tests/tcg: add more default compilers to configure.sh
[qemu.git] / block.c
1 /*
2 * QEMU System Emulator block driver
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "module_block.h"
33 #include "qemu/main-loop.h"
34 #include "qemu/module.h"
35 #include "qapi/error.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qjson.h"
38 #include "qapi/qmp/qnull.h"
39 #include "qapi/qmp/qstring.h"
40 #include "qapi/qobject-output-visitor.h"
41 #include "qapi/qapi-visit-block-core.h"
42 #include "sysemu/block-backend.h"
43 #include "sysemu/sysemu.h"
44 #include "qemu/notify.h"
45 #include "qemu/option.h"
46 #include "qemu/coroutine.h"
47 #include "block/qapi.h"
48 #include "qemu/timer.h"
49 #include "qemu/cutils.h"
50 #include "qemu/id.h"
51
52 #ifdef CONFIG_BSD
53 #include <sys/ioctl.h>
54 #include <sys/queue.h>
55 #ifndef __DragonFly__
56 #include <sys/disk.h>
57 #endif
58 #endif
59
60 #ifdef _WIN32
61 #include <windows.h>
62 #endif
63
64 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
65
66 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
67 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
68
69 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
70 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
71
72 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
73 QLIST_HEAD_INITIALIZER(bdrv_drivers);
74
75 static BlockDriverState *bdrv_open_inherit(const char *filename,
76 const char *reference,
77 QDict *options, int flags,
78 BlockDriverState *parent,
79 const 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 };
437 static const char *whitelist_ro[] = {
438 CONFIG_BDRV_RO_WHITELIST
439 };
440 const char **p;
441
442 if (!whitelist_rw[0] && !whitelist_ro[0]) {
443 return 1; /* no whitelist, anything goes */
444 }
445
446 for (p = whitelist_rw; *p; p++) {
447 if (!strcmp(format_name, *p)) {
448 return 1;
449 }
450 }
451 if (read_only) {
452 for (p = whitelist_ro; *p; p++) {
453 if (!strcmp(format_name, *p)) {
454 return 1;
455 }
456 }
457 }
458 return 0;
459 }
460
461 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
462 {
463 return bdrv_format_is_whitelisted(drv->format_name, read_only);
464 }
465
466 bool bdrv_uses_whitelist(void)
467 {
468 return use_bdrv_whitelist;
469 }
470
471 typedef struct CreateCo {
472 BlockDriver *drv;
473 char *filename;
474 QemuOpts *opts;
475 int ret;
476 Error *err;
477 } CreateCo;
478
479 static void coroutine_fn bdrv_create_co_entry(void *opaque)
480 {
481 Error *local_err = NULL;
482 int ret;
483
484 CreateCo *cco = opaque;
485 assert(cco->drv);
486
487 ret = cco->drv->bdrv_co_create_opts(cco->drv,
488 cco->filename, cco->opts, &local_err);
489 error_propagate(&cco->err, local_err);
490 cco->ret = ret;
491 }
492
493 int bdrv_create(BlockDriver *drv, const char* filename,
494 QemuOpts *opts, Error **errp)
495 {
496 int ret;
497
498 Coroutine *co;
499 CreateCo cco = {
500 .drv = drv,
501 .filename = g_strdup(filename),
502 .opts = opts,
503 .ret = NOT_DONE,
504 .err = NULL,
505 };
506
507 if (!drv->bdrv_co_create_opts) {
508 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
509 ret = -ENOTSUP;
510 goto out;
511 }
512
513 if (qemu_in_coroutine()) {
514 /* Fast-path if already in coroutine context */
515 bdrv_create_co_entry(&cco);
516 } else {
517 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
518 qemu_coroutine_enter(co);
519 while (cco.ret == NOT_DONE) {
520 aio_poll(qemu_get_aio_context(), true);
521 }
522 }
523
524 ret = cco.ret;
525 if (ret < 0) {
526 if (cco.err) {
527 error_propagate(errp, cco.err);
528 } else {
529 error_setg_errno(errp, -ret, "Could not create image");
530 }
531 }
532
533 out:
534 g_free(cco.filename);
535 return ret;
536 }
537
538 /**
539 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
540 * least the given @minimum_size.
541 *
542 * On success, return @blk's actual length.
543 * Otherwise, return -errno.
544 */
545 static int64_t create_file_fallback_truncate(BlockBackend *blk,
546 int64_t minimum_size, Error **errp)
547 {
548 Error *local_err = NULL;
549 int64_t size;
550 int ret;
551
552 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
553 &local_err);
554 if (ret < 0 && ret != -ENOTSUP) {
555 error_propagate(errp, local_err);
556 return ret;
557 }
558
559 size = blk_getlength(blk);
560 if (size < 0) {
561 error_free(local_err);
562 error_setg_errno(errp, -size,
563 "Failed to inquire the new image file's length");
564 return size;
565 }
566
567 if (size < minimum_size) {
568 /* Need to grow the image, but we failed to do that */
569 error_propagate(errp, local_err);
570 return -ENOTSUP;
571 }
572
573 error_free(local_err);
574 local_err = NULL;
575
576 return size;
577 }
578
579 /**
580 * Helper function for bdrv_create_file_fallback(): Zero the first
581 * sector to remove any potentially pre-existing image header.
582 */
583 static int create_file_fallback_zero_first_sector(BlockBackend *blk,
584 int64_t current_size,
585 Error **errp)
586 {
587 int64_t bytes_to_clear;
588 int ret;
589
590 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
591 if (bytes_to_clear) {
592 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
593 if (ret < 0) {
594 error_setg_errno(errp, -ret,
595 "Failed to clear the new image's first sector");
596 return ret;
597 }
598 }
599
600 return 0;
601 }
602
603 /**
604 * Simple implementation of bdrv_co_create_opts for protocol drivers
605 * which only support creation via opening a file
606 * (usually existing raw storage device)
607 */
608 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
609 const char *filename,
610 QemuOpts *opts,
611 Error **errp)
612 {
613 BlockBackend *blk;
614 QDict *options;
615 int64_t size = 0;
616 char *buf = NULL;
617 PreallocMode prealloc;
618 Error *local_err = NULL;
619 int ret;
620
621 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
622 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
623 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
624 PREALLOC_MODE_OFF, &local_err);
625 g_free(buf);
626 if (local_err) {
627 error_propagate(errp, local_err);
628 return -EINVAL;
629 }
630
631 if (prealloc != PREALLOC_MODE_OFF) {
632 error_setg(errp, "Unsupported preallocation mode '%s'",
633 PreallocMode_str(prealloc));
634 return -ENOTSUP;
635 }
636
637 options = qdict_new();
638 qdict_put_str(options, "driver", drv->format_name);
639
640 blk = blk_new_open(filename, NULL, options,
641 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
642 if (!blk) {
643 error_prepend(errp, "Protocol driver '%s' does not support image "
644 "creation, and opening the image failed: ",
645 drv->format_name);
646 return -EINVAL;
647 }
648
649 size = create_file_fallback_truncate(blk, size, errp);
650 if (size < 0) {
651 ret = size;
652 goto out;
653 }
654
655 ret = create_file_fallback_zero_first_sector(blk, size, errp);
656 if (ret < 0) {
657 goto out;
658 }
659
660 ret = 0;
661 out:
662 blk_unref(blk);
663 return ret;
664 }
665
666 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
667 {
668 BlockDriver *drv;
669
670 drv = bdrv_find_protocol(filename, true, errp);
671 if (drv == NULL) {
672 return -ENOENT;
673 }
674
675 return bdrv_create(drv, filename, opts, errp);
676 }
677
678 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
679 {
680 Error *local_err = NULL;
681 int ret;
682
683 assert(bs != NULL);
684
685 if (!bs->drv) {
686 error_setg(errp, "Block node '%s' is not opened", bs->filename);
687 return -ENOMEDIUM;
688 }
689
690 if (!bs->drv->bdrv_co_delete_file) {
691 error_setg(errp, "Driver '%s' does not support image deletion",
692 bs->drv->format_name);
693 return -ENOTSUP;
694 }
695
696 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
697 if (ret < 0) {
698 error_propagate(errp, local_err);
699 }
700
701 return ret;
702 }
703
704 /**
705 * Try to get @bs's logical and physical block size.
706 * On success, store them in @bsz struct and return 0.
707 * On failure return -errno.
708 * @bs must not be empty.
709 */
710 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
711 {
712 BlockDriver *drv = bs->drv;
713
714 if (drv && drv->bdrv_probe_blocksizes) {
715 return drv->bdrv_probe_blocksizes(bs, bsz);
716 } else if (drv && drv->is_filter && bs->file) {
717 return bdrv_probe_blocksizes(bs->file->bs, bsz);
718 }
719
720 return -ENOTSUP;
721 }
722
723 /**
724 * Try to get @bs's geometry (cyls, heads, sectors).
725 * On success, store them in @geo struct and return 0.
726 * On failure return -errno.
727 * @bs must not be empty.
728 */
729 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
730 {
731 BlockDriver *drv = bs->drv;
732
733 if (drv && drv->bdrv_probe_geometry) {
734 return drv->bdrv_probe_geometry(bs, geo);
735 } else if (drv && drv->is_filter && bs->file) {
736 return bdrv_probe_geometry(bs->file->bs, geo);
737 }
738
739 return -ENOTSUP;
740 }
741
742 /*
743 * Create a uniquely-named empty temporary file.
744 * Return 0 upon success, otherwise a negative errno value.
745 */
746 int get_tmp_filename(char *filename, int size)
747 {
748 #ifdef _WIN32
749 char temp_dir[MAX_PATH];
750 /* GetTempFileName requires that its output buffer (4th param)
751 have length MAX_PATH or greater. */
752 assert(size >= MAX_PATH);
753 return (GetTempPath(MAX_PATH, temp_dir)
754 && GetTempFileName(temp_dir, "qem", 0, filename)
755 ? 0 : -GetLastError());
756 #else
757 int fd;
758 const char *tmpdir;
759 tmpdir = getenv("TMPDIR");
760 if (!tmpdir) {
761 tmpdir = "/var/tmp";
762 }
763 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
764 return -EOVERFLOW;
765 }
766 fd = mkstemp(filename);
767 if (fd < 0) {
768 return -errno;
769 }
770 if (close(fd) != 0) {
771 unlink(filename);
772 return -errno;
773 }
774 return 0;
775 #endif
776 }
777
778 /*
779 * Detect host devices. By convention, /dev/cdrom[N] is always
780 * recognized as a host CDROM.
781 */
782 static BlockDriver *find_hdev_driver(const char *filename)
783 {
784 int score_max = 0, score;
785 BlockDriver *drv = NULL, *d;
786
787 QLIST_FOREACH(d, &bdrv_drivers, list) {
788 if (d->bdrv_probe_device) {
789 score = d->bdrv_probe_device(filename);
790 if (score > score_max) {
791 score_max = score;
792 drv = d;
793 }
794 }
795 }
796
797 return drv;
798 }
799
800 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
801 {
802 BlockDriver *drv1;
803
804 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
805 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
806 return drv1;
807 }
808 }
809
810 return NULL;
811 }
812
813 BlockDriver *bdrv_find_protocol(const char *filename,
814 bool allow_protocol_prefix,
815 Error **errp)
816 {
817 BlockDriver *drv1;
818 char protocol[128];
819 int len;
820 const char *p;
821 int i;
822
823 /* TODO Drivers without bdrv_file_open must be specified explicitly */
824
825 /*
826 * XXX(hch): we really should not let host device detection
827 * override an explicit protocol specification, but moving this
828 * later breaks access to device names with colons in them.
829 * Thanks to the brain-dead persistent naming schemes on udev-
830 * based Linux systems those actually are quite common.
831 */
832 drv1 = find_hdev_driver(filename);
833 if (drv1) {
834 return drv1;
835 }
836
837 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
838 return &bdrv_file;
839 }
840
841 p = strchr(filename, ':');
842 assert(p != NULL);
843 len = p - filename;
844 if (len > sizeof(protocol) - 1)
845 len = sizeof(protocol) - 1;
846 memcpy(protocol, filename, len);
847 protocol[len] = '\0';
848
849 drv1 = bdrv_do_find_protocol(protocol);
850 if (drv1) {
851 return drv1;
852 }
853
854 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
855 if (block_driver_modules[i].protocol_name &&
856 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
857 block_module_load_one(block_driver_modules[i].library_name);
858 break;
859 }
860 }
861
862 drv1 = bdrv_do_find_protocol(protocol);
863 if (!drv1) {
864 error_setg(errp, "Unknown protocol '%s'", protocol);
865 }
866 return drv1;
867 }
868
869 /*
870 * Guess image format by probing its contents.
871 * This is not a good idea when your image is raw (CVE-2008-2004), but
872 * we do it anyway for backward compatibility.
873 *
874 * @buf contains the image's first @buf_size bytes.
875 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
876 * but can be smaller if the image file is smaller)
877 * @filename is its filename.
878 *
879 * For all block drivers, call the bdrv_probe() method to get its
880 * probing score.
881 * Return the first block driver with the highest probing score.
882 */
883 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
884 const char *filename)
885 {
886 int score_max = 0, score;
887 BlockDriver *drv = NULL, *d;
888
889 QLIST_FOREACH(d, &bdrv_drivers, list) {
890 if (d->bdrv_probe) {
891 score = d->bdrv_probe(buf, buf_size, filename);
892 if (score > score_max) {
893 score_max = score;
894 drv = d;
895 }
896 }
897 }
898
899 return drv;
900 }
901
902 static int find_image_format(BlockBackend *file, const char *filename,
903 BlockDriver **pdrv, Error **errp)
904 {
905 BlockDriver *drv;
906 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
907 int ret = 0;
908
909 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
910 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
911 *pdrv = &bdrv_raw;
912 return ret;
913 }
914
915 ret = blk_pread(file, 0, buf, sizeof(buf));
916 if (ret < 0) {
917 error_setg_errno(errp, -ret, "Could not read image for determining its "
918 "format");
919 *pdrv = NULL;
920 return ret;
921 }
922
923 drv = bdrv_probe_all(buf, ret, filename);
924 if (!drv) {
925 error_setg(errp, "Could not determine image format: No compatible "
926 "driver found");
927 ret = -ENOENT;
928 }
929 *pdrv = drv;
930 return ret;
931 }
932
933 /**
934 * Set the current 'total_sectors' value
935 * Return 0 on success, -errno on error.
936 */
937 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
938 {
939 BlockDriver *drv = bs->drv;
940
941 if (!drv) {
942 return -ENOMEDIUM;
943 }
944
945 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
946 if (bdrv_is_sg(bs))
947 return 0;
948
949 /* query actual device if possible, otherwise just trust the hint */
950 if (drv->bdrv_getlength) {
951 int64_t length = drv->bdrv_getlength(bs);
952 if (length < 0) {
953 return length;
954 }
955 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
956 }
957
958 bs->total_sectors = hint;
959 return 0;
960 }
961
962 /**
963 * Combines a QDict of new block driver @options with any missing options taken
964 * from @old_options, so that leaving out an option defaults to its old value.
965 */
966 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
967 QDict *old_options)
968 {
969 if (bs->drv && bs->drv->bdrv_join_options) {
970 bs->drv->bdrv_join_options(options, old_options);
971 } else {
972 qdict_join(options, old_options, false);
973 }
974 }
975
976 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
977 int open_flags,
978 Error **errp)
979 {
980 Error *local_err = NULL;
981 char *value = qemu_opt_get_del(opts, "detect-zeroes");
982 BlockdevDetectZeroesOptions detect_zeroes =
983 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
984 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
985 g_free(value);
986 if (local_err) {
987 error_propagate(errp, local_err);
988 return detect_zeroes;
989 }
990
991 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
992 !(open_flags & BDRV_O_UNMAP))
993 {
994 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
995 "without setting discard operation to unmap");
996 }
997
998 return detect_zeroes;
999 }
1000
1001 /**
1002 * Set open flags for aio engine
1003 *
1004 * Return 0 on success, -1 if the engine specified is invalid
1005 */
1006 int bdrv_parse_aio(const char *mode, int *flags)
1007 {
1008 if (!strcmp(mode, "threads")) {
1009 /* do nothing, default */
1010 } else if (!strcmp(mode, "native")) {
1011 *flags |= BDRV_O_NATIVE_AIO;
1012 #ifdef CONFIG_LINUX_IO_URING
1013 } else if (!strcmp(mode, "io_uring")) {
1014 *flags |= BDRV_O_IO_URING;
1015 #endif
1016 } else {
1017 return -1;
1018 }
1019
1020 return 0;
1021 }
1022
1023 /**
1024 * Set open flags for a given discard mode
1025 *
1026 * Return 0 on success, -1 if the discard mode was invalid.
1027 */
1028 int bdrv_parse_discard_flags(const char *mode, int *flags)
1029 {
1030 *flags &= ~BDRV_O_UNMAP;
1031
1032 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1033 /* do nothing */
1034 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1035 *flags |= BDRV_O_UNMAP;
1036 } else {
1037 return -1;
1038 }
1039
1040 return 0;
1041 }
1042
1043 /**
1044 * Set open flags for a given cache mode
1045 *
1046 * Return 0 on success, -1 if the cache mode was invalid.
1047 */
1048 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1049 {
1050 *flags &= ~BDRV_O_CACHE_MASK;
1051
1052 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1053 *writethrough = false;
1054 *flags |= BDRV_O_NOCACHE;
1055 } else if (!strcmp(mode, "directsync")) {
1056 *writethrough = true;
1057 *flags |= BDRV_O_NOCACHE;
1058 } else if (!strcmp(mode, "writeback")) {
1059 *writethrough = false;
1060 } else if (!strcmp(mode, "unsafe")) {
1061 *writethrough = false;
1062 *flags |= BDRV_O_NO_FLUSH;
1063 } else if (!strcmp(mode, "writethrough")) {
1064 *writethrough = true;
1065 } else {
1066 return -1;
1067 }
1068
1069 return 0;
1070 }
1071
1072 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1073 {
1074 BlockDriverState *parent = c->opaque;
1075 return g_strdup(bdrv_get_device_or_node_name(parent));
1076 }
1077
1078 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1079 {
1080 BlockDriverState *bs = child->opaque;
1081 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1082 }
1083
1084 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1085 {
1086 BlockDriverState *bs = child->opaque;
1087 return bdrv_drain_poll(bs, false, NULL, false);
1088 }
1089
1090 static void bdrv_child_cb_drained_end(BdrvChild *child,
1091 int *drained_end_counter)
1092 {
1093 BlockDriverState *bs = child->opaque;
1094 bdrv_drained_end_no_poll(bs, drained_end_counter);
1095 }
1096
1097 static int bdrv_child_cb_inactivate(BdrvChild *child)
1098 {
1099 BlockDriverState *bs = child->opaque;
1100 assert(bs->open_flags & BDRV_O_INACTIVE);
1101 return 0;
1102 }
1103
1104 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1105 GSList **ignore, Error **errp)
1106 {
1107 BlockDriverState *bs = child->opaque;
1108 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1109 }
1110
1111 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1112 GSList **ignore)
1113 {
1114 BlockDriverState *bs = child->opaque;
1115 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1116 }
1117
1118 /*
1119 * Returns the options and flags that a temporary snapshot should get, based on
1120 * the originally requested flags (the originally requested image will have
1121 * flags like a backing file)
1122 */
1123 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1124 int parent_flags, QDict *parent_options)
1125 {
1126 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1127
1128 /* For temporary files, unconditional cache=unsafe is fine */
1129 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1130 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1131
1132 /* Copy the read-only and discard options from the parent */
1133 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1134 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1135
1136 /* aio=native doesn't work for cache.direct=off, so disable it for the
1137 * temporary snapshot */
1138 *child_flags &= ~BDRV_O_NATIVE_AIO;
1139 }
1140
1141 static void bdrv_backing_attach(BdrvChild *c)
1142 {
1143 BlockDriverState *parent = c->opaque;
1144 BlockDriverState *backing_hd = c->bs;
1145
1146 assert(!parent->backing_blocker);
1147 error_setg(&parent->backing_blocker,
1148 "node is used as backing hd of '%s'",
1149 bdrv_get_device_or_node_name(parent));
1150
1151 bdrv_refresh_filename(backing_hd);
1152
1153 parent->open_flags &= ~BDRV_O_NO_BACKING;
1154 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
1155 backing_hd->filename);
1156 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
1157 backing_hd->drv ? backing_hd->drv->format_name : "");
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 if (ret < 0) {
1211 error_setg_errno(errp, -ret, "Could not update backing file link");
1212 }
1213
1214 if (read_only) {
1215 bdrv_reopen_set_read_only(parent, true, NULL);
1216 }
1217
1218 return ret;
1219 }
1220
1221 /*
1222 * Returns the options and flags that a generic child of a BDS should
1223 * get, based on the given options and flags for the parent BDS.
1224 */
1225 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1226 int *child_flags, QDict *child_options,
1227 int parent_flags, QDict *parent_options)
1228 {
1229 int flags = parent_flags;
1230
1231 /*
1232 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1233 * Generally, the question to answer is: Should this child be
1234 * format-probed by default?
1235 */
1236
1237 /*
1238 * Pure and non-filtered data children of non-format nodes should
1239 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1240 * set). This only affects a very limited set of drivers (namely
1241 * quorum and blkverify when this comment was written).
1242 * Force-clear BDRV_O_PROTOCOL then.
1243 */
1244 if (!parent_is_format &&
1245 (role & BDRV_CHILD_DATA) &&
1246 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1247 {
1248 flags &= ~BDRV_O_PROTOCOL;
1249 }
1250
1251 /*
1252 * All children of format nodes (except for COW children) and all
1253 * metadata children in general should never be format-probed.
1254 * Force-set BDRV_O_PROTOCOL then.
1255 */
1256 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1257 (role & BDRV_CHILD_METADATA))
1258 {
1259 flags |= BDRV_O_PROTOCOL;
1260 }
1261
1262 /*
1263 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1264 * the parent.
1265 */
1266 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1267 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1268 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1269
1270 if (role & BDRV_CHILD_COW) {
1271 /* backing files are opened read-only by default */
1272 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1273 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1274 } else {
1275 /* Inherit the read-only option from the parent if it's not set */
1276 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1277 qdict_copy_default(child_options, parent_options,
1278 BDRV_OPT_AUTO_READ_ONLY);
1279 }
1280
1281 /*
1282 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1283 * can default to enable it on lower layers regardless of the
1284 * parent option.
1285 */
1286 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1287
1288 /* Clear flags that only apply to the top layer */
1289 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1290
1291 if (role & BDRV_CHILD_METADATA) {
1292 flags &= ~BDRV_O_NO_IO;
1293 }
1294 if (role & BDRV_CHILD_COW) {
1295 flags &= ~BDRV_O_TEMPORARY;
1296 }
1297
1298 *child_flags = flags;
1299 }
1300
1301 static void bdrv_child_cb_attach(BdrvChild *child)
1302 {
1303 BlockDriverState *bs = child->opaque;
1304
1305 if (child->role & BDRV_CHILD_COW) {
1306 bdrv_backing_attach(child);
1307 }
1308
1309 bdrv_apply_subtree_drain(child, bs);
1310 }
1311
1312 static void bdrv_child_cb_detach(BdrvChild *child)
1313 {
1314 BlockDriverState *bs = child->opaque;
1315
1316 if (child->role & BDRV_CHILD_COW) {
1317 bdrv_backing_detach(child);
1318 }
1319
1320 bdrv_unapply_subtree_drain(child, bs);
1321 }
1322
1323 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1324 const char *filename, Error **errp)
1325 {
1326 if (c->role & BDRV_CHILD_COW) {
1327 return bdrv_backing_update_filename(c, base, filename, errp);
1328 }
1329 return 0;
1330 }
1331
1332 const BdrvChildClass child_of_bds = {
1333 .parent_is_bds = true,
1334 .get_parent_desc = bdrv_child_get_parent_desc,
1335 .inherit_options = bdrv_inherited_options,
1336 .drained_begin = bdrv_child_cb_drained_begin,
1337 .drained_poll = bdrv_child_cb_drained_poll,
1338 .drained_end = bdrv_child_cb_drained_end,
1339 .attach = bdrv_child_cb_attach,
1340 .detach = bdrv_child_cb_detach,
1341 .inactivate = bdrv_child_cb_inactivate,
1342 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1343 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1344 .update_filename = bdrv_child_cb_update_filename,
1345 };
1346
1347 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1348 {
1349 int open_flags = flags;
1350
1351 /*
1352 * Clear flags that are internal to the block layer before opening the
1353 * image.
1354 */
1355 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1356
1357 return open_flags;
1358 }
1359
1360 static void update_flags_from_options(int *flags, QemuOpts *opts)
1361 {
1362 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1363
1364 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1365 *flags |= BDRV_O_NO_FLUSH;
1366 }
1367
1368 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1369 *flags |= BDRV_O_NOCACHE;
1370 }
1371
1372 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1373 *flags |= BDRV_O_RDWR;
1374 }
1375
1376 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1377 *flags |= BDRV_O_AUTO_RDONLY;
1378 }
1379 }
1380
1381 static void update_options_from_flags(QDict *options, int flags)
1382 {
1383 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1384 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1385 }
1386 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1387 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1388 flags & BDRV_O_NO_FLUSH);
1389 }
1390 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1391 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1392 }
1393 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1394 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1395 flags & BDRV_O_AUTO_RDONLY);
1396 }
1397 }
1398
1399 static void bdrv_assign_node_name(BlockDriverState *bs,
1400 const char *node_name,
1401 Error **errp)
1402 {
1403 char *gen_node_name = NULL;
1404
1405 if (!node_name) {
1406 node_name = gen_node_name = id_generate(ID_BLOCK);
1407 } else if (!id_wellformed(node_name)) {
1408 /*
1409 * Check for empty string or invalid characters, but not if it is
1410 * generated (generated names use characters not available to the user)
1411 */
1412 error_setg(errp, "Invalid node name");
1413 return;
1414 }
1415
1416 /* takes care of avoiding namespaces collisions */
1417 if (blk_by_name(node_name)) {
1418 error_setg(errp, "node-name=%s is conflicting with a device id",
1419 node_name);
1420 goto out;
1421 }
1422
1423 /* takes care of avoiding duplicates node names */
1424 if (bdrv_find_node(node_name)) {
1425 error_setg(errp, "Duplicate node name");
1426 goto out;
1427 }
1428
1429 /* Make sure that the node name isn't truncated */
1430 if (strlen(node_name) >= sizeof(bs->node_name)) {
1431 error_setg(errp, "Node name too long");
1432 goto out;
1433 }
1434
1435 /* copy node name into the bs and insert it into the graph list */
1436 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1437 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1438 out:
1439 g_free(gen_node_name);
1440 }
1441
1442 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1443 const char *node_name, QDict *options,
1444 int open_flags, Error **errp)
1445 {
1446 Error *local_err = NULL;
1447 int i, ret;
1448
1449 bdrv_assign_node_name(bs, node_name, &local_err);
1450 if (local_err) {
1451 error_propagate(errp, local_err);
1452 return -EINVAL;
1453 }
1454
1455 bs->drv = drv;
1456 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1457 bs->opaque = g_malloc0(drv->instance_size);
1458
1459 if (drv->bdrv_file_open) {
1460 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1461 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1462 } else if (drv->bdrv_open) {
1463 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1464 } else {
1465 ret = 0;
1466 }
1467
1468 if (ret < 0) {
1469 if (local_err) {
1470 error_propagate(errp, local_err);
1471 } else if (bs->filename[0]) {
1472 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1473 } else {
1474 error_setg_errno(errp, -ret, "Could not open image");
1475 }
1476 goto open_failed;
1477 }
1478
1479 ret = refresh_total_sectors(bs, bs->total_sectors);
1480 if (ret < 0) {
1481 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1482 return ret;
1483 }
1484
1485 bdrv_refresh_limits(bs, &local_err);
1486 if (local_err) {
1487 error_propagate(errp, local_err);
1488 return -EINVAL;
1489 }
1490
1491 assert(bdrv_opt_mem_align(bs) != 0);
1492 assert(bdrv_min_mem_align(bs) != 0);
1493 assert(is_power_of_2(bs->bl.request_alignment));
1494
1495 for (i = 0; i < bs->quiesce_counter; i++) {
1496 if (drv->bdrv_co_drain_begin) {
1497 drv->bdrv_co_drain_begin(bs);
1498 }
1499 }
1500
1501 return 0;
1502 open_failed:
1503 bs->drv = NULL;
1504 if (bs->file != NULL) {
1505 bdrv_unref_child(bs, bs->file);
1506 bs->file = NULL;
1507 }
1508 g_free(bs->opaque);
1509 bs->opaque = NULL;
1510 return ret;
1511 }
1512
1513 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1514 int flags, Error **errp)
1515 {
1516 BlockDriverState *bs;
1517 int ret;
1518
1519 bs = bdrv_new();
1520 bs->open_flags = flags;
1521 bs->explicit_options = qdict_new();
1522 bs->options = qdict_new();
1523 bs->opaque = NULL;
1524
1525 update_options_from_flags(bs->options, flags);
1526
1527 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1528 if (ret < 0) {
1529 qobject_unref(bs->explicit_options);
1530 bs->explicit_options = NULL;
1531 qobject_unref(bs->options);
1532 bs->options = NULL;
1533 bdrv_unref(bs);
1534 return NULL;
1535 }
1536
1537 return bs;
1538 }
1539
1540 QemuOptsList bdrv_runtime_opts = {
1541 .name = "bdrv_common",
1542 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1543 .desc = {
1544 {
1545 .name = "node-name",
1546 .type = QEMU_OPT_STRING,
1547 .help = "Node name of the block device node",
1548 },
1549 {
1550 .name = "driver",
1551 .type = QEMU_OPT_STRING,
1552 .help = "Block driver to use for the node",
1553 },
1554 {
1555 .name = BDRV_OPT_CACHE_DIRECT,
1556 .type = QEMU_OPT_BOOL,
1557 .help = "Bypass software writeback cache on the host",
1558 },
1559 {
1560 .name = BDRV_OPT_CACHE_NO_FLUSH,
1561 .type = QEMU_OPT_BOOL,
1562 .help = "Ignore flush requests",
1563 },
1564 {
1565 .name = BDRV_OPT_READ_ONLY,
1566 .type = QEMU_OPT_BOOL,
1567 .help = "Node is opened in read-only mode",
1568 },
1569 {
1570 .name = BDRV_OPT_AUTO_READ_ONLY,
1571 .type = QEMU_OPT_BOOL,
1572 .help = "Node can become read-only if opening read-write fails",
1573 },
1574 {
1575 .name = "detect-zeroes",
1576 .type = QEMU_OPT_STRING,
1577 .help = "try to optimize zero writes (off, on, unmap)",
1578 },
1579 {
1580 .name = BDRV_OPT_DISCARD,
1581 .type = QEMU_OPT_STRING,
1582 .help = "discard operation (ignore/off, unmap/on)",
1583 },
1584 {
1585 .name = BDRV_OPT_FORCE_SHARE,
1586 .type = QEMU_OPT_BOOL,
1587 .help = "always accept other writers (default: off)",
1588 },
1589 { /* end of list */ }
1590 },
1591 };
1592
1593 QemuOptsList bdrv_create_opts_simple = {
1594 .name = "simple-create-opts",
1595 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1596 .desc = {
1597 {
1598 .name = BLOCK_OPT_SIZE,
1599 .type = QEMU_OPT_SIZE,
1600 .help = "Virtual disk size"
1601 },
1602 {
1603 .name = BLOCK_OPT_PREALLOC,
1604 .type = QEMU_OPT_STRING,
1605 .help = "Preallocation mode (allowed values: off)"
1606 },
1607 { /* end of list */ }
1608 }
1609 };
1610
1611 /*
1612 * Common part for opening disk images and files
1613 *
1614 * Removes all processed options from *options.
1615 */
1616 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1617 QDict *options, Error **errp)
1618 {
1619 int ret, open_flags;
1620 const char *filename;
1621 const char *driver_name = NULL;
1622 const char *node_name = NULL;
1623 const char *discard;
1624 QemuOpts *opts;
1625 BlockDriver *drv;
1626 Error *local_err = NULL;
1627
1628 assert(bs->file == NULL);
1629 assert(options != NULL && bs->options != options);
1630
1631 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1632 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1633 ret = -EINVAL;
1634 goto fail_opts;
1635 }
1636
1637 update_flags_from_options(&bs->open_flags, opts);
1638
1639 driver_name = qemu_opt_get(opts, "driver");
1640 drv = bdrv_find_format(driver_name);
1641 assert(drv != NULL);
1642
1643 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1644
1645 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1646 error_setg(errp,
1647 BDRV_OPT_FORCE_SHARE
1648 "=on can only be used with read-only images");
1649 ret = -EINVAL;
1650 goto fail_opts;
1651 }
1652
1653 if (file != NULL) {
1654 bdrv_refresh_filename(blk_bs(file));
1655 filename = blk_bs(file)->filename;
1656 } else {
1657 /*
1658 * Caution: while qdict_get_try_str() is fine, getting
1659 * non-string types would require more care. When @options
1660 * come from -blockdev or blockdev_add, its members are typed
1661 * according to the QAPI schema, but when they come from
1662 * -drive, they're all QString.
1663 */
1664 filename = qdict_get_try_str(options, "filename");
1665 }
1666
1667 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1668 error_setg(errp, "The '%s' block driver requires a file name",
1669 drv->format_name);
1670 ret = -EINVAL;
1671 goto fail_opts;
1672 }
1673
1674 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1675 drv->format_name);
1676
1677 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1678
1679 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1680 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1681 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1682 } else {
1683 ret = -ENOTSUP;
1684 }
1685 if (ret < 0) {
1686 error_setg(errp,
1687 !bs->read_only && bdrv_is_whitelisted(drv, true)
1688 ? "Driver '%s' can only be used for read-only devices"
1689 : "Driver '%s' is not whitelisted",
1690 drv->format_name);
1691 goto fail_opts;
1692 }
1693 }
1694
1695 /* bdrv_new() and bdrv_close() make it so */
1696 assert(atomic_read(&bs->copy_on_read) == 0);
1697
1698 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1699 if (!bs->read_only) {
1700 bdrv_enable_copy_on_read(bs);
1701 } else {
1702 error_setg(errp, "Can't use copy-on-read on read-only device");
1703 ret = -EINVAL;
1704 goto fail_opts;
1705 }
1706 }
1707
1708 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1709 if (discard != NULL) {
1710 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1711 error_setg(errp, "Invalid discard option");
1712 ret = -EINVAL;
1713 goto fail_opts;
1714 }
1715 }
1716
1717 bs->detect_zeroes =
1718 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1719 if (local_err) {
1720 error_propagate(errp, local_err);
1721 ret = -EINVAL;
1722 goto fail_opts;
1723 }
1724
1725 if (filename != NULL) {
1726 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1727 } else {
1728 bs->filename[0] = '\0';
1729 }
1730 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1731
1732 /* Open the image, either directly or using a protocol */
1733 open_flags = bdrv_open_flags(bs, bs->open_flags);
1734 node_name = qemu_opt_get(opts, "node-name");
1735
1736 assert(!drv->bdrv_file_open || file == NULL);
1737 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1738 if (ret < 0) {
1739 goto fail_opts;
1740 }
1741
1742 qemu_opts_del(opts);
1743 return 0;
1744
1745 fail_opts:
1746 qemu_opts_del(opts);
1747 return ret;
1748 }
1749
1750 static QDict *parse_json_filename(const char *filename, Error **errp)
1751 {
1752 QObject *options_obj;
1753 QDict *options;
1754 int ret;
1755
1756 ret = strstart(filename, "json:", &filename);
1757 assert(ret);
1758
1759 options_obj = qobject_from_json(filename, errp);
1760 if (!options_obj) {
1761 error_prepend(errp, "Could not parse the JSON options: ");
1762 return NULL;
1763 }
1764
1765 options = qobject_to(QDict, options_obj);
1766 if (!options) {
1767 qobject_unref(options_obj);
1768 error_setg(errp, "Invalid JSON object given");
1769 return NULL;
1770 }
1771
1772 qdict_flatten(options);
1773
1774 return options;
1775 }
1776
1777 static void parse_json_protocol(QDict *options, const char **pfilename,
1778 Error **errp)
1779 {
1780 QDict *json_options;
1781 Error *local_err = NULL;
1782
1783 /* Parse json: pseudo-protocol */
1784 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1785 return;
1786 }
1787
1788 json_options = parse_json_filename(*pfilename, &local_err);
1789 if (local_err) {
1790 error_propagate(errp, local_err);
1791 return;
1792 }
1793
1794 /* Options given in the filename have lower priority than options
1795 * specified directly */
1796 qdict_join(options, json_options, false);
1797 qobject_unref(json_options);
1798 *pfilename = NULL;
1799 }
1800
1801 /*
1802 * Fills in default options for opening images and converts the legacy
1803 * filename/flags pair to option QDict entries.
1804 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1805 * block driver has been specified explicitly.
1806 */
1807 static int bdrv_fill_options(QDict **options, const char *filename,
1808 int *flags, Error **errp)
1809 {
1810 const char *drvname;
1811 bool protocol = *flags & BDRV_O_PROTOCOL;
1812 bool parse_filename = false;
1813 BlockDriver *drv = NULL;
1814 Error *local_err = NULL;
1815
1816 /*
1817 * Caution: while qdict_get_try_str() is fine, getting non-string
1818 * types would require more care. When @options come from
1819 * -blockdev or blockdev_add, its members are typed according to
1820 * the QAPI schema, but when they come from -drive, they're all
1821 * QString.
1822 */
1823 drvname = qdict_get_try_str(*options, "driver");
1824 if (drvname) {
1825 drv = bdrv_find_format(drvname);
1826 if (!drv) {
1827 error_setg(errp, "Unknown driver '%s'", drvname);
1828 return -ENOENT;
1829 }
1830 /* If the user has explicitly specified the driver, this choice should
1831 * override the BDRV_O_PROTOCOL flag */
1832 protocol = drv->bdrv_file_open;
1833 }
1834
1835 if (protocol) {
1836 *flags |= BDRV_O_PROTOCOL;
1837 } else {
1838 *flags &= ~BDRV_O_PROTOCOL;
1839 }
1840
1841 /* Translate cache options from flags into options */
1842 update_options_from_flags(*options, *flags);
1843
1844 /* Fetch the file name from the options QDict if necessary */
1845 if (protocol && filename) {
1846 if (!qdict_haskey(*options, "filename")) {
1847 qdict_put_str(*options, "filename", filename);
1848 parse_filename = true;
1849 } else {
1850 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1851 "the same time");
1852 return -EINVAL;
1853 }
1854 }
1855
1856 /* Find the right block driver */
1857 /* See cautionary note on accessing @options above */
1858 filename = qdict_get_try_str(*options, "filename");
1859
1860 if (!drvname && protocol) {
1861 if (filename) {
1862 drv = bdrv_find_protocol(filename, parse_filename, errp);
1863 if (!drv) {
1864 return -EINVAL;
1865 }
1866
1867 drvname = drv->format_name;
1868 qdict_put_str(*options, "driver", drvname);
1869 } else {
1870 error_setg(errp, "Must specify either driver or file");
1871 return -EINVAL;
1872 }
1873 }
1874
1875 assert(drv || !protocol);
1876
1877 /* Driver-specific filename parsing */
1878 if (drv && drv->bdrv_parse_filename && parse_filename) {
1879 drv->bdrv_parse_filename(filename, *options, &local_err);
1880 if (local_err) {
1881 error_propagate(errp, local_err);
1882 return -EINVAL;
1883 }
1884
1885 if (!drv->bdrv_needs_filename) {
1886 qdict_del(*options, "filename");
1887 }
1888 }
1889
1890 return 0;
1891 }
1892
1893 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1894 uint64_t perm, uint64_t shared,
1895 GSList *ignore_children,
1896 bool *tighten_restrictions, Error **errp);
1897 static void bdrv_child_abort_perm_update(BdrvChild *c);
1898 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1899
1900 typedef struct BlockReopenQueueEntry {
1901 bool prepared;
1902 bool perms_checked;
1903 BDRVReopenState state;
1904 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
1905 } BlockReopenQueueEntry;
1906
1907 /*
1908 * Return the flags that @bs will have after the reopens in @q have
1909 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1910 * return the current flags.
1911 */
1912 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1913 {
1914 BlockReopenQueueEntry *entry;
1915
1916 if (q != NULL) {
1917 QTAILQ_FOREACH(entry, q, entry) {
1918 if (entry->state.bs == bs) {
1919 return entry->state.flags;
1920 }
1921 }
1922 }
1923
1924 return bs->open_flags;
1925 }
1926
1927 /* Returns whether the image file can be written to after the reopen queue @q
1928 * has been successfully applied, or right now if @q is NULL. */
1929 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1930 BlockReopenQueue *q)
1931 {
1932 int flags = bdrv_reopen_get_flags(q, bs);
1933
1934 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1935 }
1936
1937 /*
1938 * Return whether the BDS can be written to. This is not necessarily
1939 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1940 * be written to but do not count as read-only images.
1941 */
1942 bool bdrv_is_writable(BlockDriverState *bs)
1943 {
1944 return bdrv_is_writable_after_reopen(bs, NULL);
1945 }
1946
1947 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1948 BdrvChild *c, BdrvChildRole role,
1949 BlockReopenQueue *reopen_queue,
1950 uint64_t parent_perm, uint64_t parent_shared,
1951 uint64_t *nperm, uint64_t *nshared)
1952 {
1953 assert(bs->drv && bs->drv->bdrv_child_perm);
1954 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1955 parent_perm, parent_shared,
1956 nperm, nshared);
1957 /* TODO Take force_share from reopen_queue */
1958 if (child_bs && child_bs->force_share) {
1959 *nshared = BLK_PERM_ALL;
1960 }
1961 }
1962
1963 /*
1964 * Check whether permissions on this node can be changed in a way that
1965 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1966 * permissions of all its parents. This involves checking whether all necessary
1967 * permission changes to child nodes can be performed.
1968 *
1969 * Will set *tighten_restrictions to true if and only if new permissions have to
1970 * be taken or currently shared permissions are to be unshared. Otherwise,
1971 * errors are not fatal as long as the caller accepts that the restrictions
1972 * remain tighter than they need to be. The caller still has to abort the
1973 * transaction.
1974 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1975 * encounter fatal errors even though no restrictions are to be tightened. For
1976 * example, changing a node from RW to RO will fail if the WRITE permission is
1977 * to be kept.
1978 *
1979 * A call to this function must always be followed by a call to bdrv_set_perm()
1980 * or bdrv_abort_perm_update().
1981 */
1982 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1983 uint64_t cumulative_perms,
1984 uint64_t cumulative_shared_perms,
1985 GSList *ignore_children,
1986 bool *tighten_restrictions, Error **errp)
1987 {
1988 BlockDriver *drv = bs->drv;
1989 BdrvChild *c;
1990 int ret;
1991
1992 assert(!q || !tighten_restrictions);
1993
1994 if (tighten_restrictions) {
1995 uint64_t current_perms, current_shared;
1996 uint64_t added_perms, removed_shared_perms;
1997
1998 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
1999
2000 added_perms = cumulative_perms & ~current_perms;
2001 removed_shared_perms = current_shared & ~cumulative_shared_perms;
2002
2003 *tighten_restrictions = added_perms || removed_shared_perms;
2004 }
2005
2006 /* Write permissions never work with read-only images */
2007 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2008 !bdrv_is_writable_after_reopen(bs, q))
2009 {
2010 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2011 error_setg(errp, "Block node is read-only");
2012 } else {
2013 uint64_t current_perms, current_shared;
2014 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
2015 if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
2016 error_setg(errp, "Cannot make block node read-only, there is "
2017 "a writer on it");
2018 } else {
2019 error_setg(errp, "Cannot make block node read-only and create "
2020 "a writer on it");
2021 }
2022 }
2023
2024 return -EPERM;
2025 }
2026
2027 /* Check this node */
2028 if (!drv) {
2029 return 0;
2030 }
2031
2032 if (drv->bdrv_check_perm) {
2033 return drv->bdrv_check_perm(bs, cumulative_perms,
2034 cumulative_shared_perms, errp);
2035 }
2036
2037 /* Drivers that never have children can omit .bdrv_child_perm() */
2038 if (!drv->bdrv_child_perm) {
2039 assert(QLIST_EMPTY(&bs->children));
2040 return 0;
2041 }
2042
2043 /* Check all children */
2044 QLIST_FOREACH(c, &bs->children, next) {
2045 uint64_t cur_perm, cur_shared;
2046 bool child_tighten_restr;
2047
2048 bdrv_child_perm(bs, c->bs, c, c->role, q,
2049 cumulative_perms, cumulative_shared_perms,
2050 &cur_perm, &cur_shared);
2051 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children,
2052 tighten_restrictions ? &child_tighten_restr
2053 : NULL,
2054 errp);
2055 if (tighten_restrictions) {
2056 *tighten_restrictions |= child_tighten_restr;
2057 }
2058 if (ret < 0) {
2059 return ret;
2060 }
2061 }
2062
2063 return 0;
2064 }
2065
2066 /*
2067 * Notifies drivers that after a previous bdrv_check_perm() call, the
2068 * permission update is not performed and any preparations made for it (e.g.
2069 * taken file locks) need to be undone.
2070 *
2071 * This function recursively notifies all child nodes.
2072 */
2073 static void bdrv_abort_perm_update(BlockDriverState *bs)
2074 {
2075 BlockDriver *drv = bs->drv;
2076 BdrvChild *c;
2077
2078 if (!drv) {
2079 return;
2080 }
2081
2082 if (drv->bdrv_abort_perm_update) {
2083 drv->bdrv_abort_perm_update(bs);
2084 }
2085
2086 QLIST_FOREACH(c, &bs->children, next) {
2087 bdrv_child_abort_perm_update(c);
2088 }
2089 }
2090
2091 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
2092 uint64_t cumulative_shared_perms)
2093 {
2094 BlockDriver *drv = bs->drv;
2095 BdrvChild *c;
2096
2097 if (!drv) {
2098 return;
2099 }
2100
2101 /* Update this node */
2102 if (drv->bdrv_set_perm) {
2103 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2104 }
2105
2106 /* Drivers that never have children can omit .bdrv_child_perm() */
2107 if (!drv->bdrv_child_perm) {
2108 assert(QLIST_EMPTY(&bs->children));
2109 return;
2110 }
2111
2112 /* Update all children */
2113 QLIST_FOREACH(c, &bs->children, next) {
2114 uint64_t cur_perm, cur_shared;
2115 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2116 cumulative_perms, cumulative_shared_perms,
2117 &cur_perm, &cur_shared);
2118 bdrv_child_set_perm(c, cur_perm, cur_shared);
2119 }
2120 }
2121
2122 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2123 uint64_t *shared_perm)
2124 {
2125 BdrvChild *c;
2126 uint64_t cumulative_perms = 0;
2127 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2128
2129 QLIST_FOREACH(c, &bs->parents, next_parent) {
2130 cumulative_perms |= c->perm;
2131 cumulative_shared_perms &= c->shared_perm;
2132 }
2133
2134 *perm = cumulative_perms;
2135 *shared_perm = cumulative_shared_perms;
2136 }
2137
2138 static char *bdrv_child_user_desc(BdrvChild *c)
2139 {
2140 if (c->klass->get_parent_desc) {
2141 return c->klass->get_parent_desc(c);
2142 }
2143
2144 return g_strdup("another user");
2145 }
2146
2147 char *bdrv_perm_names(uint64_t perm)
2148 {
2149 struct perm_name {
2150 uint64_t perm;
2151 const char *name;
2152 } permissions[] = {
2153 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2154 { BLK_PERM_WRITE, "write" },
2155 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2156 { BLK_PERM_RESIZE, "resize" },
2157 { BLK_PERM_GRAPH_MOD, "change children" },
2158 { 0, NULL }
2159 };
2160
2161 GString *result = g_string_sized_new(30);
2162 struct perm_name *p;
2163
2164 for (p = permissions; p->name; p++) {
2165 if (perm & p->perm) {
2166 if (result->len > 0) {
2167 g_string_append(result, ", ");
2168 }
2169 g_string_append(result, p->name);
2170 }
2171 }
2172
2173 return g_string_free(result, FALSE);
2174 }
2175
2176 /*
2177 * Checks whether a new reference to @bs can be added if the new user requires
2178 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2179 * set, the BdrvChild objects in this list are ignored in the calculations;
2180 * this allows checking permission updates for an existing reference.
2181 *
2182 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2183 *
2184 * Needs to be followed by a call to either bdrv_set_perm() or
2185 * bdrv_abort_perm_update(). */
2186 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
2187 uint64_t new_used_perm,
2188 uint64_t new_shared_perm,
2189 GSList *ignore_children,
2190 bool *tighten_restrictions,
2191 Error **errp)
2192 {
2193 BdrvChild *c;
2194 uint64_t cumulative_perms = new_used_perm;
2195 uint64_t cumulative_shared_perms = new_shared_perm;
2196
2197 assert(!q || !tighten_restrictions);
2198
2199 /* There is no reason why anyone couldn't tolerate write_unchanged */
2200 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
2201
2202 QLIST_FOREACH(c, &bs->parents, next_parent) {
2203 if (g_slist_find(ignore_children, c)) {
2204 continue;
2205 }
2206
2207 if ((new_used_perm & c->shared_perm) != new_used_perm) {
2208 char *user = bdrv_child_user_desc(c);
2209 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
2210
2211 if (tighten_restrictions) {
2212 *tighten_restrictions = true;
2213 }
2214
2215 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
2216 "allow '%s' on %s",
2217 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2218 g_free(user);
2219 g_free(perm_names);
2220 return -EPERM;
2221 }
2222
2223 if ((c->perm & new_shared_perm) != c->perm) {
2224 char *user = bdrv_child_user_desc(c);
2225 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
2226
2227 if (tighten_restrictions) {
2228 *tighten_restrictions = true;
2229 }
2230
2231 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
2232 "'%s' on %s",
2233 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2234 g_free(user);
2235 g_free(perm_names);
2236 return -EPERM;
2237 }
2238
2239 cumulative_perms |= c->perm;
2240 cumulative_shared_perms &= c->shared_perm;
2241 }
2242
2243 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
2244 ignore_children, tighten_restrictions, errp);
2245 }
2246
2247 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2248 * bdrv_child_abort_perm_update(). */
2249 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
2250 uint64_t perm, uint64_t shared,
2251 GSList *ignore_children,
2252 bool *tighten_restrictions, Error **errp)
2253 {
2254 int ret;
2255
2256 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
2257 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children,
2258 tighten_restrictions, errp);
2259 g_slist_free(ignore_children);
2260
2261 if (ret < 0) {
2262 return ret;
2263 }
2264
2265 if (!c->has_backup_perm) {
2266 c->has_backup_perm = true;
2267 c->backup_perm = c->perm;
2268 c->backup_shared_perm = c->shared_perm;
2269 }
2270 /*
2271 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2272 * same child twice during check_perm procedure
2273 */
2274
2275 c->perm = perm;
2276 c->shared_perm = shared;
2277
2278 return 0;
2279 }
2280
2281 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
2282 {
2283 uint64_t cumulative_perms, cumulative_shared_perms;
2284
2285 c->has_backup_perm = false;
2286
2287 c->perm = perm;
2288 c->shared_perm = shared;
2289
2290 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
2291 &cumulative_shared_perms);
2292 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
2293 }
2294
2295 static void bdrv_child_abort_perm_update(BdrvChild *c)
2296 {
2297 if (c->has_backup_perm) {
2298 c->perm = c->backup_perm;
2299 c->shared_perm = c->backup_shared_perm;
2300 c->has_backup_perm = false;
2301 }
2302
2303 bdrv_abort_perm_update(c->bs);
2304 }
2305
2306 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2307 Error **errp)
2308 {
2309 Error *local_err = NULL;
2310 int ret;
2311 bool tighten_restrictions;
2312
2313 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL,
2314 &tighten_restrictions, &local_err);
2315 if (ret < 0) {
2316 bdrv_child_abort_perm_update(c);
2317 if (tighten_restrictions) {
2318 error_propagate(errp, local_err);
2319 } else {
2320 /*
2321 * Our caller may intend to only loosen restrictions and
2322 * does not expect this function to fail. Errors are not
2323 * fatal in such a case, so we can just hide them from our
2324 * caller.
2325 */
2326 error_free(local_err);
2327 ret = 0;
2328 }
2329 return ret;
2330 }
2331
2332 bdrv_child_set_perm(c, perm, shared);
2333
2334 return 0;
2335 }
2336
2337 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2338 {
2339 uint64_t parent_perms, parent_shared;
2340 uint64_t perms, shared;
2341
2342 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2343 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2344 parent_perms, parent_shared, &perms, &shared);
2345
2346 return bdrv_child_try_set_perm(c, perms, shared, errp);
2347 }
2348
2349 /*
2350 * Default implementation for .bdrv_child_perm() for block filters:
2351 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2352 * filtered child.
2353 */
2354 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2355 BdrvChildRole role,
2356 BlockReopenQueue *reopen_queue,
2357 uint64_t perm, uint64_t shared,
2358 uint64_t *nperm, uint64_t *nshared)
2359 {
2360 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2361 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2362 }
2363
2364 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2365 BdrvChildRole role,
2366 BlockReopenQueue *reopen_queue,
2367 uint64_t perm, uint64_t shared,
2368 uint64_t *nperm, uint64_t *nshared)
2369 {
2370 assert(role & BDRV_CHILD_COW);
2371
2372 /*
2373 * We want consistent read from backing files if the parent needs it.
2374 * No other operations are performed on backing files.
2375 */
2376 perm &= BLK_PERM_CONSISTENT_READ;
2377
2378 /*
2379 * If the parent can deal with changing data, we're okay with a
2380 * writable and resizable backing file.
2381 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2382 */
2383 if (shared & BLK_PERM_WRITE) {
2384 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2385 } else {
2386 shared = 0;
2387 }
2388
2389 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2390 BLK_PERM_WRITE_UNCHANGED;
2391
2392 if (bs->open_flags & BDRV_O_INACTIVE) {
2393 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2394 }
2395
2396 *nperm = perm;
2397 *nshared = shared;
2398 }
2399
2400 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2401 BdrvChildRole role,
2402 BlockReopenQueue *reopen_queue,
2403 uint64_t perm, uint64_t shared,
2404 uint64_t *nperm, uint64_t *nshared)
2405 {
2406 int flags;
2407
2408 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2409
2410 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2411
2412 /*
2413 * Apart from the modifications below, the same permissions are
2414 * forwarded and left alone as for filters
2415 */
2416 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2417 perm, shared, &perm, &shared);
2418
2419 if (role & BDRV_CHILD_METADATA) {
2420 /* Format drivers may touch metadata even if the guest doesn't write */
2421 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2422 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2423 }
2424
2425 /*
2426 * bs->file always needs to be consistent because of the
2427 * metadata. We can never allow other users to resize or write
2428 * to it.
2429 */
2430 if (!(flags & BDRV_O_NO_IO)) {
2431 perm |= BLK_PERM_CONSISTENT_READ;
2432 }
2433 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2434 }
2435
2436 if (role & BDRV_CHILD_DATA) {
2437 /*
2438 * Technically, everything in this block is a subset of the
2439 * BDRV_CHILD_METADATA path taken above, and so this could
2440 * be an "else if" branch. However, that is not obvious, and
2441 * this function is not performance critical, therefore we let
2442 * this be an independent "if".
2443 */
2444
2445 /*
2446 * We cannot allow other users to resize the file because the
2447 * format driver might have some assumptions about the size
2448 * (e.g. because it is stored in metadata, or because the file
2449 * is split into fixed-size data files).
2450 */
2451 shared &= ~BLK_PERM_RESIZE;
2452
2453 /*
2454 * WRITE_UNCHANGED often cannot be performed as such on the
2455 * data file. For example, the qcow2 driver may still need to
2456 * write copied clusters on copy-on-read.
2457 */
2458 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2459 perm |= BLK_PERM_WRITE;
2460 }
2461
2462 /*
2463 * If the data file is written to, the format driver may
2464 * expect to be able to resize it by writing beyond the EOF.
2465 */
2466 if (perm & BLK_PERM_WRITE) {
2467 perm |= BLK_PERM_RESIZE;
2468 }
2469 }
2470
2471 if (bs->open_flags & BDRV_O_INACTIVE) {
2472 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2473 }
2474
2475 *nperm = perm;
2476 *nshared = shared;
2477 }
2478
2479 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2480 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2481 uint64_t perm, uint64_t shared,
2482 uint64_t *nperm, uint64_t *nshared)
2483 {
2484 if (role & BDRV_CHILD_FILTERED) {
2485 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2486 BDRV_CHILD_COW)));
2487 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2488 perm, shared, nperm, nshared);
2489 } else if (role & BDRV_CHILD_COW) {
2490 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2491 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2492 perm, shared, nperm, nshared);
2493 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2494 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2495 perm, shared, nperm, nshared);
2496 } else {
2497 g_assert_not_reached();
2498 }
2499 }
2500
2501 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2502 {
2503 static const uint64_t permissions[] = {
2504 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2505 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2506 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2507 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2508 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2509 };
2510
2511 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2512 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2513
2514 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2515
2516 return permissions[qapi_perm];
2517 }
2518
2519 static void bdrv_replace_child_noperm(BdrvChild *child,
2520 BlockDriverState *new_bs)
2521 {
2522 BlockDriverState *old_bs = child->bs;
2523 int new_bs_quiesce_counter;
2524 int drain_saldo;
2525
2526 assert(!child->frozen);
2527
2528 if (old_bs && new_bs) {
2529 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2530 }
2531
2532 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2533 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2534
2535 /*
2536 * If the new child node is drained but the old one was not, flush
2537 * all outstanding requests to the old child node.
2538 */
2539 while (drain_saldo > 0 && child->klass->drained_begin) {
2540 bdrv_parent_drained_begin_single(child, true);
2541 drain_saldo--;
2542 }
2543
2544 if (old_bs) {
2545 /* Detach first so that the recursive drain sections coming from @child
2546 * are already gone and we only end the drain sections that came from
2547 * elsewhere. */
2548 if (child->klass->detach) {
2549 child->klass->detach(child);
2550 }
2551 QLIST_REMOVE(child, next_parent);
2552 }
2553
2554 child->bs = new_bs;
2555
2556 if (new_bs) {
2557 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2558
2559 /*
2560 * Detaching the old node may have led to the new node's
2561 * quiesce_counter having been decreased. Not a problem, we
2562 * just need to recognize this here and then invoke
2563 * drained_end appropriately more often.
2564 */
2565 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2566 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2567
2568 /* Attach only after starting new drained sections, so that recursive
2569 * drain sections coming from @child don't get an extra .drained_begin
2570 * callback. */
2571 if (child->klass->attach) {
2572 child->klass->attach(child);
2573 }
2574 }
2575
2576 /*
2577 * If the old child node was drained but the new one is not, allow
2578 * requests to come in only after the new node has been attached.
2579 */
2580 while (drain_saldo < 0 && child->klass->drained_end) {
2581 bdrv_parent_drained_end_single(child);
2582 drain_saldo++;
2583 }
2584 }
2585
2586 /*
2587 * Updates @child to change its reference to point to @new_bs, including
2588 * checking and applying the necessary permisson updates both to the old node
2589 * and to @new_bs.
2590 *
2591 * NULL is passed as @new_bs for removing the reference before freeing @child.
2592 *
2593 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2594 * function uses bdrv_set_perm() to update the permissions according to the new
2595 * reference that @new_bs gets.
2596 */
2597 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2598 {
2599 BlockDriverState *old_bs = child->bs;
2600 uint64_t perm, shared_perm;
2601
2602 bdrv_replace_child_noperm(child, new_bs);
2603
2604 /*
2605 * Start with the new node's permissions. If @new_bs is a (direct
2606 * or indirect) child of @old_bs, we must complete the permission
2607 * update on @new_bs before we loosen the restrictions on @old_bs.
2608 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2609 * updating the permissions of @new_bs, and thus not purely loosen
2610 * restrictions.
2611 */
2612 if (new_bs) {
2613 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2614 bdrv_set_perm(new_bs, perm, shared_perm);
2615 }
2616
2617 if (old_bs) {
2618 /* Update permissions for old node. This is guaranteed to succeed
2619 * because we're just taking a parent away, so we're loosening
2620 * restrictions. */
2621 bool tighten_restrictions;
2622 int ret;
2623
2624 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2625 ret = bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL,
2626 &tighten_restrictions, NULL);
2627 assert(tighten_restrictions == false);
2628 if (ret < 0) {
2629 /* We only tried to loosen restrictions, so errors are not fatal */
2630 bdrv_abort_perm_update(old_bs);
2631 } else {
2632 bdrv_set_perm(old_bs, perm, shared_perm);
2633 }
2634
2635 /* When the parent requiring a non-default AioContext is removed, the
2636 * node moves back to the main AioContext */
2637 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2638 }
2639 }
2640
2641 /*
2642 * This function steals the reference to child_bs from the caller.
2643 * That reference is later dropped by bdrv_root_unref_child().
2644 *
2645 * On failure NULL is returned, errp is set and the reference to
2646 * child_bs is also dropped.
2647 *
2648 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2649 * (unless @child_bs is already in @ctx).
2650 */
2651 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2652 const char *child_name,
2653 const BdrvChildClass *child_class,
2654 BdrvChildRole child_role,
2655 AioContext *ctx,
2656 uint64_t perm, uint64_t shared_perm,
2657 void *opaque, Error **errp)
2658 {
2659 BdrvChild *child;
2660 Error *local_err = NULL;
2661 int ret;
2662
2663 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, NULL,
2664 errp);
2665 if (ret < 0) {
2666 bdrv_abort_perm_update(child_bs);
2667 bdrv_unref(child_bs);
2668 return NULL;
2669 }
2670
2671 child = g_new(BdrvChild, 1);
2672 *child = (BdrvChild) {
2673 .bs = NULL,
2674 .name = g_strdup(child_name),
2675 .klass = child_class,
2676 .role = child_role,
2677 .perm = perm,
2678 .shared_perm = shared_perm,
2679 .opaque = opaque,
2680 };
2681
2682 /* If the AioContexts don't match, first try to move the subtree of
2683 * child_bs into the AioContext of the new parent. If this doesn't work,
2684 * try moving the parent into the AioContext of child_bs instead. */
2685 if (bdrv_get_aio_context(child_bs) != ctx) {
2686 ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err);
2687 if (ret < 0 && child_class->can_set_aio_ctx) {
2688 GSList *ignore = g_slist_prepend(NULL, child);
2689 ctx = bdrv_get_aio_context(child_bs);
2690 if (child_class->can_set_aio_ctx(child, ctx, &ignore, NULL)) {
2691 error_free(local_err);
2692 ret = 0;
2693 g_slist_free(ignore);
2694 ignore = g_slist_prepend(NULL, child);
2695 child_class->set_aio_ctx(child, ctx, &ignore);
2696 }
2697 g_slist_free(ignore);
2698 }
2699 if (ret < 0) {
2700 error_propagate(errp, local_err);
2701 g_free(child);
2702 bdrv_abort_perm_update(child_bs);
2703 bdrv_unref(child_bs);
2704 return NULL;
2705 }
2706 }
2707
2708 /* This performs the matching bdrv_set_perm() for the above check. */
2709 bdrv_replace_child(child, child_bs);
2710
2711 return child;
2712 }
2713
2714 /*
2715 * This function transfers the reference to child_bs from the caller
2716 * to parent_bs. That reference is later dropped by parent_bs on
2717 * bdrv_close() or if someone calls bdrv_unref_child().
2718 *
2719 * On failure NULL is returned, errp is set and the reference to
2720 * child_bs is also dropped.
2721 *
2722 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2723 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2724 */
2725 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2726 BlockDriverState *child_bs,
2727 const char *child_name,
2728 const BdrvChildClass *child_class,
2729 BdrvChildRole child_role,
2730 Error **errp)
2731 {
2732 BdrvChild *child;
2733 uint64_t perm, shared_perm;
2734
2735 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2736
2737 assert(parent_bs->drv);
2738 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2739 perm, shared_perm, &perm, &shared_perm);
2740
2741 child = bdrv_root_attach_child(child_bs, child_name, child_class,
2742 child_role, bdrv_get_aio_context(parent_bs),
2743 perm, shared_perm, parent_bs, errp);
2744 if (child == NULL) {
2745 return NULL;
2746 }
2747
2748 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2749 return child;
2750 }
2751
2752 static void bdrv_detach_child(BdrvChild *child)
2753 {
2754 QLIST_SAFE_REMOVE(child, next);
2755
2756 bdrv_replace_child(child, NULL);
2757
2758 g_free(child->name);
2759 g_free(child);
2760 }
2761
2762 void bdrv_root_unref_child(BdrvChild *child)
2763 {
2764 BlockDriverState *child_bs;
2765
2766 child_bs = child->bs;
2767 bdrv_detach_child(child);
2768 bdrv_unref(child_bs);
2769 }
2770
2771 /**
2772 * Clear all inherits_from pointers from children and grandchildren of
2773 * @root that point to @root, where necessary.
2774 */
2775 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child)
2776 {
2777 BdrvChild *c;
2778
2779 if (child->bs->inherits_from == root) {
2780 /*
2781 * Remove inherits_from only when the last reference between root and
2782 * child->bs goes away.
2783 */
2784 QLIST_FOREACH(c, &root->children, next) {
2785 if (c != child && c->bs == child->bs) {
2786 break;
2787 }
2788 }
2789 if (c == NULL) {
2790 child->bs->inherits_from = NULL;
2791 }
2792 }
2793
2794 QLIST_FOREACH(c, &child->bs->children, next) {
2795 bdrv_unset_inherits_from(root, c);
2796 }
2797 }
2798
2799 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2800 {
2801 if (child == NULL) {
2802 return;
2803 }
2804
2805 bdrv_unset_inherits_from(parent, child);
2806 bdrv_root_unref_child(child);
2807 }
2808
2809
2810 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2811 {
2812 BdrvChild *c;
2813 QLIST_FOREACH(c, &bs->parents, next_parent) {
2814 if (c->klass->change_media) {
2815 c->klass->change_media(c, load);
2816 }
2817 }
2818 }
2819
2820 /* Return true if you can reach parent going through child->inherits_from
2821 * recursively. If parent or child are NULL, return false */
2822 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2823 BlockDriverState *parent)
2824 {
2825 while (child && child != parent) {
2826 child = child->inherits_from;
2827 }
2828
2829 return child != NULL;
2830 }
2831
2832 /*
2833 * Return the BdrvChildRole for @bs's backing child. bs->backing is
2834 * mostly used for COW backing children (role = COW), but also for
2835 * filtered children (role = FILTERED | PRIMARY).
2836 */
2837 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
2838 {
2839 if (bs->drv && bs->drv->is_filter) {
2840 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
2841 } else {
2842 return BDRV_CHILD_COW;
2843 }
2844 }
2845
2846 /*
2847 * Sets the backing file link of a BDS. A new reference is created; callers
2848 * which don't need their own reference any more must call bdrv_unref().
2849 */
2850 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2851 Error **errp)
2852 {
2853 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2854 bdrv_inherits_from_recursive(backing_hd, bs);
2855
2856 if (bdrv_is_backing_chain_frozen(bs, backing_bs(bs), errp)) {
2857 return;
2858 }
2859
2860 if (backing_hd) {
2861 bdrv_ref(backing_hd);
2862 }
2863
2864 if (bs->backing) {
2865 bdrv_unref_child(bs, bs->backing);
2866 bs->backing = NULL;
2867 }
2868
2869 if (!backing_hd) {
2870 goto out;
2871 }
2872
2873 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_of_bds,
2874 bdrv_backing_role(bs), errp);
2875 /* If backing_hd was already part of bs's backing chain, and
2876 * inherits_from pointed recursively to bs then let's update it to
2877 * point directly to bs (else it will become NULL). */
2878 if (bs->backing && update_inherits_from) {
2879 backing_hd->inherits_from = bs;
2880 }
2881
2882 out:
2883 bdrv_refresh_limits(bs, NULL);
2884 }
2885
2886 /*
2887 * Opens the backing file for a BlockDriverState if not yet open
2888 *
2889 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2890 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2891 * itself, all options starting with "${bdref_key}." are considered part of the
2892 * BlockdevRef.
2893 *
2894 * TODO Can this be unified with bdrv_open_image()?
2895 */
2896 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2897 const char *bdref_key, Error **errp)
2898 {
2899 char *backing_filename = NULL;
2900 char *bdref_key_dot;
2901 const char *reference = NULL;
2902 int ret = 0;
2903 bool implicit_backing = false;
2904 BlockDriverState *backing_hd;
2905 QDict *options;
2906 QDict *tmp_parent_options = NULL;
2907 Error *local_err = NULL;
2908
2909 if (bs->backing != NULL) {
2910 goto free_exit;
2911 }
2912
2913 /* NULL means an empty set of options */
2914 if (parent_options == NULL) {
2915 tmp_parent_options = qdict_new();
2916 parent_options = tmp_parent_options;
2917 }
2918
2919 bs->open_flags &= ~BDRV_O_NO_BACKING;
2920
2921 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2922 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2923 g_free(bdref_key_dot);
2924
2925 /*
2926 * Caution: while qdict_get_try_str() is fine, getting non-string
2927 * types would require more care. When @parent_options come from
2928 * -blockdev or blockdev_add, its members are typed according to
2929 * the QAPI schema, but when they come from -drive, they're all
2930 * QString.
2931 */
2932 reference = qdict_get_try_str(parent_options, bdref_key);
2933 if (reference || qdict_haskey(options, "file.filename")) {
2934 /* keep backing_filename NULL */
2935 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2936 qobject_unref(options);
2937 goto free_exit;
2938 } else {
2939 if (qdict_size(options) == 0) {
2940 /* If the user specifies options that do not modify the
2941 * backing file's behavior, we might still consider it the
2942 * implicit backing file. But it's easier this way, and
2943 * just specifying some of the backing BDS's options is
2944 * only possible with -drive anyway (otherwise the QAPI
2945 * schema forces the user to specify everything). */
2946 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2947 }
2948
2949 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2950 if (local_err) {
2951 ret = -EINVAL;
2952 error_propagate(errp, local_err);
2953 qobject_unref(options);
2954 goto free_exit;
2955 }
2956 }
2957
2958 if (!bs->drv || !bs->drv->supports_backing) {
2959 ret = -EINVAL;
2960 error_setg(errp, "Driver doesn't support backing files");
2961 qobject_unref(options);
2962 goto free_exit;
2963 }
2964
2965 if (!reference &&
2966 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2967 qdict_put_str(options, "driver", bs->backing_format);
2968 }
2969
2970 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2971 &child_of_bds, bdrv_backing_role(bs), errp);
2972 if (!backing_hd) {
2973 bs->open_flags |= BDRV_O_NO_BACKING;
2974 error_prepend(errp, "Could not open backing file: ");
2975 ret = -EINVAL;
2976 goto free_exit;
2977 }
2978
2979 if (implicit_backing) {
2980 bdrv_refresh_filename(backing_hd);
2981 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2982 backing_hd->filename);
2983 }
2984
2985 /* Hook up the backing file link; drop our reference, bs owns the
2986 * backing_hd reference now */
2987 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2988 bdrv_unref(backing_hd);
2989 if (local_err) {
2990 error_propagate(errp, local_err);
2991 ret = -EINVAL;
2992 goto free_exit;
2993 }
2994
2995 qdict_del(parent_options, bdref_key);
2996
2997 free_exit:
2998 g_free(backing_filename);
2999 qobject_unref(tmp_parent_options);
3000 return ret;
3001 }
3002
3003 static BlockDriverState *
3004 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3005 BlockDriverState *parent, const BdrvChildClass *child_class,
3006 BdrvChildRole child_role, bool allow_none, Error **errp)
3007 {
3008 BlockDriverState *bs = NULL;
3009 QDict *image_options;
3010 char *bdref_key_dot;
3011 const char *reference;
3012
3013 assert(child_class != NULL);
3014
3015 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3016 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3017 g_free(bdref_key_dot);
3018
3019 /*
3020 * Caution: while qdict_get_try_str() is fine, getting non-string
3021 * types would require more care. When @options come from
3022 * -blockdev or blockdev_add, its members are typed according to
3023 * the QAPI schema, but when they come from -drive, they're all
3024 * QString.
3025 */
3026 reference = qdict_get_try_str(options, bdref_key);
3027 if (!filename && !reference && !qdict_size(image_options)) {
3028 if (!allow_none) {
3029 error_setg(errp, "A block device must be specified for \"%s\"",
3030 bdref_key);
3031 }
3032 qobject_unref(image_options);
3033 goto done;
3034 }
3035
3036 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3037 parent, child_class, child_role, errp);
3038 if (!bs) {
3039 goto done;
3040 }
3041
3042 done:
3043 qdict_del(options, bdref_key);
3044 return bs;
3045 }
3046
3047 /*
3048 * Opens a disk image whose options are given as BlockdevRef in another block
3049 * device's options.
3050 *
3051 * If allow_none is true, no image will be opened if filename is false and no
3052 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3053 *
3054 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3055 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3056 * itself, all options starting with "${bdref_key}." are considered part of the
3057 * BlockdevRef.
3058 *
3059 * The BlockdevRef will be removed from the options QDict.
3060 */
3061 BdrvChild *bdrv_open_child(const char *filename,
3062 QDict *options, const char *bdref_key,
3063 BlockDriverState *parent,
3064 const BdrvChildClass *child_class,
3065 BdrvChildRole child_role,
3066 bool allow_none, Error **errp)
3067 {
3068 BlockDriverState *bs;
3069
3070 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3071 child_role, allow_none, errp);
3072 if (bs == NULL) {
3073 return NULL;
3074 }
3075
3076 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3077 errp);
3078 }
3079
3080 /*
3081 * TODO Future callers may need to specify parent/child_class in order for
3082 * option inheritance to work. Existing callers use it for the root node.
3083 */
3084 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3085 {
3086 BlockDriverState *bs = NULL;
3087 QObject *obj = NULL;
3088 QDict *qdict = NULL;
3089 const char *reference = NULL;
3090 Visitor *v = NULL;
3091
3092 if (ref->type == QTYPE_QSTRING) {
3093 reference = ref->u.reference;
3094 } else {
3095 BlockdevOptions *options = &ref->u.definition;
3096 assert(ref->type == QTYPE_QDICT);
3097
3098 v = qobject_output_visitor_new(&obj);
3099 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3100 visit_complete(v, &obj);
3101
3102 qdict = qobject_to(QDict, obj);
3103 qdict_flatten(qdict);
3104
3105 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3106 * compatibility with other callers) rather than what we want as the
3107 * real defaults. Apply the defaults here instead. */
3108 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3109 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3110 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3111 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3112
3113 }
3114
3115 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3116 obj = NULL;
3117 qobject_unref(obj);
3118 visit_free(v);
3119 return bs;
3120 }
3121
3122 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3123 int flags,
3124 QDict *snapshot_options,
3125 Error **errp)
3126 {
3127 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3128 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3129 int64_t total_size;
3130 QemuOpts *opts = NULL;
3131 BlockDriverState *bs_snapshot = NULL;
3132 Error *local_err = NULL;
3133 int ret;
3134
3135 /* if snapshot, we create a temporary backing file and open it
3136 instead of opening 'filename' directly */
3137
3138 /* Get the required size from the image */
3139 total_size = bdrv_getlength(bs);
3140 if (total_size < 0) {
3141 error_setg_errno(errp, -total_size, "Could not get image size");
3142 goto out;
3143 }
3144
3145 /* Create the temporary image */
3146 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3147 if (ret < 0) {
3148 error_setg_errno(errp, -ret, "Could not get temporary filename");
3149 goto out;
3150 }
3151
3152 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3153 &error_abort);
3154 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3155 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3156 qemu_opts_del(opts);
3157 if (ret < 0) {
3158 error_prepend(errp, "Could not create temporary overlay '%s': ",
3159 tmp_filename);
3160 goto out;
3161 }
3162
3163 /* Prepare options QDict for the temporary file */
3164 qdict_put_str(snapshot_options, "file.driver", "file");
3165 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3166 qdict_put_str(snapshot_options, "driver", "qcow2");
3167
3168 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3169 snapshot_options = NULL;
3170 if (!bs_snapshot) {
3171 goto out;
3172 }
3173
3174 /* bdrv_append() consumes a strong reference to bs_snapshot
3175 * (i.e. it will call bdrv_unref() on it) even on error, so in
3176 * order to be able to return one, we have to increase
3177 * bs_snapshot's refcount here */
3178 bdrv_ref(bs_snapshot);
3179 bdrv_append(bs_snapshot, bs, &local_err);
3180 if (local_err) {
3181 error_propagate(errp, local_err);
3182 bs_snapshot = NULL;
3183 goto out;
3184 }
3185
3186 out:
3187 qobject_unref(snapshot_options);
3188 g_free(tmp_filename);
3189 return bs_snapshot;
3190 }
3191
3192 /*
3193 * Opens a disk image (raw, qcow2, vmdk, ...)
3194 *
3195 * options is a QDict of options to pass to the block drivers, or NULL for an
3196 * empty set of options. The reference to the QDict belongs to the block layer
3197 * after the call (even on failure), so if the caller intends to reuse the
3198 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3199 *
3200 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3201 * If it is not NULL, the referenced BDS will be reused.
3202 *
3203 * The reference parameter may be used to specify an existing block device which
3204 * should be opened. If specified, neither options nor a filename may be given,
3205 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3206 */
3207 static BlockDriverState *bdrv_open_inherit(const char *filename,
3208 const char *reference,
3209 QDict *options, int flags,
3210 BlockDriverState *parent,
3211 const BdrvChildClass *child_class,
3212 BdrvChildRole child_role,
3213 Error **errp)
3214 {
3215 int ret;
3216 BlockBackend *file = NULL;
3217 BlockDriverState *bs;
3218 BlockDriver *drv = NULL;
3219 BdrvChild *child;
3220 const char *drvname;
3221 const char *backing;
3222 Error *local_err = NULL;
3223 QDict *snapshot_options = NULL;
3224 int snapshot_flags = 0;
3225
3226 assert(!child_class || !flags);
3227 assert(!child_class == !parent);
3228
3229 if (reference) {
3230 bool options_non_empty = options ? qdict_size(options) : false;
3231 qobject_unref(options);
3232
3233 if (filename || options_non_empty) {
3234 error_setg(errp, "Cannot reference an existing block device with "
3235 "additional options or a new filename");
3236 return NULL;
3237 }
3238
3239 bs = bdrv_lookup_bs(reference, reference, errp);
3240 if (!bs) {
3241 return NULL;
3242 }
3243
3244 bdrv_ref(bs);
3245 return bs;
3246 }
3247
3248 bs = bdrv_new();
3249
3250 /* NULL means an empty set of options */
3251 if (options == NULL) {
3252 options = qdict_new();
3253 }
3254
3255 /* json: syntax counts as explicit options, as if in the QDict */
3256 parse_json_protocol(options, &filename, &local_err);
3257 if (local_err) {
3258 goto fail;
3259 }
3260
3261 bs->explicit_options = qdict_clone_shallow(options);
3262
3263 if (child_class) {
3264 bool parent_is_format;
3265
3266 if (parent->drv) {
3267 parent_is_format = parent->drv->is_format;
3268 } else {
3269 /*
3270 * parent->drv is not set yet because this node is opened for
3271 * (potential) format probing. That means that @parent is going
3272 * to be a format node.
3273 */
3274 parent_is_format = true;
3275 }
3276
3277 bs->inherits_from = parent;
3278 child_class->inherit_options(child_role, parent_is_format,
3279 &flags, options,
3280 parent->open_flags, parent->options);
3281 }
3282
3283 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3284 if (ret < 0) {
3285 goto fail;
3286 }
3287
3288 /*
3289 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3290 * Caution: getting a boolean member of @options requires care.
3291 * When @options come from -blockdev or blockdev_add, members are
3292 * typed according to the QAPI schema, but when they come from
3293 * -drive, they're all QString.
3294 */
3295 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3296 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3297 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3298 } else {
3299 flags &= ~BDRV_O_RDWR;
3300 }
3301
3302 if (flags & BDRV_O_SNAPSHOT) {
3303 snapshot_options = qdict_new();
3304 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3305 flags, options);
3306 /* Let bdrv_backing_options() override "read-only" */
3307 qdict_del(options, BDRV_OPT_READ_ONLY);
3308 bdrv_inherited_options(BDRV_CHILD_COW, true,
3309 &flags, options, flags, options);
3310 }
3311
3312 bs->open_flags = flags;
3313 bs->options = options;
3314 options = qdict_clone_shallow(options);
3315
3316 /* Find the right image format driver */
3317 /* See cautionary note on accessing @options above */
3318 drvname = qdict_get_try_str(options, "driver");
3319 if (drvname) {
3320 drv = bdrv_find_format(drvname);
3321 if (!drv) {
3322 error_setg(errp, "Unknown driver: '%s'", drvname);
3323 goto fail;
3324 }
3325 }
3326
3327 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3328
3329 /* See cautionary note on accessing @options above */
3330 backing = qdict_get_try_str(options, "backing");
3331 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3332 (backing && *backing == '\0'))
3333 {
3334 if (backing) {
3335 warn_report("Use of \"backing\": \"\" is deprecated; "
3336 "use \"backing\": null instead");
3337 }
3338 flags |= BDRV_O_NO_BACKING;
3339 qdict_del(bs->explicit_options, "backing");
3340 qdict_del(bs->options, "backing");
3341 qdict_del(options, "backing");
3342 }
3343
3344 /* Open image file without format layer. This BlockBackend is only used for
3345 * probing, the block drivers will do their own bdrv_open_child() for the
3346 * same BDS, which is why we put the node name back into options. */
3347 if ((flags & BDRV_O_PROTOCOL) == 0) {
3348 BlockDriverState *file_bs;
3349
3350 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3351 &child_of_bds, BDRV_CHILD_IMAGE,
3352 true, &local_err);
3353 if (local_err) {
3354 goto fail;
3355 }
3356 if (file_bs != NULL) {
3357 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3358 * looking at the header to guess the image format. This works even
3359 * in cases where a guest would not see a consistent state. */
3360 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3361 blk_insert_bs(file, file_bs, &local_err);
3362 bdrv_unref(file_bs);
3363 if (local_err) {
3364 goto fail;
3365 }
3366
3367 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3368 }
3369 }
3370
3371 /* Image format probing */
3372 bs->probed = !drv;
3373 if (!drv && file) {
3374 ret = find_image_format(file, filename, &drv, &local_err);
3375 if (ret < 0) {
3376 goto fail;
3377 }
3378 /*
3379 * This option update would logically belong in bdrv_fill_options(),
3380 * but we first need to open bs->file for the probing to work, while
3381 * opening bs->file already requires the (mostly) final set of options
3382 * so that cache mode etc. can be inherited.
3383 *
3384 * Adding the driver later is somewhat ugly, but it's not an option
3385 * that would ever be inherited, so it's correct. We just need to make
3386 * sure to update both bs->options (which has the full effective
3387 * options for bs) and options (which has file.* already removed).
3388 */
3389 qdict_put_str(bs->options, "driver", drv->format_name);
3390 qdict_put_str(options, "driver", drv->format_name);
3391 } else if (!drv) {
3392 error_setg(errp, "Must specify either driver or file");
3393 goto fail;
3394 }
3395
3396 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3397 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3398 /* file must be NULL if a protocol BDS is about to be created
3399 * (the inverse results in an error message from bdrv_open_common()) */
3400 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3401
3402 /* Open the image */
3403 ret = bdrv_open_common(bs, file, options, &local_err);
3404 if (ret < 0) {
3405 goto fail;
3406 }
3407
3408 if (file) {
3409 blk_unref(file);
3410 file = NULL;
3411 }
3412
3413 /* If there is a backing file, use it */
3414 if ((flags & BDRV_O_NO_BACKING) == 0) {
3415 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3416 if (ret < 0) {
3417 goto close_and_fail;
3418 }
3419 }
3420
3421 /* Remove all children options and references
3422 * from bs->options and bs->explicit_options */
3423 QLIST_FOREACH(child, &bs->children, next) {
3424 char *child_key_dot;
3425 child_key_dot = g_strdup_printf("%s.", child->name);
3426 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3427 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3428 qdict_del(bs->explicit_options, child->name);
3429 qdict_del(bs->options, child->name);
3430 g_free(child_key_dot);
3431 }
3432
3433 /* Check if any unknown options were used */
3434 if (qdict_size(options) != 0) {
3435 const QDictEntry *entry = qdict_first(options);
3436 if (flags & BDRV_O_PROTOCOL) {
3437 error_setg(errp, "Block protocol '%s' doesn't support the option "
3438 "'%s'", drv->format_name, entry->key);
3439 } else {
3440 error_setg(errp,
3441 "Block format '%s' does not support the option '%s'",
3442 drv->format_name, entry->key);
3443 }
3444
3445 goto close_and_fail;
3446 }
3447
3448 bdrv_parent_cb_change_media(bs, true);
3449
3450 qobject_unref(options);
3451 options = NULL;
3452
3453 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3454 * temporary snapshot afterwards. */
3455 if (snapshot_flags) {
3456 BlockDriverState *snapshot_bs;
3457 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3458 snapshot_options, &local_err);
3459 snapshot_options = NULL;
3460 if (local_err) {
3461 goto close_and_fail;
3462 }
3463 /* We are not going to return bs but the overlay on top of it
3464 * (snapshot_bs); thus, we have to drop the strong reference to bs
3465 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3466 * though, because the overlay still has a reference to it. */
3467 bdrv_unref(bs);
3468 bs = snapshot_bs;
3469 }
3470
3471 return bs;
3472
3473 fail:
3474 blk_unref(file);
3475 qobject_unref(snapshot_options);
3476 qobject_unref(bs->explicit_options);
3477 qobject_unref(bs->options);
3478 qobject_unref(options);
3479 bs->options = NULL;
3480 bs->explicit_options = NULL;
3481 bdrv_unref(bs);
3482 error_propagate(errp, local_err);
3483 return NULL;
3484
3485 close_and_fail:
3486 bdrv_unref(bs);
3487 qobject_unref(snapshot_options);
3488 qobject_unref(options);
3489 error_propagate(errp, local_err);
3490 return NULL;
3491 }
3492
3493 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3494 QDict *options, int flags, Error **errp)
3495 {
3496 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3497 NULL, 0, errp);
3498 }
3499
3500 /* Return true if the NULL-terminated @list contains @str */
3501 static bool is_str_in_list(const char *str, const char *const *list)
3502 {
3503 if (str && list) {
3504 int i;
3505 for (i = 0; list[i] != NULL; i++) {
3506 if (!strcmp(str, list[i])) {
3507 return true;
3508 }
3509 }
3510 }
3511 return false;
3512 }
3513
3514 /*
3515 * Check that every option set in @bs->options is also set in
3516 * @new_opts.
3517 *
3518 * Options listed in the common_options list and in
3519 * @bs->drv->mutable_opts are skipped.
3520 *
3521 * Return 0 on success, otherwise return -EINVAL and set @errp.
3522 */
3523 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3524 const QDict *new_opts, Error **errp)
3525 {
3526 const QDictEntry *e;
3527 /* These options are common to all block drivers and are handled
3528 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3529 const char *const common_options[] = {
3530 "node-name", "discard", "cache.direct", "cache.no-flush",
3531 "read-only", "auto-read-only", "detect-zeroes", NULL
3532 };
3533
3534 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3535 if (!qdict_haskey(new_opts, e->key) &&
3536 !is_str_in_list(e->key, common_options) &&
3537 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3538 error_setg(errp, "Option '%s' cannot be reset "
3539 "to its default value", e->key);
3540 return -EINVAL;
3541 }
3542 }
3543
3544 return 0;
3545 }
3546
3547 /*
3548 * Returns true if @child can be reached recursively from @bs
3549 */
3550 static bool bdrv_recurse_has_child(BlockDriverState *bs,
3551 BlockDriverState *child)
3552 {
3553 BdrvChild *c;
3554
3555 if (bs == child) {
3556 return true;
3557 }
3558
3559 QLIST_FOREACH(c, &bs->children, next) {