target/arm: Convert Neon VCVT fixed-point to gvec
[qemu.git] / block.c
1 /*
2 * QEMU System Emulator block driver
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "block/module_block.h"
33 #include "qemu/main-loop.h"
34 #include "qemu/module.h"
35 #include "qapi/error.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qjson.h"
38 #include "qapi/qmp/qnull.h"
39 #include "qapi/qmp/qstring.h"
40 #include "qapi/qobject-output-visitor.h"
41 #include "qapi/qapi-visit-block-core.h"
42 #include "sysemu/block-backend.h"
43 #include "sysemu/sysemu.h"
44 #include "qemu/notify.h"
45 #include "qemu/option.h"
46 #include "qemu/coroutine.h"
47 #include "block/qapi.h"
48 #include "qemu/timer.h"
49 #include "qemu/cutils.h"
50 #include "qemu/id.h"
51
52 #ifdef CONFIG_BSD
53 #include <sys/ioctl.h>
54 #include <sys/queue.h>
55 #ifndef __DragonFly__
56 #include <sys/disk.h>
57 #endif
58 #endif
59
60 #ifdef _WIN32
61 #include <windows.h>
62 #endif
63
64 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
65
66 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
67 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
68
69 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
70 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
71
72 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
73 QLIST_HEAD_INITIALIZER(bdrv_drivers);
74
75 static BlockDriverState *bdrv_open_inherit(const char *filename,
76 const char *reference,
77 QDict *options, int flags,
78 BlockDriverState *parent,
79 const BdrvChildClass *child_class,
80 BdrvChildRole child_role,
81 Error **errp);
82
83 /* If non-zero, use only whitelisted block drivers */
84 static int use_bdrv_whitelist;
85
86 #ifdef _WIN32
87 static int is_windows_drive_prefix(const char *filename)
88 {
89 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
90 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
91 filename[1] == ':');
92 }
93
94 int is_windows_drive(const char *filename)
95 {
96 if (is_windows_drive_prefix(filename) &&
97 filename[2] == '\0')
98 return 1;
99 if (strstart(filename, "\\\\.\\", NULL) ||
100 strstart(filename, "//./", NULL))
101 return 1;
102 return 0;
103 }
104 #endif
105
106 size_t bdrv_opt_mem_align(BlockDriverState *bs)
107 {
108 if (!bs || !bs->drv) {
109 /* page size or 4k (hdd sector size) should be on the safe side */
110 return MAX(4096, qemu_real_host_page_size);
111 }
112
113 return bs->bl.opt_mem_alignment;
114 }
115
116 size_t bdrv_min_mem_align(BlockDriverState *bs)
117 {
118 if (!bs || !bs->drv) {
119 /* page size or 4k (hdd sector size) should be on the safe side */
120 return MAX(4096, qemu_real_host_page_size);
121 }
122
123 return bs->bl.min_mem_alignment;
124 }
125
126 /* check if the path starts with "<protocol>:" */
127 int path_has_protocol(const char *path)
128 {
129 const char *p;
130
131 #ifdef _WIN32
132 if (is_windows_drive(path) ||
133 is_windows_drive_prefix(path)) {
134 return 0;
135 }
136 p = path + strcspn(path, ":/\\");
137 #else
138 p = path + strcspn(path, ":/");
139 #endif
140
141 return *p == ':';
142 }
143
144 int path_is_absolute(const char *path)
145 {
146 #ifdef _WIN32
147 /* specific case for names like: "\\.\d:" */
148 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
149 return 1;
150 }
151 return (*path == '/' || *path == '\\');
152 #else
153 return (*path == '/');
154 #endif
155 }
156
157 /* if filename is absolute, just return its duplicate. Otherwise, build a
158 path to it by considering it is relative to base_path. URL are
159 supported. */
160 char *path_combine(const char *base_path, const char *filename)
161 {
162 const char *protocol_stripped = NULL;
163 const char *p, *p1;
164 char *result;
165 int len;
166
167 if (path_is_absolute(filename)) {
168 return g_strdup(filename);
169 }
170
171 if (path_has_protocol(base_path)) {
172 protocol_stripped = strchr(base_path, ':');
173 if (protocol_stripped) {
174 protocol_stripped++;
175 }
176 }
177 p = protocol_stripped ?: base_path;
178
179 p1 = strrchr(base_path, '/');
180 #ifdef _WIN32
181 {
182 const char *p2;
183 p2 = strrchr(base_path, '\\');
184 if (!p1 || p2 > p1) {
185 p1 = p2;
186 }
187 }
188 #endif
189 if (p1) {
190 p1++;
191 } else {
192 p1 = base_path;
193 }
194 if (p1 > p) {
195 p = p1;
196 }
197 len = p - base_path;
198
199 result = g_malloc(len + strlen(filename) + 1);
200 memcpy(result, base_path, len);
201 strcpy(result + len, filename);
202
203 return result;
204 }
205
206 /*
207 * Helper function for bdrv_parse_filename() implementations to remove optional
208 * protocol prefixes (especially "file:") from a filename and for putting the
209 * stripped filename into the options QDict if there is such a prefix.
210 */
211 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
212 QDict *options)
213 {
214 if (strstart(filename, prefix, &filename)) {
215 /* Stripping the explicit protocol prefix may result in a protocol
216 * prefix being (wrongly) detected (if the filename contains a colon) */
217 if (path_has_protocol(filename)) {
218 QString *fat_filename;
219
220 /* This means there is some colon before the first slash; therefore,
221 * this cannot be an absolute path */
222 assert(!path_is_absolute(filename));
223
224 /* And we can thus fix the protocol detection issue by prefixing it
225 * by "./" */
226 fat_filename = qstring_from_str("./");
227 qstring_append(fat_filename, filename);
228
229 assert(!path_has_protocol(qstring_get_str(fat_filename)));
230
231 qdict_put(options, "filename", fat_filename);
232 } else {
233 /* If no protocol prefix was detected, we can use the shortened
234 * filename as-is */
235 qdict_put_str(options, "filename", filename);
236 }
237 }
238 }
239
240
241 /* Returns whether the image file is opened as read-only. Note that this can
242 * return false and writing to the image file is still not possible because the
243 * image is inactivated. */
244 bool bdrv_is_read_only(BlockDriverState *bs)
245 {
246 return bs->read_only;
247 }
248
249 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
250 bool ignore_allow_rdw, Error **errp)
251 {
252 /* Do not set read_only if copy_on_read is enabled */
253 if (bs->copy_on_read && read_only) {
254 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
255 bdrv_get_device_or_node_name(bs));
256 return -EINVAL;
257 }
258
259 /* Do not clear read_only if it is prohibited */
260 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
261 !ignore_allow_rdw)
262 {
263 error_setg(errp, "Node '%s' is read only",
264 bdrv_get_device_or_node_name(bs));
265 return -EPERM;
266 }
267
268 return 0;
269 }
270
271 /*
272 * Called by a driver that can only provide a read-only image.
273 *
274 * Returns 0 if the node is already read-only or it could switch the node to
275 * read-only because BDRV_O_AUTO_RDONLY is set.
276 *
277 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
278 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
279 * is not NULL, it is used as the error message for the Error object.
280 */
281 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
282 Error **errp)
283 {
284 int ret = 0;
285
286 if (!(bs->open_flags & BDRV_O_RDWR)) {
287 return 0;
288 }
289 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
290 goto fail;
291 }
292
293 ret = bdrv_can_set_read_only(bs, true, false, NULL);
294 if (ret < 0) {
295 goto fail;
296 }
297
298 bs->read_only = true;
299 bs->open_flags &= ~BDRV_O_RDWR;
300
301 return 0;
302
303 fail:
304 error_setg(errp, "%s", errmsg ?: "Image is read-only");
305 return -EACCES;
306 }
307
308 /*
309 * If @backing is empty, this function returns NULL without setting
310 * @errp. In all other cases, NULL will only be returned with @errp
311 * set.
312 *
313 * Therefore, a return value of NULL without @errp set means that
314 * there is no backing file; if @errp is set, there is one but its
315 * absolute filename cannot be generated.
316 */
317 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
318 const char *backing,
319 Error **errp)
320 {
321 if (backing[0] == '\0') {
322 return NULL;
323 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
324 return g_strdup(backing);
325 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
326 error_setg(errp, "Cannot use relative backing file names for '%s'",
327 backed);
328 return NULL;
329 } else {
330 return path_combine(backed, backing);
331 }
332 }
333
334 /*
335 * If @filename is empty or NULL, this function returns NULL without
336 * setting @errp. In all other cases, NULL will only be returned with
337 * @errp set.
338 */
339 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
340 const char *filename, Error **errp)
341 {
342 char *dir, *full_name;
343
344 if (!filename || filename[0] == '\0') {
345 return NULL;
346 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
347 return g_strdup(filename);
348 }
349
350 dir = bdrv_dirname(relative_to, errp);
351 if (!dir) {
352 return NULL;
353 }
354
355 full_name = g_strconcat(dir, filename, NULL);
356 g_free(dir);
357 return full_name;
358 }
359
360 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
361 {
362 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
363 }
364
365 void bdrv_register(BlockDriver *bdrv)
366 {
367 assert(bdrv->format_name);
368 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
369 }
370
371 BlockDriverState *bdrv_new(void)
372 {
373 BlockDriverState *bs;
374 int i;
375
376 bs = g_new0(BlockDriverState, 1);
377 QLIST_INIT(&bs->dirty_bitmaps);
378 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
379 QLIST_INIT(&bs->op_blockers[i]);
380 }
381 notifier_with_return_list_init(&bs->before_write_notifiers);
382 qemu_co_mutex_init(&bs->reqs_lock);
383 qemu_mutex_init(&bs->dirty_bitmap_mutex);
384 bs->refcnt = 1;
385 bs->aio_context = qemu_get_aio_context();
386
387 qemu_co_queue_init(&bs->flush_queue);
388
389 for (i = 0; i < bdrv_drain_all_count; i++) {
390 bdrv_drained_begin(bs);
391 }
392
393 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
394
395 return bs;
396 }
397
398 static BlockDriver *bdrv_do_find_format(const char *format_name)
399 {
400 BlockDriver *drv1;
401
402 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
403 if (!strcmp(drv1->format_name, format_name)) {
404 return drv1;
405 }
406 }
407
408 return NULL;
409 }
410
411 BlockDriver *bdrv_find_format(const char *format_name)
412 {
413 BlockDriver *drv1;
414 int i;
415
416 drv1 = bdrv_do_find_format(format_name);
417 if (drv1) {
418 return drv1;
419 }
420
421 /* The driver isn't registered, maybe we need to load a module */
422 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
423 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
424 block_module_load_one(block_driver_modules[i].library_name);
425 break;
426 }
427 }
428
429 return bdrv_do_find_format(format_name);
430 }
431
432 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
433 {
434 static const char *whitelist_rw[] = {
435 CONFIG_BDRV_RW_WHITELIST
436 NULL
437 };
438 static const char *whitelist_ro[] = {
439 CONFIG_BDRV_RO_WHITELIST
440 NULL
441 };
442 const char **p;
443
444 if (!whitelist_rw[0] && !whitelist_ro[0]) {
445 return 1; /* no whitelist, anything goes */
446 }
447
448 for (p = whitelist_rw; *p; p++) {
449 if (!strcmp(format_name, *p)) {
450 return 1;
451 }
452 }
453 if (read_only) {
454 for (p = whitelist_ro; *p; p++) {
455 if (!strcmp(format_name, *p)) {
456 return 1;
457 }
458 }
459 }
460 return 0;
461 }
462
463 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
464 {
465 return bdrv_format_is_whitelisted(drv->format_name, read_only);
466 }
467
468 bool bdrv_uses_whitelist(void)
469 {
470 return use_bdrv_whitelist;
471 }
472
473 typedef struct CreateCo {
474 BlockDriver *drv;
475 char *filename;
476 QemuOpts *opts;
477 int ret;
478 Error *err;
479 } CreateCo;
480
481 static void coroutine_fn bdrv_create_co_entry(void *opaque)
482 {
483 Error *local_err = NULL;
484 int ret;
485
486 CreateCo *cco = opaque;
487 assert(cco->drv);
488
489 ret = cco->drv->bdrv_co_create_opts(cco->drv,
490 cco->filename, cco->opts, &local_err);
491 error_propagate(&cco->err, local_err);
492 cco->ret = ret;
493 }
494
495 int bdrv_create(BlockDriver *drv, const char* filename,
496 QemuOpts *opts, Error **errp)
497 {
498 int ret;
499
500 Coroutine *co;
501 CreateCo cco = {
502 .drv = drv,
503 .filename = g_strdup(filename),
504 .opts = opts,
505 .ret = NOT_DONE,
506 .err = NULL,
507 };
508
509 if (!drv->bdrv_co_create_opts) {
510 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
511 ret = -ENOTSUP;
512 goto out;
513 }
514
515 if (qemu_in_coroutine()) {
516 /* Fast-path if already in coroutine context */
517 bdrv_create_co_entry(&cco);
518 } else {
519 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
520 qemu_coroutine_enter(co);
521 while (cco.ret == NOT_DONE) {
522 aio_poll(qemu_get_aio_context(), true);
523 }
524 }
525
526 ret = cco.ret;
527 if (ret < 0) {
528 if (cco.err) {
529 error_propagate(errp, cco.err);
530 } else {
531 error_setg_errno(errp, -ret, "Could not create image");
532 }
533 }
534
535 out:
536 g_free(cco.filename);
537 return ret;
538 }
539
540 /**
541 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
542 * least the given @minimum_size.
543 *
544 * On success, return @blk's actual length.
545 * Otherwise, return -errno.
546 */
547 static int64_t create_file_fallback_truncate(BlockBackend *blk,
548 int64_t minimum_size, Error **errp)
549 {
550 Error *local_err = NULL;
551 int64_t size;
552 int ret;
553
554 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
555 &local_err);
556 if (ret < 0 && ret != -ENOTSUP) {
557 error_propagate(errp, local_err);
558 return ret;
559 }
560
561 size = blk_getlength(blk);
562 if (size < 0) {
563 error_free(local_err);
564 error_setg_errno(errp, -size,
565 "Failed to inquire the new image file's length");
566 return size;
567 }
568
569 if (size < minimum_size) {
570 /* Need to grow the image, but we failed to do that */
571 error_propagate(errp, local_err);
572 return -ENOTSUP;
573 }
574
575 error_free(local_err);
576 local_err = NULL;
577
578 return size;
579 }
580
581 /**
582 * Helper function for bdrv_create_file_fallback(): Zero the first
583 * sector to remove any potentially pre-existing image header.
584 */
585 static int create_file_fallback_zero_first_sector(BlockBackend *blk,
586 int64_t current_size,
587 Error **errp)
588 {
589 int64_t bytes_to_clear;
590 int ret;
591
592 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
593 if (bytes_to_clear) {
594 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
595 if (ret < 0) {
596 error_setg_errno(errp, -ret,
597 "Failed to clear the new image's first sector");
598 return ret;
599 }
600 }
601
602 return 0;
603 }
604
605 /**
606 * Simple implementation of bdrv_co_create_opts for protocol drivers
607 * which only support creation via opening a file
608 * (usually existing raw storage device)
609 */
610 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
611 const char *filename,
612 QemuOpts *opts,
613 Error **errp)
614 {
615 BlockBackend *blk;
616 QDict *options;
617 int64_t size = 0;
618 char *buf = NULL;
619 PreallocMode prealloc;
620 Error *local_err = NULL;
621 int ret;
622
623 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
624 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
625 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
626 PREALLOC_MODE_OFF, &local_err);
627 g_free(buf);
628 if (local_err) {
629 error_propagate(errp, local_err);
630 return -EINVAL;
631 }
632
633 if (prealloc != PREALLOC_MODE_OFF) {
634 error_setg(errp, "Unsupported preallocation mode '%s'",
635 PreallocMode_str(prealloc));
636 return -ENOTSUP;
637 }
638
639 options = qdict_new();
640 qdict_put_str(options, "driver", drv->format_name);
641
642 blk = blk_new_open(filename, NULL, options,
643 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
644 if (!blk) {
645 error_prepend(errp, "Protocol driver '%s' does not support image "
646 "creation, and opening the image failed: ",
647 drv->format_name);
648 return -EINVAL;
649 }
650
651 size = create_file_fallback_truncate(blk, size, errp);
652 if (size < 0) {
653 ret = size;
654 goto out;
655 }
656
657 ret = create_file_fallback_zero_first_sector(blk, size, errp);
658 if (ret < 0) {
659 goto out;
660 }
661
662 ret = 0;
663 out:
664 blk_unref(blk);
665 return ret;
666 }
667
668 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
669 {
670 BlockDriver *drv;
671
672 drv = bdrv_find_protocol(filename, true, errp);
673 if (drv == NULL) {
674 return -ENOENT;
675 }
676
677 return bdrv_create(drv, filename, opts, errp);
678 }
679
680 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
681 {
682 Error *local_err = NULL;
683 int ret;
684
685 assert(bs != NULL);
686
687 if (!bs->drv) {
688 error_setg(errp, "Block node '%s' is not opened", bs->filename);
689 return -ENOMEDIUM;
690 }
691
692 if (!bs->drv->bdrv_co_delete_file) {
693 error_setg(errp, "Driver '%s' does not support image deletion",
694 bs->drv->format_name);
695 return -ENOTSUP;
696 }
697
698 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
699 if (ret < 0) {
700 error_propagate(errp, local_err);
701 }
702
703 return ret;
704 }
705
706 /**
707 * Try to get @bs's logical and physical block size.
708 * On success, store them in @bsz struct and return 0.
709 * On failure return -errno.
710 * @bs must not be empty.
711 */
712 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
713 {
714 BlockDriver *drv = bs->drv;
715
716 if (drv && drv->bdrv_probe_blocksizes) {
717 return drv->bdrv_probe_blocksizes(bs, bsz);
718 } else if (drv && drv->is_filter && bs->file) {
719 return bdrv_probe_blocksizes(bs->file->bs, bsz);
720 }
721
722 return -ENOTSUP;
723 }
724
725 /**
726 * Try to get @bs's geometry (cyls, heads, sectors).
727 * On success, store them in @geo struct and return 0.
728 * On failure return -errno.
729 * @bs must not be empty.
730 */
731 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
732 {
733 BlockDriver *drv = bs->drv;
734
735 if (drv && drv->bdrv_probe_geometry) {
736 return drv->bdrv_probe_geometry(bs, geo);
737 } else if (drv && drv->is_filter && bs->file) {
738 return bdrv_probe_geometry(bs->file->bs, geo);
739 }
740
741 return -ENOTSUP;
742 }
743
744 /*
745 * Create a uniquely-named empty temporary file.
746 * Return 0 upon success, otherwise a negative errno value.
747 */
748 int get_tmp_filename(char *filename, int size)
749 {
750 #ifdef _WIN32
751 char temp_dir[MAX_PATH];
752 /* GetTempFileName requires that its output buffer (4th param)
753 have length MAX_PATH or greater. */
754 assert(size >= MAX_PATH);
755 return (GetTempPath(MAX_PATH, temp_dir)
756 && GetTempFileName(temp_dir, "qem", 0, filename)
757 ? 0 : -GetLastError());
758 #else
759 int fd;
760 const char *tmpdir;
761 tmpdir = getenv("TMPDIR");
762 if (!tmpdir) {
763 tmpdir = "/var/tmp";
764 }
765 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
766 return -EOVERFLOW;
767 }
768 fd = mkstemp(filename);
769 if (fd < 0) {
770 return -errno;
771 }
772 if (close(fd) != 0) {
773 unlink(filename);
774 return -errno;
775 }
776 return 0;
777 #endif
778 }
779
780 /*
781 * Detect host devices. By convention, /dev/cdrom[N] is always
782 * recognized as a host CDROM.
783 */
784 static BlockDriver *find_hdev_driver(const char *filename)
785 {
786 int score_max = 0, score;
787 BlockDriver *drv = NULL, *d;
788
789 QLIST_FOREACH(d, &bdrv_drivers, list) {
790 if (d->bdrv_probe_device) {
791 score = d->bdrv_probe_device(filename);
792 if (score > score_max) {
793 score_max = score;
794 drv = d;
795 }
796 }
797 }
798
799 return drv;
800 }
801
802 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
803 {
804 BlockDriver *drv1;
805
806 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
807 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
808 return drv1;
809 }
810 }
811
812 return NULL;
813 }
814
815 BlockDriver *bdrv_find_protocol(const char *filename,
816 bool allow_protocol_prefix,
817 Error **errp)
818 {
819 BlockDriver *drv1;
820 char protocol[128];
821 int len;
822 const char *p;
823 int i;
824
825 /* TODO Drivers without bdrv_file_open must be specified explicitly */
826
827 /*
828 * XXX(hch): we really should not let host device detection
829 * override an explicit protocol specification, but moving this
830 * later breaks access to device names with colons in them.
831 * Thanks to the brain-dead persistent naming schemes on udev-
832 * based Linux systems those actually are quite common.
833 */
834 drv1 = find_hdev_driver(filename);
835 if (drv1) {
836 return drv1;
837 }
838
839 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
840 return &bdrv_file;
841 }
842
843 p = strchr(filename, ':');
844 assert(p != NULL);
845 len = p - filename;
846 if (len > sizeof(protocol) - 1)
847 len = sizeof(protocol) - 1;
848 memcpy(protocol, filename, len);
849 protocol[len] = '\0';
850
851 drv1 = bdrv_do_find_protocol(protocol);
852 if (drv1) {
853 return drv1;
854 }
855
856 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
857 if (block_driver_modules[i].protocol_name &&
858 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
859 block_module_load_one(block_driver_modules[i].library_name);
860 break;
861 }
862 }
863
864 drv1 = bdrv_do_find_protocol(protocol);
865 if (!drv1) {
866 error_setg(errp, "Unknown protocol '%s'", protocol);
867 }
868 return drv1;
869 }
870
871 /*
872 * Guess image format by probing its contents.
873 * This is not a good idea when your image is raw (CVE-2008-2004), but
874 * we do it anyway for backward compatibility.
875 *
876 * @buf contains the image's first @buf_size bytes.
877 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
878 * but can be smaller if the image file is smaller)
879 * @filename is its filename.
880 *
881 * For all block drivers, call the bdrv_probe() method to get its
882 * probing score.
883 * Return the first block driver with the highest probing score.
884 */
885 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
886 const char *filename)
887 {
888 int score_max = 0, score;
889 BlockDriver *drv = NULL, *d;
890
891 QLIST_FOREACH(d, &bdrv_drivers, list) {
892 if (d->bdrv_probe) {
893 score = d->bdrv_probe(buf, buf_size, filename);
894 if (score > score_max) {
895 score_max = score;
896 drv = d;
897 }
898 }
899 }
900
901 return drv;
902 }
903
904 static int find_image_format(BlockBackend *file, const char *filename,
905 BlockDriver **pdrv, Error **errp)
906 {
907 BlockDriver *drv;
908 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
909 int ret = 0;
910
911 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
912 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
913 *pdrv = &bdrv_raw;
914 return ret;
915 }
916
917 ret = blk_pread(file, 0, buf, sizeof(buf));
918 if (ret < 0) {
919 error_setg_errno(errp, -ret, "Could not read image for determining its "
920 "format");
921 *pdrv = NULL;
922 return ret;
923 }
924
925 drv = bdrv_probe_all(buf, ret, filename);
926 if (!drv) {
927 error_setg(errp, "Could not determine image format: No compatible "
928 "driver found");
929 ret = -ENOENT;
930 }
931 *pdrv = drv;
932 return ret;
933 }
934
935 /**
936 * Set the current 'total_sectors' value
937 * Return 0 on success, -errno on error.
938 */
939 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
940 {
941 BlockDriver *drv = bs->drv;
942
943 if (!drv) {
944 return -ENOMEDIUM;
945 }
946
947 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
948 if (bdrv_is_sg(bs))
949 return 0;
950
951 /* query actual device if possible, otherwise just trust the hint */
952 if (drv->bdrv_getlength) {
953 int64_t length = drv->bdrv_getlength(bs);
954 if (length < 0) {
955 return length;
956 }
957 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
958 }
959
960 bs->total_sectors = hint;
961 return 0;
962 }
963
964 /**
965 * Combines a QDict of new block driver @options with any missing options taken
966 * from @old_options, so that leaving out an option defaults to its old value.
967 */
968 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
969 QDict *old_options)
970 {
971 if (bs->drv && bs->drv->bdrv_join_options) {
972 bs->drv->bdrv_join_options(options, old_options);
973 } else {
974 qdict_join(options, old_options, false);
975 }
976 }
977
978 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
979 int open_flags,
980 Error **errp)
981 {
982 Error *local_err = NULL;
983 char *value = qemu_opt_get_del(opts, "detect-zeroes");
984 BlockdevDetectZeroesOptions detect_zeroes =
985 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
986 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
987 g_free(value);
988 if (local_err) {
989 error_propagate(errp, local_err);
990 return detect_zeroes;
991 }
992
993 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
994 !(open_flags & BDRV_O_UNMAP))
995 {
996 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
997 "without setting discard operation to unmap");
998 }
999
1000 return detect_zeroes;
1001 }
1002
1003 /**
1004 * Set open flags for aio engine
1005 *
1006 * Return 0 on success, -1 if the engine specified is invalid
1007 */
1008 int bdrv_parse_aio(const char *mode, int *flags)
1009 {
1010 if (!strcmp(mode, "threads")) {
1011 /* do nothing, default */
1012 } else if (!strcmp(mode, "native")) {
1013 *flags |= BDRV_O_NATIVE_AIO;
1014 #ifdef CONFIG_LINUX_IO_URING
1015 } else if (!strcmp(mode, "io_uring")) {
1016 *flags |= BDRV_O_IO_URING;
1017 #endif
1018 } else {
1019 return -1;
1020 }
1021
1022 return 0;
1023 }
1024
1025 /**
1026 * Set open flags for a given discard mode
1027 *
1028 * Return 0 on success, -1 if the discard mode was invalid.
1029 */
1030 int bdrv_parse_discard_flags(const char *mode, int *flags)
1031 {
1032 *flags &= ~BDRV_O_UNMAP;
1033
1034 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1035 /* do nothing */
1036 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1037 *flags |= BDRV_O_UNMAP;
1038 } else {
1039 return -1;
1040 }
1041
1042 return 0;
1043 }
1044
1045 /**
1046 * Set open flags for a given cache mode
1047 *
1048 * Return 0 on success, -1 if the cache mode was invalid.
1049 */
1050 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1051 {
1052 *flags &= ~BDRV_O_CACHE_MASK;
1053
1054 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1055 *writethrough = false;
1056 *flags |= BDRV_O_NOCACHE;
1057 } else if (!strcmp(mode, "directsync")) {
1058 *writethrough = true;
1059 *flags |= BDRV_O_NOCACHE;
1060 } else if (!strcmp(mode, "writeback")) {
1061 *writethrough = false;
1062 } else if (!strcmp(mode, "unsafe")) {
1063 *writethrough = false;
1064 *flags |= BDRV_O_NO_FLUSH;
1065 } else if (!strcmp(mode, "writethrough")) {
1066 *writethrough = true;
1067 } else {
1068 return -1;
1069 }
1070
1071 return 0;
1072 }
1073
1074 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1075 {
1076 BlockDriverState *parent = c->opaque;
1077 return g_strdup(bdrv_get_device_or_node_name(parent));
1078 }
1079
1080 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1081 {
1082 BlockDriverState *bs = child->opaque;
1083 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1084 }
1085
1086 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1087 {
1088 BlockDriverState *bs = child->opaque;
1089 return bdrv_drain_poll(bs, false, NULL, false);
1090 }
1091
1092 static void bdrv_child_cb_drained_end(BdrvChild *child,
1093 int *drained_end_counter)
1094 {
1095 BlockDriverState *bs = child->opaque;
1096 bdrv_drained_end_no_poll(bs, drained_end_counter);
1097 }
1098
1099 static int bdrv_child_cb_inactivate(BdrvChild *child)
1100 {
1101 BlockDriverState *bs = child->opaque;
1102 assert(bs->open_flags & BDRV_O_INACTIVE);
1103 return 0;
1104 }
1105
1106 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1107 GSList **ignore, Error **errp)
1108 {
1109 BlockDriverState *bs = child->opaque;
1110 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1111 }
1112
1113 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1114 GSList **ignore)
1115 {
1116 BlockDriverState *bs = child->opaque;
1117 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1118 }
1119
1120 /*
1121 * Returns the options and flags that a temporary snapshot should get, based on
1122 * the originally requested flags (the originally requested image will have
1123 * flags like a backing file)
1124 */
1125 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1126 int parent_flags, QDict *parent_options)
1127 {
1128 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1129
1130 /* For temporary files, unconditional cache=unsafe is fine */
1131 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1132 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1133
1134 /* Copy the read-only and discard options from the parent */
1135 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1136 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1137
1138 /* aio=native doesn't work for cache.direct=off, so disable it for the
1139 * temporary snapshot */
1140 *child_flags &= ~BDRV_O_NATIVE_AIO;
1141 }
1142
1143 static void bdrv_backing_attach(BdrvChild *c)
1144 {
1145 BlockDriverState *parent = c->opaque;
1146 BlockDriverState *backing_hd = c->bs;
1147
1148 assert(!parent->backing_blocker);
1149 error_setg(&parent->backing_blocker,
1150 "node is used as backing hd of '%s'",
1151 bdrv_get_device_or_node_name(parent));
1152
1153 bdrv_refresh_filename(backing_hd);
1154
1155 parent->open_flags &= ~BDRV_O_NO_BACKING;
1156 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
1157 backing_hd->filename);
1158 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
1159 backing_hd->drv ? backing_hd->drv->format_name : "");
1160
1161 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1162 /* Otherwise we won't be able to commit or stream */
1163 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1164 parent->backing_blocker);
1165 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1166 parent->backing_blocker);
1167 /*
1168 * We do backup in 3 ways:
1169 * 1. drive backup
1170 * The target bs is new opened, and the source is top BDS
1171 * 2. blockdev backup
1172 * Both the source and the target are top BDSes.
1173 * 3. internal backup(used for block replication)
1174 * Both the source and the target are backing file
1175 *
1176 * In case 1 and 2, neither the source nor the target is the backing file.
1177 * In case 3, we will block the top BDS, so there is only one block job
1178 * for the top BDS and its backing chain.
1179 */
1180 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1181 parent->backing_blocker);
1182 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1183 parent->backing_blocker);
1184 }
1185
1186 static void bdrv_backing_detach(BdrvChild *c)
1187 {
1188 BlockDriverState *parent = c->opaque;
1189
1190 assert(parent->backing_blocker);
1191 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1192 error_free(parent->backing_blocker);
1193 parent->backing_blocker = NULL;
1194 }
1195
1196 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1197 const char *filename, Error **errp)
1198 {
1199 BlockDriverState *parent = c->opaque;
1200 bool read_only = bdrv_is_read_only(parent);
1201 int ret;
1202
1203 if (read_only) {
1204 ret = bdrv_reopen_set_read_only(parent, false, errp);
1205 if (ret < 0) {
1206 return ret;
1207 }
1208 }
1209
1210 ret = bdrv_change_backing_file(parent, filename,
1211 base->drv ? base->drv->format_name : "",
1212 false);
1213 if (ret < 0) {
1214 error_setg_errno(errp, -ret, "Could not update backing file link");
1215 }
1216
1217 if (read_only) {
1218 bdrv_reopen_set_read_only(parent, true, NULL);
1219 }
1220
1221 return ret;
1222 }
1223
1224 /*
1225 * Returns the options and flags that a generic child of a BDS should
1226 * get, based on the given options and flags for the parent BDS.
1227 */
1228 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1229 int *child_flags, QDict *child_options,
1230 int parent_flags, QDict *parent_options)
1231 {
1232 int flags = parent_flags;
1233
1234 /*
1235 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1236 * Generally, the question to answer is: Should this child be
1237 * format-probed by default?
1238 */
1239
1240 /*
1241 * Pure and non-filtered data children of non-format nodes should
1242 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1243 * set). This only affects a very limited set of drivers (namely
1244 * quorum and blkverify when this comment was written).
1245 * Force-clear BDRV_O_PROTOCOL then.
1246 */
1247 if (!parent_is_format &&
1248 (role & BDRV_CHILD_DATA) &&
1249 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1250 {
1251 flags &= ~BDRV_O_PROTOCOL;
1252 }
1253
1254 /*
1255 * All children of format nodes (except for COW children) and all
1256 * metadata children in general should never be format-probed.
1257 * Force-set BDRV_O_PROTOCOL then.
1258 */
1259 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1260 (role & BDRV_CHILD_METADATA))
1261 {
1262 flags |= BDRV_O_PROTOCOL;
1263 }
1264
1265 /*
1266 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1267 * the parent.
1268 */
1269 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1270 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1271 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1272
1273 if (role & BDRV_CHILD_COW) {
1274 /* backing files are opened read-only by default */
1275 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1276 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1277 } else {
1278 /* Inherit the read-only option from the parent if it's not set */
1279 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1280 qdict_copy_default(child_options, parent_options,
1281 BDRV_OPT_AUTO_READ_ONLY);
1282 }
1283
1284 /*
1285 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1286 * can default to enable it on lower layers regardless of the
1287 * parent option.
1288 */
1289 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1290
1291 /* Clear flags that only apply to the top layer */
1292 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1293
1294 if (role & BDRV_CHILD_METADATA) {
1295 flags &= ~BDRV_O_NO_IO;
1296 }
1297 if (role & BDRV_CHILD_COW) {
1298 flags &= ~BDRV_O_TEMPORARY;
1299 }
1300
1301 *child_flags = flags;
1302 }
1303
1304 static void bdrv_child_cb_attach(BdrvChild *child)
1305 {
1306 BlockDriverState *bs = child->opaque;
1307
1308 if (child->role & BDRV_CHILD_COW) {
1309 bdrv_backing_attach(child);
1310 }
1311
1312 bdrv_apply_subtree_drain(child, bs);
1313 }
1314
1315 static void bdrv_child_cb_detach(BdrvChild *child)
1316 {
1317 BlockDriverState *bs = child->opaque;
1318
1319 if (child->role & BDRV_CHILD_COW) {
1320 bdrv_backing_detach(child);
1321 }
1322
1323 bdrv_unapply_subtree_drain(child, bs);
1324 }
1325
1326 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1327 const char *filename, Error **errp)
1328 {
1329 if (c->role & BDRV_CHILD_COW) {
1330 return bdrv_backing_update_filename(c, base, filename, errp);
1331 }
1332 return 0;
1333 }
1334
1335 const BdrvChildClass child_of_bds = {
1336 .parent_is_bds = true,
1337 .get_parent_desc = bdrv_child_get_parent_desc,
1338 .inherit_options = bdrv_inherited_options,
1339 .drained_begin = bdrv_child_cb_drained_begin,
1340 .drained_poll = bdrv_child_cb_drained_poll,
1341 .drained_end = bdrv_child_cb_drained_end,
1342 .attach = bdrv_child_cb_attach,
1343 .detach = bdrv_child_cb_detach,
1344 .inactivate = bdrv_child_cb_inactivate,
1345 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1346 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1347 .update_filename = bdrv_child_cb_update_filename,
1348 };
1349
1350 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1351 {
1352 int open_flags = flags;
1353
1354 /*
1355 * Clear flags that are internal to the block layer before opening the
1356 * image.
1357 */
1358 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1359
1360 return open_flags;
1361 }
1362
1363 static void update_flags_from_options(int *flags, QemuOpts *opts)
1364 {
1365 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1366
1367 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1368 *flags |= BDRV_O_NO_FLUSH;
1369 }
1370
1371 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1372 *flags |= BDRV_O_NOCACHE;
1373 }
1374
1375 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1376 *flags |= BDRV_O_RDWR;
1377 }
1378
1379 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1380 *flags |= BDRV_O_AUTO_RDONLY;
1381 }
1382 }
1383
1384 static void update_options_from_flags(QDict *options, int flags)
1385 {
1386 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1387 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1388 }
1389 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1390 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1391 flags & BDRV_O_NO_FLUSH);
1392 }
1393 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1394 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1395 }
1396 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1397 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1398 flags & BDRV_O_AUTO_RDONLY);
1399 }
1400 }
1401
1402 static void bdrv_assign_node_name(BlockDriverState *bs,
1403 const char *node_name,
1404 Error **errp)
1405 {
1406 char *gen_node_name = NULL;
1407
1408 if (!node_name) {
1409 node_name = gen_node_name = id_generate(ID_BLOCK);
1410 } else if (!id_wellformed(node_name)) {
1411 /*
1412 * Check for empty string or invalid characters, but not if it is
1413 * generated (generated names use characters not available to the user)
1414 */
1415 error_setg(errp, "Invalid node name");
1416 return;
1417 }
1418
1419 /* takes care of avoiding namespaces collisions */
1420 if (blk_by_name(node_name)) {
1421 error_setg(errp, "node-name=%s is conflicting with a device id",
1422 node_name);
1423 goto out;
1424 }
1425
1426 /* takes care of avoiding duplicates node names */
1427 if (bdrv_find_node(node_name)) {
1428 error_setg(errp, "Duplicate node name");
1429 goto out;
1430 }
1431
1432 /* Make sure that the node name isn't truncated */
1433 if (strlen(node_name) >= sizeof(bs->node_name)) {
1434 error_setg(errp, "Node name too long");
1435 goto out;
1436 }
1437
1438 /* copy node name into the bs and insert it into the graph list */
1439 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1440 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1441 out:
1442 g_free(gen_node_name);
1443 }
1444
1445 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1446 const char *node_name, QDict *options,
1447 int open_flags, Error **errp)
1448 {
1449 Error *local_err = NULL;
1450 int i, ret;
1451
1452 bdrv_assign_node_name(bs, node_name, &local_err);
1453 if (local_err) {
1454 error_propagate(errp, local_err);
1455 return -EINVAL;
1456 }
1457
1458 bs->drv = drv;
1459 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1460 bs->opaque = g_malloc0(drv->instance_size);
1461
1462 if (drv->bdrv_file_open) {
1463 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1464 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1465 } else if (drv->bdrv_open) {
1466 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1467 } else {
1468 ret = 0;
1469 }
1470
1471 if (ret < 0) {
1472 if (local_err) {
1473 error_propagate(errp, local_err);
1474 } else if (bs->filename[0]) {
1475 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1476 } else {
1477 error_setg_errno(errp, -ret, "Could not open image");
1478 }
1479 goto open_failed;
1480 }
1481
1482 ret = refresh_total_sectors(bs, bs->total_sectors);
1483 if (ret < 0) {
1484 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1485 return ret;
1486 }
1487
1488 bdrv_refresh_limits(bs, &local_err);
1489 if (local_err) {
1490 error_propagate(errp, local_err);
1491 return -EINVAL;
1492 }
1493
1494 assert(bdrv_opt_mem_align(bs) != 0);
1495 assert(bdrv_min_mem_align(bs) != 0);
1496 assert(is_power_of_2(bs->bl.request_alignment));
1497
1498 for (i = 0; i < bs->quiesce_counter; i++) {
1499 if (drv->bdrv_co_drain_begin) {
1500 drv->bdrv_co_drain_begin(bs);
1501 }
1502 }
1503
1504 return 0;
1505 open_failed:
1506 bs->drv = NULL;
1507 if (bs->file != NULL) {
1508 bdrv_unref_child(bs, bs->file);
1509 bs->file = NULL;
1510 }
1511 g_free(bs->opaque);
1512 bs->opaque = NULL;
1513 return ret;
1514 }
1515
1516 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1517 int flags, Error **errp)
1518 {
1519 BlockDriverState *bs;
1520 int ret;
1521
1522 bs = bdrv_new();
1523 bs->open_flags = flags;
1524 bs->explicit_options = qdict_new();
1525 bs->options = qdict_new();
1526 bs->opaque = NULL;
1527
1528 update_options_from_flags(bs->options, flags);
1529
1530 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1531 if (ret < 0) {
1532 qobject_unref(bs->explicit_options);
1533 bs->explicit_options = NULL;
1534 qobject_unref(bs->options);
1535 bs->options = NULL;
1536 bdrv_unref(bs);
1537 return NULL;
1538 }
1539
1540 return bs;
1541 }
1542
1543 QemuOptsList bdrv_runtime_opts = {
1544 .name = "bdrv_common",
1545 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1546 .desc = {
1547 {
1548 .name = "node-name",
1549 .type = QEMU_OPT_STRING,
1550 .help = "Node name of the block device node",
1551 },
1552 {
1553 .name = "driver",
1554 .type = QEMU_OPT_STRING,
1555 .help = "Block driver to use for the node",
1556 },
1557 {
1558 .name = BDRV_OPT_CACHE_DIRECT,
1559 .type = QEMU_OPT_BOOL,
1560 .help = "Bypass software writeback cache on the host",
1561 },
1562 {
1563 .name = BDRV_OPT_CACHE_NO_FLUSH,
1564 .type = QEMU_OPT_BOOL,
1565 .help = "Ignore flush requests",
1566 },
1567 {
1568 .name = BDRV_OPT_READ_ONLY,
1569 .type = QEMU_OPT_BOOL,
1570 .help = "Node is opened in read-only mode",
1571 },
1572 {
1573 .name = BDRV_OPT_AUTO_READ_ONLY,
1574 .type = QEMU_OPT_BOOL,
1575 .help = "Node can become read-only if opening read-write fails",
1576 },
1577 {
1578 .name = "detect-zeroes",
1579 .type = QEMU_OPT_STRING,
1580 .help = "try to optimize zero writes (off, on, unmap)",
1581 },
1582 {
1583 .name = BDRV_OPT_DISCARD,
1584 .type = QEMU_OPT_STRING,
1585 .help = "discard operation (ignore/off, unmap/on)",
1586 },
1587 {
1588 .name = BDRV_OPT_FORCE_SHARE,
1589 .type = QEMU_OPT_BOOL,
1590 .help = "always accept other writers (default: off)",
1591 },
1592 { /* end of list */ }
1593 },
1594 };
1595
1596 QemuOptsList bdrv_create_opts_simple = {
1597 .name = "simple-create-opts",
1598 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1599 .desc = {
1600 {
1601 .name = BLOCK_OPT_SIZE,
1602 .type = QEMU_OPT_SIZE,
1603 .help = "Virtual disk size"
1604 },
1605 {
1606 .name = BLOCK_OPT_PREALLOC,
1607 .type = QEMU_OPT_STRING,
1608 .help = "Preallocation mode (allowed values: off)"
1609 },
1610 { /* end of list */ }
1611 }
1612 };
1613
1614 /*
1615 * Common part for opening disk images and files
1616 *
1617 * Removes all processed options from *options.
1618 */
1619 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1620 QDict *options, Error **errp)
1621 {
1622 int ret, open_flags;
1623 const char *filename;
1624 const char *driver_name = NULL;
1625 const char *node_name = NULL;
1626 const char *discard;
1627 QemuOpts *opts;
1628 BlockDriver *drv;
1629 Error *local_err = NULL;
1630
1631 assert(bs->file == NULL);
1632 assert(options != NULL && bs->options != options);
1633
1634 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1635 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1636 ret = -EINVAL;
1637 goto fail_opts;
1638 }
1639
1640 update_flags_from_options(&bs->open_flags, opts);
1641
1642 driver_name = qemu_opt_get(opts, "driver");
1643 drv = bdrv_find_format(driver_name);
1644 assert(drv != NULL);
1645
1646 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1647
1648 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1649 error_setg(errp,
1650 BDRV_OPT_FORCE_SHARE
1651 "=on can only be used with read-only images");
1652 ret = -EINVAL;
1653 goto fail_opts;
1654 }
1655
1656 if (file != NULL) {
1657 bdrv_refresh_filename(blk_bs(file));
1658 filename = blk_bs(file)->filename;
1659 } else {
1660 /*
1661 * Caution: while qdict_get_try_str() is fine, getting
1662 * non-string types would require more care. When @options
1663 * come from -blockdev or blockdev_add, its members are typed
1664 * according to the QAPI schema, but when they come from
1665 * -drive, they're all QString.
1666 */
1667 filename = qdict_get_try_str(options, "filename");
1668 }
1669
1670 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1671 error_setg(errp, "The '%s' block driver requires a file name",
1672 drv->format_name);
1673 ret = -EINVAL;
1674 goto fail_opts;
1675 }
1676
1677 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1678 drv->format_name);
1679
1680 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1681
1682 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1683 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1684 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1685 } else {
1686 ret = -ENOTSUP;
1687 }
1688 if (ret < 0) {
1689 error_setg(errp,
1690 !bs->read_only && bdrv_is_whitelisted(drv, true)
1691 ? "Driver '%s' can only be used for read-only devices"
1692 : "Driver '%s' is not whitelisted",
1693 drv->format_name);
1694 goto fail_opts;
1695 }
1696 }
1697
1698 /* bdrv_new() and bdrv_close() make it so */
1699 assert(atomic_read(&bs->copy_on_read) == 0);
1700
1701 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1702 if (!bs->read_only) {
1703 bdrv_enable_copy_on_read(bs);
1704 } else {
1705 error_setg(errp, "Can't use copy-on-read on read-only device");
1706 ret = -EINVAL;
1707 goto fail_opts;
1708 }
1709 }
1710
1711 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1712 if (discard != NULL) {
1713 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1714 error_setg(errp, "Invalid discard option");
1715 ret = -EINVAL;
1716 goto fail_opts;
1717 }
1718 }
1719
1720 bs->detect_zeroes =
1721 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1722 if (local_err) {
1723 error_propagate(errp, local_err);
1724 ret = -EINVAL;
1725 goto fail_opts;
1726 }
1727
1728 if (filename != NULL) {
1729 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1730 } else {
1731 bs->filename[0] = '\0';
1732 }
1733 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1734
1735 /* Open the image, either directly or using a protocol */
1736 open_flags = bdrv_open_flags(bs, bs->open_flags);
1737 node_name = qemu_opt_get(opts, "node-name");
1738
1739 assert(!drv->bdrv_file_open || file == NULL);
1740 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1741 if (ret < 0) {
1742 goto fail_opts;
1743 }
1744
1745 qemu_opts_del(opts);
1746 return 0;
1747
1748 fail_opts:
1749 qemu_opts_del(opts);
1750 return ret;
1751 }
1752
1753 static QDict *parse_json_filename(const char *filename, Error **errp)
1754 {
1755 QObject *options_obj;
1756 QDict *options;
1757 int ret;
1758
1759 ret = strstart(filename, "json:", &filename);
1760 assert(ret);
1761
1762 options_obj = qobject_from_json(filename, errp);
1763 if (!options_obj) {
1764 error_prepend(errp, "Could not parse the JSON options: ");
1765 return NULL;
1766 }
1767
1768 options = qobject_to(QDict, options_obj);
1769 if (!options) {
1770 qobject_unref(options_obj);
1771 error_setg(errp, "Invalid JSON object given");
1772 return NULL;
1773 }
1774
1775 qdict_flatten(options);
1776
1777 return options;
1778 }
1779
1780 static void parse_json_protocol(QDict *options, const char **pfilename,
1781 Error **errp)
1782 {
1783 QDict *json_options;
1784 Error *local_err = NULL;
1785
1786 /* Parse json: pseudo-protocol */
1787 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1788 return;
1789 }
1790
1791 json_options = parse_json_filename(*pfilename, &local_err);
1792 if (local_err) {
1793 error_propagate(errp, local_err);
1794 return;
1795 }
1796
1797 /* Options given in the filename have lower priority than options
1798 * specified directly */
1799 qdict_join(options, json_options, false);
1800 qobject_unref(json_options);
1801 *pfilename = NULL;
1802 }
1803
1804 /*
1805 * Fills in default options for opening images and converts the legacy
1806 * filename/flags pair to option QDict entries.
1807 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1808 * block driver has been specified explicitly.
1809 */
1810 static int bdrv_fill_options(QDict **options, const char *filename,
1811 int *flags, Error **errp)
1812 {
1813 const char *drvname;
1814 bool protocol = *flags & BDRV_O_PROTOCOL;
1815 bool parse_filename = false;
1816 BlockDriver *drv = NULL;
1817 Error *local_err = NULL;
1818
1819 /*
1820 * Caution: while qdict_get_try_str() is fine, getting non-string
1821 * types would require more care. When @options come from
1822 * -blockdev or blockdev_add, its members are typed according to
1823 * the QAPI schema, but when they come from -drive, they're all
1824 * QString.
1825 */
1826 drvname = qdict_get_try_str(*options, "driver");
1827 if (drvname) {
1828 drv = bdrv_find_format(drvname);
1829 if (!drv) {
1830 error_setg(errp, "Unknown driver '%s'", drvname);
1831 return -ENOENT;
1832 }
1833 /* If the user has explicitly specified the driver, this choice should
1834 * override the BDRV_O_PROTOCOL flag */
1835 protocol = drv->bdrv_file_open;
1836 }
1837
1838 if (protocol) {
1839 *flags |= BDRV_O_PROTOCOL;
1840 } else {
1841 *flags &= ~BDRV_O_PROTOCOL;
1842 }
1843
1844 /* Translate cache options from flags into options */
1845 update_options_from_flags(*options, *flags);
1846
1847 /* Fetch the file name from the options QDict if necessary */
1848 if (protocol && filename) {
1849 if (!qdict_haskey(*options, "filename")) {
1850 qdict_put_str(*options, "filename", filename);
1851 parse_filename = true;
1852 } else {
1853 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1854 "the same time");
1855 return -EINVAL;
1856 }
1857 }
1858
1859 /* Find the right block driver */
1860 /* See cautionary note on accessing @options above */
1861 filename = qdict_get_try_str(*options, "filename");
1862
1863 if (!drvname && protocol) {
1864 if (filename) {
1865 drv = bdrv_find_protocol(filename, parse_filename, errp);
1866 if (!drv) {
1867 return -EINVAL;
1868 }
1869
1870 drvname = drv->format_name;
1871 qdict_put_str(*options, "driver", drvname);
1872 } else {
1873 error_setg(errp, "Must specify either driver or file");
1874 return -EINVAL;
1875 }
1876 }
1877
1878 assert(drv || !protocol);
1879
1880 /* Driver-specific filename parsing */
1881 if (drv && drv->bdrv_parse_filename && parse_filename) {
1882 drv->bdrv_parse_filename(filename, *options, &local_err);
1883 if (local_err) {
1884 error_propagate(errp, local_err);
1885 return -EINVAL;
1886 }
1887
1888 if (!drv->bdrv_needs_filename) {
1889 qdict_del(*options, "filename");
1890 }
1891 }
1892
1893 return 0;
1894 }
1895
1896 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1897 uint64_t perm, uint64_t shared,
1898 GSList *ignore_children,
1899 bool *tighten_restrictions, Error **errp);
1900 static void bdrv_child_abort_perm_update(BdrvChild *c);
1901 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1902
1903 typedef struct BlockReopenQueueEntry {
1904 bool prepared;
1905 bool perms_checked;
1906 BDRVReopenState state;
1907 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
1908 } BlockReopenQueueEntry;
1909
1910 /*
1911 * Return the flags that @bs will have after the reopens in @q have
1912 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1913 * return the current flags.
1914 */
1915 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1916 {
1917 BlockReopenQueueEntry *entry;
1918
1919 if (q != NULL) {
1920 QTAILQ_FOREACH(entry, q, entry) {
1921 if (entry->state.bs == bs) {
1922 return entry->state.flags;
1923 }
1924 }
1925 }
1926
1927 return bs->open_flags;
1928 }
1929
1930 /* Returns whether the image file can be written to after the reopen queue @q
1931 * has been successfully applied, or right now if @q is NULL. */
1932 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1933 BlockReopenQueue *q)
1934 {
1935 int flags = bdrv_reopen_get_flags(q, bs);
1936
1937 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1938 }
1939
1940 /*
1941 * Return whether the BDS can be written to. This is not necessarily
1942 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1943 * be written to but do not count as read-only images.
1944 */
1945 bool bdrv_is_writable(BlockDriverState *bs)
1946 {
1947 return bdrv_is_writable_after_reopen(bs, NULL);
1948 }
1949
1950 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1951 BdrvChild *c, BdrvChildRole role,
1952 BlockReopenQueue *reopen_queue,
1953 uint64_t parent_perm, uint64_t parent_shared,
1954 uint64_t *nperm, uint64_t *nshared)
1955 {
1956 assert(bs->drv && bs->drv->bdrv_child_perm);
1957 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1958 parent_perm, parent_shared,
1959 nperm, nshared);
1960 /* TODO Take force_share from reopen_queue */
1961 if (child_bs && child_bs->force_share) {
1962 *nshared = BLK_PERM_ALL;
1963 }
1964 }
1965
1966 /*
1967 * Check whether permissions on this node can be changed in a way that
1968 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1969 * permissions of all its parents. This involves checking whether all necessary
1970 * permission changes to child nodes can be performed.
1971 *
1972 * Will set *tighten_restrictions to true if and only if new permissions have to
1973 * be taken or currently shared permissions are to be unshared. Otherwise,
1974 * errors are not fatal as long as the caller accepts that the restrictions
1975 * remain tighter than they need to be. The caller still has to abort the
1976 * transaction.
1977 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1978 * encounter fatal errors even though no restrictions are to be tightened. For
1979 * example, changing a node from RW to RO will fail if the WRITE permission is
1980 * to be kept.
1981 *
1982 * A call to this function must always be followed by a call to bdrv_set_perm()
1983 * or bdrv_abort_perm_update().
1984 */
1985 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1986 uint64_t cumulative_perms,
1987 uint64_t cumulative_shared_perms,
1988 GSList *ignore_children,
1989 bool *tighten_restrictions, Error **errp)
1990 {
1991 BlockDriver *drv = bs->drv;
1992 BdrvChild *c;
1993 int ret;
1994
1995 assert(!q || !tighten_restrictions);
1996
1997 if (tighten_restrictions) {
1998 uint64_t current_perms, current_shared;
1999 uint64_t added_perms, removed_shared_perms;
2000
2001 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
2002
2003 added_perms = cumulative_perms & ~current_perms;
2004 removed_shared_perms = current_shared & ~cumulative_shared_perms;
2005
2006 *tighten_restrictions = added_perms || removed_shared_perms;
2007 }
2008
2009 /* Write permissions never work with read-only images */
2010 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2011 !bdrv_is_writable_after_reopen(bs, q))
2012 {
2013 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
2014 error_setg(errp, "Block node is read-only");
2015 } else {
2016 uint64_t current_perms, current_shared;
2017 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
2018 if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
2019 error_setg(errp, "Cannot make block node read-only, there is "
2020 "a writer on it");
2021 } else {
2022 error_setg(errp, "Cannot make block node read-only and create "
2023 "a writer on it");
2024 }
2025 }
2026
2027 return -EPERM;
2028 }
2029
2030 /*
2031 * Unaligned requests will automatically be aligned to bl.request_alignment
2032 * and without RESIZE we can't extend requests to write to space beyond the
2033 * end of the image, so it's required that the image size is aligned.
2034 */
2035 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2036 !(cumulative_perms & BLK_PERM_RESIZE))
2037 {
2038 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2039 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2040 "Image size is not a multiple of request "
2041 "alignment");
2042 return -EPERM;
2043 }
2044 }
2045
2046 /* Check this node */
2047 if (!drv) {
2048 return 0;
2049 }
2050
2051 if (drv->bdrv_check_perm) {
2052 return drv->bdrv_check_perm(bs, cumulative_perms,
2053 cumulative_shared_perms, errp);
2054 }
2055
2056 /* Drivers that never have children can omit .bdrv_child_perm() */
2057 if (!drv->bdrv_child_perm) {
2058 assert(QLIST_EMPTY(&bs->children));
2059 return 0;
2060 }
2061
2062 /* Check all children */
2063 QLIST_FOREACH(c, &bs->children, next) {
2064 uint64_t cur_perm, cur_shared;
2065 bool child_tighten_restr;
2066
2067 bdrv_child_perm(bs, c->bs, c, c->role, q,
2068 cumulative_perms, cumulative_shared_perms,
2069 &cur_perm, &cur_shared);
2070 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children,
2071 tighten_restrictions ? &child_tighten_restr
2072 : NULL,
2073 errp);
2074 if (tighten_restrictions) {
2075 *tighten_restrictions |= child_tighten_restr;
2076 }
2077 if (ret < 0) {
2078 return ret;
2079 }
2080 }
2081
2082 return 0;
2083 }
2084
2085 /*
2086 * Notifies drivers that after a previous bdrv_check_perm() call, the
2087 * permission update is not performed and any preparations made for it (e.g.
2088 * taken file locks) need to be undone.
2089 *
2090 * This function recursively notifies all child nodes.
2091 */
2092 static void bdrv_abort_perm_update(BlockDriverState *bs)
2093 {
2094 BlockDriver *drv = bs->drv;
2095 BdrvChild *c;
2096
2097 if (!drv) {
2098 return;
2099 }
2100
2101 if (drv->bdrv_abort_perm_update) {
2102 drv->bdrv_abort_perm_update(bs);
2103 }
2104
2105 QLIST_FOREACH(c, &bs->children, next) {
2106 bdrv_child_abort_perm_update(c);
2107 }
2108 }
2109
2110 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
2111 uint64_t cumulative_shared_perms)
2112 {
2113 BlockDriver *drv = bs->drv;
2114 BdrvChild *c;
2115
2116 if (!drv) {
2117 return;
2118 }
2119
2120 /* Update this node */
2121 if (drv->bdrv_set_perm) {
2122 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2123 }
2124
2125 /* Drivers that never have children can omit .bdrv_child_perm() */
2126 if (!drv->bdrv_child_perm) {
2127 assert(QLIST_EMPTY(&bs->children));
2128 return;
2129 }
2130
2131 /* Update all children */
2132 QLIST_FOREACH(c, &bs->children, next) {
2133 uint64_t cur_perm, cur_shared;
2134 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2135 cumulative_perms, cumulative_shared_perms,
2136 &cur_perm, &cur_shared);
2137 bdrv_child_set_perm(c, cur_perm, cur_shared);
2138 }
2139 }
2140
2141 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2142 uint64_t *shared_perm)
2143 {
2144 BdrvChild *c;
2145 uint64_t cumulative_perms = 0;
2146 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2147
2148 QLIST_FOREACH(c, &bs->parents, next_parent) {
2149 cumulative_perms |= c->perm;
2150 cumulative_shared_perms &= c->shared_perm;
2151 }
2152
2153 *perm = cumulative_perms;
2154 *shared_perm = cumulative_shared_perms;
2155 }
2156
2157 static char *bdrv_child_user_desc(BdrvChild *c)
2158 {
2159 if (c->klass->get_parent_desc) {
2160 return c->klass->get_parent_desc(c);
2161 }
2162
2163 return g_strdup("another user");
2164 }
2165
2166 char *bdrv_perm_names(uint64_t perm)
2167 {
2168 struct perm_name {
2169 uint64_t perm;
2170 const char *name;
2171 } permissions[] = {
2172 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2173 { BLK_PERM_WRITE, "write" },
2174 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2175 { BLK_PERM_RESIZE, "resize" },
2176 { BLK_PERM_GRAPH_MOD, "change children" },
2177 { 0, NULL }
2178 };
2179
2180 GString *result = g_string_sized_new(30);
2181 struct perm_name *p;
2182
2183 for (p = permissions; p->name; p++) {
2184 if (perm & p->perm) {
2185 if (result->len > 0) {
2186 g_string_append(result, ", ");
2187 }
2188 g_string_append(result, p->name);
2189 }
2190 }
2191
2192 return g_string_free(result, FALSE);
2193 }
2194
2195 /*
2196 * Checks whether a new reference to @bs can be added if the new user requires
2197 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2198 * set, the BdrvChild objects in this list are ignored in the calculations;
2199 * this allows checking permission updates for an existing reference.
2200 *
2201 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2202 *
2203 * Needs to be followed by a call to either bdrv_set_perm() or
2204 * bdrv_abort_perm_update(). */
2205 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
2206 uint64_t new_used_perm,
2207 uint64_t new_shared_perm,
2208 GSList *ignore_children,
2209 bool *tighten_restrictions,
2210 Error **errp)
2211 {
2212 BdrvChild *c;
2213 uint64_t cumulative_perms = new_used_perm;
2214 uint64_t cumulative_shared_perms = new_shared_perm;
2215
2216 assert(!q || !tighten_restrictions);
2217
2218 /* There is no reason why anyone couldn't tolerate write_unchanged */
2219 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
2220
2221 QLIST_FOREACH(c, &bs->parents, next_parent) {
2222 if (g_slist_find(ignore_children, c)) {
2223 continue;
2224 }
2225
2226 if ((new_used_perm & c->shared_perm) != new_used_perm) {
2227 char *user = bdrv_child_user_desc(c);
2228 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
2229
2230 if (tighten_restrictions) {
2231 *tighten_restrictions = true;
2232 }
2233
2234 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
2235 "allow '%s' on %s",
2236 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2237 g_free(user);
2238 g_free(perm_names);
2239 return -EPERM;
2240 }
2241
2242 if ((c->perm & new_shared_perm) != c->perm) {
2243 char *user = bdrv_child_user_desc(c);
2244 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
2245
2246 if (tighten_restrictions) {
2247 *tighten_restrictions = true;
2248 }
2249
2250 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
2251 "'%s' on %s",
2252 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2253 g_free(user);
2254 g_free(perm_names);
2255 return -EPERM;
2256 }
2257
2258 cumulative_perms |= c->perm;
2259 cumulative_shared_perms &= c->shared_perm;
2260 }
2261
2262 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
2263 ignore_children, tighten_restrictions, errp);
2264 }
2265
2266 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2267 * bdrv_child_abort_perm_update(). */
2268 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
2269 uint64_t perm, uint64_t shared,
2270 GSList *ignore_children,
2271 bool *tighten_restrictions, Error **errp)
2272 {
2273 int ret;
2274
2275 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
2276 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children,
2277 tighten_restrictions, errp);
2278 g_slist_free(ignore_children);
2279
2280 if (ret < 0) {
2281 return ret;
2282 }
2283
2284 if (!c->has_backup_perm) {
2285 c->has_backup_perm = true;
2286 c->backup_perm = c->perm;
2287 c->backup_shared_perm = c->shared_perm;
2288 }
2289 /*
2290 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2291 * same child twice during check_perm procedure
2292 */
2293
2294 c->perm = perm;
2295 c->shared_perm = shared;
2296
2297 return 0;
2298 }
2299
2300 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
2301 {
2302 uint64_t cumulative_perms, cumulative_shared_perms;
2303
2304 c->has_backup_perm = false;
2305
2306 c->perm = perm;
2307 c->shared_perm = shared;
2308
2309 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
2310 &cumulative_shared_perms);
2311 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
2312 }
2313
2314 static void bdrv_child_abort_perm_update(BdrvChild *c)
2315 {
2316 if (c->has_backup_perm) {
2317 c->perm = c->backup_perm;
2318 c->shared_perm = c->backup_shared_perm;
2319 c->has_backup_perm = false;
2320 }
2321
2322 bdrv_abort_perm_update(c->bs);
2323 }
2324
2325 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2326 Error **errp)
2327 {
2328 Error *local_err = NULL;
2329 int ret;
2330 bool tighten_restrictions;
2331
2332 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL,
2333 &tighten_restrictions, &local_err);
2334 if (ret < 0) {
2335 bdrv_child_abort_perm_update(c);
2336 if (tighten_restrictions) {
2337 error_propagate(errp, local_err);
2338 } else {
2339 /*
2340 * Our caller may intend to only loosen restrictions and
2341 * does not expect this function to fail. Errors are not
2342 * fatal in such a case, so we can just hide them from our
2343 * caller.
2344 */
2345 error_free(local_err);
2346 ret = 0;
2347 }
2348 return ret;
2349 }
2350
2351 bdrv_child_set_perm(c, perm, shared);
2352
2353 return 0;
2354 }
2355
2356 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2357 {
2358 uint64_t parent_perms, parent_shared;
2359 uint64_t perms, shared;
2360
2361 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2362 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2363 parent_perms, parent_shared, &perms, &shared);
2364
2365 return bdrv_child_try_set_perm(c, perms, shared, errp);
2366 }
2367
2368 /*
2369 * Default implementation for .bdrv_child_perm() for block filters:
2370 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2371 * filtered child.
2372 */
2373 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2374 BdrvChildRole role,
2375 BlockReopenQueue *reopen_queue,
2376 uint64_t perm, uint64_t shared,
2377 uint64_t *nperm, uint64_t *nshared)
2378 {
2379 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2380 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2381 }
2382
2383 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2384 BdrvChildRole role,
2385 BlockReopenQueue *reopen_queue,
2386 uint64_t perm, uint64_t shared,
2387 uint64_t *nperm, uint64_t *nshared)
2388 {
2389 assert(role & BDRV_CHILD_COW);
2390
2391 /*
2392 * We want consistent read from backing files if the parent needs it.
2393 * No other operations are performed on backing files.
2394 */
2395 perm &= BLK_PERM_CONSISTENT_READ;
2396
2397 /*
2398 * If the parent can deal with changing data, we're okay with a
2399 * writable and resizable backing file.
2400 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2401 */
2402 if (shared & BLK_PERM_WRITE) {
2403 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2404 } else {
2405 shared = 0;
2406 }
2407
2408 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2409 BLK_PERM_WRITE_UNCHANGED;
2410
2411 if (bs->open_flags & BDRV_O_INACTIVE) {
2412 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2413 }
2414
2415 *nperm = perm;
2416 *nshared = shared;
2417 }
2418
2419 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2420 BdrvChildRole role,
2421 BlockReopenQueue *reopen_queue,
2422 uint64_t perm, uint64_t shared,
2423 uint64_t *nperm, uint64_t *nshared)
2424 {
2425 int flags;
2426
2427 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2428
2429 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2430
2431 /*
2432 * Apart from the modifications below, the same permissions are
2433 * forwarded and left alone as for filters
2434 */
2435 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2436 perm, shared, &perm, &shared);
2437
2438 if (role & BDRV_CHILD_METADATA) {
2439 /* Format drivers may touch metadata even if the guest doesn't write */
2440 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2441 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2442 }
2443
2444 /*
2445 * bs->file always needs to be consistent because of the
2446 * metadata. We can never allow other users to resize or write
2447 * to it.
2448 */
2449 if (!(flags & BDRV_O_NO_IO)) {
2450 perm |= BLK_PERM_CONSISTENT_READ;
2451 }
2452 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2453 }
2454
2455 if (role & BDRV_CHILD_DATA) {
2456 /*
2457 * Technically, everything in this block is a subset of the
2458 * BDRV_CHILD_METADATA path taken above, and so this could
2459 * be an "else if" branch. However, that is not obvious, and
2460 * this function is not performance critical, therefore we let
2461 * this be an independent "if".
2462 */
2463
2464 /*
2465 * We cannot allow other users to resize the file because the
2466 * format driver might have some assumptions about the size
2467 * (e.g. because it is stored in metadata, or because the file
2468 * is split into fixed-size data files).
2469 */
2470 shared &= ~BLK_PERM_RESIZE;
2471
2472 /*
2473 * WRITE_UNCHANGED often cannot be performed as such on the
2474 * data file. For example, the qcow2 driver may still need to
2475 * write copied clusters on copy-on-read.
2476 */
2477 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2478 perm |= BLK_PERM_WRITE;
2479 }
2480
2481 /*
2482 * If the data file is written to, the format driver may
2483 * expect to be able to resize it by writing beyond the EOF.
2484 */
2485 if (perm & BLK_PERM_WRITE) {
2486 perm |= BLK_PERM_RESIZE;
2487 }
2488 }
2489
2490 if (bs->open_flags & BDRV_O_INACTIVE) {
2491 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2492 }
2493
2494 *nperm = perm;
2495 *nshared = shared;
2496 }
2497
2498 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2499 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2500 uint64_t perm, uint64_t shared,
2501 uint64_t *nperm, uint64_t *nshared)
2502 {
2503 if (role & BDRV_CHILD_FILTERED) {
2504 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2505 BDRV_CHILD_COW)));
2506 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2507 perm, shared, nperm, nshared);
2508 } else if (role & BDRV_CHILD_COW) {
2509 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2510 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2511 perm, shared, nperm, nshared);
2512 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2513 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2514 perm, shared, nperm, nshared);
2515 } else {
2516 g_assert_not_reached();
2517 }
2518 }
2519
2520 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2521 {
2522 static const uint64_t permissions[] = {
2523 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2524 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2525 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2526 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2527 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2528 };
2529
2530 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2531 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2532
2533 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2534
2535 return permissions[qapi_perm];
2536 }
2537
2538 static void bdrv_replace_child_noperm(BdrvChild *child,
2539 BlockDriverState *new_bs)
2540 {
2541 BlockDriverState *old_bs = child->bs;
2542 int new_bs_quiesce_counter;
2543 int drain_saldo;
2544
2545 assert(!child->frozen);
2546
2547 if (old_bs && new_bs) {
2548 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2549 }
2550
2551 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2552 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2553
2554 /*
2555 * If the new child node is drained but the old one was not, flush
2556 * all outstanding requests to the old child node.
2557 */
2558 while (drain_saldo > 0 && child->klass->drained_begin) {
2559 bdrv_parent_drained_begin_single(child, true);
2560 drain_saldo--;
2561 }
2562
2563 if (old_bs) {
2564 /* Detach first so that the recursive drain sections coming from @child
2565 * are already gone and we only end the drain sections that came from
2566 * elsewhere. */
2567 if (child->klass->detach) {
2568 child->klass->detach(child);
2569 }
2570 QLIST_REMOVE(child, next_parent);
2571 }
2572
2573 child->bs = new_bs;
2574
2575 if (new_bs) {
2576 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2577
2578 /*
2579 * Detaching the old node may have led to the new node's
2580 * quiesce_counter having been decreased. Not a problem, we
2581 * just need to recognize this here and then invoke
2582 * drained_end appropriately more often.
2583 */
2584 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2585 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2586
2587 /* Attach only after starting new drained sections, so that recursive
2588 * drain sections coming from @child don't get an extra .drained_begin
2589 * callback. */
2590 if (child->klass->attach) {
2591 child->klass->attach(child);
2592 }
2593 }
2594
2595 /*
2596 * If the old child node was drained but the new one is not, allow
2597 * requests to come in only after the new node has been attached.
2598 */
2599 while (drain_saldo < 0 && child->klass->drained_end) {
2600 bdrv_parent_drained_end_single(child);
2601 drain_saldo++;
2602 }
2603 }
2604
2605 /*
2606 * Updates @child to change its reference to point to @new_bs, including
2607 * checking and applying the necessary permisson updates both to the old node
2608 * and to @new_bs.
2609 *
2610 * NULL is passed as @new_bs for removing the reference before freeing @child.
2611 *
2612 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2613 * function uses bdrv_set_perm() to update the permissions according to the new
2614 * reference that @new_bs gets.
2615 */
2616 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2617 {
2618 BlockDriverState *old_bs = child->bs;
2619 uint64_t perm, shared_perm;
2620
2621 bdrv_replace_child_noperm(child, new_bs);
2622
2623 /*
2624 * Start with the new node's permissions. If @new_bs is a (direct
2625 * or indirect) child of @old_bs, we must complete the permission
2626 * update on @new_bs before we loosen the restrictions on @old_bs.
2627 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2628 * updating the permissions of @new_bs, and thus not purely loosen
2629 * restrictions.
2630 */
2631 if (new_bs) {
2632 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2633 bdrv_set_perm(new_bs, perm, shared_perm);
2634 }
2635
2636 if (old_bs) {
2637 /* Update permissions for old node. This is guaranteed to succeed
2638 * because we're just taking a parent away, so we're loosening
2639 * restrictions. */
2640 bool tighten_restrictions;
2641 int ret;
2642
2643 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2644 ret = bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL,
2645 &tighten_restrictions, NULL);
2646 assert(tighten_restrictions == false);
2647 if (ret < 0) {
2648 /* We only tried to loosen restrictions, so errors are not fatal */
2649 bdrv_abort_perm_update(old_bs);
2650 } else {
2651 bdrv_set_perm(old_bs, perm, shared_perm);
2652 }
2653
2654 /* When the parent requiring a non-default AioContext is removed, the
2655 * node moves back to the main AioContext */
2656 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2657 }
2658 }
2659
2660 /*
2661 * This function steals the reference to child_bs from the caller.
2662 * That reference is later dropped by bdrv_root_unref_child().
2663 *
2664 * On failure NULL is returned, errp is set and the reference to
2665 * child_bs is also dropped.
2666 *
2667 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2668 * (unless @child_bs is already in @ctx).
2669 */
2670 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2671 const char *child_name,
2672 const BdrvChildClass *child_class,
2673 BdrvChildRole child_role,
2674 AioContext *ctx,
2675 uint64_t perm, uint64_t shared_perm,
2676 void *opaque, Error **errp)
2677 {
2678 BdrvChild *child;
2679 Error *local_err = NULL;
2680 int ret;
2681
2682 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, NULL,
2683 errp);
2684 if (ret < 0) {
2685 bdrv_abort_perm_update(child_bs);
2686 bdrv_unref(child_bs);
2687 return NULL;
2688 }
2689
2690 child = g_new(BdrvChild, 1);
2691 *child = (BdrvChild) {
2692 .bs = NULL,
2693 .name = g_strdup(child_name),
2694 .klass = child_class,
2695 .role = child_role,
2696 .perm = perm,
2697 .shared_perm = shared_perm,
2698 .opaque = opaque,
2699 };
2700
2701 /* If the AioContexts don't match, first try to move the subtree of
2702 * child_bs into the AioContext of the new parent. If this doesn't work,
2703 * try moving the parent into the AioContext of child_bs instead. */
2704 if (bdrv_get_aio_context(child_bs) != ctx) {
2705 ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err);
2706 if (ret < 0 && child_class->can_set_aio_ctx) {
2707 GSList *ignore = g_slist_prepend(NULL, child);
2708 ctx = bdrv_get_aio_context(child_bs);
2709 if (child_class->can_set_aio_ctx(child, ctx, &ignore, NULL)) {
2710 error_free(local_err);
2711 ret = 0;
2712 g_slist_free(ignore);
2713 ignore = g_slist_prepend(NULL, child);
2714 child_class->set_aio_ctx(child, ctx, &ignore);
2715 }
2716 g_slist_free(ignore);
2717 }
2718 if (ret < 0) {
2719 error_propagate(errp, local_err);
2720 g_free(child);
2721 bdrv_abort_perm_update(child_bs);
2722 bdrv_unref(child_bs);
2723 return NULL;
2724 }
2725 }
2726
2727 /* This performs the matching bdrv_set_perm() for the above check. */
2728 bdrv_replace_child(child, child_bs);
2729
2730 return child;
2731 }
2732
2733 /*
2734 * This function transfers the reference to child_bs from the caller
2735 * to parent_bs. That reference is later dropped by parent_bs on
2736 * bdrv_close() or if someone calls bdrv_unref_child().
2737 *
2738 * On failure NULL is returned, errp is set and the reference to
2739 * child_bs is also dropped.
2740 *
2741 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2742 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2743 */
2744 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2745 BlockDriverState *child_bs,
2746 const char *child_name,
2747 const BdrvChildClass *child_class,
2748 BdrvChildRole child_role,
2749 Error **errp)
2750 {
2751 BdrvChild *child;
2752 uint64_t perm, shared_perm;
2753
2754 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2755
2756 assert(parent_bs->drv);
2757 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2758 perm, shared_perm, &perm, &shared_perm);
2759
2760 child = bdrv_root_attach_child(child_bs, child_name, child_class,
2761 child_role, bdrv_get_aio_context(parent_bs),
2762 perm, shared_perm, parent_bs, errp);
2763 if (child == NULL) {
2764 return NULL;
2765 }
2766
2767 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2768 return child;
2769 }
2770
2771 static void bdrv_detach_child(BdrvChild *child)
2772 {
2773 QLIST_SAFE_REMOVE(child, next);
2774
2775 bdrv_replace_child(child, NULL);
2776
2777 g_free(child->name);
2778 g_free(child);
2779 }
2780
2781 void bdrv_root_unref_child(BdrvChild *child)
2782 {
2783 BlockDriverState *child_bs;
2784
2785 child_bs = child->bs;
2786 bdrv_detach_child(child);
2787 bdrv_unref(child_bs);
2788 }
2789
2790 /**
2791 * Clear all inherits_from pointers from children and grandchildren of
2792 * @root that point to @root, where necessary.
2793 */
2794 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child)
2795 {
2796 BdrvChild *c;
2797
2798 if (child->bs->inherits_from == root) {
2799 /*
2800 * Remove inherits_from only when the last reference between root and
2801 * child->bs goes away.
2802 */
2803 QLIST_FOREACH(c, &root->children, next) {
2804 if (c != child && c->bs == child->bs) {
2805 break;
2806 }
2807 }
2808 if (c == NULL) {
2809 child->bs->inherits_from = NULL;
2810 }
2811 }
2812
2813 QLIST_FOREACH(c, &child->bs->children, next) {
2814 bdrv_unset_inherits_from(root, c);
2815 }
2816 }
2817
2818 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2819 {
2820 if (child == NULL) {
2821 return;
2822 }
2823
2824 bdrv_unset_inherits_from(parent, child);
2825 bdrv_root_unref_child(child);
2826 }
2827
2828
2829 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2830 {
2831 BdrvChild *c;
2832 QLIST_FOREACH(c, &bs->parents, next_parent) {
2833 if (c->klass->change_media) {
2834 c->klass->change_media(c, load);
2835 }
2836 }
2837 }
2838
2839 /* Return true if you can reach parent going through child->inherits_from
2840 * recursively. If parent or child are NULL, return false */
2841 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2842 BlockDriverState *parent)
2843 {
2844 while (child && child != parent) {
2845 child = child->inherits_from;
2846 }
2847
2848 return child != NULL;
2849 }
2850
2851 /*
2852 * Return the BdrvChildRole for @bs's backing child. bs->backing is
2853 * mostly used for COW backing children (role = COW), but also for
2854 * filtered children (role = FILTERED | PRIMARY).
2855 */
2856 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
2857 {
2858 if (bs->drv && bs->drv->is_filter) {
2859 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
2860 } else {
2861 return BDRV_CHILD_COW;
2862 }
2863 }
2864
2865 /*
2866 * Sets the backing file link of a BDS. A new reference is created; callers
2867 * which don't need their own reference any more must call bdrv_unref().
2868 */
2869 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2870 Error **errp)
2871 {
2872 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2873 bdrv_inherits_from_recursive(backing_hd, bs);
2874
2875 if (bdrv_is_backing_chain_frozen(bs, backing_bs(bs), errp)) {
2876 return;
2877 }
2878
2879 if (backing_hd) {
2880 bdrv_ref(backing_hd);
2881 }
2882
2883 if (bs->backing) {
2884 bdrv_unref_child(bs, bs->backing);
2885 bs->backing = NULL;
2886 }
2887
2888 if (!backing_hd) {
2889 goto out;
2890 }
2891
2892 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_of_bds,
2893 bdrv_backing_role(bs), errp);
2894 /* If backing_hd was already part of bs's backing chain, and
2895 * inherits_from pointed recursively to bs then let's update it to
2896 * point directly to bs (else it will become NULL). */
2897 if (bs->backing && update_inherits_from) {
2898 backing_hd->inherits_from = bs;
2899 }
2900
2901 out:
2902 bdrv_refresh_limits(bs, NULL);
2903 }
2904
2905 /*
2906 * Opens the backing file for a BlockDriverState if not yet open
2907 *
2908 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2909 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2910 * itself, all options starting with "${bdref_key}." are considered part of the
2911 * BlockdevRef.
2912 *
2913 * TODO Can this be unified with bdrv_open_image()?
2914 */
2915 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2916 const char *bdref_key, Error **errp)
2917 {
2918 char *backing_filename = NULL;
2919 char *bdref_key_dot;
2920 const char *reference = NULL;
2921 int ret = 0;
2922 bool implicit_backing = false;
2923 BlockDriverState *backing_hd;
2924 QDict *options;
2925 QDict *tmp_parent_options = NULL;
2926 Error *local_err = NULL;
2927
2928 if (bs->backing != NULL) {
2929 goto free_exit;
2930 }
2931
2932 /* NULL means an empty set of options */
2933 if (parent_options == NULL) {
2934 tmp_parent_options = qdict_new();
2935 parent_options = tmp_parent_options;
2936 }
2937
2938 bs->open_flags &= ~BDRV_O_NO_BACKING;
2939
2940 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2941 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2942 g_free(bdref_key_dot);
2943
2944 /*
2945 * Caution: while qdict_get_try_str() is fine, getting non-string
2946 * types would require more care. When @parent_options come from
2947 * -blockdev or blockdev_add, its members are typed according to
2948 * the QAPI schema, but when they come from -drive, they're all
2949 * QString.
2950 */
2951 reference = qdict_get_try_str(parent_options, bdref_key);
2952 if (reference || qdict_haskey(options, "file.filename")) {
2953 /* keep backing_filename NULL */
2954 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2955 qobject_unref(options);
2956 goto free_exit;
2957 } else {
2958 if (qdict_size(options) == 0) {
2959 /* If the user specifies options that do not modify the
2960 * backing file's behavior, we might still consider it the
2961 * implicit backing file. But it's easier this way, and
2962 * just specifying some of the backing BDS's options is
2963 * only possible with -drive anyway (otherwise the QAPI
2964 * schema forces the user to specify everything). */
2965 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2966 }
2967
2968 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2969 if (local_err) {
2970 ret = -EINVAL;
2971 error_propagate(errp, local_err);
2972 qobject_unref(options);
2973 goto free_exit;
2974 }
2975 }
2976
2977 if (!bs->drv || !bs->drv->supports_backing) {
2978 ret = -EINVAL;
2979 error_setg(errp, "Driver doesn't support backing files");
2980 qobject_unref(options);
2981 goto free_exit;
2982 }
2983
2984 if (!reference &&
2985 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2986 qdict_put_str(options, "driver", bs->backing_format);
2987 }
2988
2989 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2990 &child_of_bds, bdrv_backing_role(bs), errp);
2991 if (!backing_hd) {
2992 bs->open_flags |= BDRV_O_NO_BACKING;
2993 error_prepend(errp, "Could not open backing file: ");
2994 ret = -EINVAL;
2995 goto free_exit;
2996 }
2997
2998 if (implicit_backing) {
2999 bdrv_refresh_filename(backing_hd);
3000 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
3001 backing_hd->filename);
3002 }
3003
3004 /* Hook up the backing file link; drop our reference, bs owns the
3005 * backing_hd reference now */
3006 bdrv_set_backing_hd(bs, backing_hd, &local_err);
3007 bdrv_unref(backing_hd);
3008 if (local_err) {
3009 error_propagate(errp, local_err);
3010 ret = -EINVAL;
3011 goto free_exit;
3012 }
3013
3014 qdict_del(parent_options, bdref_key);
3015
3016 free_exit:
3017 g_free(backing_filename);
3018 qobject_unref(tmp_parent_options);
3019 return ret;
3020 }
3021
3022 static BlockDriverState *
3023 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
3024 BlockDriverState *parent, const BdrvChildClass *child_class,
3025 BdrvChildRole child_role, bool allow_none, Error **errp)
3026 {
3027 BlockDriverState *bs = NULL;
3028 QDict *image_options;
3029 char *bdref_key_dot;
3030 const char *reference;
3031
3032 assert(child_class != NULL);
3033
3034 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
3035 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
3036 g_free(bdref_key_dot);
3037
3038 /*
3039 * Caution: while qdict_get_try_str() is fine, getting non-string
3040 * types would require more care. When @options come from
3041 * -blockdev or blockdev_add, its members are typed according to
3042 * the QAPI schema, but when they come from -drive, they're all
3043 * QString.
3044 */
3045 reference = qdict_get_try_str(options, bdref_key);
3046 if (!filename && !reference && !qdict_size(image_options)) {
3047 if (!allow_none) {
3048 error_setg(errp, "A block device must be specified for \"%s\"",
3049 bdref_key);
3050 }
3051 qobject_unref(image_options);
3052 goto done;
3053 }
3054
3055 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3056 parent, child_class, child_role, errp);
3057 if (!bs) {
3058 goto done;
3059 }
3060
3061 done:
3062 qdict_del(options, bdref_key);
3063 return bs;
3064 }
3065
3066 /*
3067 * Opens a disk image whose options are given as BlockdevRef in another block
3068 * device's options.
3069 *
3070 * If allow_none is true, no image will be opened if filename is false and no
3071 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3072 *
3073 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3074 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3075 * itself, all options starting with "${bdref_key}." are considered part of the
3076 * BlockdevRef.
3077 *
3078 * The BlockdevRef will be removed from the options QDict.
3079 */
3080 BdrvChild *bdrv_open_child(const char *filename,
3081 QDict *options, const char *bdref_key,
3082 BlockDriverState *parent,
3083 const BdrvChildClass *child_class,
3084 BdrvChildRole child_role,
3085 bool allow_none, Error **errp)
3086 {
3087 BlockDriverState *bs;
3088
3089 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3090 child_role, allow_none, errp);
3091 if (bs == NULL) {
3092 return NULL;
3093 }
3094
3095 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3096 errp);
3097 }
3098
3099 /*
3100 * TODO Future callers may need to specify parent/child_class in order for
3101 * option inheritance to work. Existing callers use it for the root node.
3102 */
3103 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3104 {
3105 BlockDriverState *bs = NULL;
3106 QObject *obj = NULL;
3107 QDict *qdict = NULL;
3108 const char *reference = NULL;
3109 Visitor *v = NULL;
3110
3111 if (ref->type == QTYPE_QSTRING) {
3112 reference = ref->u.reference;
3113 } else {
3114 BlockdevOptions *options = &ref->u.definition;
3115 assert(ref->type == QTYPE_QDICT);
3116
3117 v = qobject_output_visitor_new(&obj);
3118 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3119 visit_complete(v, &obj);
3120
3121 qdict = qobject_to(QDict, obj);
3122 qdict_flatten(qdict);
3123
3124 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3125 * compatibility with other callers) rather than what we want as the
3126 * real defaults. Apply the defaults here instead. */
3127 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3128 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3129 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3130 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3131
3132 }
3133
3134 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3135 obj = NULL;
3136 qobject_unref(obj);
3137 visit_free(v);
3138 return bs;
3139 }
3140
3141 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3142 int flags,
3143 QDict *snapshot_options,
3144 Error **errp)
3145 {
3146 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3147 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3148 int64_t total_size;
3149 QemuOpts *opts = NULL;
3150 BlockDriverState *bs_snapshot = NULL;
3151 Error *local_err = NULL;
3152 int ret;
3153
3154 /* if snapshot, we create a temporary backing file and open it
3155 instead of opening 'filename' directly */
3156
3157 /* Get the required size from the image */
3158 total_size = bdrv_getlength(bs);
3159 if (total_size < 0) {
3160 error_setg_errno(errp, -total_size, "Could not get image size");
3161 goto out;
3162 }
3163
3164 /* Create the temporary image */
3165 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3166 if (ret < 0) {
3167 error_setg_errno(errp, -ret, "Could not get temporary filename");
3168 goto out;
3169 }
3170
3171 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3172 &error_abort);
3173 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3174 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3175 qemu_opts_del(opts);
3176 if (ret < 0) {
3177 error_prepend(errp, "Could not create temporary overlay '%s': ",
3178 tmp_filename);
3179 goto out;
3180 }
3181
3182 /* Prepare options QDict for the temporary file */
3183 qdict_put_str(snapshot_options, "file.driver", "file");
3184 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3185 qdict_put_str(snapshot_options, "driver", "qcow2");
3186
3187 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3188 snapshot_options = NULL;
3189 if (!bs_snapshot) {
3190 goto out;
3191 }
3192
3193 /* bdrv_append() consumes a strong reference to bs_snapshot
3194 * (i.e. it will call bdrv_unref() on it) even on error, so in
3195 * order to be able to return one, we have to increase
3196 * bs_snapshot's refcount here */
3197 bdrv_ref(bs_snapshot);
3198 bdrv_append(bs_snapshot, bs, &local_err);
3199 if (local_err) {
3200 error_propagate(errp, local_err);
3201 bs_snapshot = NULL;
3202 goto out;
3203 }
3204
3205 out:
3206 qobject_unref(snapshot_options);
3207 g_free(tmp_filename);
3208 return bs_snapshot;
3209 }
3210
3211 /*
3212 * Opens a disk image (raw, qcow2, vmdk, ...)
3213 *
3214 * options is a QDict of options to pass to the block drivers, or NULL for an
3215 * empty set of options. The reference to the QDict belongs to the block layer
3216 * after the call (even on failure), so if the caller intends to reuse the
3217 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3218 *
3219 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3220 * If it is not NULL, the referenced BDS will be reused.
3221 *
3222 * The reference parameter may be used to specify an existing block device which
3223 * should be opened. If specified, neither options nor a filename may be given,
3224 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3225 */
3226 static BlockDriverState *bdrv_open_inherit(const char *filename,
3227 const char *reference,
3228 QDict *options, int flags,
3229 BlockDriverState *parent,
3230 const BdrvChildClass *child_class,
3231 BdrvChildRole child_role,
3232 Error **errp)
3233 {
3234 int ret;
3235 BlockBackend *file = NULL;
3236 BlockDriverState *bs;
3237 BlockDriver *drv = NULL;
3238 BdrvChild *child;
3239 const char *drvname;
3240 const char *backing;
3241 Error *local_err = NULL;
3242 QDict *snapshot_options = NULL;
3243 int snapshot_flags = 0;
3244
3245 assert(!child_class || !flags);
3246 assert(!child_class == !parent);
3247
3248 if (reference) {
3249 bool options_non_empty = options ? qdict_size(options) : false;
3250 qobject_unref(options);
3251
3252 if (filename || options_non_empty) {
3253 error_setg(errp, "Cannot reference an existing block device with "
3254 "additional options or a new filename");
3255 return NULL;
3256 }
3257
3258 bs = bdrv_lookup_bs(reference, reference, errp);
3259 if (!bs) {
3260 return NULL;
3261 }
3262
3263 bdrv_ref(bs);
3264 return bs;
3265 }
3266
3267 bs = bdrv_new();
3268
3269 /* NULL means an empty set of options */
3270 if (options == NULL) {
3271 options = qdict_new();
3272 }
3273
3274 /* json: syntax counts as explicit options, as if in the QDict */
3275 parse_json_protocol(options, &filename, &local_err);
3276 if (local_err) {
3277 goto fail;
3278 }
3279
3280 bs->explicit_options = qdict_clone_shallow(options);
3281
3282 if (child_class) {
3283 bool parent_is_format;
3284
3285 if (parent->drv) {
3286 parent_is_format = parent->drv->is_format;
3287 } else {
3288 /*
3289 * parent->drv is not set yet because this node is opened for
3290 * (potential) format probing. That means that @parent is going
3291 * to be a format node.
3292 */
3293 parent_is_format = true;
3294 }
3295
3296 bs->inherits_from = parent;
3297 child_class->inherit_options(child_role, parent_is_format,
3298 &flags, options,
3299 parent->open_flags, parent->options);
3300 }
3301
3302 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3303 if (ret < 0) {
3304 goto fail;
3305 }
3306
3307 /*
3308 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3309 * Caution: getting a boolean member of @options requires care.
3310 * When @options come from -blockdev or blockdev_add, members are
3311 * typed according to the QAPI schema, but when they come from
3312 * -drive, they're all QString.
3313 */
3314 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3315 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3316 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3317 } else {
3318 flags &= ~BDRV_O_RDWR;
3319 }
3320
3321 if (flags & BDRV_O_SNAPSHOT) {
3322 snapshot_options = qdict_new();
3323 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3324 flags, options);
3325 /* Let bdrv_backing_options() override "read-only" */
3326 qdict_del(options, BDRV_OPT_READ_ONLY);
3327 bdrv_inherited_options(BDRV_CHILD_COW, true,
3328 &flags, options, flags, options);
3329 }
3330
3331 bs->open_flags = flags;
3332 bs->options = options;
3333 options = qdict_clone_shallow(options);
3334
3335 /* Find the right image format driver */
3336 /* See cautionary note on accessing @options above */
3337 drvname = qdict_get_try_str(options, "driver");
3338 if (drvname) {
3339 drv = bdrv_find_format(drvname);
3340 if (!drv) {
3341 error_setg(errp, "Unknown driver: '%s'", drvname);
3342 goto fail;
3343 }
3344 }
3345
3346 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3347
3348 /* See cautionary note on accessing @options above */
3349 backing = qdict_get_try_str(options, "backing");
3350 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3351 (backing && *backing == '\0'))
3352 {
3353 if (backing) {
3354 warn_report("Use of \"backing\": \"\" is deprecated; "
3355 "use \"backing\": null instead");
3356 }
3357 flags |= BDRV_O_NO_BACKING;
3358 qdict_del(bs->explicit_options, "backing");
3359 qdict_del(bs->options, "backing");
3360 qdict_del(options, "backing");
3361 }
3362
3363 /* Open image file without format layer. This BlockBackend is only used for
3364 * probing, the block drivers will do their own bdrv_open_child() for the
3365 * same BDS, which is why we put the node name back into options. */
3366 if ((flags & BDRV_O_PROTOCOL) == 0) {
3367 BlockDriverState *file_bs;
3368
3369 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3370 &child_of_bds, BDRV_CHILD_IMAGE,
3371 true, &local_err);
3372 if (local_err) {
3373 goto fail;
3374 }
3375 if (file_bs != NULL) {
3376 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3377 * looking at the header to guess the image format. This works even
3378 * in cases where a guest would not see a consistent state. */
3379 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3380 blk_insert_bs(file, file_bs, &local_err);
3381 bdrv_unref(file_bs);
3382 if (local_err) {
3383 goto fail;
3384 }
3385
3386 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3387 }
3388 }
3389
3390 /* Image format probing */
3391 bs->probed = !drv;
3392 if (!drv && file) {
3393 ret = find_image_format(file, filename, &drv, &local_err);
3394 if (ret < 0) {
3395 goto fail;
3396 }
3397 /*
3398 * This option update would logically belong in bdrv_fill_options(),
3399 * but we first need to open bs->file for the probing to work, while
3400 * opening bs->file already requires the (mostly) final set of options
3401 * so that cache mode etc. can be inherited.
3402 *
3403 * Adding the driver later is somewhat ugly, but it's not an option
3404 * that would ever be inherited, so it's correct. We just need to make
3405 * sure to update both bs->options (which has the full effective
3406 * options for bs) and options (which has file.* already removed).
3407 */
3408 qdict_put_str(bs->options, "driver", drv->format_name);
3409 qdict_put_str(options, "driver", drv->format_name);
3410 } else if (!drv) {
3411 error_setg(errp, "Must specify either driver or file");
3412 goto fail;
3413 }
3414
3415 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3416 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3417 /* file must be NULL if a protocol BDS is about to be created
3418 * (the inverse results in an error message from bdrv_open_common()) */
3419 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3420
3421 /* Open the image */
3422 ret = bdrv_open_common(bs, file, options, &local_err);
3423 if (ret < 0) {
3424 goto fail;
3425 }
3426
3427 if (file) {
3428 blk_unref(file);
3429 file = NULL;
3430 }
3431
3432 /* If there is a backing file, use it */
3433 if ((flags & BDRV_O_NO_BACKING) == 0) {
3434 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3435 if (ret < 0) {
3436 goto close_and_fail;
3437 }
3438 }
3439
3440 /* Remove all children options and references
3441 * from bs->options and bs->explicit_options */
3442 QLIST_FOREACH(child, &bs->children, next) {
3443 char *child_key_dot;
3444 child_key_dot = g_strdup_printf("%s.", child->name);
3445 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3446 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3447 qdict_del(bs->explicit_options, child->name);
3448 qdict_del(bs->options, child->name);
3449 g_free(child_key_dot);
3450 }
3451
3452 /* Check if any unknown options were used */
3453 if (qdict_size(options) != 0) {
3454 const QDictEntry *entry = qdict_first(options);
3455 if (flags & BDRV_O_PROTOCOL) {
3456 error_setg(errp, "Block protocol '%s' doesn't support the option "
3457 "'%s'", drv->format_name, entry->key);
3458 } else {
3459 error_setg(errp,
3460 "Block format '%s' does not support the option '%s'",
3461 drv->format_name, entry->key);
3462 }
3463
3464 goto close_and_fail;
3465 }
3466
3467 bdrv_parent_cb_change_media(bs, true);
3468
3469 qobject_unref(options);
3470 options = NULL;
3471
3472 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3473 * temporary snapshot afterwards. */
3474 if (snapshot_flags) {
3475 BlockDriverState *snapshot_bs;
3476 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3477 snapshot_options, &local_err);
3478 snapshot_options = NULL;
3479 if (local_err) {
3480 goto close_and_fail;
3481 }
3482 /* We are not going to return bs but the overlay on top of it
3483 * (snapshot_bs); thus, we have to drop the strong reference to bs
3484 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3485 * though, because the overlay still has a reference to it. */
3486 bdrv_unref(bs);
3487 bs = snapshot_bs;
3488 }
3489
3490 return bs;
3491
3492 fail:
3493 blk_unref(file);
3494 qobject_unref(snapshot_options);
3495 qobject_unref(bs->explicit_options);
3496 qobject_unref(bs->options);
3497 qobject_unref(options);
3498 bs->options = NULL;
3499 bs->explicit_options = NULL;
3500 bdrv_unref(bs);
3501 error_propagate(errp, local_err);
3502 return NULL;
3503
3504 close_and_fail:
3505 bdrv_unref(bs);
3506 qobject_unref(snapshot_options);
3507 qobject_unref(options);
3508 error_propagate(errp, local_err);
3509 return NULL;
3510 }
3511
3512 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3513 QDict *options, int flags, Error **errp)
3514 {
3515 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3516 NULL, 0, errp);
3517 }
3518
3519 /* Return true if the NULL-terminated @list contains @str */
3520 static bool is_str_in_list(const char *str, const char *const *list)
3521 {
3522 if (str && list) {
3523 int i;
3524 for (i = 0; list[i] != NULL; i++) {
3525 if (!strcmp(str, list[i])) {
3526 return true;
3527 }
3528 }
3529 }
3530 return false;
3531 }
3532
3533 /*
3534 * Check that every option set in @bs->options is also set in
3535 * @new_opts.
3536 *
3537 * Options listed in the common_options list and in
3538 * @bs->drv->mutable_opts are skipped.
3539 *
3540 * Return 0 on success, otherwise return -EINVAL and set @errp.
3541 */
3542 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3543 const QDict *new_opts, Error **errp)
3544 {
3545 const QDictEntry *e;
3546 /* These options are common to all block drivers and are handled
3547 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3548 const char *const common_options[] = {
3549 "node-name", "discard", "cache.direct", "cache.no-flush",
3550 "read-only", "auto-read-only", "detect-zeroes", NULL
3551 };
3552
3553 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3554 if (!qdict_haskey(new_opts, e->key) &&
3555 !is_str_in_list(e->key, common_options) &&
3556 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3557 error_setg(errp, "Option '%s' cannot be reset "
3558