target/riscv: vector register gather instruction
[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 qemu_opts_absorb_qdict(opts, options, &local_err);
1633 if (local_err) {
1634 error_propagate(errp, local_err);
1635 ret = -EINVAL;
1636 goto fail_opts;
1637 }
1638
1639 update_flags_from_options(&bs->open_flags, opts);
1640
1641 driver_name = qemu_opt_get(opts, "driver");
1642 drv = bdrv_find_format(driver_name);
1643 assert(drv != NULL);
1644
1645 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1646
1647 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1648 error_setg(errp,
1649 BDRV_OPT_FORCE_SHARE
1650 "=on can only be used with read-only images");
1651 ret = -EINVAL;
1652 goto fail_opts;
1653 }
1654
1655 if (file != NULL) {
1656 bdrv_refresh_filename(blk_bs(file));
1657 filename = blk_bs(file)->filename;
1658 } else {
1659 /*
1660 * Caution: while qdict_get_try_str() is fine, getting
1661 * non-string types would require more care. When @options
1662 * come from -blockdev or blockdev_add, its members are typed
1663 * according to the QAPI schema, but when they come from
1664 * -drive, they're all QString.
1665 */
1666 filename = qdict_get_try_str(options, "filename");
1667 }
1668
1669 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1670 error_setg(errp, "The '%s' block driver requires a file name",
1671 drv->format_name);
1672 ret = -EINVAL;
1673 goto fail_opts;
1674 }
1675
1676 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1677 drv->format_name);
1678
1679 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1680
1681 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1682 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1683 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1684 } else {
1685 ret = -ENOTSUP;
1686 }
1687 if (ret < 0) {
1688 error_setg(errp,
1689 !bs->read_only && bdrv_is_whitelisted(drv, true)
1690 ? "Driver '%s' can only be used for read-only devices"
1691 : "Driver '%s' is not whitelisted",
1692 drv->format_name);
1693 goto fail_opts;
1694 }
1695 }
1696
1697 /* bdrv_new() and bdrv_close() make it so */
1698 assert(atomic_read(&bs->copy_on_read) == 0);
1699
1700 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1701 if (!bs->read_only) {
1702 bdrv_enable_copy_on_read(bs);
1703 } else {
1704 error_setg(errp, "Can't use copy-on-read on read-only device");
1705 ret = -EINVAL;
1706 goto fail_opts;
1707 }
1708 }
1709
1710 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1711 if (discard != NULL) {
1712 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1713 error_setg(errp, "Invalid discard option");
1714 ret = -EINVAL;
1715 goto fail_opts;
1716 }
1717 }
1718
1719 bs->detect_zeroes =
1720 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1721 if (local_err) {
1722 error_propagate(errp, local_err);
1723 ret = -EINVAL;
1724 goto fail_opts;
1725 }
1726
1727 if (filename != NULL) {
1728 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1729 } else {
1730 bs->filename[0] = '\0';
1731 }
1732 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1733
1734 /* Open the image, either directly or using a protocol */
1735 open_flags = bdrv_open_flags(bs, bs->open_flags);
1736 node_name = qemu_opt_get(opts, "node-name");
1737
1738 assert(!drv->bdrv_file_open || file == NULL);
1739 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1740 if (ret < 0) {
1741 goto fail_opts;
1742 }
1743
1744 qemu_opts_del(opts);
1745 return 0;
1746
1747 fail_opts:
1748 qemu_opts_del(opts);
1749 return ret;
1750 }
1751
1752 static QDict *parse_json_filename(const char *filename, Error **errp)
1753 {
1754 QObject *options_obj;
1755 QDict *options;
1756 int ret;
1757
1758 ret = strstart(filename, "json:", &filename);
1759 assert(ret);
1760
1761 options_obj = qobject_from_json(filename, errp);
1762 if (!options_obj) {
1763 error_prepend(errp, "Could not parse the JSON options: ");
1764 return NULL;
1765 }
1766
1767 options = qobject_to(QDict, options_obj);
1768 if (!options) {
1769 qobject_unref(options_obj);
1770 error_setg(errp, "Invalid JSON object given");
1771 return NULL;
1772 }
1773
1774 qdict_flatten(options);
1775
1776 return options;
1777 }
1778
1779 static void parse_json_protocol(QDict *options, const char **pfilename,
1780 Error **errp)
1781 {
1782 QDict *json_options;
1783 Error *local_err = NULL;
1784
1785 /* Parse json: pseudo-protocol */
1786 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1787 return;
1788 }
1789
1790 json_options = parse_json_filename(*pfilename, &local_err);
1791 if (local_err) {
1792 error_propagate(errp, local_err);
1793 return;
1794 }
1795
1796 /* Options given in the filename have lower priority than options
1797 * specified directly */
1798 qdict_join(options, json_options, false);
1799 qobject_unref(json_options);
1800 *pfilename = NULL;
1801 }
1802
1803 /*
1804 * Fills in default options for opening images and converts the legacy
1805 * filename/flags pair to option QDict entries.
1806 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1807 * block driver has been specified explicitly.
1808 */
1809 static int bdrv_fill_options(QDict **options, const char *filename,
1810 int *flags, Error **errp)
1811 {
1812 const char *drvname;
1813 bool protocol = *flags & BDRV_O_PROTOCOL;
1814 bool parse_filename = false;
1815 BlockDriver *drv = NULL;
1816 Error *local_err = NULL;
1817
1818 /*
1819 * Caution: while qdict_get_try_str() is fine, getting non-string
1820 * types would require more care. When @options come from
1821 * -blockdev or blockdev_add, its members are typed according to
1822 * the QAPI schema, but when they come from -drive, they're all
1823 * QString.
1824 */
1825 drvname = qdict_get_try_str(*options, "driver");
1826 if (drvname) {
1827 drv = bdrv_find_format(drvname);
1828 if (!drv) {
1829 error_setg(errp, "Unknown driver '%s'", drvname);
1830 return -ENOENT;
1831 }
1832 /* If the user has explicitly specified the driver, this choice should
1833 * override the BDRV_O_PROTOCOL flag */
1834 protocol = drv->bdrv_file_open;
1835 }
1836
1837 if (protocol) {
1838 *flags |= BDRV_O_PROTOCOL;
1839 } else {
1840 *flags &= ~BDRV_O_PROTOCOL;
1841 }
1842
1843 /* Translate cache options from flags into options */
1844 update_options_from_flags(*options, *flags);
1845
1846 /* Fetch the file name from the options QDict if necessary */
1847 if (protocol && filename) {
1848 if (!qdict_haskey(*options, "filename")) {
1849 qdict_put_str(*options, "filename", filename);
1850 parse_filename = true;
1851 } else {
1852 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1853 "the same time");
1854 return -EINVAL;
1855 }
1856 }
1857
1858 /* Find the right block driver */
1859 /* See cautionary note on accessing @options above */
1860 filename = qdict_get_try_str(*options, "filename");
1861
1862 if (!drvname && protocol) {
1863 if (filename) {
1864 drv = bdrv_find_protocol(filename, parse_filename, errp);
1865 if (!drv) {
1866 return -EINVAL;
1867 }
1868
1869 drvname = drv->format_name;
1870 qdict_put_str(*options, "driver", drvname);
1871 } else {
1872 error_setg(errp, "Must specify either driver or file");
1873 return -EINVAL;
1874 }
1875 }
1876
1877 assert(drv || !protocol);
1878
1879 /* Driver-specific filename parsing */
1880 if (drv && drv->bdrv_parse_filename && parse_filename) {
1881 drv->bdrv_parse_filename(filename, *options, &local_err);
1882 if (local_err) {
1883 error_propagate(errp, local_err);
1884 return -EINVAL;
1885 }
1886
1887 if (!drv->bdrv_needs_filename) {
1888 qdict_del(*options, "filename");
1889 }
1890 }
1891
1892 return 0;
1893 }
1894
1895 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1896 uint64_t perm, uint64_t shared,
1897 GSList *ignore_children,
1898 bool *tighten_restrictions, Error **errp);
1899 static void bdrv_child_abort_perm_update(BdrvChild *c);
1900 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1901
1902 typedef struct BlockReopenQueueEntry {
1903 bool prepared;
1904 bool perms_checked;
1905 BDRVReopenState state;
1906 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
1907 } BlockReopenQueueEntry;
1908
1909 /*
1910 * Return the flags that @bs will have after the reopens in @q have
1911 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1912 * return the current flags.
1913 */
1914 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1915 {
1916 BlockReopenQueueEntry *entry;
1917
1918 if (q != NULL) {
1919 QTAILQ_FOREACH(entry, q, entry) {
1920 if (entry->state.bs == bs) {
1921 return entry->state.flags;
1922 }
1923 }
1924 }
1925
1926 return bs->open_flags;
1927 }
1928
1929 /* Returns whether the image file can be written to after the reopen queue @q
1930 * has been successfully applied, or right now if @q is NULL. */
1931 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1932 BlockReopenQueue *q)
1933 {
1934 int flags = bdrv_reopen_get_flags(q, bs);
1935
1936 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1937 }
1938
1939 /*
1940 * Return whether the BDS can be written to. This is not necessarily
1941 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1942 * be written to but do not count as read-only images.
1943 */
1944 bool bdrv_is_writable(BlockDriverState *bs)
1945 {
1946 return bdrv_is_writable_after_reopen(bs, NULL);
1947 }
1948
1949 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1950 BdrvChild *c, BdrvChildRole role,
1951 BlockReopenQueue *reopen_queue,
1952 uint64_t parent_perm, uint64_t parent_shared,
1953 uint64_t *nperm, uint64_t *nshared)
1954 {
1955 assert(bs->drv && bs->drv->bdrv_child_perm);
1956 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1957 parent_perm, parent_shared,
1958 nperm, nshared);
1959 /* TODO Take force_share from reopen_queue */
1960 if (child_bs && child_bs->force_share) {
1961 *nshared = BLK_PERM_ALL;
1962 }
1963 }
1964
1965 /*
1966 * Check whether permissions on this node can be changed in a way that
1967 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1968 * permissions of all its parents. This involves checking whether all necessary
1969 * permission changes to child nodes can be performed.
1970 *
1971 * Will set *tighten_restrictions to true if and only if new permissions have to
1972 * be taken or currently shared permissions are to be unshared. Otherwise,
1973 * errors are not fatal as long as the caller accepts that the restrictions
1974 * remain tighter than they need to be. The caller still has to abort the
1975 * transaction.
1976 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1977 * encounter fatal errors even though no restrictions are to be tightened. For
1978 * example, changing a node from RW to RO will fail if the WRITE permission is
1979 * to be kept.
1980 *
1981 * A call to this function must always be followed by a call to bdrv_set_perm()
1982 * or bdrv_abort_perm_update().
1983 */
1984 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1985 uint64_t cumulative_perms,
1986 uint64_t cumulative_shared_perms,
1987 GSList *ignore_children,
1988 bool *tighten_restrictions, Error **errp)
1989 {
1990 BlockDriver *drv = bs->drv;
1991 BdrvChild *c;
1992 int ret;
1993
1994 assert(!q || !tighten_restrictions);
1995
1996 if (tighten_restrictions) {
1997 uint64_t current_perms, current_shared;
1998 uint64_t added_perms, removed_shared_perms;
1999
2000 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
2001
2002 added_perms = cumulative_perms & ~current_perms;
2003 removed_shared_perms = current_shared & ~cumulative_shared_perms;
2004
2005 *tighten_restrictions = added_perms || removed_shared_perms;
2006 }
2007
2008 /* Write permissions never work with read-only images */
2009 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2010 !bdrv_is_writable_after_reopen(bs, q))
2011 {
2012 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2013 error_setg(errp, "Block node is read-only");
2014 } else {
2015 uint64_t current_perms, current_shared;
2016 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
2017 if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
2018 error_setg(errp, "Cannot make block node read-only, there is "
2019 "a writer on it");
2020 } else {
2021 error_setg(errp, "Cannot make block node read-only and create "
2022 "a writer on it");
2023 }
2024 }
2025
2026 return -EPERM;
2027 }
2028
2029 /* Check this node */
2030 if (!drv) {
2031 return 0;
2032 }
2033
2034 if (drv->bdrv_check_perm) {
2035 return drv->bdrv_check_perm(bs, cumulative_perms,
2036 cumulative_shared_perms, errp);
2037 }
2038
2039 /* Drivers that never have children can omit .bdrv_child_perm() */
2040 if (!drv->bdrv_child_perm) {
2041 assert(QLIST_EMPTY(&bs->children));
2042 return 0;
2043 }
2044
2045 /* Check all children */
2046 QLIST_FOREACH(c, &bs->children, next) {
2047 uint64_t cur_perm, cur_shared;
2048 bool child_tighten_restr;
2049
2050 bdrv_child_perm(bs, c->bs, c, c->role, q,
2051 cumulative_perms, cumulative_shared_perms,
2052 &cur_perm, &cur_shared);
2053 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children,
2054 tighten_restrictions ? &child_tighten_restr
2055 : NULL,
2056 errp);
2057 if (tighten_restrictions) {
2058 *tighten_restrictions |= child_tighten_restr;
2059 }
2060 if (ret < 0) {
2061 return ret;
2062 }
2063 }
2064
2065 return 0;
2066 }
2067
2068 /*
2069 * Notifies drivers that after a previous bdrv_check_perm() call, the
2070 * permission update is not performed and any preparations made for it (e.g.
2071 * taken file locks) need to be undone.
2072 *
2073 * This function recursively notifies all child nodes.
2074 */
2075 static void bdrv_abort_perm_update(BlockDriverState *bs)
2076 {
2077 BlockDriver *drv = bs->drv;
2078 BdrvChild *c;
2079
2080 if (!drv) {
2081 return;
2082 }
2083
2084 if (drv->bdrv_abort_perm_update) {
2085 drv->bdrv_abort_perm_update(bs);
2086 }
2087
2088 QLIST_FOREACH(c, &bs->children, next) {
2089 bdrv_child_abort_perm_update(c);
2090 }
2091 }
2092
2093 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
2094 uint64_t cumulative_shared_perms)
2095 {
2096 BlockDriver *drv = bs->drv;
2097 BdrvChild *c;
2098
2099 if (!drv) {
2100 return;
2101 }
2102
2103 /* Update this node */
2104 if (drv->bdrv_set_perm) {
2105 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2106 }
2107
2108 /* Drivers that never have children can omit .bdrv_child_perm() */
2109 if (!drv->bdrv_child_perm) {
2110 assert(QLIST_EMPTY(&bs->children));
2111 return;
2112 }
2113
2114 /* Update all children */
2115 QLIST_FOREACH(c, &bs->children, next) {
2116 uint64_t cur_perm, cur_shared;
2117 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2118 cumulative_perms, cumulative_shared_perms,
2119 &cur_perm, &cur_shared);
2120 bdrv_child_set_perm(c, cur_perm, cur_shared);
2121 }
2122 }
2123
2124 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2125 uint64_t *shared_perm)
2126 {
2127 BdrvChild *c;
2128 uint64_t cumulative_perms = 0;
2129 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2130
2131 QLIST_FOREACH(c, &bs->parents, next_parent) {
2132 cumulative_perms |= c->perm;
2133 cumulative_shared_perms &= c->shared_perm;
2134 }
2135
2136 *perm = cumulative_perms;
2137 *shared_perm = cumulative_shared_perms;
2138 }
2139
2140 static char *bdrv_child_user_desc(BdrvChild *c)
2141 {
2142 if (c->klass->get_parent_desc) {
2143 return c->klass->get_parent_desc(c);
2144 }
2145
2146 return g_strdup("another user");
2147 }
2148
2149 char *bdrv_perm_names(uint64_t perm)
2150 {
2151 struct perm_name {
2152 uint64_t perm;
2153 const char *name;
2154 } permissions[] = {
2155 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2156 { BLK_PERM_WRITE, "write" },
2157 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2158 { BLK_PERM_RESIZE, "resize" },
2159 { BLK_PERM_GRAPH_MOD, "change children" },
2160 { 0, NULL }
2161 };
2162
2163 GString *result = g_string_sized_new(30);
2164 struct perm_name *p;
2165
2166 for (p = permissions; p->name; p++) {
2167 if (perm & p->perm) {
2168 if (result->len > 0) {
2169 g_string_append(result, ", ");
2170 }
2171 g_string_append(result, p->name);
2172 }
2173 }
2174
2175 return g_string_free(result, FALSE);
2176 }
2177
2178 /*
2179 * Checks whether a new reference to @bs can be added if the new user requires
2180 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2181 * set, the BdrvChild objects in this list are ignored in the calculations;
2182 * this allows checking permission updates for an existing reference.
2183 *
2184 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2185 *
2186 * Needs to be followed by a call to either bdrv_set_perm() or
2187 * bdrv_abort_perm_update(). */
2188 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
2189 uint64_t new_used_perm,
2190 uint64_t new_shared_perm,
2191 GSList *ignore_children,
2192 bool *tighten_restrictions,
2193 Error **errp)
2194 {
2195 BdrvChild *c;
2196 uint64_t cumulative_perms = new_used_perm;
2197 uint64_t cumulative_shared_perms = new_shared_perm;
2198
2199 assert(!q || !tighten_restrictions);
2200
2201 /* There is no reason why anyone couldn't tolerate write_unchanged */
2202 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
2203
2204 QLIST_FOREACH(c, &bs->parents, next_parent) {
2205 if (g_slist_find(ignore_children, c)) {
2206 continue;
2207 }
2208
2209 if ((new_used_perm & c->shared_perm) != new_used_perm) {
2210 char *user = bdrv_child_user_desc(c);
2211 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
2212
2213 if (tighten_restrictions) {
2214 *tighten_restrictions = true;
2215 }
2216
2217 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
2218 "allow '%s' on %s",
2219 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2220 g_free(user);
2221 g_free(perm_names);
2222 return -EPERM;
2223 }
2224
2225 if ((c->perm & new_shared_perm) != c->perm) {
2226 char *user = bdrv_child_user_desc(c);
2227 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
2228
2229 if (tighten_restrictions) {
2230 *tighten_restrictions = true;
2231 }
2232
2233 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
2234 "'%s' on %s",
2235 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2236 g_free(user);
2237 g_free(perm_names);
2238 return -EPERM;
2239 }
2240
2241 cumulative_perms |= c->perm;
2242 cumulative_shared_perms &= c->shared_perm;
2243 }
2244
2245 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
2246 ignore_children, tighten_restrictions, errp);
2247 }
2248
2249 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2250 * bdrv_child_abort_perm_update(). */
2251 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
2252 uint64_t perm, uint64_t shared,
2253 GSList *ignore_children,
2254 bool *tighten_restrictions, Error **errp)
2255 {
2256 int ret;
2257
2258 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
2259 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children,
2260 tighten_restrictions, errp);
2261 g_slist_free(ignore_children);
2262
2263 if (ret < 0) {
2264 return ret;
2265 }
2266
2267 if (!c->has_backup_perm) {
2268 c->has_backup_perm = true;
2269 c->backup_perm = c->perm;
2270 c->backup_shared_perm = c->shared_perm;
2271 }
2272 /*
2273 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2274 * same child twice during check_perm procedure
2275 */
2276
2277 c->perm = perm;
2278 c->shared_perm = shared;
2279
2280 return 0;
2281 }
2282
2283 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
2284 {
2285 uint64_t cumulative_perms, cumulative_shared_perms;
2286
2287 c->has_backup_perm = false;
2288
2289 c->perm = perm;
2290 c->shared_perm = shared;
2291
2292 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
2293 &cumulative_shared_perms);
2294 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
2295 }
2296
2297 static void bdrv_child_abort_perm_update(BdrvChild *c)
2298 {
2299 if (c->has_backup_perm) {
2300 c->perm = c->backup_perm;
2301 c->shared_perm = c->backup_shared_perm;
2302 c->has_backup_perm = false;
2303 }
2304
2305 bdrv_abort_perm_update(c->bs);
2306 }
2307
2308 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2309 Error **errp)
2310 {
2311 Error *local_err = NULL;
2312 int ret;
2313 bool tighten_restrictions;
2314
2315 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL,
2316 &tighten_restrictions, &local_err);
2317 if (ret < 0) {
2318 bdrv_child_abort_perm_update(c);
2319 if (tighten_restrictions) {
2320 error_propagate(errp, local_err);
2321 } else {
2322 /*
2323 * Our caller may intend to only loosen restrictions and
2324 * does not expect this function to fail. Errors are not
2325 * fatal in such a case, so we can just hide them from our
2326 * caller.
2327 */
2328 error_free(local_err);
2329 ret = 0;
2330 }
2331 return ret;
2332 }
2333
2334 bdrv_child_set_perm(c, perm, shared);
2335
2336 return 0;
2337 }
2338
2339 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2340 {
2341 uint64_t parent_perms, parent_shared;
2342 uint64_t perms, shared;
2343
2344 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2345 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2346 parent_perms, parent_shared, &perms, &shared);
2347
2348 return bdrv_child_try_set_perm(c, perms, shared, errp);
2349 }
2350
2351 /*
2352 * Default implementation for .bdrv_child_perm() for block filters:
2353 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2354 * filtered child.
2355 */
2356 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2357 BdrvChildRole role,
2358 BlockReopenQueue *reopen_queue,
2359 uint64_t perm, uint64_t shared,
2360 uint64_t *nperm, uint64_t *nshared)
2361 {
2362 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2363 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2364 }
2365
2366 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2367 BdrvChildRole role,
2368 BlockReopenQueue *reopen_queue,
2369 uint64_t perm, uint64_t shared,
2370 uint64_t *nperm, uint64_t *nshared)
2371 {
2372 assert(role & BDRV_CHILD_COW);
2373
2374 /*
2375 * We want consistent read from backing files if the parent needs it.
2376 * No other operations are performed on backing files.
2377 */
2378 perm &= BLK_PERM_CONSISTENT_READ;
2379
2380 /*
2381 * If the parent can deal with changing data, we're okay with a
2382 * writable and resizable backing file.
2383 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2384 */
2385 if (shared & BLK_PERM_WRITE) {
2386 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2387 } else {
2388 shared = 0;
2389 }
2390
2391 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2392 BLK_PERM_WRITE_UNCHANGED;
2393
2394 if (bs->open_flags & BDRV_O_INACTIVE) {
2395 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2396 }
2397
2398 *nperm = perm;
2399 *nshared = shared;
2400 }
2401
2402 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2403 BdrvChildRole role,
2404 BlockReopenQueue *reopen_queue,
2405 uint64_t perm, uint64_t shared,
2406 uint64_t *nperm, uint64_t *nshared)
2407 {
2408 int flags;
2409
2410 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2411
2412 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2413
2414 /*
2415 * Apart from the modifications below, the same permissions are
2416 * forwarded and left alone as for filters
2417 */
2418 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2419 perm, shared, &perm, &shared);
2420
2421 if (role & BDRV_CHILD_METADATA) {
2422 /* Format drivers may touch metadata even if the guest doesn't write */
2423 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2424 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2425 }
2426
2427 /*
2428 * bs->file always needs to be consistent because of the
2429 * metadata. We can never allow other users to resize or write
2430 * to it.
2431 */
2432 if (!(flags & BDRV_O_NO_IO)) {
2433 perm |= BLK_PERM_CONSISTENT_READ;
2434 }
2435 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2436 }
2437
2438 if (role & BDRV_CHILD_DATA) {
2439 /*
2440 * Technically, everything in this block is a subset of the
2441 * BDRV_CHILD_METADATA path taken above, and so this could
2442 * be an "else if" branch. However, that is not obvious, and
2443 * this function is not performance critical, therefore we let
2444 * this be an independent "if".
2445 */
2446
2447 /*
2448 * We cannot allow other users to resize the file because the
2449 * format driver might have some assumptions about the size
2450 * (e.g. because it is stored in metadata, or because the file
2451 * is split into fixed-size data files).
2452 */
2453 shared &= ~BLK_PERM_RESIZE;
2454
2455 /*
2456 * WRITE_UNCHANGED often cannot be performed as such on the
2457 * data file. For example, the qcow2 driver may still need to
2458 * write copied clusters on copy-on-read.
2459 */
2460 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2461 perm |= BLK_PERM_WRITE;
2462 }
2463
2464 /*
2465 * If the data file is written to, the format driver may
2466 * expect to be able to resize it by writing beyond the EOF.
2467 */
2468 if (perm & BLK_PERM_WRITE) {
2469 perm |= BLK_PERM_RESIZE;
2470 }
2471 }
2472
2473 if (bs->open_flags & BDRV_O_INACTIVE) {
2474 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2475 }
2476
2477 *nperm = perm;
2478 *nshared = shared;
2479 }
2480
2481 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2482 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2483 uint64_t perm, uint64_t shared,
2484 uint64_t *nperm, uint64_t *nshared)
2485 {
2486 if (role & BDRV_CHILD_FILTERED) {
2487 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2488 BDRV_CHILD_COW)));
2489 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2490 perm, shared, nperm, nshared);
2491 } else if (role & BDRV_CHILD_COW) {
2492 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2493 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2494 perm, shared, nperm, nshared);
2495 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2496 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2497 perm, shared, nperm, nshared);
2498 } else {
2499 g_assert_not_reached();
2500 }
2501 }
2502
2503 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2504 {
2505 static const uint64_t permissions[] = {
2506 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2507 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2508 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2509 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2510 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2511 };
2512
2513 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2514 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2515
2516 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2517
2518 return permissions[qapi_perm];
2519 }
2520
2521 static void bdrv_replace_child_noperm(BdrvChild *child,
2522 BlockDriverState *new_bs)
2523 {
2524 BlockDriverState *old_bs = child->bs;
2525 int new_bs_quiesce_counter;
2526 int drain_saldo;
2527
2528 assert(!child->frozen);
2529
2530 if (old_bs && new_bs) {
2531 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2532 }
2533
2534 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2535 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2536
2537 /*
2538 * If the new child node is drained but the old one was not, flush
2539 * all outstanding requests to the old child node.
2540 */
2541 while (drain_saldo > 0 && child->klass->drained_begin) {
2542 bdrv_parent_drained_begin_single(child, true);
2543 drain_saldo--;
2544 }
2545
2546 if (old_bs) {
2547 /* Detach first so that the recursive drain sections coming from @child
2548 * are already gone and we only end the drain sections that came from
2549 * elsewhere. */
2550 if (child->klass->detach) {
2551 child->klass->detach(child);
2552 }
2553 QLIST_REMOVE(child, next_parent);
2554 }
2555
2556 child->bs = new_bs;
2557
2558 if (new_bs) {
2559 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2560
2561 /*
2562 * Detaching the old node may have led to the new node's
2563 * quiesce_counter having been decreased. Not a problem, we
2564 * just need to recognize this here and then invoke
2565 * drained_end appropriately more often.
2566 */
2567 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2568 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2569
2570 /* Attach only after starting new drained sections, so that recursive
2571 * drain sections coming from @child don't get an extra .drained_begin
2572 * callback. */
2573 if (child->klass->attach) {
2574 child->klass->attach(child);
2575 }
2576 }
2577
2578 /*
2579 * If the old child node was drained but the new one is not, allow
2580 * requests to come in only after the new node has been attached.
2581 */
2582 while (drain_saldo < 0 && child->klass->drained_end) {
2583 bdrv_parent_drained_end_single(child);
2584 drain_saldo++;
2585 }
2586 }
2587
2588 /*
2589 * Updates @child to change its reference to point to @new_bs, including
2590 * checking and applying the necessary permisson updates both to the old node
2591 * and to @new_bs.
2592 *
2593 * NULL is passed as @new_bs for removing the reference before freeing @child.
2594 *
2595 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2596 * function uses bdrv_set_perm() to update the permissions according to the new
2597 * reference that @new_bs gets.
2598 */
2599 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2600 {
2601 BlockDriverState *old_bs = child->bs;
2602 uint64_t perm, shared_perm;
2603
2604 bdrv_replace_child_noperm(child, new_bs);
2605
2606 /*
2607 * Start with the new node's permissions. If @new_bs is a (direct
2608 * or indirect) child of @old_bs, we must complete the permission
2609 * update on @new_bs before we loosen the restrictions on @old_bs.
2610 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2611 * updating the permissions of @new_bs, and thus not purely loosen
2612 * restrictions.
2613 */
2614 if (new_bs) {
2615 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2616 bdrv_set_perm(new_bs, perm, shared_perm);
2617 }
2618
2619 if (old_bs) {
2620 /* Update permissions for old node. This is guaranteed to succeed
2621 * because we're just taking a parent away, so we're loosening
2622 * restrictions. */
2623 bool tighten_restrictions;
2624 int ret;
2625
2626 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2627 ret = bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL,
2628 &tighten_restrictions, NULL);
2629 assert(tighten_restrictions == false);
2630 if (ret < 0) {
2631 /* We only tried to loosen restrictions, so errors are not fatal */
2632 bdrv_abort_perm_update(old_bs);
2633 } else {
2634 bdrv_set_perm(old_bs, perm, shared_perm);
2635 }
2636
2637 /* When the parent requiring a non-default AioContext is removed, the
2638 * node moves back to the main AioContext */
2639 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2640 }
2641 }
2642
2643 /*
2644 * This function steals the reference to child_bs from the caller.
2645 * That reference is later dropped by bdrv_root_unref_child().
2646 *
2647 * On failure NULL is returned, errp is set and the reference to
2648 * child_bs is also dropped.
2649 *
2650 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2651 * (unless @child_bs is already in @ctx).
2652 */
2653 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2654 const char *child_name,
2655 const BdrvChildClass *child_class,
2656 BdrvChildRole child_role,
2657 AioContext *ctx,
2658 uint64_t perm, uint64_t shared_perm,
2659 void *opaque, Error **errp)
2660 {
2661 BdrvChild *child;
2662 Error *local_err = NULL;
2663 int ret;
2664
2665 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, NULL,
2666 errp);
2667 if (ret < 0) {
2668 bdrv_abort_perm_update(child_bs);
2669 bdrv_unref(child_bs);
2670 return NULL;
2671 }
2672
2673 child = g_new(BdrvChild, 1);
2674 *child = (BdrvChild) {
2675 .bs = NULL,
2676 .name = g_strdup(child_name),
2677 .klass = child_class,
2678 .role = child_role,
2679 .perm = perm,
2680 .shared_perm = shared_perm,
2681 .opaque = opaque,
2682 };
2683
2684 /* If the AioContexts don't match, first try to move the subtree of
2685 * child_bs into the AioContext of the new parent. If this doesn't work,
2686 * try moving the parent into the AioContext of child_bs instead. */
2687 if (bdrv_get_aio_context(child_bs) != ctx) {
2688 ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err);
2689 if (ret < 0 && child_class->can_set_aio_ctx) {
2690 GSList *ignore = g_slist_prepend(NULL, child);
2691 ctx = bdrv_get_aio_context(child_bs);
2692 if (child_class->can_set_aio_ctx(child, ctx, &ignore, NULL)) {
2693 error_free(local_err);
2694 ret = 0;
2695 g_slist_free(ignore);
2696 ignore = g_slist_prepend(NULL, child);
2697 child_class->set_aio_ctx(child, ctx, &ignore);
2698 }
2699 g_slist_free(ignore);
2700 }
2701 if (ret < 0) {
2702 error_propagate(errp, local_err);
2703 g_free(child);
2704 bdrv_abort_perm_update(child_bs);
2705 bdrv_unref(child_bs);
2706 return NULL;
2707 }
2708 }
2709
2710 /* This performs the matching bdrv_set_perm() for the above check. */
2711 bdrv_replace_child(child, child_bs);
2712
2713 return child;
2714 }
2715
2716 /*
2717 * This function transfers the reference to child_bs from the caller
2718 * to parent_bs. That reference is later dropped by parent_bs on
2719 * bdrv_close() or if someone calls bdrv_unref_child().
2720 *
2721 * On failure NULL is returned, errp is set and the reference to
2722 * child_bs is also dropped.
2723 *
2724 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2725 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2726 */
2727 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2728 BlockDriverState *child_bs,
2729 const char *child_name,
2730 const BdrvChildClass *child_class,
2731 BdrvChildRole child_role,
2732 Error **errp)
2733 {
2734 BdrvChild *child;
2735 uint64_t perm, shared_perm;
2736
2737 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2738
2739 assert(parent_bs->drv);
2740 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2741 perm, shared_perm, &perm, &shared_perm);
2742
2743 child = bdrv_root_attach_child(child_bs, child_name, child_class,
2744 child_role, bdrv_get_aio_context(parent_bs),
2745 perm, shared_perm, parent_bs, errp);
2746 if (child == NULL) {
2747 return NULL;
2748 }
2749
2750 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2751 return child;
2752 }
2753
2754 static void bdrv_detach_child(BdrvChild *child)
2755 {
2756 QLIST_SAFE_REMOVE(child, next);
2757
2758 bdrv_replace_child(child, NULL);
2759
2760 g_free(child->name);
2761 g_free(child);
2762 }
2763
2764 void bdrv_root_unref_child(BdrvChild *child)
2765 {
2766 BlockDriverState *child_bs;
2767
2768 child_bs = child->bs;
2769 bdrv_detach_child(child);
2770 bdrv_unref(child_bs);
2771 }
2772
2773 /**
2774 * Clear all inherits_from pointers from children and grandchildren of
2775 * @root that point to @root, where necessary.
2776 */
2777 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child)
2778 {
2779 BdrvChild *c;
2780
2781 if (child->bs->inherits_from == root) {
2782 /*
2783 * Remove inherits_from only when the last reference between root and
2784 * child->bs goes away.
2785 */
2786 QLIST_FOREACH(c, &root->children, next) {
2787 if (c != child && c->bs == child->bs) {
2788 break;
2789 }
2790 }
2791 if (c == NULL) {
2792 child->bs->inherits_from = NULL;
2793 }
2794 }
2795
2796 QLIST_FOREACH(c, &child->bs->children, next) {
2797 bdrv_unset_inherits_from(root, c);
2798 }
2799 }
2800
2801 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2802 {
2803 if (child == NULL) {
2804 return;
2805 }
2806
2807 bdrv_unset_inherits_from(parent, child);
2808 bdrv_root_unref_child(child);
2809 }
2810
2811
2812 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2813 {
2814 BdrvChild *c;
2815 QLIST_FOREACH(c, &bs->parents, next_parent) {
2816 if (c->klass->change_media) {
2817 c->klass->change_media(c, load);
2818 }
2819 }
2820 }
2821
2822 /* Return true if you can reach parent going through child->inherits_from
2823 * recursively. If parent or child are NULL, return false */
2824 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2825 BlockDriverState *parent)
2826 {
2827 while (child && child != parent) {
2828 child = child->inherits_from;
2829 }
2830
2831 return child != NULL;
2832 }
2833
2834 /*
2835 * Return the BdrvChildRole for @bs's backing child. bs->backing is
2836 * mostly used for COW backing children (role = COW), but also for
2837 * filtered children (role = FILTERED | PRIMARY).
2838 */
2839 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
2840 {
2841 if (bs->drv && bs->drv->is_filter) {
2842 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
2843 } else {
2844 return BDRV_CHILD_COW;
2845 }
2846 }
2847
2848 /*
2849 * Sets the backing file link of a BDS. A new reference is created; callers
2850 * which don't need their own reference any more must call bdrv_unref().
2851 */
2852 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2853 Error **errp)
2854 {
2855 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2856 bdrv_inherits_from_recursive(backing_hd, bs);
2857
2858 if (bdrv_is_backing_chain_frozen(bs, backing_bs(bs), errp)) {
2859 return;
2860 }
2861
2862 if (backing_hd) {
2863 bdrv_ref(backing_hd);
2864 }
2865
2866 if (bs->backing) {
2867 bdrv_unref_child(bs, bs->backing);
2868 bs->backing = NULL;
2869 }
2870
2871 if (!backing_hd) {
2872 goto out;
2873 }
2874
2875 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_of_bds,
2876 bdrv_backing_role(bs), errp);
2877 /* If backing_hd was already part of bs's backing chain, and
2878 * inherits_from pointed recursively to bs then let's update it to
2879 * point directly to bs (else it will become NULL). */
2880 if (bs->backing && update_inherits_from) {
2881 backing_hd->inherits_from = bs;
2882 }
2883
2884 out:
2885 bdrv_refresh_limits(bs, NULL);
2886 }
2887
2888 /*
2889 * Opens the backing file for a BlockDriverState if not yet open
2890 *
2891 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2892 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2893 * itself, all options starting with "${bdref_key}." are considered part of the
2894 * BlockdevRef.
2895 *
2896 * TODO Can this be unified with bdrv_open_image()?
2897 */
2898 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2899 const char *bdref_key, Error **errp)
2900 {
2901 char *backing_filename = NULL;
2902 char *bdref_key_dot;
2903 const char *reference = NULL;
2904 int ret = 0;
2905 bool implicit_backing = false;
2906 BlockDriverState *backing_hd;
2907 QDict *options;
2908 QDict *tmp_parent_options = NULL;
2909 Error *local_err = NULL;
2910
2911 if (bs->backing != NULL) {
2912 goto free_exit;
2913 }
2914
2915 /* NULL means an empty set of options */
2916 if (parent_options == NULL) {
2917 tmp_parent_options = qdict_new();
2918 parent_options = tmp_parent_options;
2919 }
2920
2921 bs->open_flags &= ~BDRV_O_NO_BACKING;
2922
2923 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2924 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2925 g_free(bdref_key_dot);
2926
2927 /*
2928 * Caution: while qdict_get_try_str() is fine, getting non-string
2929 * types would require more care. When @parent_options come from
2930 * -blockdev or blockdev_add, its members are typed according to
2931 * the QAPI schema, but when they come from -drive, they're all
2932 * QString.
2933 */
2934 reference = qdict_get_try_str(parent_options, bdref_key);
2935 if (reference || qdict_haskey(options, "file.filename")) {
2936 /* keep backing_filename NULL */
2937 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2938 qobject_unref(options);
2939 goto free_exit;
2940 } else {
2941 if (qdict_size(options) == 0) {
2942 /* If the user specifies options that do not modify the
2943 * backing file's behavior, we might still consider it the
2944 * implicit backing file. But it's easier this way, and
2945 * just specifying some of the backing BDS's options is
2946 * only possible with -drive anyway (otherwise the QAPI
2947 * schema forces the user to specify everything). */
2948 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2949 }
2950
2951 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2952 if (local_err) {
2953 ret = -EINVAL;
2954 error_propagate(errp, local_err);
2955 qobject_unref(options);
2956 goto free_exit;
2957 }
2958 }
2959
2960 if (!bs->drv || !bs->drv->supports_backing) {
2961 ret = -EINVAL;
2962 error_setg(errp, "Driver doesn't support backing files");
2963 qobject_unref(options);
2964 goto free_exit;
2965 }
2966
2967 if (!reference &&
2968 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2969 qdict_put_str(options, "driver", bs->backing_format);
2970 }
2971
2972 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2973 &child_of_bds, bdrv_backing_role(bs), errp);
2974 if (!backing_hd) {
2975 bs->open_flags |= BDRV_O_NO_BACKING;
2976 error_prepend(errp, "Could not open backing file: ");
2977 ret = -EINVAL;
2978 goto free_exit;
2979 }
2980
2981 if (implicit_backing) {
2982 bdrv_refresh_filename(backing_hd);
2983 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2984 backing_hd->filename);
2985 }
2986
2987 /* Hook up the backing file link; drop our reference, bs owns the
2988 * backing_hd reference now */
2989 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2990 bdrv_unref(backing_hd);
2991 if (local_err) {
2992 error_propagate(errp, local_err);
2993 ret = -EINVAL;
2994 goto free_exit;
2995 }
2996
2997 qdict_del(parent_options, bdref_key);
2998
2999 free_exit:
3000 g_free(backing_filename);
3001 qobject_unref(tmp_parent_options);
3002 return ret;
3003 }
3004
3005 static BlockDriverState *
3006 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3007 BlockDriverState *parent, const BdrvChildClass *child_class,
3008 BdrvChildRole child_role, bool allow_none, Error **errp)
3009 {
3010 BlockDriverState *bs = NULL;
3011 QDict *image_options;
3012 char *bdref_key_dot;
3013 const char *reference;
3014
3015 assert(child_class != NULL);
3016
3017 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3018 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3019 g_free(bdref_key_dot);
3020
3021 /*
3022 * Caution: while qdict_get_try_str() is fine, getting non-string
3023 * types would require more care. When @options come from
3024 * -blockdev or blockdev_add, its members are typed according to
3025 * the QAPI schema, but when they come from -drive, they're all
3026 * QString.
3027 */
3028 reference = qdict_get_try_str(options, bdref_key);
3029 if (!filename && !reference && !qdict_size(image_options)) {
3030 if (!allow_none) {
3031 error_setg(errp, "A block device must be specified for \"%s\"",
3032 bdref_key);
3033 }
3034 qobject_unref(image_options);
3035 goto done;
3036 }
3037
3038 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3039 parent, child_class, child_role, errp);
3040 if (!bs) {
3041 goto done;
3042 }
3043
3044 done:
3045 qdict_del(options, bdref_key);
3046 return bs;
3047 }
3048
3049 /*
3050 * Opens a disk image whose options are given as BlockdevRef in another block
3051 * device's options.
3052 *
3053 * If allow_none is true, no image will be opened if filename is false and no
3054 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3055 *
3056 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3057 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3058 * itself, all options starting with "${bdref_key}." are considered part of the
3059 * BlockdevRef.
3060 *
3061 * The BlockdevRef will be removed from the options QDict.
3062 */
3063 BdrvChild *bdrv_open_child(const char *filename,
3064 QDict *options, const char *bdref_key,
3065 BlockDriverState *parent,
3066 const BdrvChildClass *child_class,
3067 BdrvChildRole child_role,
3068 bool allow_none, Error **errp)
3069 {
3070 BlockDriverState *bs;
3071
3072 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3073 child_role, allow_none, errp);
3074 if (bs == NULL) {
3075 return NULL;
3076 }
3077
3078 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3079 errp);
3080 }
3081
3082 /*
3083 * TODO Future callers may need to specify parent/child_class in order for
3084 * option inheritance to work. Existing callers use it for the root node.
3085 */
3086 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3087 {
3088 BlockDriverState *bs = NULL;
3089 QObject *obj = NULL;
3090 QDict *qdict = NULL;
3091 const char *reference = NULL;
3092 Visitor *v = NULL;
3093
3094 if (ref->type == QTYPE_QSTRING) {
3095 reference = ref->u.reference;
3096 } else {
3097 BlockdevOptions *options = &ref->u.definition;
3098 assert(ref->type == QTYPE_QDICT);
3099
3100 v = qobject_output_visitor_new(&obj);
3101 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3102 visit_complete(v, &obj);
3103
3104 qdict = qobject_to(QDict, obj);
3105 qdict_flatten(qdict);
3106
3107 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3108 * compatibility with other callers) rather than what we want as the
3109 * real defaults. Apply the defaults here instead. */
3110 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3111 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3112 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3113 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3114
3115 }
3116
3117 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3118 obj = NULL;
3119 qobject_unref(obj);
3120 visit_free(v);
3121 return bs;
3122 }
3123
3124 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3125 int flags,
3126 QDict *snapshot_options,
3127 Error **errp)
3128 {
3129 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3130 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3131 int64_t total_size;
3132 QemuOpts *opts = NULL;
3133 BlockDriverState *bs_snapshot = NULL;
3134 Error *local_err = NULL;
3135 int ret;
3136
3137 /* if snapshot, we create a temporary backing file and open it
3138 instead of opening 'filename' directly */
3139
3140 /* Get the required size from the image */
3141 total_size = bdrv_getlength(bs);
3142 if (total_size < 0) {
3143 error_setg_errno(errp, -total_size, "Could not get image size");
3144 goto out;
3145 }
3146
3147 /* Create the temporary image */
3148 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3149 if (ret < 0) {
3150 error_setg_errno(errp, -ret, "Could not get temporary filename");
3151 goto out;
3152 }
3153
3154 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3155 &error_abort);
3156 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3157 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3158 qemu_opts_del(opts);
3159 if (ret < 0) {
3160 error_prepend(errp, "Could not create temporary overlay '%s': ",
3161 tmp_filename);
3162 goto out;
3163 }
3164
3165 /* Prepare options QDict for the temporary file */
3166 qdict_put_str(snapshot_options, "file.driver", "file");
3167 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3168 qdict_put_str(snapshot_options, "driver", "qcow2");
3169
3170 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3171 snapshot_options = NULL;
3172 if (!bs_snapshot) {
3173 goto out;
3174 }
3175
3176 /* bdrv_append() consumes a strong reference to bs_snapshot
3177 * (i.e. it will call bdrv_unref() on it) even on error, so in
3178 * order to be able to return one, we have to increase
3179 * bs_snapshot's refcount here */
3180 bdrv_ref(bs_snapshot);
3181 bdrv_append(bs_snapshot, bs, &local_err);
3182 if (local_err) {
3183 error_propagate(errp, local_err);
3184 bs_snapshot = NULL;
3185 goto out;
3186 }
3187
3188 out:
3189 qobject_unref(snapshot_options);
3190 g_free(tmp_filename);
3191 return bs_snapshot;
3192 }
3193
3194 /*
3195 * Opens a disk image (raw, qcow2, vmdk, ...)
3196 *
3197 * options is a QDict of options to pass to the block drivers, or NULL for an
3198 * empty set of options. The reference to the QDict belongs to the block layer
3199 * after the call (even on failure), so if the caller intends to reuse the
3200 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3201 *
3202 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3203 * If it is not NULL, the referenced BDS will be reused.
3204 *
3205 * The reference parameter may be used to specify an existing block device which
3206 * should be opened. If specified, neither options nor a filename may be given,
3207 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3208 */
3209 static BlockDriverState *bdrv_open_inherit(const char *filename,
3210 const char *reference,
3211 QDict *options, int flags,
3212 BlockDriverState *parent,
3213 const BdrvChildClass *child_class,
3214 BdrvChildRole child_role,
3215 Error **errp)
3216 {
3217 int ret;
3218 BlockBackend *file = NULL;
3219 BlockDriverState *bs;
3220 BlockDriver *drv = NULL;
3221 BdrvChild *child;
3222 const char *drvname;
3223 const char *backing;
3224 Error *local_err = NULL;
3225 QDict *snapshot_options = NULL;
3226 int snapshot_flags = 0;
3227
3228 assert(!child_class || !flags);
3229 assert(!child_class == !parent);
3230
3231 if (reference) {
3232 bool options_non_empty = options ? qdict_size(options) : false;
3233 qobject_unref(options);
3234
3235 if (filename || options_non_empty) {
3236 error_setg(errp, "Cannot reference an existing block device with "
3237 "additional options or a new filename");
3238 return NULL;
3239 }
3240
3241 bs = bdrv_lookup_bs(reference, reference, errp);
3242 if (!bs) {
3243 return NULL;
3244 }
3245
3246 bdrv_ref(bs);
3247 return bs;
3248 }
3249
3250 bs = bdrv_new();
3251
3252 /* NULL means an empty set of options */
3253 if (options == NULL) {
3254 options = qdict_new();
3255 }
3256
3257 /* json: syntax counts as explicit options, as if in the QDict */
3258 parse_json_protocol(options, &filename, &local_err);
3259 if (local_err) {
3260 goto fail;
3261 }
3262
3263 bs->explicit_options = qdict_clone_shallow(options);
3264
3265 if (child_class) {
3266 bool parent_is_format;
3267
3268 if (parent->drv) {
3269 parent_is_format = parent->drv->is_format;
3270 } else {
3271 /*
3272 * parent->drv is not set yet because this node is opened for
3273 * (potential) format probing. That means that @parent is going
3274 * to be a format node.
3275 */
3276 parent_is_format = true;
3277 }
3278
3279 bs->inherits_from = parent;
3280 child_class->inherit_options(child_role, parent_is_format,
3281 &flags, options,
3282 parent->open_flags, parent->options);
3283 }
3284
3285 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3286 if (ret < 0) {
3287 goto fail;
3288 }
3289
3290 /*
3291 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3292 * Caution: getting a boolean member of @options requires care.
3293 * When @options come from -blockdev or blockdev_add, members are
3294 * typed according to the QAPI schema, but when they come from
3295 * -drive, they're all QString.
3296 */
3297 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3298 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3299 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3300 } else {
3301 flags &= ~BDRV_O_RDWR;
3302 }
3303
3304 if (flags & BDRV_O_SNAPSHOT) {
3305 snapshot_options = qdict_new();
3306 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3307 flags, options);
3308 /* Let bdrv_backing_options() override "read-only" */
3309 qdict_del(options, BDRV_OPT_READ_ONLY);
3310 bdrv_inherited_options(BDRV_CHILD_COW, true,
3311 &flags, options, flags, options);
3312 }
3313
3314 bs->open_flags = flags;
3315 bs->options = options;
3316 options = qdict_clone_shallow(options);
3317
3318 /* Find the right image format driver */
3319 /* See cautionary note on accessing @options above */
3320 drvname = qdict_get_try_str(options, "driver");
3321 if (drvname) {
3322 drv = bdrv_find_format(drvname);
3323 if (!drv) {
3324 error_setg(errp, "Unknown driver: '%s'", drvname);
3325 goto fail;
3326 }
3327 }
3328
3329 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3330
3331 /* See cautionary note on accessing @options above */
3332 backing = qdict_get_try_str(options, "backing");
3333 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3334 (backing && *backing == '\0'))
3335 {
3336 if (backing) {
3337 warn_report("Use of \"backing\": \"\" is deprecated; "
3338 "use \"backing\": null instead");
3339 }
3340 flags |= BDRV_O_NO_BACKING;
3341 qdict_del(bs->explicit_options, "backing");
3342 qdict_del(bs->options, "backing");
3343 qdict_del(options, "backing");
3344 }
3345
3346 /* Open image file without format layer. This BlockBackend is only used for
3347 * probing, the block drivers will do their own bdrv_open_child() for the
3348 * same BDS, which is why we put the node name back into options. */
3349 if ((flags & BDRV_O_PROTOCOL) == 0) {
3350 BlockDriverState *file_bs;
3351
3352 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3353 &child_of_bds, BDRV_CHILD_IMAGE,
3354 true, &local_err);
3355 if (local_err) {
3356 goto fail;
3357 }
3358 if (file_bs != NULL) {
3359 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3360 * looking at the header to guess the image format. This works even
3361 * in cases where a guest would not see a consistent state. */
3362 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3363 blk_insert_bs(file, file_bs, &local_err);
3364 bdrv_unref(file_bs);
3365 if (local_err) {
3366 goto fail;
3367 }
3368
3369 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3370 }
3371 }
3372
3373 /* Image format probing */
3374 bs->probed = !drv;
3375 if (!drv && file) {
3376 ret = find_image_format(file, filename, &drv, &local_err);
3377 if (ret < 0) {
3378 goto fail;
3379 }
3380 /*
3381 * This option update would logically belong in bdrv_fill_options(),
3382 * but we first need to open bs->file for the probing to work, while
3383 * opening bs->file already requires the (mostly) final set of options
3384 * so that cache mode etc. can be inherited.
3385 *
3386 * Adding the driver later is somewhat ugly, but it's not an option
3387 * that would ever be inherited, so it's correct. We just need to make
3388 * sure to update both bs->options (which has the full effective
3389 * options for bs) and options (which has file.* already removed).
3390 */
3391 qdict_put_str(bs->options, "driver", drv->format_name);
3392 qdict_put_str(options, "driver", drv->format_name);
3393 } else if (!drv) {
3394 error_setg(errp, "Must specify either driver or file");
3395 goto fail;
3396 }
3397
3398 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3399 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3400 /* file must be NULL if a protocol BDS is about to be created
3401 * (the inverse results in an error message from bdrv_open_common()) */
3402 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3403
3404 /* Open the image */
3405 ret = bdrv_open_common(bs, file, options, &local_err);
3406 if (ret < 0) {
3407 goto fail;
3408 }
3409
3410 if (file) {
3411 blk_unref(file);
3412 file = NULL;
3413 }
3414
3415 /* If there is a backing file, use it */
3416 if ((flags & BDRV_O_NO_BACKING) == 0) {
3417 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3418 if (ret < 0) {
3419 goto close_and_fail;
3420 }
3421 }
3422
3423 /* Remove all children options and references
3424 * from bs->options and bs->explicit_options */
3425 QLIST_FOREACH(child, &bs->children, next) {
3426 char *child_key_dot;
3427 child_key_dot = g_strdup_printf("%s.", child->name);
3428 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3429 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3430 qdict_del(bs->explicit_options, child->name);
3431 qdict_del(bs->options, child->name);
3432 g_free(child_key_dot);
3433 }
3434
3435 /* Check if any unknown options were used */
3436 if (qdict_size(options) != 0) {
3437 const QDictEntry *entry = qdict_first(options);
3438 if (flags & BDRV_O_PROTOCOL) {
3439 error_setg(errp, "Block protocol '%s' doesn't support the option "
3440 "'%s'", drv->format_name, entry->key);
3441 } else {
3442 error_setg(errp,
3443 "Block format '%s' does not support the option '%s'",
3444 drv->format_name, entry->key);
3445 }
3446
3447 goto close_and_fail;
3448 }
3449
3450 bdrv_parent_cb_change_media(bs, true);
3451
3452 qobject_unref(options);
3453 options = NULL;
3454
3455 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3456 * temporary snapshot afterwards. */
3457 if (snapshot_flags) {
3458 BlockDriverState *snapshot_bs;
3459 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3460 snapshot_options, &local_err);
3461 snapshot_options = NULL;
3462 if (local_err) {
3463 goto close_and_fail;
3464 }
3465 /* We are not going to return bs but the overlay on top of it
3466 * (snapshot_bs); thus, we have to drop the strong reference to bs
3467 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3468 * though, because the overlay still has a reference to it. */
3469 bdrv_unref(bs);
3470 bs = snapshot_bs;
3471 }
3472
3473 return bs;
3474
3475 fail:
3476 blk_unref(file);
3477 qobject_unref(snapshot_options);
3478 qobject_unref(bs->explicit_options);
3479 qobject_unref(bs->options);
3480 qobject_unref(options);
3481 bs->options = NULL;
3482 bs->explicit_options = NULL;
3483 bdrv_unref(bs);
3484 error_propagate(errp, local_err);
3485 return NULL;
3486
3487 close_and_fail:
3488 bdrv_unref(bs);
3489 qobject_unref(snapshot_options);
3490 qobject_unref(options);
3491 error_propagate(errp, local_err);
3492 return NULL;
3493 }
3494
3495 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3496 QDict *options, int flags, Error **errp)
3497 {
3498 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3499 NULL, 0, errp);
3500 }
3501
3502 /* Return true if the NULL-terminated @list contains @str */
3503 static bool is_str_in_list(const char *str, const char *const *list)
3504 {
3505 if (str && list) {
3506 int i;
3507 for (i = 0; list[i] != NULL; i++) {
3508 if (!strcmp(str, list[i])) {
3509 return true;
3510 }
3511 }
3512 }
3513 return false;
3514 }
3515
3516 /*
3517 * Check that every option set in @bs->options is also set in
3518 * @new_opts.
3519 *
3520 * Options listed in the common_options list and in
3521 * @bs->drv->mutable_opts are skipped.
3522 *
3523 * Return 0 on success, otherwise return -EINVAL and set @errp.
3524 */
3525 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3526 const QDict *new_opts, Error **errp)
3527 {
3528 const QDictEntry *e;
3529 /* These options are common to all block drivers and are handled
3530 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3531 const char *const common_options[] = {
3532 "node-name", "discard", "cache.direct", "cache.no-flush",
3533 "read-only", "auto-read-only", "detect-zeroes", NULL
3534 };
3535
3536 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3537 if (!qdict_haskey(new_opts, e->key) &&
3538 !is_str_in_list(e->key, common_options) &&
3539 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3540 error_setg(errp, "Option '%s' cannot be reset "
3541 "to its default value", e->key);
3542 return -EINVAL;
3543 }
3544 }
3545
3546 return 0;
3547 }
3548
3549 /*
3550 * Returns true if @child can be reached recursively from @bs
3551 */
3552 static bool bdrv_recurse_has_child(BlockDriverState *bs,
3553 BlockDriverState *child)
3554 {
3555 BdrvChild *c;
3556
3557 if (bs == child) {
3558 return true;
3559 }