block: Introduce 'bdrv_reopen_commit_post' step
[qemu.git] / block.c
1 /*
2 * QEMU System Emulator block driver
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "module_block.h"
33 #include "qemu/main-loop.h"
34 #include "qemu/module.h"
35 #include "qapi/error.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qjson.h"
38 #include "qapi/qmp/qnull.h"
39 #include "qapi/qmp/qstring.h"
40 #include "qapi/qobject-output-visitor.h"
41 #include "qapi/qapi-visit-block-core.h"
42 #include "sysemu/block-backend.h"
43 #include "sysemu/sysemu.h"
44 #include "qemu/notify.h"
45 #include "qemu/option.h"
46 #include "qemu/coroutine.h"
47 #include "block/qapi.h"
48 #include "qemu/timer.h"
49 #include "qemu/cutils.h"
50 #include "qemu/id.h"
51
52 #ifdef CONFIG_BSD
53 #include <sys/ioctl.h>
54 #include <sys/queue.h>
55 #ifndef __DragonFly__
56 #include <sys/disk.h>
57 #endif
58 #endif
59
60 #ifdef _WIN32
61 #include <windows.h>
62 #endif
63
64 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
65
66 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
67 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
68
69 static QTAILQ_HEAD(, BlockDriverState) all_bdrv_states =
70 QTAILQ_HEAD_INITIALIZER(all_bdrv_states);
71
72 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
73 QLIST_HEAD_INITIALIZER(bdrv_drivers);
74
75 static BlockDriverState *bdrv_open_inherit(const char *filename,
76 const char *reference,
77 QDict *options, int flags,
78 BlockDriverState *parent,
79 const BdrvChildRole *child_role,
80 Error **errp);
81
82 /* If non-zero, use only whitelisted block drivers */
83 static int use_bdrv_whitelist;
84
85 #ifdef _WIN32
86 static int is_windows_drive_prefix(const char *filename)
87 {
88 return (((filename[0] >= 'a' && filename[0] <= 'z') ||
89 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
90 filename[1] == ':');
91 }
92
93 int is_windows_drive(const char *filename)
94 {
95 if (is_windows_drive_prefix(filename) &&
96 filename[2] == '\0')
97 return 1;
98 if (strstart(filename, "\\\\.\\", NULL) ||
99 strstart(filename, "//./", NULL))
100 return 1;
101 return 0;
102 }
103 #endif
104
105 size_t bdrv_opt_mem_align(BlockDriverState *bs)
106 {
107 if (!bs || !bs->drv) {
108 /* page size or 4k (hdd sector size) should be on the safe side */
109 return MAX(4096, qemu_real_host_page_size);
110 }
111
112 return bs->bl.opt_mem_alignment;
113 }
114
115 size_t bdrv_min_mem_align(BlockDriverState *bs)
116 {
117 if (!bs || !bs->drv) {
118 /* page size or 4k (hdd sector size) should be on the safe side */
119 return MAX(4096, qemu_real_host_page_size);
120 }
121
122 return bs->bl.min_mem_alignment;
123 }
124
125 /* check if the path starts with "<protocol>:" */
126 int path_has_protocol(const char *path)
127 {
128 const char *p;
129
130 #ifdef _WIN32
131 if (is_windows_drive(path) ||
132 is_windows_drive_prefix(path)) {
133 return 0;
134 }
135 p = path + strcspn(path, ":/\\");
136 #else
137 p = path + strcspn(path, ":/");
138 #endif
139
140 return *p == ':';
141 }
142
143 int path_is_absolute(const char *path)
144 {
145 #ifdef _WIN32
146 /* specific case for names like: "\\.\d:" */
147 if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
148 return 1;
149 }
150 return (*path == '/' || *path == '\\');
151 #else
152 return (*path == '/');
153 #endif
154 }
155
156 /* if filename is absolute, just return its duplicate. Otherwise, build a
157 path to it by considering it is relative to base_path. URL are
158 supported. */
159 char *path_combine(const char *base_path, const char *filename)
160 {
161 const char *protocol_stripped = NULL;
162 const char *p, *p1;
163 char *result;
164 int len;
165
166 if (path_is_absolute(filename)) {
167 return g_strdup(filename);
168 }
169
170 if (path_has_protocol(base_path)) {
171 protocol_stripped = strchr(base_path, ':');
172 if (protocol_stripped) {
173 protocol_stripped++;
174 }
175 }
176 p = protocol_stripped ?: base_path;
177
178 p1 = strrchr(base_path, '/');
179 #ifdef _WIN32
180 {
181 const char *p2;
182 p2 = strrchr(base_path, '\\');
183 if (!p1 || p2 > p1) {
184 p1 = p2;
185 }
186 }
187 #endif
188 if (p1) {
189 p1++;
190 } else {
191 p1 = base_path;
192 }
193 if (p1 > p) {
194 p = p1;
195 }
196 len = p - base_path;
197
198 result = g_malloc(len + strlen(filename) + 1);
199 memcpy(result, base_path, len);
200 strcpy(result + len, filename);
201
202 return result;
203 }
204
205 /*
206 * Helper function for bdrv_parse_filename() implementations to remove optional
207 * protocol prefixes (especially "file:") from a filename and for putting the
208 * stripped filename into the options QDict if there is such a prefix.
209 */
210 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
211 QDict *options)
212 {
213 if (strstart(filename, prefix, &filename)) {
214 /* Stripping the explicit protocol prefix may result in a protocol
215 * prefix being (wrongly) detected (if the filename contains a colon) */
216 if (path_has_protocol(filename)) {
217 QString *fat_filename;
218
219 /* This means there is some colon before the first slash; therefore,
220 * this cannot be an absolute path */
221 assert(!path_is_absolute(filename));
222
223 /* And we can thus fix the protocol detection issue by prefixing it
224 * by "./" */
225 fat_filename = qstring_from_str("./");
226 qstring_append(fat_filename, filename);
227
228 assert(!path_has_protocol(qstring_get_str(fat_filename)));
229
230 qdict_put(options, "filename", fat_filename);
231 } else {
232 /* If no protocol prefix was detected, we can use the shortened
233 * filename as-is */
234 qdict_put_str(options, "filename", filename);
235 }
236 }
237 }
238
239
240 /* Returns whether the image file is opened as read-only. Note that this can
241 * return false and writing to the image file is still not possible because the
242 * image is inactivated. */
243 bool bdrv_is_read_only(BlockDriverState *bs)
244 {
245 return bs->read_only;
246 }
247
248 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
249 bool ignore_allow_rdw, Error **errp)
250 {
251 /* Do not set read_only if copy_on_read is enabled */
252 if (bs->copy_on_read && read_only) {
253 error_setg(errp, "Can't set node '%s' to r/o with copy-on-read enabled",
254 bdrv_get_device_or_node_name(bs));
255 return -EINVAL;
256 }
257
258 /* Do not clear read_only if it is prohibited */
259 if (!read_only && !(bs->open_flags & BDRV_O_ALLOW_RDWR) &&
260 !ignore_allow_rdw)
261 {
262 error_setg(errp, "Node '%s' is read only",
263 bdrv_get_device_or_node_name(bs));
264 return -EPERM;
265 }
266
267 return 0;
268 }
269
270 /*
271 * Called by a driver that can only provide a read-only image.
272 *
273 * Returns 0 if the node is already read-only or it could switch the node to
274 * read-only because BDRV_O_AUTO_RDONLY is set.
275 *
276 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
277 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
278 * is not NULL, it is used as the error message for the Error object.
279 */
280 int bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
281 Error **errp)
282 {
283 int ret = 0;
284
285 if (!(bs->open_flags & BDRV_O_RDWR)) {
286 return 0;
287 }
288 if (!(bs->open_flags & BDRV_O_AUTO_RDONLY)) {
289 goto fail;
290 }
291
292 ret = bdrv_can_set_read_only(bs, true, false, NULL);
293 if (ret < 0) {
294 goto fail;
295 }
296
297 bs->read_only = true;
298 bs->open_flags &= ~BDRV_O_RDWR;
299
300 return 0;
301
302 fail:
303 error_setg(errp, "%s", errmsg ?: "Image is read-only");
304 return -EACCES;
305 }
306
307 /*
308 * If @backing is empty, this function returns NULL without setting
309 * @errp. In all other cases, NULL will only be returned with @errp
310 * set.
311 *
312 * Therefore, a return value of NULL without @errp set means that
313 * there is no backing file; if @errp is set, there is one but its
314 * absolute filename cannot be generated.
315 */
316 char *bdrv_get_full_backing_filename_from_filename(const char *backed,
317 const char *backing,
318 Error **errp)
319 {
320 if (backing[0] == '\0') {
321 return NULL;
322 } else if (path_has_protocol(backing) || path_is_absolute(backing)) {
323 return g_strdup(backing);
324 } else if (backed[0] == '\0' || strstart(backed, "json:", NULL)) {
325 error_setg(errp, "Cannot use relative backing file names for '%s'",
326 backed);
327 return NULL;
328 } else {
329 return path_combine(backed, backing);
330 }
331 }
332
333 /*
334 * If @filename is empty or NULL, this function returns NULL without
335 * setting @errp. In all other cases, NULL will only be returned with
336 * @errp set.
337 */
338 static char *bdrv_make_absolute_filename(BlockDriverState *relative_to,
339 const char *filename, Error **errp)
340 {
341 char *dir, *full_name;
342
343 if (!filename || filename[0] == '\0') {
344 return NULL;
345 } else if (path_has_protocol(filename) || path_is_absolute(filename)) {
346 return g_strdup(filename);
347 }
348
349 dir = bdrv_dirname(relative_to, errp);
350 if (!dir) {
351 return NULL;
352 }
353
354 full_name = g_strconcat(dir, filename, NULL);
355 g_free(dir);
356 return full_name;
357 }
358
359 char *bdrv_get_full_backing_filename(BlockDriverState *bs, Error **errp)
360 {
361 return bdrv_make_absolute_filename(bs, bs->backing_file, errp);
362 }
363
364 void bdrv_register(BlockDriver *bdrv)
365 {
366 QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
367 }
368
369 BlockDriverState *bdrv_new(void)
370 {
371 BlockDriverState *bs;
372 int i;
373
374 bs = g_new0(BlockDriverState, 1);
375 QLIST_INIT(&bs->dirty_bitmaps);
376 for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
377 QLIST_INIT(&bs->op_blockers[i]);
378 }
379 notifier_with_return_list_init(&bs->before_write_notifiers);
380 qemu_co_mutex_init(&bs->reqs_lock);
381 qemu_mutex_init(&bs->dirty_bitmap_mutex);
382 bs->refcnt = 1;
383 bs->aio_context = qemu_get_aio_context();
384
385 qemu_co_queue_init(&bs->flush_queue);
386
387 for (i = 0; i < bdrv_drain_all_count; i++) {
388 bdrv_drained_begin(bs);
389 }
390
391 QTAILQ_INSERT_TAIL(&all_bdrv_states, bs, bs_list);
392
393 return bs;
394 }
395
396 static BlockDriver *bdrv_do_find_format(const char *format_name)
397 {
398 BlockDriver *drv1;
399
400 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
401 if (!strcmp(drv1->format_name, format_name)) {
402 return drv1;
403 }
404 }
405
406 return NULL;
407 }
408
409 BlockDriver *bdrv_find_format(const char *format_name)
410 {
411 BlockDriver *drv1;
412 int i;
413
414 drv1 = bdrv_do_find_format(format_name);
415 if (drv1) {
416 return drv1;
417 }
418
419 /* The driver isn't registered, maybe we need to load a module */
420 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
421 if (!strcmp(block_driver_modules[i].format_name, format_name)) {
422 block_module_load_one(block_driver_modules[i].library_name);
423 break;
424 }
425 }
426
427 return bdrv_do_find_format(format_name);
428 }
429
430 static int bdrv_format_is_whitelisted(const char *format_name, bool read_only)
431 {
432 static const char *whitelist_rw[] = {
433 CONFIG_BDRV_RW_WHITELIST
434 };
435 static const char *whitelist_ro[] = {
436 CONFIG_BDRV_RO_WHITELIST
437 };
438 const char **p;
439
440 if (!whitelist_rw[0] && !whitelist_ro[0]) {
441 return 1; /* no whitelist, anything goes */
442 }
443
444 for (p = whitelist_rw; *p; p++) {
445 if (!strcmp(format_name, *p)) {
446 return 1;
447 }
448 }
449 if (read_only) {
450 for (p = whitelist_ro; *p; p++) {
451 if (!strcmp(format_name, *p)) {
452 return 1;
453 }
454 }
455 }
456 return 0;
457 }
458
459 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
460 {
461 return bdrv_format_is_whitelisted(drv->format_name, read_only);
462 }
463
464 bool bdrv_uses_whitelist(void)
465 {
466 return use_bdrv_whitelist;
467 }
468
469 typedef struct CreateCo {
470 BlockDriver *drv;
471 char *filename;
472 QemuOpts *opts;
473 int ret;
474 Error *err;
475 } CreateCo;
476
477 static void coroutine_fn bdrv_create_co_entry(void *opaque)
478 {
479 Error *local_err = NULL;
480 int ret;
481
482 CreateCo *cco = opaque;
483 assert(cco->drv);
484
485 ret = cco->drv->bdrv_co_create_opts(cco->filename, cco->opts, &local_err);
486 error_propagate(&cco->err, local_err);
487 cco->ret = ret;
488 }
489
490 int bdrv_create(BlockDriver *drv, const char* filename,
491 QemuOpts *opts, Error **errp)
492 {
493 int ret;
494
495 Coroutine *co;
496 CreateCo cco = {
497 .drv = drv,
498 .filename = g_strdup(filename),
499 .opts = opts,
500 .ret = NOT_DONE,
501 .err = NULL,
502 };
503
504 if (!drv->bdrv_co_create_opts) {
505 error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
506 ret = -ENOTSUP;
507 goto out;
508 }
509
510 if (qemu_in_coroutine()) {
511 /* Fast-path if already in coroutine context */
512 bdrv_create_co_entry(&cco);
513 } else {
514 co = qemu_coroutine_create(bdrv_create_co_entry, &cco);
515 qemu_coroutine_enter(co);
516 while (cco.ret == NOT_DONE) {
517 aio_poll(qemu_get_aio_context(), true);
518 }
519 }
520
521 ret = cco.ret;
522 if (ret < 0) {
523 if (cco.err) {
524 error_propagate(errp, cco.err);
525 } else {
526 error_setg_errno(errp, -ret, "Could not create image");
527 }
528 }
529
530 out:
531 g_free(cco.filename);
532 return ret;
533 }
534
535 /**
536 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
537 * least the given @minimum_size.
538 *
539 * On success, return @blk's actual length.
540 * Otherwise, return -errno.
541 */
542 static int64_t create_file_fallback_truncate(BlockBackend *blk,
543 int64_t minimum_size, Error **errp)
544 {
545 Error *local_err = NULL;
546 int64_t size;
547 int ret;
548
549 ret = blk_truncate(blk, minimum_size, false, PREALLOC_MODE_OFF, &local_err);
550 if (ret < 0 && ret != -ENOTSUP) {
551 error_propagate(errp, local_err);
552 return ret;
553 }
554
555 size = blk_getlength(blk);
556 if (size < 0) {
557 error_free(local_err);
558 error_setg_errno(errp, -size,
559 "Failed to inquire the new image file's length");
560 return size;
561 }
562
563 if (size < minimum_size) {
564 /* Need to grow the image, but we failed to do that */
565 error_propagate(errp, local_err);
566 return -ENOTSUP;
567 }
568
569 error_free(local_err);
570 local_err = NULL;
571
572 return size;
573 }
574
575 /**
576 * Helper function for bdrv_create_file_fallback(): Zero the first
577 * sector to remove any potentially pre-existing image header.
578 */
579 static int create_file_fallback_zero_first_sector(BlockBackend *blk,
580 int64_t current_size,
581 Error **errp)
582 {
583 int64_t bytes_to_clear;
584 int ret;
585
586 bytes_to_clear = MIN(current_size, BDRV_SECTOR_SIZE);
587 if (bytes_to_clear) {
588 ret = blk_pwrite_zeroes(blk, 0, bytes_to_clear, BDRV_REQ_MAY_UNMAP);
589 if (ret < 0) {
590 error_setg_errno(errp, -ret,
591 "Failed to clear the new image's first sector");
592 return ret;
593 }
594 }
595
596 return 0;
597 }
598
599 static int bdrv_create_file_fallback(const char *filename, BlockDriver *drv,
600 QemuOpts *opts, Error **errp)
601 {
602 BlockBackend *blk;
603 QDict *options;
604 int64_t size = 0;
605 char *buf = NULL;
606 PreallocMode prealloc;
607 Error *local_err = NULL;
608 int ret;
609
610 size = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
611 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
612 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
613 PREALLOC_MODE_OFF, &local_err);
614 g_free(buf);
615 if (local_err) {
616 error_propagate(errp, local_err);
617 return -EINVAL;
618 }
619
620 if (prealloc != PREALLOC_MODE_OFF) {
621 error_setg(errp, "Unsupported preallocation mode '%s'",
622 PreallocMode_str(prealloc));
623 return -ENOTSUP;
624 }
625
626 options = qdict_new();
627 qdict_put_str(options, "driver", drv->format_name);
628
629 blk = blk_new_open(filename, NULL, options,
630 BDRV_O_RDWR | BDRV_O_RESIZE, errp);
631 if (!blk) {
632 error_prepend(errp, "Protocol driver '%s' does not support image "
633 "creation, and opening the image failed: ",
634 drv->format_name);
635 return -EINVAL;
636 }
637
638 size = create_file_fallback_truncate(blk, size, errp);
639 if (size < 0) {
640 ret = size;
641 goto out;
642 }
643
644 ret = create_file_fallback_zero_first_sector(blk, size, errp);
645 if (ret < 0) {
646 goto out;
647 }
648
649 ret = 0;
650 out:
651 blk_unref(blk);
652 return ret;
653 }
654
655 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
656 {
657 BlockDriver *drv;
658
659 drv = bdrv_find_protocol(filename, true, errp);
660 if (drv == NULL) {
661 return -ENOENT;
662 }
663
664 if (drv->bdrv_co_create_opts) {
665 return bdrv_create(drv, filename, opts, errp);
666 } else {
667 return bdrv_create_file_fallback(filename, drv, opts, errp);
668 }
669 }
670
671 /**
672 * Try to get @bs's logical and physical block size.
673 * On success, store them in @bsz struct and return 0.
674 * On failure return -errno.
675 * @bs must not be empty.
676 */
677 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
678 {
679 BlockDriver *drv = bs->drv;
680
681 if (drv && drv->bdrv_probe_blocksizes) {
682 return drv->bdrv_probe_blocksizes(bs, bsz);
683 } else if (drv && drv->is_filter && bs->file) {
684 return bdrv_probe_blocksizes(bs->file->bs, bsz);
685 }
686
687 return -ENOTSUP;
688 }
689
690 /**
691 * Try to get @bs's geometry (cyls, heads, sectors).
692 * On success, store them in @geo struct and return 0.
693 * On failure return -errno.
694 * @bs must not be empty.
695 */
696 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
697 {
698 BlockDriver *drv = bs->drv;
699
700 if (drv && drv->bdrv_probe_geometry) {
701 return drv->bdrv_probe_geometry(bs, geo);
702 } else if (drv && drv->is_filter && bs->file) {
703 return bdrv_probe_geometry(bs->file->bs, geo);
704 }
705
706 return -ENOTSUP;
707 }
708
709 /*
710 * Create a uniquely-named empty temporary file.
711 * Return 0 upon success, otherwise a negative errno value.
712 */
713 int get_tmp_filename(char *filename, int size)
714 {
715 #ifdef _WIN32
716 char temp_dir[MAX_PATH];
717 /* GetTempFileName requires that its output buffer (4th param)
718 have length MAX_PATH or greater. */
719 assert(size >= MAX_PATH);
720 return (GetTempPath(MAX_PATH, temp_dir)
721 && GetTempFileName(temp_dir, "qem", 0, filename)
722 ? 0 : -GetLastError());
723 #else
724 int fd;
725 const char *tmpdir;
726 tmpdir = getenv("TMPDIR");
727 if (!tmpdir) {
728 tmpdir = "/var/tmp";
729 }
730 if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
731 return -EOVERFLOW;
732 }
733 fd = mkstemp(filename);
734 if (fd < 0) {
735 return -errno;
736 }
737 if (close(fd) != 0) {
738 unlink(filename);
739 return -errno;
740 }
741 return 0;
742 #endif
743 }
744
745 /*
746 * Detect host devices. By convention, /dev/cdrom[N] is always
747 * recognized as a host CDROM.
748 */
749 static BlockDriver *find_hdev_driver(const char *filename)
750 {
751 int score_max = 0, score;
752 BlockDriver *drv = NULL, *d;
753
754 QLIST_FOREACH(d, &bdrv_drivers, list) {
755 if (d->bdrv_probe_device) {
756 score = d->bdrv_probe_device(filename);
757 if (score > score_max) {
758 score_max = score;
759 drv = d;
760 }
761 }
762 }
763
764 return drv;
765 }
766
767 static BlockDriver *bdrv_do_find_protocol(const char *protocol)
768 {
769 BlockDriver *drv1;
770
771 QLIST_FOREACH(drv1, &bdrv_drivers, list) {
772 if (drv1->protocol_name && !strcmp(drv1->protocol_name, protocol)) {
773 return drv1;
774 }
775 }
776
777 return NULL;
778 }
779
780 BlockDriver *bdrv_find_protocol(const char *filename,
781 bool allow_protocol_prefix,
782 Error **errp)
783 {
784 BlockDriver *drv1;
785 char protocol[128];
786 int len;
787 const char *p;
788 int i;
789
790 /* TODO Drivers without bdrv_file_open must be specified explicitly */
791
792 /*
793 * XXX(hch): we really should not let host device detection
794 * override an explicit protocol specification, but moving this
795 * later breaks access to device names with colons in them.
796 * Thanks to the brain-dead persistent naming schemes on udev-
797 * based Linux systems those actually are quite common.
798 */
799 drv1 = find_hdev_driver(filename);
800 if (drv1) {
801 return drv1;
802 }
803
804 if (!path_has_protocol(filename) || !allow_protocol_prefix) {
805 return &bdrv_file;
806 }
807
808 p = strchr(filename, ':');
809 assert(p != NULL);
810 len = p - filename;
811 if (len > sizeof(protocol) - 1)
812 len = sizeof(protocol) - 1;
813 memcpy(protocol, filename, len);
814 protocol[len] = '\0';
815
816 drv1 = bdrv_do_find_protocol(protocol);
817 if (drv1) {
818 return drv1;
819 }
820
821 for (i = 0; i < (int)ARRAY_SIZE(block_driver_modules); ++i) {
822 if (block_driver_modules[i].protocol_name &&
823 !strcmp(block_driver_modules[i].protocol_name, protocol)) {
824 block_module_load_one(block_driver_modules[i].library_name);
825 break;
826 }
827 }
828
829 drv1 = bdrv_do_find_protocol(protocol);
830 if (!drv1) {
831 error_setg(errp, "Unknown protocol '%s'", protocol);
832 }
833 return drv1;
834 }
835
836 /*
837 * Guess image format by probing its contents.
838 * This is not a good idea when your image is raw (CVE-2008-2004), but
839 * we do it anyway for backward compatibility.
840 *
841 * @buf contains the image's first @buf_size bytes.
842 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
843 * but can be smaller if the image file is smaller)
844 * @filename is its filename.
845 *
846 * For all block drivers, call the bdrv_probe() method to get its
847 * probing score.
848 * Return the first block driver with the highest probing score.
849 */
850 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
851 const char *filename)
852 {
853 int score_max = 0, score;
854 BlockDriver *drv = NULL, *d;
855
856 QLIST_FOREACH(d, &bdrv_drivers, list) {
857 if (d->bdrv_probe) {
858 score = d->bdrv_probe(buf, buf_size, filename);
859 if (score > score_max) {
860 score_max = score;
861 drv = d;
862 }
863 }
864 }
865
866 return drv;
867 }
868
869 static int find_image_format(BlockBackend *file, const char *filename,
870 BlockDriver **pdrv, Error **errp)
871 {
872 BlockDriver *drv;
873 uint8_t buf[BLOCK_PROBE_BUF_SIZE];
874 int ret = 0;
875
876 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
877 if (blk_is_sg(file) || !blk_is_inserted(file) || blk_getlength(file) == 0) {
878 *pdrv = &bdrv_raw;
879 return ret;
880 }
881
882 ret = blk_pread(file, 0, buf, sizeof(buf));
883 if (ret < 0) {
884 error_setg_errno(errp, -ret, "Could not read image for determining its "
885 "format");
886 *pdrv = NULL;
887 return ret;
888 }
889
890 drv = bdrv_probe_all(buf, ret, filename);
891 if (!drv) {
892 error_setg(errp, "Could not determine image format: No compatible "
893 "driver found");
894 ret = -ENOENT;
895 }
896 *pdrv = drv;
897 return ret;
898 }
899
900 /**
901 * Set the current 'total_sectors' value
902 * Return 0 on success, -errno on error.
903 */
904 int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
905 {
906 BlockDriver *drv = bs->drv;
907
908 if (!drv) {
909 return -ENOMEDIUM;
910 }
911
912 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
913 if (bdrv_is_sg(bs))
914 return 0;
915
916 /* query actual device if possible, otherwise just trust the hint */
917 if (drv->bdrv_getlength) {
918 int64_t length = drv->bdrv_getlength(bs);
919 if (length < 0) {
920 return length;
921 }
922 hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
923 }
924
925 bs->total_sectors = hint;
926 return 0;
927 }
928
929 /**
930 * Combines a QDict of new block driver @options with any missing options taken
931 * from @old_options, so that leaving out an option defaults to its old value.
932 */
933 static void bdrv_join_options(BlockDriverState *bs, QDict *options,
934 QDict *old_options)
935 {
936 if (bs->drv && bs->drv->bdrv_join_options) {
937 bs->drv->bdrv_join_options(options, old_options);
938 } else {
939 qdict_join(options, old_options, false);
940 }
941 }
942
943 static BlockdevDetectZeroesOptions bdrv_parse_detect_zeroes(QemuOpts *opts,
944 int open_flags,
945 Error **errp)
946 {
947 Error *local_err = NULL;
948 char *value = qemu_opt_get_del(opts, "detect-zeroes");
949 BlockdevDetectZeroesOptions detect_zeroes =
950 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup, value,
951 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF, &local_err);
952 g_free(value);
953 if (local_err) {
954 error_propagate(errp, local_err);
955 return detect_zeroes;
956 }
957
958 if (detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP &&
959 !(open_flags & BDRV_O_UNMAP))
960 {
961 error_setg(errp, "setting detect-zeroes to unmap is not allowed "
962 "without setting discard operation to unmap");
963 }
964
965 return detect_zeroes;
966 }
967
968 /**
969 * Set open flags for aio engine
970 *
971 * Return 0 on success, -1 if the engine specified is invalid
972 */
973 int bdrv_parse_aio(const char *mode, int *flags)
974 {
975 if (!strcmp(mode, "threads")) {
976 /* do nothing, default */
977 } else if (!strcmp(mode, "native")) {
978 *flags |= BDRV_O_NATIVE_AIO;
979 #ifdef CONFIG_LINUX_IO_URING
980 } else if (!strcmp(mode, "io_uring")) {
981 *flags |= BDRV_O_IO_URING;
982 #endif
983 } else {
984 return -1;
985 }
986
987 return 0;
988 }
989
990 /**
991 * Set open flags for a given discard mode
992 *
993 * Return 0 on success, -1 if the discard mode was invalid.
994 */
995 int bdrv_parse_discard_flags(const char *mode, int *flags)
996 {
997 *flags &= ~BDRV_O_UNMAP;
998
999 if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
1000 /* do nothing */
1001 } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
1002 *flags |= BDRV_O_UNMAP;
1003 } else {
1004 return -1;
1005 }
1006
1007 return 0;
1008 }
1009
1010 /**
1011 * Set open flags for a given cache mode
1012 *
1013 * Return 0 on success, -1 if the cache mode was invalid.
1014 */
1015 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough)
1016 {
1017 *flags &= ~BDRV_O_CACHE_MASK;
1018
1019 if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
1020 *writethrough = false;
1021 *flags |= BDRV_O_NOCACHE;
1022 } else if (!strcmp(mode, "directsync")) {
1023 *writethrough = true;
1024 *flags |= BDRV_O_NOCACHE;
1025 } else if (!strcmp(mode, "writeback")) {
1026 *writethrough = false;
1027 } else if (!strcmp(mode, "unsafe")) {
1028 *writethrough = false;
1029 *flags |= BDRV_O_NO_FLUSH;
1030 } else if (!strcmp(mode, "writethrough")) {
1031 *writethrough = true;
1032 } else {
1033 return -1;
1034 }
1035
1036 return 0;
1037 }
1038
1039 static char *bdrv_child_get_parent_desc(BdrvChild *c)
1040 {
1041 BlockDriverState *parent = c->opaque;
1042 return g_strdup(bdrv_get_device_or_node_name(parent));
1043 }
1044
1045 static void bdrv_child_cb_drained_begin(BdrvChild *child)
1046 {
1047 BlockDriverState *bs = child->opaque;
1048 bdrv_do_drained_begin_quiesce(bs, NULL, false);
1049 }
1050
1051 static bool bdrv_child_cb_drained_poll(BdrvChild *child)
1052 {
1053 BlockDriverState *bs = child->opaque;
1054 return bdrv_drain_poll(bs, false, NULL, false);
1055 }
1056
1057 static void bdrv_child_cb_drained_end(BdrvChild *child,
1058 int *drained_end_counter)
1059 {
1060 BlockDriverState *bs = child->opaque;
1061 bdrv_drained_end_no_poll(bs, drained_end_counter);
1062 }
1063
1064 static void bdrv_child_cb_attach(BdrvChild *child)
1065 {
1066 BlockDriverState *bs = child->opaque;
1067 bdrv_apply_subtree_drain(child, bs);
1068 }
1069
1070 static void bdrv_child_cb_detach(BdrvChild *child)
1071 {
1072 BlockDriverState *bs = child->opaque;
1073 bdrv_unapply_subtree_drain(child, bs);
1074 }
1075
1076 static int bdrv_child_cb_inactivate(BdrvChild *child)
1077 {
1078 BlockDriverState *bs = child->opaque;
1079 assert(bs->open_flags & BDRV_O_INACTIVE);
1080 return 0;
1081 }
1082
1083 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1084 GSList **ignore, Error **errp)
1085 {
1086 BlockDriverState *bs = child->opaque;
1087 return bdrv_can_set_aio_context(bs, ctx, ignore, errp);
1088 }
1089
1090 static void bdrv_child_cb_set_aio_ctx(BdrvChild *child, AioContext *ctx,
1091 GSList **ignore)
1092 {
1093 BlockDriverState *bs = child->opaque;
1094 return bdrv_set_aio_context_ignore(bs, ctx, ignore);
1095 }
1096
1097 /*
1098 * Returns the options and flags that a temporary snapshot should get, based on
1099 * the originally requested flags (the originally requested image will have
1100 * flags like a backing file)
1101 */
1102 static void bdrv_temp_snapshot_options(int *child_flags, QDict *child_options,
1103 int parent_flags, QDict *parent_options)
1104 {
1105 *child_flags = (parent_flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
1106
1107 /* For temporary files, unconditional cache=unsafe is fine */
1108 qdict_set_default_str(child_options, BDRV_OPT_CACHE_DIRECT, "off");
1109 qdict_set_default_str(child_options, BDRV_OPT_CACHE_NO_FLUSH, "on");
1110
1111 /* Copy the read-only and discard options from the parent */
1112 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1113 qdict_copy_default(child_options, parent_options, BDRV_OPT_DISCARD);
1114
1115 /* aio=native doesn't work for cache.direct=off, so disable it for the
1116 * temporary snapshot */
1117 *child_flags &= ~BDRV_O_NATIVE_AIO;
1118 }
1119
1120 /*
1121 * Returns the options and flags that bs->file should get if a protocol driver
1122 * is expected, based on the given options and flags for the parent BDS
1123 */
1124 static void bdrv_inherited_options(int *child_flags, QDict *child_options,
1125 int parent_flags, QDict *parent_options)
1126 {
1127 int flags = parent_flags;
1128
1129 /* Enable protocol handling, disable format probing for bs->file */
1130 flags |= BDRV_O_PROTOCOL;
1131
1132 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
1133 * the parent. */
1134 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1135 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1136 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1137
1138 /* Inherit the read-only option from the parent if it's not set */
1139 qdict_copy_default(child_options, parent_options, BDRV_OPT_READ_ONLY);
1140 qdict_copy_default(child_options, parent_options, BDRV_OPT_AUTO_READ_ONLY);
1141
1142 /* Our block drivers take care to send flushes and respect unmap policy,
1143 * so we can default to enable both on lower layers regardless of the
1144 * corresponding parent options. */
1145 qdict_set_default_str(child_options, BDRV_OPT_DISCARD, "unmap");
1146
1147 /* Clear flags that only apply to the top layer */
1148 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ |
1149 BDRV_O_NO_IO);
1150
1151 *child_flags = flags;
1152 }
1153
1154 const BdrvChildRole child_file = {
1155 .parent_is_bds = true,
1156 .get_parent_desc = bdrv_child_get_parent_desc,
1157 .inherit_options = bdrv_inherited_options,
1158 .drained_begin = bdrv_child_cb_drained_begin,
1159 .drained_poll = bdrv_child_cb_drained_poll,
1160 .drained_end = bdrv_child_cb_drained_end,
1161 .attach = bdrv_child_cb_attach,
1162 .detach = bdrv_child_cb_detach,
1163 .inactivate = bdrv_child_cb_inactivate,
1164 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1165 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1166 };
1167
1168 /*
1169 * Returns the options and flags that bs->file should get if the use of formats
1170 * (and not only protocols) is permitted for it, based on the given options and
1171 * flags for the parent BDS
1172 */
1173 static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
1174 int parent_flags, QDict *parent_options)
1175 {
1176 child_file.inherit_options(child_flags, child_options,
1177 parent_flags, parent_options);
1178
1179 *child_flags &= ~(BDRV_O_PROTOCOL | BDRV_O_NO_IO);
1180 }
1181
1182 const BdrvChildRole child_format = {
1183 .parent_is_bds = true,
1184 .get_parent_desc = bdrv_child_get_parent_desc,
1185 .inherit_options = bdrv_inherited_fmt_options,
1186 .drained_begin = bdrv_child_cb_drained_begin,
1187 .drained_poll = bdrv_child_cb_drained_poll,
1188 .drained_end = bdrv_child_cb_drained_end,
1189 .attach = bdrv_child_cb_attach,
1190 .detach = bdrv_child_cb_detach,
1191 .inactivate = bdrv_child_cb_inactivate,
1192 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1193 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1194 };
1195
1196 static void bdrv_backing_attach(BdrvChild *c)
1197 {
1198 BlockDriverState *parent = c->opaque;
1199 BlockDriverState *backing_hd = c->bs;
1200
1201 assert(!parent->backing_blocker);
1202 error_setg(&parent->backing_blocker,
1203 "node is used as backing hd of '%s'",
1204 bdrv_get_device_or_node_name(parent));
1205
1206 bdrv_refresh_filename(backing_hd);
1207
1208 parent->open_flags &= ~BDRV_O_NO_BACKING;
1209 pstrcpy(parent->backing_file, sizeof(parent->backing_file),
1210 backing_hd->filename);
1211 pstrcpy(parent->backing_format, sizeof(parent->backing_format),
1212 backing_hd->drv ? backing_hd->drv->format_name : "");
1213
1214 bdrv_op_block_all(backing_hd, parent->backing_blocker);
1215 /* Otherwise we won't be able to commit or stream */
1216 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_COMMIT_TARGET,
1217 parent->backing_blocker);
1218 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_STREAM,
1219 parent->backing_blocker);
1220 /*
1221 * We do backup in 3 ways:
1222 * 1. drive backup
1223 * The target bs is new opened, and the source is top BDS
1224 * 2. blockdev backup
1225 * Both the source and the target are top BDSes.
1226 * 3. internal backup(used for block replication)
1227 * Both the source and the target are backing file
1228 *
1229 * In case 1 and 2, neither the source nor the target is the backing file.
1230 * In case 3, we will block the top BDS, so there is only one block job
1231 * for the top BDS and its backing chain.
1232 */
1233 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_SOURCE,
1234 parent->backing_blocker);
1235 bdrv_op_unblock(backing_hd, BLOCK_OP_TYPE_BACKUP_TARGET,
1236 parent->backing_blocker);
1237
1238 bdrv_child_cb_attach(c);
1239 }
1240
1241 static void bdrv_backing_detach(BdrvChild *c)
1242 {
1243 BlockDriverState *parent = c->opaque;
1244
1245 assert(parent->backing_blocker);
1246 bdrv_op_unblock_all(c->bs, parent->backing_blocker);
1247 error_free(parent->backing_blocker);
1248 parent->backing_blocker = NULL;
1249
1250 bdrv_child_cb_detach(c);
1251 }
1252
1253 /*
1254 * Returns the options and flags that bs->backing should get, based on the
1255 * given options and flags for the parent BDS
1256 */
1257 static void bdrv_backing_options(int *child_flags, QDict *child_options,
1258 int parent_flags, QDict *parent_options)
1259 {
1260 int flags = parent_flags;
1261
1262 /* The cache mode is inherited unmodified for backing files; except WCE,
1263 * which is only applied on the top level (BlockBackend) */
1264 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
1265 qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
1266 qdict_copy_default(child_options, parent_options, BDRV_OPT_FORCE_SHARE);
1267
1268 /* backing files always opened read-only */
1269 qdict_set_default_str(child_options, BDRV_OPT_READ_ONLY, "on");
1270 qdict_set_default_str(child_options, BDRV_OPT_AUTO_READ_ONLY, "off");
1271 flags &= ~BDRV_O_COPY_ON_READ;
1272
1273 /* snapshot=on is handled on the top layer */
1274 flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
1275
1276 *child_flags = flags;
1277 }
1278
1279 static int bdrv_backing_update_filename(BdrvChild *c, BlockDriverState *base,
1280 const char *filename, Error **errp)
1281 {
1282 BlockDriverState *parent = c->opaque;
1283 bool read_only = bdrv_is_read_only(parent);
1284 int ret;
1285
1286 if (read_only) {
1287 ret = bdrv_reopen_set_read_only(parent, false, errp);
1288 if (ret < 0) {
1289 return ret;
1290 }
1291 }
1292
1293 ret = bdrv_change_backing_file(parent, filename,
1294 base->drv ? base->drv->format_name : "");
1295 if (ret < 0) {
1296 error_setg_errno(errp, -ret, "Could not update backing file link");
1297 }
1298
1299 if (read_only) {
1300 bdrv_reopen_set_read_only(parent, true, NULL);
1301 }
1302
1303 return ret;
1304 }
1305
1306 const BdrvChildRole child_backing = {
1307 .parent_is_bds = true,
1308 .get_parent_desc = bdrv_child_get_parent_desc,
1309 .attach = bdrv_backing_attach,
1310 .detach = bdrv_backing_detach,
1311 .inherit_options = bdrv_backing_options,
1312 .drained_begin = bdrv_child_cb_drained_begin,
1313 .drained_poll = bdrv_child_cb_drained_poll,
1314 .drained_end = bdrv_child_cb_drained_end,
1315 .inactivate = bdrv_child_cb_inactivate,
1316 .update_filename = bdrv_backing_update_filename,
1317 .can_set_aio_ctx = bdrv_child_cb_can_set_aio_ctx,
1318 .set_aio_ctx = bdrv_child_cb_set_aio_ctx,
1319 };
1320
1321 static int bdrv_open_flags(BlockDriverState *bs, int flags)
1322 {
1323 int open_flags = flags;
1324
1325 /*
1326 * Clear flags that are internal to the block layer before opening the
1327 * image.
1328 */
1329 open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_PROTOCOL);
1330
1331 return open_flags;
1332 }
1333
1334 static void update_flags_from_options(int *flags, QemuOpts *opts)
1335 {
1336 *flags &= ~(BDRV_O_CACHE_MASK | BDRV_O_RDWR | BDRV_O_AUTO_RDONLY);
1337
1338 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
1339 *flags |= BDRV_O_NO_FLUSH;
1340 }
1341
1342 if (qemu_opt_get_bool_del(opts, BDRV_OPT_CACHE_DIRECT, false)) {
1343 *flags |= BDRV_O_NOCACHE;
1344 }
1345
1346 if (!qemu_opt_get_bool_del(opts, BDRV_OPT_READ_ONLY, false)) {
1347 *flags |= BDRV_O_RDWR;
1348 }
1349
1350 if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
1351 *flags |= BDRV_O_AUTO_RDONLY;
1352 }
1353 }
1354
1355 static void update_options_from_flags(QDict *options, int flags)
1356 {
1357 if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
1358 qdict_put_bool(options, BDRV_OPT_CACHE_DIRECT, flags & BDRV_O_NOCACHE);
1359 }
1360 if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
1361 qdict_put_bool(options, BDRV_OPT_CACHE_NO_FLUSH,
1362 flags & BDRV_O_NO_FLUSH);
1363 }
1364 if (!qdict_haskey(options, BDRV_OPT_READ_ONLY)) {
1365 qdict_put_bool(options, BDRV_OPT_READ_ONLY, !(flags & BDRV_O_RDWR));
1366 }
1367 if (!qdict_haskey(options, BDRV_OPT_AUTO_READ_ONLY)) {
1368 qdict_put_bool(options, BDRV_OPT_AUTO_READ_ONLY,
1369 flags & BDRV_O_AUTO_RDONLY);
1370 }
1371 }
1372
1373 static void bdrv_assign_node_name(BlockDriverState *bs,
1374 const char *node_name,
1375 Error **errp)
1376 {
1377 char *gen_node_name = NULL;
1378
1379 if (!node_name) {
1380 node_name = gen_node_name = id_generate(ID_BLOCK);
1381 } else if (!id_wellformed(node_name)) {
1382 /*
1383 * Check for empty string or invalid characters, but not if it is
1384 * generated (generated names use characters not available to the user)
1385 */
1386 error_setg(errp, "Invalid node name");
1387 return;
1388 }
1389
1390 /* takes care of avoiding namespaces collisions */
1391 if (blk_by_name(node_name)) {
1392 error_setg(errp, "node-name=%s is conflicting with a device id",
1393 node_name);
1394 goto out;
1395 }
1396
1397 /* takes care of avoiding duplicates node names */
1398 if (bdrv_find_node(node_name)) {
1399 error_setg(errp, "Duplicate node name");
1400 goto out;
1401 }
1402
1403 /* Make sure that the node name isn't truncated */
1404 if (strlen(node_name) >= sizeof(bs->node_name)) {
1405 error_setg(errp, "Node name too long");
1406 goto out;
1407 }
1408
1409 /* copy node name into the bs and insert it into the graph list */
1410 pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
1411 QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
1412 out:
1413 g_free(gen_node_name);
1414 }
1415
1416 static int bdrv_open_driver(BlockDriverState *bs, BlockDriver *drv,
1417 const char *node_name, QDict *options,
1418 int open_flags, Error **errp)
1419 {
1420 Error *local_err = NULL;
1421 int i, ret;
1422
1423 bdrv_assign_node_name(bs, node_name, &local_err);
1424 if (local_err) {
1425 error_propagate(errp, local_err);
1426 return -EINVAL;
1427 }
1428
1429 bs->drv = drv;
1430 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1431 bs->opaque = g_malloc0(drv->instance_size);
1432
1433 if (drv->bdrv_file_open) {
1434 assert(!drv->bdrv_needs_filename || bs->filename[0]);
1435 ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
1436 } else if (drv->bdrv_open) {
1437 ret = drv->bdrv_open(bs, options, open_flags, &local_err);
1438 } else {
1439 ret = 0;
1440 }
1441
1442 if (ret < 0) {
1443 if (local_err) {
1444 error_propagate(errp, local_err);
1445 } else if (bs->filename[0]) {
1446 error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
1447 } else {
1448 error_setg_errno(errp, -ret, "Could not open image");
1449 }
1450 goto open_failed;
1451 }
1452
1453 ret = refresh_total_sectors(bs, bs->total_sectors);
1454 if (ret < 0) {
1455 error_setg_errno(errp, -ret, "Could not refresh total sector count");
1456 return ret;
1457 }
1458
1459 bdrv_refresh_limits(bs, &local_err);
1460 if (local_err) {
1461 error_propagate(errp, local_err);
1462 return -EINVAL;
1463 }
1464
1465 assert(bdrv_opt_mem_align(bs) != 0);
1466 assert(bdrv_min_mem_align(bs) != 0);
1467 assert(is_power_of_2(bs->bl.request_alignment));
1468
1469 for (i = 0; i < bs->quiesce_counter; i++) {
1470 if (drv->bdrv_co_drain_begin) {
1471 drv->bdrv_co_drain_begin(bs);
1472 }
1473 }
1474
1475 return 0;
1476 open_failed:
1477 bs->drv = NULL;
1478 if (bs->file != NULL) {
1479 bdrv_unref_child(bs, bs->file);
1480 bs->file = NULL;
1481 }
1482 g_free(bs->opaque);
1483 bs->opaque = NULL;
1484 return ret;
1485 }
1486
1487 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
1488 int flags, Error **errp)
1489 {
1490 BlockDriverState *bs;
1491 int ret;
1492
1493 bs = bdrv_new();
1494 bs->open_flags = flags;
1495 bs->explicit_options = qdict_new();
1496 bs->options = qdict_new();
1497 bs->opaque = NULL;
1498
1499 update_options_from_flags(bs->options, flags);
1500
1501 ret = bdrv_open_driver(bs, drv, node_name, bs->options, flags, errp);
1502 if (ret < 0) {
1503 qobject_unref(bs->explicit_options);
1504 bs->explicit_options = NULL;
1505 qobject_unref(bs->options);
1506 bs->options = NULL;
1507 bdrv_unref(bs);
1508 return NULL;
1509 }
1510
1511 return bs;
1512 }
1513
1514 QemuOptsList bdrv_runtime_opts = {
1515 .name = "bdrv_common",
1516 .head = QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts.head),
1517 .desc = {
1518 {
1519 .name = "node-name",
1520 .type = QEMU_OPT_STRING,
1521 .help = "Node name of the block device node",
1522 },
1523 {
1524 .name = "driver",
1525 .type = QEMU_OPT_STRING,
1526 .help = "Block driver to use for the node",
1527 },
1528 {
1529 .name = BDRV_OPT_CACHE_DIRECT,
1530 .type = QEMU_OPT_BOOL,
1531 .help = "Bypass software writeback cache on the host",
1532 },
1533 {
1534 .name = BDRV_OPT_CACHE_NO_FLUSH,
1535 .type = QEMU_OPT_BOOL,
1536 .help = "Ignore flush requests",
1537 },
1538 {
1539 .name = BDRV_OPT_READ_ONLY,
1540 .type = QEMU_OPT_BOOL,
1541 .help = "Node is opened in read-only mode",
1542 },
1543 {
1544 .name = BDRV_OPT_AUTO_READ_ONLY,
1545 .type = QEMU_OPT_BOOL,
1546 .help = "Node can become read-only if opening read-write fails",
1547 },
1548 {
1549 .name = "detect-zeroes",
1550 .type = QEMU_OPT_STRING,
1551 .help = "try to optimize zero writes (off, on, unmap)",
1552 },
1553 {
1554 .name = BDRV_OPT_DISCARD,
1555 .type = QEMU_OPT_STRING,
1556 .help = "discard operation (ignore/off, unmap/on)",
1557 },
1558 {
1559 .name = BDRV_OPT_FORCE_SHARE,
1560 .type = QEMU_OPT_BOOL,
1561 .help = "always accept other writers (default: off)",
1562 },
1563 { /* end of list */ }
1564 },
1565 };
1566
1567 static QemuOptsList fallback_create_opts = {
1568 .name = "fallback-create-opts",
1569 .head = QTAILQ_HEAD_INITIALIZER(fallback_create_opts.head),
1570 .desc = {
1571 {
1572 .name = BLOCK_OPT_SIZE,
1573 .type = QEMU_OPT_SIZE,
1574 .help = "Virtual disk size"
1575 },
1576 {
1577 .name = BLOCK_OPT_PREALLOC,
1578 .type = QEMU_OPT_STRING,
1579 .help = "Preallocation mode (allowed values: off)"
1580 },
1581 { /* end of list */ }
1582 }
1583 };
1584
1585 /*
1586 * Common part for opening disk images and files
1587 *
1588 * Removes all processed options from *options.
1589 */
1590 static int bdrv_open_common(BlockDriverState *bs, BlockBackend *file,
1591 QDict *options, Error **errp)
1592 {
1593 int ret, open_flags;
1594 const char *filename;
1595 const char *driver_name = NULL;
1596 const char *node_name = NULL;
1597 const char *discard;
1598 QemuOpts *opts;
1599 BlockDriver *drv;
1600 Error *local_err = NULL;
1601
1602 assert(bs->file == NULL);
1603 assert(options != NULL && bs->options != options);
1604
1605 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
1606 qemu_opts_absorb_qdict(opts, options, &local_err);
1607 if (local_err) {
1608 error_propagate(errp, local_err);
1609 ret = -EINVAL;
1610 goto fail_opts;
1611 }
1612
1613 update_flags_from_options(&bs->open_flags, opts);
1614
1615 driver_name = qemu_opt_get(opts, "driver");
1616 drv = bdrv_find_format(driver_name);
1617 assert(drv != NULL);
1618
1619 bs->force_share = qemu_opt_get_bool(opts, BDRV_OPT_FORCE_SHARE, false);
1620
1621 if (bs->force_share && (bs->open_flags & BDRV_O_RDWR)) {
1622 error_setg(errp,
1623 BDRV_OPT_FORCE_SHARE
1624 "=on can only be used with read-only images");
1625 ret = -EINVAL;
1626 goto fail_opts;
1627 }
1628
1629 if (file != NULL) {
1630 bdrv_refresh_filename(blk_bs(file));
1631 filename = blk_bs(file)->filename;
1632 } else {
1633 /*
1634 * Caution: while qdict_get_try_str() is fine, getting
1635 * non-string types would require more care. When @options
1636 * come from -blockdev or blockdev_add, its members are typed
1637 * according to the QAPI schema, but when they come from
1638 * -drive, they're all QString.
1639 */
1640 filename = qdict_get_try_str(options, "filename");
1641 }
1642
1643 if (drv->bdrv_needs_filename && (!filename || !filename[0])) {
1644 error_setg(errp, "The '%s' block driver requires a file name",
1645 drv->format_name);
1646 ret = -EINVAL;
1647 goto fail_opts;
1648 }
1649
1650 trace_bdrv_open_common(bs, filename ?: "", bs->open_flags,
1651 drv->format_name);
1652
1653 bs->read_only = !(bs->open_flags & BDRV_O_RDWR);
1654
1655 if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
1656 if (!bs->read_only && bdrv_is_whitelisted(drv, true)) {
1657 ret = bdrv_apply_auto_read_only(bs, NULL, NULL);
1658 } else {
1659 ret = -ENOTSUP;
1660 }
1661 if (ret < 0) {
1662 error_setg(errp,
1663 !bs->read_only && bdrv_is_whitelisted(drv, true)
1664 ? "Driver '%s' can only be used for read-only devices"
1665 : "Driver '%s' is not whitelisted",
1666 drv->format_name);
1667 goto fail_opts;
1668 }
1669 }
1670
1671 /* bdrv_new() and bdrv_close() make it so */
1672 assert(atomic_read(&bs->copy_on_read) == 0);
1673
1674 if (bs->open_flags & BDRV_O_COPY_ON_READ) {
1675 if (!bs->read_only) {
1676 bdrv_enable_copy_on_read(bs);
1677 } else {
1678 error_setg(errp, "Can't use copy-on-read on read-only device");
1679 ret = -EINVAL;
1680 goto fail_opts;
1681 }
1682 }
1683
1684 discard = qemu_opt_get(opts, BDRV_OPT_DISCARD);
1685 if (discard != NULL) {
1686 if (bdrv_parse_discard_flags(discard, &bs->open_flags) != 0) {
1687 error_setg(errp, "Invalid discard option");
1688 ret = -EINVAL;
1689 goto fail_opts;
1690 }
1691 }
1692
1693 bs->detect_zeroes =
1694 bdrv_parse_detect_zeroes(opts, bs->open_flags, &local_err);
1695 if (local_err) {
1696 error_propagate(errp, local_err);
1697 ret = -EINVAL;
1698 goto fail_opts;
1699 }
1700
1701 if (filename != NULL) {
1702 pstrcpy(bs->filename, sizeof(bs->filename), filename);
1703 } else {
1704 bs->filename[0] = '\0';
1705 }
1706 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), bs->filename);
1707
1708 /* Open the image, either directly or using a protocol */
1709 open_flags = bdrv_open_flags(bs, bs->open_flags);
1710 node_name = qemu_opt_get(opts, "node-name");
1711
1712 assert(!drv->bdrv_file_open || file == NULL);
1713 ret = bdrv_open_driver(bs, drv, node_name, options, open_flags, errp);
1714 if (ret < 0) {
1715 goto fail_opts;
1716 }
1717
1718 qemu_opts_del(opts);
1719 return 0;
1720
1721 fail_opts:
1722 qemu_opts_del(opts);
1723 return ret;
1724 }
1725
1726 static QDict *parse_json_filename(const char *filename, Error **errp)
1727 {
1728 QObject *options_obj;
1729 QDict *options;
1730 int ret;
1731
1732 ret = strstart(filename, "json:", &filename);
1733 assert(ret);
1734
1735 options_obj = qobject_from_json(filename, errp);
1736 if (!options_obj) {
1737 error_prepend(errp, "Could not parse the JSON options: ");
1738 return NULL;
1739 }
1740
1741 options = qobject_to(QDict, options_obj);
1742 if (!options) {
1743 qobject_unref(options_obj);
1744 error_setg(errp, "Invalid JSON object given");
1745 return NULL;
1746 }
1747
1748 qdict_flatten(options);
1749
1750 return options;
1751 }
1752
1753 static void parse_json_protocol(QDict *options, const char **pfilename,
1754 Error **errp)
1755 {
1756 QDict *json_options;
1757 Error *local_err = NULL;
1758
1759 /* Parse json: pseudo-protocol */
1760 if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
1761 return;
1762 }
1763
1764 json_options = parse_json_filename(*pfilename, &local_err);
1765 if (local_err) {
1766 error_propagate(errp, local_err);
1767 return;
1768 }
1769
1770 /* Options given in the filename have lower priority than options
1771 * specified directly */
1772 qdict_join(options, json_options, false);
1773 qobject_unref(json_options);
1774 *pfilename = NULL;
1775 }
1776
1777 /*
1778 * Fills in default options for opening images and converts the legacy
1779 * filename/flags pair to option QDict entries.
1780 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1781 * block driver has been specified explicitly.
1782 */
1783 static int bdrv_fill_options(QDict **options, const char *filename,
1784 int *flags, Error **errp)
1785 {
1786 const char *drvname;
1787 bool protocol = *flags & BDRV_O_PROTOCOL;
1788 bool parse_filename = false;
1789 BlockDriver *drv = NULL;
1790 Error *local_err = NULL;
1791
1792 /*
1793 * Caution: while qdict_get_try_str() is fine, getting non-string
1794 * types would require more care. When @options come from
1795 * -blockdev or blockdev_add, its members are typed according to
1796 * the QAPI schema, but when they come from -drive, they're all
1797 * QString.
1798 */
1799 drvname = qdict_get_try_str(*options, "driver");
1800 if (drvname) {
1801 drv = bdrv_find_format(drvname);
1802 if (!drv) {
1803 error_setg(errp, "Unknown driver '%s'", drvname);
1804 return -ENOENT;
1805 }
1806 /* If the user has explicitly specified the driver, this choice should
1807 * override the BDRV_O_PROTOCOL flag */
1808 protocol = drv->bdrv_file_open;
1809 }
1810
1811 if (protocol) {
1812 *flags |= BDRV_O_PROTOCOL;
1813 } else {
1814 *flags &= ~BDRV_O_PROTOCOL;
1815 }
1816
1817 /* Translate cache options from flags into options */
1818 update_options_from_flags(*options, *flags);
1819
1820 /* Fetch the file name from the options QDict if necessary */
1821 if (protocol && filename) {
1822 if (!qdict_haskey(*options, "filename")) {
1823 qdict_put_str(*options, "filename", filename);
1824 parse_filename = true;
1825 } else {
1826 error_setg(errp, "Can't specify 'file' and 'filename' options at "
1827 "the same time");
1828 return -EINVAL;
1829 }
1830 }
1831
1832 /* Find the right block driver */
1833 /* See cautionary note on accessing @options above */
1834 filename = qdict_get_try_str(*options, "filename");
1835
1836 if (!drvname && protocol) {
1837 if (filename) {
1838 drv = bdrv_find_protocol(filename, parse_filename, errp);
1839 if (!drv) {
1840 return -EINVAL;
1841 }
1842
1843 drvname = drv->format_name;
1844 qdict_put_str(*options, "driver", drvname);
1845 } else {
1846 error_setg(errp, "Must specify either driver or file");
1847 return -EINVAL;
1848 }
1849 }
1850
1851 assert(drv || !protocol);
1852
1853 /* Driver-specific filename parsing */
1854 if (drv && drv->bdrv_parse_filename && parse_filename) {
1855 drv->bdrv_parse_filename(filename, *options, &local_err);
1856 if (local_err) {
1857 error_propagate(errp, local_err);
1858 return -EINVAL;
1859 }
1860
1861 if (!drv->bdrv_needs_filename) {
1862 qdict_del(*options, "filename");
1863 }
1864 }
1865
1866 return 0;
1867 }
1868
1869 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
1870 uint64_t perm, uint64_t shared,
1871 GSList *ignore_children,
1872 bool *tighten_restrictions, Error **errp);
1873 static void bdrv_child_abort_perm_update(BdrvChild *c);
1874 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared);
1875 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1876 uint64_t *shared_perm);
1877
1878 typedef struct BlockReopenQueueEntry {
1879 bool prepared;
1880 bool perms_checked;
1881 BDRVReopenState state;
1882 QTAILQ_ENTRY(BlockReopenQueueEntry) entry;
1883 } BlockReopenQueueEntry;
1884
1885 /*
1886 * Return the flags that @bs will have after the reopens in @q have
1887 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1888 * return the current flags.
1889 */
1890 static int bdrv_reopen_get_flags(BlockReopenQueue *q, BlockDriverState *bs)
1891 {
1892 BlockReopenQueueEntry *entry;
1893
1894 if (q != NULL) {
1895 QTAILQ_FOREACH(entry, q, entry) {
1896 if (entry->state.bs == bs) {
1897 return entry->state.flags;
1898 }
1899 }
1900 }
1901
1902 return bs->open_flags;
1903 }
1904
1905 /* Returns whether the image file can be written to after the reopen queue @q
1906 * has been successfully applied, or right now if @q is NULL. */
1907 static bool bdrv_is_writable_after_reopen(BlockDriverState *bs,
1908 BlockReopenQueue *q)
1909 {
1910 int flags = bdrv_reopen_get_flags(q, bs);
1911
1912 return (flags & (BDRV_O_RDWR | BDRV_O_INACTIVE)) == BDRV_O_RDWR;
1913 }
1914
1915 /*
1916 * Return whether the BDS can be written to. This is not necessarily
1917 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1918 * be written to but do not count as read-only images.
1919 */
1920 bool bdrv_is_writable(BlockDriverState *bs)
1921 {
1922 return bdrv_is_writable_after_reopen(bs, NULL);
1923 }
1924
1925 static void bdrv_child_perm(BlockDriverState *bs, BlockDriverState *child_bs,
1926 BdrvChild *c, const BdrvChildRole *role,
1927 BlockReopenQueue *reopen_queue,
1928 uint64_t parent_perm, uint64_t parent_shared,
1929 uint64_t *nperm, uint64_t *nshared)
1930 {
1931 assert(bs->drv && bs->drv->bdrv_child_perm);
1932 bs->drv->bdrv_child_perm(bs, c, role, reopen_queue,
1933 parent_perm, parent_shared,
1934 nperm, nshared);
1935 /* TODO Take force_share from reopen_queue */
1936 if (child_bs && child_bs->force_share) {
1937 *nshared = BLK_PERM_ALL;
1938 }
1939 }
1940
1941 /*
1942 * Check whether permissions on this node can be changed in a way that
1943 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1944 * permissions of all its parents. This involves checking whether all necessary
1945 * permission changes to child nodes can be performed.
1946 *
1947 * Will set *tighten_restrictions to true if and only if new permissions have to
1948 * be taken or currently shared permissions are to be unshared. Otherwise,
1949 * errors are not fatal as long as the caller accepts that the restrictions
1950 * remain tighter than they need to be. The caller still has to abort the
1951 * transaction.
1952 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1953 * encounter fatal errors even though no restrictions are to be tightened. For
1954 * example, changing a node from RW to RO will fail if the WRITE permission is
1955 * to be kept.
1956 *
1957 * A call to this function must always be followed by a call to bdrv_set_perm()
1958 * or bdrv_abort_perm_update().
1959 */
1960 static int bdrv_check_perm(BlockDriverState *bs, BlockReopenQueue *q,
1961 uint64_t cumulative_perms,
1962 uint64_t cumulative_shared_perms,
1963 GSList *ignore_children,
1964 bool *tighten_restrictions, Error **errp)
1965 {
1966 BlockDriver *drv = bs->drv;
1967 BdrvChild *c;
1968 int ret;
1969
1970 assert(!q || !tighten_restrictions);
1971
1972 if (tighten_restrictions) {
1973 uint64_t current_perms, current_shared;
1974 uint64_t added_perms, removed_shared_perms;
1975
1976 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
1977
1978 added_perms = cumulative_perms & ~current_perms;
1979 removed_shared_perms = current_shared & ~cumulative_shared_perms;
1980
1981 *tighten_restrictions = added_perms || removed_shared_perms;
1982 }
1983
1984 /* Write permissions never work with read-only images */
1985 if ((cumulative_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) &&
1986 !bdrv_is_writable_after_reopen(bs, q))
1987 {
1988 if (!bdrv_is_writable_after_reopen(bs, NULL)) {
1989 error_setg(errp, "Block node is read-only");
1990 } else {
1991 uint64_t current_perms, current_shared;
1992 bdrv_get_cumulative_perm(bs, &current_perms, &current_shared);
1993 if (current_perms & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) {
1994 error_setg(errp, "Cannot make block node read-only, there is "
1995 "a writer on it");
1996 } else {
1997 error_setg(errp, "Cannot make block node read-only and create "
1998 "a writer on it");
1999 }
2000 }
2001
2002 return -EPERM;
2003 }
2004
2005 /* Check this node */
2006 if (!drv) {
2007 return 0;
2008 }
2009
2010 if (drv->bdrv_check_perm) {
2011 return drv->bdrv_check_perm(bs, cumulative_perms,
2012 cumulative_shared_perms, errp);
2013 }
2014
2015 /* Drivers that never have children can omit .bdrv_child_perm() */
2016 if (!drv->bdrv_child_perm) {
2017 assert(QLIST_EMPTY(&bs->children));
2018 return 0;
2019 }
2020
2021 /* Check all children */
2022 QLIST_FOREACH(c, &bs->children, next) {
2023 uint64_t cur_perm, cur_shared;
2024 bool child_tighten_restr;
2025
2026 bdrv_child_perm(bs, c->bs, c, c->role, q,
2027 cumulative_perms, cumulative_shared_perms,
2028 &cur_perm, &cur_shared);
2029 ret = bdrv_child_check_perm(c, q, cur_perm, cur_shared, ignore_children,
2030 tighten_restrictions ? &child_tighten_restr
2031 : NULL,
2032 errp);
2033 if (tighten_restrictions) {
2034 *tighten_restrictions |= child_tighten_restr;
2035 }
2036 if (ret < 0) {
2037 return ret;
2038 }
2039 }
2040
2041 return 0;
2042 }
2043
2044 /*
2045 * Notifies drivers that after a previous bdrv_check_perm() call, the
2046 * permission update is not performed and any preparations made for it (e.g.
2047 * taken file locks) need to be undone.
2048 *
2049 * This function recursively notifies all child nodes.
2050 */
2051 static void bdrv_abort_perm_update(BlockDriverState *bs)
2052 {
2053 BlockDriver *drv = bs->drv;
2054 BdrvChild *c;
2055
2056 if (!drv) {
2057 return;
2058 }
2059
2060 if (drv->bdrv_abort_perm_update) {
2061 drv->bdrv_abort_perm_update(bs);
2062 }
2063
2064 QLIST_FOREACH(c, &bs->children, next) {
2065 bdrv_child_abort_perm_update(c);
2066 }
2067 }
2068
2069 static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms,
2070 uint64_t cumulative_shared_perms)
2071 {
2072 BlockDriver *drv = bs->drv;
2073 BdrvChild *c;
2074
2075 if (!drv) {
2076 return;
2077 }
2078
2079 /* Update this node */
2080 if (drv->bdrv_set_perm) {
2081 drv->bdrv_set_perm(bs, cumulative_perms, cumulative_shared_perms);
2082 }
2083
2084 /* Drivers that never have children can omit .bdrv_child_perm() */
2085 if (!drv->bdrv_child_perm) {
2086 assert(QLIST_EMPTY(&bs->children));
2087 return;
2088 }
2089
2090 /* Update all children */
2091 QLIST_FOREACH(c, &bs->children, next) {
2092 uint64_t cur_perm, cur_shared;
2093 bdrv_child_perm(bs, c->bs, c, c->role, NULL,
2094 cumulative_perms, cumulative_shared_perms,
2095 &cur_perm, &cur_shared);
2096 bdrv_child_set_perm(c, cur_perm, cur_shared);
2097 }
2098 }
2099
2100 static void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
2101 uint64_t *shared_perm)
2102 {
2103 BdrvChild *c;
2104 uint64_t cumulative_perms = 0;
2105 uint64_t cumulative_shared_perms = BLK_PERM_ALL;
2106
2107 QLIST_FOREACH(c, &bs->parents, next_parent) {
2108 cumulative_perms |= c->perm;
2109 cumulative_shared_perms &= c->shared_perm;
2110 }
2111
2112 *perm = cumulative_perms;
2113 *shared_perm = cumulative_shared_perms;
2114 }
2115
2116 static char *bdrv_child_user_desc(BdrvChild *c)
2117 {
2118 if (c->role->get_parent_desc) {
2119 return c->role->get_parent_desc(c);
2120 }
2121
2122 return g_strdup("another user");
2123 }
2124
2125 char *bdrv_perm_names(uint64_t perm)
2126 {
2127 struct perm_name {
2128 uint64_t perm;
2129 const char *name;
2130 } permissions[] = {
2131 { BLK_PERM_CONSISTENT_READ, "consistent read" },
2132 { BLK_PERM_WRITE, "write" },
2133 { BLK_PERM_WRITE_UNCHANGED, "write unchanged" },
2134 { BLK_PERM_RESIZE, "resize" },
2135 { BLK_PERM_GRAPH_MOD, "change children" },
2136 { 0, NULL }
2137 };
2138
2139 GString *result = g_string_sized_new(30);
2140 struct perm_name *p;
2141
2142 for (p = permissions; p->name; p++) {
2143 if (perm & p->perm) {
2144 if (result->len > 0) {
2145 g_string_append(result, ", ");
2146 }
2147 g_string_append(result, p->name);
2148 }
2149 }
2150
2151 return g_string_free(result, FALSE);
2152 }
2153
2154 /*
2155 * Checks whether a new reference to @bs can be added if the new user requires
2156 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2157 * set, the BdrvChild objects in this list are ignored in the calculations;
2158 * this allows checking permission updates for an existing reference.
2159 *
2160 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2161 *
2162 * Needs to be followed by a call to either bdrv_set_perm() or
2163 * bdrv_abort_perm_update(). */
2164 static int bdrv_check_update_perm(BlockDriverState *bs, BlockReopenQueue *q,
2165 uint64_t new_used_perm,
2166 uint64_t new_shared_perm,
2167 GSList *ignore_children,
2168 bool *tighten_restrictions,
2169 Error **errp)
2170 {
2171 BdrvChild *c;
2172 uint64_t cumulative_perms = new_used_perm;
2173 uint64_t cumulative_shared_perms = new_shared_perm;
2174
2175 assert(!q || !tighten_restrictions);
2176
2177 /* There is no reason why anyone couldn't tolerate write_unchanged */
2178 assert(new_shared_perm & BLK_PERM_WRITE_UNCHANGED);
2179
2180 QLIST_FOREACH(c, &bs->parents, next_parent) {
2181 if (g_slist_find(ignore_children, c)) {
2182 continue;
2183 }
2184
2185 if ((new_used_perm & c->shared_perm) != new_used_perm) {
2186 char *user = bdrv_child_user_desc(c);
2187 char *perm_names = bdrv_perm_names(new_used_perm & ~c->shared_perm);
2188
2189 if (tighten_restrictions) {
2190 *tighten_restrictions = true;
2191 }
2192
2193 error_setg(errp, "Conflicts with use by %s as '%s', which does not "
2194 "allow '%s' on %s",
2195 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2196 g_free(user);
2197 g_free(perm_names);
2198 return -EPERM;
2199 }
2200
2201 if ((c->perm & new_shared_perm) != c->perm) {
2202 char *user = bdrv_child_user_desc(c);
2203 char *perm_names = bdrv_perm_names(c->perm & ~new_shared_perm);
2204
2205 if (tighten_restrictions) {
2206 *tighten_restrictions = true;
2207 }
2208
2209 error_setg(errp, "Conflicts with use by %s as '%s', which uses "
2210 "'%s' on %s",
2211 user, c->name, perm_names, bdrv_get_node_name(c->bs));
2212 g_free(user);
2213 g_free(perm_names);
2214 return -EPERM;
2215 }
2216
2217 cumulative_perms |= c->perm;
2218 cumulative_shared_perms &= c->shared_perm;
2219 }
2220
2221 return bdrv_check_perm(bs, q, cumulative_perms, cumulative_shared_perms,
2222 ignore_children, tighten_restrictions, errp);
2223 }
2224
2225 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2226 * bdrv_child_abort_perm_update(). */
2227 static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q,
2228 uint64_t perm, uint64_t shared,
2229 GSList *ignore_children,
2230 bool *tighten_restrictions, Error **errp)
2231 {
2232 int ret;
2233
2234 ignore_children = g_slist_prepend(g_slist_copy(ignore_children), c);
2235 ret = bdrv_check_update_perm(c->bs, q, perm, shared, ignore_children,
2236 tighten_restrictions, errp);
2237 g_slist_free(ignore_children);
2238
2239 if (ret < 0) {
2240 return ret;
2241 }
2242
2243 if (!c->has_backup_perm) {
2244 c->has_backup_perm = true;
2245 c->backup_perm = c->perm;
2246 c->backup_shared_perm = c->shared_perm;
2247 }
2248 /*
2249 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2250 * same child twice during check_perm procedure
2251 */
2252
2253 c->perm = perm;
2254 c->shared_perm = shared;
2255
2256 return 0;
2257 }
2258
2259 static void bdrv_child_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared)
2260 {
2261 uint64_t cumulative_perms, cumulative_shared_perms;
2262
2263 c->has_backup_perm = false;
2264
2265 c->perm = perm;
2266 c->shared_perm = shared;
2267
2268 bdrv_get_cumulative_perm(c->bs, &cumulative_perms,
2269 &cumulative_shared_perms);
2270 bdrv_set_perm(c->bs, cumulative_perms, cumulative_shared_perms);
2271 }
2272
2273 static void bdrv_child_abort_perm_update(BdrvChild *c)
2274 {
2275 if (c->has_backup_perm) {
2276 c->perm = c->backup_perm;
2277 c->shared_perm = c->backup_shared_perm;
2278 c->has_backup_perm = false;
2279 }
2280
2281 bdrv_abort_perm_update(c->bs);
2282 }
2283
2284 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
2285 Error **errp)
2286 {
2287 Error *local_err = NULL;
2288 int ret;
2289 bool tighten_restrictions;
2290
2291 ret = bdrv_child_check_perm(c, NULL, perm, shared, NULL,
2292 &tighten_restrictions, &local_err);
2293 if (ret < 0) {
2294 bdrv_child_abort_perm_update(c);
2295 if (tighten_restrictions) {
2296 error_propagate(errp, local_err);
2297 } else {
2298 /*
2299 * Our caller may intend to only loosen restrictions and
2300 * does not expect this function to fail. Errors are not
2301 * fatal in such a case, so we can just hide them from our
2302 * caller.
2303 */
2304 error_free(local_err);
2305 ret = 0;
2306 }
2307 return ret;
2308 }
2309
2310 bdrv_child_set_perm(c, perm, shared);
2311
2312 return 0;
2313 }
2314
2315 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp)
2316 {
2317 uint64_t parent_perms, parent_shared;
2318 uint64_t perms, shared;
2319
2320 bdrv_get_cumulative_perm(bs, &parent_perms, &parent_shared);
2321 bdrv_child_perm(bs, c->bs, c, c->role, NULL, parent_perms, parent_shared,
2322 &perms, &shared);
2323
2324 return bdrv_child_try_set_perm(c, perms, shared, errp);
2325 }
2326
2327 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
2328 const BdrvChildRole *role,
2329 BlockReopenQueue *reopen_queue,
2330 uint64_t perm, uint64_t shared,
2331 uint64_t *nperm, uint64_t *nshared)
2332 {
2333 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
2334 *nshared = (shared & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
2335 }
2336
2337 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
2338 const BdrvChildRole *role,
2339 BlockReopenQueue *reopen_queue,
2340 uint64_t perm, uint64_t shared,
2341 uint64_t *nperm, uint64_t *nshared)
2342 {
2343 bool backing = (role == &child_backing);
2344 assert(role == &child_backing || role == &child_file);
2345
2346 if (!backing) {
2347 int flags = bdrv_reopen_get_flags(reopen_queue, bs);
2348
2349 /* Apart from the modifications below, the same permissions are
2350 * forwarded and left alone as for filters */
2351 bdrv_filter_default_perms(bs, c, role, reopen_queue, perm, shared,
2352 &perm, &shared);
2353
2354 /* Format drivers may touch metadata even if the guest doesn't write */
2355 if (bdrv_is_writable_after_reopen(bs, reopen_queue)) {
2356 perm |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2357 }
2358
2359 /* bs->file always needs to be consistent because of the metadata. We
2360 * can never allow other users to resize or write to it. */
2361 if (!(flags & BDRV_O_NO_IO)) {
2362 perm |= BLK_PERM_CONSISTENT_READ;
2363 }
2364 shared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
2365 } else {
2366 /* We want consistent read from backing files if the parent needs it.
2367 * No other operations are performed on backing files. */
2368 perm &= BLK_PERM_CONSISTENT_READ;
2369
2370 /* If the parent can deal with changing data, we're okay with a
2371 * writable and resizable backing file. */
2372 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2373 if (shared & BLK_PERM_WRITE) {
2374 shared = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2375 } else {
2376 shared = 0;
2377 }
2378
2379 shared |= BLK_PERM_CONSISTENT_READ | BLK_PERM_GRAPH_MOD |
2380 BLK_PERM_WRITE_UNCHANGED;
2381 }
2382
2383 if (bs->open_flags & BDRV_O_INACTIVE) {
2384 shared |= BLK_PERM_WRITE | BLK_PERM_RESIZE;
2385 }
2386
2387 *nperm = perm;
2388 *nshared = shared;
2389 }
2390
2391 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm)
2392 {
2393 static const uint64_t permissions[] = {
2394 [BLOCK_PERMISSION_CONSISTENT_READ] = BLK_PERM_CONSISTENT_READ,
2395 [BLOCK_PERMISSION_WRITE] = BLK_PERM_WRITE,
2396 [BLOCK_PERMISSION_WRITE_UNCHANGED] = BLK_PERM_WRITE_UNCHANGED,
2397 [BLOCK_PERMISSION_RESIZE] = BLK_PERM_RESIZE,
2398 [BLOCK_PERMISSION_GRAPH_MOD] = BLK_PERM_GRAPH_MOD,
2399 };
2400
2401 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions) != BLOCK_PERMISSION__MAX);
2402 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions) != BLK_PERM_ALL + 1);
2403
2404 assert(qapi_perm < BLOCK_PERMISSION__MAX);
2405
2406 return permissions[qapi_perm];
2407 }
2408
2409 static void bdrv_replace_child_noperm(BdrvChild *child,
2410 BlockDriverState *new_bs)
2411 {
2412 BlockDriverState *old_bs = child->bs;
2413 int new_bs_quiesce_counter;
2414 int drain_saldo;
2415
2416 assert(!child->frozen);
2417
2418 if (old_bs && new_bs) {
2419 assert(bdrv_get_aio_context(old_bs) == bdrv_get_aio_context(new_bs));
2420 }
2421
2422 new_bs_quiesce_counter = (new_bs ? new_bs->quiesce_counter : 0);
2423 drain_saldo = new_bs_quiesce_counter - child->parent_quiesce_counter;
2424
2425 /*
2426 * If the new child node is drained but the old one was not, flush
2427 * all outstanding requests to the old child node.
2428 */
2429 while (drain_saldo > 0 && child->role->drained_begin) {
2430 bdrv_parent_drained_begin_single(child, true);
2431 drain_saldo--;
2432 }
2433
2434 if (old_bs) {
2435 /* Detach first so that the recursive drain sections coming from @child
2436 * are already gone and we only end the drain sections that came from
2437 * elsewhere. */
2438 if (child->role->detach) {
2439 child->role->detach(child);
2440 }
2441 QLIST_REMOVE(child, next_parent);
2442 }
2443
2444 child->bs = new_bs;
2445
2446 if (new_bs) {
2447 QLIST_INSERT_HEAD(&new_bs->parents, child, next_parent);
2448
2449 /*
2450 * Detaching the old node may have led to the new node's
2451 * quiesce_counter having been decreased. Not a problem, we
2452 * just need to recognize this here and then invoke
2453 * drained_end appropriately more often.
2454 */
2455 assert(new_bs->quiesce_counter <= new_bs_quiesce_counter);
2456 drain_saldo += new_bs->quiesce_counter - new_bs_quiesce_counter;
2457
2458 /* Attach only after starting new drained sections, so that recursive
2459 * drain sections coming from @child don't get an extra .drained_begin
2460 * callback. */
2461 if (child->role->attach) {
2462 child->role->attach(child);
2463 }
2464 }
2465
2466 /*
2467 * If the old child node was drained but the new one is not, allow
2468 * requests to come in only after the new node has been attached.
2469 */
2470 while (drain_saldo < 0 && child->role->drained_end) {
2471 bdrv_parent_drained_end_single(child);
2472 drain_saldo++;
2473 }
2474 }
2475
2476 /*
2477 * Updates @child to change its reference to point to @new_bs, including
2478 * checking and applying the necessary permisson updates both to the old node
2479 * and to @new_bs.
2480 *
2481 * NULL is passed as @new_bs for removing the reference before freeing @child.
2482 *
2483 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2484 * function uses bdrv_set_perm() to update the permissions according to the new
2485 * reference that @new_bs gets.
2486 */
2487 static void bdrv_replace_child(BdrvChild *child, BlockDriverState *new_bs)
2488 {
2489 BlockDriverState *old_bs = child->bs;
2490 uint64_t perm, shared_perm;
2491
2492 bdrv_replace_child_noperm(child, new_bs);
2493
2494 /*
2495 * Start with the new node's permissions. If @new_bs is a (direct
2496 * or indirect) child of @old_bs, we must complete the permission
2497 * update on @new_bs before we loosen the restrictions on @old_bs.
2498 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2499 * updating the permissions of @new_bs, and thus not purely loosen
2500 * restrictions.
2501 */
2502 if (new_bs) {
2503 bdrv_get_cumulative_perm(new_bs, &perm, &shared_perm);
2504 bdrv_set_perm(new_bs, perm, shared_perm);
2505 }
2506
2507 if (old_bs) {
2508 /* Update permissions for old node. This is guaranteed to succeed
2509 * because we're just taking a parent away, so we're loosening
2510 * restrictions. */
2511 bool tighten_restrictions;
2512 int ret;
2513
2514 bdrv_get_cumulative_perm(old_bs, &perm, &shared_perm);
2515 ret = bdrv_check_perm(old_bs, NULL, perm, shared_perm, NULL,
2516 &tighten_restrictions, NULL);
2517 assert(tighten_restrictions == false);
2518 if (ret < 0) {
2519 /* We only tried to loosen restrictions, so errors are not fatal */
2520 bdrv_abort_perm_update(old_bs);
2521 } else {
2522 bdrv_set_perm(old_bs, perm, shared_perm);
2523 }
2524
2525 /* When the parent requiring a non-default AioContext is removed, the
2526 * node moves back to the main AioContext */
2527 bdrv_try_set_aio_context(old_bs, qemu_get_aio_context(), NULL);
2528 }
2529 }
2530
2531 /*
2532 * This function steals the reference to child_bs from the caller.
2533 * That reference is later dropped by bdrv_root_unref_child().
2534 *
2535 * On failure NULL is returned, errp is set and the reference to
2536 * child_bs is also dropped.
2537 *
2538 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2539 * (unless @child_bs is already in @ctx).
2540 */
2541 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
2542 const char *child_name,
2543 const BdrvChildRole *child_role,
2544 AioContext *ctx,
2545 uint64_t perm, uint64_t shared_perm,
2546 void *opaque, Error **errp)
2547 {
2548 BdrvChild *child;
2549 Error *local_err = NULL;
2550 int ret;
2551
2552 ret = bdrv_check_update_perm(child_bs, NULL, perm, shared_perm, NULL, NULL,
2553 errp);
2554 if (ret < 0) {
2555 bdrv_abort_perm_update(child_bs);
2556 bdrv_unref(child_bs);
2557 return NULL;
2558 }
2559
2560 child = g_new(BdrvChild, 1);
2561 *child = (BdrvChild) {
2562 .bs = NULL,
2563 .name = g_strdup(child_name),
2564 .role = child_role,
2565 .perm = perm,
2566 .shared_perm = shared_perm,
2567 .opaque = opaque,
2568 };
2569
2570 /* If the AioContexts don't match, first try to move the subtree of
2571 * child_bs into the AioContext of the new parent. If this doesn't work,
2572 * try moving the parent into the AioContext of child_bs instead. */
2573 if (bdrv_get_aio_context(child_bs) != ctx) {
2574 ret = bdrv_try_set_aio_context(child_bs, ctx, &local_err);
2575 if (ret < 0 && child_role->can_set_aio_ctx) {
2576 GSList *ignore = g_slist_prepend(NULL, child);
2577 ctx = bdrv_get_aio_context(child_bs);
2578 if (child_role->can_set_aio_ctx(child, ctx, &ignore, NULL)) {
2579 error_free(local_err);
2580 ret = 0;
2581 g_slist_free(ignore);
2582 ignore = g_slist_prepend(NULL, child);
2583 child_role->set_aio_ctx(child, ctx, &ignore);
2584 }
2585 g_slist_free(ignore);
2586 }
2587 if (ret < 0) {
2588 error_propagate(errp, local_err);
2589 g_free(child);
2590 bdrv_abort_perm_update(child_bs);
2591 return NULL;
2592 }
2593 }
2594
2595 /* This performs the matching bdrv_set_perm() for the above check. */
2596 bdrv_replace_child(child, child_bs);
2597
2598 return child;
2599 }
2600
2601 /*
2602 * This function transfers the reference to child_bs from the caller
2603 * to parent_bs. That reference is later dropped by parent_bs on
2604 * bdrv_close() or if someone calls bdrv_unref_child().
2605 *
2606 * On failure NULL is returned, errp is set and the reference to
2607 * child_bs is also dropped.
2608 *
2609 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2610 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2611 */
2612 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
2613 BlockDriverState *child_bs,
2614 const char *child_name,
2615 const BdrvChildRole *child_role,
2616 Error **errp)
2617 {
2618 BdrvChild *child;
2619 uint64_t perm, shared_perm;
2620
2621 bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
2622
2623 assert(parent_bs->drv);
2624 bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
2625 perm, shared_perm, &perm, &shared_perm);
2626
2627 child = bdrv_root_attach_child(child_bs, child_name, child_role,
2628 bdrv_get_aio_context(parent_bs),
2629 perm, shared_perm, parent_bs, errp);
2630 if (child == NULL) {
2631 return NULL;
2632 }
2633
2634 QLIST_INSERT_HEAD(&parent_bs->children, child, next);
2635 return child;
2636 }
2637
2638 static void bdrv_detach_child(BdrvChild *child)
2639 {
2640 QLIST_SAFE_REMOVE(child, next);
2641
2642 bdrv_replace_child(child, NULL);
2643
2644 g_free(child->name);
2645 g_free(child);
2646 }
2647
2648 void bdrv_root_unref_child(BdrvChild *child)
2649 {
2650 BlockDriverState *child_bs;
2651
2652 child_bs = child->bs;
2653 bdrv_detach_child(child);
2654 bdrv_unref(child_bs);
2655 }
2656
2657 /**
2658 * Clear all inherits_from pointers from children and grandchildren of
2659 * @root that point to @root, where necessary.
2660 */
2661 static void bdrv_unset_inherits_from(BlockDriverState *root, BdrvChild *child)
2662 {
2663 BdrvChild *c;
2664
2665 if (child->bs->inherits_from == root) {
2666 /*
2667 * Remove inherits_from only when the last reference between root and
2668 * child->bs goes away.
2669 */
2670 QLIST_FOREACH(c, &root->children, next) {
2671 if (c != child && c->bs == child->bs) {
2672 break;
2673 }
2674 }
2675 if (c == NULL) {
2676 child->bs->inherits_from = NULL;
2677 }
2678 }
2679
2680 QLIST_FOREACH(c, &child->bs->children, next) {
2681 bdrv_unset_inherits_from(root, c);
2682 }
2683 }
2684
2685 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
2686 {
2687 if (child == NULL) {
2688 return;
2689 }
2690
2691 bdrv_unset_inherits_from(parent, child);
2692 bdrv_root_unref_child(child);
2693 }
2694
2695
2696 static void bdrv_parent_cb_change_media(BlockDriverState *bs, bool load)
2697 {
2698 BdrvChild *c;
2699 QLIST_FOREACH(c, &bs->parents, next_parent) {
2700 if (c->role->change_media) {
2701 c->role->change_media(c, load);
2702 }
2703 }
2704 }
2705
2706 /* Return true if you can reach parent going through child->inherits_from
2707 * recursively. If parent or child are NULL, return false */
2708 static bool bdrv_inherits_from_recursive(BlockDriverState *child,
2709 BlockDriverState *parent)
2710 {
2711 while (child && child != parent) {
2712 child = child->inherits_from;
2713 }
2714
2715 return child != NULL;
2716 }
2717
2718 /*
2719 * Sets the backing file link of a BDS. A new reference is created; callers
2720 * which don't need their own reference any more must call bdrv_unref().
2721 */
2722 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
2723 Error **errp)
2724 {
2725 bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
2726 bdrv_inherits_from_recursive(backing_hd, bs);
2727
2728 if (bdrv_is_backing_chain_frozen(bs, backing_bs(bs), errp)) {
2729 return;
2730 }
2731
2732 if (backing_hd) {
2733 bdrv_ref(backing_hd);
2734 }
2735
2736 if (bs->backing) {
2737 bdrv_unref_child(bs, bs->backing);
2738 }
2739
2740 if (!backing_hd) {
2741 bs->backing = NULL;
2742 goto out;
2743 }
2744
2745 bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing,
2746 errp);
2747 /* If backing_hd was already part of bs's backing chain, and
2748 * inherits_from pointed recursively to bs then let's update it to
2749 * point directly to bs (else it will become NULL). */
2750 if (bs->backing && update_inherits_from) {
2751 backing_hd->inherits_from = bs;
2752 }
2753
2754 out:
2755 bdrv_refresh_limits(bs, NULL);
2756 }
2757
2758 /*
2759 * Opens the backing file for a BlockDriverState if not yet open
2760 *
2761 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2762 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2763 * itself, all options starting with "${bdref_key}." are considered part of the
2764 * BlockdevRef.
2765 *
2766 * TODO Can this be unified with bdrv_open_image()?
2767 */
2768 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
2769 const char *bdref_key, Error **errp)
2770 {
2771 char *backing_filename = NULL;
2772 char *bdref_key_dot;
2773 const char *reference = NULL;
2774 int ret = 0;
2775 bool implicit_backing = false;
2776 BlockDriverState *backing_hd;
2777 QDict *options;
2778 QDict *tmp_parent_options = NULL;
2779 Error *local_err = NULL;
2780
2781 if (bs->backing != NULL) {
2782 goto free_exit;
2783 }
2784
2785 /* NULL means an empty set of options */
2786 if (parent_options == NULL) {
2787 tmp_parent_options = qdict_new();
2788 parent_options = tmp_parent_options;
2789 }
2790
2791 bs->open_flags &= ~BDRV_O_NO_BACKING;
2792
2793 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2794 qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
2795 g_free(bdref_key_dot);
2796
2797 /*
2798 * Caution: while qdict_get_try_str() is fine, getting non-string
2799 * types would require more care. When @parent_options come from
2800 * -blockdev or blockdev_add, its members are typed according to
2801 * the QAPI schema, but when they come from -drive, they're all
2802 * QString.
2803 */
2804 reference = qdict_get_try_str(parent_options, bdref_key);
2805 if (reference || qdict_haskey(options, "file.filename")) {
2806 /* keep backing_filename NULL */
2807 } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
2808 qobject_unref(options);
2809 goto free_exit;
2810 } else {
2811 if (qdict_size(options) == 0) {
2812 /* If the user specifies options that do not modify the
2813 * backing file's behavior, we might still consider it the
2814 * implicit backing file. But it's easier this way, and
2815 * just specifying some of the backing BDS's options is
2816 * only possible with -drive anyway (otherwise the QAPI
2817 * schema forces the user to specify everything). */
2818 implicit_backing = !strcmp(bs->auto_backing_file, bs->backing_file);
2819 }
2820
2821 backing_filename = bdrv_get_full_backing_filename(bs, &local_err);
2822 if (local_err) {
2823 ret = -EINVAL;
2824 error_propagate(errp, local_err);
2825 qobject_unref(options);
2826 goto free_exit;
2827 }
2828 }
2829
2830 if (!bs->drv || !bs->drv->supports_backing) {
2831 ret = -EINVAL;
2832 error_setg(errp, "Driver doesn't support backing files");
2833 qobject_unref(options);
2834 goto free_exit;
2835 }
2836
2837 if (!reference &&
2838 bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
2839 qdict_put_str(options, "driver", bs->backing_format);
2840 }
2841
2842 backing_hd = bdrv_open_inherit(backing_filename, reference, options, 0, bs,
2843 &child_backing, errp);
2844 if (!backing_hd) {
2845 bs->open_flags |= BDRV_O_NO_BACKING;
2846 error_prepend(errp, "Could not open backing file: ");
2847 ret = -EINVAL;
2848 goto free_exit;
2849 }
2850
2851 if (implicit_backing) {
2852 bdrv_refresh_filename(backing_hd);
2853 pstrcpy(bs->auto_backing_file, sizeof(bs->auto_backing_file),
2854 backing_hd->filename);
2855 }
2856
2857 /* Hook up the backing file link; drop our reference, bs owns the
2858 * backing_hd reference now */
2859 bdrv_set_backing_hd(bs, backing_hd, &local_err);
2860 bdrv_unref(backing_hd);
2861 if (local_err) {
2862 error_propagate(errp, local_err);
2863 ret = -EINVAL;
2864 goto free_exit;
2865 }
2866
2867 qdict_del(parent_options, bdref_key);
2868
2869 free_exit:
2870 g_free(backing_filename);
2871 qobject_unref(tmp_parent_options);
2872 return ret;
2873 }
2874
2875 static BlockDriverState *
2876 bdrv_open_child_bs(const char *filename, QDict *options, const char *bdref_key,
2877 BlockDriverState *parent, const BdrvChildRole *child_role,
2878 bool allow_none, Error **errp)
2879 {
2880 BlockDriverState *bs = NULL;
2881 QDict *image_options;
2882 char *bdref_key_dot;
2883 const char *reference;
2884
2885 assert(child_role != NULL);
2886
2887 bdref_key_dot = g_strdup_printf("%s.", bdref_key);
2888 qdict_extract_subqdict(options, &image_options, bdref_key_dot);
2889 g_free(bdref_key_dot);
2890
2891 /*
2892 * Caution: while qdict_get_try_str() is fine, getting non-string
2893 * types would require more care. When @options come from
2894 * -blockdev or blockdev_add, its members are typed according to
2895 * the QAPI schema, but when they come from -drive, they're all
2896 * QString.
2897 */
2898 reference = qdict_get_try_str(options, bdref_key);
2899 if (!filename && !reference && !qdict_size(image_options)) {
2900 if (!allow_none) {
2901 error_setg(errp, "A block device must be specified for \"%s\"",
2902 bdref_key);
2903 }
2904 qobject_unref(image_options);
2905 goto done;
2906 }
2907
2908 bs = bdrv_open_inherit(filename, reference, image_options, 0,
2909 parent, child_role, errp);
2910 if (!bs) {
2911 goto done;
2912 }
2913
2914 done:
2915 qdict_del(options, bdref_key);
2916 return bs;
2917 }
2918
2919 /*
2920 * Opens a disk image whose options are given as BlockdevRef in another block
2921 * device's options.
2922 *
2923 * If allow_none is true, no image will be opened if filename is false and no
2924 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2925 *
2926 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2927 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2928 * itself, all options starting with "${bdref_key}." are considered part of the
2929 * BlockdevRef.
2930 *
2931 * The BlockdevRef will be removed from the options QDict.
2932 */
2933 BdrvChild *bdrv_open_child(const char *filename,
2934 QDict *options, const char *bdref_key,
2935 BlockDriverState *parent,
2936 const BdrvChildRole *child_role,
2937 bool allow_none, Error **errp)
2938 {
2939 BlockDriverState *bs;
2940
2941 bs = bdrv_open_child_bs(filename, options, bdref_key, parent, child_role,
2942 allow_none, errp);
2943 if (bs == NULL) {
2944 return NULL;
2945 }
2946
2947 return bdrv_attach_child(parent, bs, bdref_key, child_role, errp);
2948 }
2949
2950 /* TODO Future callers may need to specify parent/child_role in order for
2951 * option inheritance to work. Existing callers use it for the root node. */
2952 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp)
2953 {
2954 BlockDriverState *bs = NULL;
2955 Error *local_err = NULL;
2956 QObject *obj = NULL;
2957 QDict *qdict = NULL;
2958 const char *reference = NULL;
2959 Visitor *v = NULL;
2960
2961 if (ref->type == QTYPE_QSTRING) {
2962 reference = ref->u.reference;
2963 } else {
2964 BlockdevOptions *options = &ref->u.definition;
2965 assert(ref->type == QTYPE_QDICT);
2966
2967 v = qobject_output_visitor_new(&obj);
2968 visit_type_BlockdevOptions(v, NULL, &options, &local_err);
2969 if (local_err) {
2970 error_propagate(errp, local_err);
2971 goto fail;
2972 }
2973 visit_complete(v, &obj);
2974
2975 qdict = qobject_to(QDict, obj);
2976 qdict_flatten(qdict);
2977
2978 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2979 * compatibility with other callers) rather than what we want as the
2980 * real defaults. Apply the defaults here instead. */
2981 qdict_set_default_str(qdict, BDRV_OPT_CACHE_DIRECT, "off");
2982 qdict_set_default_str(qdict, BDRV_OPT_CACHE_NO_FLUSH, "off");
2983 qdict_set_default_str(qdict, BDRV_OPT_READ_ONLY, "off");
2984 qdict_set_default_str(qdict, BDRV_OPT_AUTO_READ_ONLY, "off");
2985
2986 }
2987
2988 bs = bdrv_open_inherit(NULL, reference, qdict, 0, NULL, NULL, errp);
2989 obj = NULL;
2990
2991 fail:
2992 qobject_unref(obj);
2993 visit_free(v);
2994 return bs;
2995 }
2996
2997 static BlockDriverState *bdrv_append_temp_snapshot(BlockDriverState *bs,
2998 int flags,
2999 QDict *snapshot_options,
3000 Error **errp)
3001 {
3002 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3003 char *tmp_filename = g_malloc0(PATH_MAX + 1);
3004 int64_t total_size;
3005 QemuOpts *opts = NULL;
3006 BlockDriverState *bs_snapshot = NULL;
3007 Error *local_err = NULL;
3008 int ret;
3009
3010 /* if snapshot, we create a temporary backing file and open it
3011 instead of opening 'filename' directly */
3012
3013 /* Get the required size from the image */
3014 total_size = bdrv_getlength(bs);
3015 if (total_size < 0) {
3016 error_setg_errno(errp, -total_size, "Could not get image size");
3017 goto out;
3018 }
3019
3020 /* Create the temporary image */
3021 ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
3022 if (ret < 0) {
3023 error_setg_errno(errp, -ret, "Could not get temporary filename");
3024 goto out;
3025 }
3026
3027 opts = qemu_opts_create(bdrv_qcow2.create_opts, NULL, 0,
3028 &error_abort);
3029 qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size, &error_abort);
3030 ret = bdrv_create(&bdrv_qcow2, tmp_filename, opts, errp);
3031 qemu_opts_del(opts);
3032 if (ret < 0) {
3033 error_prepend(errp, "Could not create temporary overlay '%s': ",
3034 tmp_filename);
3035 goto out;
3036 }
3037
3038 /* Prepare options QDict for the temporary file */
3039 qdict_put_str(snapshot_options, "file.driver", "file");
3040 qdict_put_str(snapshot_options, "file.filename", tmp_filename);
3041 qdict_put_str(snapshot_options, "driver", "qcow2");
3042
3043 bs_snapshot = bdrv_open(NULL, NULL, snapshot_options, flags, errp);
3044 snapshot_options = NULL;
3045 if (!bs_snapshot) {
3046 goto out;
3047 }
3048
3049 /* bdrv_append() consumes a strong reference to bs_snapshot
3050 * (i.e. it will call bdrv_unref() on it) even on error, so in
3051 * order to be able to return one, we have to increase
3052 * bs_snapshot's refcount here */
3053 bdrv_ref(bs_snapshot);
3054 bdrv_append(bs_snapshot, bs, &local_err);
3055 if (local_err) {
3056 error_propagate(errp, local_err);
3057 bs_snapshot = NULL;
3058 goto out;
3059 }
3060
3061 out:
3062 qobject_unref(snapshot_options);
3063 g_free(tmp_filename);
3064 return bs_snapshot;
3065 }
3066
3067 /*
3068 * Opens a disk image (raw, qcow2, vmdk, ...)
3069 *
3070 * options is a QDict of options to pass to the block drivers, or NULL for an
3071 * empty set of options. The reference to the QDict belongs to the block layer
3072 * after the call (even on failure), so if the caller intends to reuse the
3073 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3074 *
3075 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3076 * If it is not NULL, the referenced BDS will be reused.
3077 *
3078 * The reference parameter may be used to specify an existing block device which
3079 * should be opened. If specified, neither options nor a filename may be given,
3080 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3081 */
3082 static BlockDriverState *bdrv_open_inherit(const char *filename,
3083 const char *reference,
3084 QDict *options, int flags,
3085 BlockDriverState *parent,
3086 const BdrvChildRole *child_role,
3087 Error **errp)
3088 {
3089 int ret;
3090 BlockBackend *file = NULL;
3091 BlockDriverState *bs;
3092 BlockDriver *drv = NULL;
3093 BdrvChild *child;
3094 const char *drvname;
3095 const char *backing;
3096 Error *local_err = NULL;
3097 QDict *snapshot_options = NULL;
3098 int snapshot_flags = 0;
3099
3100 assert(!child_role || !flags);
3101 assert(!child_role == !parent);
3102
3103 if (reference) {
3104 bool options_non_empty = options ? qdict_size(options) : false;
3105 qobject_unref(options);
3106
3107 if (filename || options_non_empty) {
3108 error_setg(errp, "Cannot reference an existing block device with "
3109 "additional options or a new filename");
3110 return NULL;
3111 }
3112
3113 bs = bdrv_lookup_bs(reference, reference, errp);
3114 if (!bs) {
3115 return NULL;
3116 }
3117
3118 bdrv_ref(bs);
3119 return bs;
3120 }
3121
3122 bs = bdrv_new();
3123
3124 /* NULL means an empty set of options */
3125 if (options == NULL) {
3126 options = qdict_new();
3127 }
3128
3129 /* json: syntax counts as explicit options, as if in the QDict */
3130 parse_json_protocol(options, &filename, &local_err);
3131 if (local_err) {
3132 goto fail;
3133 }
3134
3135 bs->explicit_options = qdict_clone_shallow(options);
3136
3137 if (child_role) {
3138 bs->inherits_from = parent;
3139 child_role->inherit_options(&flags, options,
3140 parent->open_flags, parent->options);
3141 }
3142
3143 ret = bdrv_fill_options(&options, filename, &flags, &local_err);
3144 if (local_err) {
3145 goto fail;
3146 }
3147
3148 /*
3149 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3150 * Caution: getting a boolean member of @options requires care.
3151 * When @options come from -blockdev or blockdev_add, members are
3152 * typed according to the QAPI schema, but when they come from
3153 * -drive, they're all QString.
3154 */
3155 if (g_strcmp0(qdict_get_try_str(options, BDRV_OPT_READ_ONLY), "on") &&
3156 !qdict_get_try_bool(options, BDRV_OPT_READ_ONLY, false)) {
3157 flags |= (BDRV_O_RDWR | BDRV_O_ALLOW_RDWR);
3158 } else {
3159 flags &= ~BDRV_O_RDWR;
3160 }
3161
3162 if (flags & BDRV_O_SNAPSHOT) {
3163 snapshot_options = qdict_new();
3164 bdrv_temp_snapshot_options(&snapshot_flags, snapshot_options,
3165 flags, options);
3166 /* Let bdrv_backing_options() override "read-only" */
3167 qdict_del(options, BDRV_OPT_READ_ONLY);
3168 bdrv_backing_options(&flags, options, flags, options);
3169 }
3170
3171 bs->open_flags = flags;
3172 bs->options = options;
3173 options = qdict_clone_shallow(options);
3174
3175 /* Find the right image format driver */
3176 /* See cautionary note on accessing @options above */
3177 drvname = qdict_get_try_str(options, "driver");
3178 if (drvname) {
3179 drv = bdrv_find_format(drvname);
3180 if (!drv) {
3181 error_setg(errp, "Unknown driver: '%s'", drvname);
3182 goto fail;
3183 }
3184 }
3185
3186 assert(drvname || !(flags & BDRV_O_PROTOCOL));
3187
3188 /* See cautionary note on accessing @options above */
3189 backing = qdict_get_try_str(options, "backing");
3190 if (qobject_to(QNull, qdict_get(options, "backing")) != NULL ||
3191 (backing && *backing == '\0'))
3192 {
3193 if (backing) {
3194 warn_report("Use of \"backing\": \"\" is deprecated; "
3195 "use \"backing\": null instead");
3196 }
3197 flags |= BDRV_O_NO_BACKING;
3198 qdict_del(bs->explicit_options, "backing");
3199 qdict_del(bs->options, "backing");
3200 qdict_del(options, "backing");
3201 }
3202
3203 /* Open image file without format layer. This BlockBackend is only used for
3204 * probing, the block drivers will do their own bdrv_open_child() for the
3205 * same BDS, which is why we put the node name back into options. */
3206 if ((flags & BDRV_O_PROTOCOL) == 0) {
3207 BlockDriverState *file_bs;
3208
3209 file_bs = bdrv_open_child_bs(filename, options, "file", bs,
3210 &child_file, true, &local_err);
3211 if (local_err) {
3212 goto fail;
3213 }
3214 if (file_bs != NULL) {
3215 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3216 * looking at the header to guess the image format. This works even
3217 * in cases where a guest would not see a consistent state. */
3218 file = blk_new(bdrv_get_aio_context(file_bs), 0, BLK_PERM_ALL);
3219 blk_insert_bs(file, file_bs, &local_err);
3220 bdrv_unref(file_bs);
3221 if (local_err) {
3222 goto fail;
3223 }
3224
3225 qdict_put_str(options, "file", bdrv_get_node_name(file_bs));
3226 }
3227 }
3228
3229 /* Image format probing */
3230 bs->probed = !drv;
3231 if (!drv && file) {
3232 ret = find_image_format(file, filename, &drv, &local_err);
3233 if (ret < 0) {
3234 goto fail;
3235 }
3236 /*
3237 * This option update would logically belong in bdrv_fill_options(),
3238 * but we first need to open bs->file for the probing to work, while
3239 * opening bs->file already requires the (mostly) final set of options
3240 * so that cache mode etc. can be inherited.
3241 *
3242 * Adding the driver later is somewhat ugly, but it's not an option
3243 * that would ever be inherited, so it's correct. We just need to make
3244 * sure to update both bs->options (which has the full effective
3245 * options for bs) and options (which has file.* already removed).
3246 */
3247 qdict_put_str(bs->options, "driver", drv->format_name);
3248 qdict_put_str(options, "driver", drv->format_name);
3249 } else if (!drv) {
3250 error_setg(errp, "Must specify either driver or file");
3251 goto fail;
3252 }
3253
3254 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3255 assert(!!(flags & BDRV_O_PROTOCOL) == !!drv->bdrv_file_open);
3256 /* file must be NULL if a protocol BDS is about to be created
3257 * (the inverse results in an error message from bdrv_open_common()) */
3258 assert(!(flags & BDRV_O_PROTOCOL) || !file);
3259
3260 /* Open the image */
3261 ret = bdrv_open_common(bs, file, options, &local_err);
3262 if (ret < 0) {
3263 goto fail;
3264 }
3265
3266 if (file) {
3267 blk_unref(file);
3268 file = NULL;
3269 }
3270
3271 /* If there is a backing file, use it */
3272 if ((flags & BDRV_O_NO_BACKING) == 0) {
3273 ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
3274 if (ret < 0) {
3275 goto close_and_fail;
3276 }
3277 }
3278
3279 /* Remove all children options and references
3280 * from bs->options and bs->explicit_options */
3281 QLIST_FOREACH(child, &bs->children, next) {
3282 char *child_key_dot;
3283 child_key_dot = g_strdup_printf("%s.", child->name);
3284 qdict_extract_subqdict(bs->explicit_options, NULL, child_key_dot);
3285 qdict_extract_subqdict(bs->options, NULL, child_key_dot);
3286 qdict_del(bs->explicit_options, child->name);
3287 qdict_del(bs->options, child->name);
3288 g_free(child_key_dot);
3289 }
3290
3291 /* Check if any unknown options were used */
3292 if (qdict_size(options) != 0) {
3293 const QDictEntry *entry = qdict_first(options);
3294 if (flags & BDRV_O_PROTOCOL) {
3295 error_setg(errp, "Block protocol '%s' doesn't support the option "
3296 "'%s'", drv->format_name, entry->key);
3297 } else {
3298 error_setg(errp,
3299 "Block format '%s' does not support the option '%s'",
3300 drv->format_name, entry->key);
3301 }
3302
3303 goto close_and_fail;
3304 }
3305
3306 bdrv_parent_cb_change_media(bs, true);
3307
3308 qobject_unref(options);
3309 options = NULL;
3310
3311 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3312 * temporary snapshot afterwards. */
3313 if (snapshot_flags) {
3314 BlockDriverState *snapshot_bs;
3315 snapshot_bs = bdrv_append_temp_snapshot(bs, snapshot_flags,
3316 snapshot_options, &local_err);
3317 snapshot_options = NULL;
3318 if (local_err) {
3319 goto close_and_fail;
3320 }
3321 /* We are not going to return bs but the overlay on top of it
3322 * (snapshot_bs); thus, we have to drop the strong reference to bs
3323 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3324 * though, because the overlay still has a reference to it. */
3325 bdrv_unref(bs);
3326 bs = snapshot_bs;
3327 }
3328
3329 return bs;
3330
3331 fail:
3332 blk_unref(file);
3333 qobject_unref(snapshot_options);
3334 qobject_unref(bs->explicit_options);
3335 qobject_unref(bs->options);
3336 qobject_unref(options);
3337 bs->options = NULL;
3338 bs->explicit_options = NULL;
3339 bdrv_unref(bs);
3340 error_propagate(errp, local_err);
3341 return NULL;
3342
3343 close_and_fail:
3344 bdrv_unref(bs);
3345 qobject_unref(snapshot_options);
3346 qobject_unref(options);
3347 error_propagate(errp, local_err);
3348 return NULL;
3349 }
3350
3351 BlockDriverState *bdrv_open(const char *filename, const char *reference,
3352 QDict *options, int flags, Error **errp)
3353 {
3354 return bdrv_open_inherit(filename, reference, options, flags, NULL,
3355 NULL, errp);
3356 }
3357
3358 /* Return true if the NULL-terminated @list contains @str */
3359 static bool is_str_in_list(const char *str, const char *const *list)
3360 {
3361 if (str && list) {
3362 int i;
3363 for (i = 0; list[i] != NULL; i++) {
3364 if (!strcmp(str, list[i])) {
3365 return true;
3366 }
3367 }
3368 }
3369 return false;
3370 }
3371
3372 /*
3373 * Check that every option set in @bs->options is also set in
3374 * @new_opts.
3375 *
3376 * Options listed in the common_options list and in
3377 * @bs->drv->mutable_opts are skipped.
3378 *
3379 * Return 0 on success, otherwise return -EINVAL and set @errp.
3380 */
3381 static int bdrv_reset_options_allowed(BlockDriverState *bs,
3382 const QDict *new_opts, Error **errp)
3383 {
3384 const QDictEntry *e;
3385 /* These options are common to all block drivers and are handled
3386 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3387 const char *const common_options[] = {
3388 "node-name", "discard", "cache.direct", "cache.no-flush",
3389 "read-only", "auto-read-only", "detect-zeroes", NULL
3390 };
3391
3392 for (e = qdict_first(bs->options); e; e = qdict_next(bs->options, e)) {
3393 if (!qdict_haskey(new_opts, e->key) &&
3394 !is_str_in_list(e->key, common_options) &&
3395 !is_str_in_list(e->key, bs->drv->mutable_opts)) {
3396 error_setg(errp, "Option '%s' cannot be reset "
3397 "to its default value", e->key);
3398 return -EINVAL;
3399 }
3400 }
3401
3402 return 0;
3403 }
3404
3405 /*
3406 * Returns true if @child can be reached recursively from @bs
3407 */
3408 static bool bdrv_recurse_has_child(BlockDriverState *bs,
3409 BlockDriverState *child)
3410 {
3411 BdrvChild *c;
3412
3413 if (bs == child) {
3414 return true;
3415 }
3416
3417 QLIST_FOREACH(c, &bs->children, next) {
3418 if (bdrv_recurse_has_child(c->bs, child)) {
3419 return true;
3420 }
3421 }
3422
3423 return false;
3424 }
3425
3426 /*
3427 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3428 * reopen of multiple devices.
3429 *
3430 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3431 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3432 * be created and initialized. This newly created BlockReopenQueue should be
3433 * passed back in for subsequent calls that are intended to be of the same
3434 * atomic 'set'.
3435 *
3436 * bs is the BlockDriverState to add to the reopen queue.
3437 *
3438 * options contains the changed options for the associated bs
3439 * (the BlockReopenQueue takes ownership)
3440 *
3441 * flags contains the open flags for the associated bs
3442 *
3443 * returns a pointer to bs_queue, which is either the newly allocated
3444 * bs_queue, or the existing bs_queue being used.
3445 *
3446 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3447 */
3448 static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
3449 BlockDriverState *bs,
3450 QDict *options,
3451 const BdrvChildRole *role,
3452 QDict *parent_options,
3453 int parent_flags,
3454 bool keep_old_opts)
3455 {
3456 assert(bs != NULL);
3457
3458 BlockReopenQueueEntry *bs_entry;
3459 BdrvChild *child;
3460 QDict *old_options, *explicit_options, *options_copy;
3461 int flags;
3462 QemuOpts *opts;
3463
3464 /* Make sure that the caller remembered to use a drained section. This is
3465 * important to avoid graph changes between the recursive queuing here and
3466 * bdrv_reopen_multiple(). */
3467 assert(bs->quiesce_counter > 0);
3468
3469 if (bs_queue == NULL) {
3470 bs_queue = g_new0(BlockReopenQueue, 1);
3471 QTAILQ_INIT(bs_queue);
3472 }
3473
3474 if (!options) {
3475 options = qdict_new();
3476 }
3477
3478 /* Check if this BlockDriverState is already in the queue */
3479 QTAILQ_FOREACH(bs_entry, bs_queue, entry) {
3480 if (bs == bs_entry->state.bs) {
3481 break;
3482 }
3483 }
3484
3485 /*
3486 * Precedence of options:
3487 * 1. Explicitly passed in options (highest)
3488 * 2. Retained from explicitly set options of bs
3489 * 3. Inherited from parent node
3490 * 4. Retained from effective options of bs
3491 */
3492
3493 /* Old explicitly set values (don't overwrite by inherited value) */
3494 if (bs_entry || keep_old_opts) {
3495 old_options = qdict_clone_shallow(bs_entry ?
3496 bs_entry->state.explicit_options :
3497 bs->explicit_options);
3498 bdrv_join_options(bs, options, old_options);
3499 qobject_unref(old_options);
3500 }
3501
3502 explicit_options = qdict_clone_shallow(options);
3503
3504 /* Inherit from parent node */
3505 if (parent_options) {
3506 flags = 0;
3507 role->inherit_options(&flags, options, parent_flags, parent_options);
3508 } else {
3509 flags = bdrv_get_flags(bs);
3510 }
3511
3512 if (keep_old_opts) {
3513 /* Old values are used for options that aren't set yet */
3514 old_options = qdict_clone_shallow(bs->options);
3515 bdrv_join_options(bs, options, old_options);
3516 qobject_unref(old_options);
3517 }
3518
3519 /* We have the final set of options so let's update the flags */
3520 options_copy = qdict_clone_shallow(options);
3521 opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
3522 qemu_opts_absorb_qdict(opts, options_copy, NULL);
3523 update_flags_from_options(&flags, opts);
3524 qemu_opts_del(opts);
3525 qobject_unref(options_copy);
3526
3527 /* bdrv_open_inherit() sets and clears some additional flags internally */
3528 flags &= ~BDRV_O_PROTOCOL;
3529 if (flags & BDRV_O_RDWR) {
3530 flags |= BDRV_O_ALLOW_RDWR;
3531 }
3532
3533 if (!bs_entry) {
3534 bs_entry = g_new0(BlockReopenQueueEntry, 1);
3535 QTAILQ_INSERT_TAIL(bs_queue, bs_entry, entry);
3536 } else {
3537 qobject_unref(bs_entry->state.options);
3538 qobject_unref(bs_entry->state.explicit_options);
3539 }
3540
3541 bs_entry->state.bs = bs;
3542 bs_entry->state.options = options;
3543 bs_entry->state.explicit_options = explicit_options;
3544 bs_entry->state.flags = flags;
3545
3546 /* This needs to be overwritten in bdrv_reopen_prepare() */
3547 bs_entry->state.perm = UINT64_MAX;
3548 bs_entry->state.shared_perm = 0;
3549
3550 /*
3551 * If keep_old_opts is false then it means that unspecified
3552 * options must be reset to their original value. We don't allow
3553