_flatten(qdict);
3127
3128 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3129 * compatibility with other callers) rather than what we want as the
3130 * real defaults. Apply the defaults here instead. */
3131 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3132 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3133 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3134 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3135
3136 }
3137
3138 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3139 obj = NULL;
3140 qobject_unref(obj);
3141 visit_free(v);
3142 return bs;
3143 }
3144
3145 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3146 int flags,
3147 QDict *snapshot_options,
3148 Error **errp)
3149 {
3150 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3151 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3152 int64_t total_size;
3153 QemuOpts *opts = NULL;
3154 BlockDriverState *bs_snapshot = NULL;
3155 Error *local_err = NULL;
3156 int ret;
3157
3158 /* if snapshot, we create a temporary backing file and open it
3159 instead of opening 'filename' directly */
3160
3161 /* Get the required size from the image */
3162 total_size = bdrv_getlength(bs);
3163 if (total_size < 0) {
3164 error_setg_errno(errp, -total_size, "Could not get image size");
3165 goto out;
3166 }
3167
3168 /* Create the temporary image */
3169 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3170 if (ret < 0) {
3171 error_setg_errno(errp, -ret, "Could not get temporary filename");
3172 goto out;
3173 }
3174
3175 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3176 &error_abort);
3177 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3178 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3179 qemu_opts_del(opts);
3180 if (ret < 0) {
3181 error_prepend(errp, "Could not create temporary overlay '%s': ",
3182 tmp_filename);
3183 goto out;
3184 }
3185
3186 /* Prepare options QDict for the temporary file */
3187 qdict_put_str(snapshot_options, "file.driver", "file");
3188 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3189 qdict_put_str(snapshot_options, "driver", "qcow2");
3190
3191 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3192 snapshot_options = NULL;
3193 if (!bs_snapshot) {
3194 goto out;
3195 }
3196
3197 /* bdrv_append() consumes a strong reference to bs_snapshot
3198 * (i.e. it will call bdrv_unref() on it) even on error, so in
3199 * order to be able to return one, we have to increase
3200 * bs_snapshot's refcount here */
3201 bdrv_ref(bs_snapshot);
3202 bdrv_append(bs_snapshot, bs, &local_err);
3203 if (local_err) {
3204 error_propagate(errp, local_err);
3205 bs_snapshot = NULL;
3206 goto out;
3207 }
3208
3209 out:
3210 qobject_unref(snapshot_options);
3211 g_free(tmp_filename);
3212 return bs_snapshot;
3213 }
3214
3215 /*
3216 * Opens a disk image (raw, qcow2, vmdk, ...)
3217 *
3218 * options is a QDict of options to pass to the block drivers, or NULL for an
3219 * empty set of options. The reference to the QDict belongs to the block layer
3220 * after the call (even on failure), so if the caller intends to reuse the
3221 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3222 *
3223 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3224 * If it is not NULL, the referenced BDS will be reused.
3225 *
3226 * The reference parameter may be used to specify an existing block device which
3227 * should be opened. If specified, neither options nor a filename may be given,
3228 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3229 */
3230 static BlockDriverState *bdrv_open_inherit(const char *filename,
3231 const char *reference,
3232 QDict *options, int flags,
3233 BlockDriverState *parent,
3234 const BdrvChildClass *child_class,
3235 BdrvChildRole child_role,
3236 Error **errp)
3237 {
3238 int ret;
3239 BlockBackend *file = NULL;
3240 BlockDriverState *bs;
3241 BlockDriver *drv = NULL;
3242 BdrvChild *child;
3243 const char *drvname;
3244 const char *backing;
3245 Error *local_err = NULL;
3246 QDict *snapshot_options = NULL;
3247 int snapshot_flags = 0;
3248
3249 assert(!child_class || !flags);
3250 assert(!child_class == !parent);
3251
3252 if (reference) {
3253 bool options_non_empty = options ? qdict_size(options) : false;
3254 qobject_unref(options);
3255
3256 if (filename || options_non_empty) {
3257 error_setg(errp, "Cannot reference an existing block device with "
3258 "additional options or a new filename");
3259 return NULL;
3260 }
3261
3262 bs = bdrv_lookup_bs(reference, reference, errp);
3263 if (!bs) {
3264 return NULL;
3265 }
3266
3267 bdrv_ref(bs);
3268 return bs;
3269 }
3270
3271 bs = bdrv_new();
3272
3273 /* NULL means an empty set of options */
3274 if (options == NULL) {
3275 options = qdict_new();
3276 }
3277
3278 /* json: syntax counts as explicit options, as if in the QDict */
3279 parse_json_protocol(options, &filename, &local_err);
3280 if (local_err) {
3281 goto fail;
3282 }
3283
3284 bs->explicit_options = qdict_clone_shallow(options);
3285
3286 if (child_class) {
3287 bool parent_is_format;
3288
3289 if (parent->drv) {
3290 parent_is_format = parent->drv->is_format;
3291 } else {
3292 /*
3293 * parent->drv is not set yet because this node is opened for
3294 * (potential) format probing. That means that @parent is going
3295 * to be a format node.
3296 */
3297 parent_is_format = true;
3298 }
3299
3300 bs->inherits_from = parent;
3301 child_class->inherit_options(child_role, parent_is_format,
3302 &flags, options,
3303 parent->open_flags, parent->options);
3304 }
3305
3306 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3307 if (ret < 0) {
3308 goto fail;
3309 }
3310
3311 /*
3312 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3313 * Caution: getting a boolean member of @options requires care.
3314 * When @options come from -blockdev or blockdev_add, members are
3315 * typed according to the QAPI schema, but when they come from
3316 * -drive, they're all QString.
3317 */
3318 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3319 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3320 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3321 } else {
3322 flags &= ~BDRV_O_RDWR;
3323 }
3324
3325 if (flags & BDRV_O_SNAPSHOT) {
3326 snapshot_options = qdict_new();
3327 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3328 flags, options);
3329 /* Let bdrv_backing_options() override "read-only" */
3330 qdict_del(options, BDRV_OPT_READ_ONLY);
3331 bdrv_inherited_options(BDRV_CHILD_COW, true,
3332 &flags, options, flags, options);
3333 }
3334
3335 bs->open_flags = flags;
3336 bs->options = options;
3337 options = qdict_clone_shallow(options);
3338
3339 /* Find the right image format driver */
3340 /* See cautionary note on accessing @options above */
3341 drvname = qdict_get_try_str(options, "driver");
3342 if (drvname) {
3343 drv = bdrv_find_format(drvname);
3344 if (!drv) {
3345 error_setg(errp, "Unknown driver: '%s'", drvname);
3346 goto fail;
3347 }
3348 }
3349
3350 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3351
3352 /* See cautionary note on accessing @options above */
3353 backing = qdict_get_try_str(options, "backing");
3354 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3355 (backing && *backing == '\0'))
3356 {
3357 if (backing) {
3358 warn_report("Use of \"backing\": \"\" is deprecated; "
3359 "use \"backing\": null instead");
3360 }
3361 flags |= BDRV_O_NO_BACKING;
3362 qdict_del(bs->explicit_options, "backing");
3363 qdict_del(bs->options, "backing");
3364 qdict_del(options, "backing");
3365 }
3366
3367 /* Open image file without format layer. This BlockBackend is only used for
3368 * probing, the block drivers will do their own bdrv_open_child() for the
3369 * same BDS, which is why we put the node name back into options. */
3370 if ((flags & BDRV_O_PROTOCOL) == 0) {
3371 BlockDriverState *file_bs;
3372
3373 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3374 &child_of_bds, BDRV_CHILD_IMAGE,
3375 true, &local_err);
3376 if (local_err) {
3377 goto fail;
3378 }
3379 if (file_bs != NULL) {
3380 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3381 * looking at the header to guess the image format. This works even
3382 * in cases where a guest would not see a consistent state. */
3383 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3384 blk_insert_bs(file, file_bs, &local_err);
3385 bdrv_unref(file_bs);
3386 if (local_err) {
3387 goto fail;
3388 }
3389
3390 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3391 }
3392 }
3393
3394 /* Image format probing */
3395 bs->probed = !drv;
3396 if (!drv && file) {
3397 ret = find_image_format(file, filename, &drv, &local_err);
3398 if (ret < 0) {
3399 goto fail;
3400 }
3401 /*
3402 * This option update would logically belong in bdrv_fill_options(),
3403 * but we first need to open bs->file for the probing to work, while
3404 * opening bs->file already requires the (mostly) final set of options
3405 * so that cache mode etc. can be inherited.
3406 *
3407 * Adding the driver later is somewhat ugly, but it's not an option
3408 * that would ever be inherited, so it's correct. We just need to make
3409 * sure to update both bs->options (which has the full effective
3410 * options for bs) and options (which has file.* already removed).
3411 */
3412 qdict_put_str(bs->options, "driver", drv->format_name);
3413 qdict_put_str(options, "driver", drv->format_name);
3414 } else if (!drv) {
3415 error_setg(errp, "Must specify either driver or file");
3416 goto fail;
3417 }
3418
3419 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3420 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3421 /* file must be NULL if a protocol BDS is about to be created
3422 * (the inverse results in an error message from bdrv_open_common()) */
3423 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3424
3425 /* Open the image */
3426 ret = bdrv_open_common(bs, file, options, &local_err);
3427 if (ret < 0) {
3428 goto fail;
3429 }
3430
3431 if (file) {
3432 blk_unref(file);
3433 file = NULL;
3434 }
3435
3436 /* If there is a backing file, use it */
3437 if ((flags & BDRV_O_NO_BACKING) == 0) {
3438 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3439 if (ret < 0) {
3440 goto close_and_fail;
3441 }
3442 }
3443
3444 /* Remove all children options and references
3445 * from bs->options and bs->explicit_options */
3446 QLIST_FOREACH(child, &bs->children, next) {
3447 char *child_key_dot;
3448 child_key_dot = g_strdup_printf("%s.", child->name);
3449 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3450 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3451 qdict_del(bs->explicit_options, child->name);
3452 qdict_del(bs->options, child->name);
3453 g_free(child_key_dot);
3454 }
3455
3456 /* Check if any unknown options were used */
3457 if (qdict_size(options) != 0) {
3458 const QDictEntry *entry = qdict_first(options);
3459 if (flags & BDRV_O_PROTOCOL) {
3460 error_setg(errp, "Block protocol '%s' doesn't support the option "
3461 "'%s'", drv->format_name, entry->key);
3462 } else {
3463 error_setg(errp,
3464 "Block format '%s' does not support the option '%s'",
3465 drv->format_name, entry->key);
3466 }
3467
3468 goto close_and_fail;
3469 }
3470
3471 bdrv_parent_cb_change_media(bs, true);
3472
3473 qobject_unref(options);
3474 options = NULL;
3475
3476 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3477 * temporary snapshot afterwards. */
3478 if (snapshot_flags) {
3479 BlockDriverState *snapshot_bs;
3480 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3481 snapshot_options, &local_err);
3482 snapshot_options = NULL;
3483 if (local_err) {
3484 goto close_and_fail;
3485 }
3486 /* We are not going to return bs but the overlay on top of it
3487 * (snapshot_bs); thus, we have to drop the strong reference to bs
3488 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3489 * though, because the overlay still has a reference to it. */
3490 bdrv_unref(bs);
3491 bs = snapshot_bs;
3492 }
3493
3494 return bs;
3495
3496 fail:
3497 blk_unref(file);
3498 qobject_unref(snapshot_options);
3499 qobject_unref(bs->explicit_options);
3500 qobject_unref(bs->options);
3501 qobject_unref(options);
3502 bs->options = NULL;
3503 bs->explicit_options = NULL;
3504 bdrv_unref(bs);
3505 error_propagate(errp, local_err);
3506 return NULL;
3507
3508 close_and_fail:
3509 bdrv_unref(bs);
3510 qobject_unref(snapshot_options);
3511 qobject_unref(options);
3512 error_propagate(errp, local_err);
3513 return NULL;
3514 }
3515
3516 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3517 QDict *options, int flags, Error **errp)
3518 {
3519 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3520 NULL, 0, errp);
3521 }
3522
3523 /* Return true if the NULL-terminated @list contains @str */
3524 static bool is_str_in_list(const char *str, const char *const *list)
3525 {
3526 if (str && list) {
3527 int i;
3528 for (i = 0; list[i] != NULL; i++) {
3529 if (!strcmp(str, list[i])) {
3530 return true;
3531 }
3532 }
3533 }
3534 return false;
3535 }
3536
3537 /*
3538 * Check that every option set in @bs->options is also set in
3539 * @new_opts.
3540 *
3541 * Options listed in the common_options list and in
3542 * @bs->drv->mutable_opts are skipped.
3543 *
3544 * Return 0 on success, otherwise return -EINVAL and set @errp.
3545 */
3546 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3547 const QDict *new_opts, Error **errp)
3548 {
3549 const QDictEntry *e;
3550 /* These options are common to all block drivers and are handled
3551 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3552 const char *const common_options[] = {
3553 "node-name", "discard", "cache.direct", "cache.no-flush",
3554 "read-only", "auto-read-only", "detect-zeroes", NULL
3555 };
3556
3557 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3558 if (!qdict_haskey(new_opts, e->key) &&
3559 !is_str_in_list(e->key, common_options) &&
3560 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3561 error_setg(errp, "Option '%s' cannot be reset "
<