Merge remote-tracking branch 'remotes/cohuck-gitlab/tags/s390x-20210121' into staging
[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/fuse.h"
30 #include "block/nbd.h"
31 #include "block/qdict.h"
32 #include "qemu/error-report.h"
33 #include "block/module_block.h"
34 #include "qemu/main-loop.h"
35 #include "qemu/module.h"
36 #include "qapi/error.h"
37 #include "qapi/qmp/qdict.h"
38 #include "qapi/qmp/qjson.h"
39 #include "qapi/qmp/qnull.h"
40 #include "qapi/qmp/qstring.h"
41 #include "qapi/qobject-output-visitor.h"
42 #include "qapi/qapi-visit-block-core.h"
43 #include "sysemu/block-backend.h"
44 #include "sysemu/sysemu.h"
45 #include "qemu/notify.h"
46 #include "qemu/option.h"
47 #include "qemu/coroutine.h"
48 #include "block/qapi.h"
49 #include "qemu/timer.h"
50 #include "qemu/cutils.h"
51 #include "qemu/id.h"
52 #include "block/coroutines.h"
53
54 #ifdef CONFIG_BSD
55 #include <sys/ioctl.h>
56 #include <sys/queue.h>
57 #ifndef __DragonFly__
58 #include <sys/disk.h>
59 #endif
60 #endif
61
62 #ifdef _WIN32
63 #include <windows.h>
64 #endif
65
66 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
67
68 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
69 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
70
71 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
72 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
73
74 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
75 QLIST_HEAD_INITIALIZER(bdrv_drivers);
76
77 static BlockDriverState *bdrv_open_inherit(const char *filename,
78 const char *reference,
79 QDict *options, int flags,
80 BlockDriverState *parent,
81 const BdrvChildClass *child_class,
82 BdrvChildRole child_role,
83 Error **errp);
84
85 /* If non-zero, use only whitelisted block drivers */
86 static int use_bdrv_whitelist;
87
88 #ifdef _WIN32
89 static int is_windows_drive_prefix(const char *filename)
90 {
91 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
92 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
93 filename[1] == ':');
94 }
95
96 int is_windows_drive(const char *filename)
97 {
98 if (is_windows_drive_prefix(filename) &&
99 filename[2] == '\0')
100 return 1;
101 if (strstart(filename, "\\\\.\\", NULL) ||
102 strstart(filename, "//./", NULL))
103 return 1;
104 return 0;
105 }
106 #endif
107
108 size_t bdrv_opt_mem_align(BlockDriverState *bs)
109 {
110 if (!bs || !bs->drv) {
111 /* page size or 4k (hdd sector size) should be on the safe side */
112 return MAX(4096, qemu_real_host_page_size);
113 }
114
115 return bs->bl.opt_mem_alignment;
116 }
117
118 size_t bdrv_min_mem_align(BlockDriverState *bs)
119 {
120 if (!bs || !bs->drv) {
121 /* page size or 4k (hdd sector size) should be on the safe side */
122 return MAX(4096, qemu_real_host_page_size);
123 }
124
125 return bs->bl.min_mem_alignment;
126 }
127
128 /* check if the path starts with "<protocol>:" */
129 int path_has_protocol(const char *path)
130 {
131 const char *p;
132
133 #ifdef _WIN32
134 if (is_windows_drive(path) ||
135 is_windows_drive_prefix(path)) {
136 return 0;
137 }
138 p = path + strcspn(path, ":/\\");
139 #else
140 p = path + strcspn(path, ":/");
141 #endif
142
143 return *p == ':';
144 }
145
146 int path_is_absolute(const char *path)
147 {
148 #ifdef _WIN32
149 /* specific case for names like: "\\.\d:" */
150 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
151 return 1;
152 }
153 return (*path == '/' || *path == '\\');
154 #else
155 return (*path == '/');
156 #endif
157 }
158
159 /* if filename is absolute, just return its duplicate. Otherwise, build a
160 path to it by considering it is relative to base_path. URL are
161 supported. */
162 char *path_combine(const char *base_path, const char *filename)
163 {
164 const char *protocol_stripped = NULL;
165 const char *p, *p1;
166 char *result;
167 int len;
168
169 if (path_is_absolute(filename)) {
170 return g_strdup(filename);
171 }
172
173 if (path_has_protocol(base_path)) {
174 protocol_stripped = strchr(base_path, ':');
175 if (protocol_stripped) {
176 protocol_stripped++;
177 }
178 }
179 p = protocol_stripped ?: base_path;
180
181 p1 = strrchr(base_path, '/');
182 #ifdef _WIN32
183 {
184 const char *p2;
185 p2 = strrchr(base_path, '\\');
186 if (!p1 || p2 > p1) {
187 p1 = p2;
188 }
189 }
190 #endif
191 if (p1) {
192 p1++;
193 } else {
194 p1 = base_path;
195 }
196 if (p1 > p) {
197 p = p1;
198 }
199 len = p - base_path;
200
201 result = g_malloc(len + strlen(filename) + 1);
202 memcpy(result, base_path, len);
203 strcpy(result + len, filename);
204
205 return result;
206 }
207
208 /*
209 * Helper function for bdrv_parse_filename() implementations to remove optional
210 * protocol prefixes (especially "file:") from a filename and for putting the
211 * stripped filename into the options QDict if there is such a prefix.
212 */
213 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
214 QDict *options)
215 {
216 if (strstart(filename, prefix, &filename)) {
217 /* Stripping the explicit protocol prefix may result in a protocol
218 * prefix being (wrongly) detected (if the filename contains a colon) */
219 if (path_has_protocol(filename)) {
220 GString *fat_filename;
221
222 /* This means there is some colon before the first slash; therefore,
223 * this cannot be an absolute path */
224 assert(!path_is_absolute(filename));
225
226 /* And we can thus fix the protocol detection issue by prefixing it
227 * by "./" */
228 fat_filename = g_string_new("./");
229 g_string_append(fat_filename, filename);
230
231 assert(!path_has_protocol(fat_filename->str));
232
233 qdict_put(options, "filename",
234 qstring_from_gstring(fat_filename));
235 } else {
236 /* If no protocol prefix was detected, we can use the shortened
237 * filename as-is */
238 qdict_put_str(options, "filename", filename);
239 }
240 }
241 }
242
243
244 /* Returns whether the image file is opened as read-only. Note that this can
245 * return false and writing to the image file is still not possible because the
246 * image is inactivated. */
247 bool bdrv_is_read_only(BlockDriverState *bs)
248 {
249 return bs->read_only;
250 }
251
252 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
253 bool ignore_allow_rdw, Error **errp)
254 {
255 /* Do not set read_only if copy_on_read is enabled */
256 if (bs->copy_on_read && read_only) {
257 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
258 bdrv_get_device_or_node_name(bs));
259 return -EINVAL;
260 }
261
262 /* Do not clear read_only if it is prohibited */
263 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
264 !ignore_allow_rdw)
265 {
266 error_setg(errp, "Node '%s' is read only",
267 bdrv_get_device_or_node_name(bs));
268 return -EPERM;
269 }
270
271 return 0;
272 }
273
274 /*
275 * Called by a driver that can only provide a read-only image.
276 *
277 * Returns 0 if the node is already read-only or it could switch the node to
278 * read-only because BDRV_O_AUTO_RDONLY is set.
279 *
280 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
281 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
282 * is not NULL, it is used as the error message for the Error object.
283 */
284 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
285 Error **errp)
286 {
287 int ret = 0;
288
289 if (!(bs->open_flags & BDRV_O_RDWR)) {
290 return 0;
291 }
292 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
293 goto fail;
294 }
295
296 ret = bdrv_can_set_read_only(bs, true, false, NULL);
297 if (ret < 0) {
298 goto fail;
299 }
300
301 bs->read_only = true;
302 bs->open_flags &= ~BDRV_O_RDWR;
303
304 return 0;
305
306 fail:
307 error_setg(errp, "%s", errmsg ?: "Image is read-only");
308 return -EACCES;
309 }
310
311 /*
312 * If @backing is empty, this function returns NULL without setting
313 * @errp. In all other cases, NULL will only be returned with @errp
314 * set.
315 *
316 * Therefore, a return value of NULL without @errp set means that
317 * there is no backing file; if @errp is set, there is one but its
318 * absolute filename cannot be generated.
319 */
320 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
321 const char *backing,
322 Error **errp)
323 {
324 if (backing[0] == '\0') {
325 return NULL;
326 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
327 return g_strdup(backing);
328 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
329 error_setg(errp, "Cannot use relative backing file names for '%s'",
330 backed);
331 return NULL;
332 } else {
333 return path_combine(backed, backing);
334 }
335 }
336
337 /*
338 * If @filename is empty or NULL, this function returns NULL without
339 * setting @errp. In all other cases, NULL will only be returned with
340 * @errp set.
341 */
342 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
343 const char *filename, Error **errp)
344 {
345 char *dir, *full_name;
346
347 if (!filename || filename[0] == '\0') {
348 return NULL;
349 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
350 return g_strdup(filename);
351 }
352
353 dir = bdrv_dirname(relative_to, errp);
354 if (!dir) {
355 return NULL;
356 }
357
358 full_name = g_strconcat(dir, filename, NULL);
359 g_free(dir);
360 return full_name;
361 }
362
363 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
364 {
365 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
366 }
367
368 void bdrv_register(BlockDriver *bdrv)
369 {
370 assert(bdrv->format_name);
371 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
372 }
373
374 BlockDriverState *bdrv_new(void)
375 {
376 BlockDriverState *bs;
377 int i;
378
379 bs = g_new0(BlockDriverState, 1);
380 QLIST_INIT(&bs->dirty_bitmaps);
381 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
382 QLIST_INIT(&bs->op_blockers[i]);
383 }
384 notifier_with_return_list_init(&bs->before_write_notifiers);
385 qemu_co_mutex_init(&bs->reqs_lock);
386 qemu_mutex_init(&bs->dirty_bitmap_mutex);
387 bs->refcnt = 1;
388 bs->aio_context = qemu_get_aio_context();
389
390 qemu_co_queue_init(&bs->flush_queue);
391
392 for (i = 0; i < bdrv_drain_all_count; i++) {
393 bdrv_drained_begin(bs);
394 }
395
396 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
397
398 return bs;
399 }
400
401 static BlockDriver *bdrv_do_find_format(const char *format_name)
402 {
403 BlockDriver *drv1;
404
405 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
406 if (!strcmp(drv1->format_name, format_name)) {
407 return drv1;
408 }
409 }
410
411 return NULL;
412 }
413
414 BlockDriver *bdrv_find_format(const char *format_name)
415 {
416 BlockDriver *drv1;
417 int i;
418
419 drv1 = bdrv_do_find_format(format_name);
420 if (drv1) {
421 return drv1;
422 }
423
424 /* The driver isn't registered, maybe we need to load a module */
425 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
426 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
427 block_module_load_one(block_driver_modules[i].library_name);
428 break;
429 }
430 }
431
432 return bdrv_do_find_format(format_name);
433 }
434
435 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
436 {
437 static const char *whitelist_rw[] = {
438 CONFIG_BDRV_RW_WHITELIST
439 NULL
440 };
441 static const char *whitelist_ro[] = {
442 CONFIG_BDRV_RO_WHITELIST
443 NULL
444 };
445 const char **p;
446
447 if (!whitelist_rw[0] && !whitelist_ro[0]) {
448 return 1; /* no whitelist, anything goes */
449 }
450
451 for (p = whitelist_rw; *p; p++) {
452 if (!strcmp(format_name, *p)) {
453 return 1;
454 }
455 }
456 if (read_only) {
457 for (p = whitelist_ro; *p; p++) {
458 if (!strcmp(format_name, *p)) {
459 return 1;
460 }
461 }
462 }
463 return 0;
464 }
465
466 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
467 {
468 return bdrv_format_is_whitelisted(drv->format_name, read_only);
469 }
470
471 bool bdrv_uses_whitelist(void)
472 {
473 return use_bdrv_whitelist;
474 }
475
476 typedef struct CreateCo {
477 BlockDriver *drv;
478 char *filename;
479 QemuOpts *opts;
480 int ret;
481 Error *err;
482 } CreateCo;
483
484 static void coroutine_fn bdrv_create_co_entry(void *opaque)
485 {
486 Error *local_err = NULL;
487 int ret;
488
489 CreateCo *cco = opaque;
490 assert(cco->drv);
491
492 ret = cco->drv->bdrv_co_create_opts(cco->drv,
493 cco->filename, cco->opts, &local_err);
494 error_propagate(&cco->err, local_err);
495 cco->ret = ret;
496 }
497
498 int bdrv_create(BlockDriver *drv, const char* filename,
499 QemuOpts *opts, Error **errp)
500 {
501 int ret;
502
503 Coroutine *co;
504 CreateCo cco = {
505 .drv = drv,
506 .filename = g_strdup(filename),
507 .opts = opts,
508 .ret = NOT_DONE,
509 .err = NULL,
510 };
511
512 if (!drv->bdrv_co_create_opts) {
513 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
514 ret = -ENOTSUP;
515 goto out;
516 }
517
518 if (qemu_in_coroutine()) {
519 /* Fast-path if already in coroutine context */
520 bdrv_create_co_entry(&cco);
521 } else {
522 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
523 qemu_coroutine_enter(co);
524 while (cco.ret == NOT_DONE) {
525 aio_poll(qemu_get_aio_context(), true);
526 }
527 }
528
529 ret = cco.ret;
530 if (ret < 0) {
531 if (cco.err) {
532 error_propagate(errp, cco.err);
533 } else {
534 error_setg_errno(errp, -ret, "Could not create image");
535 }
536 }
537
538 out:
539 g_free(cco.filename);
540 return ret;
541 }
542
543 /**
544 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
545 * least the given @minimum_size.
546 *
547 * On success, return @blk's actual length.
548 * Otherwise, return -errno.
549 */
550 static int64_t create_file_fallback_truncate(BlockBackend *blk,
551 int64_t minimum_size, Error **errp)
552 {
553 Error *local_err = NULL;
554 int64_t size;
555 int ret;
556
557 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, 0,
558 &local_err);
559 if (ret < 0 && ret != -ENOTSUP) {
560 error_propagate(errp, local_err);
561 return ret;
562 }
563
564 size = blk_getlength(blk);
565 if (size < 0) {
566 error_free(local_err);
567 error_setg_errno(errp, -size,
568 "Failed to inquire the new image file's length");
569 return size;
570 }
571
572 if (size < minimum_size) {
573 /* Need to grow the image, but we failed to do that */
574 error_propagate(errp, local_err);
575 return -ENOTSUP;
576 }
577
578 error_free(local_err);
579 local_err = NULL;
580
581 return size;
582 }
583
584 /**
585 * Helper function for bdrv_create_file_fallback(): Zero the first
586 * sector to remove any potentially pre-existing image header.
587 */
588 static int create_file_fallback_zero_first_sector(BlockBackend *blk,
589 int64_t current_size,
590 Error **errp)
591 {
592 int64_t bytes_to_clear;
593 int ret;
594
595 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
596 if (bytes_to_clear) {
597 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
598 if (ret < 0) {
599 error_setg_errno(errp, -ret,
600 "Failed to clear the new image's first sector");
601 return ret;
602 }
603 }
604
605 return 0;
606 }
607
608 /**
609 * Simple implementation of bdrv_co_create_opts for protocol drivers
610 * which only support creation via opening a file
611 * (usually existing raw storage device)
612 */
613 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
614 const char *filename,
615 QemuOpts *opts,
616 Error **errp)
617 {
618 BlockBackend *blk;
619 QDict *options;
620 int64_t size = 0;
621 char *buf = NULL;
622 PreallocMode prealloc;
623 Error *local_err = NULL;
624 int ret;
625
626 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
627 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
628 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
629 PREALLOC_MODE_OFF, &local_err);
630 g_free(buf);
631 if (local_err) {
632 error_propagate(errp, local_err);
633 return -EINVAL;
634 }
635
636 if (prealloc != PREALLOC_MODE_OFF) {
637 error_setg(errp, "Unsupported preallocation mode '%s'",
638 PreallocMode_str(prealloc));
639 return -ENOTSUP;
640 }
641
642 options = qdict_new();
643 qdict_put_str(options, "driver", drv->format_name);
644
645 blk = blk_new_open(filename, NULL, options,
646 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
647 if (!blk) {
648 error_prepend(errp, "Protocol driver '%s' does not support image "
649 "creation, and opening the image failed: ",
650 drv->format_name);
651 return -EINVAL;
652 }
653
654 size = create_file_fallback_truncate(blk, size, errp);
655 if (size < 0) {
656 ret = size;
657 goto out;
658 }
659
660 ret = create_file_fallback_zero_first_sector(blk, size, errp);
661 if (ret < 0) {
662 goto out;
663 }
664
665 ret = 0;
666 out:
667 blk_unref(blk);
668 return ret;
669 }
670
671 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
672 {
673 BlockDriver *drv;
674
675 drv = bdrv_find_protocol(filename, true, errp);
676 if (drv == NULL) {
677 return -ENOENT;
678 }
679
680 return bdrv_create(drv, filename, opts, errp);
681 }
682
683 int coroutine_fn bdrv_co_delete_file(BlockDriverState *bs, Error **errp)
684 {
685 Error *local_err = NULL;
686 int ret;
687
688 assert(bs != NULL);
689
690 if (!bs->drv) {
691 error_setg(errp, "Block node '%s' is not opened", bs->filename);
692 return -ENOMEDIUM;
693 }
694
695 if (!bs->drv->bdrv_co_delete_file) {
696 error_setg(errp, "Driver '%s' does not support image deletion",
697 bs->drv->format_name);
698 return -ENOTSUP;
699 }
700
701 ret = bs->drv->bdrv_co_delete_file(bs, &local_err);
702 if (ret < 0) {
703 error_propagate(errp, local_err);
704 }
705
706 return ret;
707 }
708
709 /**
710 * Try to get @bs's logical and physical block size.
711 * On success, store them in @bsz struct and return 0.
712 * On failure return -errno.
713 * @bs must not be empty.
714 */
715 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
716 {
717 BlockDriver *drv = bs->drv;
718 BlockDriverState *filtered = bdrv_filter_bs(bs);
719
720 if (drv && drv->bdrv_probe_blocksizes) {
721 return drv->bdrv_probe_blocksizes(bs, bsz);
722 } else if (filtered) {
723 return bdrv_probe_blocksizes(filtered, bsz);
724 }
725
726 return -ENOTSUP;
727 }
728
729 /**
730 * Try to get @bs's geometry (cyls, heads, sectors).
731 * On success, store them in @geo struct and return 0.
732 * On failure return -errno.
733 * @bs must not be empty.
734 */
735 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
736 {
737 BlockDriver *drv = bs->drv;
738 BlockDriverState *filtered = bdrv_filter_bs(bs);
739
740 if (drv && drv->bdrv_probe_geometry) {
741 return drv->bdrv_probe_geometry(bs, geo);
742 } else if (filtered) {
743 return bdrv_probe_geometry(filtered, geo);
744 }
745
746 return -ENOTSUP;
747 }
748
749 /*
750 * Create a uniquely-named empty temporary file.
751 * Return 0 upon success, otherwise a negative errno value.
752 */
753 int get_tmp_filename(char *filename, int size)
754 {
755 #ifdef _WIN32
756 char temp_dir[MAX_PATH];
757 /* GetTempFileName requires that its output buffer (4th param)
758 have length MAX_PATH or greater. */
759 assert(size >= MAX_PATH);
760 return (GetTempPath(MAX_PATH, temp_dir)
761 && GetTempFileName(temp_dir, "qem", 0, filename)
762 ? 0 : -GetLastError());
763 #else
764 int fd;
765 const char *tmpdir;
766 tmpdir = getenv("TMPDIR");
767 if (!tmpdir) {
768 tmpdir = "/var/tmp";
769 }
770 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
771 return -EOVERFLOW;
772 }
773 fd = mkstemp(filename);
774 if (fd < 0) {
775 return -errno;
776 }
777 if (close(fd) != 0) {
778 unlink(filename);
779 return -errno;
780 }
781 return 0;
782 #endif
783 }
784
785 /*
786 * Detect host devices. By convention, /dev/cdrom[N] is always
787 * recognized as a host CDROM.
788 */
789 static BlockDriver *find_hdev_driver(const char *filename)
790 {
791 int score_max = 0, score;
792 BlockDriver *drv = NULL, *d;
793
794 QLIST_FOREACH(d, &bdrv_drivers, list) {
795 if (d->bdrv_probe_device) {
796 score = d->bdrv_probe_device(filename);
797 if (score > score_max) {
798 score_max = score;
799 drv = d;
800 }
801 }
802 }
803
804 return drv;
805 }
806
807 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
808 {
809 BlockDriver *drv1;
810
811 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
812 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
813 return drv1;
814 }
815 }
816
817 return NULL;
818 }
819
820 BlockDriver *bdrv_find_protocol(const char *filename,
821 bool allow_protocol_prefix,
822 Error **errp)
823 {
824 BlockDriver *drv1;
825 char protocol[128];
826 int len;
827 const char *p;
828 int i;
829
830 /* TODO Drivers without bdrv_file_open must be specified explicitly */
831
832 /*
833 * XXX(hch): we really should not let host device detection
834 * override an explicit protocol specification, but moving this
835 * later breaks access to device names with colons in them.
836 * Thanks to the brain-dead persistent naming schemes on udev-
837 * based Linux systems those actually are quite common.
838 */
839 drv1 = find_hdev_driver(filename);
840 if (drv1) {
841 return drv1;
842 }
843
844 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
845 return &bdrv_file;
846 }
847
848 p = strchr(filename, ':');
849 assert(p != NULL);
850 len = p - filename;
851 if (len > sizeof(protocol) - 1)
852 len = sizeof(protocol) - 1;
853 memcpy(protocol, filename, len);
854 protocol[len] = '\0';
855
856 drv1 = bdrv_do_find_protocol(protocol);
857 if (drv1) {
858 return drv1;
859 }
860
861 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
862 if (block_driver_modules[i].protocol_name &&
863 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
864 block_module_load_one(block_driver_modules[i].library_name);
865 break;
866 }
867 }
868
869 drv1 = bdrv_do_find_protocol(protocol);
870 if (!drv1) {
871 error_setg(errp, "Unknown protocol '%s'", protocol);
872 }
873 return drv1;
874 }
875
876 /*
877 * Guess image format by probing its contents.
878 * This is not a good idea when your image is raw (CVE-2008-2004), but
879 * we do it anyway for backward compatibility.
880 *
881 * @buf contains the image's first @buf_size bytes.
882 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
883 * but can be smaller if the image file is smaller)
884 * @filename is its filename.
885 *
886 * For all block drivers, call the bdrv_probe() method to get its
887 * probing score.
888 * Return the first block driver with the highest probing score.
889 */
890 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
891 const char *filename)
892 {
893 int score_max = 0, score;
894 BlockDriver *drv = NULL, *d;
895
896 QLIST_FOREACH(d, &bdrv_drivers, list) {
897 if (d->bdrv_probe) {
898 score = d->bdrv_probe(buf, buf_size, filename);
899 if (score > score_max) {
900 score_max = score;
901 drv = d;
902 }
903 }
904 }
905
906 return drv;
907 }
908
909 static int find_image_format(BlockBackend *file, const char *filename,
910 BlockDriver **pdrv, Error **errp)
911 {
912 BlockDriver *drv;
913 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
914 int ret = 0;
915
916 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
917 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
918 *pdrv = &bdrv_raw;
919 return ret;
920 }
921
922 ret = blk_pread(file, 0, buf, sizeof(buf));
923 if (ret < 0) {
924 error_setg_errno(errp, -ret, "Could not read image for determining its "
925 "format");
926 *pdrv = NULL;
927 return ret;
928 }
929
930 drv = bdrv_probe_all(buf, ret, filename);
931 if (!drv) {
932 error_setg(errp, "Could not determine image format: No compatible "
933 "driver found");
934 ret = -ENOENT;
935 }
936 *pdrv = drv;
937 return ret;
938 }
939
940 /**
941 * Set the current 'total_sectors' value
942 * Return 0 on success, -errno on error.
943 */
944 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
945 {
946 BlockDriver *drv = bs->drv;
947
948 if (!drv) {
949 return -ENOMEDIUM;
950 }
951
952 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
953 if (bdrv_is_sg(bs))
954 return 0;
955
956 /* query actual device if possible, otherwise just trust the hint */
957 if (drv->bdrv_getlength) {
958 int64_t length = drv->bdrv_getlength(bs);
959 if (length < 0) {
960 return length;
961 }
962 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
963 }
964
965 bs->total_sectors = hint;
966
967 if (bs->total_sectors * BDRV_SECTOR_SIZE > BDRV_MAX_LENGTH) {
968 return -EFBIG;
969 }
970
971 return 0;
972 }
973
974 /**
975 * Combines a QDict of new block driver @options with any missing options taken
976 * from @old_options, so that leaving out an option defaults to its old value.
977 */
978 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
979 QDict *old_options)
980 {
981 if (bs->drv && bs->drv->bdrv_join_options) {
982 bs->drv->bdrv_join_options(options, old_options);
983 } else {
984 qdict_join(options, old_options, false);
985 }
986 }
987
988 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
989 int open_flags,
990 Error **errp)
991 {
992 Error *local_err = NULL;
993 char *value = qemu_opt_get_del(opts, "detect-zeroes");
994 BlockdevDetectZeroesOptions detect_zeroes =
995 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
996 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
997 g_free(value);
998 if (local_err) {
999 error_propagate(errp, local_err);
1000 return detect_zeroes;
1001 }
1002
1003 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
1004 !(open_flags & BDRV_O_UNMAP))
1005 {
1006 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
1007 "without setting discard operation to unmap");
1008 }
1009
1010 return detect_zeroes;
1011 }
1012
1013 /**
1014 * Set open flags for aio engine
1015 *
1016 * Return 0 on success, -1 if the engine specified is invalid
1017 */
1018 int bdrv_parse_aio(const char *mode, int *flags)
1019 {
1020 if (!strcmp(mode, "threads")) {
1021 /* do nothing, default */
1022 } else if (!strcmp(mode, "native")) {
1023 *flags |= BDRV_O_NATIVE_AIO;
1024 #ifdef CONFIG_LINUX_IO_URING
1025 } else if (!strcmp(mode, "io_uring")) {
1026 *flags |= BDRV_O_IO_URING;
1027 #endif
1028 } else {
1029 return -1;
1030 }
1031
1032 return 0;
1033 }
1034
1035 /**
1036 * Set open flags for a given discard mode
1037 *
1038 * Return 0 on success, -1 if the discard mode was invalid.
1039 */
1040 int bdrv_parse_discard_flags(const char *mode, int *flags)
1041 {
1042 *flags &= ~BDRV_O_UNMAP;
1043
1044 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1045 /* do nothing */
1046 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1047 *flags |= BDRV_O_UNMAP;
1048 } else {
1049 return -1;
1050 }
1051
1052 return 0;
1053 }
1054
1055 /**
1056 * Set open flags for a given cache mode
1057 *
1058 * Return 0 on success, -1 if the cache mode was invalid.
1059 */
1060 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1061 {
1062 *flags &= ~BDRV_O_CACHE_MASK;
1063
1064 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1065 *writethrough = false;
1066 *flags |= BDRV_O_NOCACHE;
1067 } else if (!strcmp(mode, "directsync")) {
1068 *writethrough = true;
1069 *flags |= BDRV_O_NOCACHE;
1070 } else if (!strcmp(mode, "writeback")) {
1071 *writethrough = false;
1072 } else if (!strcmp(mode, "unsafe")) {
1073 *writethrough = false;
1074 *flags |= BDRV_O_NO_FLUSH;
1075 } else if (!strcmp(mode, "writethrough")) {
1076 *writethrough = true;
1077 } else {
1078 return -1;
1079 }
1080
1081 return 0;
1082 }
1083
1084 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1085 {
1086 BlockDriverState *parent = c->opaque;
1087 return g_strdup(bdrv_get_device_or_node_name(parent));
1088 }
1089
1090 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1091 {
1092 BlockDriverState *bs = child->opaque;
1093 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1094 }
1095
1096 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1097 {
1098 BlockDriverState *bs = child->opaque;
1099 return bdrv_drain_poll(bs, false, NULL, false);
1100 }
1101
1102 static void bdrv_child_cb_drained_end(BdrvChild *child,
1103 int *drained_end_counter)
1104 {
1105 BlockDriverState *bs = child->opaque;
1106 bdrv_drained_end_no_poll(bs, drained_end_counter);
1107 }
1108
1109 static int bdrv_child_cb_inactivate(BdrvChild *child)
1110 {
1111 BlockDriverState *bs = child->opaque;
1112 assert(bs->open_flags & BDRV_O_INACTIVE);
1113 return 0;
1114 }
1115
1116 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1117 GSList **ignore, Error **errp)
1118 {
1119 BlockDriverState *bs = child->opaque;
1120 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1121 }
1122
1123 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1124 GSList **ignore)
1125 {
1126 BlockDriverState *bs = child->opaque;
1127 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1128 }
1129
1130 /*
1131 * Returns the options and flags that a temporary snapshot should get, based on
1132 * the originally requested flags (the originally requested image will have
1133 * flags like a backing file)
1134 */
1135 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1136 int parent_flags, QDict *parent_options)
1137 {
1138 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1139
1140 /* For temporary files, unconditional cache=unsafe is fine */
1141 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1142 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1143
1144 /* Copy the read-only and discard options from the parent */
1145 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1146 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1147
1148 /* aio=native doesn't work for cache.direct=off, so disable it for the
1149 * temporary snapshot */
1150 *child_flags &= ~BDRV_O_NATIVE_AIO;
1151 }
1152
1153 static void bdrv_backing_attach(BdrvChild *c)
1154 {
1155 BlockDriverState *parent = c->opaque;
1156 BlockDriverState *backing_hd = c->bs;
1157
1158 assert(!parent->backing_blocker);
1159 error_setg(&parent->backing_blocker,
1160 "node is used as backing hd of '%s'",
1161 bdrv_get_device_or_node_name(parent));
1162
1163 bdrv_refresh_filename(backing_hd);
1164
1165 parent->open_flags &= ~BDRV_O_NO_BACKING;
1166
1167 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1168 /* Otherwise we won't be able to commit or stream */
1169 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1170 parent->backing_blocker);
1171 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1172 parent->backing_blocker);
1173 /*
1174 * We do backup in 3 ways:
1175 * 1. drive backup
1176 * The target bs is new opened, and the source is top BDS
1177 * 2. blockdev backup
1178 * Both the source and the target are top BDSes.
1179 * 3. internal backup(used for block replication)
1180 * Both the source and the target are backing file
1181 *
1182 * In case 1 and 2, neither the source nor the target is the backing file.
1183 * In case 3, we will block the top BDS, so there is only one block job
1184 * for the top BDS and its backing chain.
1185 */
1186 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1187 parent->backing_blocker);
1188 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1189 parent->backing_blocker);
1190 }
1191
1192 static void bdrv_backing_detach(BdrvChild *c)
1193 {
1194 BlockDriverState *parent = c->opaque;
1195
1196 assert(parent->backing_blocker);
1197 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1198 error_free(parent->backing_blocker);
1199 parent->backing_blocker = NULL;
1200 }
1201
1202 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1203 const char *filename, Error **errp)
1204 {
1205 BlockDriverState *parent = c->opaque;
1206 bool read_only = bdrv_is_read_only(parent);
1207 int ret;
1208
1209 if (read_only) {
1210 ret = bdrv_reopen_set_read_only(parent, false, errp);
1211 if (ret < 0) {
1212 return ret;
1213 }
1214 }
1215
1216 ret = bdrv_change_backing_file(parent, filename,
1217 base->drv ? base->drv->format_name : "",
1218 false);
1219 if (ret < 0) {
1220 error_setg_errno(errp, -ret, "Could not update backing file link");
1221 }
1222
1223 if (read_only) {
1224 bdrv_reopen_set_read_only(parent, true, NULL);
1225 }
1226
1227 return ret;
1228 }
1229
1230 /*
1231 * Returns the options and flags that a generic child of a BDS should
1232 * get, based on the given options and flags for the parent BDS.
1233 */
1234 static void bdrv_inherited_options(BdrvChildRole role, bool parent_is_format,
1235 int *child_flags, QDict *child_options,
1236 int parent_flags, QDict *parent_options)
1237 {
1238 int flags = parent_flags;
1239
1240 /*
1241 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1242 * Generally, the question to answer is: Should this child be
1243 * format-probed by default?
1244 */
1245
1246 /*
1247 * Pure and non-filtered data children of non-format nodes should
1248 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1249 * set). This only affects a very limited set of drivers (namely
1250 * quorum and blkverify when this comment was written).
1251 * Force-clear BDRV_O_PROTOCOL then.
1252 */
1253 if (!parent_is_format &&
1254 (role & BDRV_CHILD_DATA) &&
1255 !(role & (BDRV_CHILD_METADATA | BDRV_CHILD_FILTERED)))
1256 {
1257 flags &= ~BDRV_O_PROTOCOL;
1258 }
1259
1260 /*
1261 * All children of format nodes (except for COW children) and all
1262 * metadata children in general should never be format-probed.
1263 * Force-set BDRV_O_PROTOCOL then.
1264 */
1265 if ((parent_is_format && !(role & BDRV_CHILD_COW)) ||
1266 (role & BDRV_CHILD_METADATA))
1267 {
1268 flags |= BDRV_O_PROTOCOL;
1269 }
1270
1271 /*
1272 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1273 * the parent.
1274 */
1275 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1276 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1277 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1278
1279 if (role & BDRV_CHILD_COW) {
1280 /* backing files are opened read-only by default */
1281 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1282 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1283 } else {
1284 /* Inherit the read-only option from the parent if it's not set */
1285 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1286 qdict_copy_default(child_options, parent_options,
1287 BDRV_OPT_AUTO_READ_ONLY);
1288 }
1289
1290 /*
1291 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1292 * can default to enable it on lower layers regardless of the
1293 * parent option.
1294 */
1295 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1296
1297 /* Clear flags that only apply to the top layer */
1298 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
1299
1300 if (role & BDRV_CHILD_METADATA) {
1301 flags &= ~BDRV_O_NO_IO;
1302 }
1303 if (role & BDRV_CHILD_COW) {
1304 flags &= ~BDRV_O_TEMPORARY;
1305 }
1306
1307 *child_flags = flags;
1308 }
1309
1310 static void bdrv_child_cb_attach(BdrvChild *child)
1311 {
1312 BlockDriverState *bs = child->opaque;
1313
1314 if (child->role & BDRV_CHILD_COW) {
1315 bdrv_backing_attach(child);
1316 }
1317
1318 bdrv_apply_subtree_drain(child, bs);
1319 }
1320
1321 static void bdrv_child_cb_detach(BdrvChild *child)
1322 {
1323 BlockDriverState *bs = child->opaque;
1324
1325 if (child->role & BDRV_CHILD_COW) {
1326 bdrv_backing_detach(child);
1327 }
1328
1329 bdrv_unapply_subtree_drain(child, bs);
1330 }
1331
1332 static int bdrv_child_cb_update_filename(BdrvChild *c, BlockDriverState *base,
1333 const char *filename, Error **errp)
1334 {
1335 if (c->role & BDRV_CHILD_COW) {
1336 return bdrv_backing_update_filename(c, base, filename, errp);
1337 }
1338 return 0;
1339 }
1340
1341 const BdrvChildClass child_of_bds = {
1342 .parent_is_bds = true,
1343 .get_parent_desc = bdrv_child_get_parent_desc,
1344 .inherit_options = bdrv_inherited_options,
1345 .drained_begin = bdrv_child_cb_drained_begin,
1346 .drained_poll = bdrv_child_cb_drained_poll,
1347 .drained_end = bdrv_child_cb_drained_end,
1348 .attach = bdrv_child_cb_attach,
1349 .detach = bdrv_child_cb_detach,
1350 .inactivate = bdrv_child_cb_inactivate,
1351 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1352 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1353 .update_filename = bdrv_child_cb_update_filename,
1354 };
1355
1356 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1357 {
1358 int open_flags = flags;
1359
1360 /*
1361 * Clear flags that are internal to the block layer before opening the
1362 * image.
1363 */
1364 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1365
1366 return open_flags;
1367 }
1368
1369 static void update_flags_from_options(int *flags, QemuOpts *opts)
1370 {
1371 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1372
1373 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1374 *flags |= BDRV_O_NO_FLUSH;
1375 }
1376
1377 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1378 *flags |= BDRV_O_NOCACHE;
1379 }
1380
1381 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1382 *flags |= BDRV_O_RDWR;
1383 }
1384
1385 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1386 *flags |= BDRV_O_AUTO_RDONLY;
1387 }
1388 }
1389
1390 static void update_options_from_flags(QDict *options, int flags)
1391 {
1392 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1393 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1394 }
1395 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1396 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1397 flags & BDRV_O_NO_FLUSH);
1398 }
1399 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1400 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1401 }
1402 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1403 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1404 flags & BDRV_O_AUTO_RDONLY);
1405 }
1406 }
1407
1408 static void bdrv_assign_node_name(BlockDriverState *bs,
1409 const char *node_name,
1410 Error **errp)
1411 {
1412 char *gen_node_name = NULL;
1413
1414 if (!node_name) {
1415 node_name = gen_node_name = id_generate(ID_BLOCK);
1416 } else if (!id_wellformed(node_name)) {
1417 /*
1418 * Check for empty string or invalid characters, but not if it is
1419 * generated (generated names use characters not available to the user)
1420 */
1421 error_setg(errp, "Invalid node name");
1422 return;
1423 }
1424
1425 /* takes care of avoiding namespaces collisions */
1426 if (blk_by_name(node_name)) {
1427 error_setg(errp, "node-name=%s is conflicting with a device id",
1428 node_name);
1429 goto out;
1430 }
1431
1432 /* takes care of avoiding duplicates node names */
1433 if (bdrv_find_node(node_name)) {
1434 error_setg(errp, "Duplicate node name");
1435 goto out;
1436 }
1437
1438 /* Make sure that the node name isn't truncated */
1439 if (strlen(node_name) >= sizeof(bs->node_name)) {
1440 error_setg(errp, "Node name too long");
1441 goto out;
1442 }
1443
1444 /* copy node name into the bs and insert it into the graph list */
1445 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1446 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1447 out:
1448 g_free(gen_node_name);
1449 }
1450
1451 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1452 const char *node_name, QDict *options,
1453 int open_flags, Error **errp)
1454 {
1455 Error *local_err = NULL;
1456 int i, ret;
1457
1458 bdrv_assign_node_name(bs, node_name, &local_err);
1459 if (local_err) {
1460 error_propagate(errp, local_err);
1461 return -EINVAL;
1462 }
1463
1464 bs->drv = drv;
1465 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1466 bs->opaque = g_malloc0(drv->instance_size);
1467
1468 if (drv->bdrv_file_open) {
1469 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1470 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1471 } else if (drv->bdrv_open) {
1472 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1473 } else {
1474 ret = 0;
1475 }
1476
1477 if (ret < 0) {
1478 if (local_err) {
1479 error_propagate(errp, local_err);
1480 } else if (bs->filename[0]) {
1481 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1482 } else {
1483 error_setg_errno(errp, -ret, "Could not open image");
1484 }
1485 goto open_failed;
1486 }
1487
1488 ret = refresh_total_sectors(bs, bs->total_sectors);
1489 if (ret < 0) {
1490 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1491 return ret;
1492 }
1493
1494 bdrv_refresh_limits(bs, &local_err);
1495 if (local_err) {
1496 error_propagate(errp, local_err);
1497 return -EINVAL;
1498 }
1499
1500 assert(bdrv_opt_mem_align(bs) != 0);
1501 assert(bdrv_min_mem_align(bs) != 0);
1502 assert(is_power_of_2(bs->bl.request_alignment));
1503
1504 for (i = 0; i < bs->quiesce_counter; i++) {
1505 if (drv->bdrv_co_drain_begin) {
1506 drv->bdrv_co_drain_begin(bs);
1507 }
1508 }
1509
1510 return 0;
1511 open_failed:
1512 bs->drv = NULL;
1513 if (bs->file != NULL) {
1514 bdrv_unref_child(bs, bs->file);
1515 bs->file = NULL;
1516 }
1517 g_free(bs->opaque);
1518 bs->opaque = NULL;
1519 return ret;
1520 }
1521
1522 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1523 int flags, Error **errp)
1524 {
1525 BlockDriverState *bs;
1526 int ret;
1527
1528 bs = bdrv_new();
1529 bs->open_flags = flags;
1530 bs->explicit_options = qdict_new();
1531 bs->options = qdict_new();
1532 bs->opaque = NULL;
1533
1534 update_options_from_flags(bs->options, flags);
1535
1536 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1537 if (ret < 0) {
1538 qobject_unref(bs->explicit_options);
1539 bs->explicit_options = NULL;
1540 qobject_unref(bs->options);
1541 bs->options = NULL;
1542 bdrv_unref(bs);
1543 return NULL;
1544 }
1545
1546 return bs;
1547 }
1548
1549 QemuOptsList bdrv_runtime_opts = {
1550 .name = "bdrv_common",
1551 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1552 .desc = {
1553 {
1554 .name = "node-name",
1555 .type = QEMU_OPT_STRING,
1556 .help = "Node name of the block device node",
1557 },
1558 {
1559 .name = "driver",
1560 .type = QEMU_OPT_STRING,
1561 .help = "Block driver to use for the node",
1562 },
1563 {
1564 .name = BDRV_OPT_CACHE_DIRECT,
1565 .type = QEMU_OPT_BOOL,
1566 .help = "Bypass software writeback cache on the host",
1567 },
1568 {
1569 .name = BDRV_OPT_CACHE_NO_FLUSH,
1570 .type = QEMU_OPT_BOOL,
1571 .help = "Ignore flush requests",
1572 },
1573 {
1574 .name = BDRV_OPT_READ_ONLY,
1575 .type = QEMU_OPT_BOOL,
1576 .help = "Node is opened in read-only mode",
1577 },
1578 {
1579 .name = BDRV_OPT_AUTO_READ_ONLY,
1580 .type = QEMU_OPT_BOOL,
1581 .help = "Node can become read-only if opening read-write fails",
1582 },
1583 {
1584 .name = "detect-zeroes",
1585 .type = QEMU_OPT_STRING,
1586 .help = "try to optimize zero writes (off, on, unmap)",
1587 },
1588 {
1589 .name = BDRV_OPT_DISCARD,
1590 .type = QEMU_OPT_STRING,
1591 .help = "discard operation (ignore/off, unmap/on)",
1592 },
1593 {
1594 .name = BDRV_OPT_FORCE_SHARE,
1595 .type = QEMU_OPT_BOOL,
1596 .help = "always accept other writers (default: off)",
1597 },
1598 { /* end of list */ }
1599 },
1600 };
1601
1602 QemuOptsList bdrv_create_opts_simple = {
1603 .name = "simple-create-opts",
1604 .head = QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple.head),
1605 .desc = {
1606 {
1607 .name = BLOCK_OPT_SIZE,
1608 .type = QEMU_OPT_SIZE,
1609 .help = "Virtual disk size"
1610 },
1611 {
1612 .name = BLOCK_OPT_PREALLOC,
1613 .type = QEMU_OPT_STRING,
1614 .help = "Preallocation mode (allowed values: off)"
1615 },
1616 { /* end of list */ }
1617 }
1618 };
1619
1620 /*
1621 * Common part for opening disk images and files
1622 *
1623 * Removes all processed options from *options.
1624 */
1625 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1626 QDict *options, Error **errp)
1627 {
1628 int ret, open_flags;
1629 const char *filename;
1630 const char *driver_name = NULL;
1631 const char *node_name = NULL;
1632 const char *discard;
1633 QemuOpts *opts;
1634 BlockDriver *drv;
1635 Error *local_err = NULL;
1636
1637 assert(bs->file == NULL);
1638 assert(options != NULL && bs->options != options);
1639
1640 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1641 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
1642 ret = -EINVAL;
1643 goto fail_opts;
1644 }
1645
1646 update_flags_from_options(&bs->open_flags, opts);
1647
1648 driver_name = qemu_opt_get(opts, "driver");
1649 drv = bdrv_find_format(driver_name);
1650 assert(drv != NULL);
1651
1652 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1653
1654 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1655 error_setg(errp,
1656 BDRV_OPT_FORCE_SHARE
1657 "=on can only be used with read-only images");
1658 ret = -EINVAL;
1659 goto fail_opts;
1660 }
1661
1662 if (file != NULL) {
1663 bdrv_refresh_filename(blk_bs(file));
1664 filename = blk_bs(file)->filename;
1665 } else {
1666 /*
1667 * Caution: while qdict_get_try_str() is fine, getting
1668 * non-string types would require more care. When @options
1669 * come from -blockdev or blockdev_add, its members are typed
1670 * according to the QAPI schema, but when they come from
1671 * -drive, they're all QString.
1672 */
1673 filename = qdict_get_try_str(options, "filename");
1674 }
1675
1676 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1677 error_setg(errp, "The '%s' block driver requires a file name",
1678 drv->format_name);
1679 ret = -EINVAL;
1680 goto fail_opts;
1681 }
1682
1683 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1684 drv->format_name);
1685
1686 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1687
1688 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1689 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1690 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1691 } else {
1692 ret = -ENOTSUP;
1693 }
1694 if (ret < 0) {
1695 error_setg(errp,
1696 !bs->read_only && bdrv_is_whitelisted(drv, true)
1697 ? "Driver '%s' can only be used for read-only devices"
1698 : "Driver '%s' is not whitelisted",
1699 drv->format_name);
1700 goto fail_opts;
1701 }
1702 }
1703
1704 /* bdrv_new() and bdrv_close() make it so */
1705 assert(qatomic_read(&bs->copy_on_read) == 0);
1706
1707 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1708 if (!bs->read_only) {
1709 bdrv_enable_copy_on_read(bs);
1710 } else {
1711 error_setg(errp, "Can't use copy-on-read on read-only device");
1712 ret = -EINVAL;
1713 goto fail_opts;
1714 }
1715 }
1716
1717 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1718 if (discard != NULL) {
1719 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1720 error_setg(errp, "Invalid discard option");
1721 ret = -EINVAL;
1722 goto fail_opts;
1723 }
1724 }
1725
1726 bs->detect_zeroes =
1727 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1728 if (local_err) {
1729 error_propagate(errp, local_err);
1730 ret = -EINVAL;
1731 goto fail_opts;
1732 }
1733
1734 if (filename != NULL) {
1735 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1736 } else {
1737 bs->filename[0] = '\0';
1738 }
1739 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1740
1741 /* Open the image, either directly or using a protocol */
1742 open_flags = bdrv_open_flags(bs, bs->open_flags);
1743 node_name = qemu_opt_get(opts, "node-name");
1744
1745 assert(!drv->bdrv_file_open || file == NULL);
1746 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1747 if (ret < 0) {
1748 goto fail_opts;
1749 }
1750
1751 qemu_opts_del(opts);
1752 return 0;
1753
1754 fail_opts:
1755 qemu_opts_del(opts);
1756 return ret;
1757 }
1758
1759 static QDict *parse_json_filename(const char *filename, Error **errp)
1760 {
1761 QObject *options_obj;
1762 QDict *options;
1763 int ret;
1764
1765 ret = strstart(filename, "json:", &filename);
1766 assert(ret);
1767
1768 options_obj = qobject_from_json(filename, errp);
1769 if (!options_obj) {
1770 error_prepend(errp, "Could not parse the JSON options: ");
1771 return NULL;
1772 }
1773
1774 options = qobject_to(QDict, options_obj);
1775 if (!options) {
1776 qobject_unref(options_obj);
1777 error_setg(errp, "Invalid JSON object given");
1778 return NULL;
1779 }
1780
1781 qdict_flatten(options);
1782
1783 return options;
1784 }
1785
1786 static void parse_json_protocol(QDict *options, const char **pfilename,
1787 Error **errp)
1788 {
1789 QDict *json_options;
1790 Error *local_err = NULL;
1791
1792 /* Parse json: pseudo-protocol */
1793 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1794 return;
1795 }
1796
1797 json_options = parse_json_filename(*pfilename, &local_err);
1798 if (local_err) {
1799 error_propagate(errp, local_err);
1800 return;
1801 }
1802
1803 /* Options given in the filename have lower priority than options
1804 * specified directly */
1805 qdict_join(options, json_options, false);
1806 qobject_unref(json_options);
1807 *pfilename = NULL;
1808 }
1809
1810 /*
1811 * Fills in default options for opening images and converts the legacy
1812 * filename/flags pair to option QDict entries.
1813 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1814 * block driver has been specified explicitly.
1815 */
1816 static int bdrv_fill_options(QDict **options, const char *filename,
1817 int *flags, Error **errp)
1818 {
1819 const char *drvname;
1820 bool protocol = *flags & BDRV_O_PROTOCOL;
1821 bool parse_filename = false;
1822 BlockDriver *drv = NULL;
1823 Error *local_err = NULL;
1824
1825 /*
1826 * Caution: while qdict_get_try_str() is fine, getting non-string
1827 * types would require more care. When @options come from
1828 * -blockdev or blockdev_add, its members are typed according to
1829 * the QAPI schema, but when they come from -drive, they're all
1830 * QString.
1831 */
1832 drvname = qdict_get_try_str(*options, "driver");
1833 if (drvname) {
1834 drv = bdrv_find_format(drvname);
1835 if (!drv) {
1836 error_setg(errp, "Unknown driver '%s'", drvname);
1837 return -ENOENT;
1838 }
1839 /* If the user has explicitly specified the driver, this choice should
1840 * override the BDRV_O_PROTOCOL flag */
1841 protocol = drv->bdrv_file_open;
1842 }
1843
1844 if (protocol) {
1845 *flags |= BDRV_O_PROTOCOL;
1846 } else {
1847 *flags &= ~BDRV_O_PROTOCOL;
1848 }
1849
1850 /* Translate cache options from flags into options */
1851 update_options_from_flags(*options, *flags);
1852
1853 /* Fetch the file name from the options QDict if necessary */
1854 if (protocol && filename) {
1855 if (!qdict_haskey(*options, "filename")) {
1856 qdict_put_str(*options, "filename", filename);
1857 parse_filename = true;
1858 } else {
1859 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1860 "the same time");
1861 return -EINVAL;
1862 }
1863 }
1864
1865 /* Find the right block driver */
1866 /* See cautionary note on accessing @options above */
1867 filename = qdict_get_try_str(*options, "filename");
1868
1869 if (!drvname && protocol) {
1870 if (filename) {
1871 drv = bdrv_find_protocol(filename, parse_filename, errp);
1872 if (!drv) {
1873 return -EINVAL;
1874 }
1875
1876 drvname = drv->format_name;
1877 qdict_put_str(*options, "driver", drvname);
1878 } else {
1879 error_setg(errp, "Must specify either driver or file");
1880 return -EINVAL;
1881 }
1882 }
1883
1884 assert(drv || !protocol);
1885
1886 /* Driver-specific filename parsing */
1887 if (drv && drv->bdrv_parse_filename && parse_filename) {
1888 drv->bdrv_parse_filename(filename, *options, &local_err);
1889 if (local_err) {
1890 error_propagate(errp, local_err);
1891 return -EINVAL;
1892 }
1893
1894 if (!drv->bdrv_needs_filename) {
1895 qdict_del(*options, "filename");
1896 }
1897 }
1898
1899 return 0;
1900 }
1901
1902 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1903 uint64_t perm, uint64_t shared,
1904 GSList *ignore_children, Error **errp);
1905 static void bdrv_child_abort_perm_update(BdrvChild *c);
1906 static void bdrv_child_set_perm(BdrvChild *c);
1907
1908 typedef struct BlockReopenQueueEntry {
1909 bool prepared;
1910 bool perms_checked;
1911 BDRVReopenState state;
1912 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
1913 } BlockReopenQueueEntry;
1914
1915 /*
1916 * Return the flags that @bs will have after the reopens in @q have
1917 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1918 * return the current flags.
1919 */
1920 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1921 {
1922 BlockReopenQueueEntry *entry;
1923
1924 if (q != NULL) {
1925 QTAILQ_FOREACH(entry, q, entry) {
1926 if (entry->state.bs == bs) {
1927 return entry->state.flags;
1928 }
1929 }
1930 }
1931
1932 return bs->open_flags;
1933 }
1934
1935 /* Returns whether the image file can be written to after the reopen queue @q
1936 * has been successfully applied, or right now if @q is NULL. */
1937 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1938 BlockReopenQueue *q)
1939 {
1940 int flags = bdrv_reopen_get_flags(q, bs);
1941
1942 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1943 }
1944
1945 /*
1946 * Return whether the BDS can be written to. This is not necessarily
1947 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1948 * be written to but do not count as read-only images.
1949 */
1950 bool bdrv_is_writable(BlockDriverState *bs)
1951 {
1952 return bdrv_is_writable_after_reopen(bs, NULL);
1953 }
1954
1955 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1956 BdrvChild *c, BdrvChildRole role,
1957 BlockReopenQueue *reopen_queue,
1958 uint64_t parent_perm, uint64_t parent_shared,
1959 uint64_t *nperm, uint64_t *nshared)
1960 {
1961 assert(bs->drv && bs->drv->bdrv_child_perm);
1962 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1963 parent_perm, parent_shared,
1964 nperm, nshared);
1965 /* TODO Take force_share from reopen_queue */
1966 if (child_bs && child_bs->force_share) {
1967 *nshared = BLK_PERM_ALL;
1968 }
1969 }
1970
1971 /*
1972 * Check whether permissions on this node can be changed in a way that
1973 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1974 * permissions of all its parents. This involves checking whether all necessary
1975 * permission changes to child nodes can be performed.
1976 *
1977 * A call to this function must always be followed by a call to bdrv_set_perm()
1978 * or bdrv_abort_perm_update().
1979 */
1980 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1981 uint64_t cumulative_perms,
1982 uint64_t cumulative_shared_perms,
1983 GSList *ignore_children, Error **errp)
1984 {
1985 BlockDriver *drv = bs->drv;
1986 BdrvChild *c;
1987 int ret;
1988
1989 /* Write permissions never work with read-only images */
1990 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1991 !bdrv_is_writable_after_reopen(bs, q))
1992 {
1993 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
1994 error_setg(errp, "Block node is read-only");
1995 } else {
1996 uint64_t current_perms, current_shared;
1997 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
1998 if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
1999 error_setg(errp, "Cannot make block node read-only, there is "
2000 "a writer on it");
2001 } else {
2002 error_setg(errp, "Cannot make block node read-only and create "
2003 "a writer on it");
2004 }
2005 }
2006
2007 return -EPERM;
2008 }
2009
2010 /*
2011 * Unaligned requests will automatically be aligned to bl.request_alignment
2012 * and without RESIZE we can't extend requests to write to space beyond the
2013 * end of the image, so it's required that the image size is aligned.
2014 */
2015 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
2016 !(cumulative_perms & BLK_PERM_RESIZE))
2017 {
2018 if ((bs->total_sectors * BDRV_SECTOR_SIZE) % bs->bl.request_alignment) {
2019 error_setg(errp, "Cannot get 'write' permission without 'resize': "
2020 "Image size is not a multiple of request "
2021 "alignment");
2022 return -EPERM;
2023 }
2024 }
2025
2026 /* Check this node */
2027 if (!drv) {
2028 return 0;
2029 }
2030
2031 if (drv->bdrv_check_perm) {
2032 ret = drv->bdrv_check_perm(bs, cumulative_perms,
2033 cumulative_shared_perms, errp);
2034 if (ret < 0) {
2035 return ret;
2036 }
2037 }
2038
2039 /* Drivers that never have children can omit .bdrv_child_perm() */
2040 if (!drv->bdrv_child_perm) {
2041 assert(QLIST_EMPTY(&bs->children));
2042 return 0;
2043 }
2044
2045 /* Check all children */
2046 QLIST_FOREACH(c, &bs->children, next) {
2047 uint64_t cur_perm, cur_shared;
2048
2049 bdrv_child_perm(bs, c->bs, c, c->role, q,
2050 cumulative_perms, cumulative_shared_perms,
2051 &cur_perm, &cur_shared);
2052 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children,
2053 errp);
2054 if (ret < 0) {
2055 return ret;
2056 }
2057 }
2058
2059 return 0;
2060 }
2061
2062 /*
2063 * Notifies drivers that after a previous bdrv_check_perm() call, the
2064 * permission update is not performed and any preparations made for it (e.g.
2065 * taken file locks) need to be undone.
2066 *
2067 * This function recursively notifies all child nodes.
2068 */
2069 static void bdrv_abort_perm_update(BlockDriverState *bs)
2070 {
2071 BlockDriver *drv = bs->drv;
2072 BdrvChild *c;
2073
2074 if (!drv) {
2075 return;
2076 }
2077
2078 if (drv->bdrv_abort_perm_update) {
2079 drv->bdrv_abort_perm_update(bs);
2080 }
2081
2082 QLIST_FOREACH(c, &bs->children, next) {
2083 bdrv_child_abort_perm_update(c);
2084 }
2085 }
2086
2087 static void bdrv_set_perm(BlockDriverState *bs)
2088 {
2089 uint64_t cumulative_perms, cumulative_shared_perms;
2090 BlockDriver *drv = bs->drv;
2091 BdrvChild *c;
2092
2093 if (!drv) {
2094 return;
2095 }
2096
2097 bdrv_get_cumulative_perm(bs, &cumulative_perms, &cumulative_shared_perms);
2098
2099 /* Update this node */
2100 if (drv->bdrv_set_perm) {
2101 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2102 }
2103
2104 /* Drivers that never have children can omit .bdrv_child_perm() */
2105 if (!drv->bdrv_child_perm) {
2106 assert(QLIST_EMPTY(&bs->children));
2107 return;
2108 }
2109
2110 /* Update all children */
2111 QLIST_FOREACH(c, &bs->children, next) {
2112 bdrv_child_set_perm(c);
2113 }
2114 }
2115
2116 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2117 uint64_t *shared_perm)
2118 {
2119 BdrvChild *c;
2120 uint64_t cumulative_perms = 0;
2121 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2122
2123 QLIST_FOREACH(c, &bs->parents, next_parent) {
2124 cumulative_perms |= c->perm;
2125 cumulative_shared_perms &= c->shared_perm;
2126 }
2127
2128 *perm = cumulative_perms;
2129 *shared_perm = cumulative_shared_perms;
2130 }
2131
2132 static char *bdrv_child_user_desc(BdrvChild *c)
2133 {
2134 if (c->klass->get_parent_desc) {
2135 return c->klass->get_parent_desc(c);
2136 }
2137
2138 return g_strdup("another user");
2139 }
2140
2141 char *bdrv_perm_names(uint64_t perm)
2142 {
2143 struct perm_name {
2144 uint64_t perm;
2145 const char *name;
2146 } permissions[] = {
2147 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2148 { BLK_PERM_WRITE, "write" },
2149 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2150 { BLK_PERM_RESIZE, "resize" },
2151 { BLK_PERM_GRAPH_MOD, "change children" },
2152 { 0, NULL }
2153 };
2154
2155 GString *result = g_string_sized_new(30);
2156 struct perm_name *p;
2157
2158 for (p = permissions; p->name; p++) {
2159 if (perm & p->perm) {
2160 if (result->len > 0) {
2161 g_string_append(result, ", ");
2162 }
2163 g_string_append(result, p->name);
2164 }
2165 }
2166
2167 return g_string_free(result, FALSE);
2168 }
2169
2170 /*
2171 * Checks whether a new reference to @bs can be added if the new user requires
2172 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2173 * set, the BdrvChild objects in this list are ignored in the calculations;
2174 * this allows checking permission updates for an existing reference.
2175 *
2176 * Needs to be followed by a call to either bdrv_set_perm() or
2177 * bdrv_abort_perm_update(). */
2178 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
2179 uint64_t new_used_perm,
2180 uint64_t new_shared_perm,
2181 GSList *ignore_children,
2182 Error **errp)
2183 {
2184 BdrvChild *c;
2185 uint64_t cumulative_perms = new_used_perm;
2186 uint64_t cumulative_shared_perms = new_shared_perm;
2187
2188
2189 /* There is no reason why anyone couldn't tolerate write_unchanged */
2190 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
2191
2192 QLIST_FOREACH(c, &bs->parents, next_parent) {
2193 if (g_slist_find(ignore_children, c)) {
2194 continue;
2195 }
2196
2197 if ((new_used_perm & c->shared_perm) != new_used_perm) {
2198 char *user = bdrv_child_user_desc(c);
2199 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
2200
2201 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
2202 "allow '%s' on %s",
2203 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2204 g_free(user);
2205 g_free(perm_names);
2206 return -EPERM;
2207 }
2208
2209 if ((c->perm & new_shared_perm) != c->perm) {
2210 char *user = bdrv_child_user_desc(c);
2211 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
2212
2213 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
2214 "'%s' on %s",
2215 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2216 g_free(user);
2217 g_free(perm_names);
2218 return -EPERM;
2219 }
2220
2221 cumulative_perms |= c->perm;
2222 cumulative_shared_perms &= c->shared_perm;
2223 }
2224
2225 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
2226 ignore_children, errp);
2227 }
2228
2229 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2230 * bdrv_child_abort_perm_update(). */
2231 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
2232 uint64_t perm, uint64_t shared,
2233 GSList *ignore_children, Error **errp)
2234 {
2235 int ret;
2236
2237 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
2238 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children, errp);
2239 g_slist_free(ignore_children);
2240
2241 if (ret < 0) {
2242 return ret;
2243 }
2244
2245 if (!c->has_backup_perm) {
2246 c->has_backup_perm = true;
2247 c->backup_perm = c->perm;
2248 c->backup_shared_perm = c->shared_perm;
2249 }
2250 /*
2251 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2252 * same child twice during check_perm procedure
2253 */
2254
2255 c->perm = perm;
2256 c->shared_perm = shared;
2257
2258 return 0;
2259 }
2260
2261 static void bdrv_child_set_perm(BdrvChild *c)
2262 {
2263 c->has_backup_perm = false;
2264
2265 bdrv_set_perm(c->bs);
2266 }
2267
2268 static void bdrv_child_abort_perm_update(BdrvChild *c)
2269 {
2270 if (c->has_backup_perm) {
2271 c->perm = c->backup_perm;
2272 c->shared_perm = c->backup_shared_perm;
2273 c->has_backup_perm = false;
2274 }
2275
2276 bdrv_abort_perm_update(c->bs);
2277 }
2278
2279 static int bdrv_refresh_perms(BlockDriverState *bs, Error **errp)
2280 {
2281 int ret;
2282 uint64_t perm, shared_perm;
2283
2284 bdrv_get_cumulative_perm(bs, &perm, &shared_perm);
2285 ret = bdrv_check_perm(bs, NULL, perm, shared_perm, NULL, errp);
2286 if (ret < 0) {
2287 bdrv_abort_perm_update(bs);
2288 return ret;
2289 }
2290 bdrv_set_perm(bs);
2291
2292 return 0;
2293 }
2294
2295 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2296 Error **errp)
2297 {
2298 Error *local_err = NULL;
2299 int ret;
2300
2301 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL, &local_err);
2302 if (ret < 0) {
2303 bdrv_child_abort_perm_update(c);
2304 if ((perm & ~c->perm) || (c->shared_perm & ~shared)) {
2305 /* tighten permissions */
2306 error_propagate(errp, local_err);
2307 } else {
2308 /*
2309 * Our caller may intend to only loosen restrictions and
2310 * does not expect this function to fail. Errors are not
2311 * fatal in such a case, so we can just hide them from our
2312 * caller.
2313 */
2314 error_free(local_err);
2315 ret = 0;
2316 }
2317 return ret;
2318 }
2319
2320 bdrv_child_set_perm(c);
2321
2322 return 0;
2323 }
2324
2325 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2326 {
2327 uint64_t parent_perms, parent_shared;
2328 uint64_t perms, shared;
2329
2330 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2331 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2332 parent_perms, parent_shared, &perms, &shared);
2333
2334 return bdrv_child_try_set_perm(c, perms, shared, errp);
2335 }
2336
2337 /*
2338 * Default implementation for .bdrv_child_perm() for block filters:
2339 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2340 * filtered child.
2341 */
2342 static void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2343 BdrvChildRole role,
2344 BlockReopenQueue *reopen_queue,
2345 uint64_t perm, uint64_t shared,
2346 uint64_t *nperm, uint64_t *nshared)
2347 {
2348 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2349 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2350 }
2351
2352 static void bdrv_default_perms_for_cow(BlockDriverState *bs, BdrvChild *c,
2353 BdrvChildRole role,
2354 BlockReopenQueue *reopen_queue,
2355 uint64_t perm, uint64_t shared,
2356 uint64_t *nperm, uint64_t *nshared)
2357 {
2358 assert(role & BDRV_CHILD_COW);
2359
2360 /*
2361 * We want consistent read from backing files if the parent needs it.
2362 * No other operations are performed on backing files.
2363 */
2364 perm &= BLK_PERM_CONSISTENT_READ;
2365
2366 /*
2367 * If the parent can deal with changing data, we're okay with a
2368 * writable and resizable backing file.
2369 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2370 */
2371 if (shared & BLK_PERM_WRITE) {
2372 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2373 } else {
2374 shared = 0;
2375 }
2376
2377 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2378 BLK_PERM_WRITE_UNCHANGED;
2379
2380 if (bs->open_flags & BDRV_O_INACTIVE) {
2381 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2382 }
2383
2384 *nperm = perm;
2385 *nshared = shared;
2386 }
2387
2388 static void bdrv_default_perms_for_storage(BlockDriverState *bs, BdrvChild *c,
2389 BdrvChildRole role,
2390 BlockReopenQueue *reopen_queue,
2391 uint64_t perm, uint64_t shared,
2392 uint64_t *nperm, uint64_t *nshared)
2393 {
2394 int flags;
2395
2396 assert(role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA));
2397
2398 flags = bdrv_reopen_get_flags(reopen_queue, bs);
2399
2400 /*
2401 * Apart from the modifications below, the same permissions are
2402 * forwarded and left alone as for filters
2403 */
2404 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2405 perm, shared, &perm, &shared);
2406
2407 if (role & BDRV_CHILD_METADATA) {
2408 /* Format drivers may touch metadata even if the guest doesn't write */
2409 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2410 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2411 }
2412
2413 /*
2414 * bs->file always needs to be consistent because of the
2415 * metadata. We can never allow other users to resize or write
2416 * to it.
2417 */
2418 if (!(flags & BDRV_O_NO_IO)) {
2419 perm |= BLK_PERM_CONSISTENT_READ;
2420 }
2421 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2422 }
2423
2424 if (role & BDRV_CHILD_DATA) {
2425 /*
2426 * Technically, everything in this block is a subset of the
2427 * BDRV_CHILD_METADATA path taken above, and so this could
2428 * be an "else if" branch. However, that is not obvious, and
2429 * this function is not performance critical, therefore we let
2430 * this be an independent "if".
2431 */
2432
2433 /*
2434 * We cannot allow other users to resize the file because the
2435 * format driver might have some assumptions about the size
2436 * (e.g. because it is stored in metadata, or because the file
2437 * is split into fixed-size data files).
2438 */
2439 shared &= ~BLK_PERM_RESIZE;
2440
2441 /*
2442 * WRITE_UNCHANGED often cannot be performed as such on the
2443 * data file. For example, the qcow2 driver may still need to
2444 * write copied clusters on copy-on-read.
2445 */
2446 if (perm & BLK_PERM_WRITE_UNCHANGED) {
2447 perm |= BLK_PERM_WRITE;
2448 }
2449
2450 /*
2451 * If the data file is written to, the format driver may
2452 * expect to be able to resize it by writing beyond the EOF.
2453 */
2454 if (perm & BLK_PERM_WRITE) {
2455 perm |= BLK_PERM_RESIZE;
2456 }
2457 }
2458
2459 if (bs->open_flags & BDRV_O_INACTIVE) {
2460 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2461 }
2462
2463 *nperm = perm;
2464 *nshared = shared;
2465 }
2466
2467 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
2468 BdrvChildRole role, BlockReopenQueue *reopen_queue,
2469 uint64_t perm, uint64_t shared,
2470 uint64_t *nperm, uint64_t *nshared)
2471 {
2472 if (role & BDRV_CHILD_FILTERED) {
2473 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA |
2474 BDRV_CHILD_COW)));
2475 bdrv_filter_default_perms(bs, c, role, reopen_queue,
2476 perm, shared, nperm, nshared);
2477 } else if (role & BDRV_CHILD_COW) {
2478 assert(!(role & (BDRV_CHILD_DATA | BDRV_CHILD_METADATA)));
2479 bdrv_default_perms_for_cow(bs, c, role, reopen_queue,
2480 perm, shared, nperm, nshared);
2481 } else if (role & (BDRV_CHILD_METADATA | BDRV_CHILD_DATA)) {
2482 bdrv_default_perms_for_storage(bs, c, role, reopen_queue,
2483 perm, shared, nperm, nshared);
2484 } else {
2485 g_assert_not_reached();
2486 }
2487 }
2488
2489 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2490 {
2491 static const uint64_t permissions[] = {
2492 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2493 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2494 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2495 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2496 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2497 };
2498
2499 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2500 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2501
2502 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2503
2504 return permissions[qapi_perm];
2505 }
2506
2507 static void bdrv_replace_child_noperm(BdrvChild *child,
2508 BlockDriverState *new_bs)
2509 {
2510 BlockDriverState *old_bs = child->bs;
2511 int new_bs_quiesce_counter;
2512 int drain_saldo;
2513
2514 assert(!child->frozen);
2515
2516 if (old_bs && new_bs) {
2517 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2518 }
2519
2520 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2521 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2522
2523 /*
2524 * If the new child node is drained but the old one was not, flush
2525 * all outstanding requests to the old child node.
2526 */
2527 while (drain_saldo > 0 && child->klass->drained_begin) {
2528 bdrv_parent_drained_begin_single(child, true);
2529 drain_saldo--;
2530 }
2531
2532 if (old_bs) {
2533 /* Detach first so that the recursive drain sections coming from @child
2534 * are already gone and we only end the drain sections that came from
2535 * elsewhere. */
2536 if (child->klass->detach) {
2537 child->klass->detach(child);
2538 }
2539 QLIST_REMOVE(child, next_parent);
2540 }
2541
2542 child->bs = new_bs;
2543
2544 if (new_bs) {
2545 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2546
2547 /*
2548 * Detaching the old node may have led to the new node's
2549 * quiesce_counter having been decreased. Not a problem, we
2550 * just need to recognize this here and then invoke
2551 * drained_end appropriately more often.
2552 */
2553 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2554 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2555
2556 /* Attach only after starting new drained sections, so that recursive
2557 * drain sections coming from @child don't get an extra .drained_begin
2558 * callback. */
2559 if (child->klass->attach) {
2560 child->klass->attach(child);
2561 }
2562 }
2563
2564 /*
2565 * If the old child node was drained but the new one is not, allow
2566 * requests to come in only after the new node has been attached.
2567 */
2568 while (drain_saldo < 0 && child->klass->drained_end) {
2569 bdrv_parent_drained_end_single(child);
2570 drain_saldo++;
2571 }
2572 }
2573
2574 /*
2575 * Updates @child to change its reference to point to @new_bs, including
2576 * checking and applying the necessary permission updates both to the old node
2577 * and to @new_bs.
2578 *
2579 * NULL is passed as @new_bs for removing the reference before freeing @child.
2580 *
2581 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2582 * function uses bdrv_set_perm() to update the permissions according to the new
2583 * reference that @new_bs gets.
2584 *
2585 * Callers must ensure that child->frozen is false.
2586 */
2587 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2588 {
2589 BlockDriverState *old_bs = child->bs;
2590
2591 /* Asserts that child->frozen == false */
2592 bdrv_replace_child_noperm(child, new_bs);
2593
2594 /*
2595 * Start with the new node's permissions. If @new_bs is a (direct
2596 * or indirect) child of @old_bs, we must complete the permission
2597 * update on @new_bs before we loosen the restrictions on @old_bs.
2598 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2599 * updating the permissions of @new_bs, and thus not purely loosen
2600 * restrictions.
2601 */
2602 if (new_bs) {
2603 bdrv_set_perm(new_bs);
2604 }
2605
2606 if (old_bs) {
2607 /*
2608 * Update permissions for old node. We're just taking a parent away, so
2609 * we're loosening restrictions. Errors of permission update are not
2610 * fatal in this case, ignore them.
2611 */
2612 bdrv_refresh_perms(old_bs, NULL);
2613
2614 /* When the parent requiring a non-default AioContext is removed, the
2615 * node moves back to the main AioContext */
2616 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2617 }
2618 }
2619
2620 /*
2621 * This function steals the reference to child_bs from the caller.
2622 * That reference is later dropped by bdrv_root_unref_child().
2623 *
2624 * On failure NULL is returned, errp is set and the reference to
2625 * child_bs is also dropped.
2626 *
2627 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2628 * (unless @child_bs is already in @ctx).
2629 */
2630 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2631 const char *child_name,
2632 const BdrvChildClass *child_class,
2633 BdrvChildRole child_role,
2634 AioContext *ctx,
2635 uint64_t perm, uint64_t shared_perm,
2636 void *opaque, Error **errp)
2637 {
2638 BdrvChild *child;
2639 Error *local_err = NULL;
2640 int ret;
2641
2642 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, errp);
2643 if (ret < 0) {
2644 bdrv_abort_perm_update(child_bs);
2645 bdrv_unref(child_bs);
2646 return NULL;
2647 }
2648
2649 child = g_new(BdrvChild, 1);
2650 *child = (BdrvChild) {
2651 .bs = NULL,
2652 .name = g_strdup(child_name),
2653 .klass = child_class,
2654 .role = child_role,
2655 .perm = perm,
2656 .shared_perm = shared_perm,
2657 .opaque = opaque,
2658 };
2659
2660 /* If the AioContexts don't match, first try to move the subtree of
2661 * child_bs into the AioContext of the new parent. If this doesn't work,
2662 * try moving the parent into the AioContext of child_bs instead. */
2663 if (bdrv_get_aio_context(child_bs) != ctx) {
2664 ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err);
2665 if (ret < 0 && child_class->can_set_aio_ctx) {
2666 GSList *ignore = g_slist_prepend(NULL, child);
2667 ctx = bdrv_get_aio_context(child_bs);
2668 if (child_class->can_set_aio_ctx(child, ctx, &ignore, NULL)) {
2669 error_free(local_err);
2670 ret = 0;
2671 g_slist_free(ignore);
2672 ignore = g_slist_prepend(NULL, child);
2673 child_class->set_aio_ctx(child, ctx, &ignore);
2674 }
2675 g_slist_free(ignore);
2676 }
2677 if (ret < 0) {
2678 error_propagate(errp, local_err);
2679 g_free(child);
2680 bdrv_abort_perm_update(child_bs);
2681 bdrv_unref(child_bs);
2682 return NULL;
2683 }
2684 }
2685
2686 /* This performs the matching bdrv_set_perm() for the above check. */
2687 bdrv_replace_child(child, child_bs);
2688
2689 return child;
2690 }
2691
2692 /*
2693 * This function transfers the reference to child_bs from the caller
2694 * to parent_bs. That reference is later dropped by parent_bs on
2695 * bdrv_close() or if someone calls bdrv_unref_child().
2696 *
2697 * On failure NULL is returned, errp is set and the reference to
2698 * child_bs is also dropped.
2699 *
2700 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2701 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2702 */
2703 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2704 BlockDriverState *child_bs,
2705 const char *child_name,
2706 const BdrvChildClass *child_class,
2707 BdrvChildRole child_role,
2708 Error **errp)
2709 {
2710 BdrvChild *child;
2711 uint64_t perm, shared_perm;
2712
2713 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2714
2715 assert(parent_bs->drv);
2716 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2717 perm, shared_perm, &perm, &shared_perm);
2718
2719 child = bdrv_root_attach_child(child_bs, child_name, child_class,
2720 child_role, bdrv_get_aio_context(parent_bs),
2721 perm, shared_perm, parent_bs, errp);
2722 if (child == NULL) {
2723 return NULL;
2724 }
2725
2726 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2727 return child;
2728 }
2729
2730 static void bdrv_detach_child(BdrvChild *child)
2731 {
2732 QLIST_SAFE_REMOVE(child, next);
2733
2734 bdrv_replace_child(child, NULL);
2735
2736 g_free(child->name);
2737 g_free(child);
2738 }
2739
2740 /* Callers must ensure that child->frozen is false. */
2741 void bdrv_root_unref_child(BdrvChild *child)
2742 {
2743 BlockDriverState *child_bs;
2744
2745 child_bs = child->bs;
2746 bdrv_detach_child(child);
2747 bdrv_unref(child_bs);
2748 }
2749
2750 /**
2751 * Clear all inherits_from pointers from children and grandchildren of
2752 * @root that point to @root, where necessary.
2753 */
2754 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child)
2755 {
2756 BdrvChild *c;
2757
2758 if (child->bs->inherits_from == root) {
2759 /*
2760 * Remove inherits_from only when the last reference between root and
2761 * child->bs goes away.
2762 */
2763 QLIST_FOREACH(c, &root->children, next) {
2764 if (c != child && c->bs == child->bs) {
2765 break;
2766 }
2767 }
2768 if (c == NULL) {
2769 child->bs->inherits_from = NULL;
2770 }
2771 }
2772
2773 QLIST_FOREACH(c, &child->bs->children, next) {
2774 bdrv_unset_inherits_from(root, c);
2775 }
2776 }
2777
2778 /* Callers must ensure that child->frozen is false. */
2779 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2780 {
2781 if (child == NULL) {
2782 return;
2783 }
2784
2785 bdrv_unset_inherits_from(parent, child);
2786 bdrv_root_unref_child(child);
2787 }
2788
2789
2790 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2791 {
2792 BdrvChild *c;
2793 QLIST_FOREACH(c, &bs->parents, next_parent) {
2794 if (c->klass->change_media) {
2795 c->klass->change_media(c, load);
2796 }
2797 }
2798 }
2799
2800 /* Return true if you can reach parent going through child->inherits_from
2801 * recursively. If parent or child are NULL, return false */
2802 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2803 BlockDriverState *parent)
2804 {
2805 while (child && child != parent) {
2806 child = child->inherits_from;
2807 }
2808
2809 return child != NULL;
2810 }
2811
2812 /*
2813 * Return the BdrvChildRole for @bs's backing child. bs->backing is
2814 * mostly used for COW backing children (role = COW), but also for
2815 * filtered children (role = FILTERED | PRIMARY).
2816 */
2817 static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
2818 {
2819 if (bs->drv && bs->drv->is_filter) {
2820 return BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
2821 } else {
2822 return BDRV_CHILD_COW;
2823 }
2824 }
2825
2826 /*
2827 * Sets the bs->backing link of a BDS. A new reference is created; callers
2828 * which don't need their own reference any more must call bdrv_unref().
2829 */
2830 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2831 Error **errp)
2832 {
2833 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2834 bdrv_inherits_from_recursive(backing_hd, bs);
2835
2836 if (bdrv_is_backing_chain_frozen(bs, child_bs(bs->backing), errp)) {
2837 return;
2838 }
2839
2840 if (backing_hd) {
2841 bdrv_ref(backing_hd);
2842 }
2843
2844 if (bs->backing) {
2845 /* Cannot be frozen, we checked that above */
2846 bdrv_unref_child(bs, bs->backing);
2847 bs->backing = NULL;
2848 }
2849
2850 if (!backing_hd) {
2851 goto out;
2852 }
2853
2854 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_of_bds,
2855 bdrv_backing_role(bs), errp);
2856 /* If backing_hd was already part of bs's backing chain, and
2857 * inherits_from pointed recursively to bs then let's update it to
2858 * point directly to bs (else it will become NULL). */
2859 if (bs->backing && update_inherits_from) {
2860 backing_hd->inherits_from = bs;
2861 }
2862
2863 out:
2864 bdrv_refresh_limits(bs, NULL);
2865 }
2866
2867 /*
2868 * Opens the backing file for a BlockDriverState if not yet open
2869 *
2870 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2871 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2872 * itself, all options starting with "${bdref_key}." are considered part of the
2873 * BlockdevRef.
2874 *
2875 * TODO Can this be unified with bdrv_open_image()?
2876 */
2877 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2878 const char *bdref_key, Error **errp)
2879 {
2880 char *backing_filename = NULL;
2881 char *bdref_key_dot;
2882 const char *reference = NULL;
2883 int ret = 0;
2884 bool implicit_backing = false;
2885 BlockDriverState *backing_hd;
2886 QDict *options;
2887 QDict *tmp_parent_options = NULL;
2888 Error *local_err = NULL;
2889
2890 if (bs->backing != NULL) {
2891 goto free_exit;
2892 }
2893
2894 /* NULL means an empty set of options */
2895 if (parent_options == NULL) {
2896 tmp_parent_options = qdict_new();
2897 parent_options = tmp_parent_options;
2898 }
2899
2900 bs->open_flags &= ~BDRV_O_NO_BACKING;
2901
2902 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2903 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2904 g_free(bdref_key_dot);
2905
2906 /*
2907 * Caution: while qdict_get_try_str() is fine, getting non-string
2908 * types would require more care. When @parent_options come from
2909 * -blockdev or blockdev_add, its members are typed according to
2910 * the QAPI schema, but when they come from -drive, they're all
2911 * QString.
2912 */
2913 reference = qdict_get_try_str(parent_options, bdref_key);
2914 if (reference || qdict_haskey(options, "file.filename")) {
2915 /* keep backing_filename NULL */
2916 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2917 qobject_unref(options);
2918 goto free_exit;
2919 } else {
2920 if (qdict_size(options) == 0) {
2921 /* If the user specifies options that do not modify the
2922 * backing file's behavior, we might still consider it the
2923 * implicit backing file. But it's easier this way, and
2924 * just specifying some of the backing BDS's options is
2925 * only possible with -drive anyway (otherwise the QAPI
2926 * schema forces the user to specify everything). */
2927 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2928 }
2929
2930 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2931 if (local_err) {
2932 ret = -EINVAL;
2933 error_propagate(errp, local_err);
2934 qobject_unref(options);
2935 goto free_exit;
2936 }
2937 }
2938
2939 if (!bs->drv || !bs->drv->supports_backing) {
2940 ret = -EINVAL;
2941 error_setg(errp, "Driver doesn't support backing files");
2942 qobject_unref(options);
2943 goto free_exit;
2944 }
2945
2946 if (!reference &&
2947 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2948 qdict_put_str(options, "driver", bs->backing_format);
2949 }
2950
2951 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2952 &child_of_bds, bdrv_backing_role(bs), errp);
2953 if (!backing_hd) {
2954 bs->open_flags |= BDRV_O_NO_BACKING;
2955 error_prepend(errp, "Could not open backing file: ");
2956 ret = -EINVAL;
2957 goto free_exit;
2958 }
2959
2960 if (implicit_backing) {
2961 bdrv_refresh_filename(backing_hd);
2962 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2963 backing_hd->filename);
2964 }
2965
2966 /* Hook up the backing file link; drop our reference, bs owns the
2967 * backing_hd reference now */
2968 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2969 bdrv_unref(backing_hd);
2970 if (local_err) {
2971 error_propagate(errp, local_err);
2972 ret = -EINVAL;
2973 goto free_exit;
2974 }
2975
2976 qdict_del(parent_options, bdref_key);
2977
2978 free_exit:
2979 g_free(backing_filename);
2980 qobject_unref(tmp_parent_options);
2981 return ret;
2982 }
2983
2984 static BlockDriverState *
2985 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2986 BlockDriverState *parent, const BdrvChildClass *child_class,
2987 BdrvChildRole child_role, bool allow_none, Error **errp)
2988 {
2989 BlockDriverState *bs = NULL;
2990 QDict *image_options;
2991 char *bdref_key_dot;
2992 const char *reference;
2993
2994 assert(child_class != NULL);
2995
2996 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2997 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2998 g_free(bdref_key_dot);
2999
3000 /*
3001 * Caution: while qdict_get_try_str() is fine, getting non-string
3002 * types would require more care. When @options come from
3003 * -blockdev or blockdev_add, its members are typed according to
3004 * the QAPI schema, but when they come from -drive, they're all
3005 * QString.
3006 */
3007 reference = qdict_get_try_str(options, bdref_key);
3008 if (!filename && !reference && !qdict_size(image_options)) {
3009 if (!allow_none) {
3010 error_setg(errp, "A block device must be specified for \"%s\"",
3011 bdref_key);
3012 }
3013 qobject_unref(image_options);
3014 goto done;
3015 }
3016
3017 bs = bdrv_open_inherit(filename, reference, image_options, 0,
3018 parent, child_class, child_role, errp);
3019 if (!bs) {
3020 goto done;
3021 }
3022
3023 done:
3024 qdict_del(options, bdref_key);
3025 return bs;
3026 }
3027
3028 /*
3029 * Opens a disk image whose options are given as BlockdevRef in another block
3030 * device's options.
3031 *
3032 * If allow_none is true, no image will be opened if filename is false and no
3033 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3034 *
3035 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3036 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3037 * itself, all options starting with "${bdref_key}." are considered part of the
3038 * BlockdevRef.
3039 *
3040 * The BlockdevRef will be removed from the options QDict.
3041 */
3042 BdrvChild *bdrv_open_child(const char *filename,
3043 QDict *options, const char *bdref_key,
3044 BlockDriverState *parent,
3045 const BdrvChildClass *child_class,
3046 BdrvChildRole child_role,
3047 bool allow_none, Error **errp)
3048 {
3049 BlockDriverState *bs;
3050
3051 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_class,
3052 child_role, allow_none, errp);
3053 if (bs == NULL) {
3054 return NULL;
3055 }
3056
3057 return bdrv_attach_child(parent, bs, bdref_key, child_class, child_role,
3058 errp);
3059 }
3060
3061 /*
3062 * TODO Future callers may need to specify parent/child_class in order for
3063 * option inheritance to work. Existing callers use it for the root node.
3064 */
3065 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
3066 {
3067 BlockDriverState *bs = NULL;
3068 QObject *obj = NULL;
3069 QDict *qdict = NULL;
3070 const char *reference = NULL;
3071 Visitor *v = NULL;
3072
3073 if (ref->type == QTYPE_QSTRING) {
3074 reference = ref->u.reference;
3075 } else {
3076 BlockdevOptions *options = &ref->u.definition;
3077 assert(ref->type == QTYPE_QDICT);
3078
3079 v = qobject_output_visitor_new(&obj);
3080 visit_type_BlockdevOptions(v, NULL, &options, &error_abort);
3081 visit_complete(v, &obj);
3082
3083 qdict = qobject_to(QDict, obj);
3084 qdict_flatten(qdict);
3085
3086 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3087 * compatibility with other callers) rather than what we want as the
3088 * real defaults. Apply the defaults here instead. */
3089 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
3090 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
3091 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
3092 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
3093
3094 }
3095
3096 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, 0, errp);
3097 obj = NULL;
3098 qobject_unref(obj);
3099 visit_free(v);
3100 return bs;
3101 }
3102
3103 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
3104 int flags,
3105 QDict *snapshot_options,
3106 Error **errp)
3107 {
3108 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3109 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3110 int64_t total_size;
3111 QemuOpts *opts = NULL;
3112 BlockDriverState *bs_snapshot = NULL;
3113 Error *local_err = NULL;
3114 int ret;
3115
3116 /* if snapshot, we create a temporary backing file and open it
3117 instead of opening 'filename' directly */
3118
3119 /* Get the required size from the image */
3120 total_size = bdrv_getlength(bs);
3121 if (total_size < 0) {
3122 error_setg_errno(errp, -total_size, "Could not get image size");
3123 goto out;
3124 }
3125
3126 /* Create the temporary image */
3127 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3128 if (ret < 0) {
3129 error_setg_errno(errp, -ret, "Could not get temporary filename");
3130 goto out;
3131 }
3132
3133 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3134 &error_abort);
3135 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3136 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3137 qemu_opts_del(opts);
3138 if (ret < 0) {
3139 error_prepend(errp, "Could not create temporary overlay '%s': ",
3140 tmp_filename);
3141 goto out;
3142 }
3143
3144 /* Prepare options QDict for the temporary file */
3145 qdict_put_str(snapshot_options, "file.driver", "file");
3146 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3147 qdict_put_str(snapshot_options, "driver", "qcow2");
3148
3149 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3150 snapshot_options = NULL;
3151 if (!bs_snapshot) {
3152 goto out;
3153 }
3154
3155 /* bdrv_append() consumes a strong reference to bs_snapshot
3156 * (i.e. it will call bdrv_unref() on it) even on error, so in
3157 * order to be able to return one, we have to increase
3158 * bs_snapshot's refcount here */
3159 bdrv_ref(bs_snapshot);
3160 bdrv_append(bs_snapshot, bs, &local_err);
3161 if (local_err) {
3162 error_propagate(errp, local_err);
3163 bs_snapshot = NULL;
3164 goto out;
3165 }
3166
3167 out:
3168 qobject_unref(snapshot_options);
3169 g_free(tmp_filename);
3170 return bs_snapshot;
3171 }
3172
3173 /*
3174 * Opens a disk image (raw, qcow2, vmdk, ...)
3175 *
3176 * options is a QDict of options to pass to the block drivers, or NULL for an
3177 * empty set of options. The reference to the QDict belongs to the block layer
3178 * after the call (even on failure), so if the caller intends to reuse the
3179 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3180 *
3181 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3182 * If it is not NULL, the referenced BDS will be reused.
3183 *
3184 * The reference parameter may be used to specify an existing block device which
3185 * should be opened. If specified, neither options nor a filename may be given,
3186 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3187 */
3188 static BlockDriverState *bdrv_open_inherit(const char *filename,
3189 const char *reference,
3190 QDict *options, int flags,
3191 BlockDriverState *parent,
3192 const BdrvChildClass *child_class,
3193 BdrvChildRole child_role,
3194 Error **errp)
3195 {
3196 int ret;
3197 BlockBackend *file = NULL;
3198 BlockDriverState *bs;
3199 BlockDriver *drv = NULL;
3200 BdrvChild *child;
3201 const char *drvname;
3202 const char *backing;
3203 Error *local_err = NULL;
3204 QDict *snapshot_options = NULL;
3205 int snapshot_flags = 0;
3206
3207 assert(!child_class || !flags);
3208 assert(!child_class == !parent);
3209
3210 if (reference) {
3211 bool options_non_empty = options ? qdict_size(options) : false;
3212 qobject_unref(options);
3213
3214 if (filename || options_non_empty) {
3215 error_setg(errp, "Cannot reference an existing block device with "
3216 "additional options or a new filename");
3217 return NULL;
3218 }
3219
3220 bs = bdrv_lookup_bs(reference, reference, errp);
3221 if (!bs) {
3222 return NULL;
3223 }
3224
3225 bdrv_ref(bs);
3226 return bs;
3227 }
3228
3229 bs = bdrv_new();
3230
3231 /* NULL means an empty set of options */
3232 if (options == NULL) {
3233 options = qdict_new();
3234 }
3235
3236 /* json: syntax counts as explicit options, as if in the QDict */
3237 parse_json_protocol(options, &filename, &local_err);
3238 if (local_err) {
3239 goto fail;
3240 }
3241
3242 bs->explicit_options = qdict_clone_shallow(options);
3243
3244 if (child_class) {
3245 bool parent_is_format;
3246
3247 if (parent->drv) {
3248 parent_is_format = parent->drv->is_format;
3249 } else {
3250 /*
3251 * parent->drv is not set yet because this node is opened for
3252 * (potential) format probing. That means that @parent is going
3253 * to be a format node.
3254 */
3255 parent_is_format = true;
3256 }
3257
3258 bs->inherits_from = parent;
3259 child_class->inherit_options(child_role, parent_is_format,
3260 &flags, options,
3261 parent->open_flags, parent->options);
3262 }
3263
3264 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3265 if (ret < 0) {
3266 goto fail;
3267 }
3268
3269 /*
3270 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3271 * Caution: getting a boolean member of @options requires care.
3272 * When @options come from -blockdev or blockdev_add, members are
3273 * typed according to the QAPI schema, but when they come from
3274 * -drive, they're all QString.
3275 */
3276 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3277 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3278 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3279 } else {
3280 flags &= ~BDRV_O_RDWR;
3281 }
3282
3283 if (flags & BDRV_O_SNAPSHOT) {
3284 snapshot_options = qdict_new();
3285 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3286 flags, options);
3287 /* Let bdrv_backing_options() override "read-only" */
3288 qdict_del(options, BDRV_OPT_READ_ONLY);
3289 bdrv_inherited_options(BDRV_CHILD_COW, true,
3290 &flags, options, flags, options);
3291 }
3292
3293 bs->open_flags = flags;
3294 bs->options = options;
3295 options = qdict_clone_shallow(options);
3296
3297 /* Find the right image format driver */
3298 /* See cautionary note on accessing @options above */
3299 drvname = qdict_get_try_str(options, "driver");
3300 if (drvname) {
3301 drv = bdrv_find_format(drvname);
3302 if (!drv) {
3303 error_setg(errp, "Unknown driver: '%s'", drvname);
3304 goto fail;
3305 }
3306 }
3307
3308 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3309
3310 /* See cautionary note on accessing @options above */
3311 backing = qdict_get_try_str(options, "backing");
3312 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3313 (backing && *backing == '\0'))
3314 {
3315 if (backing) {
3316 warn_report("Use of \"backing\": \"\" is deprecated; "
3317 "use \"backing\": null instead");
3318 }
3319 flags |= BDRV_O_NO_BACKING;
3320 qdict_del(bs->explicit_options, "backing");
3321 qdict_del(bs->options, "backing");
3322 qdict_del(options, "backing");
3323 }
3324
3325 /* Open image file without format layer. This BlockBackend is only used for
3326 * probing, the block drivers will do their own bdrv_open_child() for the
3327 * same BDS, which is why we put the node name back into options. */
3328 if ((flags & BDRV_O_PROTOCOL) == 0) {
3329 BlockDriverState *file_bs;
3330
3331 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3332 &child_of_bds, BDRV_CHILD_IMAGE,
3333 true, &local_err);
3334 if (local_err) {
3335 goto fail;
3336 }
3337 if (file_bs != NULL) {
3338 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3339 * looking at the header to guess the image format. This works even
3340 * in cases where a guest would not see a consistent state. */
3341 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3342 blk_insert_bs(file, file_bs, &local_err);
3343 bdrv_unref(file_bs);
3344 if (local_err) {
3345 goto fail;
3346 }
3347
3348 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3349 }
3350 }
3351
3352 /* Image format probing */
3353 bs->probed = !drv;
3354 if (!drv && file) {
3355 ret = find_image_format(file, filename, &drv, &local_err);
3356 if (ret < 0) {
3357 goto fail;
3358 }
3359 /*
3360 * This option update would logically belong in bdrv_fill_options(),
3361 * but we first need to open bs->file for the probing to work, while
3362 * opening bs->file already requires the (mostly) final set of options
3363 * so that cache mode etc. can be inherited.
3364 *
3365 * Adding the driver later is somewhat ugly, but it's not an option
3366 * that would ever be inherited, so it's correct. We just need to make
3367 * sure to update both bs->options (which has the full effective
3368 * options for bs) and options (which has file.* already removed).
3369 */
3370 qdict_put_str(bs->options, "driver", drv->format_name);
3371 qdict_put_str(options, "driver", drv->format_name);
3372 } else if (!drv) {
3373 error_setg(errp, "Must specify either driver or file");
3374 goto fail;
3375 }
3376
3377 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3378 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3379 /* file must be NULL if a protocol BDS is about to be created
3380 * (the inverse results in an error message from bdrv_open_common()) */
3381 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3382
3383 /* Open the image */
3384 ret = bdrv_open_common(bs, file, options, &local_err);
3385 if (ret < 0) {
3386 goto fail;
3387 }
3388
3389 if (file) {
3390 blk_unref(file);
3391 file = NULL;
3392 }
3393
3394 /* If there is a backing file, use it */
3395 if ((flags & BDRV_O_NO_BACKING) == 0) {
3396 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3397 if (ret < 0) {
3398 goto close_and_fail;
3399 }
3400 }
3401
3402 /* Remove all children options and references
3403 * from bs->options and bs->explicit_options */
3404 QLIST_FOREACH(child, &bs->children, next) {
3405 char *child_key_dot;
3406 child_key_dot = g_strdup_printf("%s.", child->name);
3407 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3408 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3409 qdict_del(bs->explicit_options, child->name);
3410 qdict_del(bs->options, child->name);
3411 g_free(child_key_dot);
3412 }
3413
3414 /* Check if any unknown options were used */
3415 if (qdict_size(options) != 0) {
3416 const QDictEntry *entry = qdict_first(options);
3417 if (flags & BDRV_O_PROTOCOL) {
3418 error_setg(errp, "Block protocol '%s' doesn't support the option "
3419 "'%s'", drv->format_name, entry->key);
3420 } else {
3421 error_setg(errp,
3422 "Block format '%s' does not support the option '%s'",
3423 drv->format_name, entry->key);
3424 }
3425
3426 goto close_and_fail;
3427 }
3428
3429 bdrv_parent_cb_change_media(bs, true);
3430
3431 qobject_unref(options);
3432 options = NULL;
3433
3434 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3435 * temporary snapshot afterwards. */
3436 if (snapshot_flags) {
3437 BlockDriverState *snapshot_bs;
3438 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3439 snapshot_options, &local_err);
3440 snapshot_options = NULL;
3441 if (local_err) {
3442 goto close_and_fail;
3443 }
3444 /* We are not going to return bs but the overlay on top of it
3445 * (snapshot_bs); thus, we have to drop the strong reference to bs
3446 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3447 * though, because the overlay still has a reference to it. */
3448 bdrv_unref(bs);
3449 bs = snapshot_bs;
3450 }
3451
3452 return bs;
3453
3454 fail:
3455 blk_unref(file);
3456 qobject_unref(snapshot_options);
3457 qobject_unref(bs->explicit_options);
3458 qobject_unref(bs->options);
3459 qobject_unref(options);
3460 bs->options = NULL;
3461 bs->explicit_options = NULL;
3462 bdrv_unref(bs);
3463 error_propagate(errp, local_err);
3464 return NULL;
3465
3466 close_and_fail:
3467 bdrv_unref(bs);
3468 qobject_unref(snapshot_options);
3469 qobject_unref(options);
3470 error_propagate(errp, local_err);
3471 return NULL;
3472 }
3473
3474 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3475 QDict *options, int flags, Error **errp)
3476 {
3477 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3478 NULL, 0, errp);
3479 }
3480
3481 /* Return true if the NULL-terminated @list contains @str */
3482 static bool is_str_in_list(const char *str, const char *const *list)
3483 {
3484 if (str && list) {
3485 int i;
3486 for (i = 0; list[i] != NULL; i++) {
3487 if (!strcmp(str, list[i])) {
3488 return true;
3489 }
3490 }
3491 }
3492 return false;
3493 }
3494
3495 /*
3496 * Check that every option set in @bs->options is also set in
3497 * @new_opts.
3498 *
3499 * Options listed in the common_options list and in
3500 * @bs->drv->mutable_opts are skipped.
3501 *
3502 * Return 0 on success, otherwise return -EINVAL and set @errp.
3503 */
3504 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3505 const QDict *new_opts, Error **errp)
3506 {
3507 const QDictEntry *e;
3508 /* These options are common to all block drivers and are handled
3509 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3510 const char *const common_options[] = {
3511 "node-name", "discard", "cache.direct", "cache.no-flush",
3512 "read-only", "auto-read-only", "detect-zeroes", NULL
3513 };
3514
3515 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3516 if (!qdict_haskey(new_opts, e->key) &&
3517 !is_str_in_list(e->key, common_options) &&
3518 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3519 error_setg(errp, "Option '%s' cannot be reset "
3520 "to its default value", e->key);
3521 return -EINVAL;
3522 }
3523 }
3524
3525 return 0;
3526 }
3527
3528 /*
3529 * Returns true if @child can be reached recursively from @bs
3530 */
3531 static bool bdrv_recurse_has_child(BlockDriverState *bs,
3532 BlockDriverState *child)
3533 {
3534 BdrvChild *c;
3535
3536 if (bs == child) {
3537 return true;
3538 }
3539
3540 QLIST_FOREACH(c, &bs->children, next) {
3541 if (bdrv_recurse_has_child(c->bs, child)) {
3542 return true;
3543 }
3544 }
3545
3546 return false;
3547 }
3548
3549 /*
3550 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3551 * reopen of multiple devices.
3552 *
3553 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3554 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3555 * be created and initialized. This newly created BlockReopenQueue should be
3556 * passed back in for subsequent calls that are intended to be of the same
3557 * atomic 'set'.
3558 *
3559 * bs is the BlockDriverState to add to the reopen queue.
3560 *
3561 * options contains the changed options for the associated bs
3562 * (the BlockReopenQueue takes ownership)
3563 *
3564 * flags contains the open flags for the associated bs
3565 *
3566 * returns a pointer to bs_queue, which is either the newly allocated
3567 * bs_queue, or the existing bs_queue being used.
3568 *
3569 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3570 */